C. Barry Carter · David B.

Williams
Editors

Transmission Electron
Microscopy
Diffraction, Imaging, and Spectrometry

Foreword by Sir John Meurig Thomas

Transmission Electron Microscopy

Licensed to ([email protected])
C. Barry Carter · David B. Williams (Eds.)

Transmission Electron
Microscopy
Diffraction, Imaging, and Spectrometry
Editors
C. Barry Carter David B. Williams
University of Connecticut The Ohio State University
Storrs, USA Columbus, USA

ISBN 978-3-319-26649-7              ISBN 978-3-319-26651-0 (eBook)

DOI 10.1007/978-3-319-26651-0

Library of Congress Control Number: 2016945257

© Springer International Publishing Switzerland 2016

This work is subject to copyright. All rights are reserved by the Publisher, whether the whole or part of the material is
concerned, specifically the rights of translation, reprinting, reuse of illustrations, recitation, broadcasting, reproduction on
microfilms or in any other physical way, and transmission or information storage and retrieval, electronic adaptation, com-
puter software, or by similar or dissimilar methodology now known or hereafter developed.
The use of general descriptive names, registered names, trademarks, service marks, etc. in this publication does not imply,
even in the absence of a specific statement, that such names are exempt from the relevant protective laws and regulations
and therefore free for general use.
The publisher, the authors and the editors are safe to assume that the advice and information in this book are believed to be
true and accurate at the date of publication. Neither the publisher nor the authors or the editors give a warranty, express or
implied, with respect to the material contained herein or for any errors or omissions that may have been made.

Printed on acid-free paper

This Springer International Publishing Switzerland imprint is published by Springer Nature

The registered company is Springer-Verlag GmbH Berlin Heidelberg
 

Dedicated to Bryony Carter and Margie Williams

Foreword by Sir John Thomas of the Companion
to Williams and Carter’s book on TEM

Ever since 1996, when the first edition of Transmission Electron Microscopy: A Textbook for Materials
Science (by David Williams and Barry Carter) appeared, this became the favoured and standard text for
all those interested in mastering the electron microscopic examination of materials. It was, and remains,
the vade mecum of choice. With its massive repository of highly relevant information and advice, teem-
ing with attractive pedagogical accoutrements, this text gained almost instant, worldwide popularity.

The second edition, published in 2009, contained many new features, prompted primarily by the growth
of the subject and the arrival of a range of powerful additional variants of transmission electron micros-
copy (TEM). It had become fully apparent at that time that TEM is not just a widely useful investigative
tool, but also, in its most modern form, a near complete chemical and structural laboratory, from which
a multitude of properties pertaining to condensed matter may be extracted.

In the intervening years it has become possible to add yet further, powerful variants for identification
and characterization of both solids and, increasingly of liquids and surfaces, at evermore impressive
resolutions – spatial, spectral and temporal. For example, modern scanning transmission electron mi-
croscopes (STEM), used principally for the retrieval of energy-dispersive X-rays (EDX) and electron
energy loss (EEL), as may be gleaned from Chap. 11 of this Companion, can routinely generate spectra
of images in which behind every image pixel lies a complete (EDX or EEL) spectrum. Moreover, by
recording a tilt series of such spectrum images, 4D ‘spectrum tomograms’ produce spectral (and have
chemical) information at every real space voxel.

One of the most dramatic advances made since the second edition of W&C appeared is the vast im-
provement (by nearly ten orders of magnitude) in temporal resolution that has been accomplished,
largely through the innovative work of Zewail and colleagues in California. Up until less than a decade
ago, electron microscopists recorded dynamic changes of specimens in TEMs at around millisecond
resolution. Nowadays, sub-picosecond resolutions can be attained, thereby uncovering a whole host of
new condensed-matter phenomena.

The emergence of major new advances such as this, along with those in tomography, holography,
EELS and in ‘aberration corrected’ imaging, has, inter alia, persuaded W&C to enlist a collection of
world-renowned experts to expatiate their knowledge following the same admirable pedagogic approach
of the progenitor text. Nowadays, thanks to modern versions of electron microscopes, the structure and
composition of matter in all realms can be elucidated at unprecedented resolution in both space and
time; one can ‘see’ individual atoms, and follow the acts of bond formation and rupture. All this opens
up ways, outlined by contributors to this admirable Companion, to even greater insights and informa-
tion in the years to come. It is reassuring to learn that W&C intend to produce further editions of this
Companion when the time is ripe to do so.

When I decided to turn to TEM as an investigative tool, more than fifty years ago, other chemists re-
garded my decision with incredulity and perplexity. ‘Of what possible use could TEM be to the chem-
ist?’ was the refrain I heard. That was in the era when chemists believed that only X-ray crystallography
could enlighten chemists about the nature of solids. It was fondly believed by most chemists of that
generation that all crystals were, in effect, a paradise of faultless regularity. But I had already discovered
that the reactivity of solids, as well as their electronic and excitonic behaviour, is intimately associated
with crystalline defects, and TEM, in its numerous variants, could answer most of my questions. These

vii
viii Foreword by Sir John Thomas of the Companion to Williams and Carter’s book on TEM

days, no self-respecting Department of Chemistry anywhere in the world can function properly with-
out an electron microscope. Indeed, nanoscience and nanotechnology, which are of importance to all
physical and biological sciences, are impossible to pursue without a TEM. Today, such is the power of
direct-detection cameras (described in Chap. 2), that many X-ray crystallographers have now turned
to the TEM to solve very complicated biologically significant structures like large ribosomal subunits
from human mitochondria. It is most gratifying that this Companion text is likely to satisfy the scientific
appetites of all complexions of investigators of condensed matter.

I warmly commend this Companion and, in particular, the way in which it follows the ethos of the
original W&C volumes by posing two different levels of questions (Q and T), and the use “traffic light”
boxes.

Sir John Meurig Thomas

Hon. Professor at the Department of Materials Science and Metallurgy,
University of Cambridge, and Former Director of the Royal Institution of G.B., London
Transmission Electron Microscopy: Diffraction, Imaging,
and Spectrometry

Preface

In the prefaces to the first and second editions of Williams & Carter, we asked the same question: “How
is this book different from the many other TEM books on the market?” Our answer was, in essence, that
these volumes are true textbooks, written to be studied by undergraduates and graduates, constructed in
lecture-size segments and (in the softbound versions) able to be used at the console of the microscope.
Perhaps the most distinguishing feature of the books was that, unlike all previous TEM texts, we wrote
them as we taught in the classroom: in an informal style, interspersed with side comments and the occa-
sional attempted joke – which rendered translation into other languages too challenging. Around 20,000
hard copies have been printed to date, and hundreds of thousands of chapters have been downloaded.

So why do we need a Companion Volume, and why do we refer to it as the Companion Volume? The
answer requires recounting a little history, which, as we’ve indicated in prior prefaces, is something
we enjoy. We see this book as filling a need that is really a reflection on what has happened to the TEM
field in the decades since Williams first suggested the need for such a book to Carter in the living room
of the Carter home in Ithaca, NY, one cold spring morning in 1985, and which is why Carter proposed
that it should be W&C. Buoyed by excess coffee (see Section 34.2 of the second edition) we decided
to go ahead and write. At first, we decided that a group of four complementary experts could cover the
principal aspects of the TEM field at that time (hence the division of the book into four parts). Such
a team was assembled. However, we found that textbook writing was not for the faint of heart. Soon
after we started putting fingers to the keyboards of our $2,500(!) Apple Macintoshes and mailing floppy
disks to each other containing our initial attempts, we were reduced to just W&C.

Despite our lack of expertise in certain key areas of the TEM field, we managed, by 1996, to complete
W&CI. Then, no doubt due to the same excesses of caffeine, we decided in 2003 to rewrite the text, and
so W&CII was born in 2009. When starting W&CII we recognized that there was not much in W&CI
that we could omit! But it was already a macrotome. We also realized that writing another full book on
all the new things that were happening in TEM was not what we had time to do – it would take another
10 years at least. (It did.) So we invited a few of our long-time friends and colleagues who were won-
derfully qualified to write particular chapters; the ones whose photos you see in these pages are still our
friends, we hope. So the Companion Volume was conceived as a collection of chapters that would be
written by world experts on topics that are either perennially important or really new and fascinating,
and that we could keep current without sacrificing material in W&C. It may not all be conventional
wisdom but we decided that this volume would be CW, or C&W for consistency.

Our own careers have mirrored the rapid evolution of TEM. If you have nothing better to do than read
the previous prefaces to our texts, you’ll know that W&C have both evolved from our professorial
positions at Lehigh and Cornell back in 1985. We have both taken on roles that required us (like all
TEM operators) to broaden our skills and face more challenging responsibilities. We have each moved
on to (two) different universities. Our children of the 1970 s and 80 s are now independent adults with
their own twenty-first-century families. Our parents (to whom W&C is dedicated) have all passed on.
Yet, particularly important to us, our wives remain with us. Without them, none of this would ever have
happened. Thousands of successful TEM students owe their thanks to Bryony and Margie, as do we.

So, taking a leaf out of previous prefaces, we ask anew, why is this Companion Volume different from
the many multi-author TEM review texts that are published at regular intervals? First, it builds directly

ix
x Preface

on W&C, which is extensively referenced throughout. Second, we have taken the high-quality writing
of our friends and attempted to give it a W&C flavor, by changing it into informal script where possible.
We apologize to all of them when, in doing so, we have destroyed some of their brilliant phraseology.
Nevertheless, we believe that, compared to similar texts, we have managed to bring a greater degree of
homogeneity to the writings of multiple authors from many different countries. We apologize that a few
of the chapters require two or three (or four) lectures and that the 17 chapters will take a full semester
to cover. A consequence of this approach is that, except in the About the Authors pages, rather than
ascribe individual chapters to individual author(s), we simply acknowledge all our friends up front. If,
or when, you are well versed in the TEM field, it should not be difficult to work out who wrote what!
Perhaps a good exercise for all you students who are relatively new to TEM would be to search the
literature and come to your own conclusions.

We’ll mention a few idiosyncrasies that you’ll find in the text. One of us has a thing about hyphens and
loves Lynne Truss’s book about the panda. In spite of this, we have tried to omit all punctuation when
writing equations: punctuation is supposed to help, not confuse. We’ve used our traffic-light boxes
again: if it’s red, be sure to stop for a moment; if it’s amber – you’ve been warned; if it’s green, take
note but move on.

While we may not have succeeded in making this Companion Volume as comprehensive and uniform as
we made the original W&C editions, we do hope that all you students who enjoyed those first volumes
will find this one to be equally valuable. It has been one heck of a ride, and this text is the technical
apex of both our professional careers, opening doors that we never dreamed of 30 years ago. TEM
continues to grow as a discipline and as an instrument: it is the essential tool for studying nanomaterials.
What happens on the nanoscale determines what happens, period, and only the TEM can tell us about
structure and composition with accuracy and precision at the nanoscale.

Finally, we thank our friend Sir John Thomas for writing the Foreword to this volume and remember
our friend the late Gareth Thomas who wrote the Foreword to W&C.

Albuquerque, Storrs and Columbus, September 2015

Thanks to Our Friends

Most of the authors have discussed their chapters with colleagues, received suggestions and crit-
icisms, and then updated and improved their chapters. As you will appreciate from our ‘People’
sections, we also recognize that we are all building on the contributions of earlier writers and re-
searchers. The editors and authors have been privileged to know many of them personally – this is
still a young field. Here we recognize those who specifically helped us in writing these chapters.

Alwyn Eades

thanks Jean-Paul Morniroli, Mike Kaufman, and John Spence, who read an early version of his chapter;
each of them made suggestions that resulted in important improvements.

Pieter Kruit

chapter is based on knowledge that was acquired when he and his then PhD student Merijn Bronsgeest
worked closely with Greg Schwind and Lyn Swanson of FEI Beamtech.

Katie Jungjohann

and CBC acknowledge the Center for Integrated Nanotechnologies (CINT), a DOE-BES supported
national user facility located at both Sandia and Los Alamos National Laboratories. Sandia, where KJ
is personally located, is a multiprogram laboratory operated by Sandia Corporation, a Lockheed Martin
Company, for the United States Department of Energy’s (DOE) National Nuclear Security Adminis-
tration (NNSA) under contract DE-AC0494AL85000. They thank Matt Janish, Yang Liu, Jiangwei
Wang, Joseph Grogan, Madeline Dukes, Paul Kotula, Joe Michael, Bill Mook, Khalid Hattar, Blythe
Clark, Katie Harrison, Andrew Leenheer, Eric Stach, Dave Mitlin, Grant Norton, Jessica Vanderburg,
Kevin Zavadil, Neal Shinn, Sean Hearne, and Charles Barbour for their support and their help in so
many ways.

Grace Burke

would like to thank her many colleagues and collaborators over the years, especially Tom Nuhfer
(Carnegie-Mellon University) and Ram Bajaj (Bettis) for their long-time collaborations and friendship,
and to Joven Lim (University of Manchester). The examples used in this chapter relied on the excellent
TEM specimens prepared by Jim Haugh (retired) of Westinghouse, whose expertise and friendship have
been so important over the years. She would also like to thank Mike for his patience and proof-reading.

xi
xii Thanks to Our Friends

Stephen J. Pennycook

is very grateful to all his collaborators in the work described in his chapter: E. Abe, B. Alen, L. J. Allen,
H. S. Baik, B. Beaumont, P. F. Becher, A. Y. Borisevich, N. D. Browning, M. F. Chisholm, A. J. D’Al-
fonso, N. Dellby, S. Doh, G. Duscher, M. M. Erwin, J. P. Faurie, L. C. Feldman, A. G. Ferridge, S. D.
Findlay, D. Fuster, P. L. Galindo, P. Gibart, M. V. Glazoff, L. Gonzalez, Y. Gonzalez, T. R. Gosnell,
J.-C. Idrobo, D. E. Jesson, A. V. Kadavanich, M. Kim, T. C. Kippeny, O. L. Krivanek, J. H. Lee, J. T.
Luck, A. R. Lupini, J. R. McBride, A. J. McGibbon, M. M. McGibbon, S. I. Molina, M. F. Murfitt, P.
D. Nellist, P. D. Nellist, F. Omnes, M. P. Oxley, G. S. Painter, S. T. Pantelides, Y. Peng, R. C. Puetter,
S. N. Rashkeev, K. G. Roberts, S. J. Rosenthal, D. L. Sales, M. J. Seddon, W. A. Shelton, N. Shibata,
W. H. Sides, S. Sivananthan, K. Sohlberg, Z. S. Szilagyi, M. Tanaka, M. Terauchi, J. Treadway, A. P.
Tsai, K. van Benthem, M. Varela, L. G. Wang, S. W. Wang, Y. Xin, A. Yahil, Y. F. Yan.

Paul Midgley

thanks all his collaborators, past and present, for their contributions to the work described in the chapters.

Matthew Weyland

thanks Chris Boothroyd, Owen Saxton, Ron Broom, Rafal Dunin-Borkowski, and Sir John Meurig
Thomas for their critical input to the early development of electron tomography in materials science.

Paul Kotula

thanks Peter Duncumb for pioneering X-ray mapping, Legge and Hammond for developing true
spectral imaging, and Rick Mott and the late John Friel for bringing modern spectral imaging to the
microanalysis community. He would like to thank Michael Rye of Sandia for the preparation of most
of the FIB specimens presented here. As we all know, the specimen preparation can be the most critical
part of any materials analysis. He would also like to thank Michael Keenan (retired) and Mark van
Benthem from Sandia National Laboratories for critical insight and helpful discussions about MSA
over the years. Thanks also to Chad Parish (Oak Ridge National Laboratory), Josh Sugar (Sandia),
Katie Jungjohann (Sandia), Yang Liu (North Carolina State University), Shreyas Rajasekhara (Intel)
for insightful comments that improved this chapter. Sandia is a multiprogram laboratory operated by
Sandia Corporation, a Lockheed Martin Company, for the United States National Nuclear Security
Administration, part of the Department of Energy (DOE), under contract DE-AC04 94AL85000.

Masashi Watanabe

wishes to thank Prof. David Williams for his thoughtful supervision for many years. In collaboration
with Prof. Williams, the ζ-factor method and MSA plug-ins were developed. In addition, the author
would like to thank Prof. Christopher Kiely, Mr. David Ackland, Mr. Bill Mushock, Dr. Rob Keyse
and colleagues at Lehigh, Prof. Zenji Horita at Kyushu University, Prof. Ray Egerton at University
of Alberta, Dr. Kazuo Ishizuka at HREM Research Inc., Dr. Hidetaka Sawada, Mr. Eiji Okunishi and
Mr. Masahiko Kanno at JEOL, Mr. Shintaro Yazuka, Mr. Toshihiro Aoki, Mr. Toshihiro Nomura, Dr.
Masahiro Kawasaki and Dr. Tom Isabell at JEOL USA, Dr. Toshie Yaguchi at Hitachi Hitechnologies,
and Dr. Nestor Zaluzec at Algonne National Laboratory for their collaboration.
List of Contributors

C. Barry Carter Editor and Chapter 2

University of Connecticut C. Barry Carter is the Editor-in-Chief of the Journal of Materials

[email protected] Science and a CINT Distinguished Affiliate Scientist. He ist the
Past President of the International Federation of Societies for
Microscopy. He teaches at UConn.

David B. Williams Editor

The Ohio State University, David B. Williams is the Monte Ahuja Endowed Dean’s Chair,
College of Engineering Executive Dean of The Professional Colleges and Dean of the College
[email protected] of Engineering at The Ohio State University.

Pieter Kruit Chapter 1

Delft University of Technology, Pieter Kruit is Professor of Physics at Delft University of

Faculty of Applied Sciences Technology in the Netherlands and has 30 years of experience
[email protected] in developing novel electron- and ion-optical instruments for
microscopy and lithography.

Katie Jungjohann Chapter 2

CINT, Katie Jungjohann is the TEM Research Leader at CINT, Sandia/LANL.

Sandia National Her research is focused on the development and use of operando TEM
Laboratories techniques to understand the structural/chemical changes to nanomate-
[email protected] rials, especially in the electrochemical cell discovery platform.

Grace Burke Chapter 3

University of Manchester, Grace Burke is a Professor in the School of Materials at the

School of Materials University of Manchester, having spent many years in industry,
[email protected] where her research involved microstructural characterization
applied to understanding materials behavior. She is Director
of both the Materials Performance Centre and the Electron
Microscopy Centre.

xiii

Licensed to ([email protected])
xiv List of Contributors

Alwyn Eades Chapter 4

Lehigh University, Alwyn Eades worked on electron microscopy and, especially,

Department of Materials electron diffraction for more than half a century and now enjoys
Science & Engineering retirement.
[email protected]

John Spence Chapter 5

Arizona State University, John Spence FRS (ForMem) is the Richard Snell Professor of Physics
Department of Physics at Arizona State University, author of ‘High Resolution Electron
[email protected] Microscopy’, recipient of the Cowley and Buerger Medals, and Director
of Science for the NSF BioXFEL Science and Technology Center.

Bernhard Schaffer Chapter 6

Gatan Inc. Bernhard Schaffer did his PhD in developing EFTEM SI

[email protected] techniques at FELMI/ZFE Graz before working with Cs
corrected STEM at SuperSTEM Daresbury/UK. He is currently
working as application developer at Gatan Inc.

Michael Lehmann Chapters 7 and 8

TU Berlin, Michael Lehmann studied physics in Tübingen, was Postdoc at the

Institute for Optics Triebenberg Lab, and is now full professor at TU Berlin.
and Atomic Physics
lehmann@physik.
tu-­berlin.de

Hannes Lichte Chapters 7 and 8

TU Dresden, Hannes Lichte studied physics at Kiel and Tübingen universities,

Triebenberg Laboratory where he finished the Dr.rer.nat. in 1977 under supervision
[email protected] of Gottfried Möllenstedt. He pioneered holographic imaging
mainly of atomic structures. Since 1994 he has been professor at
the Technische Universität Dresden and founder and director of
the Triebenberg Laboratory.

Andreas Thust Chapter 9

Forschungszentrum Andreas Thust is an Austrian physicist working at Forschungszentrum

Jülich GmbH, Jülich, Germany. In 2005 he was awarded together with Wim Coene
Ernst Ruska-Centre the internationally prestigious Ernst Ruska Price for their outstanding
[email protected] achievements in the field of focal-series reconstruction.
 List of Contributors xv

Sandra Van Aert Chapter 10

University of Antwerp, Sandra Van Aert is a Senior Lecturer at the University of
Electron Microscopy for Antwerp (Belgium). Within the Electron Microscopy for
Materials Science (EMAT) Materials Research (EMAT) group, her research focuses on new
[email protected] developments in the field of model-based electron microscopy
aiming at quantitative measurements of atomic positions, atomic
types, and chemical concentrations with the highest possible
precision.

Dirk Van Dyck Chapter 10

University of Antwerp, Dirk Van Dyck is Emeritus Professor in Physics of the University of
Electron Microscopy Antwerp. He is known for his pioneering work in dynamical electron
for Materials Science diffraction, exit wave reconstruction, electron channeling, and electron
(EMAT) tomography.
dirk.vandyck@
uantwerpen.be

Stephen J. Pennycook Chapter 11

National University Stephen J. Pennycook is a Professor in the Materials Science

of Singapore, and Engineering Department, National University of Singapore,
Department of Materials an adjunct Professor at the University of Tennessee, USA, and
Science & Engineering an adjoint Professor at Vanderbilt University, USA.
[email protected]

Paul Midgley Chapters 12 and 13

University of Cambridge, Paul Midgley is Professor of Materials Science at the Department of

Department of Materials Materials Science and Metallurgy at the University of Cambridge and a
Science and Metallurgy Professorial Fellow at Peterhouse.
[email protected]

Matthew Weyland Chapter 12

Monash University, Matthew Weyland is an Associate Professor at the Monash

Department of Materials Centre for Electron Microscopy and in the Department of
Science and Engineering Materials Science and Engineering at Monash University,
[email protected] Melbourne.

Paul Thomas Chapter 13

Gatan Inc. Paul Thomas graduated with a PhD in EFTEM Technique Development
[email protected] from the University of Cambridge in 2000. Ever since he has worked for
Gatan, where he is passionate about software for analytical EM.
xvi List of Contributors

Vicki Keast Chapter 14

The University of Newcastle, Vicki Keast is currently Associate Professor in the School
School of Mathematical of Mathematical and Physical Science at The University of
and Physical Science Newcastle, Australia.
[email protected]

Ian Jones Chapter 15

University Ian Jones is an electron microscopist and Professor of Physical Metal-

of Birmingham, lurgy at the University of Birmingham, UK.
School of Metallurgy
and Materials
[email protected]

Paul Kotula Chapter 16

Sandia National Laboratories, Paul Kotula is a staff member in the Materials Characteri-
Materials Characterization zation Department at Sandia National Laboratories on the
Department Albuquerque, NM campus. His research involves analytical
[email protected] electron microscopy, and in particular acquisition and statistical
analysis of X-ray and electron energy-loss spectral images.

Masashi Watanabe Chapter 17

Lehigh University, Masashi Watanabe is an Associate Professor of Materials Science and

Materials Science Engineering at Lehigh University. Masashi’s research emphasizes
and Engineering materials characterization using AEMs.
masashi.watanabe@
lehigh.edu
List of Initials and Acronyms

The field of TEM is a rich source of initials and acronyms (these are words formed by the initials),
behind which we hide both simple and esoteric concepts. While the generation of new initials and
acronyms can be a source of original thinking (e.g., see ALCHEMl), it undoubtedly makes for easier
communication in many cases and certainly reduces the length of voluminous textbooks. You have
to master this strange language before being accepted into the community of microscopists, so we
present a comprehensive listing that you should memorize.

ABF annular bright field

ABLAD all-beam large-angle diffraction
ACF absorption-correction factor
ACT automated crystallography for TEM
ACTF amplitude contrast transfer function
A/D analog to digital (converter)
ADF annular dark field
AEM analytical electron microscope/microscopy
AES Auger electron spectrometer/spectroscopy
AFF aberration-free focus
AFM atomic force microscope/microscopy
ALCHEMI atom location by channeling-enhanced microanalysis
ALS alternating least squares
ANL Argonne National Laboratory
AOs atomic-like orbitals
APB anti-phase (domain) boundary
APFIM atom-probe field ion microscope/microscopy
APW augmented plane wave
ART algebraic reconstruction technique
ASW augmented spherical wave
ATW atmospheric thin window
AWG arbitrary waveform generator
AXSIA automated expert spectral image analysis
BF bright field
BFP back-focal plane
BGPL beam-gas path length
BSE backscattered electron
BZB Brillouin-zone boundary
C1,2 condenser 1,2, etc., lens
CASTEP Cambridge Serial Total Energy Package
CAT computerized axial tomography
CB coherent bremsstrahlung
CBED convergent-beam electron diffraction
CBIM convergent-beam imaging
CCD charged-coupled device
CCF cross-correlation function
CCM charge-collection microscopy
CDF centered dark field
CF coherent Fresnel/Foucault
CFE cold field emission

xvii
xviii List of Initials and Acronyms

CL cathodoluminescence
CNTs carbon nanotubes
cps counts per second
CRT cathode-ray tube
CS crystallographic shear
CSET compressed sensing electron tomography
CSL coincident-site lattice
CSTEM confocal scanning transmission electron microscopy
CV cyclic voltammetry
CVD chemical vapor deposition
DADF displaced-aperture dark field
DART discrete algebraic reconstruction technique
DC direct current
DDF diffuse dark field
DF dark field
DFT density-functional theory
DOS density of states
DP diffraction pattern
DQE detection quantum efficiency
DSTEM dedicated scanning transmission electron microscope/microscopy
DTEM dynamical TEM
DTSA desktop spectrum analyzer
EBD electron beam deposition
EBIC electron beam-induced current/conductivity
EBSD electron-backscatter diffraction
EDS energy dispersive spectroscopy
EDX energy dispersive X-ray
EELS electron energy-loss spectrometry
EFI energy-filtered imaging
EFTEM energy filtered transmission electron microscopy
ELNES energy-loss near-edge structure
ELP™ energy-loss program (Gatan)
EMMA electron microscope microanalyzer
EMS electron microscope image simulation
(E)MSA (Electron) Microscopy Society of America
EPMA electron-probe microanalyzer
ESCA electron spectroscopy for chemical analysis
ESI electron-spectroscopic imaging
EST equally-sloped tomography
ETEM controlled environment TEM
EXAFS extended X-ray-absorption fine structure
EXELFS extended energy-loss fine structure
FEFF ab-initio multiple scattering software
FEG field-emission gun
FET field-effect transistor
FFP front-focal plane
FFT fast Fourier transform
FIB focused-ion beam
FIM field ion microscopy
FLAPW full-potential linearized augmented plane wave
FOLZ first-order Laue zone
FTP file-transfer protocol
FTs Fourier transformations
FWHM full width at half maximum
FWTM full width at tenth maximum
GB grain boundary
GGA generalized gradient approximation
GIF Gatan Image Filter™
GIGO garbage in garbage out
 List of Initials and Acronyms xix

GM Lines Gjonnes–Moodie lines

GMS Gatan Microscopy Suite™
GOS generalized oscillator strength
GS lines “glide and screw” lines
HAADF high-angle annular dark-field
HOLZ higher-order Laue zone
HPGe high-purity germanium
HREELS high-resolution electron energy-loss spectrometer/spectrometry
HRTEM high resolution transmission electron microscopy
HV high vacuum
HVEM high-voltage electron microscope/microscopy
IBD ion-beam deposition
IBF incoherent bright field
ICC incomplete charge collection
ICDD International Center for Diffraction Data
ICP incoherent channeling patterns
ID identification (of peaks in spectrum)
IDB inversion domain boundary
IEEE International Electronics and Electrical Engineering
IG intrinsic Ge
IHC inverse hole-count
ITRS International Technology Roadmap for Semiconductors
IVEM intermediate-voltage electron microscope/microscopy
K-M Kossel–Möllenstedt
KKR Korringa–Kohn–Rostoker
LACBED large-angle convergent-beam electron diffraction
LCAO linear combination of atomic orbitals
LCD liquid-crystal display
LDA local-density approximation
LEED low-energy electron diffraction
LKKR layered Korringa–Kohn–Rostoker
LSDA local spin density approximation
MAC mass absorption coefficient
MAS Microanalysis Society
MBE molecular-beam epitaxy
MC minimum contrast
MCA multichannel analyzer
MCDF multi-configuration Dirac–Fock
MCR multivariate curve resolution
MDA minimum detectable atoms
MDM minimum detectable mass
MEMS microelectromechanical systems
MLLS multiple linear least squares
MMF minimum mass fraction
MO molecular orbital
MRS Materials Research Society
MS multiple scattering
MSA multivariate statistical analysis
MT muffin tin
MTF modulation transfer function
MV megavolt
NCEMSS National Center for Electron Microscopy simulation system
NIH National Institutes of Health
NIST National Institute of Standards and Technology
NPL National Physical Laboratory
OIM orientation-imaging microscopy
OR orientation relationship
OTL ordering tie line
PARODI parallel recording of dark-field images
xx List of Initials and Acronyms

PAW projector augmented wave

PB phase boundary
P/B peak to background ratio
PCA principal component analysis
PCTF phase contrast transfer function
PDA photo-diode array
PEELS parallel electron energy-loss spectrometer/spectrometry
PIPS Precision Ion-Polishing System™
PIXE proton-induced X-ray emission
PM photomultiplier
POA phase-object approximation
ppb/m parts per billion/million
PSF point spread function
PTS position-tagged spectrometry
QCBED quantitative convergent-beam method
QHRTEM quantitative high-resolution transmission electron microscopy
RB translation boundary (yes, it does!)
RDF radial distribution function
REM reflection electron microscope/microscopy
RGA residual gas analyzer
RHEED reflection high-energy electron diffraction
ROI region of interest
RPA random phase approximation
SACT small-angle cleaving technique
SAD(P) selected area diffraction (pattern)
SCF self-consistent field
SDD silicon-drift detector
SDK software development kit
SE secondary electron
SEELS serial electron energy-loss spectrometer/spectrometry
SEM scanning electron microscope/microscopy
SESAMe sub-eV sub-Å microscope
SF stacking fault
SHRLI simulated high-resolution lattice images
SI spectrum imaging
SI Système Internationale
SIGMAK K-edge quantification software
SIGMAL L-edge quantification software
SIMS secondary ion mass spectroscopy
SIRT simultaneous iterative reconstructive techniques
S/N signal-to-noise ratio
SOLZ second-order Laue zone
SRM standard reference material
STEM scanning transmission electron microscope
STM scanning tunneling microscope/microscopy
TB twin boundary
TCC transmission-cross coefficient
TED transmission electron diffraction
TEM transmission electron microscope/microscopy
TFE thermal field emission
UEM ultrafast electron microscopes
UHV ultra high vacuum
URL uniform resource locator
UTW ultra-thin window
V/F voltage to frequency (converter)
VLM visible-light microscope/microscopy
VUV vacuum ultra violet
WB weak beam
WBDF weak-beam dark field
 List of Initials and Acronyms xxi

WDS wavelength-dispersive spectrometer/spectrometry

WP whole pattern
WPOA weak phase-object approximation
XANES X-ray absorption near-edge structure
XEDS X-ray energy-dispersive spectrometer/spectrometry
XPS X-ray photoelectron spectrometer/spectrometry
XRD/F X-ray diffraction/fluorescence
YAG yttrium-aluminum garnet
YBCO yttrium-barium-copper oxide
YSZ yttria-stabilized zirconia
ZAF atomic number/absorption/fluorescence correction
ZAP zone-axis pattern
ZLP zero-loss peak
ZOLZ zero-order Laue zone
Contents

1 Electron Sources . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.1 Introduction and Definitions of Parameters . . . . . . . . . . . . . . . . 2
1.2 Schottky Sources . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.2.1 Emission Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.2.2 Coulomb Interactions . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.2.3 Practical Aspects . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.3 Field Emission Sources . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
1.3.1 Emission Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
1.3.2 Practical Aspects . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
1.4 Photo-Emission Sources . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
1.5 Effect of the Electron Source Parameters on Resolution in STEM . . 11
1.5.1 Contributions to the Probe Size . . . . . . . . . . . . . . . . . 11
1.5.2 Current in a Probe . . . . . . . . . . . . . . . . . . . . . . . . . . 12
Appendix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
2 In situ and Operando . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.1 General Principles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
2.2 Some history . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
2.3 The Possibilities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
2.3.1 Post-Mortem Characterization . . . . . . . . . . . . . . . . . . 20
2.3.2 Statistics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
2.4 Time . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
2.4.1 Recording the Data . . . . . . . . . . . . . . . . . . . . . . . . . 22
2.4.2 The CCD Camera . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
2.4.3 Direct-Detection Cameras . . . . . . . . . . . . . . . . . . . . . 22
2.4.4 Software and Data Handling . . . . . . . . . . . . . . . . . . . 23
2.4.5 Drift Correction . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
2.4.6 Ultrafast Electron Microscopy . . . . . . . . . . . . . . . . . . . 25
2.5 The Environment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29

xxiii
xxiv Contents

2.5.1 Ultrahigh Vacuum . . . . . . . . . . . . . . . . . . . . . . . . . . 36

2.5.2 Working in a Gas Cell . . . . . . . . . . . . . . . . . . . . . . . . 37
2.5.3 Working in a Liquid Cell . . . . . . . . . . . . . . . . . . . . . . 39
2.6 The Temperature . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
2.6.1 Temperature Measurement . . . . . . . . . . . . . . . . . . . . 41
2.6.2 Heating . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
2.6.3 Cooling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
2.7 Other Stimuli . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
2.7.1 Deformation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
2.7.2 Magnetic Fields . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
2.7.3 Electric Fields . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
2.7.4 Photons . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
2.8 Adding or Removing Material . . . . . . . . . . . . . . . . . . . . . . . 67
2.8.1 Depositing Layers/Particles . . . . . . . . . . . . . . . . . . . . . 67
2.8.2 Deposition Energy: Electron and Ion Irradiation . . . . . . . 68
2.9 The Future . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
Appendix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
3 Electron Diffraction and Phase Identification . . . . . . . . . . . . . . . . . . 81
3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
3.2 Spinodal Alloys . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
3.2.1 Example: Ordered FeBe Phases and A2 Matrix . . . . . . . . . 83
3.3 Superalloys with Ordered Precipitates . . . . . . . . . . . . . . . . . . 85
3.3.1 Example: γ″ and γ′ Precipitation in Alloy 718 . . . . . . . . . 87
3.3.2 Example: D0a-Ordered δ Precipitation in Alloy 718 . . . . . . 89
3.4 Carbide Precipitation in fcc Alloys . . . . . . . . . . . . . . . . . . . . . 93
3.4.1 Example: M23C6 Precipitation in a Ni–Base Alloy . . . . . . . . 93
3.4.2 Example: MC Carbides in a Ni–Base Alloy . . . . . . . . . . . . 94
3.5 Ferritic Steels . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96
3.5.1 Relationships Between Austenite and Ferrite, Austenite and
Martensite (fcc/bcc) . . . . . . . . . . . . . . . . . . . . . . . . . 96
3.5.2 Relationship Between Cementite (Orthorhombic Fe3C or
M3C) and Ferrite/Tempered Martensite . . . . . . . . . . . . . 97
3.5.3 Relationships Between Alloy Carbides and Ferrite . . . . . . 97
3.5.4 Precipitation in Ferritic Structures . . . . . . . . . . . . . . . . 98
3.6 Epitaxial Oxide on Metal: Presence of Fe3O4 on Steel Foils . . . . . . 99
Appendix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102

Licensed to ([email protected])
 Contents xxv

4 Convergent-Beam Electron Diffraction: Symmetry and Large-Angle

Patterns . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
4.1 Symmetry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104
4.2 Point-Group Determination . . . . . . . . . . . . . . . . . . . . . . . . 104
4.3 Space-Group Determination . . . . . . . . . . . . . . . . . . . . . . . 109
4.3.1 Forbidden Reflections . . . . . . . . . . . . . . . . . . . . . . . 109
4.3.2 Black Crosses . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111
4.3.3 Complete Procedure for Space-Group Determination . . . 113
4.4 Ni3Mo – A Worked Example . . . . . . . . . . . . . . . . . . . . . . . . 114
4.4.1 Ni3Mo – a Worked Example, Part I: Point Group . . . . . . . 114
4.4.2 Qualifications . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118
4.4.3 Ni3Mo – a Worked Example, Part II: Space Group . . . . . . 119
4.5 Additional and Alternative Symmetry Methods . . . . . . . . . . . . 120
4.5.1 Symmetry Determination from Off-Axis Patterns . . . . . . 120
4.5.2 Symmetry from Precession Patterns . . . . . . . . . . . . . . 122
4.6 More on Glide Planes . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123
4.6.1 GM Lines in HOLZ Reflections . . . . . . . . . . . . . . . . . . 124
4.6.2 Glide Planes Normal to the Beam . . . . . . . . . . . . . . . 124
4.7 Beyond Symmetry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 124
4.7.1 Enantiomorphous Pairs: Handedness . . . . . . . . . . . . . 126
4.7.2 Polarity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 126
4.7.3 Coherent Convergent-Beam Diffraction . . . . . . . . . . . 127
4.8 Tanaka Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127
4.9 LACBED . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127
4.9.1 The Nature of LACBED Patterns . . . . . . . . . . . . . . . . . 129
4.9.2 Obtaining LACBED Patterns in Practice . . . . . . . . . . . . 130
4.9.3 Choosing the Parameters . . . . . . . . . . . . . . . . . . . . . 131
4.10 Spherical Aberration and LACBED . . . . . . . . . . . . . . . . . . . . 132
4.11 Crystal Defects in LACBED Patterns: Dislocations . . . . . . . . . . . 132
4.12 Crystal Defects in LACBED Patterns: Stacking Faults and Antiphase
Boundaries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 134
4.13 Other Tanaka Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . 134
4.13.1 Bright- and Dark-Field LACBED . . . . . . . . . . . . . . . . . 134
4.13.2 Convergent-Beam Imaging (CBIM) . . . . . . . . . . . . . . . 136
4.13.3 Rastering Techniques . . . . . . . . . . . . . . . . . . . . . . . 137
Appendix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 141
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 142
xxvi Contents

5 Electron Crystallography, Charge-Density Mapping, and Nanodiffraction 145

5.1 Can We Quantify Electron Diffraction Data? . . . . . . . . . . . . . 146
5.2 Quantitative CBED for Charge-Density Mapping . . . . . . . . . . . 147
5.3 Strain Mapping, High Voltage, Lattice Parameters Measured
by QCBED . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 153
5.4 Spot Patterns – Solving Crystal Structures . . . . . . . . . . . . . . . 155
5.5 The Precession Method . . . . . . . . . . . . . . . . . . . . . . . . . . . 157
5.6 Diffuse Scattering, Defects, Phonons, and Phase Transitions . . . . 158
5.7 Diffractive Imaging, Ptychography, STEM Holography, Ronchigrams,
and All That . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 159
5.8 Equipment for Quantitative Electron Diffraction . . . . . . . . . . . 162
Appendix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 163
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 164
6 DigitalMicrograph . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 167
6.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 168
6.1.1 What Is DigitalMicrograph? . . . . . . . . . . . . . . . . . . . 168
6.1.2 Installing DigitalMicrograph Offline . . . . . . . . . . . . . . 168
6.1.3 A (Very) Quick Overview . . . . . . . . . . . . . . . . . . . . . 168
6.2 Understanding Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . 170
6.2.1 What is an Image? . . . . . . . . . . . . . . . . . . . . . . . . . 170
6.2.2 Image Display . . . . . . . . . . . . . . . . . . . . . . . . . . . . 171
6.2.3 Number Formats . . . . . . . . . . . . . . . . . . . . . . . . . . 173
6.2.4 Image Calibration and Image Tags . . . . . . . . . . . . . . . 178
6.2.5 Some Simple Tools . . . . . . . . . . . . . . . . . . . . . . . . . 180
6.2.6 Extracting Subsets of Data . . . . . . . . . . . . . . . . . . . . 181
6.3 Digital Image Processing . . . . . . . . . . . . . . . . . . . . . . . . . . 183
6.3.1 Image ‘Filters’ . . . . . . . . . . . . . . . . . . . . . . . . . . . . 185
6.3.2 Fourier Transformation in Images . . . . . . . . . . . . . . . 187
6.3.3 Fourier Filtering . . . . . . . . . . . . . . . . . . . . . . . . . . 189
6.3.4 Coordinate Transformations . . . . . . . . . . . . . . . . . . . 192
6.4 Scripting and Plugins . . . . . . . . . . . . . . . . . . . . . . . . . . . . 193
Appendix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 195
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 195
7 
Electron Waves, Interference, and Coherence . . . . . . . . . . . . . . . . . 197
7.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 198
7.2 Description of Waves . . . . . . . . . . . . . . . . . . . . . . . . . . . . 198
7.2.1 Plane Wave . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 199
 Contents xxvii

7.2.2 Spherical Wave . . . . . . . . . . . . . . . . . . . . . . . . . . . 199

7.2.3 Modulated Wave . . . . . . . . . . . . . . . . . . . . . . . . . 199
7.3 Interference . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 200
7.4 Modulation of a Wave by an Object . . . . . . . . . . . . . . . . . . . 201
7.5 Propagation of Waves . . . . . . . . . . . . . . . . . . . . . . . . . . . 201
7.5.1 Fresnel Approximation in the Near-Field of the Object . . 202
7.5.2 Fraunhofer Approximation in the Far-Field of the Object . 202
7.6 Imaging: Formation of the Image Wave . . . . . . . . . . . . . . . . 203
7.6.1 Fourier Transform of the Object Exit Wave . . . . . . . . . . 203
7.6.2 Building the Image Wave by Inverse Fourier Transform of
the Fourier Spectrum . . . . . . . . . . . . . . . . . . . . . . . 203
7.7 Electron Wave Function . . . . . . . . . . . . . . . . . . . . . . . . . . 204
7.8 Electron Interference . . . . . . . . . . . . . . . . . . . . . . . . . . . . 205
7.9 Findings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 206
7.10 Interpretation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 207
7.11 Coherence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 207
7.11.1 Spatial Coherence . . . . . . . . . . . . . . . . . . . . . . . . . 208
7.11.2 Coherent Current . . . . . . . . . . . . . . . . . . . . . . . . . 210
7.11.3 Temporal Coherence . . . . . . . . . . . . . . . . . . . . . . . 211
7.11.4 Total Degree of Coherence . . . . . . . . . . . . . . . . . . . 211
7.11.5 A Generalization . . . . . . . . . . . . . . . . . . . . . . . . . . 211
7.11.6 Coherence at Inelastic Interaction . . . . . . . . . . . . . . . 211
Appendix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 213
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 213
8 Electron Holography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 215
8.1 Big Problem with TEM: Phase Contrast . . . . . . . . . . . . . . . . . 216
8.2 Wave Modulation and Conventional Imaging . . . . . . . . . . . . . 216
8.2.1 Amplitude Modulation . . . . . . . . . . . . . . . . . . . . . . 216
8.2.2 Phase Modulation . . . . . . . . . . . . . . . . . . . . . . . . . 217
8.2.3 What Do We See in an Electron Image? . . . . . . . . . . . . 218
8.3 Principle of Image-Plane Off-Axis Holography . . . . . . . . . . . . 219
8.3.1 Recording a Hologram . . . . . . . . . . . . . . . . . . . . . . 219
8.3.2 Reconstructing the Object Exit-Wave . . . . . . . . . . . . . 220
8.3.3 What Have We Achieved so Far? . . . . . . . . . . . . . . . . 223
8.4 Properties of the Reconstructed Object Exit-Wave . . . . . . . . . . 223
8.5 Requirements of Holography . . . . . . . . . . . . . . . . . . . . . . . 224
8.6 Quality Criteria . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 224
xxviii Contents

8.7 Application to Electric Potentials on Nanometer Scale . . . . . . . 225

8.7.1 Phase Shift Due to Electrostatic Potentials . . . . . . . . . . 225
8.7.2 Experimental Considerations . . . . . . . . . . . . . . . . . . 226
8.7.3 Application Example: p–n Junctions . . . . . . . . . . . . . . 227
8.8 Further Derivatives of Electron Holography . . . . . . . . . . . . . . 227
8.8.1 Holographic Tomography . . . . . . . . . . . . . . . . . . . . 227
8.8.2 Dark-Field Holography . . . . . . . . . . . . . . . . . . . . . . 228
Appendix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 230
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 230
9 Focal-Series Reconstruction . . . . . . . . . . . . . . . . . . . . . . . . . . . . 233
9.1 Motivation: Why the Effort? . . . . . . . . . . . . . . . . . . . . . . . 234
9.2 Quick Walk Through Electron Diffraction . . . . . . . . . . . . . . . 235
9.3 From the Wavefunction to the Image . . . . . . . . . . . . . . . . . . 237
9.3.1 Imaging with a ‘Neutral’ Microscope . . . . . . . . . . . . . 238
9.3.2 Linear Imaging with a Constant-Phase-Shift Microscope . 240
9.3.3 Linear Imaging with a Real Microscope . . . . . . . . . . . . 241
9.3.4 From Oscillations to Windings: an Integral View on Linear
Imaging . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 247
9.4 From the Images to the Wavefunction . . . . . . . . . . . . . . . . . 249
9.4.1 Tomographic Interpretation of Focal Series . . . . . . . . . 249
9.4.2 Fundamental Properties of Focal Series . . . . . . . . . . . . 250
9.4.3 An Explicit Solution to the Linear Inversion Problem . . . . 253
9.4.4 Nonlinear Reconstruction . . . . . . . . . . . . . . . . . . . . 255
9.4.5 Numerical Correction of Residual Aberrations . . . . . . . . 256
9.5 Application Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . 257
9.5.1 Twin Boundaries in BaTiO3 . . . . . . . . . . . . . . . . . . . . 258
9.5.2 Stacking Fault in GaAs . . . . . . . . . . . . . . . . . . . . . . 260
Appendix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 263
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 264
10 
Direct Methods for Image Interpretation . . . . . . . . . . . . . . . . . . . 267
10.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 268
10.2 Basics of Image Formation . . . . . . . . . . . . . . . . . . . . . . . . . 268
10.2.1 Real imaging . . . . . . . . . . . . . . . . . . . . . . . . . . . . 268
10.2.2 Successive Imaging Steps . . . . . . . . . . . . . . . . . . . . . 269
10.2.3 Coherent Imaging . . . . . . . . . . . . . . . . . . . . . . . . . 269
10.2.4 High-Resolution Imaging in the TEM . . . . . . . . . . . . . 270
10.3 Focal-Series Reconstruction of the Exit Wave . . . . . . . . . . . . . 271
 Contents xxix

10.4 Interpretation of the Reconstructed Exit Wave . . . . . . . . . . . . 271

10.4.1 Electron Channeling . . . . . . . . . . . . . . . . . . . . . . . . 272
10.4.2 Argand Plot . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 273
10.5 Quantitative Structure Refinement . . . . . . . . . . . . . . . . . . . 274
10.5.1 Precision Versus Resolution . . . . . . . . . . . . . . . . . . . 276
10.5.2 Quantitative Model-Based Structure Determination . . . . 276
Appendix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 280
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 280
11 
Imaging in STEM . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 283
11.1 Z-Contrast STEM: an Introduction . . . . . . . . . . . . . . . . . . . . 284
11.1.1 Independent Scatterers . . . . . . . . . . . . . . . . . . . . . . 284
11.1.2 An Array of Scatterers . . . . . . . . . . . . . . . . . . . . . . 284
11.1.3 As the Crystal Thickens . . . . . . . . . . . . . . . . . . . . . . 284
11.1.4 Inside and Outside . . . . . . . . . . . . . . . . . . . . . . . . . 286
11.1.5 The Effect of Defects . . . . . . . . . . . . . . . . . . . . . . . 287
11.1.6 Quasicrystals . . . . . . . . . . . . . . . . . . . . . . . . . . . . 288
11.2 An Electron’s Eye View of STEM . . . . . . . . . . . . . . . . . . . . . 288
11.2.1 Plane Waves and Probes . . . . . . . . . . . . . . . . . . . . . 291
11.2.2 Rayleigh, Airy and Resolution . . . . . . . . . . . . . . . . . . 292
11.3 Lens Aberrations for STEM . . . . . . . . . . . . . . . . . . . . . . . . 293
11.3.1 The Benefits of Aberration Correction . . . . . . . . . . . . 295
11.3.2 Resolution in the Third Dimension – Depth Resolution . . 300
11.4 Spatial and Temporal Incoherence . . . . . . . . . . . . . . . . . . . . 305
11.4.1 Spatial Incoherence . . . . . . . . . . . . . . . . . . . . . . . . 305
11.4.2 Temporal Incoherence . . . . . . . . . . . . . . . . . . . . . . 306
11.4.3 “How Do I Know if I Have a Coherent Probe?” The
Ronchigram . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 306
11.5 Coherent or Incoherent Imaging . . . . . . . . . . . . . . . . . . . . . 310
11.5.1 A Point Detector; Coherent Imaging . . . . . . . . . . . . . . 311
11.5.2 An Infinite Detector: Incoherent Imaging . . . . . . . . . . 312
11.5.3 An Annular Detector: Incoherent Dark-Field or Bright-Field
Imaging . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 314
11.5.4 Atoms Are Smaller in HAADF STEM . . . . . . . . . . . . . . 315
11.5.5 Transverse Coherence . . . . . . . . . . . . . . . . . . . . . . . 316
11.5.6 The Origin of Contrast in the Scanned Image . . . . . . . . 317
11.5.7 Transfer Function and Damping Function . . . . . . . . . . . 318
11.5.8 Longitudinal Coherence . . . . . . . . . . . . . . . . . . . . . 319
xxx Contents

11.6 Dynamical Diffraction . . . . . . . . . . . . . . . . . . . . . . . . . . . 323

11.7 Other Sources of Image Contrast . . . . . . . . . . . . . . . . . . . . . 326
11.8 Image Processing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 329
11.9 Image Simulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 332
11.9.1 Bloch Waves . . . . . . . . . . . . . . . . . . . . . . . . . . . . 333
11.9.2 Multislice . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 333
11.9.3 Bloch Waves with Absorption . . . . . . . . . . . . . . . . . . 333
11.9.4 There Is No Stobb’s Factor in HAADF . . . . . . . . . . . . . 334
11.10 Future Directions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 335
Appendix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 337
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 338
12 Electron Tomography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 343
12.1 Theory of Projection . . . . . . . . . . . . . . . . . . . . . . . . . . . . 344
12.2 Back-Projection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 346
12.3 Constrained Reconstruction . . . . . . . . . . . . . . . . . . . . . . . . 347
12.3.1 Constraint by Projection Consistency . . . . . . . . . . . . . 347
12.3.2 Constraint by Discrete Methods . . . . . . . . . . . . . . . . . 348
12.3.3 Constraint by Symmetry . . . . . . . . . . . . . . . . . . . . . 348
12.3.4 Metric-Based Constraint . . . . . . . . . . . . . . . . . . . . . 348
12.4 Other Reconstruction Approaches . . . . . . . . . . . . . . . . . . . . 350
12.5 Meeting the Projection Requirement . . . . . . . . . . . . . . . . . . 350
12.6 STEM Tomography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 351
12.7 Element-Selected Tomography . . . . . . . . . . . . . . . . . . . . . . 354
12.8 Dark-Field TEM Tomography . . . . . . . . . . . . . . . . . . . . . . . 356
12.9 Holographic Tomography . . . . . . . . . . . . . . . . . . . . . . . . . 358
12.10 Atomistic Tomography . . . . . . . . . . . . . . . . . . . . . . . . . . . 359
12.11 Experimental Limitations . . . . . . . . . . . . . . . . . . . . . . . . . 360
12.12 Beam Damage and Contamination . . . . . . . . . . . . . . . . . . . 364
12.13 Automated Acquisition . . . . . . . . . . . . . . . . . . . . . . . . . . 365
12.14 Tilt-Series Alignment . . . . . . . . . . . . . . . . . . . . . . . . . . . . 366
12.15 Visualization of Three-Dimensional Datasets . . . . . . . . . . . . . 368
12.16 Segmentation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 369
12.17 Quantitative Analysis of Volumetric Data . . . . . . . . . . . . . . . 371
Appendix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 373
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 373
13 EFTEM . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 377
13.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 378
13.2 Why Use EFTEM? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 378
 Contents xxxi

13.3 Instrumentation for EFTEM . . . . . . . . . . . . . . . . . . . . . . . . 379

13.3.1 General TEM Considerations . . . . . . . . . . . . . . . . . . 379
13.3.2 The Imaging Filter . . . . . . . . . . . . . . . . . . . . . . . . . 379
13.3.3 Detector Considerations . . . . . . . . . . . . . . . . . . . . . 380
13.4 Limitations and Artefacts . . . . . . . . . . . . . . . . . . . . . . . . . 381
13.4.1 Spatial Resolution in EFTEM Images . . . . . . . . . . . . . . 381
13.4.2 Non-Isochromaticity . . . . . . . . . . . . . . . . . . . . . . . . 383
13.4.3 Sample Drift . . . . . . . . . . . . . . . . . . . . . . . . . . . . 383
13.4.4 Diffraction Contrast . . . . . . . . . . . . . . . . . . . . . . . . 384
13.4.5 Illumination Convergence . . . . . . . . . . . . . . . . . . . . 384
13.5 Application of EFTEM . . . . . . . . . . . . . . . . . . . . . . . . . . . 385
13.5.1 Zero-Loss Imaging and Diffraction . . . . . . . . . . . . . . . 385
13.5.2 Measuring Relative Thickness (t/λ Mapping) . . . . . . . . . 386
13.6 Core-Loss Elemental Mapping . . . . . . . . . . . . . . . . . . . . . . 387
13.6.1 Elemental Mapping (Three-Window Method) . . . . . . . . 387
13.6.2 Jump-Ratio Mapping (Two-Window Method) . . . . . . . . 388
13.7 EFTEM Spectrum-Imaging . . . . . . . . . . . . . . . . . . . . . . . . . 389
13.8 Low-Loss Imaging . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 392
13.9 Alternative Imaging Techniques for Biological Specimens . . . . . 393
13.10 Quantitative Elemental Mapping . . . . . . . . . . . . . . . . . . . . 394
13.11 Chemical State Mapping Using ELNES . . . . . . . . . . . . . . . . . . 396
13.12 Hybrid EFTEM Modes(ω-q, Line Spectrum EFTEM) . . . . . . . . . . 397
13.13 EFTEM Tomography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 398
Appendix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 401
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 401
14 Calculating EELS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 405
14.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 406
14.2 Density Functional Theory (DFT) . . . . . . . . . . . . . . . . . . . . . 407
14.2.1 Introduction to DFT . . . . . . . . . . . . . . . . . . . . . . . . 407
14.2.2 The Exchange Correlation Potential . . . . . . . . . . . . . . 409
14.2.3 Approximations to the Potential . . . . . . . . . . . . . . . . 409
14.2.4 Basis Sets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 410
14.2.5 The Korringa–Kohn–Rostoker (KKR) Method . . . . . . . . 412
14.3 Calculations of the ELNES . . . . . . . . . . . . . . . . . . . . . . . . . 412
14.3.1 ELNES Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . 412
14.3.2 The Core Hole . . . . . . . . . . . . . . . . . . . . . . . . . . . 414
14.3.3 Multiplet Theory . . . . . . . . . . . . . . . . . . . . . . . . . . 415
14.3.4 Multiple Scattering (MS) Methods . . . . . . . . . . . . . . . 416
xxxii Contents

14.4 Calculating Low-Loss EELS . . . . . . . . . . . . . . . . . . . . . . . . . 417

Appendix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 421
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 422
15 Diffraction & X-ray Excitation . . . . . . . . . . . . . . . . . . . . . . . . . . 425
15.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 426
15.2 ALCHEMI . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 426
15.3 Gedanken ALCHEMI . . . . . . . . . . . . . . . . . . . . . . . . . . . . 426
15.4 Two Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 428
15.4.1 Dilute Solution/Partition Coefficient Analysis . . . . . . . . 428
15.4.2 Concentrated Solution/OTL Analysis . . . . . . . . . . . . . . 430
15.5 Delocalization and Axial Channeling . . . . . . . . . . . . . . . . . . 431
15.6 Optimizing ALCHEMI: ‘Statistical’ ALCHEMI . . . . . . . . . . . . . . 432
15.7 Incoherent Channeling Patterns . . . . . . . . . . . . . . . . . . . . . 432
15.8 Vacancies and Interstitials . . . . . . . . . . . . . . . . . . . . . . . . . 432
15.9 Chemistry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 434
Appendix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 435
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 436
16 X-ray and EELS Imaging . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 439
16.1 What Are Spectral Images and Why Should We Collect Them? . . 440
16.2 Some History . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 441
16.3 Acquisition and Analysis of Spectral Images . . . . . . . . . . . . . . 442
16.3.1 Sampling and the Effect of Probe Versus Pixel Size (STEM-
XEDS/EELS) or Magnification (EFTEM) . . . . . . . . . . . . . 442
16.3.2 Signal: Count Rate, Dwell Time, Spectral Image Size, and
Acquisition Time . . . . . . . . . . . . . . . . . . . . . . . . . . 443
16.3.3 Drift Correction and Beam Damage . . . . . . . . . . . . . . 446
16.3.4 Conventional Data Analysis Methods . . . . . . . . . . . . . 446
16.4 Multivariate Statistical Analysis Methods . . . . . . . . . . . . . . . . 451
16.4.1 Principal Components Analysis (PCA) . . . . . . . . . . . . . 454
16.4.2 Factor Rotations . . . . . . . . . . . . . . . . . . . . . . . . . . 455
16.4.3 Multivariate Curve Resolution (MCR) . . . . . . . . . . . . . 456
16.4.4 Quantification . . . . . . . . . . . . . . . . . . . . . . . . . . . 457
16.5 Example of X-ray and Electron Energy-Loss Spectral Image
Acquisition and Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . 458
16.5.1 Fe-Ni Spectral Image Acquisition and Quantification . . . . 458
16.5.2 Mn-Doped SrTiO3 Grain Boundary Spectral Image
Acquisition and Quantification . . . . . . . . . . . . . . . . . 459
16.5.3 Plasmon Mapping of AG Nanorods: EELS Spectral Image
Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 462
 Contents xxxiii

Appendix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 464
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 464
17 
Practical Aspects and Advanced Applications of XEDS . . . . . . . . . . . 467
17.1 Performance Parameters of XEDS Detectors . . . . . . . . . . . . . . 468
17.1.1 Detector, Fundamental Parameters . . . . . . . . . . . . . . 468
17.1.2 Monitoring Detector Contamination . . . . . . . . . . . . . 470
17.1.3 Software to Determine Detector Parameters . . . . . . . . 471
17.2 X-ray Spectrum Simulation – a Tutorial and Applications of DTSA 472
17.2.1 What Is DTSA? . . . . . . . . . . . . . . . . . . . . . . . . . . . 473
17.2.2 A Brief Tutorial of X-ray Spectrum Simulation for a Thin
Specimen Using DTSA . . . . . . . . . . . . . . . . . . . . . . . 475
17.2.3 Details of X-ray Simulation in DTSA . . . . . . . . . . . . . . 477
17.2.4 Application 1: Confirmation of Peak Overlap . . . . . . . . 481
17.2.5 Application 2: Evaluation of X-ray Absorption into a Thin
Specimen . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 482
17.2.6 Application 3: Evaluation of the AEM-XEDS Interface . . . 483
17.2.7 Application 4: Estimation of the Detectability Limits . . . . 483
17.3 The ζ-factor Method: a New Approach for Quantitative X-ray
Analysis of Thin Specimens . . . . . . . . . . . . . . . . . . . . . . . . 486
17.3.1 Why Bother with Quantification? . . . . . . . . . . . . . . . 486
17.3.2 What Is the ζ-factor? . . . . . . . . . . . . . . . . . . . . . . . 487
17.3.3 Quantification Procedure in the ζ-factor Method . . . . . . 488
17.3.4 Determination of ζ factors . . . . . . . . . . . . . . . . . . . . 489
17.3.5 Applications of ζ-factor Method . . . . . . . . . . . . . . . . 490
17.4 Contemporary Aplications of X-ray Analysis . . . . . . . . . . . . . . 492
17.4.1 Renaissance of X-ray Analysis . . . . . . . . . . . . . . . . . . 493
17.4.2 XEDS Tomography for 3D Elemental Distribution . . . . . . 494
17.4.3 Atomic Resolution X-ray Mapping . . . . . . . . . . . . . . . 495
Appendix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 500
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 501
Figure and Table Credits . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 505

Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 515
Electron Sources
1

Chapter 1
Chapter Preview

Without an electron source there is no electron microscope, the two most important source types (Schottky sources and
that’s obvious. For the electron microscopist there are sev- cold field emission sources) starts with the physics theory
eral reasons to know something about the source. Someone of electron emission that is behind their properties. This in-
knowledgeable decides which microscope to buy; someone volves a bit of mathematics that some readers may enjoy and
equally knowledgeable has to use it when it has been pur- others may want to skip to go directly to practical aspects of
chased. How is the choice of source reflected in the quality of the sources. The last part contains a discussion on the rela-
the images? Being able to use the microscope with a specific tion between source parameters and resolution in STEM that
source to its optimal performance is another reason. How to will allow the reader to find the optimum balance between
operate the microscope without destroying the source? How current in the probe and resolution in the image.
to recognize that a source does not live up to specification
anymore? This chapter starts by introducing the parameters We introduced the topic in Chap. 5 of W&C. The present
used for describing the quality of electron sources and links chapter looks at the topic in much more depth, emphasizing
these directly to their effect on imaging. The description of why you should know about the source.

© Springer International Publishing Switzerland 2016

C. B. Carter, D. B. Williams (Eds.), Transmission Electron Microscopy, DOI 10.1007/978-3-319-26651-0_1 1

Licensed to ([email protected])
 2 1 Electron Sources

1.1 Introduction and Definitions and thus the emittance is constant. However, in electron micro-
scopes there are many apertures and there is acceleration, so the
of Parameters emittance is not a constant.
Chapter 1

Electron microscopy starts with the electron beam, a stream of There is one parameter that does not change at a beam-limiting
particles that travels at incredible speed through an almost per- aperture: it is the differential brightness of a beam. The differen-
fect vacuum to impact and scatter on the object in its path. Or tial brightness of an electron beam is defined as
should we say it is a wave that moves smoothly towards the
dI
sample to change its phase ever so slightly before traveling on, Bdiff = (1.1)
creating intricate interference patterns on a screen? Whatever dAd˝
this beam is, it needs to be created in the electron source. For
the best microscope we obviously want the best electron source, where dI is the current passing through a surface dA within a solid
but how to describe the quality of an electron source, and what angle dΩ. Clearly, when an aperture is placed in the beam, either the
is the exact relation between the properties of the source and the area or the angle is reduced, but the current is reduced proportionally,
performance of the microscope? so the brightness remains constant. Acceleration or deceleration,
however, does not keep the brightness constant: usually the beam
Electron sources supply the electrons to the system, so the main angle changes – it reduces with acceleration and increases with
property of a source seems to be the number of electrons per deceleration. By defining a reduced brightness of the beam as
second, or current, it supplies. However, we need to distinguish
dI
between the total current that the emitter emits, the beam current Br = (1.2)
that passes through the extractor aperture, and the probe current dAd˝ V
or specimen current that arrives at the sample through the beam-­
limiting apertures of the microscope. that problem is solved. Reduced brightness in SI units is ex-
pressed in A/(m2 · sr · V). Because this beam parameter is constant
After the initial acceleration, the beam seems to come from an in the whole system, we can also use this as a parameter of the
area with a diameter that we call virtual source size or just source source. In fact, for microscopic applications of electron sources, it
diameter dv. It is called a ‘virtual source’ because it is different is the single most important parameter of a source. We’ll see later
from the surface that is actually emitting electrons. If a cathode that there are very few physical phenomena that can change the
has a spherical shape, all electrons that come out of the surface beam brightness: stochastic Coulomb interactions between the
perpendicular to the surface, seem to come from a single point at individual electrons is one, energy filtering is another. The con-
the center of the sphere. Electrons are emitted with a slight lateral cept of brightness is easiest understood in the plane of a source
velocity: in Schottky and thermal emitters this is a result of the image, where the area is the size of the source image and the
thermal energy kT; in field emission it is considerably larger. These angle is the aperture angle. However, we need to be careful here.
electrons seem to come from a circle around the center of the sphere The definition given above for the differential brightness intrin-
and are thus responsible for the finite size of the virtual source. The sically assumes a homogeneous current density in the area dA. In
electric field that causes the acceleration may also partly act like a source image, there is a current distribution with a maximum
a lens effect that magnifies and shifts the virtual source position density in the center and often long tails. Although this is not
along the optical axis. The convention is to define the virtual source a formal definition, we then often use the concept of practical
size looking back from the plane of the extractor. Thus, the virtual brightness B = I/AΩ, which makes it necessary to define the size
source size depends on the extractor voltage. Interactions between of the source image. We shall show later that the best definition
the electrons may increase the effective size of the source. is the full width that contains 50 % of the current: FW50. Because
we still use the full current, practical brightness is larger than
The virtual source size is usually small compared to the size of differential brightness (Bronsgeest et al. 2008):
the extractor aperture, so we can speak of ‘the’ emission angle of
I
the beam and thus of the angular current density, I’ which is de- Bpract =  2 = 1:44 Bdiff (1.3)
fined as the current divided by the emission solid angle, or better: d
4 FW50
 ˛2
the beam current divided by the solid angle that is accepted by
the extraction aperture. We shall encounter many more current Brightness in an electron microscope can best be measured in
densities, such as the emission current density j, which is the spot mode, at relatively large spots. It is important to find a mode
emitted current per unit area at the emitter surface. in which the size of the probe is dominated by the size of the
source image, so at relatively small aperture angle to avoid ab-
In the general literature on electron emitters, you will often en- erration contributions to the probe size and at sufficient current
counter the parameter emittance, which is defined as the prod- to avoid a diffraction contribution to the spot size. In TEM, the
uct of the source size and the emission angle. If there are no spot at the sample plane is magnified with a known magnification
beam-limiting apertures and no acceleration, this is a useful pa- to the CCD camera. The angle at the sample can be measured in
rameter, because it stays constant from one crossover to the next: diffraction mode, and the current is measured in a Faraday cup,
if the crossover size is magnified, the beam angle is demagnified, or on the microscope screen.
 1.1 Introduction and Definitions of Parameters 3

The last parameter that is used to describe an electron source

is the energy spread ∆E in the beam. Electrons can be emitted Remember that electrons are fermions, and fermions, un-
with slightly different energies. Whatever the acceleration is af- like bosons such as photons, cannot interfere with each

Chapter 1
terwards, the energy conservation law tells us that this original other. This means that different electrons are never mu-
energy difference will stay. In fact there is a whole distribution tually coherent, so coherence is a phenomenon that plays
of different energies, the energy distribution. Usually the dis- within individual electrons. When there is coherence be-
tribution is asymmetric and it is a little difficult to characterize tween two points in a beam, it means that the electron
it with a single number. Some use full width at half maximum wave of an individual electron has spread out over those
FWHM, others use full width that contains 50 % of the current two points. 9
FW50. Even for a nice distribution such as a Gaussian, the FW50
is different from the FWHM (FWHM = 1.75 × FW50), so we have
to be careful regarding what to use in calculations of, e.g., the Imax and Imin are the maximum and minimum intensities in the
chromatic aberration. It turns out that the FW50 value is less interference pattern. The distance over which an electron wave
sensitive to the exact energy distribution than the FWHM and is coherent is inversely proportional to the internal angle in the
thus gives a more consistent result in calculations (Barth and beam, that is, the angle towards the nearest source image. For a
Nykerk 1999). large coherence length, this source image must be small. In the
full Van Cittert–Zernike theory, it is found that the degree of
The energy spread can only be increased by stochastic inter- coherence γ(x1,2) is directly proportional to the Fourier transform
actions between individual electrons, where a collision may of the source image intensity function. For instance, when the
accelerate one electron and decelerate the other. That effect is source image is a homogeneously filled disk, the Fourier trans-
called the Boersch effect. In practice, we also add the variations form is the Airy function. In a target plane illuminated from that
in accelerator voltage to the energy spread, because they have crossover, the function γ has fallen by 12 % when
the same effect when averaged over time, but strictly this is not
a parameter of the electron source. The energy spread may be

x1;2 = = Xc (1.6)
reduced by an electron monochromator – also called an energy 2c
filter. In the filtering process, some electrons are stopped, so the
act of filtering decreases the beam current and also the bright- where θc is the semi-angle from the target to the crossover disk.
ness. What is constant is the brightness per electron volt energy Xc is usually called the coherence width. At a distance
spread. So when a monochromator is present, the single param-
eter that we need to know about the source is Br/∆E.

x1;2 =
c (1.7)
So far, we have been able to limit our view of an electron gun as
a source of charged particles, little balls that fly through space! the illumination is generally considered fully incoherent.
However, electrons have wave properties and these wave prop-
erties are crucial for electron microscopy, both in TEM, where At a distance
we measure phase contrast, and in STEM, where diffraction ul-
timately limits resolution. The electron wavelength is

x1;2 =
2c
= Xpc (1.8)
r
h 1
(1.4)
−9
 = = 1:226  10
p V there is still some coherence left and this is often taken as a
practical coherence width, although the definition is somewhat
when λ is given in m and V*, the relativity-corrected acceleration arbitrary. The exact degree of coherence obtained between points
voltage of the beam, is given in volt, V  = V.1 + 10−6 V/. For at this distance will depend on the current density distribution
TEM, we are often interested in the coherence of the beam. If in the source image and on the definition of the diameter of the
there is coherence between two points in the beam, it means that source image. Probably it is best to define the crossover diameter
there is a constant phase relation between the electron waves at as the FW50 value.
the two points: the amplitudes move as two pendulums in perfect
unison. Only when there is coherence can interference occur There is a simple relationship between coherence and brightness:
between the waves coming from these points. If only part of the the current within the coherent area is
amplitude moves in that unison, we speak of partial coherence.
This is mathematically expressed by the degree of coherence, 2  2
 
 2
Icoh = Br V Xpc c2 = Br  V = 0:93  10−18 Br (1.9)
which is the correlation between the wave amplitudes ψ at points 4 4 2
r1 and r2 (at a distance x1,2).
ˇ ˇ This is a surprisingly simple relation that tells us we have a
Z
coherent beam as soon as we limit the current to smaller than
ˇ ˇ
ˇ ˇ Imax − Imin
.x1;2 / = ˇˇ lim   .r1 ; t/ .r2 ; t/dtˇˇ = (1.5) 10−18 Br. This is important to realize: if the current is smaller than
ˇ !1 ˇ Imax + Imin
− 10−18 Br the beam is coherent: it is a necessary and sufficient con-
 4 1 Electron Sources

dition! In TEM, there is coherence over an area that contains this 1.2 Schottky Sources
current. Thus, spatial coherence is not related to the size of the
source, but rather to the size of the last source image before the
Chapter 1

target and the distance between that source image and the target. Most high-resolution (S)TEMs now have either a Schottky elec-
Note that lenses between that crossover and the target change the tron source or a cold field emission source. The reason for that
effective size and distance, but we do not need to know these is mostly the superior brightness of these sources. At the highest
exact relations: in practice, simply reducing the current density resolution they easily deliver a factor of a 1,000 more current in
on the target increases the coherence. a probe or in a coherent area than a simple thermionic source.
In other words, at the same current they give a 30-fold larger
coherent width. Basically, these are the sources that have en-
This implies that every beam can be made coherent by abled routine atomic-resolution microscopy. Tungsten hairpin
sufficient aperturing of the beam. 9 sources and LaB6 or CeB6 thermionic sources do exist, but for a
description we refer to the original W&C where they have been
described in sufficient detail.
To talk about electron sources as being either coherent or not is
generally wrong. The difference between sources is the amount
of current that can be put in a coherent beam, but that is just a
different way, and not a bad way either, of expressing the reduced 1.2.1 Emission Theory
brightness of a source.

Note that this analysis is limited to sources that have an emitting Although often erroneously labeled as such, a Schottky source is
area larger than the coherent area on the emitting surface itself. not a field emission source, nor is it even a ‘thermal field emission’
For single-atom emitters and carbon nanotubes, this may not be source. A Schottky source in normal operation is a purely thermi-
true, in which case the brightness goes effectively to infinity, as onic source where the effective work function of the material has
it does in laser beams (Kruit et al. 2006). been lowered by the Schottky effect. To understand the emission
properties we first need to summarize the theory of electron emis-
sion. Figure 1.1 gives an illustration of the theory by showing the
The temporal coherence length or longitudinal coherence energy diagram (energy in eV versus distance in nm) for electrons
is the distance along the z-axis over which the wave is near the boundary between a conductor and vacuum while there
coherent. It can be imagined as the length of the wave is an electric field on the surface. Simply to exit a conductor elec-
package that represents the electron. 9 trons must overcome an energy barrier called the work function.
To find out the emission current density we need to determine
the occupation of the states in 3D momentum space (left in the
A perfectly monochromatic beam has a very long temporal energy diagram) and then calculate the quantum-mechanical prob-
coherence length. Different energies, however, have different ability of their escaping from each state into vacuum, either over
wavelengths which interfere destructively, so a beam with en- the potential barrier, or tunneling through the barrier. Thermionic
ergy spread has limited temporal coherence. From the uncer- emission occurs when, by heating, the energies of some of the
tainty principle: conductance electrons become higher than the work function.
When the work function is denoted by φ, the current density at
h h
E  t = ; so zcoh =   t = (1.10) the surface is expressed by the Richardson equation
2 2  E  ' 
j = A.1 − r/T 2 exp − (1.11)
kT
It is not directly obvious what the definition of this energy spread
should be; it must be related to something intrinsic in the emis- where A is the emission constant, 1 − r is the transmission coef-
sion process and not contain contributions such as the high-volt- ficient of the surface barrier for electrons, T is the temperature,
age instability. Typical coherence length values are in the order and k is the Boltzmann constant (1.38 · 10−23 J/K = 86 μeV/K). φ
of 1 µm (Niklaus and Hasselbach 1993). If two parts of the beam and kT are both in eV. The quantum-statistical derivation of this
follow very different paths, it may occur that at the point of in- equation leads to
terference, the different parts of the wave function do not arrive
simultaneously, which will affect the interference contrast. In 4mek2
A= = 1:204  106 A=.m2 K2 / (1.12)
principle, this may even happen in a lens: although the optical h3
path lengths for all paths are equal, the lengths traveled by the
wave packages on the different paths may be different. For most In general, 1 − r is close to one. Furthermore, there is a slight
practical purposes we assume that the beam has sufficient long­ temperature dependence of φ, so the work function will have
itudinal coherence. the general form
' = '0 + cT (1.13)
 1.2 Schottky Sources 5

Schottky plot

1600K

Chapter 1
1700K
98 1800K

96

94

log (l )
92

90

88

0 10 20 30 40 50 60 70
√ VE
Fig. 1.1 Energy diagram (energy in eV versus distance in nm) illustrating the
basics of emission theory Fig. 1.2 Schottky plot

Actually, the work function also varies with crystal orientation. The magnitude of the Schottky effect can be analyzed by cal-
culating the potential curve based on the image-force potential
In practice, a simpler equation is often used for the current density resulting from the attraction between an electron outside the sur-
 '  face and its image behind the surface. The net potential is now
j = AT 2 exp − (1.14) the sum of the zero-field image-force potential and the potential
kT
due to the field. For the lowering of the potential barrier we find
(in eV)
in which the values of A and φ are then adapted for the measured s
current densities. eF
' = − (1.17)
4"0
At the surface of the emitter, the mean energy in the plane of
the surface is kT and as long as the acceleration is perpendicular
to the surface, this is conserved, so ˛ 2  e  V = kT (α is the half This increases the thermi onic emission to
aperture angle in the beam). p !
2 ' − eF=4"0
The intrinsic reduced differential brightness can then be ex-
j = AT exp −
kT (1.18)
pressed as (where dA is the emission surface)
dI ej
Br = = (1.15) Experimental plots of ln I´ (the angular current density) versus
p
dAd˝  V kT VE are called ‘Schottky plots,’ see Fig. 1.2. Since F = βVE and
I´ = j(r/Mα)2 (Mα is the angular magnification from the emission
if we take the FW50 value (diameter that contains 50 % of the surface to the measurement plane), the slopes of these plots yield
current) for the source size. The energy distribution of pure ther- the value of the field-enhancement factor β:
mionic emission is given by
4"0
  ˇ= .kB T/2 slope2 = 5:157.T  slope/2 (1.19)
A E+' e3
dj.E/ = 2  E exp −
k kT
dE (1.16)
Since r is the tip radius and Mα is the angular magnification of
The Schottky effect is the effective decrease of the work func- the extraction field, with estimates for Mα and φ, the Schottky
tion when an external field F is applied at the metal surface (see plot also yields the tip radius r. The Schottky plot can also give
Fig. 1.1). The necessary fields are very high and can only be the work function
obtained by sharpening a tip to submicron size, so that the field  
in front of the tip is enhanced by a large factor. The field-en- 4em 2 r
 = kB T ln .k B T/ + 2 ln − axis intercept (1.20)
hancement factor β is defined through F = βVE, where VE is the h3 M˛
extractor potential.
 6 1 Electron Sources

12 0.5
0.3 V/nm 1600 K
0.5 V/nm 1700 K
0.7 V/nm 1800 K
Chapter 1

10

8 0.4

4 0.3

0.2 7
1 1.5 2 2.5 3 3.5 10 10 8 10 9
Br (A/m2eV)

Fig. 1.3 Energy distributions of electrons emitted from a Schottky emitter Fig. 1.4 Intrinsic energy spread (FW50 values) versus intrinsic brightness for
Schottky emitters

Note that there are several assumptions in the model used to The energy distribution broadens to:
produce the Schottky plot. There is no tunneling through the 
barrier, the work function is independent of the field, and the tip djES .E/ = A  p  
is a sphere, not a flat facet. Even with approximate equations, 1 + exp E + ' − eF=4"0 =kT
the Schottky plot is still useful (Swanson et al. 2008; Bahm et al. ln Œ1 + exp.E=/ dE (1.23)
2008). For example, we take into account that Mα is itself a func­

tion of β and r as in Mα = 0.525(βr)0.42. In order to keep the Mα  djS .E/  ln Œ1 + exp.E=/
constant, we should scale the voltage of the suppressor with the E
changing extractor voltage. It is also possible to fit the Schottky with djs(E) the energy distribution for pure Schottky emission.
plot to a more complete model and extract the parameters from These approximations are valid up to a value of q of about 0.7
the fit (Swanson et al. 2012). when extracting the energy spectrum, but only up to q = 0.25 for the
current density jES. For stronger fields, only numerical solutions are
The energy distribution for pure Schottky emission is the same available. Figure 1.3 shows how the energy distribution broadens
as for pure thermionic emitters with the lower work function. and shifts as the field on the surface increases. Figure 1.4 plots the
The position of the energy spectrum shifts with increasing field full width containing 50 % of the current value of the energy spread
to lower energies (Fransen et al. 1999b). When the field on distribution as a function of brightness for different temperatures.
the emitter surface is sufficiently increased, electron tunneling
through the top of the potential barrier occurs. The emission is As in thermionic emitters, the practical brightness is
then in the ‘extended Schottky regime.’ In order to find analytical
ejES
equations for the emission current and energy distribution, the Br = 1:44 (1.24)
form of the energy barrier is approximated by a parabolic func- kT
tion. The emission current density is then found from
p !
2 ' − eF=4"0 q
jES = AT exp −
kT

sin q (1.21) 1.2.2 Coulomb Interactions

with Schottky emitters yield particle densities that are sufficiently

high for stochastic electron–electron interactions to occur. It is
h.4"0 e/1=4 F 3=4  −3 F
3=4
customary to distinguish between effects that increase the virtual
q= = = 0:118  10 (1.22)
2 2 m1=2 kT kT T source size (trajectory displacements) and effects that increase
the energy spread (the Boersch effect). Monte Carlo simulations
can give reasonably accurate estimates of these effects, but are
cumbersome. The effects can also be approximated by analytical
equations, but these are strictly speaking only valid for sections
in the optics column where the beam has a constant energy. Es-
 1.2 Schottky Sources 7

pecially in the source region this is not the case, and we should 1.8
Energy spread including Boersch effect versus brightness
use better approximations. The whole analytical theory is fairly
involved, and here we will only touch on it.

Chapter 1
1.6

Which equations we use will depend on the local beam param- 1.4
eters such as the current density. For the typical parameters in
a TEM we distinguish between the Holtsmark regime and the 1.2
pencil-beam regime. In the pencil-beam regime, the electrons
have a nearest-neighbor distance that is larger than the diameter 1
of the beam, so this regime is appropriate after the beam has been
apertured. In the absence of apertures, and before the first ex­ 0.8
traction aperture, the crossover size is blurred by a point spread
function with a Holtsmark distribution of FW50: 0.6

m1=3 I 2=3 L2=3 0.4

FW50H = 0:172  (1.25)
"0 V 4=3 ˛ 4=3
0.2 7
10 108 109
Here L is the length of the segment under consideration between
the crossover where the blur is calculated and a lens or aperture
at the other end of the segment. V is the acceleration potential, Fig. 1.5 Full energy spread versus intrinsic brightness for Schottky emitters with
α the half angle of the beam, m is the electron mass, and ε0 the various tip radii
dielectric constant. If the aperture angle is sufficiently small, the
section is in the pencil-beam regime, with a FW50
fectively travels only through one side of the ‘crossover,’ the pref-
m3=2 I 3 ˛L3 actors for the energy spread should be divided by 2. As a rule of
FW50PB = 0:145 7=2  5=2 (1.26)
e "0 V thumb, when the regime may be either Holtsmark or Lorentzian,
the contribution to the energy spread can be found from
As a rule of thumb, when the regime may be either Holtsmark
4 #1=4
or pencil-beam, the contribution to the crossover size can be
" 4 
E E E
found from = + (1.30)
EFW50T EFW50L EFW50H
" 6=7  6=7 #−7=6
1 1
FW50T = + (1.27) To get a first estimate for the contribution of the Coulomb inter-
FW50H FW50PB
actions to the virtual-source size and energy spread of the source,
we simply take the values for the parameters in the equations
In Schottky sources and cold field emitters, the trajectory (V, α, rc) as they are at the extraction electrode and compare the
displace­ment reaches values comparable to the virtual-source FW50 with the virtual source size as seen in that plane. A more
image more easily than it does in thermionic emitters. The result sophisticated approach can take the acceleration into account.
is effectively a lower brightness of the source. For brightness values up to a few times 108 A/(m2 sr V), the tra-
jectory displacement in the source region is still small compared
The energy spread resulting from Coulomb interactions occurs to the virtual source size. The main danger of a virtual-source
mainly in the crossovers, or in the source very close to the tip. size increase and brightness loss lies in the section between the
For the Boersch effect, the Lorentzian and the Holtsmark regime extractor and the accelerator. However, the energy broadening
are most appropriate, with equations for the contribution to the near the tip is important. Figure 1.5 shows how the energy spread
energy spread (in eV) in one beam crossover can increase from intrinsic values of about 0.3 eV up to a value
close to 1 eV, especially for large tip radii because these emit
m1=2 I large total currents.
EFW50L = 1:41 (1.28)
"0 e ˛V 1=2
1=2

m1=3 I 2=3
EFW50H = 0:891 1=3 (1.29) 1.2.3 Practical Aspects
"0 rc ˛V 1=3

where rc is the size of the crossover. When the energy spread Schottky emission may occur from any heated material in a
comes from the segment that includes the emitter, rc must be inter- strong field. Nevertheless, we shall limit the rest of our discus-
preted as the size of the emission surface. For Schottky emitters, sion to the most commonly used materials: single-crystal tung-
this is the size of the low work-function facet, usually about one sten with a (100) surface at the tip end, which is coated with
third of the tip diameter. Because the beam from the emitter ef- zirconium oxide (referred to as ZrO), see Fig. 1.6.
 8 1 Electron Sources

especially detrimental in this respect. Another ‘contamination’

effect that sometimes occurs is a hindering of the flow of ZrO
towards the facet, making the emission unstable. After a vacuum
Chapter 1

‘accident,’ the continuous renewal of the ZrO layer may restore

the original properties of the source.

A serious complication in operating a Schottky source comes from

the fact that the shape of the tungsten tip depends strongly on its
history: the shape depends on the temperatures and fields that the
tip has experienced (Fujita and Shimoyama 2007; Bronsgeest and
Kruit 2008; Bronsgeest and Kruit 2009, 2010; Liu et al. 2010;
Bahm et al. 2011). Figure 1.7 shows some different shapes that a
tip can have. The creation of a flat facet on an etched, round tip, is a
natural result of the thermodynamics of the surface: at elevated tem-
peratures, tungsten surface atoms are mobile and diffuse over the
surface. They attach and detach to surface steps such that terraces
grow or shrink. At installation, the manufacturer has pretreated the
tip to have a certain shape, but this shape may change during oper-
ation. To keep this ‘ideal’ shape as long as possible, the extraction
voltage should be kept within a narrow range while the tip is hot.

At high fields, the end facet shrinks and becomes more square-­
shaped. The sensitivity of the shape to the extraction field is also
the intrinsic limitation to the brightness of Schottky sources. A
higher brightness, measured up to 2 × 108 A/(m2 sr V), can be ob-
tained by increasing the extraction field, but this results in further
faceting, including faceting of the sides of the tip, and ultimately
in unstable operation (van Veen et al. 2001).

At low fields, the atoms tend to move away from terraces of

Fig. 1.6 a Schottky emitter with the blob of ZrO clearly visible on the shank. smaller radius towards steps of larger radius. Thus the center
b Tip end of the emitter with the (100) facet. (From Bronsgeest PhD thesis TU terraces on the tip disappear until a flat facet is obtained. In oper-
Delft 2009) ation, this effect may continue when the extraction voltage is too
low. When a number of these terraces pile up and shrink together,
this is recognizable as a moving ring in the angular emission pat-
At a temperature above 1,500 K, the ZrO forms an electric dipole tern. The effect is known as ‘ring collapse.’ This phenomenon is
layer at the tungsten surface that lowers the work function from important, because you may be tempted to operate the Schottky
about 4.5 eV to about 2.9 eV. The tip radius ranges from 0.3 to source at low extraction field in order to minimize the energy
2 μm. The front face forms a flat-facet surface, which ideally has spread. This only works for a short time, until the ring collapse
a diameter of about one third of the tip diameter. The operating sets in with its accompanying change of current and brightness.
temperature is about 1,850 K. A typical suppressor voltage is
−300 V and a typical extractor voltage is 5 kV. For a 1-μm-diam- At intermediate fields, the facet is reasonably stable, although in
eter tip with a 300-nm-diameter facet, a typical angular current the course of many months, the tip head always grows in diam-
density is 0.7 mA/sr, total emission from the facet 40 μA, emis- eter. This dependence of the tip shape on temperature and field
sion from the shank 100 μA, virtual source size 30 nm. makes it important to follow the correct start-up procedures for
a tip and not operate it at temperatures or fields far away from
In the early years of the Schottky emitter the brightness was the recommended values.
limited to about 2 × 107 A/(m2 sr V). Nowadays, brightness values
of 5 × 107 A/(m2 sr V) (Maunders et al. 2011) and even above 108
are commercially available with energy spreads as given above.
The lifetime of a well-treated Schottky tip can be several years, 1.3 Field Emission Sources
limited mainly by the supply of ZrO. The rate of evaporation is
of course a steep function of temperature: if a tip has a lifetime
of 10,000 h at 1,750 K, this reduces to 1,000 h at 1,900 K. 1.3.1 Emission Theory
The ZrO-coated tungsten must be operated in pressures below
10−7 Pa. If the pressure is too high, this may influence the work By applying an electric field of very high field strength at the
function and thus reduce the emission current. O2 and H2O are surface of a metal, electrons are emitted even without heating the
 1.3 Field Emission Sources 9

Chapter 1
Fig. 1.7 Shapes of the W(100) facet in a Schottky emitter after different operating conditions. Magnification markers: a, b 500 nm; c, d 1 µm. (From Bronsgeest PhD
thesis TU Delft 2009)

metal: this is known as cold field emission. In cold field emis- A plot of ln(I/VE2) as a function of 1/VE, called the Fowler–Nor-
sion, electrons tunnel straight from the Fermi level through the dheim plot, yields approximately a straight line in the zero-tem-
potential barrier into free space. perature equation, see Fig. 1.8. For a given work function, it
yields the size of the emitting area A = I/j and the field-enhance-
Just as in Schottky emission, the high field strength is realized ment factor β = F/VE.
by using a nanometer-size pointed emitter facing an extraction
electrode. For these sharp tips, the field-enhancement factor β is The energy distribution, shown in Fig. 1.9, is given by:
assumed inversely proportional to the tip radius:
4me3 2
 
2.y/' exp.E=d/
VE djCF .E/ = d exp −  dE
F = ˇVE = (1.31) h3 3t.y/d 1 + exp.E=kT/
kr
(1.37)
where k is a factor depending on the exact configuration: it’s The practical brightness is given by:
typically between 1 and 5.
jCF
Br = 1:44 (1.38)
The current-density equation is derived in what is known as the d
‘Fowler–Nordheim theory,’ for a model cathode that is flat. For ex-
tremely sharp tips, deviations occur which we shall not treat here. Coulomb interactions do not deteriorate the energy spread as
p ! they do in Schottky emitters because the total emission current
"3 k 2 F 2 8' 3=2 2m p is much smaller. However, limitations to the brightness occur
sin p (1.32)
jCF = exp − .y/ 
8h't2 .y/ 3ehF sooner because of the small intrinsic size of the virtual source
(Cook et al. 2010).
with
y = '=' (1.33)
–12
the relative decrease of the work function by the field, and t(y)
and υ(y) slowly varying functions:
–14
t.y/ = 1 + 0:11y1:33 and .y/ = 1 − y1:69 (1.34)
–16
The parameter p is a measure for the current emitted above the
Fermi level because T is above zero:
–18
p = kT=d (1.35)

–20
The parameter d is a kind of measure for internal energy, when
φ and d are given in eV:
hF F 0 0.005 0.01 0.015 0.02 0.025
d= p  9:76  10−11 3=2 (1.36)
4t.y/ 2me' 3 '
Fig. 1.8 Fowler–Nordheim plot
 10 1 Electron Sources

15
4.0 V/nm larger and thus require us to use a larger demagnification, which
4.1 V/nm in turn has the same effect as lowering the brightness. When
4.2 V/nm more than a few nanoamperes are drawn from a field emission
Chapter 1

4.3 V/nm
4.4 V/nm
tip, there is a danger that gun-lens aberrations will dominate the
4.5 V/nm performance of the source, which can be minimized by careful
10 design of the gun lens (Kasuya et al. 2014).

Besides pure tungsten emitters in different orientations, many

other FEG cathodes have been tested such as: oxygen-processed
tungsten tips, carbon and carbon-coated tips, tantalum carbide
tips, and semiconductor tips. Soon after the discovery of car-
5
bon nanotubes (CNTs), work started on using these as electron
sources (de Heer et al. 1995; Fransen et al. 1999a). The main
advantage expected of the CNT is that the surface is very sta-
ble and rather insensitive to contamination. At slightly elevated
temperatures they emit in the order of 200 nA with only about
0
–0.6 –0.4 –0.2 0 0.2 1 % variations (De Jonge 2004). Metal tips with protrusions of
only 3 or even 1 atom were pioneered in the early nineties, but
efforts to create such ‘single-atom tips’ have recently increased.
Fig. 1.9 Energy spectra for a cold field emitter for various extraction fields The intrinsic brightness of these tips for all practical purposes
is infinity because the virtual source size is always smaller than
other contributions to the probe size (diffraction, trajectory dis-
1.3.2 Practical Aspects placement, gun-lens aberrations). For TEM, the beams from such
sources are fully coherent. Efforts to use these innovations in a
commercial machine continue (Houdellier et al. 2015) at the time
Although tungsten has a high work function (approximately of writing, but have failed in the past either due to insufficient
4.5 eV), it is chosen for its rigidity, high melting point (3,655 K), current for analytical applications, or fabrication issues, or un-
and good etching characteristics. The high melting temperature is expected instabilities, or limited lifetime, or any other practical
important for the cleaning procedure, which consists of heating aspect.
the tip to ~2,500 K for a short period to vaporize all contami-
nants. This is known as flashing the tip. Since the tip of the field
emission source (FEG) is spot-welded to a standard microscope
filament, this temperature is simply achieved by passing a heat- 1.4 Photo-Emission Sources
ing current through the filament.

A typical cold field emission tip in a TEM may have a diameter The photo-emission effect can be used to create an electron
of 10–20 nm and give an angular current density 5 μA/sr, total source. However, the photons should have sufficient energy: it
emission 20–40 μA, virtual source size about 2 nm. There aren’t should be at least as high as the work function, and preferably a
very many measured brightness values for FEGs, but a represen- little higher, in order to have any efficiency. For unprepared sur-
tative one is 1.8 × 1010 A/(cm2 sr) at 1 MV (Kawasaki et al. 2000), faces this means that the light must be far into the ultraviolet, for
so a generally accepted value for Br is 2 × 108 to 4 × 108 A/(m2 sr which there are no good direct-current (DC) sources. So usually
V). An advantage of field emission sources is the intrinsic low people go to low work-function surfaces or to pulsed emission.
energy spread with a FW50 value smaller than 0.3 eV. We can produce low work-function surfaces by, e.g., covering a
semiconductor with a few monolayers of cesium, but it is very
The presence of even single-molecule contaminants may criti- hard to keep the surface under the right conditions so this has not
cally affect the shape and the work function of the tip, so the FEG yet led to a source for TEM. Using high-intensity pulsed lasers,
must operate in a very clean high vacuum, in the range 10−8 Pa, a single electron may absorb the energy from several photons
but preferably lower. The general rule is that the emitter is more so that photo-emission can occur at visible wavelengths without
stable if the vacuum is better. Contaminants diffuse along the sur- having to lower the work function of the surface.
face of the tip towards the emission surface, so it is important to
keep the whole tip clean. Traditionally, a tip needed to be cleaned For TEM, the interest in pulsed emission has recently increased
at least once per day, because the emission showed a continuous in order to create ‘ultrafast electron microscopes’ (UEMs – see
decrease of 10–20 %/h, but recent advances seem to find an op- Chap. 2) for the study of effects at the nanoscale and at the
erational mode that is stable for longer times (Cho et al. 2013). sub-picosecond time scale (Zewail 2006). As for normal TEM
sources, it is necessary to go to pointed cathodes for obtaining
The small source size makes the FEG quite sensitive to stray high brightness. When hitting a cold field emission tip with a
fields and vibration. These may cause the effective diameter to be femtosecond laser, all kinds of interesting processes occur: field
 1.5 Effect of the Electron Source Parameters on Resolution in STEM 11

enhancement from the E-field of the laser light, multiphoton ab- practical for optimizing the probe size. So what we usually do
sorption, combined photon absorption and field emission, etc. is to calculate the different contributions to the probe size sep-
(Hommelhoff et al. 2006). The characteristics of these sources arately. These contributions include, e.g., diffraction, spherical

Chapter 1
while in operation in a TEM are not yet well established. aberration, and source image. Now if all these contributions
were independent and had a Gaussian intensity distribution, it
would be simple to add them and find the full probe size: we
could add them in quadrature. However, when the distributions
1.5 Effect of the Electron Source are non-Gaussian this does not work. For some specific analyt-
ical forms of the distribution, an addition rule can be derived;
Parameters on Resolution for other distributions only a numerical approach can give an
in STEM approximate addition rule.

The equation that yields the best fit to the minimum FW50 re-
The resolution in STEM is determined predominantly by the size sults of the fundamentally correct procedure is (Barth and Kruit
of the electron probe, with subtle influences from the coherence 1996):
in the probe. Of course, there are also effects from the interaction 1 12
between the beam and the specimen. Before continuing, we must
00
1:3 ! 1:31 2
1

define what we mean by ‘probe size.’ One parameter, the size,
1
dP = @@ dI1:3 + dA4 + ds4 4 A + dc2 C (1.39)
B  
cannot, of course, define every aspect of a focused beam. Even
A
for a perfectly round beam, what we would like to know is the
full intensity distribution – the current density as a function of
the distance to the center of the probe. That distribution is also where dI, dA, ds, and dc are the FW50 values of the contributions
called the point spread function (PSF – more on this too in later from the source image, the diffraction disk, the spherical aberra-
chapters). If the distribution is Gaussian, it is all fairly simple and tion, and the chromatic aberration, respectively. So we first add
any measure of size can be used. The full width at half maximum the diffraction to the spherical aberration with a fourth-power
(FWHM) for a Gaussian is the same as the diameter that contains rule, then add the source size contribution with a 1.3-power rule,
50 % of the total current (FW50) and the same as the distance and finally add the chromatic aberration effect with a simple
between the 12 % point and the 88 % point of the current that is square-power rule. For spherical-aberration-corrected systems
measured when scanning over a knife edge – known as edge res- we do not have such an equation yet, but a slight adaptation to
olution d12-88. Many distributions, however, are not Gaussian. One the ds term should yield a useful equivalent.
example is the distribution in a spherical-aberration-dominated
probe: at underfocus, this distribution has a very bright central Optimization of the probe size consists of finding the aperture
spot on a broad background and at overfocus, the highest intensity angle and source-to-probe magnification that gives the smallest
is at the edge of the probe. The FWHM may then be small, but it probe size, or the largest current in a specific probe size.
contains hardly any current so it’s not a good measure for probe
size! We have found that the FW50 is the most consistent measure
to characterize a probe, also when it is non-Gaussian. One of the
reasons is that the conversion between 25 and 75 % edge resolu- 1.5.1 Contributions to the Probe Size
tion and probe size, FW50 = 1.76d25-75, is practically independent of
the current density distribution in the probe. There are situations,
however, where it is useful to optimize for a minimum FW20 (size To form a probe, the virtual source is imaged on the target with
of the probe that contains 20 % of the current), which leads to a magnification M to an FW50 size Mdv. Since Br is a conserved
higher resolution at the cost of a lower contrast. quantity through the whole system, the current in the probe is
calculated from
Fundamentally, the correct way to calculate the distribution 
function in an electron probe is to start with the wave-optical IP = Br .Mdv /2  ˛ 2 V (1.40)
4
calculation of the diffraction distribution as modified by the
spherical aberration of the lens. It is necessary to do this for
a range of defocus values, since we do not yet know which where α is the half-aperture angle at the probe, V is the accel-
defocus will give the smallest probe size. Subsequently, these erating voltage at the target, and Br the practical brightness as
distributions must be convoluted with the distribution of the discussed earlier. From this, the contribution to the probe size
(defocused) source image. Finally, we can take account of the can be expressed as:
chromatic aberration effects by convoluting the distributions s
in the defocus direction with the energy-spread function. The 2 IP 1
minimum size of the probe is not found in the Gaussian image
d1 = Mdv = (1.41)
 Br V ˛ 2
plane, but, due to the spherical aberration, at a small defocus.
This procedure is worthwhile for a single calculation, but not

Licensed to ([email protected])
 12 1 Electron Sources

The size of the diffraction spot is often given as 1.22λ/α; however,

this is not the FW50 size, but the size of the first zero in the ring
pattern. The FW50 value is
Chapter 1

dp
10 0 for
  1 I =5
dA = 0:54 = 0:54 1=2 (1.42) x10 –
˛ V ˛ dp 18. B
for r
I =0
.5x1
with Λ = 1.226 × 10 m V .
−9 1/2 0 18
–
. Br

d [nm]
Spherical aberration causes the outer rays in the beam to be de- dI f
or I
focused by an amount dz = Csα2, so at the Gaussian image plane, =0
.5x1 dA
they have a distance Csα3 from the axis. This leads to a ‘disk of 0 –18 dc
–1 . Br
least confusion’ or minimum width of the total beam (FW100) of 10
0.5Csα3. However, the FW50 value has a much smaller minimum:

ds
ds = 0:18Cs ˛ 3 (1.43)

In probe-size calculations, we can often only use the aberra- 0.005 0.01 0.02
tion coefficients of the last probe-forming lens for the analysis. alpha [rad]
This is because usually the last lens demagnifies all aberration
contributions of the other lenses in the system. However, for Fig. 1.10 Optimization of aperture angle (Br = 108, V = 200 keV, dU = 0.8 eV,
emitters with a very small virtual source size, such as the CNT Cs = 0.7 mm, Cc = 2 mm)
emitter, we must also explicitly take account of the aberration
coefficients of the gun lens. In a microscope with an aberration
corrector for the probe lens, this equation should be replaced by 1.5.2 Current in a Probe
a different one that also contains a term increasing with aperture
angle. The problem is that we cannot give a general equation
valid for each microscope or even for one particular microscope. At very small currents, the size of the source image is negligible.
The reason is that once the third-order aberration (and nowadays If we assume that the spherical-aberration contribution is much
sometimes even the fifth-order aberration) is corrected, the re- smaller than the chromatic-aberration contribution, the probe
maining aberrations depend on the settings of the corrector and size is
the alignment of the beam. Often the latter are no longer rota­
 !1=2
tionally symmetric.  1 2 ıU 2
  
dP = 0:54 1=2  + 0:6Cc ˛ (1.45)
V ˛ V
Chromatic aberration causes the electrons with lower energy to
be focused closer to the lens than electrons with higher energy.
This leads to a blurring contribution for FW50: For optimized α this is
ıU 1=2
Cc ıU 1=2
dc = 0:6Cc ˛ (1.44) .dP /min = dAC = 2:81  10−5 at
V V 3=4
(1.46)
s
0:54V 12
with Cc the chromatic aberration coefficient of the system and δU ˛=
the FW50 of the energy distribution of the source. If the energy 0:6Cc ıU
distribution is Gaussian, the FWHM of the distribution is very
different from the FW50 (in contrast with a 2D Gaussian distri- Thus, the minimum probe size is not dependent on the brightness
bution, where FW50 = FWHM). If you were to use that FWHM or the coherence of the emitter; the only emitter property that
value, the prefactor in the chromatic aberration contribution enters the equation is the energy spread!
would be 0.34 instead of 0.6.
If we assume the chromatic aberration contribution to be much
Figure 1.10 shows all these contributions as a function of the smaller than the spherical aberration contribution, the probe size is
aperture angle α and the addition according to Eq. 1.39. ! 14
 1 4
 
dP = 0:54 1=2  3 4
+ .0:18Cs ˛ / (1.47)
V ˛
 1.5 Effect of the Electron Source Parameters on Resolution in STEM 13

For optimized α this is where KΛ = 1.08 × 10−18 m2 sr V. This is a very useful equation,
1=4 1=4 because it gives a simple estimate of how much current there
is in a probe which is close to the minimum size, whatever

Cs

Chapter 1

.dP /min = dAS = 1:03  10−8 at ˛ = 1:23
V 3=8 Cs V 1=2 causes the minimum size: chromatic aberration, spherical
(1.48) aberration, or any other remaining aberration in a corrected
system. When the FW50 of the virtual-source image equals the
FW50 of the diffraction spot (dI = dA) we call the current IΛ, with
Thus, the minimum probe size is not dependent on the brightness IΛ = KΛBr. For example, for a Schottky emitter at Br = 108 A/(m2
or the coherence of the emitter; here it does not depend on any sr V), this current is 100 pA, independent of the beam energy
emitter property. or the lens aberrations! Of course, beam energy and lens ab-
errations do determine the size of the probe that contains this
For aberration-corrected systems the minimum probe size will current.
depend on the remaining aberrations and the defocus.

However, a probe without current is of no use, so we have to

allow a contribution to the probe size that is related to the cur-
rent in the probe. By expressing the probe-size contribution in
terms of the diffraction contribution, the probe current can be
rewritten as

 2 dI 2
   2
 2 2  dI
IP = Br dI  ˛ V = Br 0:54  = Br K
4 2 dA dA
 (1.49)

Chapter Summary

The single most important parameter of an electron source field emitters operate at an even higher field than Schottky
is its brightness: the current that is emitted into a certain emitters, which enables field emission, the tunneling of the
solid angle (i.e., the extraction aperture) divided by the vir- electrons through the potential barrier of the work function.
tual size of the source, normalized for the acceleration volt- The differences in how emission occurs cause all kinds of
age. We have seen that the brightness is directly related to differences in source parameters and practical issues during
the current that can be obtained in a coherent beam in TEM, operation. We included a short ‘excursion’ on the theory of
and in a probe in STEM. The second most important param- stochastic Coulomb interactions because this effect can be
eter is the energy spread, related to the effects of chromatic of influence both on the energy spread and the brightness
aberration and to energy resolution in EELS. of a source.

The core of this chapter is the description of Schottky elec- We ended the chapter with an analysis of the relation be-
tron sources and cold field emission sources. Schottky emit- tween source properties and resolution in STEM, which of
ters are really thermionic emitters, where the emission is course is the same as an analysis of how small we can focus
enhanced by the Schottky effect, which is simply a lowering the beam and how much current we can get in the probe. A
of the work function by the strong field at the surface. Cold few useful rules of thumb are explained.
 14 Appendix Effect of the Electron Source Parameters on Resolution in STEM

Appendix
Chapter 1

People Q1.13 Give two reasons why the energy spread of a beam from
a Schottky source increases when more current is ex-
tracted from the tip.
Walter Schottky was born in Zurich on 23 July 1886 and died Q1.14 What limits the lifetime of a Schottky source if all goes
in Pretzfeld, Germany, on 4 March 1976. He contributed to well?
many technological inventions and discoveries, including the Q1.15 Give at least three incidents that can end the life of a
Schottky defect, Schottky barrier and Schottky diode. Schottky source prematurely.
Q1.16 What is a Fowler–Nordheim plot and which informa-
tion on a field emission source can be obtained from it?
Q1.17 Why is the short-term current stability of a cold field
Self-Assessment Questions emission source intrinsically less than that of a Schottky
source?
Q1.18 Why is the energy spread of a cold field emission source
Q1.1 Why is the total emission current not a good parameter intrinsically smaller than that of a Schottky source?
to characterize a source? Q1.19 What is the difference between a FWHM and a FW50 of
Q1.2 What is the difference between the virtual source size an electron probe, in definition, and in practice?
and the physical size of the emitting area of an electron Q1.20 What is the numerical difference between the FWHM,
source? the FW50 of a probe and the 25–75 edge resolution for
Q1.3 Why can we assume that the angular current density in a Gaussian current density distribution in the probe?
a microscope is uniform, while the spatial density in the Q1.21 Compute the difference in total probe size after adding
virtual source is approximately Gaussian? two equal distributions either linearly or by the two-
Q1.4 Explain the difference between differential brightness, power rule or by the four-power rule.
reduced brightness, and practical brightness. Q1.22 Consider a lens system that images a Schottky electron
Q1.5 Compute the electron wavelength for acceleration volt- source (effective dsource = 30 nm, Br,pract = 108 A/(m2. sr. V),
age 1 kV, 30 kV, 100 kV, 300 kV. FW50dE = 1 V) onto a sample at electron energy 80 kV.
Q1.6 Use the interdependencies of the parameters in the Assume aberration coefficients Cs = 1 mm, Cc = 2 mm.
equation for reduced brightness to show that the re- Calculate all contributions to the probe size for an ap-
duced brightness is not affected by: erture angle of 10 mrad and a source to probe magnifi-
a. the choice of the magnification from source to probe cation of M = 1/150. Also calculate the total probe size
b. the size of the beam-limiting aperture and the current in the probe.
c. acceleration or deceleration (for ease of explanation Q1.23 (Advanced) Use your favorite math program to calcu-
assume the acceleration in the plane of a lens) from late and plot all contributions to the probe size as a
energy V1 to V2. function of half angle α and determine the optimum
Q1.7 Why are there three different spatial coherence lengths angle. Use realistic estimates for all parameters.
defined? Q1.24 How does the current in a probe at the highest resolu-
Q1.8 Derive the relation between current within a coherent tion setting in STEM depend on the aberrations of the
area Icoh and the reduced brightness. microscope?
Q1.9 Explain how electrons are forced to leave the emitter in Q1.25 (Advanced) Construct the (I, d) function for parameters
respectively a Schottky emitter and a cold field emitter. of your choice, optimizing both M and α. If applicable
Q1.10 What is the relation between brightness and temperature for your parameters, show where the dominant aberra-
in respectively a Schottky emitter and a field emitter? tion changes from Cc to Cs.
Q1.11 (Advanced) Use your favorite math program to perform
a simulation that gives you practical brightness and the
full-width-50 value of the energy spread of a thermionic
electron source as a function of temperature.
Q1.12 If a Schottky source in the electron microscope seems
not to deliver the expected brightness, how can a
Schottky plot show whether this is a result of a blunt
tip or an increased work function or neither?
 Appendix References 15

Cook B, Verduin T, Hagen C, Kruit P (2010) Brightness limita-

References tions of cold field emitters caused by Coulomb interactions.
J Vac Sci Technol B28(6):C6C74–C6C79

Chapter 1
Fransen MJ, van Rooy TL, Kruit P (1999a) Field emission en-
ergy distributions from individual multiwalled carbon nano-
General References tubes. Appl Surf Sci 146:312–327
Fransen MJ, van Rooy TL, Tiemeijer PC, Overwijk MHF, Faber
Bronsgeest MS (2014) Physics of Schottky Electron Sources, JS, Kruit P (1999b) On the Electron-Optical Properties of
Theory and Optimum Operation. Pan Stanford Publishing the ZrO/W Schottky Electron Emitter. Adv Imaging Electron
Pte, Singapore Phys 111:91–166
Fursey G (2005) Field Emission in Vacuum Microelectronics. Fujita S, Shimoyama H (2007) Mechanism of surface-tension
Kluwer Academic, New York reduction by electric-field application: Shape changes in sin-
Jansen GH (1990) Coulomb Interactions in Particle Beams. gle-crystal field emitters under thermal-field treatment. Phys
Advances in Imaging and Electron Physics, Supplement 21, Rev B75(23):75, 235431
Academic Press, Inc, San Diego de Heer WA, Chatelein A, Ugarte D (1995) A Carbon nanotube
Jensen KL (2007) Electron Emission Physics. Advances in Imag- Field Emission Electron Source. Science 270:1179–1180
ing and Electron Physics 149 Academic Press Inc, San Diego Hommelhoff P, Sortais Y, Aghajani-Talesh A, Kasevich MA
Kruit P, Jansen GH (1997 and 2008). Space Charge and Statis- (2006) Field Emission Tip as a Nanometer Source of Free
tical Coulomb Effects. Handbook of Charged Particle Optics, Electron Femtosecond Pulses. Phys Rev Lett 96:077401
edited by Orloff, J., CRC press. Houdellier F, de Knoop L, Gatel C, Masseboeuf A, Mamishin S,
Swanson LW, Schwind GA (1997 and 2008). Review of ZrO/W Taniguchi Y, Delmas M, Monthioux M, Hytch MJ, Snoeck E
Schottky Cathode. Handbook of Charged Particle Optics, ed- (2015) Development of TEM and SEM high brightness elec-
ited by Orloff, J., CRC press. tron guns using cold-field emission from a carbon nanotip.
Swanson LW, Schwind GA (2009) A Review of the Cold-Field Ultramicroscopy 151:107–115
Electron cathode. Advances in Imaging and Electron Physics De Jonge N (2004) Brightness of carbon nanotube electron
159:63–101 (Elsevier Academic Press) sources. J Appl Phys 95(2):673–681
Kawasaki T, Matsui I, Yoshida T, Katsuta T, Hayashi S, Onai
T, Furutsu T, Myochin K, Numata M, Mogaki H, Gorai M,
Akashi T, Kamimura O, Matsuda T, Osakabe N, Tonomura
Specific References AA, Kitazawa K (2000) Development of a 1 MV field emis-
sion transmission electron microscope. J Electron Microsc
Bahm A, Schwind G, Swanson L (2008) Range of validity of field 49(6):711–718
emission equations. J Vac Sci Technol B26(6):2080–2084 Kasuya K, Kawasaki T, Moriya N, Arai M, Furutsu T (2014)
Bahm A, Schwind G, Swanson L (2011) The ZrO/W(100) Magnetic field superimposed cold field emission gun under
Schottky cathode: Morphological modification and its effect extreme-high vacuum. J Vac Sci Technol B 031802
on long term operation. J Appl Phys 110 054322 Kruit P, Bezuijen M, Barth JE (2006) Source brightness and use-
Barth JE, Kruit P (1996) Addition of different contributions to ful beam current of carbon nanotubes and other very small
the charged particle probe size. Optik 101(3):101–109 emitters. J Appl Phys 99, 024315
Barth JE, Nykerk MD (1999) Dependence of the chromatic Liu K, Schwind G, Swanson L, Campbell J (2010) Field induced
aberration spot size on the form of the energy distribution shape and work function modification for the ZrO/W(100)
of the charged particles. Nuclear Instr Methods Phys Res A Schottky cathode. J Vac Sci Technol B28(6):C6C2–C6C33
427(1–2):86–90 Maunders C, Dwyer C, Tiemeijer P, Etheridge J (2011) Practical
Bronsgeest M, Kruit P (2008) Effect of the electric field on the methods for the measurement of spatial coherence-A compar-
form stability of a Schottky electron emitter: A step model. J ative study. Ultramicroscopy 111(8):1437–1446
Vac Sci Technol B26(6):2073–2079 Niklaus M, Hasselbach F (1993) Wien filter: A wave-pack-
Bronsgeest M, Barth J, Swanson L, Kruit P (2008) Probe current, et-shifting device for restoring longitudinal coherence in
probe size, and the practical brightness for probe forming charged-matter-wave interferometers. Phys Rev A 48:152–
systems. J Vac Sci Technol B26(3):949–955 160
Bronsgeest M, Kruit P (2009) Reversible shape changes of the Swanson LW, Schwind GA, Kellogg SM, Liu K (2012) Com-
end facet on Schottky electron emitters. J Vac Sci Technol puter modeling of the Schottky electron source. J Vac Sci
B27(6):2524–2531 Technol B30 06F603
Bronsgeest M, Kruit P (2010) ‘Collapsing rings’ on Schottky van Veen AHV, Hagen CW, Barth JE, Kruit P (2001) Reduced
electron emitters. Ultramicroscopy 110(9):1243–1254 brightness of the ZrO/W Schottky electron emitter. J Vac Sci
Cho B, Shigeru K, Oshima C (2013) W(310) cold-field emission Technol B19:2038–2044
characteristics reflecting the vacuum states of an extreme high Zewail AH (2006) 4D Ultrafast Electron Diffraction, Crystallog-
vacuum electron gun. Review of Scientific Instruments 84, raphy and Microscopy. Annual Rev Phys Chem 57:65–103
013305
In situ and Operando
2

Chapter 1
Chapter Preview

Chapter 2
This topic is a very broad one since it involves everything that uum is to prevent contamination of the sample and to avoid
has been covered in W&C and then allows the specimen to gases streaming up the column and degrading the vacuum of
change while you image it, record the spectra, and/or measure the electron gun or depositing on the window of the XEDS
what changes are taking place. We are combining two topics detector. Now we have two ways of changing the environ-
that could be treated separately, namely in-situ experimenta- ment: we can use special holders, or we can actually modify
tion and controlled-environment TEM (which we’ll call an the column around the specimen using extra apertures and
ETEM, but you’ll also see E-TEM or eTEM). The reason for differential pumping.
combining the two is that changes in the environment about
the sample can change the material as we examine it in the The field of in-situ TEM is changing rapidly for two rea-
TEM. With TEM, it’s always an in-situ study! We’re also sons: nanotechnology is being used to build better, more
including operando with in situ; operando implies that the flexible holders, and we have better, faster ways of col-
material is behaving as it would do in “real life”; the transla- lecting and recording the information from the changing
tion is “working.” For example, if a catalyst particle is actu- specimen. Of course, the TEMs are also better and more
ally acting as a catalyst when we observe it, it’s an operando stable, but that is not what is presently driving the progress.
study; in general such a particle can’t act as a catalyst unless
there is something to catalyze! You are often measuring some- So in this chapter we will outline what the holders can, and
thing (there is a metric) that relates to the performance of the maybe can’t, do. We’ll discuss how the environment can be
materials – so that you know it’s performing! controlled, how we can record the changes, and how we can
correct for drift when it is occurring. We’ll also mention the
In the traditional TEM, the environment is always very re- cost of holders, why you might not have the one you want,
ducing – traditionally we use the highest vacuum we can; and what you can do about it. There is no difference between
historically we deposit a layer of carbon on the sample too an in situ experiment and an in-situ one – it’s just grammar
(whether we wanted to or not). The reason for the clean vac- and both are correct.

© Springer International Publishing Switzerland 2016

C. B. Carter, D. B. Williams (Eds.), Transmission Electron Microscopy, DOI 10.1007/978-3-319-26651-0_2 17
 18 2 In situ and Operando

2.1 General Principles

Chapter 1

If you are asking ‘why do we want to do in-situ microscopy?’

you should probably skip this chapter for now. The purpose of
the chapter is to show you the potential of this type of micros-
copy and to give you an idea of how you can apply the tech-
niques. For example, if you are processing nanoparticles, you
might want to understand how they change during processing.
Similarly, if you study catalysis, you know that the environment
is a critical factor in determining and controlling the processes
that take place; most catalysis does not take place in a vacuum,
especially not in an ultrahigh vacuum (UHV). So we need to be
Chapter 2

able to change the environment around the specimen, but the

atmosphere in a regular TEM is very reducing. Old TEMs tend Fig. 2.1 The power of in-situ TEM: controlled manipulation of nanosized objects,
to deposit carbon on the sample as the electron beam cracks oil and seeing what you are doing! Carbon nanotubes are being used to relocate
vapor in the column. Of course, the vacuum can be an advan- iron nanoparticles
tage if we want to study film growth in situ, e.g., for molecular
beam epitaxy (MBE), because we are already close to the actual
growth conditions. So the question then becomes, can you con- Materials change as they are used, often due to exposure to cor-
trollably change the environment if you must? rosive environments. The changes take place at the surface of the
sample and propagate into the bulk material. If the material is a
The realization that nanomaterials are important and that they nanoparticle, there is no bulk! Atomistic mechanisms are involved
often have properties differing from bulk materials has led to whenever a material interacts with the environment. In-situ TEM
enormous interest in TEM characterization. Gradually, we have can directly reveal how these processes occur – especially if you
realized that it is not enough to be able to image these small parti- can obtain the necessary resolution (temporal and spatial).
cles because they change: nanoparticles of ‘inert’ materials can be
very reactive. We want to understand how and why they change. In-situ manipulation of a nanoparticle is illustrated in Fig. 2.1;
this figure really shows the power of in-situ experimentation –
Now, it is not enough to synthesize a novel material, we must we can see what we are doing to a nanoparticle!
understand the origin of novelty. In order to understand the ori-
gin of novelty, it is important to understand how the atoms come Many other components of the TEM will be close to the in-situ
together to form a cluster of atoms, and at a later stage, how they holder, including the objective lenses, the objective aperture,
combine together to form the final novel material. This is where the cold-finger, and possibly the XEDS detector(s). When de-
in-situ microscopy can play a unique role, using careful experi- signing a new holder, the proximity and interaction with these
mental design of the process in the TEM: in-situ TEM provides microscope components must be considered. The effect of the
the link between structure, processing, and performance. Tradi- magnetic field (~ 2 T) of the pole-piece on the specimen has al-
tionally, in materials science and biology, a series of samples are ways created a challenge for imaging magnetic samples even if
synthesized by varying the processing parameters and then they they are fixed tightly in the holder; many older TEMs have such
are characterized under the microscope. What we want to do now samples decorating the interior of the column.
is avoid this post-mortem approach by using in-situ studies and
continuously vary the processing parameters while the material is It takes time and effort to learn an in-situ technique; attention to
still inside the microscope. This can allow us to analyze transition detail and persistence is the key to success.
states that we wouldn’t otherwise see! We say that in-situ exper-
iments allow us to study the full parameter space. We will only
give examples as illustrations, not as an end in themselves – we’ll
give references for you to pursue the applications. 2.2 Some history
Operando
The history of in-situ TEM studies is relevant, because you
will read and use reports that are in the literature. The in-situ
You can read about the origin of the term operando. It is specimen holders used to be top-entry and side-entry. Now they
the ablative form of the gerund of the Latin verb operari are nearly all side-entry. We often have two competing ‘moti-
so we italicize it and do not add ‘in’! It has same meaning vations’: i) improve the vacuum so that there is less contam-
in Spanish and Italian. The other rule is that we stop using ination, and ii) find ways to envelop the specimen in a liquid
italics when the word or phrase is in common use. 9 or a gas. A FEG source has to be in the best vacuum possible
to maximize its stability and lifetime. Remember: the gun may
 2.3 The Possibilities 19

have to go back to the factory if the tip is destroyed. To minimize

contamination and back-streaming to the gun, the vacuum within
the column for a modern TEM is ~ 1 × 10−7 torr.

Chapter 1
The traditional problems for in-situ TEM studies have always
been specimen drift, contamination and alterations caused by
the electron beam; and they still are! We can minimize the first
two, but the third is unavoidable. Historically, we had three types
of in-situ holders: straining, heating, and cooling, but we could
generally only have one of these at any one time. Some in-situ
studies were carried out in high-voltage (1 to 3 MV) machines
so as to see through thicker specimens at high resolution, which
more closely represent bulk material. Often these high-kV ma-

Chapter 2
chines were used to simulate the effect of ion irradiation, which
could be carried out in combination with straining, heating, or
cooling. With the development of nanotechnology, new holders Fig. 2.2 Schematic summarizing the different sections of in-situ TEM
have become available; this allows us to be more quantitative
and to combine stimuli.

The main emphasis in W&C was often to optimize all the com- 2.3 The Possibilities
ponents of the microscope to maximize the spatial resolution.
We usually worked hard to obtain a high vacuum (aka a low
pressure) and a contamination-free environment. But now our Since this topic can take up several books itself, let’s look at the
philosophy is going to be slightly different. We now want to ob- scope of the topic first. The schematic in Fig. 2.2 summarizes the
serve changes in the specimen/thin foil as some external stimulus possibilities we are discussing. We may want to combine these
is changed. topics. We may want to strain the sample while heating it. We
may want to apply the strain locally (indentation) rather than
The emphasis here is to be able to see things as they happen, and globally (compress the specimen). We may even want to strain
so, we may compromise on the resolution to be able to achieve the specimen while heating it in a gas and irradiating it. We must
this. While in a conventional TEM, we take elaborate steps to also think about how fast or slow we want the stimulus to be, and
minimize thermal gradients (to minimize drift), we actually heat for how long. How quickly do we need to record the specimen’s
the specimen (often to fairly high temperatures) when we want reaction after stimulation?
to study thermal transformations in situ. Since the thermally in-
duced transitions often depend upon the environment, we may TEM as a Lab
also decide to deliberately contaminate the TEM by passing
gases into the sample chamber. Of course, we try to do this in a
controlled fashion such that this contamination is restricted to a The basic idea is to use the TEM as an in-situ experimen-
region very close to the specimen. tal laboratory with atomic resolution. 9

The Stimulus
Some points to keep in mind as we discuss this topic:

This stimulus could be chemical, the environment, a tem- zz Always keep safety in mind – both your own and that of the
perature change, stress or staining, laser illumination, an TEM and specimen holder.
applied external electric or magnetic field, etc. 9 zz You will certainly be able to extract data from an in-situ ex-
periment; the challenge becomes understanding the data and
its relevance.
The big challenge is how do we fit in-situ capabilities within zz Before you begin an experiment, think about what the sample
the pole-piece gap of the TEM? In the future, we will see that will ‘see’ during the experiment; that includes the environ-
one solution is to use aberration correction to increase the pole- ment (chemical, pressure, temperature, stress, magnetic and
piece gap without forfeiting the obtainable resolution. That’s the electric fields), any grids, the holder, or anything else that
future, but in many labs it is already here. might influence the reaction/mechanism.
zz What accelerating voltage should we use in the experiments?
For thicker samples, a higher accelerating voltage is prefera-
ble. For liquid-cell studies or indentation on bulk-like struc-
tures, a medium voltage is preferred. If we want to minimize
the effect of the beam, a low voltage may be necessary. In any
 20 2 In situ and Operando

Fig. 2.3 A grain boundary in α-Ti

building up dislocations during in-si-
tu loading, followed by the eventual
Chapter 1

ejection of dislocations with several

slip systems shown as B, C and D,
where C was dominant. The initial (a)
and final (c) states would not give a
clear understanding of intermediate
(b) state without in-situ observation
Chapter 2

case, how long the beam irradiates the samples may well be 2.3.1 Post-Mortem Characterization
critical. Beam control can be automated in STEM. Of course,
you always want to know what the beam current is. Use a
Faraday cup to check it. Post-mortem analysis of deformation only provides the end
zz It almost goes without saying that we are always interested result from a nanoindentation experiment. In-situ observation
in the kinetics of processes; some changes, like the phase provides real-time imaging for a complete story of transition
changes in phase-change materials (PCMs) are very fast, but states that can be related to the stress-strain measurements. This
we can now be quite quick too. can be seen in Fig. 2.3, where dislocation accumulation at a ran-
zz What environment do we want or need? The environment dom grain boundary in α-Ti produced elastic distortion, until
will be critical for catalysis, but also the in-situ growth of slip systems were activated to minimize the accumulated strain
thin films. energy. This is why in-situ microscopy is so powerful. You see
zz What temperature will we need? We like room temperature what happens and the order in which it happens, instead of trying
because that is where the sample is generally most stable. to interpret what has happened during a reaction or process by
zz We are often most concerned about how materials respond to comparing the initial and final states. We can view the processes
stimuli – to applied fields – this is the basis of the electronics as they occur within the microscope with nanoscale resolution
industry, the memory industry, etc. We’d like to be able to see (or better). Real-time imaging of dynamic processes enables us
directly the impact from an applied field. to view the effect that defects have on processes and structures.
zz Whether we are growing thin films, simply depositing parti- In a post-mortem analysis, you may overlook many factors. You
cles, or implanting atoms, we are often adding material. Of could simulate the process, but hopefully you would remember
course, material will also be removed from a structure by to check what really happened. However, post mortem analysis
sputtering or by displacement, such as bubble formation in is essential for many in-situ studies; e.g., a reactive environment
a liquid cell. might not allow you to analyze the specimen completely – you
zz There is always an advantage to using energy-filtered imaging hope to stop the reaction and then probe it in a stable state. This
to decrease the effect of chromatic aberration. is especially important when chemical analysis is not possible
in the configuration you may be working with, such as older
We are not going to list every variant of every experiment, but designs of liquid-cell holders.
we want to help you see the possibilities. The general question
is ‘how does the material change over time?

Keep in mind that we should always carry out the corresponding 2.3.2 Statistics
ex-situ analysis to be sure that beam effects have not completely
changed the experimental response. Results from the ex-situ
experiment will help us know whether or not the study can be Before you perform in-situ TEM experiments, consider the sta-
completed in situ within the TEM. The experimental study on tistical significance of the experiment and its reproducibility.
a microelectromechanical system (MEMS) heating stage could Questions you should be asked at every presentation of your
be carried out ex situ in an external vacuum chamber to repro- work will concern the set-up and control of your in-situ TEM ex-
duce the experimental conditions in the microscope without the periment: the imaging conditions, beam effects, the defects in the
electron beam. Overall, it might be easier to leave part of the material, especially when caused by specimen preparation, and
specimen unexposed to the electron beam until the end of the how these considerations influenced your findings. It is so im-
test, to compare characterization with the exposed region. portant to spend time with your experimental design so that you
don’t end up repeating all your hard work in an effort to confirm
your interpretation, but you need to be able to when necessary for
the scientific process! Reporting the complete characterization of
the parameters defining your experiment will leave a legacy for
 2.3 The Possibilities 21

Fig. 2.4 Measurement of hundreds of inter-nanoparticle spacings

of DNA-assembled Au nanoparticles, where the samples dried
configured to spacing far below that which was predicted by the

Chapter 1
DNA attachment in solution. Rows show the time that particles
were allowed for attachment. a Hybridized Au nanoparticles im-
aged on a dry surface in vacuum. b Hybridized Au nanoparticles
imaged in a continuous liquid

Chapter 2
your published results, even if your printed interpretation turns smaller inter-nanoparticle spacing than those you would observe
out to be incorrect. if the arrays were in solution, Fig. 2.4. This systematic approach
provided enough statistical evidence to prove the conclusion,
Give some thought to the relevance of the properties observed though reproducibility in the experimental design was required.
within the TEM to the bulk materials that we are trying to charac-
terize. The electrical, mechanical, magnetic, optical, and reaction We do not want to introduce artifacts that will affect our in-situ
properties of materials measured in situ will all be affected by measurements; otherwise we are only studying sample-prepara-
the free surfaces available, particularly when in contact with gas/ tion artifacts. Pay attention to the preparation of the specimen
liquid. This is especially important when you consider the sub- – it’s at least as important for in-situ analysis as it is in conven-
strates used to support small catalyst particles or reactive species. tional characterization. An example is in preparation of a semi-
In TEM, we cannot avoid having relatively large surface areas conductor specimen for characterizing the electrical properties
(especially of heavy elements like metals), since the samples are of the material: if you use an ion mill to thin the sample, you will
thinned to electron transparency or are nanoscale to begin with. implant ions into the material. You may not see these dopants
The advantage TEM provides is location-specific detail. Spe- in the image or detect them in a spectrum, but they will highly
cific test cases could be set up to determine the effect of defects, influence the electrical behavior of the material. Therefore, once
grain boundary orientations, doping, and crack dependence on again, the success of the TEM experiment for in-situ analysis
the nanostructure in comparison to an averaged response in bulk will be based on how well the specimens are prepared before
studies. As an example, many researchers have used conventional you put them in the TEM. We have to apply statistics to in-situ
TEM analysis on carbon grids to study nanostructured assemblies, studies so that we can compare them to other measurements; you
though it has been shown through controlled experiments that may find the article by Spitzer et al. (2014) interesting. (Remem-
during drying, the free surfaces cause these assemblies to have ber the meaning of ‘typical image’.)
 22 2 In situ and Operando

2.4 Time The goal is to use computers to identify the data that you are
looking for without manually searching through thousands of
images. Still, it is important to plan the experiment well before
Chapter 1

The experiment defines the temporal resolution required to cap- you start recording it, as you can quickly create more data than
ture the events of interest happening in a material. For exam- a person is able to examine. While doing the experiments, you
ple, atom migration and transient states during phase transitions store data directly onto the TEM’s computer and then transfer
require sub-microsecond timeframes to capture those events, it to a removable drive; many managers don’t want their TEM
whereas corrosion and radioactive decay occur on much longer directly connected to the internet.
timescales of hours to months. So, when approaching the char-
acterization of a material process or mechanism within the TEM, You may even just want to use video to allow you to correct for
we should first define the process duration and then choose the drift of the sample. To do this you’ll need to find the software
instrumentation. that is compatible with your TEM. But remember, our repeated
warning: having the beam on so that you can see the specimen
Chapter 2

may be enough to change what happens to your specimen! So,

blank the beam for various intervals upon repeating the experi-
2.4.1 Recording the Data ment to check how bad the beam effects are.

The first in-situ experiments were captured by pointing a video

camera directly at the phosphor screen in the viewing chamber. 2.4.2 The CCD Camera
The first videos of dislocations moving were recorded in this
way. There could well be a situation where this will still be the
best approach! In this case, you will be limited by how fast your CCD cameras are now attached to all research-grade TEMs. The
phosphor can respond to changes in the intensity of the electron CCD cameras may be side-entry or bottom-entry. With a side-en-
beam and what resolution it can give. You record changes in your try camera, we do not compromise on the field of view. The
specimen as they take place to prove to others that you understand bottom-entry camera does compromise on the field of view, but
how the changes occurred. Recording technologies have changed, the magnification is ~ 10× larger. In either case the data is stored
along with the development of microscopy instrumentation. digitally. The parameters you should know for in-situ micros-
copy imaging are the digital size of the camera, the sensitivity
Of course, you have to record useful and correct data while ac- of the pixels, the threshold electron dose that a pixel can tolerate,
tually doing the in-situ experiment. You will often be looking and the lifetime electron dose per pixel of the camera. Cameras
at one particular region of a sample while the most interesting traditionally use an optical-fiber-based stack technology with
things are happening at the other end of the specimen. This could multiple read-out ports for high-contrast resolution. While doing
be due to a beam effect, non-uniformities in the environment the in-situ microscopy, you will always work to optimize the
across your specimen, or just luck. So you need to keep your eyes camera parameters. Better image quality usually means poorer
open for anything that you are not seeing! We can use several dif- time resolution. Of course, the exposure time depends on the
ferent approaches. One comes with experience. You can switch quality of the sample, conditions of the experiment, etc.
from the CCD camera to the microscope screen for a wider field
of view, but this may interrupt the video recording. During in-
situ microscopy, the sample and the environment may be chang-
ing continuously; the focus condition also keeps on changing. 2.4.3 Direct-Detection Cameras
You need to keep adjusting the focus while recording the images.

The ideal recording media would have a detector quantum ef- Direct-detection cameras are quite new and currently very ex-
ficiency (DQE) of 1, with a large number of pixels and rapid pensive. The principle is rather like the operation of a second-
readout rate. As more data is recorded, storing and processing the ary-electron detector in SEM. CCD cameras work by converting
data becomes challenging; TEM is now a producer of ‘big data’. an electron signal into photons using a scintillator; direct elec-
With the new high-speed 2k-by-2k cameras, 60 seconds of im- tron detectors detect each electron that reaches each pixel, so the
ages acquired at 400 frames per second (FPS) produces ~ 160 gi- DQE is much higher. Examples (2015) are the DE-series (up to
gabytes of data. To convert this data from DM format to indi- 67.1 megapixels) from Direct Electron, Falcon II (16 megapix-
vidual tiffs would require a full day using the DM batch convert els) from FEI, and the K2 (57 megapixels) from Gatan. Since
software. (See Chap. 6 for more on DM – digital micrograph.) the DQE is higher (in principle, it is 1, but in practice between
With the new direct detection cameras, described in Sect. 2.4.3, 0.35 and 0.8), the ‘exposure time’ can be much shorter. The first
15 minutes of recording can produce up to 5 terabytes of data. Do users were those in the biosciences, where low dose is recog-
the calculation: even 30 FPS at 1k by 1k per frame for 15 minutes nized as being critical and money is more plentiful. Now these
produces around 27 GB of data. Researchers and companies are cameras are being applied to materials science questions, such as
addressing this issue of too much data (big, but redundant, data). understanding the interface structure between silicon and a NiSi2

Licensed to ([email protected])
 2.4 Time 23

Chapter 1
Chapter 2
Fig. 2.5 Si nanowire growth at an interface with NiSi2 nanocrystal at 400 ˚C, images acquired at 400 frames per second. Angle between crystals is indicated, where
rotation for epitaxial alignment is achieved over the 37.1-s period. These high-speed cameras can resolve atomic-scale information when using ‘enough’ electrons

nanocrystal during Si nanowire growth, Fig. 2.5. These cameras 2.4.4 Software and Data Handling
generally will cost over 500,000 (2015) US dollars. For our in-
situ studies, the video frame rate of 1,600 FPS is the enormous
improvement, but the amount of data that you must store and The most important part of in-situ TEM is often considered as
transfer is correspondingly large. (Think about this challenge!) an afterthought! You know you will use an electronic device
A recent advance using these high-speed cameras is the ability to record the image, but what software will you use? We list
to collect full tomographic data from a structure within seconds, some of the manufacturers at the end of the chapter. (Notice
though stability of the holder and focus conditions are the key we are already no longer saying CCD camera.) One important
factor in the quality of images obtained. aspect that we have to discuss is the computer and its processing
 24 2 In situ and Operando

time. Image recording, processing, and saving, depends on the Processing Data
computer that you are using. The computers that are generally
interfaced with the microscope should be state-of-the-art and fit
Chapter 1

the hardware configuration when the instrument is purchased. You must report any changes that are not just correcting
Shortly after it’s purchased, it will no longer be state-of-the-art; levels (brightness/contrast) or cropping the image, even
it will still be adequate for running the microscope – but not for in video. Be sure to record and report any modification
doing state-of-the-art image processing, etc. done to the video data just as you would for images. 9

Many in-situ experiments require the use of multiple computers

to operate the TEM separately from the holder or external stim-
ulus controller. Currently, there are ways to use both hardware
(master clock) and plotting software (LabVIEW, Python, Matlab, 2.4.5 Drift Correction
Mathematica, Igor, Origin, and Excel) to synchronize the quan-
Chapter 2

titative property measurement from the holder with the video/

images collected on the TEM. To correctly overlay the video Depending on the task at hand, you have to set the working
onto the plot, it becomes necessary to input the exact moments range of magnification. If you’re working in nanoscience and
each image/data point was acquired at. If you are working on this technology, the magnification is generally high. This worsens the
analysis, you should refer to the website spacetime.uithetblauw. drift problem for in-situ microscopy. General methods to reduce
nl (08/2015) to synchronize your external data with your image. drift in a series of images include using different support grids,
Camera software can usually place a timestamp directly on the piezo-controlled goniometers, drift-corrected holders, and auto-
image, so that this information cannot be separated from the mated drift-correction software. Hardware may use electronic
data during the time it takes between collecting the results and drift compensation with piezo-driven stage control or image-shift
publishing them. At present, we usually use LabVIEW to control deflection coils to correct for drift in the images (Banhart 2012),
laser beams and ion beams incident on the specimen. Many use though the drift-correction software and drift-corrected holders
Comsol® software to model the various in-situ systems, pro- on the market are not really drift-corrected for in-situ micros-
cesses, and environments. copy. The sources of drift for in-situ microscopy are expansion
or contraction of the sample, specimen charging, temperature
The camera you use will have its own image-grabbing software variations induced by the beam, reaction of the samples with
package, but you’ll find that different users have different philos- the environment, flowing water inside the holder for cooling,
ophies. In practice, you’ll optimize the camera parameters and and occasional bubble burst inside the cooling channels; or the
then start the auto image-grabbing software: record images con- sample may change and thus move in most in-situ experiments.
tinuously. The negative part of this is that you can only see the
images while they are being recorded! If your microscope does The drift thus caused can only be corrected manually by observ-
not have software to record videos, you can load software onto ing the images while they are being recorded, Fig. 2.6. Auto-
the microscope control computer (with consent of the instru- mated image-drift correction works by keeping a pre-selected
ment manager) to collect screen captures of the search screen or reference region (fiducial) in the field of view. The amount of
scanning frames. Examples include VirtualDub and CamStudio,
which allow you to select a region of the monitor to collect a
screen shot with a predetermined capture rate (usually synched
with the frame rate while searching).

Post-processing of the videos and images can be completed using

the same software you use for conventional TEM images. ImageJ,
Fiji, Python and DM are free; Matlab, Mathematica, Camtasia
Corel are not. (Check out youtube.com for instructional videos
on using the software, especially for ImageJ.) For example, you
can use image registration to correct for drift in an image series
or group of video frames. Scripts can also be found through var-
ious sources; there are collections for DM or for Matlab such as
Smart Align (lewysjones.com (08/2015)). Some image-process-
ing techniques may be tricky for in-situ experiments when the
background is not constant, such as with bowed membranes in
liquid-cell experiments. In such images, it may be useful to apply
more specialized algorithms to gain information from the exper- Fig. 2.6 Tracking of the thermal-drift velocity, as the displacement rate between
imental data, such as adaptive threshholding. Images recorded consecutive images (dark blue), of a heating experiment from room temperature
on high-frame-rate cameras may require their own software to to 501 ˚C (red). Light-blue points show the manual drift corrections during the
be able to convert the bulk data into a format that is compatible series of images. Displacement spikes observed at each step increase in tempera-
with more conventional image-analysis software. ture, larger displacement values for larger steps
 2.4 Time 25

drift in an image may be related to the electrical conductivity of pulse, but the temporal resolution is determined by the duration
the specimen/grid and the beam dose on the specimen (related of the acquired pulse.
to beam-induced heating). Remember, when only a small region

Chapter 1
of the sample is exposed at one time, charge build-up is limited Fast vs Ultrafast
and beam-induced movement is reduced. Drift performance for
in-situ holders should have been tested and recorded for your
instrument when they were installed, but that might have been This high-speed electron microscopy has come to be
10 (or 30) years ago! known as ultrafast electron microscopy. (1,600 FPS is
fast but not ultrafast!) 9
Good Practice

The development first targeted imaging and diffraction at high tem-

For all TEM, in situ or not, keep the o-ring on the spec- poral resolution by Bostanjoglo in 2000, framed in the use of laser

Chapter 2
imen rod properly lubricated. Tap the base of the holder pump-probe measurements. In this design, an external stimulus
to seat the o-ring and improve contact between the holder induced a modification to the specimen; the imaging signal, has a
and goniometer, but be aware of when you shouldn’t!9 predetermined timed delay, allowing us to collect the information
of the sample’s response. In this experimental design, the tempo-
ral resolution is only limited by the pulse duration. Based on this
Note: be careful when tapping to seat the o-ring! Don’t tap the pump-probe approach, two methods have been developed where
holder if doing so will move your carefully aligned specimen or repeated diffraction or imaging compiles single-electron pulse
probe tip (e.g., if you’re using an STM-TEM holder). Drift or events for reversible processes – this is the stroboscopic UEM,
mechanical vibrations may be caused by the cabling connected to or an individual packet of electrons is used to produce a single
the base of the holder (common for in-situ and double-tilt hold- image or diffraction pattern for irreversible processes – this is sin-
ers). Pay careful attention to stabilizing this cabling and avoid gle-shot UEM (commonly known as dynamic TEM (DTEM)). The
moving cables during the experiment. Electrical experiments can single-electron pulse actually consists of 1–10 electrons, whereas
be destroyed if static discharge from moving the cables causes a the packet of electrons is generally ~ 109 electrons. Although the
sharp pulse to flow through a nanomaterial sample. DTEM is a single-shot process, we can record a series of such
images with a defined time delay to make a short video!

Using the stroboscopic approach, a femtosecond pulse continu-

2.4.6 Ultrafast Electron Microscopy ously collects information from events that occur on an ultrafast
(< 1 ns) timescale. For this to work, the specimen must return to
the unperturbed state prior to interaction with the next pulse. In
What happens when we want to observe dynamic events in mate- the single-shot approach, a single packet of electrons collects
rials that occur in very short times? For example, phase transfor- information from a sample in the nanosecond time regime (the
mations, transient states, or nanostructure nucleation can happen time needed to fit in enough electrons without significant repul-
in millisecond or shorter times. We need to be able to acquire sion), which is important for irreversible processes to produce
images in series that are representative of the timescale upon an image/diffraction pattern with reduced noise. In both of these
which the dynamic events are taking place. methods, the pulse duration of the incident laser is about the
same as the emitted electron pulse. Bostanjoglo’s single-shot
During TEM imaging, images can be acquired at standard video DTEM was able to collect an image using 7- to 11-ns pulse du-
frame rates of 33 ms per frame (30 FPS), or with an advanced ration, with times between pulses of 20 ns to several microsec-
high-speed camera down to 625 µs per frame (1,600 FPS as onds, producing an image with ~ 100-nm spatial resolution. This
we saw in Sect. 2.4.3). Even the fastest cameras still limit the microscope was also the first to use an electrostatic beam-shift,
kinds of dynamic processes in materials that we can study using located after the specimen, to deflect a pulse train onto different
nanometer-resolution real-space imaging, and we always need regions of a CCD to produce a ‘movie’, i.e., a short sequence
to consider the electron beam dose necessary to collect enough of images, of the irreversible process. The TEM modifications
information from the sample with the high-frame-rate cameras. of the stroboscopic and single-shot systems are similar, though
A change in electronic states and atomic-scale fluctuations in the number of electrons released from the gun differs. The sin-
materials can occur within a femtosecond (1 fs is 10−15 s), which gle-shot approach uses greater laser beam energy with a low
is much too fast for standard TEM imaging. Ultrafast electron repetition rate; the stroboscopic mode uses the opposite. The
microscopy (UEM) has set out to overcome these barriers and photoelectron source can be used as a thermionic emitter, pro-
to increase the temporal (time) resolution for TEM imaging/dif- viding conventional TEM imaging when desired.
fraction down below the nanosecond time regime. The key de-
velopment in UEM is that the camera speed no longer defines the A standard TEM can actually be retrofitted to become a UEM by
temporal resolution; instead the beam is limited to emit electrons adding at least one laser system to the microscope column that is
for a very short defined pulse. The camera records the data over a deflected onto the tip of the photoemission gun. A second laser
much longer timescale by leaving the shutter open to wait for the is commonly used to induce synchronized heating or ablation of
 26 2 In situ and Operando

toemission source. Synchronizing the sample- and cathode-drive

lasers provides the ability to observe changes occurring within
1 ns from its stimulated initiation. If we increase the rate of image
Chapter 1

acquisition, we’ll degrade the coherency of the beam and lose

high-resolution information; it’s a trade-off between spatial and
temporal resolution.

It’s interesting to think about the physics of the single-shot UEM.

The number of electrons that can be emitted from the surface of
the cathode within a short pulse is limited by the Child-Lang-
muir effect (the space-charge-limited emission of electrons from
a source). As the emitted packet of electrons travels down the
electron column, individual charges repel one another, so the
Chapter 2

packet is broadened, causing the spatial resolution to degrade.

This space-charge effect limits the number of electrons collected
by the condenser system. To reduce these space-charge effects,
we sacrifice some of the temporal resolution and extend the
pulse duration. This should increase both the number of elec-
trons contributing to the image and the spatial resolution. The
original single-shot UEM used a Ta disk cathode that emitted
~ 3–4 × 107 electrons per 10-nm pulse from a 200-μJ laser. Re-
member, mass-thickness and diffraction-contrast imaging modes
are less dependent on coherence.

Irreversible processes, such as the melting of thin metal films

Fig. 2.7 Bringing laser beams into the TEM; the experimental set-up. One la-
ser system is incorporated above the condenser lens system to deflect onto a
(Al, Cr, Ni, Ni-P, Co, and Pt) and phase transformations, have
photoemission cathode, while a second laser is directed with mirrors onto the been studied with the DTEM (Fig. 2.9). We can use laser-heating
specimen to stimulate processes in the material to reach high temperatures within short pulse durations. Dark-
field imaging has provided temporally resolved images of grain
boundary, stacking fault, and dislocation motion.
the sample. This modification is now commercially available,
Fig. 2.7. It’s easy to bring light into the TEM – we can use win- We can expect improvements in the design of photoemission
dows – and not affect the vacuum! guns as DTEMs become more popular. Already we see brighter
sources, and plans for back-illumination of the gun to remove
During an experiment, the laser that is directed onto the sample the aperture created by the mirror. As in any gun, the material
can stimulate a reaction or change in the sample. After a pre- used for the photoemission source used can greatly affect the
defined time delay, the second laser is directed onto a photoemis- brightness and current density of the emitted electron pulses.
sion source producing emitted electrons, which are then focused Materials that have already been considered include a combi-
to interact with the specimen and form the image/diffraction nation of photoelectron emitters (CeB6, LaB6, CeS, ZrC, Ce,
pattern. In contrast to high-speed cameras, which still require Tb, Ti, and Zr) on hairpin metals (Ir-W, Nb, Re, Ta, and W).
a high electron dose to provide enough contrast in an image, A Zr-coated Re hairpin cathode is producing the highest electron
UEM is inherently a low-dose technique, evidenced by the fact current densities and is operating as a thermionic emitter. Pho-
that only a relatively small number of electrons are emitted to to-excitation would allow the cathode material to be optimized
interact with the specimen and the time of this interaction is lim- for quantum efficiency without worrying about stability at high
ited. UEM thus combines a relatively high spatial and temporal temperatures.
resolution with low-dose imaging! The information transferred
from the sample is a result of the laser-initiated process, not an You can make movies using a high-speed electrostatic deflector
electron-beam effect (Fig. 2.8). to deflect a defined number of pulses on different regions of
the CCD; In 2015, the number of frames is 9, but that’s just the
beginning (Fig. 2.10). (If you’re interested, check how many
Irreversible Processes frames can currently be collected with an SEM using parallel
The single-shot UEM (or DTEM) can be used to study phase beams.) An arbitrary waveform generator (AWG) for microsec-
transformations, crystallization, and nanowire growth. Currently ond pulsed imaging, aberration correction, and more control over
(in 2015), the record for the temporal and spatial resolution is a the parasitic electron–electron interactions within the electron
15-ns pulse with 9-nm features resolved. The size of the electron packet are planned (or hoped for). The AWG can extend the
packet is related to the bias on the cathode and the laser pulse du- pulse duration of the laser on the photoemission source to vary
ration. A 15-ns packet will contain ~ 109 electrons from the pho- from 10 ns to several microseconds. With an increase in the pulse
 2.4 Time 27

Chapter 1
Chapter 2
Fig. 2.8 Imaging with one electron at a time allows for standard imaging, diffraction, STEM, and EELS. This is demonstrated by the pristine atomic-scale images,
EFTEM boron map, atomic-resolution diffraction patterns, and defined EEL spectra from a BN specimen

duration and a brighter electron source, imaging down to 1 Å for One instrumental aspect of Zewail’s work is that the group ini-
a 1-µs pulse is theoretically possible. The importance of using tially used a 120-kV TEM (an FEI T12 Tecnai) with a LaB6 elec-
an AWG is that it allows us to circumvent the laser design where tron source for this pioneering work, and the LLNL group used
a long pulse path length is required; a 1-µs time delay means an old JEOL 2000FX! Such microscopes are found in every char-
that the pulses will be 300 m apart. Instead of housing the laser acterization lab. The point we’re making is that groundbreaking
tables in another room, this allows it to all be set up right next TEM designs are often not made on state-of-the-art equipment.
to the TEM column.

Reversible Processes
Stroboscopic Pump-Probe Method Electron–electron interactions can be avoided by having a single
The advance pioneered by Nobel Laureate Ahmed Zewail re- electron emitted per pulse: hence the stroboscopic UEM. The
sulted from his work on femtosecond chemistry. His work took pulses last for femtoseconds: hence ultrafast and stroboscopic.
the UEM to a faster regime using femtosecond lasers. The In the stroboscopic UEM, space-charge effects are eliminated:
stroboscopic UEM is not widely available, but can achieve there is only one electron (or a small number (< 103) of electrons)
unique results. Examples of applications with a time resolution excited in each pulse. If you continuously cycle the pump-probe
of ~ 100 fs include observing the grain growth in Al, the met- lasers at a high frequency, you generate enough electrons (> 1012)
al-to-insulator phase transformation in VO2, the flow of liquid to sum them together to form an image or diffraction pattern
along a nanotube, and the axial and radial expansion in carbon (~ 107–109 cycles). Stroboscopic images with resolution equal
nanotubes. The latter can occur in a time of, e.g., 5.0 ± 0.3 ps! to that obtained using constant illumination have been demon-
 28 2 In situ and Operando
Chapter 1
Chapter 2

Fig. 2.9 Melting thin Al metal films on a Si3N4 substrate, a microsecond-scale irreversible process, images, and diffraction patterns collected in the DTEM and ana-
lyzed using the radial average intensity
 2.5 The Environment 29

Chapter 1
Chapter 2
Fig. 2.11 Stroboscopic UEM can sum single electron-scattering events to
form an image or diffraction pattern at varied time delays (t1 = 0, t2 = 3 ps, and
t3 = 6 ps). This study revealed that the phase transition of polycrystalline vana-
dium dioxide from monoclinic to tetragonal is mediated by a nonequilibrium
metastable structure

Fig. 2.10 Movie-mode DTEM currently deflects a nine pulse-train series of 17-
ns pulses onto separate regions of the same CCD exposure. Images of a liquid
state (dark line at reaction front) during the reaction of a multilayered 2Ti-3B foil
2.5 The Environment

In principle, any TEM can now be modified to produce a con-

strated. Diffraction patterns can be collected with ~ 3-ps temporal trolled environment. We can either use in-situ holders from com-
resolution (Fig. 2.11). Since image/diffraction pattern formation mercial vendors (details at the end of the chapter) or modify the
requires around a billion cycles, the process being imaged must stage of the TEM to produce an environmental TEM (ETEM).
be perfectly reversible. With the introduction of microelectromechanical systems
(MEMS) fabrication technology, specimen holders that fit within
The variable time delay between pulses should provide enough the pole-piece gap of any TEM are available for in-situ studies.
time between pump-probe cycles for the structure to return to its
unperturbed state. Processes that have been studied using this
technique include phase transformations, structural dynamics, Start with a Little History and a Caution
electronic phase transitions (including nucleation and crystalliza- Remember when you are reading the literature that historically
tion), isolated chemical reactions, optoelectronic nanostructures, individual groups built their own holders or modified their own
photo-injection of electron–hole pairs (chemical doping), and microscopes. Commercial ETEMs that are dedicated to envi-
nanoscale mechanical vibrations. ronmental studies are available in National User Facilities and
are set up with gas inlet/outlet lines leading into the column to
In addition to ultrafast imaging and diffraction, this technique expose the specimen to gases or water vapor, so you don’t have
actually allows us to collect enough electrons over tens of sec- to build your own. Alternatively, in-situ holders allow you to
onds to produce an electron energy-loss spectrum (EELS) that flow gas and liquid into the TEM while controlling the higher
relates to changes in the electronic structure that occurred in the pressures, composition, mixing, temperature, flow rates, and
femtosecond time regime. Tomography is discussed in Chap. 12; transmission path lengths. Be sure to consider the gas pressure
imagine what stroboscopic imaging might do for time-resolved required for the experiment: a pressure of 0.1 Pa can convert to
tomography, where you would capture 3D events with different 1,000 monolayers per second of molecules to the surface of the
time delays between tomographic reconstructions. In conclusion, TEM sample. So, it may not be necessary to use a high pressure
single-electron imaging allows much more versatility for ultra- (e.g., atmospheric) to image the reaction mechanisms or struc-
fast imaging than the single-shot DTEM, and can be operated in tural transformations that occur in air. Figure 2.12 summarizes
STEM mode, used for tomography, and chemical analysis is still these ideas.
possible. Remember the word ‘reversible’.
Studies with in-situ holders used to be limited by the size of the
pole-piece gap, so the geometry of the sample holder was very
constrained. In medium-voltage TEMs (200–400 kV), the largest
 30 2 In situ and Operando

Fig. 2.12 The different pressure ranges that are of interest and
Pressures giving Maximum Resolution Commercial
which can be achieved in the TEM The Pressure Gap
Reactors
Chapter 1

‘High 0.3
Pressure’ mbar E-Cells
N2

Surface Science ETEM

10-6 10-4 10-2 1 102 104 mbar

10-4 10-2 1 102 104 106 Pa
1 102 104 106 monolayers of gas per s
Chapter 2

pole-piece gap is usually 7–9 mm. High-voltage TEMs (1 MV)

have an even larger pole-piece gap of 13–17 mm. Of course, the
thickness of bulky windowed cell holders or the requirements of
differential pumping apertures still mean that specimen tilt will
be limited when compared to standard single-tilt and double-tilt
holders (Chap. 8 W&C). If the TEM is a Cs-corrected instrument,
the pole-piece gap can be increased to ~ 10 mm for a 300-kV
instrument without degrading the resolution.

The Beam Can Change Everything

Always remember that the electron beam can change

your specimen and it even changes the environment (gas
or liquid) that you are using in the ETEM. Chemical
reactions or phase changes may take place at tempera-
tures that are much lower than you anticipate if they are
beam-assisted. 9
Fig. 2.13 Schematic of the pole-piece region within an ETEM for controlled
gas exposure to the sample with pumping above and below the specimen to
The rates of reactions in environmental studies can be very maintain high vacuum at the gun and camera chamber
fast. So you have to consider what electron dose rate you’ll use
and which signals you’ll collect ahead of time. In particular,
high-temperature studies such as nucleation and growth, co- ETEMs, the region close to the sample is at a higher pressure
alescence, Ostwald ripening, particle migration, reaction mech- while the rest of the column is still maintained at a low pres-
anisms, and kinetics all really need high-temporal-resolution sure using differential apertures. The design of these apertures
imaging. As the structure/phase/morphology is changing within is so that the rate of pumping is similar to the rate of leaking
your sample due to an external stimulus, such as heating, you through the differential apertures; the low pressure is thus
need to ensure that all the critical information is recorded for maintained in a dynamic fashion.
analysis. One of the worst mistakes you can make is not record-
ing all the data while doing the experiment. Some early environmental TEMs simply used a fine nozzle to in-
troduce gas molecules locally near the sample region. Then you
just pumped very hard! This entailed no other special modifica-
Change the Holder or the Stage tions to either the TEM or the holder. Of course, the problem was
There are two methods that we can use to introduce gas or liquid that if you pumped too fast you lost any hope of specimen stabil-
to the sample. ity, too slow and you compromised the vacuum in the column.

zz The first method is to use a TEM that has been modified using zz The second method is to use a modified holder, which is of
apertures that allow differential pumping: the vacuum around course the easier and less expensive method (only 100,000–
the specimen is kept at a pressure that is different from the 200,000 (2015) US dollars). (See Fig. 2.14.) (It’s the only
rest of the TEM; of course this is for gases (and water vapor), method if you don’t have an ETEM!) However, you need to
not liquids! (See Fig. 2.13.) This approach has long been suc- exercise great care, especially if you are a new graduate stu-
cessfully used for environmental SEMs (ESEMs). In these dent trying to do a new experiment! The first rule of the TEM
 2.5 The Environment 31

keep the path length of the gas-containing region to a minimum,

to minimize electron scattering by the gas or liquid, since that
could quickly become the main contribution to the image. You

Chapter 1
may also want to heat the specimen or apply other stimuli at the
same time (such as ion irradiation). The stability of the holder
materials at elevated temperatures and in different environments
places major constraints on the design and development of such
holders. Figure 2.15 displays the tip of a custom-built holder that
supports a closed cell for gas reactions with temperature control.

ETEM
The original ETEM was also known as the controlled-atmo-

Chapter 2
sphere TEM, and dates back to the early 1940s when Ernst Ruska
proposed the idea of incorporating gases into the objective lens
of the microscope; i.e., to have gas around the sample. This is
Fig. 2.14 E-cell holders commercially available from a Protochips: Poseidon elec- still the basis for the design of environmental TEMs used today.
trochemical cell; b Hummingbird Scientific: liquid flow holder; c Hitachi: gas stage The first design of the differentially pumped gas-reaction speci-
men chamber was introduced in the 1950s. To limit the exposure
of the electron source to the gases in the column, which would
is: don’t break it by being careless! Accidents can happen, but damage the gun, vacuum-pumped apertures were placed above
if you are careless and pump liquid into the column of your and below the objective lens. The gas flows into the center of
FEG-TEM, you may only be allowed to use XRD. the objective lens and around the specimen; any gas that leaks
from the apertures surrounding the pole-piece is removed from
Using this approach, the sample sits within two electron-trans- the column, Fig. 2.13. Holes were actually drilled through the
parent windows in the holder (the windows are typically silicon objective lens.
nitride, but they could be silica, amorphous silicon, or graphene).
This holder can be used with any conventional TEM, as the col- At first read, you might think ‘won’t drilling holes through the
umn is still maintained at a low pressure. The challenge is to pole-piece of the objective lens have significant impact on imag-

Fig. 2.15 Custom-built gas stage with a MEMS heater patterned on the membrane window
 32 2 In situ and Operando

Ideally, you’d like to have several ETEMs that operate at different

voltages, where each is aberration-corrected and monochromated.
Alternatively, you may look for an ETEM that can operate be-
Chapter 1

tween say 80 and 300 kV. It should then be able to achieve ~ 0.1-
nm resolution for structure or chemical analysis, and have an
energy spread of less than 200 meV. You need to be able to use
Fourier transforms of lattice fringe images to monitor the resolu-
tion and you should be able to monitor the stability of the system
at the various gas compositions and pressures that you’ll use.

The space you have for the specimen will depend on the micro-
scope and the voltage you’re using; remember that the pole-piece
gap may only be 5 mm or so, even with aberration correction.
Chapter 2

The column must be pumped in several stages so that you can

maintain a high vacuum at the gun. Most ETEMs will use a
combination of turbomolecular and ion-getter pumps. Typically,
if the pressure at the sample is 1,500 Pa, it will drop to 1 Pa after
the first pumping stage, then to 10−3 Pa after the second stage, and
will retain 10−6 Pa at the gun. The pressure at the gun is critical
for a FEG, which requires a high vacuum (10−9 torr), otherwise
the lifetime and performance will degrade.

zz A reminder: 1 Pa = 7.5 mtorr; 1 torr = 133 Pa; 1 atm = 760 torr;

Fig. 2.16 Surface structural characterization of a Pt catalyst particle when 1 bar = 105 Pa.
oxidized with exposure to O2, and reduced when CO or H2O is introduced at
significant pressures The size of the apertures close to the specimen region will de-
pend on the gas leak rate, the angular range you need for diffrac-
tion imaging, and the field of view you need to see the sample!
ing?’ The answer is yes, but to balance these gaps in the lens, sym- Larger apertures can be used in the second pumping stage, since
metrical holes are drilled on the opposite side of the pole-piece. the pressure is already lower compared to the first stage apertures
The sizes of the apertures are calculated based on the gas compo- (usually 100–200 µm in diameter). The third pumping stage is
sition, final pressures, and pumping speed. The highest pressures located before the viewing chamber and in the section between
achieved have been 50 torr. Using the differential pumped speci- the gun chamber and condenser aperture. Pumping at the view-
men region in the ETEM, liquids are not stable at these pressures ing chamber is important to prevent contamination build-up and
– the achievable pressure is still too low – so they can’t be imaged reacting on the scintillator. The column can be closed off from
using an ETEM unless we also use specially designed holders. the gas inlet/outlet valves and pumps by using pneumatic valves.
Alternatively, we can use water vapor to stabilize hydrated ma- The leak rate should be proportional to the aperture diameter
terials. The upper limit of the temperature and pressure in the squared and the pressure. Dependent on the composition of the
differentially pumped ETEM is 900 ˚C and 103 Pa, respectively. gas, pressures up to 2,000 Pa can be maintained at the specimen
within the ETEM.
An early application of ETEM was for observing catalyst par-
ticles during exposure to reactive gases. As you’ll see in later The ETEM typically has a lower maximum cutoff angle for
chapters, a Cs-corrected system has the advantage of minimizing high-angle scattering due to the aperture pumping design. So, mi-
image delocalization, which can be a problem when you need to croscopes that operate in both TEM and STEM modes will gen-
interpret images of crystal edges and facet structures, Fig. 2.16. erally not be able to use a scattering angle above ~ 50 mrad for
We’ll summarize some features of this type of instrument. There HAADF STEM imaging, but new ETEM designs for uncompro-
will soon be several different models available; we leave you to mised STEM imaging are demonstrated by Boyes and Gai (2014a
check the literature at the end of the chapter and maybe work out and b). For monochromated EELS, in the core-loss region, the de-
which TEM is which. creased beam intensity will degrade the signal-to-background ratio.

The system developed by Gai and Boyes at DuPont and devel- When you are imaging the specimen, everything must remain
oped further at Haldor Topsøe, used a two-stage differentially constant even if it’s changing! So you need a constant flow of gas
pumped environmental cell. This early cell allowed heating to minimize drift of the specimen while imaging. The composi-
in a controlled gaseous environment (T up to 900 °C; P up to tion and cleanliness of the gas at the sample is measured using
~ 20 mbar) while maintaining resolution at better than 0.16 nm. a residual gas analyzer (RGA). This is located close to the tip
It even used a GIF to record EELS and EFTEM images. of the sample holder. You should always consider the amount of

Licensed to ([email protected])
 2.5 The Environment 33

Fig. 2.17 Diffraction patterns of Pd nanocrystals in 4 bar of

H2 at two different temperatures, demonstrating hydrogen
gas influence on the lattice constants

Chapter 1
gas being produced in a reaction and the sensitivity of the RGA. barrier between the atmospheric conditions around the specimen
If you change the gas flow or composition, the response time of in the holder and the high vacuum in the microscope column.

Chapter 2
the RGA is on the order of seconds. When the gas composition The membrane material must be nondiffracting, or amorphous
into the column is fixed, you must control the gas flow to retain (so as to limit diffraction information contributing to the back-
stability around the sample. Drift and the achievable resolution ground in the image), nonporous, and not be degraded by the en-
are determined by the stability of the gas flow. If you are using vironment or electron beam. In this design, the specimen within
several gases, flow is controlled by a digital mass-flow controller a gas or liquid layer is placed between the sandwich of two thin-
on each gas line to provide high accuracy of gas injection into film membranes, with the internal region of the holder confined
the column. Modeling the gas flow through the system may help or accessible to an external (gas/liquid) supply through the base
identify the amount of back diffusion in your system as an effect of the holder, Fig. 2.14.
of temperature and geometry. Minimal drift with high stability
of the gas flow allows you to observe the same region of the Such windows were actually first used in the mid-1930s. They
specimen over a prolonged time. were initially thin Al films (~ 500 nm thick), then they became
carbon/polymer films (1940s), then carbon/SiO films (1960s);
Tracking the composition of the gases in the system can be a finally, the preferred thin-membrane material of today’s stan-
guide to the cleanliness of the specimen region, or it can provide dard is silicon nitride – SiN is SiNx, not necessarily Si3N4 (early
clues to potential artifacts that are occurring during imaging. 2000s, although single films of SiN were used in the 1980s).
Between experiments, you continue to pump everything and use SiN provides high mechanical durability, chemical stability, and
standard UHV techniques. Cleaning the column after in-situ gas low background scattering. The negative aspects of using SiN
experiments is routine. Bake the chamber (~ 70 ˚C for 24 h is membranes are possible charging in the electron beam, which
good) or plasma clean to remove carbon contamination. Water can cause drift of the specimen during imaging, and outward
contamination may be removed by baking the system. You want bowing of the membrane when they are used as pressure barriers.
to prevent contamination of the next sample; and always check Any bowing will depend on the strain within the thin membrane.
who is booked on the instrument before you!
The outward bulging illustrated schematically in Fig. 2.18 is
Future designs will achieve greater stability and larger pole-piece observed in both gas- and liquid-cell studies. The result is an
gaps for higher tilt angles so that tomography will be possible. increased gas/liquid path length in the center of the cell, which
Decreasing the voltage and the beam currents will reduce elec- can be a problem for liquids – they have a higher density. Some
tron irradiation artifacts. researchers have used the formation of a large bubble in the center
of the membrane cell to provide an ultrathin layer of water that
Figure 2.17 gives you an idea of what we can do when we are gives a decreased background signal while maintaining liquid
not limited to UHV conditions. Here Pd particles have been contact. In experiments to determine reaction mechanisms, this
imaged while in 4 bar of H2 gas (the ideal gas for TEM – it is technique will provide better images, but the mass transport and
very light!). The diffraction patterns are from the specimen at kinetics will be affected. When imaging through a continuous
240 °C and room temperature. When the temperature increased, liquid layer, it is best to localize imaging to the edges of the win-
the amount of H2 in the metal nanoparticles decreased, causing dow to keep the thickness of the thin liquid layer more uniform.
the lattice spacing to decrease (so the g-spacings became larger).
The metal then expands as the specimen is cooled. Remember SiN membrane windows are produced using optical lithography
that SAD is not great for measuring spacings, but it is good when techniques for patterning the window from a low-stress SiN thin
comparing two spacings. You can’t see the H2, but you can see film deposited on a Si substrate. The thickness of the Si substrate
its structural effect. will impact the signals into and out of the thin cell. This process-
ing is carried out in a cleanroom facility (< 1,000 particles/m3)
to minimize surface contamination. In general, the SiN mem-
Using an Environmental Holder branes on Si substrates are disposable (consumable) for each
The development of environmental holders required the produc- experiment. (The cost is $15–$60 per window: each experiment
tion of thin-film electron-transparent windows that provided a requires two windows.)
 34 2 In situ and Operando

possibly improved resolution, but the mechanical durability will

also decrease, which increases the likelihood for leaking gases
or liquids into the TEM column. How serious this is will depend
Chapter 1

on the flow through the cell (how much solution will leak) and
the properties of the solution (its volatility). This is the main
deterrent that has held back environmental imaging for 80 years
– the fear of permanently damaging the TEM. Today, if you’re
using a static cell or an aqueous solution with a slow flow rate,
and the window fractures, there should only be a small increase
in the column pressure. This is a manageable incident.

In practice, to assemble the reaction ‘chamber’: you will manu-

ally place the top and bottom windows into the specimen holder
Chapter 2

for commercial cells using o-ring seals, and then likely worry if
the alignment of the windows is not perfect. It doesn’t have to be.
Many researchers load rectangular membrane windows perpen-
dicularly to make sure that there is a known transparent area of
Fig. 2.18 Schematic diagram suggesting how the bulging SiN membranes in
closed cells are managed to image in thin-liquid layers either through a con-
overlap. The problem when trying to load square or rectangular
tinuous liquid layer at the edges of the window (a) or a thin layer created by windows identically is that your hand–eye coordination may not
liquid displacement from bubble formation (b). Optimal imaging is performed at be good enough to control the window overlap to < 5 µm. Any
locations identified by the red arrows variation can have a large impact on the distance between the
membranes, in the case where the membrane bows. Then if, like
others, you work close to the edges of the window so that the gas
zz Remember: in TEM imaging you get the highest resolution or fluid thickness is smallest, the minimum path length can easily
when the particle sits on the bottom membrane window – the vary by > 500 nm when you move 5 µm out from the membrane/
image depends on the exit-wave; in STEM imaging you get substrate edge interface! Therefore, this is a concern for liquid
the highest resolution when the particle sits on the top mem- cells and not very concerning for gas cells. When you are doing
brane window because probe-broadening occurs in the liquid. in-situ testing, first plan your experiment and then work to reduce
zz Beware: in a liquid, metal nanoparticles, such as Au, can inconsistencies; e.g., load the samples in the same way between
become attracted to the membrane surface which becomes different runs to maximize reproducibility. New and future cell
charged in the electron beam. This attraction can overcome designs will include self-alignment features to automatically
Brownian motion in the liquid. This effect will continue until align the membrane windows.
the electron beam changes the potential environment around
the particle, which in turn will destabilize the attractive force You would like the thickness of the cell to be defined by the
and allows the particle to repel the electron beam. The smaller distance between the top and bottom membranes. There will still
the particles are, the more they move, which makes it difficult be bowing of the thin-film-membrane windows, but in liquid
to image, e.g., 2 -nm Au particles without the beam inducing cells, this path length is generally below 1 µm. Actually, whole
movement. Soft materials and biomaterials are usually very biological cells have been imaged using spacers as tall as 7 µm.
beam sensitive, which is why they are often imaged using The spacer materials used are generally metal, ceramic, or poly-
cryo-EM. Some researchers have used ‘affinity surfaces’ or mer, depending on chemical compatibility and tolerance for
etched wells to deliberately localize (attract/attach) materials to compression. Many different spacer designs have been demon-
the SiN membrane surface to increase stability during imaging. strated including metal or ceramic pillars, photoresist beams, and
polystyrene or metal spheres dried on window chips that are not
The window parameters, including: window area, membrane patterned. Flow channels have been designed to control the gas/
thickness, membrane stress, membrane composition, and stack- liquid path during the experiment – the microfluidics approach.
ing orientation between the two windows, can all impact the
image quality and experimental design. The initial stresses in The resolution that you can obtain depends on the cell thickness,
the membrane are determined by the SiN growth process: the pressure, and composition. At this time, atomic-scale resolution
membranes are generally under tension after growth. The area of of metal nanoparticles has been demonstrated in liquids up to
the window will determine how much the thin membranes bow 300 nm in aqueous environments and within a gas cell 5–10-
outwards towards the high vacuum. Larger windows provide μm thick at 700 torr of N2 and 500 °C. So, it is best to quantify
a greater area for imaging the sample, which may be essential these parameters based on your own design for each new exper-
if the feature you’re studying is large (like a whole biological iment you do. Check the environmental conditions within the
cell). When the membranes are thinner, the bowing increases cell before and during TEM imaging: the beam can alter these
the tension in the membrane – how much depends on the me- conditions by forming bubbles in liquids, or the gas may not
chanical properties of the film. Thinner membranes will result be flowing at the pressure you set. Of course, you won’t form
in less background scattering, allowing for higher contrast and bubbles if there is no liquid present!
 2.5 The Environment 35

The thin-film membrane material will interact directly with the the thickness of the membranes you are using. As you know, in
system and the gas/liquid cell. Therefore, we have to be careful XEDS, the p/b ratio scales linearly with the increase in sample
to remove any contaminants from the membrane surface that thickness; EELS really needs thin specimens. Low-loss EELS

Chapter 1
could affect the environment or degrade the imaging conditions. is generally used in liquid-cell studies to estimate the specific
Contaminants could be chemicals or just Si dust particles left thickness of the fluid path; this thickness can vary greatly due
over from the window fabrication; such Si dust can determine to the bulging of the membranes. Depending on the geometry
the path length between the windows if it is larger than the de- of the window enclosures, these systems can be combined with
posited spacer. Rinse the chips in acetone, then methanol, then XEDS since the environment is completely contained within the
dry carefully with nitrogen; follow this with plasma cleaning sealed cell, which is not true for an ETEM. Tilting the specimen
for a couple minutes to remove any residual carbon. Use carbon holder to increase the signal on an XEDS detector may increase
fiber tweezers for handling to prevent fracturing of the silicon the effective sample thickness depending on the sample location
substrate. where you are collecting X-rays. Think of which point in the
sample will generate the most X-rays that will reach the detector

Chapter 2
In liquid-cell studies, we can change the membrane surface char- (see Zaluzec et al. 2014.)
acteristics from hydrophobic to hydrophilic for finite periods by
either heating to 150 °C or via oxygen plasma cleaning. We can The sealing mechanism for these environmental cells has cre-
treat the surface of the SiN membrane windows to immobilize ated limits for the tilting capabilities. Generally, the thin-film-
nano-/bio-materials on the membrane surface; this prevents mo- membrane windows are exposed by back-etching using KOH
tion of nanomaterials (long-range lateral diffusion, Brownian wet-etch methods from a silicon substrate (100–500 μm thick)
motion, beam-induced and short-ranged diffusion) during an along the (111) orientation at an angle of 54.7˚. These Si walls
experiment. New designs implement pillars and wells to con- will block the scattering signal at high tilt angles for imaging/
trol the thickness of the cell or to trap samples in a confined diffraction and collection on the XEDS detector. The thinner
region to limit diffusion. When using high-resolution imaging, 100-μm-thick Si substrates are thus better for providing higher
any small motion of the materials during image acquisition can tilt angles, but processing these wafers using photolithography
drastically reduce the structural information. Companies are pro- techniques can be challenging: the thin Si wafers are very fragile.
viding membrane coatings for MEMS-based gas/liquid cells to
define the solid–gas or solid–liquid interface, immobilize nano-/ Remember that contamination can build up on the cell windows
bio-materials, and purify biospecies. with prolonged imaging. Other difficulties arise during S/TEM
imaging due to thickness variations across the cell. And remem-
Windowed cells did not used to be compatible with heating sys- ber – if you have particles, you need to know if they are located
tems until we made them using MEMS technology; now we can on the top or bottom membrane and how they are interacting
make a ‘lab on a chip’. With MEMS devices, we can change the with one another.
heating rates with much better control and with response times
below one second. Unlike furnace-based heating units that can
take 0.5 h to stabilize at high temperature, these MEMS devices Measuring Chemistry in the ETEM
(illustrated in Fig. 2.15) can achieve high temperatures and be Of course, we want to use chemical analysis in the ETEM. What
stable within minutes. If a resistive heater is in direct contact reaction processes are occurring in the sample? Is the environ-
with a gas or liquid environment, unintended processes may take ment actually changing chemically, not just the specimen? The
place! It is best to use heating elements that are buried under a di- gas around the specimen and additional apertures for differential
electric layer or patterned on the opposite side of the membrane pumping in the column means that we can’t use XEDS analysis;
from the specimen. Infrared imaging can be used to calibrate the the X-ray background signal is increased too much. But, we can
heating components for specific atmospheric conditions, unless use EELS to follow the evolution at gas–solid interfaces spa-
you can use multiple electrodes for resistance thermometry. An tially and spectroscopically, and monitor the composition and
initial drift rate of 3.3 nm/s at 500 ˚C can readily be achieved with bonding information during gas–solid interactions. For example,
a MEMS gas/heating chip. Rapid heating causes a displacement this allows us to determine the initial and final gas compositions
that is proportional to the temperature increase (~ 0.40 nm/˚C). and the reaction products from catalyst particles in the ETEM.
Liquid cells are expected to achieve temperature stability up to A downstream RGA will only be able to measure bulk gas prod-
80–100 ˚C; maximum temperatures are limited by the stability of ucts from the reaction, in the best case.
the membrane window supporting the heater. How good we are
at measuring the temperature is always something that concerns
us (see Sect. 2.6.1). The Gases
For ETEM experiments, we must pay careful attention to the
Chemical analysis using EELS and X-ray energy dispersive interaction of exposed gases with the materials that will be used
spectroscopy (XEDS) can be possible using these windowed in the experiment, and that are present on the grid, holder, and
environmental cells, dependent on the design of the enclosure. within the TEM column. It is a common mistake to select a grid
You can make EELS measurements in both gaseous and liquid that will form an alloy with the material under investigation,
environments, although the quality of the spectra will depend on especially when changes in the gas atmosphere/pressure and
 36 2 In situ and Operando

temperature are occurring; it’s like heating Si with a Pt washer When heating specimens in the ETEM at high gas pressures,
in vacuum, but worse. Look up the phase diagrams for the ma- you’ll have to supply more power to reach a given tempera-
terials that are within the reactor (TEM column); watch out for ture than when you do the same experiment in vacuum, because
Chapter 1

reactions between CO with transition metals used in many hold- you’ll lose heat due to conduction. The required extra power will
ers (forming carbonyls). The support material may also be very depend on the thermal conductivity and the pressure of the gas
active in the reaction environment: test different grid materials within the cell. For gases with high thermal conductivity, like
such as carbon, boron nitride, or silicon nitride. You might test H2, you can mix the gas with inert gases that have lower thermal
an experimental design, exposing all components to the gas but conductivity. In any experiment, you should normally stabilize
without the sample present (a control study). For example, Cu the environment around the specimen before you heat it.
nanoparticles can form by diffusion from the grid at tempera-
tures much below what you expect. It is important to know and
understand the conditions of an in-situ microscopy experiment
beyond the microscope alignment. 2.5.1 Ultrahigh Vacuum
Chapter 2

The gases that are typically used in ETEMs include O2, H2, N2,
CO, CO2, and NH3. Be very careful about safety: some gases In a typical TEM, the specimen sits in a vacuum but not UHV.
such as acetylene, nitrous oxide, halogens, and disilane are really UHV-TEMs are special in that the specimen does indeed sit in
dangerous. If you are using these, you should make sure that ev- a UHV environment. So we can say that UHV is a special envi-
eryone working in the lab knows it. Better yet, go to a lab where ronment – sometimes thought of as the absence of environment,
they use them frequently. The materials of the holder or the grids but it’s really not. Most UHV-TEMs are in fact custom-designed
you’re using may not be appropriate for the environmental con- instruments operating with a UHV chamber (which might be at
ditions in a particular ETEM. If you’re using pure oxygen, you 5 × 10−9 torr) attached to the specimen loading region of the TEM
should only use Au or Pt grids; other metals will corrode. to transfer samples directly from the UHV chamber into the TEM
for analysis without a change in pressure or exposure to contam-
The length of a pressurized gas path is limited if you want to inants. A dedicated UHV-TEM will have an improved vacuum
keep a high vacuum in the other regions of the column. This design and a direct sample-transfer system so the specimen is
length is related to the beam–gas path length (BGPL), which not exposed to air. We can bring the gas within a few millimeters
is sometimes referred to as the GPL. It is the distance that the of the specimen using a capillary tube. This approach is very
beam travels in the gas from the specimen to the aperture where useful if you want to study highly reactive mechanisms that may
it exits the cell. This topic is very familiar to those working in be unsuitable for an ETEM or gas closed cell. Typical applica-
ESEM. tions would be to study surface processes or the growth of thin
films. This design may have a differentially pumped specimen
This BGPL affects the imaging conditions within the microscope chamber to prevent gas leakage into the rest of the microscope.
because the incident electrons have an inelastic-mean-free-path, We can, of course, use higher gas pressures in the column if
which will generally be less than the BGPL. A shorter BGPL we turn the electron beam off! Since it is UHV, you will have
(determined by the pole-piece gap) is preferred, even though a to be able to bake it out, so the vacuum pumps and seals must
smaller pole-piece gap will limit the specimen-tilt range. The in- be bakeable. Bake-out is limited for a TEM because the lenses
creased inelastic scattering caused by a long BGPL will degrade are water-cooled; the temperatures should never go above 80 ˚C.
resolution by increasing the effect of chromatic aberration. The Figure 2.19 gives you an idea of what can happen to your TEM
highest achievable gas pressures around the specimen will be when you take the idea of in situ and UHV seriously. The Nion
determined by the need to maintain a high vacuum environment STEM is the exception to all of the above: it is a UHV system
in the rest of the column. Beam coherence and intensity will just as the old VG STEMs were, although some may be fitted
decrease due to the primary electrons being scattered by the gas with a ‘leak valve’ to allow gas to be intentionally leaked into
molecules. These interactions may also introduce new features the region near the specimen.
between 0 and 25 eV in the measured EELS. Delocalization in
the image may look like monolayers on the surface of metal The airlock separating the TEM specimen chamber from the
nanoparticles – you can find such published images. Simulate the UHV chamber must be UHV too. The goniometer and speci-
images to check what you are seeing experimentally. men holder also have to be designed to allow the specimen to
be transferred through the UHV chamber, and then inserted into
Remember that the cleanliness of the gas lines can impact the the TEM column. Since these instruments tend not to be off-the-
reactions studied during the experiment. Take care to purge the shelf, you may still encounter a top-entry design; these stages
gas lines before and after an in-situ gas experiment to avoid gas used to be standard whenever you needed to minimize drift; drift
contamination within the ETEM. It’s a good plan to purge the rates of ~ 0.5 Å/s might be achieved on a good day.
system with the gas that will be used in the experiment for some
time to stabilize the environment within the column. If possible, The UHV instrument can incorporate in-situ cleaning and depo-
bake the gas lines after the experiment to further reduce gas sition capabilities as attachments. We can study chemical vapor
contamination. deposition (CVD) growth in a UHV-TEM by introducing oxy-
 2.5 The Environment 37

Chapter 1
Chapter 2
Fig. 2.19 The TEM as part of a large capability for studying thin films. The JEOL TEM is just visible; the specimen can be cleaned in UHV and a thin film deposited
without breaking the vacuum.

gen, germane, and disilane, as has been done for the growth of high-temperature annealing, and even ion milling. It is even pos-
Ge nanowires from an Si substrate. In this experiment, first the sible to add a solid-source system (hence MBE) into the TEM
sample is cleaned to provide an oxide-defect-free surface, and by using a port above the upper pole-piece. In Sect. 2.8.1, we’ll
then a thin film is deposited onto the clean surface. The deposi- discuss how a UHV system can be used to study the growth of
tion can be done in a UHV system connected to the TEM (if you thin films and nanowires.
have one), or in the TEM during imaging. Vapor–liquid–solid
(VLS) growth can be imaged, by reaction of the gas at the liquid
surface and then allowing diffusion to the catalyst–wire interface.
Other surface processes such as growth, etching, chemisorption, 2.5.2 Working in a Gas Cell
surface roughening, heterogeneous nucleation and growth at
variable temperatures, and oxidation processes (native, thermal,
and electron-beam-induced) can also be observed using such an The advantage of using a thin-membrane gas cell rather than the
instrument. ETEM is that we can then achieve pressures up to 5 atm within
the microscope and drastically reduce the effective BGPL. The
Different UHV-TEMs usually were designed with different ca- ETEM is limited to maximum pressure of ~ 0.1 atm with a con-
pabilities or different vacuum-enclosed attachments including trolled BGPL. In the past, the main deterrent for using an envi-
ion-beam deposition/irradiation, MBE deposition chamber, Au- ronmental gas-cell holder was that the thin membrane and the gas
ger electron spectroscopy, X-ray photoemission spectroscopy in the enclosure degraded the images. Window rupture is still a
(XPS), SEM/FIB, sputter chambers, electron-beam evaporator, concern, especially when using thinner membranes to maximize
 38 2 In situ and Operando

resolution. New window designs can use thin SiN membranes

that are only 5 nm thick and atomic-scale resolution images have
been obtained with BGPLs as large as 20 µm. The spacing be-
Chapter 1

tween the thin-film membranes can be much larger than in liquid

cells, since the inelastic mean free path for electron scattering is
much longer in the lower density gas phase than in the liquid.

These gas-cell holders are designed with inlet/outlet gas lines

connected to mass flow controllers to control the composition
and pressure of the gases flowing between the thin-membrane
windows. The gas supply/return lines are 360-μm-OD stainless
steel capillary tubing; they are generally chemically inert. The
cleanliness of the gas lines in the windowed holder is a primary
Chapter 2

concern for these experiments. The gas lines should be pumped

down (~ 10 torr) then purged with the gas that you will use in
the experiment. Always plasma clean the windows and the cell
components prior to use.

In addition to gas control, temperature may also be controlled

using MEMS-patterned resistive heaters. The heating devices
are calibrated for vacuum by the manufacturer at 50 and 750 torr
in H2/Ar atmosphere using infrared imaging. The current that
we need to heat a specimen to 700 ˚C in vacuum, rather than in
300 torr of N2, can decrease from 5 mA to 4.5 mA because there
is no convection or conduction in a vacuum. Temperatures of
up to 800 ˚C in 1 atm gas, ramping rates of 106 °C/s with sample Fig. 2.20 Commercial in-situ holder for elevated-temperature gas environ-
stability have been achieved after 1 min. This device has used ments with a laser beam traveling along the rod and deflected onto the speci-
built-in sensors to calibrate the temperature during the experi- men with a mirror
ment. Remember that exothermic reactions may locally increase
the temperature even in a TEM. You can use microcalorimetry to
measure changes in temperature relative to the applied current lamellae on a grid) is possible using a specimen holder with a
through a resistive heater with a sensitivity of 0.5 ˚C and make very large BGPL of up to 250 μm. In this design, the BGPL is
corrections within milliseconds. Heating these windowed cells controlled using 15 μm spacer titanium disks that are positioned
increases the likelihood that the membrane will fail. between the membrane windows In addition, this design offers
site-specific heating using a translatable fiber-based 1,085-nm
In the MEMS heater design, the window area lies inside the Pt laser mounted on the specimen holder (total power output 8.5 W).
heating unit that is embedded below the SiN membrane; open- The laser is focused onto the specimen, adjustable from 30- to
ings of ~ 10 μm diameter form the window viewing regions. The 300-μm diameter spot, using a 60˚ gold-plated titanium mirror.
small size of the heating unit located just outside the viewing The temperature is expected to be able to crystallize amorphous
area has many benefits: it produces a much smaller thermal ex- SiN membranes (> 2,000 ˚C) – which you should never do! All
pansion and there is little thermal conduction from the surfaces of these reside within a 2-mm-thick frame that sits within the
during heating. External electrical connection is made by tung- pole-piece gap. This design, shown in Fig. 2.20, demonstrated an
sten probes that contact the heating unit and thermal sensor. The image resolution within the 35-μm-thick, 1.2 bar of H2 gas envi-
temperature of this system is measured using local electrical ronment that was close to the resolution limit of the microscope
resistance that is measured from four points on the chip, giving used: 0.18 nm was obtained (0.14 nm would have been the limit).
an estimated value within 10 % of the actual temperature. (Your Temperature calibration for the laser heating will be critical.
question: “Is that good enough?”)
Some other thoughts and odd facts:
A basic gas-cell design has been created by JEOL for high-tempera-
ture heating within a windowed enclosure. The cell uses Cu disks zz Water vapor can be introduced by bubbling a gas through a
with small holes sealed using thin carbon films. The 1-mm gap heated water reservoir.
encases a W–Re heating filament, where samples can be deposited zz Remember that flowing gas will conduct heat away from the
directly onto the filament. (See the temperature control Sect. 2.6.2 heating device in the TEM.
for more information on these types of heating elements.) zz The chips are mounted on leak-tight tubing for the inlet/outlet
gases.
Increased temperature range and loading of conventional TEM
specimens (e.g., dimpled 3 mm disks, mesh grids, and FIB
 2.5 The Environment 39

Chapter 1
Chapter 2
Fig. 2.21 Liquid-cell TEM imaging of the Kirkendall effect in the formation of hollow bismuth oxide nanoparticles. Observed at a room temperature, and b 192 ˚C

zz With thin (< 20 nm) SiN membrane windows, we need to 2.5.3 Working in a Liquid Cell
increase the pressure very gradually so as to minimize the
mechanical shock that could cause failure of the membranes.
zz The amount of gas actually passing through the membrane Imaging materials in a liquid has long been the dream for struc-
region has been reported to only account for ~ 1 % of the total tural biology where macromolecules including proteins, viruses,
volume; keep this in mind when modeling reactions within bacteria, vesicles, microtubules, and DNA always exist in the
the cell. presence of water. Of course, there are many other material sci-
ence reactions that take place in a liquid environment. You can
It is difficult to measure the gas composition accurately and flow see the ‘potential’ for liquid-state studies in the TEM: nucleation
rate within the cell at any given time. Portable gas mixing units and growth of nanoparticles or nanowires, templated nanopar-
are generally included with the purchase of a gas-cell holder, ticle growth, nanoparticle self-assembly, nanoparticle assembly
though you’ll have to calibrate it and check for consistency. The during drying, corrosion, and soft/biological material imaging,
gas flow is controlled using mass flow controllers and monitored galvanic replacement, Ostwald ripening, and the Kirkendall ef-
with pressure regulators. The gases are mixed before flowing fect (Fig. 2.21). We are now able to investigate the mechanisms
them into the TEM holder. Commercial systems generally use and kinetics controlling these phenomena by directly imaging
stainless-steel gas tubes, although nylon gas tubes can be better the reaction within a closed liquid cell.
for chemical and mechanical stability as the gases flow to and
from the holder. Residual gas analyzers or mass spectrometers MEMS-based designs allow us to have a continuous liquid layer
have been used to detect the outlet gases from the closed-cell in the TEM confined within a hermetically sealed closed cell.
reactor. Using these down-line analysis techniques involves a Now we can image fully hydrated structures without freezing the
time delay between setting the gas composition for the cell and sample to immobilize it. Since the samples are not frozen solid,
measuring it with a mass spectrometer; a 5-s delay has been ob- we can study the dynamics of bio- and nanomaterials in real liq-
served, but space–time broadening can add another 20 s. uids, even highly volatile ones. We can study liquid–solid reac-
tions, how nanoparticles move in a liquid, how particles nucleate
and grow in a liquid, and even the dynamics of liquid–liquid
mixing reactions. A static MEMS-based liquid cell is illustrated
in Fig. 2.22a; this one was constructed at a microfabrication fa-
cility, though generally a graduate student in a cleanroom can
 40 2 In situ and Operando

just float the graphene sheets off from the substrate where it was
deposited, Fig. 2.22b. Graphene static cells are used to explore
the structure of materials in a confined liquid environment, for
Chapter 1

chemical analysis, or to explore beam-induced processes such as

metal-ion reduction to nucleate and grow metal nanostructures.

Microfluidic pumping holders can control the flow of liquid

through channels and can control how liquids mix to stimulate
chemical reactions. Generally microfluidic tubing, such as PEEK
tubing, is super-glued in place; o-rings seal the top and bottom
membranes to maintain the liquid enclosure. These holders pro-
vide relatively fast sample loading and can be thoroughly cleaned
after each use to prevent contamination. We can use microfluidic
Chapter 2

pumping during an experiment to reduce the amount of beam

damage to the solvent for high-exposure imaging, or to allow us
to image the same region for longer. Kinetics will be different for
static cells compared to the flow cell, where the reactants are not
being continuously replenished. Note: when liquid flows through
the cell, small particles may move or become unstable.

Chemistry at this very small scale will be influenced by any

changes in composition or reaction with surface species on the
cell components. So, salts or changes in pH will determine under
what conditions materials will react. For example, Au nanoparti-
cles in aqueous media have been observed to dissolve at high pH
values, and increased beam-induced motion has been observed
at very low pH values. Note: pH can change during a reaction,
especially locally! To reduce bubble formation, take consider-
ation of the amount of gas in the solution (sonicating the liquid
for several minutes will reduce the gas concentration). Of course,
you must always consider chemical compatibility, as it is pos-
sible for the liquid-cell windows or holder to irreversibly react;
Fig. 2.22 a CINT’s electrochemical liquid-cell design for quantitative electro- e.g., Si substrates are etched in basic environments. These effects
chemistry of 10 individual electrodes within a 150-nm fluid gap with picoampere also can be triggered by the electron beam itself, depending on
current control. b Home-built graphene liquid cells, demonstrating pockets of the chemistry (see Sect. 2.8.2).
liquid being hermetically trapped for high-resolution imaging. c Annular bright-
field STEM image of a liquid pocket outlined in black dashed lines. d Atomic Electrochemical TEM cells are generally designed based on the
resolution STEM imaging of hydrated biomolecules same principles and components as the static and microfluidic
cells with the difference being that the electrodes are integrated
fabricate a similar design. Liquid-cell designs are available for into the MEMS windows. Variations on this approach have been
static cells, microfluidic pumping, and electrochemical control. successful with commercial cells using 3–4 electrodes of de-
fined electrode materials (Au, Pt, Cu, C), and areas (1–100 μm2).
Static cells are generally composed of two membrane windows You can customize the basic designs to change the electrode
with a fixed spacer height and are epoxy-sealed. Such cells can areas and materials (anything that can be e-beam evaporated,
completely enclose the immersed material and then be loaded deposited using ALD (atomic layer deposition), or sputtered)
into a standard single- or double-tilting TEM holder. The dis- and separate it from the solution/electrolyte with a passivating
advantage of such a cell is that the liquid is usually in contact layer or bury it below the SiN membrane layer on the MEMS
with the epoxy; this may contaminate the liquid if the epoxy is device. The active region of the working electrode(s) is gener-
not completely inert during, or after, curing. (Test it! Also, take ally contained within the window region, so that we can image
note of outgassing and what this will do to your clean surfaces.) the entire electrochemical reaction, any electrochemical data
Once this type of cell has been sealed, there is no way to refill that can be acquired simultaneously with imaging can be quan-
or replace the sample solution if a bubble forms between the titatively correlated with the images. Depending on the size of
membrane windows. In a modification of this design, graphene your electrode, you may or may not detect the variations in local
has been used to avoid the ‘bulky’ support window. You can, in current density where you were imaging. Mass transport through
principle, construct graphene ‘pouches’ by encasing the sample the cell is a concern, since the thin liquid layer can constrain
(in small pockets of stabilized liquid) between two graphene the local environment around the electrode. The design of these
sheets and sealing the edges on a TEM grid: for aqueous solvents MEMS devices can be quite complex! You need to connect the
 2.6 The Temperature 41

stream of liquid. (Picture water streaming from a hosepipe and

remember the Rayleigh instability.) The thickness emerging
from the liquid jets have been produced below 800 nm while

Chapter 1
keeping the pressure below 1 torr in the specimen chamber of
the ETEM. The difficulty with this design is the control over
the liquid layer thickness; liquid droplets are controlled by the
geometry and pressure of a gas nozzle surrounding the liquid
nozzle. This technique will be limited by the type of liquids that
your lab director is willing to have dispersed in their ETEM!
Also, there currently is no commercial holder on the market
with this capability.

Enclosing a liquid layer between two thin membranes is the most

Chapter 2
stable way to image liquids, but other methods can have their
advantage. Some methods may not be available to everyone! So
we mention this to encourage you to be creative.

Fig. 2.23 Electrochemical growth of Pb dendrites in an aqueous solution con-

tained in a closed-cell during STEM imaging
2.6 The Temperature
electrode on the window (1–100 μm2) to a current or voltage
control unit so you can apply and measure the electrochemical 2.6.1 Temperature Measurement
profiles (electrochemical impedance spectroscopy, potentio-
static hold, galvanostatic hold, cyclic voltammetry, etc.). The
electrode sample sizes in the TEM are small, so the currents may We heat the specimen to temperature T. How well do we know
be at the femtoamp level. Larger electrodes at microamp-level T? Companies that produce commercial heating holders should
currents may cause an unwanted resistive voltage drop through provide a calibration for you; but when was that done, under
the electrolyte. We’ll have more details in Sect. 2.7.3. Factors what conditions, and on what specimen? Almost certainly it
that you should quantify during an experiment include the water was carried out under vacuum, but how good was the vacuum?
content (for aprotic solvents), electrolyte thickness, electrode You want an accurate measurement of temperature, taking into
surface contamination/oxidation, and electron-beam effects for account the thermal conductivity of the sample material and ra-
the given electrochemical environment. We can operate in a diative heat losses. (See Banhart (2008) for more details.) The
range where the electron beam causes deposition and/or dis- temperature of filament-wire and MEMS-based heating holders
solution of many species to compete during electrodeposition are calibrated using optical pyrometry. Spatially resolved tem-
of intended materials, and the electron beam may even apply perature maps of a heated MEMS chips in STEM have been
current to the working electrode during imaging! produced using plasmon shifts in EELS and Raman spectroscopy
(see Sect. 2.7.4, Fig. 2.51), though neither of these techniques is
In electrochemical TEM studies, experimentalists may be inter- commonly used for most heating experiments, but they will be.
ested in batteries, supercapacitors, fuel cells, electrodeposition, Mapping shifts in the plasmon energy in EELS may be extended
corrosion, or electrocatalysis. The possibilities are clearly illus- in the future for measuring temperature variations in gases and
trated by Fig. 2.23, which shows Pb dendrites growing in an aque- liquids in situ. See the paper by Mecklenburg et al. (2015) for a
ous nitrate solution while a voltage is applied. More information foretaste of what is to come.
on electrochemical studies in the TEM is located in Sect. 2.7.3.
Most heating experiments in the literature take place within the
In all liquid-state studies the most important concerns are the high vacuum of the TEM column; so optical pyrometry is very
presence and quantity of liquid within the closed cell and the fitting, though when the experiment is taking place under atmo-
effect of the electron beam on the solution. You’ll find that every spheric conditions or with exposure to gas mixtures, the calibra-
in-situ experiment faces the same question: “What is the effect tion is not standard or simple. In ETEM heating experiments, the
of the electron beam?” So be prepared. temperature may continuously decrease as you increase the gas
pressure around the MEMS heater. A simulation of the tempera-
ture profile within the ETEM is shown in Fig. 2.24.
Hydrated State – Liquid Jets
We can study liquid–solid interactions using a liquid injection A demonstration of nanothermometers has shown a linear re-
holder (Deponte et al. 2011). This method was developed for sponse to increases in temperature from 50 ˚C to over 500 ˚C
X-ray crystallography within a synchrotron, has been success- (Fig. 2.25), so this is a promising method to measure the tem-
fully used in the TEM. The idea is to use an ultralow-volume perature at the viewing region accurately. Unfortunately, making
 42 2 In situ and Operando

Fig. 2.24 Temperature profiles within the specimen region

of an ETEM in a y–z plane and b x–y plane as contour plots.
The H2 gas inlet flow of 10 NmL/min, furnace heating holder
Chapter 1

temperature set to 500 ˚C. c H2 gas pressure drop from inlet

to outlet and the maximum temperature change over the grid
relative to the furnace temperature as a function of inlet flow
Chapter 2

Fig. 2.25 a The “nanothermometer”

with many applications in principle.
The tube is a silica nanotube and the
liquid inside the nanotube is indium.
b Indium expansion as a function of
temperature: heating (circles) and
cooling (triangles)

the nanothermometer may be more difficult than doing the ex- 2.6.2 Heating
periment. Designs have been demonstrated using liquid gallium
(29.78–2,403 ˚C) in carbon nanotubes or MgO nanotubes for
measurements at higher temperature. The assumption is that the Heating the sample inside the TEM is a well-established proce-
thermal expansion coefficient of liquid gallium in the nanotube dure; details of the heating holder have already been described
is the same as for bulk Ga. The difficulty with this method for in W&C. Here we are concentrating more on what the chal-
measuring the temperature during an experiment is not just how lenges and possibilities are. If you’re heating the specimen, you
do you place the thermometer so that it doesn’t interfere with the need to know the temperature; sometimes you want to know it
experiment, but are these nanothermometers readily available? accurately. One prevailing question is: how accurately can you
know T?

Under high-resolution-imaging conditions, you can watch atoms

move around on the specimen. Of course, you don’t see this if
you are not at high magnification – but you can be sure it still
happens! The electron beam can certainly increase the tempera-

Licensed to ([email protected])
 2.6 The Temperature 43

ture of the specimen, but by how much? Prolonged exposure

may increase the movement (Sinclair et al. 1981). Generally,
beam heating is not what we want: we want controlled heating

Chapter 1
and quantitative analysis of, e.g., heat-induced phase transforma-
tions, sublimation, melting, sintering, diffusion, recrystallization,
and annealing. We can study phase transformations at either low
or high temperature using specialized stages that control the tem-
perature of the specimen within the microscope.

Several types of heating holders have been developed to allow

this. The main challenge is always the same: we want to know
the temperature and we don’t want the specimen to move. We Fig. 2.26 Dewetting of Ni on sapphire after in-situ reduction of NiO to Ni in a
want to eliminate, or at least limit, thermal drift. Most heating standard TEM vacuum at ~ 827 °C; note the changes at X and Y, which occurred

Chapter 2
holders use resistive-heating – like an electric fire – and tradi- between frames
tionally about that sophisticated. Now the heating elements may
be MEMS, but the principle is the same.
cling. Water cooling is used to minimize the heating of other
Specimen drift during heating can be controlled by several fac- parts of the holder and to thus reduce specimen drift. With these
tors: holders, it takes quite some time for the temperature to stabilize
and the maximum temperature is limited to 1,300 ˚C for sin-
zz If you change the temperature, then thermal expansion or gle-tilt holders and 1,000 °C for double-tilt.
contraction will occur for everything that changes tem-
perature. The specimen changes dimensions, but so does The melting point of the grid material may be a concern in these
the specimen holder. So we want to change the tempera- studies; grid atoms become mobile at temperatures half that of
ture of the smallest amount of material possible. When their melting point. Copper melts around ~ 1,300 K, so beware.
we use a 3-mm disk and a conventional ‘furnace’, it all Carbon thin films will tear at higher temperatures – another effect
changes dimension: 3 mm is 3 × 106 nm, and Cu, say has of thermal expansion. If you are going to use very high tempera-
an expansion coefficient of 16.7 μm/m/˚C. Should you ture, think first and plan.
clamp the specimen or not? If you use a MEMS device,
you should only be heating a much smaller volume: much
better for stability. Filament Wire Heating Stages
zz If you’re using a gas cell, then gas–solid interactions will If your sample is a powder, you can achieve a very high tempera-
contribute. So heating in the ETEM is challenging since the ture using a thin tungsten wire filament: the sample is attached
heat transfer to the gas and the temperature of the specimen directly on the heating element. Saka et al. 2008 pioneered this
depend on the gas pressure and composition. Small pressure approach using a battery-operated holder. The holder could stay
and convection variations can result in large temperature vari- at 1,500 ˚C for almost 10 hours, and could reach > 1,700 ˚C. The
ations. filament of this holder can be quite small, which gives more
zz Light is emitted when you heat specimens above ~ 700 ˚C, so stability and faster heating than a furnace heater. The thickness
we can’t use optical measurements below this temperature. of the tip of this holder in the pole-piece gap is ~ 2 mm. A gas
zz Consider what might happen if you heat a valuable Si speci- injector can be added to such a holder to allow environmental
men in a heating holder that uses a Pt wire for the heating. To heating in a conventional TEM. Caution: the pressure in the col-
answer this question, you should check the equilibrium phase umn will slowly rise, Fig. 2.27. In Saka’s design, the gas injector
diagram. The answer will remind you that you should know of 0.5-mm-ID stainless steel pipe was positioned 1 mm from the
what materials are used in the holder. heating wire. You may have seen this approach used before – it
zz Furnace elements might outgas. In a TEM, this can be very replicates the tungsten filament in an incandescent light bulb.
unfortunate!
A similar holder design, reaching very high temperatures (to
Remember, the environment under which the sample is heated ~ 2,000 ˚C), has been developed for studies on the formation
plays a critical role in controlling the microstructural changes at of ceramics and carbon nanotubes in the ETEM, Fig. 2.28. The
elevated temperatures. For instance, if we heat NiO in the TEM holder is composed of stainless steel and brass as the base, with
in a conventional vacuum to observe the reaction with Al2O3, it a tungsten filament. The temperatures were calibrated as current
transforms to metallic Ni before it can react, and then the metal (mA) vs temperature (˚C) using an optical pyrometer in a test
dewets the sapphire as shown in Fig. 2.26. rig under different atmospheres. This holder design was demon-
strated by converting cellulose to graphite by removing H2O, etc.,
at high temperatures. The samples are attached directly to the
Crucible Heating Stages filament, which may limit the samples and geometries that may
Many heating holders are composed of tantalum, due to its high be studied. Heating in gases can cause damage to nearby parts
melting temperature and structural stability during thermal cy- in the TEM, including the apertures and the pole-piece, so it’s
 44 2 In situ and Operando
Chapter 1
Chapter 2

Fig. 2.27 In-situ heating on a wire filament. a–e The original approach of Saka et al. f Example of melting Al metal on a sapphire ‘substrate’ using such a holder

Fig. 2.28 a A home-built resistive heating stage, composed of two electrodes with a wire filament conductively connected. b Current (mA) vs temperature (˚C) plot
of the holder calibrated using an optical pyrometer; red and black lines denote measurements from two different filaments

always best to make sure no damage will occur before starting a zz They minimize the distance between the heating filament and
new experiment. the sample.
zz They only heat a small volume.
One lesson to take away: if you have a workshop, you can make zz The platforms are customizable using lithographic processing.
an original holder, Fig. 2.28.
The sample is deposited directly on thin-film windows, or next
to holes in the membrane, with the heating filament at the edge
MEMS Heating Stages of the window. Such a holder can reach temperatures above
MEMS heating holders do several things right: 1,000 ˚C within milliseconds (a rate of 106 ˚C/s [Note: the ac-
curacy of measuring the time here relies on knowing the frame
rate of high-speed camera that was used to record the images]).
 2.6 The Temperature 45

Chapter 1
Chapter 2
Fig. 2.29 a Protochips MEMS heating stage composed of metal contacts patterned onto different conductivity ceramics supported by a silicon substrate with an
insulating barrier. b This stage was used for a comparison between theory and experimental results for subliming Ag nanoparticles using arrays of initially identical
particles

A MEMS-based heating stage can give atomic-resolution imag-

ing at temperatures > 800 ˚C.

Most MEMS heating holders use a disposable MEMS heating

chip: the device acts as both the sample support and the heating
element. The window region is a conducting ceramic film: typ-
ical window dimensions would be 150 nm × 500 µm × 500 µm,
supported on a ~ 4 × 6-mm Si chip. In the Protochips version,
2-µm-diameter holes are patterned onto the ceramic membrane
and carbon (~ 20 nm), SiN (~ 50 nm), or empty space covers the
holes (Fig. 2.29). You place your specimen inside a ‘hole’, or
at the edge. Each chip is calibrated by the supplier (using an
optical pyrometer) and provided with a chip-specific calibration
Fig. 2.30 Commercial heating holder in Fig. 2.29 was used to obtain HAADF-
file. Then you use this current-vs-temperature curve to determine
STEM images of ultrasmall FePt nanoparticles during structural transition
or set the temperature. Just remember that the temperature is
a to a partially ordered structure after 24 min heating at 400 ˚C, 1 min at 500 ˚C
calibrated under vacuum, and you have to consider the spatial
and 55 min at 400 ˚C, and b to an ordered structure
resolution of the optical pyrometer.

There will always be an initial thermal drift when the tempera-

ture changes. For the MEMS holders, when the temperature One study of nanoparticles heated in a MEMS holder found a
only changes over a small region, high-resolution imaging is 2 ˚C temperature difference between the expected temperature
obtainable, Fig. 2.30. At 800 ˚C, the sample drift rate has been and that from the calibration curve; theoretical modeling of the
measured at < 1 nm/s with temperature stability within 0.03 ˚C for nanoparticle response suggested a slightly lower temperature,
a DENSsolutions holder. You can buy a double-tilt holder giving but the resultant temperature was found to be much higher than
± 30˚ in two directions. the ‘driven’ temperature because of beam heating. See Fig. 2.29.

The main drawback for MEMS-based heating holders is that

you must be able to place your sample in a small window. These Environment and Heating Holders
holders are great for thin-film studies, where the film or nano- In a conventional TEM with a heating holder, the sample is
materials can be directly deposited from solution or sputtered heated under very low pressures which may not correspond
onto the windows and analyzed in vacuum. FIB cross-sectioned to the actual environment the material experiences in service.
samples can be placed on the window membrane, especially if Oxidizing or reducing conditions can change the nature and
you have a Giannuzzi system (EXpressLO) for controlled mi- concentration of charged defects in ionic materials and can thus
cromechanical placement of the sample in air. influence the thermal stability and/or properties. When we study
supported metal nanoparticles (like those used in catalysis) the
 46 2 In situ and Operando
Chapter 1
Chapter 2

Fig. 2.31 Refaceting of a Pt nanoparticle during oscillatory CO oxidation held at 727 K in an ETEM

temperature will allow them to change their morphology and to a vitreous instead of crystalline state. Automated systems are
coarsen (or conversely disappear). This can also happen when a commercially available that provide temperature and humidity
gas is present and would affect their catalytic activity. TEM has control with controlled wicking. Or you can build your own (see
shown the differences in shapes of supported metal nanoparticles Bellare et al. 1988)!
under different gas environments (Fig. 2.31). Elevated tempera-
ture control in a liquid cell has not yet been standardized, though Radiolysis
companies are offering these products.

Radiolysis is the decomposition of a material at the mo-

lecular level. It is caused by ionizing radiation, in our
2.6.3 Cooling case the electron beam, due to Coulomb interactions;
don’t confuse it with displacement damage. 9

Cryo Stages
A little more history first: encasing the sample within a thin layer
of vitreous ice was developed to study beam-sensitive biological
materials in a hydrated state by Taylor and Glaeser (1974). Even
then, low-dose imaging (< 10 e-/Å2) could give images with res-
olution better than 0.5 nm, although the contrast was very low.
They then used Klug’s approach to improve the s/n ratio, aver-
aging hundreds to 1,000+ images to give a clear structure (van
Heel et al. 2000; Renault et al. 2006). Since their specimens were
biological, they would enhance the contrast by defocussing the
image (see, e.g., Adrian et al. 1998). If the electron dose is too
high, hydrogen gas bubbles form in the vitreous ice from the
build-up of radiolysis damage.

Today, these samples are prepared by wetting a droplet of the

sample solution onto a holey carbon grid attached to a plunge Fig. 2.32 Three different imaging methods for viewing nanolipoprotein disks.
freezer, wicking away the excess solution from the grid to form a Conventional TEM image of a negative-stained dry sample. b Cryogenic TEM
a thin liquid layer, then ‘plunging’ the grid into liquid ethane image of a negative-stained hydrated sample. c Liquid-cell HAADF-STEM image
(kept cool with liquid N2). The thin water layer rapidly freezes of an unstained sample. Scale bars represent 20 nm
 2.6 The Temperature 47

Fig. 2.33 Vitrified poly(3-hexylthio-

phene) nanowires in organic solvents
were imaged using cryo-TEM with

Chapter 1
added Au nanoparticles as markers
for electron tomography. Sharp and
diffuse interfaces are observed within
the same frame for these structures

Chapter 2
Low-dose imaging is performed in three steps, namely ‘search’, fluid) He that is itself cooled by liquid N2. The temperature at the
‘focus’ and ‘photo’, as summarized by Chou et al. (2007). The specimen holder may reach 4 K, although it will be higher with
idea is to set up the low-dose image aligning the photo mode the electron beam on the specimen.
(where the image is collected) to the focus mode (a nearby re-
gion to photo mode alignment, to correct the focus in the final
image). Use the search step to locate the region of interest at low Cooling Holders
magnification to minimize the total electron dose on the region Most low-temperature stages will cool the specimen to liquid
of interest. You want to record focused images from regions that N2 temperatures (~ 77 K); liquid-He holders are also available,
have never (or hardly) seen the beam! but much more expensive. A heating element is included in the
design so that you can reach all the values between the minimum
As noted earlier, cryo-TEM imaging is expected to preserve the and room temperature. These stages require a source of the liquid
structure of the specimen since it remains in a hydrated state; the N2 or He, which is usually held in an attached dewar. Once the
price we pay is the images have low contrast. If you need to im- stage is loaded with cooling liquid in the dewar, the temperature
age biological specimens and cannot accept the low contrast, you is controlled via an external control unit. The performance of the
can use negative staining: heavy-metal atoms are diffused into cooling stage depends on the condition of the zeolite material
the material to produce an outline of the structure. (See Fig. 2.32 within the dewar, which must be regenerated frequently. The
for comparison to HAADF-STEM liquid-cell images. Then you cryopump allows for stabilization of the liquid N2 in the dewar to
should go to your biomaterials TEM books.) Alternatively, if you minimize the boil-off rate, which also decreases the mechanical
are having a hard time focusing or keeping track of beam sen- vibrations at the stage tip.
sitive samples, you can add metal nanoparticles to the solution
prior to plunge freezing, though a high nanoparticle concentra- The dewar at the back of the sample holder is connected to the
tion may get in the way during imaging, Fig. 2.33. sample at the tip by a thermal conductive rod that is contained
within the interior of the holder. These holders come in single and
Why Cool It? double-tilt side-entry designs. New holders generally come with a
loading unit to be able to transfer plunge-frozen cryo-EM speci-
mens without exposing the grid to air. A sheath on the outside of
The fundamental idea of cooling the specimen is to stop the tip is used to prevent crystalline ice build-up on the sample
thermally activated processes from occurring or changing during quick transfer through the air into the TEM. Once the
the state of the specimen. These processes include surface holder is stable in the TEM, it will act like another cold-finger!
diffusion of contamination. 9

There is one more step in cryo-TEM. We can reduce the effects

of radiation damage on beam-sensitive materials by using an
ultralow-temperature stage. Such a stage will use liquid (super-
 48 2 In situ and Operando

rod reaches a minimal value. A transparent lid covers the loading

station about the tip to prevent ice build-up. Secure the specimen
in place using a hexring tool: the sample and hexring must both
Chapter 1

be kept cold so that you don’t heat the grid. Cover the specimen
tip with the sheath on the holder to stabilize it as you transfer it
into the TEM. Caution: when loading the holder you may well
empty the dewar of liquid N2 onto the floor or worse onto the
microscope desktop!

A special holder design offered by FEI, then allows you to detach

Fig. 2.34 Custom-built design of a static gas- or liquid-cell holder with the inte- the specimen tip from the specimen rod – then the specimen tip
rior Cu rod attached to a cryogen dewar for low-temperature e-cell experiments. is thermally isolated inside the TEM.
Pt heater enables temperatures between liquid nitrogen and room temperature
Chapter 2

Cryogenics 2.7 Other Stimuli

Materials science at T < −150 °C or < 123 K.
N2 is liquid at T < 77 K. Originally, as shown in Chap. 7 of the W&C book, there were
(Cryo, though, just means icy cold.) 9 both top-entry and side-entry holders. Essentially all holders for
in-situ studies are now side-entry. The majority of in-situ hold-
ers in use today are what we might refer to as ‘conventional’.
During imaging with the conventional cryo holder, mechanical One reason that such holders are, and will continue to be, much
stability depends on how well you control the liquid N2 bubbling more widely available is cost. The external stimulus designs
in the dewar. Keep the liquid N2 level below the height of the can cost six times the conventional in-situ holder. If you are
copper rod. You might use a debubbler to remove ice crystals in bold and have access to the right workshop, you can modify a
the dewar and some of the liquid N2. Of course, inserting a warm holder or even modify the stage; few people will ever do this,
metal rod into the liquid N2 will also cause removal of liquid N2 but it is possible.
from the dewar.

Recently (2014), a different type of cryo-holder design was used

to study phase evolution in ice, Fig. 2.34. This holder takes ad- 2.7.1 Deformation
vantage of an environmental cell to begin the experiment in sta-
ble liquid (or gas), then transitions to a low-temperature phase by
cooling the cell with a copper cold-finger. The dewar at the back Studies of mechanical properties broadly cover three categories,
end of the holder, which is connected to the copper rod, may be namely elastic deformation (reversible deformation including
filled with liquid N2. Temperature control above the dewar tem- measurements of modulus), plastic deformation (permanent de-
perature was provided with a Pt heating wire controlled through formation including measurement of yield stress, flow stress,
a thermocouple. The message: innovative design of cooling hold- work hardening, stress relaxation), and fracture (the final break-
ers is possible too. ing for fracture toughness, etc.).

In-situ mechanical testing in the TEM has the potential to impact

Cryo Transfer two areas of research in particular. It can link specific nanoscale
Preparing plunge-frozen samples is generally practiced for soft/ deformation mechanisms to larger-scale (micro-, meso-, and
biomaterials (Bellare et al. 1988), though it can be used to study macro-) deformation, and it can be used to test volumes of ma-
other materials that are in contact with a liquid. In contrast to a terial that can be simulated theoretically using similar boundary
liquid-cell holder, this technique provides the advantage of struc- constraints, e.g., with molecular dynamics (MD) models. The
tural stability – the orientation is literally frozen in. You must earlier studies were mainly empirical because the stress could not
transfer the grid to a grid holder while keeping it under liquid N2 be measured quantitatively. That is now changing – the applied
to prevent ice formation on the sample surface. Once the grid is in load can now be measured (in principle, but beware). Four types
the grid holder for cryo storage, it is moved to the holder loading of load cases to investigate a material’s mechanical response: i)
station or into cryo storage. You can also prepare cryo microtome uniaxial (tension, compression); ii) bending (cantilever bending,
samples for transfer into the cryo holder by cutting frozen thin nanowire bending, resonance); iii) indentation (technically not
sections of larger samples (of frozen soft materials), keeping the indentation since there is not a hydrostatic component of stress
cut sample confined within a clam-grid under liquid N2. beneath the contact point); iv) buckling. (Buckling may occur un-
intentionally when testing using any of i–iii. Actually, buckling
To load the grid in the holder, you submerge the whole tip of the while performing mechanical tests in the TEM would be difficult
holder under liquid N2 with the dewar cooled so that the copper to analyze, but it could be quite interesting.)
 2.7 Other Stimuli 49

Limitations because the TEM sample must be thin can present

difficulties in relation to bulk properties. In the cleanest load
case, i.e., the one that is most consistent and uniform within the

Chapter 1
electron transparent region of interest, is the tension test; the
stress state in the electron transparent portion of the sample can
then be plane stress.

Motivation
So, we have two purposes for studying deformation in the TEM.
The first is to be able to deform small samples in a very con-
trolled manner. The second is to measure the load during the de-
formation process. In W&C, we mentioned deforming a sample

Chapter 2
that has been thinned to behave like a tensile specimen. Such
holders have little or no control over the actual load applied.
One advantage of the newer holders is that they can measure the
applied load. We can also pull a sample, compress it, fracture it,
or even study tribology. Piezo drives allow us to do such exper-
iments quantitatively.

Deformation mechanisms cannot be fully understood by

post-mortem analysis. The mechanical changes in materials
may lead to deformation or fracture, but we want to understand Fig. 2.35 A MEMS design for in-situ tensile testing of nanoscale structures. The
the mechanisms. In-situ studies allow real-time observation of strain is deduced from the displacements
the deformation events at various scales (down to atomic scale)
correlated with stress–strain curves. The TEM provides direct
imaging of dislocation behavior and characterization of the de- dislocations behave in confined spaces. Worse still, some re-
fects. Diffraction and dark-field TEM imaging can be used to sults from in-situ deformation experiments do not match theory
capture phase transformations during deformation. In principle, because the very high stress fields are much higher under the
in-situ nanoindentation gives a method for directly monitoring indenter – there are large stress gradients. In fact, for nanoin-
the deformation events as indentation proceeds. dentation, the stress fields are always inhomogeneous; the stress
gradients are much larger than when we try to use a uniform
tension or compression. Such stress gradients may even cause
Microcompression (Nanoindentation) or Microtensile grain boundaries to move at room temperature. Remember that
Testing the modulus of elasticity and fracture toughness depend on the
size of the material, therefore nanoparticles will have different
Classical nanoindentation studies are carried out on bulk samples values compared to a bulk film. This is related to the behavior of
where the indent only affects a very small region close to the sur- dislocations, the effect of surfaces, and the size of the specimen
face. When we image defects in the TEM during indentation, the (nanoparticles are small).
specimen must be electron transparent. Although we are calling
this in-situ nanoindentation we are actually just using a holder to An early MEMS design is shown in Fig. 2.35.
apply a local stress to the TEM specimen. One big advantage of
this technique is that we can ‘see’ the area that the stress actually Mechanical properties depend on the geometry and orienta-
acts on. Even though we observe the indenter in 2D we are able tion of the sample. The specimen thickness will be important.
to estimate the 3D geometry. Atomic-scale studies of, e.g., grain boundaries require you to
orient the grains along the zone axis so that you can observe
Mechanical testing has been widely used in SEM. There is much lattice changes during loading. Specimens for nanoindentation
more space in the specimen chamber, but we can neither see nor experiments are usually aligned perpendicular to the electron
fully characterize lattice defects at the resolution of the SEM. So beam for optimized imaging. We can thin bulk samples using
the challenge is to explore the mechanical properties when the mechanical grinding, polishing, and ion milling to create a
material is electron transparent. The maximum thickness of the wedge structure. Better still, but not without drawbacks, is
samples for these studies depends on the scale of the features the FIB. You may FIB-mill pillars of material with different
we want to observe during testing and the properties that will dimensions (diameters down to 140 nm with a length/diameter
control the mechanisms for a given material. In intermediate-kV ratio between 2 and 3) from bulk samples, Fig. 2.36. Unfor-
TEMs, the specimen thickness can be up to 5 µm, though it tunately, a FIB can modify the original microstructure, espe-
must be much less for atomic-resolution studies. The response cially in the all-important surface region. FIB creates a high
to deformation in nanostructured materials is very different than density of defects close to the surface of the specimen; surface
for bulk (large-grain) materials due to internal interfaces or how damage or contamination can never be neglected for shallow
 50 2 In situ and Operando

In a nanoindentation experiment, the sample faces the indenter

perpendicular to the direction of the electron beam; you can then
move either the indenter or the specimen. In modern systems
Chapter 1

where we expect quantitative measurements, a transducer is

used to measure the load vs displacement during TEM imaging.
Generally you want an indenter that will not deform when the
load is applied to the specimen during testing, so we usually use
a tungsten or diamond tip. However, as you’ll see in the TEM,
the indenter will also deform, at least elastically! The geometry
of the indenter will certainly influence the deformation mecha-
nisms, as in any other deformation experiment. FIB may be used
on the indenter tip to modify its geometry.
Fig. 2.36 FIB preparation of a NiTi thin film into a pillar for in-situ nanocom-
Chapter 2

pression. a Low-magnification image of the experimental set-up. b Higher-mag-

nification dark-field TEM image Deformation Holders
You can actually make a deformation holder in your workshop,
It is relatively easy to make a holder that can accurately apply
indents in nanoindentation. Therefore, during FIB preparation, a force to the sample in the correct geometry. The difficulty is
protecting the surface during milling may be essential, though measuring the force accurately.
nontrivial.
One of the two most popular deformation holders is made by
Local Stress Fields Hysitron, Fig. 2.37d. The Hysitron TEM Picoindenter system
has a nominally flat diamond punch that contacts the specimen
normal to the imaging direction. The indenter is mounted to a
It can be difficult to calculate local stress fields due to tip miniature transducer that is coupled to a piezo tube. To over-
imperfections, surface roughness and uneven specimen come the problems with hysteresis the piezo is kept stationary
geometry. 9 during indentation. This holder has demonstrated loading up to
1,300 ± 0.1 µN and displacements up to 5,000 ± 0.4 nm. The dis-
placement signal generated from the transducer directly mea-
The wedge approach uses micromachining. We can thin Si (001) sures the penetration depth of the nanoindenter, which can be
slices to give 100-nm wedges, ‘sharpening’ the wedge by repeated programmed with a predefined displacement profile as a function
annealing and oxide removal. We can then deposit thin films of of time to use the feedback system.
just about any material onto the wedged end of this Si substrate.
Of course, the substrate material will influence the measurements, When testing the modulus of a nanowire, we can use its mechan-
but you can correct for it by also testing the bare substrate. As a ical resonance. When the nanowire is positioned perpendicular
rule of thumb: the influence of the substrate is not significant for to the beam direction, we can use the modes of vibration of a
soft films on a hard substrate until the depth of indentation ex- nanowire to extract the mechanical response and then interpret
ceeds 50 % of the film thickness. If the FIB damage is a pro­blem, this using continuum and statistical mechanical models. In fact,
try using lithography to deposit materials on defined regions of a we can use either thermal or electrostatic resonance in the TEM
substrate, then etch away the unwanted material. to determine the elastic moduli of a nanostructure using a custom
built design, Fig. 2.37a (Treacy et al. 1996; Espinosa et al. 2012).
Although we have emphasized how to prepare and load a de-
fect-free specimen, you may actually want a preformed crack or A second holder (now produced by FEI – but still called the
other imperfection in the specimen, e.g., for stress concentration. Nanofactory holder) is configured with the diamond tip stable
Studying such controlled defects can aid in our understanding of on one end and the sample attached to the piezo tube on the
these processes at the atomic scale. opposite side, Fig. 2.37b. The sample is controlled in 3D with a
piezoelectric actuator, giving fine movements down to 0.1 nm/
Micromanipulators such as tungsten probes, used in sample step. The maximum load depth is set at ~ 200 nm to protect the tip
transfer during FIB preparation, may be used to place FIB cut- from damage. The force sensor is made from Si that is anodically
outs or other small structures (nanowires, nanofibers, and ma- bonded to glass with a replaceable tip. Forces up to 100 µN with
terials) onto MEMS platforms. This ‘pick-and-place’ technique a sensitivity of 0.27 pF/0.1 µN have been demonstrated, with a
for attaching the material of interest to a specific location on the maximum load of 10 mN. This holder is able to produce curves
mechanical testing device is now commercialized (EXpressLO). for load and displacement vs time. To operate in tension, this
Remember: the electron beam in the TEM can be used as a pow­ holder can be loaded with a micro-mechanical device on the
erful tool in specimen preparation. You can use it for welding, sample side to produce quantitative stress vs strain curves. An
depositing materials, or drilling holes. extra feature is the double-tilt capability, allowing you to orient
 2.7 Other Stimuli 51

Fig. 2.37 Schematics of experi-

mental stages used for mechanical
testing within a TEM. a Electrostatic

Chapter 1
resonance design. b Nanofactory
STM-TEM holder. c Nanofactory AFM-
TEM holder. d Hysitron compression/
tension testing designs. e Collodion
thin film on a grid that stretches un-
der α-beam irradiation. f Microfab-
ricated platform on a TEM grid for
4-point electrical measurements

Chapter 2
the specimen along a particular zone axis; this option is currently
(in 2015) not available for the Hysitron picoindenter.

The Nanofactory STM holder (see the section on electrical prop-

erties) is also widely used for piezo-controlled movement of a
small tip ~ 20–50 nm for the deformation of nanoscale materials.
The tip, which can be electrochemically etched to make it really
small (which is necessary for testing individual nanoparticles), is
moved at a constant velocity. Like the Nanofactory nanoindenta-
tion holder, this design allows us to control movement of the tung-
sten probe to ensure optimum contact with the specimen, but also Fig. 2.38 Fracturing a nanoparticle using an ‘indentation’ holder; the “indent-
allows for electromechanical testing of materials (for piezoelec- er” is actually larger than the nanoparticles. a Before and b after compression
tricity and piezoresistance). The mechanical measurements made and fracture
using this design are generally only qualitative, because only dis-
placement data by imaging can be tracked with force sensing.
Figure 2.38 shows a small spherical particle of Si that has been
AFM-TEM holders have been fabricated in various designs with compressed until it actually fractured. The same phenomenon
the commercialized holder by Nanofactory Instruments (FEI Co.) was observed in an AFM-based indenter but the fracture was
being equipped with a MEMS force sensor — a microfabricated ‘inferred’ rather than ‘observed’ – that’s the advantage of TEM.
device with a Si force-sensing ‘springboard’ cantilever, Fig. 2.37c. Incidentally, the fracture occurred in less than the time between
The sensor has piezo-resistive elements arranged as a Wheatstone frames when recording at 30 FPS.
bridge (so you can measure unknown electrical resistances) to en-
able a high degree of sensitivity with forces down to 10 nN. The Deformation mechanisms can be investigated through load
Si tip can be micromachined, but you have to know the spring or displacement control. The latter allows for load relaxation
constant of the cantilever. Other designs may use imaging of the (which occurs naturally), whereas load control requires the in-
displacement of the cantilever (requires low-resolution imaging) put of additional force to compensate for the relaxation. If there
during the test, or using the traditional laser-optical lever element is no force compensation during relaxation, the specimen may
that is used in other AFM designs; this is complicated here because undergo catastrophic failure. Similarly, in fatigue studies, the
we need sub-nN resolution. The AFM-TEM holder has been used forces must be compensated to relieve stresses for cyclability.
successfully in tribological studies and can be modified to allow
combined mechanical and electrical measurements.
 52 2 In situ and Operando
Chapter 1
Chapter 2

Fig. 2.39 MEMS platform for coupled tension testing and 4-point electrical characterization, as shown for testing individual nanowires. Scale bars: left top 200 µm;
left bottom 40 µm; center top: 4 µm; center middle 300 nm; center bottom 100 nm.

MEMS Instrumentation MEMS designs have the advantage of incorporating multiple ca-
From reading the previous section you’ll appreciate that the align- pabilities into one platform, such that temperature control, elec-
ment of the indenter and the specimen is going to be very import- trical measurements, and mechanical testing may be completed
ant. You’ll also realize that the tips used for nanoindentation may while imaging in the TEM. This means that we can, e.g., heat
easily be larger than the specimen we are interested in indenting not only the specimen but also the indenter tip, which will be
or the specimen may be inhomogeneous so we need to identify important at elevated temperature. Thermal drift is minimized
very small regions and test those explicitly. When we are in the na- as only a small volume of material is at temperature. Four-point
no-regime, we have the challenge of how to move a nanoindenter electrical measurements with leads may be performed on struc-
onto the sample within the TEM. The holder must be designed tures placed across a gap for coupled stress–strain analysis (Han
to provide the right geometry of the specimen and nanoindenter and Saif 2006); you can add the leads using ion-beam or elec-
perpendicular to the imaging direction. How can we fit all of the tron-beam deposition (IBD or EBD), or simply pattern them us-
necessary components/sensors into the tip of a side-entry holder ing lithography.
that can fit within the small pole-piece gap inside the microscope?
MEMS devices have greatly advanced the capabilities for in-situ Electrical Contact
holder designs. Now we can obtain stress vs displacement curves
using a picoindenter while imaging in the TEM.
Each electrical connection on a MEMS platform requires
A custom MEMS design holder might include an actuator (a another electrical feed-through on the TEM holder to
comb-drive electrostatic actuator – force-controlled, or an in- allow you to control/monitor each electrode separately
plane thermal actuator – displacement-controlled up to tens of during the experiment. 9
µm) with a load sensor (previously calibrated) based on a capac-
itance measurement (see Fig. 2.39). This design is controlled by
contact pads for electrical actuation/sensing with an integrated The other advantage of using a stand-alone MEMS device is
sensor circuit chip located outside the TEM to mitigate the effect that you can use it in the TEM, and then externally in another
of parasitic capacitance and to minimize stray capacitance and system allowing optimum measurement using, e.g., VLM, XPS,
electromagnetic interference. This type of set-up has achieved a STXM,Raman, or NMR. (VLM – visible light microscopy; XPS:
capacitance change of 0.05 fF, load sensor stiffness of 11.8 N/m, X-ray photoelectron spectroscopy; STXM: scanning transmis-
and load resolution of 12 nN (increased by decreasing the stiff- sion X-ray microscopy; NMR: nuclear magnetic resonance).
ness) (Zhu and Espinosa 2005).

Licensed to ([email protected])
 2.7 Other Stimuli 53

Table 2.1 Mechanical and electromechanical properties that can be obtained, loading mode, resolution and throughput of various in-situ holders/methods

Method Loading mode Measurement Capabilities Electronic read- Force resolution Testing

Chapter 1
out in-situ TEM throughput

Mechanical Electrical

Resonance Bending E N/A No N/A High

TEM/SPM

Nanoindenter Compression/ E, σ, ε, σf , εf N/A Yes Sub μN Low

based Tension

TEM/STM Bending/Tension E, ε, εf R Yes N/A High

TEM/AFM Bending/Tension E, σ, ε, σf , εf R In some setups 15 nN Medium

Chapter 2
MEMS Compression/ E, σ, ε, σf , εf R, ρ Yes 12 nN Low
Tension

Straining grid Tension (not well ε, εf N/A No N/A High

controlled)

Four-point grid N/A N/A R, ρ No N/A Low

Experimental Design to the beam direction, then the variability in the step motion will
The type of TEM holder required has to be suitable for the ex- define the standard deviation.
periment, and, of course, it has to fit your TEM! (It must fit into
the pole-piece gap.) The holder may be a single or double-tilt Various mechanical testing TEM holder designs are shown in
design. Double-tilt mechanical testing stages generally have a Fig. 2.37. The data shown in Table 2.1 is from the same paper
very low β-tilt range (approx. ± 5 °C) because you’re not just and summarizes the different loading modes that can be achieved.
tilting a 3-mm grid. Stability of the holder is a critical issue
in nanoindentation experiments to retain the region of interest For tensile tests, we can deposit specimens directly onto a
within the field of view. Typical indentation rates during these MEMS testing platform, such as the push-to-pull design by
experiments are around 5 nm/s for 50–150-nm thin films. Hysitron, then subsequently etch some of the film away using
FIB, thus selecting the region we want to test. Alternatively,
As with essentially all in-situ studies, you have to image the we can use the classic ‘dog-bone’-shaped specimen, produced
area of interest! So you have to be able to control the location by FIB, to create interlocking segments between the two sides
of the indenter, or the sample relative to the indenter. It may of the platform and then deform the thin middle section of the
be that you need to find a defect-free region of the specimen, a ‘dog-bone’ to localize the strain, Figure 2.37d. Always keep
specific crystal orientation, or a particular grain boundary. As in mind that the FIB can change the near-surface region of the
always, you set up the indenter perpendicular to the electron specimen.
beam to maximize the field of view. Using a piezoelectric tube
gives you the ability to move the tip in three dimensions with The quality of the results will depend on the quality of the speci-
nanometer precision. Older designs used a manual screw that men (as always), but also on having accurate control of the spec-
would move the indenter assembly against the vacuum bellows. imen holder, and being able to image at the appropriate time and
In principle, we can use any type of indenter that we use for ex- length scales. You may decide to emphasize recording the best
situ studies. For example, we can use a Berkovich-type indenter quantitative deformation data even if the images are not the best
(three-sided pyramid) made from boron-doped diamond so that you’ve seen – they might be just ‘typical’. If they contain the in-
it is electrically conductive in the TEM. The displacement of the formation you need, that may be sufficient. Generally, it is much
indenter, relative to the specimen, is controlled by the voltage more difficult and time intensive to acquire force–displacement
applied to the piezotube during indentation. The displacement of curves from an in-situ nanoindentation test than an ex-situ test.
the material depends on the applied voltage and, of course, the
response of the material to the indenter. Hysteresis of the piezo It will take time for even the experienced TEM user to become
control, creep, and saturation cause the correlation between the acquainted with positioning the indenter tip on the sample in 3D,
applied voltage and measured displacement of the piezo element while observing the 2D image. The holder should allow you to
to be nonlinear, making it difficult to calculate the load during move the specimen in all three directions with nanometer preci-
displace­ment. However, you can calibrate the piezo drive by sion, though you’ll need patience to contact the specimen with
applying a constant potential in steps to move the indenter lateral the probe tip without destroying the specimen in the process.
 54 2 In situ and Operando

Generally, the technique is to bring the sample edge to eucentric deformation holder. Most in-situ mechanical stages cannot pro-
height, then to use the alpha wobbler to ± 5 to 10° at low mag vide a high strain rate; you may want that for larger nanopar-
while moving the probe tip closer to the specimen until it is ticles. Typical elongation rates (strain rates) for TEM holders
Chapter 1

within ~ 500 nm and the probe tip is stationary. You can then range from 0.1 to 0.01 μm/s. Soon you will want to consider
go to higher magnification, making sure that the probe tip and other variables while nanoindenting e.g., changing the envi-
the specimen are both at eucentric height and moving the tip to ronment, the temperature, magnetic or electric fields, etc. You
contact the specimen. will want to combine deformation with precession imaging to
map out grain structures and orientations before and after the
It may be difficult to set up the two-beam condition for dif- experiment. At present, this technique cannot be used during
fraction contrast imaging when you are using an in-situ nanoindentation, in part because you need a high sample stabil-
nanoindentation holder, especially when you want to deform ity and long scan durations to acquire the precession electron
nanoparticles! You’re likely to lose the particle. Image inter- diffraction maps.
pretation can be more difficult than usual because the indenter
Chapter 2

will elastically deform the specimen so you’ll see lots of bend

contours and changes in thickness fringes (as the specimen Mechanical Properties of Interest
rotates), before you see dislocations moving; we refer you There are many more topics that we’d like to discuss, but we’ll
to Chaps. 25–27 of W&C for more discussion on this topic. just list a few. In general, there will be many future Ph.D. disser-
So, the applied stress can change the specimen in many ways tations on in-situ deformation, especially as the holders improve
– in some cases, it will simply relocate a nanoparticle rather and become more accessible.
than deform it. You can also use phase-contrast imaging, but
in practice, the specimen is rarely stable enough and tends to zz Shape memory effect in, e.g., Ni–Ti alloys (stress-induced,
rotate, changing the diffraction conditions enough for you to reversible, martensitic, phase transformations), compression
lose the lattice fringes. behavior, size effects and mechanical properties (plastic and
elastic deformation), stress vs strain curves, deformation char-
The advantage that we have in using a TEM holder for acteristics (e.g., Al–Mg films), dislocation and grain boundary
mechanical testing is that we can also carry out chemical analysis motion (e.g., in polycrystalline Al films), shear strength at
on the specimen at the same time. This may be helpful in identi- interfaces, effect of vacancies and interstitials at interfaces,
fying changes in the nanostructure due to mobility and reorder- formation of stacking faults at interfaces, dislocation plas-
ing. XEDS and EELS can trace chemical shifts and thickness ticity, grain-boundary strengthening.
variations as a result of the mechanical loading. For example, zz Using differences in thermal expansion coefficients between
chemical analysis can show when the mechanical disordering materials to study plasticity in metal and semiconductor films.
of the atomic structure in Si is caused by dislocation movement zz Effects of temperature, strain, strain rate, and impurity con-
rather than oxygen diffusion. centration.
zz Solute atoms are barriers to mobile dislocations – can we
Some points to consider: study the Suzuki effect finally, diffusion of solute atoms to
dislocations, serrations (stress drops or steps in the stress–
zz Are you setting up a uniaxial stress state (with the stress nor- strain curve).
mal to the interface)? What are the load measurement capa- zz When is plastic deformation controlled by dislocation nucle-
bilities of the holder you’re using and are they sufficient to ation?
measure the processes accurately? zz There can be unique bonding configurations coupling me-
zz The size of the indent may affect your measurement; smaller chanical and electronic properties in nanomaterials. Fig-
indents give higher hardness readings. ure 2.39 details a MEMS platform for these experiments.
zz The direction of grain boundary motion has been observed
As a rule: perform indentation using displacement control, and both at and away from the indenter; it depends on the stress
indent to the same depth each time to ensure that the dimen- gradients and the grain boundary structures. Can we correlate
sions are not influencing your result. For example, if you are the grain boundary motion and the loading? Is softening a
studying ion irradiation, deformation will be load-controlled loading response?
and will result in different indent sizes for the same maximum zz Size effect (smaller is stronger) has been shown for nano-
load. Worse still, the specimen size imposes boundary condi- crystalline and single crystals. Dislocations in small crystals
tions which can create local strain gradients in the specimen; escape more quickly at free surfaces – they’re closer.
these variations can be minimized by miniaturizing the tensile zz The specimen is critical: microstructural defects impact the
testing geometry. deformation of ion-irradiated materials.
zz Use a MEMS device to measure carbon nanotubes, biofibrils,
thin films, and nanowires.
Limitations zz Link hardening, yield strength, ultimate tensile strength, ul-
It may be difficult to prepare specimens that are compatible timate yield strength, and ductile-to-brittle transition temper­
with the additional limitation of having a small tilt range for a ature.
 2.7 Other Stimuli 55

zz Use nanoindentation for thin metal films; use nanocompres- We can image magnetic structures in the TEM using the Lorentz
sion for metal pillars, explore bending/buckling, resonance/ force (Sect. 23.7 of W&C). This is similar to how we use mag-
vibration of nanowires. Orientation is critical for tensile netic lenses to focus the electrons onto the sample. The Lorentz

Chapter 1
tests. force deflects the electrons causing a beam tilt with the deflection
zz Place the indenter tip to minimize surface effects. angle proportional to the induction perpendicular to the beam tra-
zz Strain can be characterized by the displacement of the di- jectory. A magnetic field in the specimen can deflect the electron
amond indenter divided by the initial length of the sample beam and cause contrast in the image. The contrast is modified by
structure. focusing above or below the sample (underfocus or overfocus) so
zz Post-mortem TEM analysis cannot track dislocations that that the deflected beams interfere with the beams that were either
have escaped the crystal during deformation! not deflected or deflected differently. In this imaging mode, we
zz Quantitative nanoindentation in the TEM requires calibration generate a vertical magnetic field so that tilting the specimen re-
of the force sensor; combine voltage and displacement. An sults in an in-plane change in the magnetic field. We can optimize
actuator is linearly related to the displacement. this approach by placing a special lens where the objective lens

Chapter 2
was, giving us low magnification images of the different magnetic
domains or domain walls within the sample; this phase contrast
Future Developments imaging technique is known as Lorentz microscopy.
Currently, many of the high-stability deformation holders only
operate on a single-tilt platform; future developments could For regular imaging in the TEM, the magnetic field in the objec-
provide double-tilt capabilities to view the specimen on a zone tive lens is set between 1 and 2 T. You can operate the TEM in
axis. Orientation mapping in these types of studies with ultra- low magnification mode which automatically turns off the field
fine grains is an extremely useful technique for understanding in the objective lens, giving the poor-man’s alternative to using
the motion of grains within the structure before and after in- a Lorentz lens. However, even when the objective lens is not in
dentation. Of course, phase mapping, like chemical mapping, use, there are residual fields about the specimen that should be
is limited by the high temporal resolution and high specimen accounted for: e.g., a value of 25 Oe has been reported at 300 kV
stability required. Increased temporal resolution afforded by for Ni81Fe19.
the DTEM would allow for high-strain-rate experiments if the
DTEM was able to obtain increased spatial resolution beyond In electron-beam holography (Sect. 29.11 of W&C and Chap. 8
its current abilities. Displacement control during tensile testing here), the incident electron beam is split above the sample us-
would afford experiments on phase transformations and discrete ing a biprism which effectively produces two virtual sources. In
dislocation yield events (Espinosa et al. 2012). Each advance in this section, we will concentrate on using an external magnetic
instrumentation opens up new experimental realms for under- stimulus, separate from the TEM imaging system, to create an
standing challenging science questions that had not previously in-plane magnetic field which we can use to probe the magnetic
been testable. properties and structures in materials. We want to observe the
domain structure when acted on by an applied magnetic field
without degrading the image.

2.7.2 Magnetic Fields In TEM, the sample geometry generally forces the magnetic
moments to lie in the plane: TEM specimens are thin. We can
produce an in-plane, applied, magnetic flux using a magnetiz-
As you know from many sections of W&C (e.g., 6.3, 23.7, 29.3, ing TEM holder to study magnetic-domain switching and do-
and 29.11), the lenses in a TEM use strong magnetic fields and main-wall motion at a microstructure level. The first account
these influence the images we record and even how we handle of this in-situ magnetization holder was published in 2003. The
magnetic specimens (Sect. 8.8). When we place such a specimen holder was designed to generate an in-plane field, up to 5 kA/m
in a large magnet (the imaging lens), it can be pulled out of the with four mini-coils, configured with two orthogonal ‘Helm-
holder if it is not firmly held in place. If the sample is not lost in holtz-like arrangements’. With this holder, demagnetization
the microscope, the native magnetic structure may be affected experiments could be carried out by either alternating the field
by placing it within the center of the objective lens. One way along a fixed direction while applying a decreasing amplitude to
around this problem is to turn off the objective lens when imag- the specimen, or by decreasing the field’s amplitude while the
ing magnetic samples, which will have the additional advantage direction is rotated in-plane (Uhlig et al. 2003).
of preventing the lens fields moving magnetic domains in the
specimen. Of course, it will also limit high magnification im- Another original design modified the Gatan double-tilt holder to
aging. We can summarize before we even begin the discussion: apply a 26-gauss pure in-plane magnetic field to the specimen
magnetic materials can be difficult to image in the TEM. without any remnant field (the solenoid does not have a magnetic
core) (Lau et al. 2003). This design consisted of a single solenoid
Material systems of interest include magnetic data storage at with the specimen positioned at the center Fig. 2.40. The applied
small scale, dynamics of domain motion behavior, and magnetic magnetic field from the current applied could be measured by
shape memory alloys. using a microstructured Hall device or by the beam deflection in
 56 2 In situ and Operando
Chapter 1
Chapter 2

Fig. 2.40 A modified Gatan double-tilt in-plane magnetic-field specimen holder. Fig. 2.41 Lorentz TEM magnetization studies on an Fe–Pd–Co martensitic alloy.
The red part is the solenoid, which forms part of the holder The use of Fresnel images; white arrow indicates a magnetic domain wall that
coincides with the twin boundary and black arrow indicates a zig-zag wall that
runs across the twin boundaries. c Color map of the integrated magnetic induc-
tion of the region in a imaged underfocus, and b imaged overfocus
diffraction mode. It is also possible to modify a TEM holder to
be a Hall probe that can accurately measure the remnant field at
the specimen location within the TEM (Lau et al. 2007). In fact,
magnetization holder designs are now capable of producing an ine nanometer-scale sensors and actuators, or probe nanoelectro-
in-plane field without significant beam deflection and no hyster­ mechanical systems (NEMS).
esis effects (Yi et al. 2004).
If you want to know about electrical properties of bulk structures,
A magnetically shielded objective lens will limit the effect of the testing in the TEM may be problematic since the beam can change
magnetization holder on the electron-beam deflection, though a their electronic properties, and the change in the ratio of surface
deflection system can be implemented to correct for holder-in- area to volume can seriously impact the electrical response. So,
duced deflection up to 200 Oe (Inoue et al. 2005; Arita et al. you are more likely to see nanowires and nanoparticles tested in
2014). We have to consider the heat produced by current passing the TEM – the structure of the nanomaterial is then already de-
through a solenoid, since this may affect the specimen depend- fining its carrier properties. We can probe electrical properties of
ing on the holder design. Hummingbird Scientific currently of- nanomaterials and thin films using an electrical probe to contact
fers a holder that allows control up to ±900 gauss (depending various regions of the nanostructure that are in contact with a con-
on the TEM and the pole-pieces) with an applied field between ducting substrate. Nanowires provide an ideal structure to probe
–500 and +300 Oe, while you are actually imaging your speci- the electrical properties of materials as there is one direction for
men. To compensate for the deflection of the electron beam by conduction through the structure, which can be used to define the
the applied magnetic field, the design has an integrated passive carrier transport mechanism through the material in one dimension.
magnetic compensator, meaning that the specimen can still be
tilted by up to ±45˚. This holder has been used to investigate, Using the TEM, nanostructure characteristics we can determine
e.g., the depinning of magnetic domain walls from twin bound- are the length and diameter, surface structure and roughness,
aries in a Fe68Pd30Co2 martensite alloy, Fig. 2.41. In-situ obser- crystal structure and orientation, composition, and purity of such
vations were able to provide an understanding of the behavior nanomaterials. These characteristics must be properly defined
of magnetic domain walls and their interactions with micro- for a nanostructure’s conducting properties to be quantitatively
structural features under the influence of an applied external determined, as they will be strongly influenced by the surface
magnetic field. composition and surface defects/states. Scattering along the sur-
face of a nanoscale material can change the transport properties
in the structure compared with larger structures (> 100 nm) where,
e.g., grain boundaries might dominate. The final factor strongly
2.7.3 Electric Fields controlling carrier transport is the crystal structure of the nano-
material.

We might ask, which of the following can be examined in the Using current–voltage sweeps we can produce I–V curves,
TEM? Nanoelectrodes, opto-electronic devices, field-emitters, where the resistivity of the structure may be used to determine
memory devices, electromigration, piezoelectric devices, trans- the conductivity of the nanostructure by accurately measuring
parent conductor materials, chemical and biosensors, hydrogen the carrier path length in the TEM. Then by analyzing the I–V
storage, electrochemistry, catalysis, or solar cells (photovoltaics). curves, a calculation of the carrier concentration and mobility
The answer of course is ‘all of them.’ We can also use TEM to can be produced. So, once again, in-situ TEM characterization is
look for materials with high electron-conducting abilities, exam- ideal for obtaining quantitative electrical data with atomic-scale
 2.7 Other Stimuli 57

Chapter 1
Chapter 2
Fig. 2.42 Joining together two electron-transparent pieces of tungsten using
the STM holder with a 3 V potential applied, and then pulling the joined material
to observe how it deformed. Essentially, the process is electromechanical testing

structural and compositional characterization from individual

nanostructures.

Nanowelding

Apply a relatively small current between a structure and

probe, but is it welding? In TEM you can see! 9

Fig. 2.43 a Design of an STM holder that can simultaneously be used for re-
We can make qualitative electrical measurements while we me- flection electron microscopy (REM), reflected electron beam shown by 100 kV
arrow scattering form the specimen surface. b–d Images acquired from tunnel-
chanically strain a structure using an STM-TEM holder; e.g., we
ing during scanning
can apply a force to the material to induce bending during the
electrical measurement.

Figure 2.42 illustrates how we can use an electric current to AFM/STM

weld together two wires (or a thin protrusion and a wire) and The first construction of a scanning tunneling microscopy (STM)
then apply a stress to observe the mechanical deformation with tip in a side-entry TEM holder (by John Spence in 1988) used
atomic resolution; remember that your STM-TEM holder (see reflection electron microscopy (REM – see W&C Sect. 29.6.A)
Sect. 2.7.1) may only have one tilt axis, so aligning the grains to for imaging the surface of a material: the high topographic reso-
a zone axis can be tricky. Actually the process was sintering since lution of REM can be combined with spectroscopic information.
the material was being imaged during the joining and no melting The prototype was designed with the STM tip axis horizontal
was observed – the TEM tells you that immediately. to the specimen so that the REM imaging and diffraction pat-
terns could be obtained from the same region as illustrated in
Fig. 2.43a. A tube scanner was used, where a voltage applied to
the inner electrodes relative to an outer electrode will create a
 58 2 In situ and Operando

length change along that side, resulting in x or y displacement of in the resistors, is relative to the electron-beam dose which will
a tip placed at the end of the tube. Displacement along the Z-axis impact the intensity of backscattered electrons from the imaged
was controlled by applying a common bias to all of the inner structure/substrate.
Chapter 1

electrodes. Thermal drift in this design was minimized using

phosphor-bronze alloy components. REM images of the tungsten For I–V measurements using an individual nanowire, the current
STM tip scanning a graphite sample are shown in Fig. 2.43b; will be in the picoampere to nanoampere range. Voltage control
the high-vacuum environment of the TEM is suitable for such and sensitivity of the current reading will greatly influence the
surface studies (Spence 1988). The STM-TEM holder has been noise in the I–V curves, and thus, the quality of this measure-
used less frequently for actually studying surfaces; rather, it has ment. We should really reevaluate the design of the commercial
gained most interest and notoriety for its ability to apply voltage/ holders to incorporate extra shielding for the connectors between
current to nanospecific areas on a TEM specimen. the probe tip and sample mount of the holder; with the inter-
nal and external cabling from the holder, unshielded wires will
A newer STM-TEM design probes the electrical properties of produce noise in the current measurement which may easily be
Chapter 2

materials using nanometer positioning with a piezo-scanner for in the nano- to microampere range. When you set up this type
manipulating the probe in all three dimensions (Nafari et al. 2011) of measurement, you should consider everything that will be
Fig. 2.37b. The tip and substrate are generally composed of Au, in contact with the holder/specimen during loading and testing
Pt, Al, or W. The probes can be electrochemically etched, which in the TEM. Think about grounding or floating the electrodes,
is a delicate procedure for obtaining tips below 100 nm diameter grounding yourself, possible static discharge, any charging of the
(ethanol in place of water has been effective in reducing bubbling cabling when you change its location, and contact with charged
for Au etching in HCl (Lopes et al. 2013)). Samples are generally surfaces. In the latter case, an example is the inside of the go-
attached to a metal wire (~ 0.25 mm diameter) with a conductive niometer, which may sit at –2 V – it creates a circuit from the
epoxy for loading in the holder. At the end of the piezo-scanner holder to the objective lens to activate the ‘touch alarm’ in newer
is a sapphire ball, where the probe is mounted into a six-elastic- TEMs. Also, if you use currents that are too large during elec-
legged hemisphere holder with very small screws. (Don’t lose the trical testing, you may cause unwanted resistive heating (Joule
screws! This is very easy to do!) This design allows for millime- heating), though increasing the voltage may aid in the removal
ter-scale coarse motion (stick–slip inertial sliding actuators) and of samples from the end of the probe tip (☺).
subnanometer fine motion (solid deformation of piezo-actuators)
in all three directions. We mount the specimens in the holder When you design an experiment, the equipment must give re-
and position them by eye to the same z height with the minimal producible results; if the W probe has a significant oxide during
possible gap between them. Through touching the conductive tip one test, and a clean metal surface on a different test, then the
(generally W) to the end of a nanoparticle/nanowire on a conduc- voltage measurements will change to reflect the overpotential
tive substrate, you can obtain an I–V curve from the individual required to break through the passivating surface oxide on the
nanostructure. You might think that maintaining good electrical probe. You could easily misinterpret the difference as a charac-
contact between the probe and the nanomaterial will likely re- teristic of the nanomaterial sample being tested. To avoid this,
quire applying a stress to the nanomaterial. In addition, once the clean the W probe in the microscope just before beginning the
probe and nanomaterial are in contact, the sample position must electrical measurements. One method we use is to short the probe
remain stable to maintain contact during testing, so image shift tip to the sample substrate, passing a large current through the W
can be used to center the image on the camera. If you sweep the probe to remove the oxide barrier on the probe surface. Be aware
voltages, the current measured will depend on the set-up of the that too high a current may reshape the probe and change the
circuit: contact between the sample substrate and the holder, the < 100-nm tip into a poorly defined tip; this reshaping is also the
surface characteristics of the conductive STM tip, and sample reason to mount the nanomaterial samples on a W or Al support
surface will all impact the impedance of the measurement. that we’ll use for discharging the W tip and removing the surface
oxide prior to contact with the nanomaterial of interest. It is very
You can use AFM-TEM holders (see Sect. 2.7.1) for electrical important that you know the shape of the probe if you really want
measurements with slightly different sample loading from the to understand the electrostatic and chemical forces between the
STM-TEM holder, Fig. 2.37c. The AFM tip resides on the static tip and the sample relative to the tip radius and tip-to-sample
side of the chip, so the sample is loaded onto the probe tip on the distance. To summarize, you must prevent electrostatic discharge
piezo-tube end. You can mount the sample by dipping a freshly between the probe and sample if they are at different potentials
cut Au or Al wire, preloaded into the hat, into a dried material after being loaded into the TEM column.
powder or by scraping along the surface of samples grown from
a substrate (Golberg et al. 2009). Metal wires should be freshly Electron-beam-induced deposition (EBID) of free carbon in the
cut to minimize the amount of surface oxide that forms on the TEM column is often used to ‘weld’ nanowires to the end of a
wire, which will create impedance in the electrical measure- probe for manipulation and controlled contact between the con-
ment. The force measurements are calibrated prior to sample ductive probe and the sample. This technique generally takes
measurements for the spring constant of the Si cantilever, which some time depending on the amount of carbon contamination
may be made using soft indentation of cut Au wire. Noise in you have in the TEM column! Some researchers have been
these experiments, above the inherent 1 nN from electronic noise known to add in contaminants to speed up deposition rates. Note:
 2.7 Other Stimuli 59

localizing carbon using EBID isn’t always confined to the illu- of individual nanostructures during TEM observation because it
minated region; contamination can coat the nanowire surface, allows us to measure currents. The majority of electrochemical
affecting the desired electrical properties for the final experiment. experiments within the TEM have been committed to research-

Chapter 1
Using this technique you can attach a single nanowire on the end ing materials for batteries.
of a probe and use this probe for electrical or electrochemical
measurements and even for mechanical testing (see Sect. 2.7.1).
A clever technique you can use to place an individual nano- Battery Materials
structures on the end of a probe is to dip a sharp tip into a dou- The open-cell set-up is ideal for TEM, of course, since the elec-
ble-sided silver conductive tape to create ‘sticky probes’ and trolyte is a solid or a low-vapor-pressure (LVP) liquid. An exam-
then move this sticky probe across the sample surface to attach ple of the latter is the ionic liquid used in many studies of battery
nanomaterials to the probe surface (Dong et al. 2008). materials; LVP liquids are stable in UHV. With this design, we
can test a complete two-electrode cell by loading the material
Charging induced by the electron beam is very common during of interest on the stationary side of the STM-TEM holder, and

Chapter 2
STEM imaging, as a focused probe (< 1 nm diameter) contains the conducting tip with the solid or LVP liquid electrolyte, on
all the electrons used for imaging at each pixel and they must the piezo-controlled conducting tip with or without an opposing
be stopped on the material, which may cause damage, but also material supported on the wire base.
charges up the structure. Of course, we should always calibrate
the electron-beam current – the screen current reading is usu- In the study of lithium-ion batteries, the open-cell design allows
ally not as accurate as that obtained using a Faraday cup. Note: us to test nano-anode materials such as Sn, SnO2, or Si nanow-
if the beam current is high enough, it may register when you ires in the TEM using a Li salt containing LVP electrolyte on a
read the current values while imaging a nanostructure; such typical cathode material such as LiFePO4 or LiCoO2, Fig. 2.45.
spikes will result in a ‘galvanostatic hold’ when taking ‘snap- We can examine Na-, Mg-, or K-ion battery materials using the
shots’ of a structure. Given the value provided by a Faraday same approach, just replacing the counter electrode as an ion
cup, you can calculate the electron flux passing through the source. Solid-state batteries have also been tested in the TEM
structure. All materials will have a threshold electron flux that using Li metal. If the Li metal is exposed to air, a thin layer
will initiate damage in the structure. So, you have to consider of LiO2 forms on the surface; this layer is a solid electrolyte.
not only beam-induced charging, but also beam heating, which Actually, this oxide layer (~ 1 μm) will form on the surface of
can change the response from individual nanowires under a the Li while you carry it from the glove box to the TEM using
high beam current. Mechanical strain on the material can in- an air-tight bag; the brief exposure to air as you load the holder
directly occur when attempting to make good contact between into the TEM is long enough for the oxide to form. The Li2O
the nanostructure and a conductive tip, as this may drift over layer is electrically insulating but conducts Li+ ions when driven
the duration of the experiment (causing strain or disconnection at about –2 V, with a potential barrier of 0.4 eV, Fig. 2.45. This
from the sample). design is fairly straightforward and can be successfully extended
to prepare other air-sensitive samples, including Li-ion battery
The STM-TEM holder has been used to study the effect of materials. The main challenge is that you have to load the STM
electron-beam irradiation when testing electrical properties of holder in the inert atmosphere in a glove box! A special caution
nanostructures. For example, in ZnO nanowires, electron irra- to those interested in Li–S chemistries, there are glove boxes and
diation can actually lead to the formation of Schottky contacts TEMs dedicated to working with sulfur, as it readily diffuses and
at the ends of the nanowires; i.e., where the nanowires connect cannot be removed!
a metal support to the probe tip, Fig. 2.44. Electron–hole pairs
were generated by the beam and caused an increased density of When you are testing anode nanomaterials, 2D materials, or
carriers, which in turn improved electrical conduction through thin films for Li-ion batteries, using the open-cell approach, you
the nanowires. generally don’t submerse the electrode material in an electrolyte
as you would want to for an operando study. What this geometry
does give you is the ability to reach the resolution limits of the
Electrochemistry TEM – there is no background scattering from electrolyte or
Studying electrochemical systems in the TEM is a growing inter- from the membrane windows that are required in the closed-
est since the processes occurring at the interface between an elec- cell studies. If you do, in fact, immerse materials in the LVP
trode and electrolyte are still poorly understood in many systems. liquid during an experiment, concurrent imaging will not be
Electrochemical processes that we can study in the TEM include possible because the electrolyte layer will be too thick so that
intercalation, deposition, stripping, corrosion, electrolysis, and the background scattering would be overwhelming. Some ex-
the growth and assembly of nanostructures. Electrochemistry in perimentalists have, instead, used ‘dipped’ experiments to cycle
the TEM uses two different set-ups: open cell or closed cell. In an the structure and remove the electrode from the electrolyte for
open cell the electrode and electrolyte are exposed to the vacuum high-resolution imaging. Although the electron-beam scattering
in the TEM; in a closed cell the liquid is hermetically sealed at is not compromised for open-cell studies, you have to be careful
1 atm (see Sect. 2.5.3). Since 2000, the STM-TEM holder has to minimize electron-beam damage on the battery electrodes
been used increasingly for testing the electrochemical activity (especially once lithiated) and the solid or LVP liquid electro-
 60 2 In situ and Operando
Chapter 1
Chapter 2

Fig. 2.44 Electrical probing of ZnO nanostructures within the TEM. Au tip created a Shottky barrier with the ZnO nanowire due to electron-beam induced excitation
at the contact region

lytes: solid electrolytes, such as Li2O will rapidly reduce to the possibly failure in batteries, but the experimental challenges are
pure metal under the electron beam, and this metal can cause a always present.
short circuit for these experiments. LVP ionic liquid electrolytes
have been seen to ‘gel’ when exposed to the electron beam; this TEM studies of battery materials have revealed different mech-
is clearly evidence of electron-beam damage and polymerization anisms for Li insertion into a host structure, including alloying,
of carbon-containing electrolytes. As an example, for the case intercalation, and a conversion reaction. Li insertion mecha-
of LiTFSI in P14TFSI, the critical electron dose threshold was nisms have been explored for materials such as SnO2, ZnO, Si,
only ~ 10−3 A cm−2. Ge, Ru, Al, Sn, Fe2O3, Co9S8, Ag2S, and carbonaceous species
(graphite, CNTs, and graphene) using the open-cell geometry.
It will be obvious that open-cell experiments are preferable to In the TEM, we obtain direct observations of changes in the lat-
closed-cell studies when you want to use EELS or XEDS since tice structure during the electrochemical activity. An example is
there is no membrane window. Overall, in-situ electrochemical the lithiation front moving across an amorphous-silicon-coated
TEM has enormous advantages in improving our understanding carbon nanofiber composite anode, Fig. 2.46. The electrode
of reaction mechanisms, electrode–electrolyte interface chemis- material can be studied at high resolution to clarify swelling,
try, and degradation mechanisms that lead to capacity loss and
 2.7 Other Stimuli 61

elongation, fully charged structures, and electrochemical-me-

chanical coupled effects (dislocation nucleation and fracture).

Chapter 1
As an example, Li insertion into the structure has been shown
to be isotropic for Ge nanowire anodes and anisotropic for Si
nanowire anodes. Fully lithiated materials have been character-
ized by diffraction by identifying the phase of the structure and
the relative loading of Li into the initial lattice or amorphous
material; fully lithiated Si structures contain 3.75 Li atoms per
Si atom at room temperature. Controlling the interface of the
electrode has implications for engineering anode materials where
we need to control morphological changes; such studies are an
illustration for the Ge–Si core–shell structures in Fig. 2.47. An-

Chapter 2
other route to controlling the interface is to use a solid ionic con-
ductor such as LiAlSiO4 to protect the anode from direct contact
with the electrolyte, which then reduces parasitic reactions on the
electrode surface that may limit electrochemical performance.
Many questions concerning electrode interfaces for batteries
are focused on the solid–electrolyte interface (SEI). (An SEI
Fig. 2.45 Schematic of the electrochemical open-cell study of materials used forms when a battery operates: it is a solid layer between the
in lithium ion batteries; designs for using an iconic liquid or solid (Li) electrolyte electrolyte and the electrode.) We want to know about the ionic
are shown and electronic transporting properties, the structural composition
and components, and the evolution of the SEI layer during the
operation of an electrochemical cell. Closed-cell studies have a
big advantage for characterizing the SEI: industrially relevant

Fig. 2.46 Open-cell lithiation of an amorphous silicon coat-

ed carbon nanofiber during charging. The structural evolution
during charging of an amorphous alloy followed by a crystal-
line phase was captured with images and diffraction patterns
 62 2 In situ and Operando
Chapter 1
Chapter 2

Fig. 2.47 Atomic-scale imaging of the dynamic lithiation front in core-shell Ge–Si nanowires in comparison to pure Ge nanowires. Interfacial engineering can be
used to control the lithiation mechanism and kinetics

Licensed to ([email protected])
 2.7 Other Stimuli 63

Chapter 1
Chapter 2
Fig. 2.48 Quantitative Li deposition and stripping on a < 1-μm2 Ti electrode for two cycles. Note the Li morphology differences between the two cycles, and the
impact of the electron beam (image taken at each arrow) on the measured electrode voltage

carbonate electrolytes can be tested while observing the native morphological and electrochemical performance, and different
SEI that exists in working Li-ion batteries. beam-induced electrolyte degradation.

To illustrate the use of a closed cell for volatile electrolytes, Note that when we test the electrochemical behavior of individ-
we show an example of picoampere current control for elec- ual nanowires, the currents are femtoamps to picoamps, so the
trodeposition and subsequent stripping of Li on ultramicroelec- potentiostat requires a femtoammeter to measure and control the
trodes. Noise in the voltage plateau for a galvanostatic hold can current through the nanostructure, quantitatively. Very few open-
be directly correlated to the nucleation of new Li grains on the cell studies have provided quantitative electrochemical data with
electrode surface (Fig. 2.48). Take note of the initial Li morphol- simultaneous structural imaging within the TEM, but this will
ogy and electrochemical plateau prior to beam exposure (only change.
one or two Li grains). Give special consideration to data and
image contrast interpretation, because mass-thickness contrast
can be difficult to interpret when you have elements that scatter
less than the background, such as bubbles or Li metal deposits.
Different salt–solvent combinations are likely to show different
 64 2 In situ and Operando

Fig. 2.49 a, b STM-TEM holder equipped with an optical fiber

directed at the specimen. Electrical characterization and use of
CdS (c, e, g) and CdS/ZnO (d, f, h) nanostructures as photode-
Chapter 1

tectors in the absence of the electron beam. In this experiment,

the STM-TEM holder was used to probe the I–V character of the
specimen (e, f), and separately, to detect photocurrent stimulated
by the laser source on the same nanostructure (g, h)
Chapter 2

which can be swapped out for the objective aperture in the TEM.
2.7.4 Photons We are not going to go into detail on this technique, but instead
we will discuss how to probe the sample in the TEM with light
rather than the electron beam.
The creation of photons during a TEM experiment occurs be-
cause we have inelastic scattering in the sample; we generate Irradiation of TEM samples by light is fairly uncommon out­side
electron–hole pairs, which recombine to emit photons. Since the of the UEM community. Depending on the experiment, you
photons are emitted when the electrons excite the sample, we can must consider first what photon energies are needed to produce
collect optical information with high spatial resolution. For spec- the intended effect. We use optical probing in the TEM to study
troscopy studies using cathodoluminescence, we place a mirror photovoltaics, photocatalysts, piezo-phototronic effect, surface
close to the sample so that the signal is focused onto a fiber optic acoustic wave devices, opto-electronic devices, and transparent
to be monochromated and detected with a photomultiplier tube conducting materials. We can incorporate the light source either
(Zagonel et al. 2010). Gatan and Attolight sell these instruments, by using a specially designed holder or the beam can be focused
 2.7 Other Stimuli 65

Chapter 1
Chapter 2
Fig. 2.50 a Schematic of optical fiber incorporation into the pole-piece re-
gion of the TEM through the objective aperture port. b Comparison of the light
intensity of the sun on the earth’s surface to that of the external optical fiber
on the TEM sample. c ETEM images of un-electron-irradiated TiO2 nanocrystals
after 40 hours of exposure to water vapor and light (UV range intensity at
122 mW/cm2) while being held at 150 ˚C

through a port above the specimen without limiting the speci- In the study of photocatalysts, we want to use broadband light
men holder, so you may still use heating, cooling, mechanical to simulate conditions experienced by particles interacting with
straining, or electrical-biasing capabilities. Flexible optical fi- sunlight. Incorporating a photon emitter, in place of the objective
bers can guide the light more precisely than a lens-based system aperture strip, gives us increased flexibility in the light source,
composed of lenses and mirrors. Though fiber optics makes Fig. 2.50a. Miller and Crozier (2013) used a xenon lamp pow-
it easy to incorporate and change the light sources, they give ered by a laser to provide the light; the xenon plasma is respon-
an illumination cone and must be coated with an electrically sible for the light emission (1,460 mW/cm2). The high brightness
conducting layer to dissipate charge from the electron beam of this emitter is achieved by making the source diameter small,
during operation. On the other hand, a stable optical system namely 100 µm. The light from the source is guided to the sample
can give much greater power transmission focused down to a using an optical fiber with a 6-mm-diameter core, giving minimal
smaller area. intensity loss. Directing the light so that it fully illuminates a
TEM specimen positioned in the pole-piece gap of the micro-
Cavalca et al. (2012) have described two ways to modify a sam- scope is not a trivial alignment: you use micromanipulators for
ple holder so that it can be used with any compatible goniometer fine adjustments while the fiber optic is positioned in column.
configurations. Therefore, light incorporation isn’t limited by Remember that the XEDS detectors, the cold trap, and in an
any microscope modification – the light passes along the speci- ETEM, the gas inlet and outlet ports, are also located close to
men rod. These holders are not yet commercially available, but the specimen. Keeping the optics outside the TEM gives more
can be soon – especially if you know a good machinist. One flexibility in controlling the light that reaches the sample. An
design used a system of lenses and the other fiber optics. The aperture can then be used to control the light intensity, a filter
latter integrated a laser diode into the tip of the holder so that to select the spectral energy range, and tunable filters to select
photons can be directed onto a tilted TEM sample grid. Such a a narrow band at any wavelength. Which materials you can use
customized holder can allow you to combine heating and elec- to bring a light source near the sample will depend on the tem-
trical biasing capabilities to the UV-vis and IR-vis optical fibers. perature you expect to use (when using a heating holder), the
A design demonstrated by Zhang et al. (2013) used a multimodal electrical conductivity of the guide (electrons will be scattered
optical fiber connected to a laser diode such that the light travels onto the source), the reactivity of the guide with gases within an
90° to the specimen rod, Fig. 2.49. The sample sitting on the ETEM, any possible magnetic properties (which may distort the
piezo-controlled STM tip is then illuminated while individual electron beam), and the vapor pressure which will need to be
nanostructures can be mechanically strained or characterized low for a UHV-TEM.
electrically.
 66 2 In situ and Operando
Chapter 1
Chapter 2

Fig. 2.51 a Parabolic mirror for photon collection from the sample placed above the lower objective lens. b Schematic of the fiber optic containing specimen holder
in the TEM. c Temperature mapping a MEMS heater using Raman spectroscopy, demonstrating high temperature variability from the center to the edge of the device

Characterize the beam diameter and set the intensity of light If we can add a light source to the TEM, the obvious extension
reaching the TEM sample holder outside of the column to make is to perform other complementary in-situ measurements such
measurements easier; a photodiode positioned at the end of a as X-ray diffraction, X-ray photoelectron spectroscopy (XPS)
sample holder can directly determine the intensity and position and/or Raman spectroscopy. We encounter difficulties when
of the light in the column. We can align the highest intensity light transporting a platform between instruments, although it can be
point and the electron optical axis using an internal photodiode. done. So new efforts are being focused on combinatorial tools,
This alignment is critical because the light intensity will vary which allow for both TEM imaging and in-situ optical measure-
over the TEM grid; during any experiment you will find that only ments. For example, Picher et al. (2015) have demonstrated that
small regions within a defined area of the TEM grid are exposed we can add Raman spectroscopy to a TEM by inserting a para-
to intensities above 90 % relative to the emitted intensity. You’ll bolic mirror between the sample holder and the lower objective
have to perform this alignment before carrying out the in-situ lens pole-piece so as to focus light on the sample or collect the
experiment – using the specialized photodiode holder. emitted/scattered photons, as illustrated in Fig. 2.51. The broad-
 2.8 Adding or Removing Material 67

band high-efficiency light source is directed onto the sample,

and Raman or CL signals can be acquired during environmental
STEM imaging. Making and placing the parabolic mirror cor-

Chapter 1
rectly directs the laser focal point to coincide with the electron
beam. Note that the area in the STEM field of view and in the
Raman field of view will be different, because in this system
the light covers a 100- ± 60-μm2 area. This means that the Ra-
man signal will result from microscale features in the sample,
Fig. 2.52 Ge growth at the tip of a GaAs nanowire via a VLS mechanism in situ
whereas typical high-resolution STEM imaging will cover an in 7.6 × 10−7 torr Ge2H6 at 470 ˚C, normal to the electron beam. Dark-field TEM
area that is smaller by a factor of ~ 105. Such a system has been images acquired at time delay after gas exposure of 53, 60.2, 68.1, 81.4, and
used to measure the temperature when heating with a MEMS 87.4 min
platform in an ETEM; shifts in the Raman peaks should give an
accurate measure of temperature. However, it has been reported

Chapter 2
that measuring the temperature in different holes in a MEMS zz We can use a higher gas pressure in the column when the
heating chip can give significant variations (±80 ˚C) at different electron beam is off. This might be useful to study the initial
holes, as shown in Fig. 2.51c. This means that reproducibility in stages of nanowire growth.
heating experiments when using the MEMS system is dependent
on where the data is acquired.
Surfaces and the Growth of Thin Films
As always, we have to consider the effects of the electron beam Surface processes such as growth, etching, chemisorption, and
in these experiments; inelastic scattering in the sample can gen- roughening can be observed using the UHV-TEM. It has been
erate electron–hole pairs in photocatalysts just as light can. The used for imaging the crystallographic structure of small metal
study of structural changes in anatase TiO2 in 1 torr of water nanoparticles at the initial stages of formation by vapor deposi-
vapor at 150 °C (see Fig. 2.50c) found that the total energy de- tion. This is so impressive that it is easily forgotten that catalyst
posited during electron-beam exposure can be comparable to particles aren’t used in UHV.
extended irradiation (40 hours at 1,460 mW/cm2) in broadband
light. For reference, in terms of absorption of photons from elec- Some TEMs have a UHV chamber added to the specimen-load-
tron scattering, the external light source produces about 10 suns’ ing region. This allows you to deposit or process a specimen
worth of light intensity (Fig. 2.50b), where the 200 kV electron prior to loading it into the TEM – you don’t have to remove
beam can produce a light intensity of 100,000 suns. it from the UHV environment. The external chamber may also
have in-situ cleaning capabilities. As an example: you can study
CVD growth in the UHV-TEM by introducing oxygen, germane,
or disilane. The growth of Ge nanowires on an Si substrate can
2.8 Adding or Removing Material be carried out like this, but you have to clean the substrate by
resistive heating to 1,250 ˚C prior to depositing the Au catalyst
particles on the surface. So you do the cleaning and Au deposi-
2.8.1 Depositing Layers/Particles tion in the attached UHV chamber.

When you are interacting the specimen with a gas, one side of Growth of Nanowires or Nanorods
the specimen (usually the side that’s upper in the TEM) is usually We can study vapor–liquid–solid (VLS) growth directly in the
directly exposed to the gas by a capillary tube at a distance a few TEM; introduce the gas so it can contact the surface of the liq-
millimeters away. The capillary tube is particularly important uid and then diffuse to the catalyst/wire interface and continue
if the gas is very reactive, so you don’t corrode the rest of the the growth. A TEM can be attached to other instruments, es-
holder. pecially when connected by a load-lock to keep the specimen
under vacuum. For thin-film deposition it could connect to, e.g.,
zz The challenge: how can we characterize the gases that are a magnetron sputtering system, an MBE deposition chamber, or
contacting the specimen? We use a quadrupole mass (resid- an e-beam evaporator.
ual-gas) analyzer.
zz The temperature of the sample is calibrated using infrared We can incorporate a solid-source system to the TEM; one route
pyrometry (measures thermal radiation). is to use a port above the upper pole-piece to allow a molecu-
zz UHV-TEM systems always run with lower gas pressures in lar beam through the bore. Figure 2.52 illustrates the growth
comparison to ETEMs even when you’re growing thin films; of GaAs nanowires. Note that the specimen geometry has been
an ETEM will have a more stable environment at the speci- optimized for the experiment.
men.
 68 2 In situ and Operando

Fig. 2.53 Pt/Al2O3 catalyst exposed

to 10 mbar of air at 650 ˚C. Time-lapse
images paired with particle size dis-
Chapter 1

tributions to analyze Pt ripening in (I)

the same sample region consecutive-
ly irradiated with the electron beam,
and (II) unexposed sample regions.
Statistical evidence supports the
claim that the electron beam directly
impacts the Pt particle size evolution
during the process
Chapter 2

2.8.2 Deposition Energy: the change. Our intention for this section is to concentrate on
Electron and Ion Irradiation how to consider the effect of the electron beam on your in-situ
TEM experiment; we emphasize that you must not ignore this
effect – it’s incredibly important.
Electron-Beam-Induced Processes
In all of the experimental set-ups we described above, we have All in-situ experiments should be compared to the same mea-
kept in mind the effect of the electron beam on the processes surements in the absence of the electron beam, Fig. 2.53. This
being investigated. What we mean here is we must watch for will clearly define the observable electron-beam-induced effects.
any change produced by the beam that would not have occurred Generally, the goal is for noninvasive characterization of the
if the beam had not been present. This covers beam-induced in-situ processes and properties, unless beam-induced processes
changes for in-situ characterization where the key difference is are the subject of the experiment. In most cases, we are not able
that the operator is not controlling or measuring what is causing to neglect the impact of the electron beam; even if we cannot
 2.8 Adding or Removing Material 69

observe the effect of the electron beam on the specimen, any Electron dose, dose rate, and accumulated dose should be re-
exposure during imaging will cause an effect, however small ported with each image published. Collective researchers have
it may be. To image a specimen without changing it with the expressed thoughts on recording the time and dose of each

Chapter 1
electron beam, the beam energy must be below a threshold en- frame/scan over the sample to get a total accumulated dose at
ergy for atomic displacements. One example of electron-beam the point of sample in the reported image. This would be much
influence on a sample/substrate can be sputtering of atoms by easier to record for STEM imaging in comparison to TEM im-
the electron beam. A method of beam-damage detection can be aging, due to the defined scan region. R. Unocic at ORNL has
to image the same region of a specimen in series for the obser- developed an external control of the STEM probe to accurately
vation of high-resolution information disappearing in the FFT control the scan patterns and dwell times of the probe for the
with continuous irradiation. experiment. This level of beam control could also be used to
track the electron exposure to the sample throughout an experi-
How do we properly measure the electron-beam current in our ment by tracking the beam voltage, beam current, convergence
microscope? The fluorescent screen measurements reported may angle, specimen position, dwell time in STEM, unblanked beam

Chapter 2
not be properly calibrated, so it is best to calibrate the beam cur- in TEM, and search/focus/preview/acquire modes of imaging.
rent yourself. To do this, you will need a holder with a Faraday This software and system is not publicly available, though sim-
cup (Gatan Inc.), where you can plot the fluorescent screen cur- ilar software for electron-beam lithography on SEMs could be
rent vs the measured beam current with a picoammeter to create integrated for the TEM.
a correction factor for both the small and large screens (the small
screen should be more sensitive to low beam currents). A systematic study of changing the electron dose can separate
out the effects of the electron beam and generate a dose threshold
Generally, it is best to operate the TEM for in-situ experiments for damage of a given experimental set-up, to enable imaging
using the lowest electron dose possible to minimize the influ- with minimal beam-induced artifacts. Comparable modeling can
ence of the electron beam on the processes occurring within the be used to determine what damage products may be produced
TEM, as long as this does not degrade the information you ob- and to what concentration they would start to significantly im-
tain. The typical dose in a HRTEM image can be on the order of pact the experimental results. For example, parts-per-million
5 × 103 e−/Å2. Low-dose imaging used for biomaterials in vitreous concentrations of gas phase contamination can cause a difference
ice is kept below 10 e−/Å2, to prevent hydrogen gas evolution. At in catalysis research.
these low doses used in cryo-TEM of biomaterials, the contrast
in each individual image is significantly decreased, which is why Analytical methods generally require localized probing of the
the reproducibility in the structure of these materials requires specimen to acquire chemical and electronic structure infor-
researchers to collect thousands of images and build contrast mation from the region of interest, though placing a focused
by averaging over the frames. Therefore, one can figure out the probe on the specimen will locally induce damage. Techniques
relationship between electron dose with contrast and resolution of Aloof beam mode allow for probing the surface states from
under general conditions to determine the lowest beam current the region of interest without focusing the probe directly down
that can be used to retain the visibility and information in the on the sample.
collected images. Low-dose imaging also assumes that you focus
somewhere other than the region of interest, otherwise, the total
dose will include that involved in both focusing and imaging. It Experimental Effects from the Electron Beam
is quite difficult to keep track of the dose, especially since you Experimentalists have used the electron beam as an irradiation
don’t usually know ahead of time where or when something re- source to study defect formation, coalescence, annealing, sinter-
ally interesting is going to happen during the in-situ TEM exper- ing, sputtering, and atom mobility for more than 50 years. Today,
iment. Frame-averaging techniques highly degrade the temporal we can directly image vacancies and adatoms in 2D materials
resolution but also may not be feasible, depending on how much such as graphene or boron nitride. We can also restructure them
structural information is changing over time. The use of direct with electron-beam irradiation. We now routinely see surface
electron detection cameras now allows us to collect the same migration of atoms during TEM imaging, and shape changes
amount of information from the specimen with a much-reduced of nanomaterials or even Ostwald ripening. Irradiating mate-
electron dose (see Sect. 2.43). Commonly for in-situ TEM stud- rials with the electron beam may also induce structural reorga-
ies, the beam is blanked during most of the experiment, with nization: amorphous materials can crystallize and crystals can
only snapshots recorded to monitor the progress of the mecha- amorphize. The TEM is the best platform we have for irradiation
nism under study. At the end of the study, comparison between studies because we immediately see the effect.
regions of the sample that experienced electron irradiation to
those without provides evidence to the impact that imaging had Electron-beam effects are always important when you are ob-
on the processes being studied. Ideally, we would like to image serving processes or materials in gaseous or liquid environments,
the in-situ processes with minimal influence from the electron because irradiation of the environmental molecules will create
beam (low-dose imaging) with high-contrast, spatial, and tem- ionization products that may react with all other components
poral resolution. within the sample/substrate/closed cell. This can be beneficial
 70 2 In situ and Operando
Chapter 1
Chapter 2

Fig. 2.54 Electron-beam-controlled generation of aqueous electrons in STEM, by controlling the a scan pixel size and b dwell time at set beam currents and fluid
path lengths. c Bright-field STEM images of Ag nanoparticle nucleation and growth in an aqueous liquid cell by e-beam induced generation of aqueous electrons.
d Nanoparticle growth data for images in c, dotted lines represent the median nucleation induction time and median induction dose. Scale bar in c is 100 nm

when using the electron beam to stimulate a reaction, as seen in periments, what are the diffusion characteristics of these damage
the generation of aqueous electrons for the formation of metal products, and how far away from the irritated area must one
nanoparticles in a liquid cell, Fig. 2.54. The amount of ioniza- move to image an unperturbed region of the specimen? This is
tion ‘damage’ (radiolysis) will scale with the beam voltage and critical; as you watch one region of the specimen during a movie,
current; the most damage occurs when we acquire atomic-scale you may not realize that the specimen region being irradiated
images due to the high electron dose required when focusing the with the electron beam is changing differently from regions that
electrons down on small regions. Aqueous solution ionization were not irradiated, Fig. 2.53.
products are fairly well characterized, though the cascade of
processes that result is dependent on the substrate/membrane The electron beam has been known to alter the reaction rate or
window material, solutes, and other materials in contact with mechanism of an in-situ process. This has been observed in the
the aqueous environment, Fig. 2.55 (Schneider et al., 2014). CVD of Au nanoparticles within an ETEM. Ex-situ comparison
It has been suggested that graphene liquid cells can reduce the is the best means to verify the validity of the in-situ experimental
amount of ionization damage in comparison to SiN membrane results. The effect of the electron beam has shown to be de-
liquid cells. pendent on the lifetime of the precursor molecule after being
absorbed on the substrate.
Aqueous systems provide one of the most forgiving environ-
ments for ionization damage, as the products can quickly react
to reform water molecules, whereas organic solvents are likely Ion Irradiation
to react to form polymerized carbon chains, which is irrevers- As a caveat for this section, keep in mind that we might, for
ible decomposition of the solvent. In gaseous environments, the example, be implanting ions and at the same time sputtering
electron beam has been observed to impede oxidation of metal material from the specimen.
nanoparticles due to the reducing effect of electron irradiation.
In some experiments, you may be able to tailor the beam current Some researchers have been interested in irradiation of species
to balance oxidative and reductive processes. In liquid-cell ex- larger than an electron, whereby atoms may be impacted with the
 2.8 Adding or Removing Material 71

ions can be used for high numbers of atomic displacements, or

self-irradiation (with the same ions that are contained within the
TEM sample). These heavy ions will supply enough energy to

Chapter 1
create dense collision cascades, generally in the range of 0.8 to
48 MeV ion energies. Implantation of ions is used to study the
effects of changes in chemical composition by irradiation, using
lower energy (0.8–30 keV) or lighter ions. He ions can be used to
simulate alpha-particle radiation. The light elements, such as H
and He, are prone to loss of ion beam flux, and therefore, require
separate short drift tubes energies of 500 eV to 3,000 keV. Sput-
tering of atoms is observed when enough energy is transferred to
the sample to displace surface atoms from the material (Banhart,
2008). Of course, the electron beam may also serve as a source

Chapter 2
for irradiation. Therefore, generally electron irradiation during
these experiments is minimized by operating the TEM below the
incident electron threshold energy. The ions used and energy val-
ues are limited by the maximum magnetic fields of the ion-beam
directing parts, though 48-MeV Si8+ and 14-MeV Au12+ have been
accelerated into a TEM (Hattar et al. 2014). Ion beams may be
focused down to small (and decreasing) diameters (as with a FIB
column attached to a TEM), although often the whole TEM sam-
ple is irradiated with the beam.

Fig. 2.55 Deaerated water electron-beam radiolysis products. a Concentrations The set-up of the ion source/accelerator is dependent on the
as a function of time at a constant dose rate of 7.5 × 107 Gy/s. b Steady-state types of ions, energies, and fluxes needed (Hinks 2009). Nu-
concentrations as a function of dose rate. c Steady-state pH as a function of dose clear reactions involve both heavy ions as well as alpha-particle
rate from initial unexposed conditions irradiation. Therefore, multibeam systems are used to simulate
this environment. Dual ion-beam designs usually require two
ion beamlines into the microscope, as the energies and masses of
sample to observe radiation damage to materials and their change ions will differ. Commonly, the TEM resides within an ion-beam
in properties. The radiation will generally cause displacement or facility, where different instrumentation is connected to the same
ionization damage to the material of interest. If you want to do ion source about the facility. There have been instances where a
controlled studies of ion irradiation, you’ll want to use an ion FIB can be attached to the side-entry of a TEM column, to study
beam source that is interfaced directly with the specimen region structural FIB damage during microfabrication.
within the TEM. The first studies of this nature within the TEM
were conducted in the 1960s. Today, this combined instrumenta- Ion-beam experiments within a TEM require some microscope
tion is quite rare, with only two operational facilities in the USA, modification to properly direct the ion beam onto the TEM spec-
Fig. 2.56, of a dozen throughout the world. Researchers use this imen. Ports about the specimen region are used to connect the ion
technique to study defect/bubble formation, radiation-induced beamline to the TEM, where a wide pole-piece gap or modifica-
phase changes and segregation, changes in grain size, sputtering tion to the objective lens is necessary. Side-entry incorporation of
or removal of material, growth, diffusion, and collision cascades the ion-beam onto the sample requires tilting of the TEM sample
(Hinks 2009). Funding for these facilities is generally tied to to a high angle for irradiation. Consideration of shadowing from
sources interested in materials for nuclear reactors, radioactive the specimen rod is important when setting up this geometry
waste storage containers, materials used in space, and the semi- or when using a new specimen holder. In-situ sample holders
conductor industry. The effect of bombardment with energetic for temperature, environmental cells, stress, and electrical and
particles to produce atomic displacements per atom is used to magnetic property measurements can be operated within these
study the materials response. Materials processes may then be microscopes if they are compatible with the tilt geometries for
characterized under irradiative conditions, which are not simple irradiation with the ion beam. Modification to the upper objective
to replicate. lens may allow for the ion beam to be directed onto the sample
with minimal sample tilt, although imaging performance is the
Ions are used instead of neutrons for radiation testing of mate- consequence. The other TEM modification is to use electrostatic
rials, as they are easier to control and measure the energy and deflectors to reposition the beam onto the specimen. Alignment
flux for accurate reporting of the dose subjected to the specimen of the ion beam onto the specimen is not trivial, as consider-
(Hinks 2009). Ions are supplied from a plasma produced from ation of the effect of the magnetic field from the objective lens
a gas, liquid, or solid source that is then accelerated to defined must be factored in. Measurement of the ion beam current at the
energies using an ion accelerator and beamline, generally with sample position within the TEM can be completed using a Fara-
fluxes ranging from ~ 1 × 107 to 1 × 1013 ions cm−2 s−1. Heavy day cup. Ion-beam current densities generally range from 10−9 to
 72 2 In situ and Operando
Chapter 1
Chapter 2

Fig. 2.56 I3TEM facility at Sandia National Laboratories for ion irradiation studies in the TEM. This may be the reason for so few operational facilities in the world

5 × 10−5 A cm−2. This can be used to aid alignment of the ion beam Commonly these facilities were constructed using high-voltage
on the specimen. Ion energies and fluxes of the instruments are TEMs to enable imaging through thick specimens to replicate
unique to each design, therefore we refer readers to the specific the bulk nature of the material’s response. When studying de-
instrumental details for individual facilities in a review article fects in materials, consideration to the amount of free surfaces
by Hinks (2009); most of these instruments are located in Japan, is important as they can act as sinks for mobile defects (Hinks
with others scattered in France, China, UK, and the USA. An 2009). In addition, it is important to ensure that the sample is
schematic of an ion beam TEM facility is shown in Fig. 2.56. free of defects that could be misinterpreted as radiation-induced
As you can imagine, these facilities are expensive to maintain. defects. Irradiation will cause heating in the specimen. Therefore
you must consider the amount of energy being deposited in the
Image quality can be reduced by incorporation of one or multiple sample and what fraction dissipates as heat. Modeling and simu-
ion beamlines into the TEM column. But not necessarily. Imag- lation of the environmental impact on the specimen is critical in
ing specimens tilted to high angles during ion irradiation will ef- these experiments for proper characterization of the mechanisms
fectively increase the sample thickness. Therefore, the specimen dominating the sample dynamics (Vetrano et al. 1989). Com-
tilt during analysis should be considered initially when preparing monly, the radiation damage effects being studied are the for-
the sample at a specific crystal orientation or thickness. Mechan- mation of Frenkel pairs (vacancy-interstitial pairs) and cascade
ical vibration may be a concern for connection of the beamline damage, leading to structural changes in the material. Voids and
to the TEM column, where vacuum pumps may induce increased dislocation loops are visible in these in-situ TEM experiments as
noise during imaging. Many techniques can be used to dampen agglomerates of point defects. Work by Hattar et al. (2014) has
these vibrations with careful planning during instrument instal- shown self-irradiation of gold thin films with simultaneous alpha
lation, such as use of low-vibration pumps or cushioning below radiation to produce dislocation loops from a single Au cascade
the equipment to absorb vibrations. High-temporal resolution in resulting in a long dislocation line, Fig. 2.57.
these experiments would provide information on the close-pair
recombination (10−6 s), requiring UEM techniques. Generally, the
mobility of individual point defects is too high for most image
capture techniques.

Licensed to ([email protected])
 2.9 The Future 73

over the specimen and the environment around it, to obtain truly
quantitative data. An emphasis on relating what we observe in
the TEM with application space and bulk samples will always

Chapter 1
be a priority. With improved software we will be able to collect
more of the information and sort through it faster, even to the
point where the computer tells us where to look and which
events are rare. Those rare events will be studied to understand
the origin of processes such as nucleation. The electron-beam
damage will become more manageable as we develop strategies
to efficiently introduce electrons to the specimen and never let
any go undetected. Emphasis on antidrift and anticontamination
methods will be revisited, so that enhanced stability will bring
us enhanced imaging/data collection. We will more broadly

Chapter 2
exploit the increased pole-piece gap possible (> 1 cm) with
Cs-corrected instruments to fit more detectors/probes around
the specimen. The branching of in-situ TEM combined with
other in-situ techniques will provide us a broader scope for
characterizing processes within materials. Finally, we want to
have 3D information; so the future will involve 3D imaging and
spectroscopy while you carry out the operando studies. This
chapter is not a stand-alone exercise!

Fig. 2.57 a–c Selected TEM images from a video of concurrent 3.6 MeV Au6+ and
10 keV He1+ irradiation of an Au foil. d Distance of a dislocation loop (red arrow) Environment
from a grain boundary was tracked as a function of time during irradiation

Change Change
TEM

2.9 The Future

Stimuli TEM Record TEM Time

Simply put, the future is in-situ and operando TEM. As long as TEM
we can see through the specimen, and sometimes if we can’t, Change Change
the TEM allows us to control the environment and record im-
ages and spectra. So what does the future hold? The answer is
much more of everything. We will be pairing instrumentation to Temperature
create a laboratory within the pole-piece gap of the TEM. Now
that the foundations of many of these techniques have been
established, we can tailor our abilities to allow precise control
 74 2 In situ and Operando

Chapter Summary
Chapter 1

This chapter does not tell you everything about in-situ the right holder and the right TEM (refer to the list of user
and operando TEM. The purpose is to make suggestions, facilities), and the results will be publishable in the Journal
and hopefully to inspire you to try other ways of studying of Materials Science or another leading journal.
your specimen. The biggest obstacle is the usual one – the
specimen has to be prepared for the experiments (and we Microscopes with the objective lens Cs-corrected are becom-
mean experiments, plural). The TEM becomes your lab. ing much more widely available. One big advantage of Cs-cor-
This means you have to be even more careful than usual rected TEMs is the resolution, of course. For in-situ studies,
because there are many opportunities to damage the TEM. an even bigger advantage for the future will be that the pole-
You will almost invariably want to combine techniques and piece gap can be made quite large (> 1 cm) while maintaining
Chapter 2

the conditions will probably not be ideal. acceptable resolution. In the future, all state-of-the-art centers
for in-situ TEM studies will use Cs-corrected TEMs.
The key factor in whether or not you can do an experiment
in situ, is the holder. The holder can be very expensive and Generally, you’ll find that although you may not be able to
you can’t have every holder, nor should you. Maintaining obtain the data you were hoping for, you will find interesting
equipment is time consuming and expensive. But if you have new results, and you might forget what information you
the specimens, and you really know what you want to do, sought in the first place. Of course, if you get an unexpected
you will be able to collaborate with someone who does have result, you have to confirm that it is real!
 Appendix The Future 75

Appendix

Part I
People Q2.11 For environmental TEM, we can use a holder or an
ETEM. Explain the difference, suggest which you would
prefer and explain your reasons.
Gertrude Fleming Rempfer was born in Seattle in 1912 and died Q2.12 What voltage is preferred for operando studies?
on October 4, 2011, in Portland, OR, aged 99. She worked in her Q2.13 What is the challenge when you want to perform XEDS
lab with her microscopes until weeks before she died. She was in an ETEM?
working on photoemission TEMs when most people had forgot- Q2.14 For high-resolution imaging in an operando study, do you

Chapter 2
ten that they existed. See Ultramicroscopy 119, pages 1–110 to want to use STEM or TEM?
learn more of her work. Q2.15 Operando TEM is new. When were TEMs first used to
study reactions in gases?
Peter Roland Swann (February 4, 1935–July 14, 2013) made an Q2.16 What are the windows made from in modern gas or liquid
e-cell TEM holder in ~1960. Interested in straining experiments operando holders?
at controlled temperatures when most wanted one or the other! Q2.17 Why does window bowing happen when you are using a
He introduced CBC to TEM in 1970 and MGB to TEM a little “liquid” holder.
later. Q2.18 What are the advantages of using a MEMS device for
heating your specimen.
Ugo Valdrè was born on February 4, 1936 in Faenza Italy (also Q2.19 What is the BGPL?
the birthplace of faience pottery), trained as a physicist and de- Q2.20 Why do we include UHV in the discussion of environ-
signed many early holders for in-situ studies; in 1971 he wrote ment?
“… we cherish the time when, reversing the present criteria, the Q2.21 Why do we call it Klug’s approach?
electron microscope will be made to fit a predesigned universal Q2.22 Why might you want to cool your specimen? Is this op-
specimen stage containing the facilities of a laboratory for spec- erando?
imen treatment.” Prescient. Q2.23 Can you do operando experiments in specimens prepared
by plunge freezing?
Ahmed Hassan Zewail (born February 26, 1946 in Daman- Q2.24 How cold can the specimen be in cryo-TEM?
hur, Egypt, died 2 August 2016) won the 1999 Nobel Prize for
Chemistry for his work on femtochemistry and made UEM work.

Text-Specific Questions
Self-Assessment Questions
T2.1 If you were to custom build a TEM holder, what con-
siderations to the TEM are necessary to prevent damage
Q2.1 What is the pressure at the specimen in a conventional to the instrument and have a operational stage? What
TEM – the one you use most days? additional requirements would you consider based on
Q2.2 Define nanoparticle. safety?
Q2.3 What is the difference between in situ and operando? T2.2 What is adaptive threshholding?
What does ex situ TEM mean? T2.3 Discuss the state-of-the-art in drift correction, identify
Q2.4 You’ve seen several Latin terms, including those in the which method you favor (and why) and guess how this
title. Do you think in silico experiments have a place in will improve in the future.
materials research? T2.4 Compare and contrast DTEM and UEM.
Q2.5 Why are nearly all TEM made for side-entry holders? T2.5 Compare the difficulties of operando studies in gases
Which are top-entry? versus in liquids.
Q2.6 Summarize the most important features of recording the T2.6 Operando studies present a major challenge to the TEM
data. manufacturers. Explain this statement.
Q2.7 What is a direct-detection camera and why are these cam- T2.7 How do we measure the partial pressure around the spec-
eras expensive? imen in a TEM.
Q2.8 What is ImageJ and who provides it? T2.8 How much energy is needed to heat a 3-mm TEM spec-
Q2.9 What is DTEM and what is UEM? imen to 500 °C? Be sure to state your assumptions about
Q2.10 How many electrons do you need to produce a DP? the specimen geometry.
 76 2 In situ and Operando

T2.9 How much energy must be removed to change the tem- Gai PL (ed) (1997) In-situ Microscopy in Materials Research:
perature of a 1-µm-long nanowire of Si. Leading International Research in Electron and Scanning
T2.10 Discuss window bowing when you are using a liquid-cell Probe Microscopies. Springer, NY (You need to have access
Chapter 1

holder and what to do about it. to this book)

T2.11 Consider the schematic in Fig. 2.12. Why is there a pres- Gai PL, Sharma R, Ross FM (2008) Environmental (S)TEM
sure gap? Studies of Gas-Liquid-Solid Interations under Reaction Con-
T2.12 Discuss the reliability and accuracy of temperature mea- ditions. MRS Bull 33:107–114
surement in the TEM. Panciera F, Chou Y-C, Reuter MC, Zakharov D, Stach EA,
T2.13 Discuss how you would use the nanothermometer shown Hofmann S, Ross FM (2015) Synthesis of nanostructures in
in Fig. 2.25. nanowires using sequential catalyst reactions. Nature Mate-
T2.14 Discuss how to stop the reaction and dewetting shown in rials 14:820–826
Fig. 2.26. Poppa H (2004) High resolution, high speed ultrahigh vacuum
T2.15 Discuss the artifacts that might occur in the experiment microscopy. J Vac Sci Technol A 22(5):1931–1947 (A nice
Chapter 2

illustrated in Fig. 2.36. historical overview explaining the development of UHV)

T2.16 Compare the pros and cons of the crucible and fila- Robertson IM, Ferreira PJ, Dehm G, Hull R, Stach EA (2008)
ment-wire heating stages. Visualizing the behavior of dislocations – seeing is believing.
T2.17 Discuss the advantages and disadvantages of using dog- MRS Bulletin 33:122–131
bone specimens versus MEMS specimens for studying Rudolph AR, Jungjohann KL, Wheeler DR, Brozik SM (2014)
mechanical behavior. Drying Effect Creates False Assemblies in DNA-coated Gold
T2.18 What is a comb-drive electrostatic actuator and how does Nanoparticles as Determined Through In Situ Liquid Cell
it function? STEM. Microscopy and Microanalysis 20:437–444
T2.19 Summarize with bullets the challenges and consider- Sharma R, Gai PL, Gajdardziska-Josifovska M, Sinclair R, Whit-
ations you must keep in mind when studying mechanical man L (eds) (1996) In situ electron and tunneling microscopy
properties in the TEM. of dynamic processes. MRS Symp Proc, vol. 404.
T2.20 Explain the contrast seen in Fig. 2.41. Spitzer M, Wildenhain J, Rappsilber J, Tyers M (2014) Box-
T2.21 Use bullets to summarize the pros and cons of studying PlotR: a web tool for generation of box plots. Nature Methods
mechanical properties using the method illustrated in 11:121–122
Fig. 2.39. van Tendeloo G, van Dyck D, Pennycook SJ (eds) (2012) Hand-
T2.22 Explain the contrast that we see in Fig. 2.43. book of Nanoscopy. Wiley-VCH, Weinheim
T2.23 Why is the contrast in Fig. 2.47 in the lithiation of Ge so Weckhuysen BM (2002) Chem Commun 2:97–110 (See ref. 11
different from the Ge–Si core–shell structure? for the origin of ‘operando’)
T2.24 Using the literature describe five other systems that have Yao N, Wang ZL (eds) (2005) Handbook of Microscopy for
used the geometry shown in Fig. 2.52 to study the growth Nanotechnology. Kluwer Academic Publishers., Boston, USA
of nanowires. Zaluzec NJ, Burke MG, Haigh SJ, Kulzick MA (2014) X-ray
energy-dispersive spectrometry during in-situ liquid cell stud-
ies using an analytical electron microscope. Micro Microanal
20:323–329
References Zewail AH, Thomas JM (2009) 4D Electron Microscopy: Im-
aging in Space and Time. Imperial College Press, London
(See section 2.4.3. The book is co-authored by a Nobel Prize
We’ll list a selection of papers and put others on the web but you winner and a leader in TEM for solid-state chemistry. The
can see with a quick search on ‘in-situ’ plus ‘TEM’ on Web of fourth D is time – hence it’s DTEM. Essential reading if your
Science – which all research Universities have – that the number emphasis is femtosecond spectroscopy)
of papers is growing rapidly.

Specific References – In-Situ TEM Collections

General References
Journal of Materials Research Vol 20(7) 2005.
Banhart F (ed) (2008) In situ Electron Microscopy at High Res- Microscopy and Microanalysis, Vol 7(6), 2001.
olution. World Scientific Publishers, Hackensack, NJ USA Microscopy and Microanalysis, Vol 20(2), 2014. TEM in Liq-
Butler EP, Hale KF (1981) Dynamic Experiments in the Electron uids.
Microscope in Practical Methods in Electron Microscopy vol. Microscopy Research and Technique, Vol 72, 2009.
9. North-Holland, Amsterdam (Aging but still useful) MRS Bulletin, Vol 33(2), 2009.
Dehm G, Howe JM, Zweck J (eds) (2012) In-situ Electron Mi-
croscopy; Applications in Physics, Chemistry and Materials
Science. Wiley-VCH, Weinheim (Check if the chapters cover
your interest)
 Appendix References 77

ETEM dislocation-grain boundary interactions: a tribute to David

Brandon. J Mater Sci 41:7704–7719
Boyes ED, Gai PL (1997) Environmental high resolution elec- Espinosa HD, Bernal RA, Filleter T (2012) In Situ TEM Elec-

Chapter 1
tron microscopy and applications to chemical science. Ultra- tromechanical Testing of Nanowires and Nanotubes. Small
microscopy 67:219–232 (ETEM) 8(21):3233–3252
Boyes ED, Gai PL (2014a) Aberration corrected environmental Han JN, Saif MTA (2006) In situ microtensile stage for electro-
STEM (AC ESTEM) for dynamic in-situ gas reaction stud- mechanical characterization of nanoscale freestanding films.
ies of nanoparticle catalysts. Journal of Physics: Conference Rev Sci Instrum 77:045102
Series 522:012004 Nafari A, Angenete J, Svensson K, Sanz-Velasco A, Olinm H
Boyes ED, Gai PL (2014b) Visualising reacting single atoms (2011) Scanning Probe Microscopy and Transmission Elec-
under controlled conditions: Advances in atomic resolution tron Microscopy. Nanoscience and Nanotechnology 2:59–99
in situ Environmental (Scanning) Transmission Electron Mi- (NanoScience and Technology, Part 1)
croscopy. C R Physique 15:200–213 Stach EA, Freeman T, Minor AM, Owen DK, Cumings J, Wall

Chapter 2
Crozier PA, Chenna S (2011) In Situ Analysis of Gas Compo- MA, Chraska T, Hull R, Morris JM Jr, Zettl A, Dahmen U
sition by Electron Energy-Loss Spectroscopy for Environ- (2001) Development of a Nanoindenter for In Situ Trans-
mental Transmission Electron Microscopy. Ultramicroscopy mission Electron Microscopy. Microsc Microanal 7:507–517
111:177–185 (Measuring catalysis by operando TEM) Treacy MMJ, Ebbesen TW, Gibson JM (1996) Exceptionally
Gai PL, Calvino JJ (2005) Electron microscopy in the catalysis high Young’s modulus observed for carbon nanotubes. Nature
of alkane oxidation, environmental control, and alternative 381:678–680
energy sources. Ann Rev Mat Sci 35:465–504 Zhu Y, Espinosa HD (2005) An electromechanical material test-
Gai PL (2002) Development of Wet Environmental TEM (WET- ing system for in situ electron microscopy and applications.
ETEM) for In Situ Studies of Liquid-Catalyst Reactions on Proc Natl Acad Sci USA 102:14503–14508
the Nanoscale. Microsc Microanal 8:21–28 (Specially de-
signed holders)
Hansen PL, Wagner JB, Helveg S, Rostrup-Nielsen JR, Clausen DTEM and UEM
BS, Topsøe H (2011) Atom-Resolved Imaging of Dynamic
Shape Changes in Supported Copper Nanocrystals. Science Bostanjoglo O, Elschner R, Mao Z, Nink T, Weingärtner M
295:2053–2055 (2000) Nanosecond electron microscopes. Ultramicroscopy
Hansen TW, Wagner JB, Hansen PL, Dahl S, Topsøe H, Jacobsen 81:141–147
CJH (2001) Atomic-resolution in-situ transmission electron Deponte DP, Mckeown JT, Weierstall U, Doak RB, Spence JCH
microscopy of a promoter of a heterogeneous catalyst. Sci- (2011) Towards ETEM serial crystallography: Electron dif-
ence 16:294 (1508) fraction from liquid jets. Ultramicrosc 111:824–827
Hansen TW, Wagner JB, Dunin-Borkowski RE (2010) Aber- Grinolds MS, Lobastov VA, Weissenrieder J, Zewail AH (2006)
ration corrected and monochromated environmental trans- Four-dimensional ultrafast electron microscopy of phase tran-
mission electron microscopy: challenges and prospects for sitions. Proc Nat Acad Sci 103(49):18427–18431
materials science. Mater Sci Tech Ser 26(11):1338–1344 LaGrange T, Reed BW, Santala MK, McKeown JT, Kulovits A,
Haque MA, Saif MTA (2002) In-situ tensile testing of nano-scale Wiezorek JMK, Nikolovac L, Roseic F, Siwick BJ, Camp-
specimens in SEM and TEM. Experimental Mech 42(1):123– bell GH (2012) Approaches for ultrafast imaging of transient
128 materials processes in the transmission electron microscope.
Jinschek JR, Helveg S (2012) Image resolution and sensitivity in Micron 43:1108–1120
an environmental transmission electron microscope. Micron Lobastov VA, Srinivasan R, Zewail AH (2005) Four-dimen-
43:1156–1168 sional ultrafast electron microscopy. Proc Nat Acad Sci
Mortensen PM, Hasen TW, Wagner JB, Jensen AD (2015) Mod- 102(20):7069–7073
eling of temperature profiles in an environmental transmis- Vanacore GM, van der Veen RM, Zewail AH (2015) Origin
sion electron microscope using computational fluid dynamics. of Axial and Radial Expansions in Carbon Nanotubes Re-
Ultramicroscopy 152:1–9 vealed by Ultrafast Diffraction and Spectroscopy. ACS Nano
Wu Y, Yang P (2001) Direct observation of vapor-liquid-solid 9(2):1721–1729
nanowire growth. J Am Chem Soc 123:3165–3166 Zewail AH (2010) Four-Dimensional Electron Microscopy. Sci-
ence 328(5975):187–193

Deformation
Temperature Control
Carlton CE, Ferreira PJ (2012) In situ TEM nanoindentation of
nanoparticles. Micron 43:1134–1139 Adrian M, Dubochet J, Fuller SD, Harris JR (1998) Cryo-nega-
De Hosson JTM, Soer WA, Minor AM, Shan AZ, Stach EA, Asif tive Staining. Micron 29:145–160
SAS, Warren OL (2006) In situ TEM nanoindentation and
 78 2 In situ and Operando

Asoro MA, Kovar D, Ferreira PJ (2013) In Situ Transmission Saka H, Kamino T, Arai S, Sasaki K (2008) In Situ Heating
Electron Microscopy Observations of Sublimation in Silver Transmission Electron Microscopy. MRS Bull 33:93–100
Nanoparticles. ACS Nano 7(9):7844–7852 (Nice description of the filament heating holder)
Chapter 1

Bellare JR, Davis HT, Scriven LE, Talmon Y (1988) Controlled Sinclair R, Yamashita T, Ponce FA (1981) Atomic Motion on
Environment Vitrification System: An Improved Sample the Surface of a Cadmium Telluride Single Crystal. Nature
Preparation Technique. J Electron Micr Tech 10:87–111 290:386–388
Chou HT, Evans JE, Stahlberg H (2007) In: Kuo J (ed) Elec- Tai K, Liy Y, Dillon SJ (2014) In Situ Cryogenic Transmission
tron Microscopy Methods and Protocols. Humana Press Inc., Electron Microscopy for Characterizing the Evolution of So-
Totowa, NJ, pp 331–343 lidifying Water Ice in Colloidal Systems. Microsc Microanal
Delalande M, Guinel MJ, Allard LF, Delattre A, Le Bris R, Sam- 20:330–337
son Y, Bayle-Guillemaud P, Reiss P (2012) L10 Ordering Taylor KA, Glaeser RM (1974) Electron Diffraction of Frozen,
of Ultrasmall FePt Nanoparticles Revealed by TEM In Situ Hydrated Protein Crystals. Science 186:1036–1037
Annealing. J Phys Chem C 116:6866–6872 van Heel M, Gowen B, Matadeen R, Orlova EV, Finn R, Pape
Chapter 2

Fujiyoshi Y, Mizusaki T, Morikawa K, Yamagishi H, Aoki Y, T, Cohen D, Stark H, Schmidt R, Schatz M, Patwardhan A
Kihara H, Harada Y (1991) Development of a Superfluid He- (2000) Single-Particle Electron Cryo-Microscopy: Towards
lium Stage for High-Resolution Electron Microscopy. Ultra- Atomic Resolution. Q Rev Biophys 33(4):307–369
microscopy 38:241–251 (See section 2.6.3)
Gai PL, Boyes ED (2009) Advances in Atomic Resolution In
Situ Environmental Transmission Electron Microscopy and Magnetic Fields (section 2.7.2)
1Å Aberration Corrected In Situ Electron Microscopy. Mi-
crosc Res Techn 72:153–164 Arita M, Tokuda R, Hamada K, Takahashi Y (2014) Development
Gao Y, Bando Y (2002) Carbon Nanothermometer Containing of TEM Holder Generating In-Plane Magnetic Field Used for
Gallium. Nature 415:599 In-Situ TEM Observation. Mater Transact 55(3):403–409
Kasama T, Dunin-Borkowski RE, Matsuya L, Broom RF, Twitch- Budruk A, Phatak C, Petford-Long AK, De Graef M (2011) In
ett AC, Midgley PA, Newcomb SB, Robins AC, Smith DW, situ Lorentz TEM magnetization studies on a Fe-Pd-Co mar-
Gronsky JJ, Thomas CA, Fischione PE (2005) A versatile tensitic alloy. Acta Mater 59:6646–6657
three-contact electrical biasing transmission electron micro- Budruk A, Phatak C, Petford-Long AK, De Graef M (2011) In
scope specimen holder for electron holography and electron situ Lorentz TEM magnetization study of a Ni-Mn-Ga fer-
tomography of working devices. In: Ferreira PJ, Robertson romagnetic shape memory alloy. Acta Mater 59:4895–4906
IM, Dehm G, Saka H (eds) In-situ Electron Microscopy of Inoue M, Tomita T, Naruse M, Akase Z, Murakami Y, Shindo D
Materials. Mater. Res. Soc. Symp. Proc. 907E, vol MM13.02. (2005) Development of a magnetizing stage for in situ obser-
Li YB, Bando Y, Golberg D, Liu ZW (2003) Ga-Filled Single vations with electron holography and Lorentz microscopy. J
Crystalline MgO Nanotube: Wide-Temperature Range Nan- Electron Microsc 54(6):509–513
othermometer. Appl Phys Lett 83(5):999–1001 Lau JW, Schofield MA, Zhu Y (2007) A straightforward speci-
Mecklenburg M, Hubbard WA, White ER, Dhall R, Cronin SB, men holder modification for remnant magnetic-field measure-
Aloni S, Regan BC (2015) Nanoscale temperature mapping in ment in TEM. Ultramicroscopy 107:396–400
operating microelectronic devices. Science 347(6222):629– Lau JW, Schofield MA, Zhu Y, Neumarl GF (2003) In-situ Mag-
632 netodynamic Experiments Achieved with the Design of an In-
Mortensen PM, Hansen TW, Wagner JB, Jansen AD (2015) plane Magnetic Field Specimen Holder. Microsc Microanal
Modeling of temperature profiles in an environmental trans- 9(Suppl 2):130–131
mission electronmicroscope using computational fluid dy- Uhlig T, Heumann M, Zweck J (2003) Development of a spec-
namics. Ultramicroscopy 152:1–9 (Measuring T in a TEM imen holder for in situ generation of pure in-plane magnetic
with gas around the specimen) fields in a transmission electron microscope. Ultramicroscopy
Oh SH, Kauffmann Y, Scheu C, Kaplan WD, Rühle M (2005) 94(3–4):193–196 (Beam deflection using the Lorentz force;
Ordered Liquid Aluminum at the Interface with Sapphire. using an in-situ magnetization holder)
Science 310:661–663 Yi G, Nicholson WAP, Lim CK, Chapman JN, McVitie S,
Reguer A, Bedu F, Nitsche S, Chaudanson D, Detailleur B, Dal- Wilkinson CDW (2004) A new design of specimen stage for
laporta H (2009) Probing the Local Temperature by In Situ in situ magnetising experiments in the transmission electron
Electron Microscopy on a Heated Si3N4 Membrane. Ultrami- microscope. Ultramicroscopy 99:65–72
croscopy 110:61–66 (Fast ramping and very high tempera- Zweck J (2012) Lorentz Microscopy. In: Dehm G, Howe JM,
tures) Zweck J (eds) In-Situ Electron Microscopy: Applications in
Renault L, Chou HT, Chiu PL, Hill RM, Zeng X, Gipson B, Physics, Chemistry and Materials Science. Wiley-VCH Ver-
Zhang ZY, Cheng A, Unger V, Stahlberg H (2006) Mile- lag & Co KGaA, Weinheim, Germany (Specific numbers for
stones in Electron Crystallography. J Comput Aided Mol the field in a 300 kV TEM)
Des 20:519–527
 Appendix References 79

Electric Fields (section 2.7.3) TEM specimen holder equipped with a piezodriving probe
and a laser irradiation port. J Elect Microsc 58(4):245–249
Bonifacio CS, Rufner JF, Holland TB, van Benthem K (2012) Yoshida K, Yamasaki J, Tanaka N (2004) In situ high-resolution

Chapter 1
In situ transmission electron microscopy study of dielectric transmission electron microscopy observation of photode-
breakdown of surface oxides during electric field-assisted sin- composition process of poly-hydrocarbons on catalytic TiO
tering of nickel nanoparticles. Appl Phys Lett 101(5):093107– 2 films. Appl Phys Lett 84(14):2542–2544
093101 Yoshida K, Nozaki T, Hirayama T, Tanaka N (2007) In situ
Costas PMFJ, Golberg D, Shen G, Mitome M, Bando Y (2008) high-resolution transmission electron microscopy of photo-
ZnO low-dimensional structures: electrical properties mea- catalytic reactions by excited electrons in ionic liquid. J Elect
sured inside a transmission electron microscope. J Mater Sci Microsc 56(5):177–180
43:1460–1470 Zagonel LF, Mazzucco S, Tencé M, March K, Bernard R, Laslier
Dong LX, Shou KY, Frutiger DR, Subramanian A, Zhang L, Tao B, Jacopin G, Tchernycheva M, Rigutti L, Julien FH, Song-
XL, Zhang XB (2008) Engineering multiwalled carbon nano- muang R, Kociak M (2010) Nanometer Scale Spectral Imag-

Chapter 2
tubes inside a transmission electron microscope using nano- ing of Quantum Emitters in Nanowires and Its Correlation to
robotic manipulation. IEEE Trans Nanotech 7(4):508–517 Their Atomically Resolved Structure. Nano Lett 11:568–573
Golberg D, Costa PMFJ, Mitome M, Bando Y (2009) Proper- Zhang L, Miller BK, Crozier PA (2013) Atomic Level In Situ
ties and engineering of individual inorganic nanotubes in a Observation of Surface Amorphization in Anatase Nano-
transmission electron microscope. J Mater Chem 19:909–920 crystals During Light Irradiation in Water Vapor. Nano Lett
Kizuka T, Ohmi H, Sumi T, Kumazawa K, Deguchi S, Naruse 13:679–684
M, Fujisawa S, Sasaki S, Yabe A, Enomoto Y (2001) Simul- Zhang C, Tian W, Xu Z, Wang X, Liu J, Li S-L, Tang D-M, Liu
taneous observation of millisecond dynamics in atomistic D, Liao M, Bando Y, Golberg D (2014) Photosensing perfor-
structure, force and conductance on the basis of transmission mance of branched CdS/ZnO heterostructures as revealed by
electron microscopy. Japanese J Appl Phys 40:L170–L173 in situ TEM and photodetector tests. Nanoscale 6:8084–8090
Liu XH, Huang JY (2011) In situ TEM electrochemistry of anode
materials in lithium ion batteries, Energy Environ. Sci 4:3844
Lopes M, Toury T, de La Chapelle ML, Bonaccorso F, Gucciardi Electron and Ion Irradiation
PG (2013) Fast and reliable fabrication of gold tips with sub-
50 nm radius of curvature for tip-enhanced Raman spectros- Hattar K, Bufford DC, Buller DL (2014) Concurrent in situ ion
copy. Rev Sci Instrum 84:073702 irradiation transmission electron microscope. Nucl Instrum
Nafari A, Angenete J, Svensson K, Sanz-Velasco A, Olinm H Meth B 338:56–65
(2011) Scanning Probe Microscopy and Transmission Elec- Hinks JA (2009) A review of transmission electron microscopes
tron Microscopy. Nanoscience and Nanotechnology 2:59–99 with in situ ion irradiation. Nucl Instrum Meth B 267:3652–
(NanoScience and Technology, Part 1) 3662
Spence JCH (1988) A scanning tunneling microscope in a Schneider MN, Norton MM, Mendel BJ, Grogan JM, Ross FM,
side-entry holder for reflection electron-microscopy in the Bau HH (2014) Electron–Water Interactions and Implica-
Philips EM400. Ultramicroscopy 25:165–170 tions for Liquid Cell Electron Microscopy. J Phys Chem C
Spence JCH, Lo W, Kuwabara M (1990) Observation of the 118(38):22373–22382
graphite surface by reflection electron microscopy during Vetrano JS, Bench MW, Robertson IMKMA, Kirk MA (1989)
STM operation. Ultramicroscopy 33:69–82 In Situ Studies of Ion Irradiation Effects in an Electron Mi-
croscope. Metall Trans A 20:2673–2680

Optical Effects
Image Capture Software
Cavalca F, Laursen AB, Kardynal BE, Dunin-Borkowski RE,
Dahl S, Wagner JB, Hansen TW (2012) In situ transmission Camtasia allows you to record the images on your monitor as
electron microscopy of light-induced photocatalytic reac- you see them with 1,024-by-1,024 resolution. Versions for
tions. Nanotechnology 23:(075705-1-6) Mac and PC are available. http://www.techsmith.com/cam-
Miller BK, Crozier PA (2013) System for In Situ UV-Visible tasia.html
Illumination of Environmental Transmission Electron Mi- VirtualDub is a video capture and processing utility that works
croscopy Samples. Microsc Microanal 19:461–469 on the PC that came with your TEM. It runs on both 32-bit
Picher M, Mazzucco S, Blankenship S, Sharma R (2015) Vi- and 64-bit Windows platforms, but sadly not on a Mac. http://
brational and optical spectroscopies integrated with environ- www.virtualdub.org
mental transmission electron microscopy. Ultramicroscopy Commercial programs like VideoStudio Pro X8, or the Apple
150:10–15 equivalent, can be used to polish your videos for the supple-
Shindo D, Takahashi K, Murakami Y, Yamazaki K, Deguchi S, mentary material section. www.videostudiopro.com
Suga H, Kondo Y (2009) Development of a multifunctional Spacetime is a code that can link data from an external stimulus
to the corresponding images in time. spacetime.uithetblauw.nl
 80 2 In situ and Operando

Different research groups are willing to share their own codes User Facilities
for image analysis, Lewys Jones included. http://www.lewys-
jones.com https://nsrcportal.sandia.gov Nanoscale Science Research Cen-
Chapter 1

ters: source of in-situ TEM expertise and equipment through

accepted peer-reviewed user proposals
Companies http://www.nist.gov/cnst/index.cfm Center for Nanoscale Sci-
ence and Technology: in-situ TEM and Raman-ETEM
http://www.attolight.com Manufactures cathodoluminescence https://www.emsl.pnl.gov/emslweb/ Environmental Molecu-
detectors for S/TEM. Accessed in August 2015 lar Sciences Laboratory: in-situ TEM and DTEM available
http://www.comsol.com Multiphysics modeling software. Ac- through accepted peer-reviewed user proposals
cessed in August 2015 http://le-csss.asu.edu LeRoy Eyring Center for Solid State Sci-
http://www.denssolutions.com Solutions for in-situ microscopy: ence: in-situ TEM, ETEM and optical properties
Stages for MEMS heating, gas and liquid cells. Accessed in http://www.mrl.illinois.edu/facilities/center-microanalysis-ma-
Chapter 2

August 2015 terials Center for Microanalysis of Materials: high-pressure

http://www.directelectron.com Producing high-speed and high environmental cell TEM
DQE cameras for TEM. Accessed in August 2015
http://www.expresslo.com EXpressLO LLC: Quick FIB sample
lift out and grid placement using visible optics. Accessed in
August 2015
http://www.fei.com Producing ETEMs, temperature control
stages, and former Nanofactory STM-TEM and Nanoinden-
tation stage designs. Accessed in August 2015
http://www.fischione.com/products/holders Tomography holders
and a heated gas stage. Accessed in August 2015
http://www.gatan.com/products/tem-specimen-holders Stages
for temperature control, vacuum transfer, straining, and to-
mography. Producing high-frame-rate direct-electron-de-
tection cameras, in addition to software and spectrometers.
Accessed in August 2015
http://www.hitachi-hightech.com E-TEM products: MEMS heat-
ing gas injection, bulk heating gas injection, MEMS biasing,
liquid-cell electrochemistry/static and flow, and zone holders.
Accessed in August 2015
http://www.hummingbirdscientific.com Hummingbird Scientific
designs and manufactures pioneering in-situ TEM specimen
holders that enable electrical probing, sample heating, cooling
and magnetizing, tomographic imaging, nano-manipulation
and environmental TEM in liquids and gases. Accessed in
August 2015
http://www.hysitron.com/products/pi-series Quantitative nano-
mechanics stages. The design for JEOL and FEI instruments
have an important difference, which results from the diameter
of the rod, and which we leave you to discover. Accessed in
August 2015
http://www.IDES-inc.com Retrofitting TEMs for ultrafast imag-
ing. Accessed in August 2015
http://www.nanofactory-user-group.org NanoFactory is now
owned by FEI; this group supports users of non-FEI equip-
ment. Accessed in August 2015
http://www.ni.com/labview System design software, graphical
programming to code engineering systems. Accessed in Au-
gust 2015
http://www.protochips.com Protochips provides complete in-
situ TEM systems including Atmosphere for gas control and
heating, Poseidon for liquid and electrochemistry, and Aduro
for heating and electrical analysis. Accessed in August 2015
 Electron Diffraction
3

Part I
and Phase Identification

Chapter Preview

A unique strength of TEM is its ability to obtain crystallo- lographic orientation between the different phases. These
graphic data from specific locations in electron-transparent crystallographic relationships often determine the properties
specimens. As described in W&C, Chaps. 16–21, we can of alloys (such as their mechanical behavior and their cor-
accomplish this via several techniques: selected-area dif- rosion behavior) and so this is very useful information. We
fraction (SAD), Kikuchi patterns, convergent-beam electron usually describe such a crystallographic orientation is terms
diffraction (CBED), and microdiffraction. Despite being the of an orientation relationship (OR) which identifies parallel
oldest of these techniques, SAD is still probably the most planes and parallel directions within each crystalline phase.
commonly used technique in the TEM for structural identifi-
cation. We govern the area selected for generation of the dif- In this chapter we’ll take you through the various steps nec-
fraction pattern by our choice of the field-limiting (or select- essary to measure these ORs, using classic examples, often

Chapter 3
ed-area) aperture. Microdiffraction and CBED techniques from some decades ago when this approach was first devel-
permit you to target specific locations within your specimen oped. Consequently, the examples are generally from metals
for analysis, and so you can (generally) avoid contributions and alloys, because such specimens were easy to thin and,
to your patterns from an adjacent phase or the parent/matrix at that time, constituted the vast majority of engineering
phase (in the case of an embedded particle). materials – as in fact they still do! Nevertheless, you can
use many of the techniques described in the chapter in the
However, you’ll frequently want to obtain diffraction infor- study of other materials such as ceramics, semiconductors,
mation from both the feature of interest and its surrounding and composites, as a quick search through the literature will
matrix or adjacent phase so that you can assess the crystal- confirm.

© Springer International Publishing Switzerland 2016

C. B. Carter, D. B. Williams (Eds.), Transmission Electron Microscopy, DOI 10.1007/978-3-319-26651-0_3 81
 82 3 Electron Diffraction and Phase Identification

3.1 Introduction
Part I

In many materials systems, particularly in metals and in alloy

systems, several different phases can be present. This is espe-
cially true for engineering materials, such as steels and super-
alloys, wherein we obtain the desired properties from the pres-
ence, the distribution, and the specific relative orientations of the
various second phases. Although we can identify the structure
of phases through single-crystal diffraction patterns (e.g., via
CBED or microdiffraction), the OR relating the phases can also
provide very useful information. For example we can carry out
phase identification and discern the distribution of phases within
the microstructure, as well as the sequence of formation.

You saw in Chap. 18 of W&C that the ORs between phases (par-
ticularly between a parent/matrix and a second phase/transforma-
tion product) are frequently the way that we can gain evidence to
support phase identification and to understand the mechanisms
of phase formation. When we characterize materials it is often
important for us not only to identify constituent phases, but also
to assess the size and distribution of the phases. When the phases
of interest exhibit ORs with either the matrix or parent phase,
Chapter 3

we can easily form diffraction-contrast images of all (or most)

of the second phases (or precipitates) that share the common
orientation.

For crystalline materials, the interface between crystals/grains

or between two phases can be incoherent (the crystal planes do
not ‘match’), semicoherent (some ‘matching’ of crystal planes
between phases with mismatch taken up by dislocations), or co-
herent (‘matching’ of crystal planes). For semicoherent interfaces
and some coherent interfaces, the orientation of the two phases
will be influenced by the crystal structure of each phase and ther-
modynamic considerations. By analyzing the crystallographic
orientation of each phase using electron diffraction, it is then pos-
sible to identify parallel crystallographic directions and planes
Fig. 3.1 a High-resolution TEM image of two ordered precipitates in a matrix.
in the two phases. An example is shown in the high-resolution
Note that the {100} planes of the matrix are parallel to a set of planes in the
TEM image in Fig. 3.1, in which the ‘a’ set of lattice planes in
precipitate. b More detailed view of ordered precipitate and matrix
precipitate 1 exhibits good ‘matching’ with the parallel lattice
planes in the matrix phase. Similarly, precipitate 2 (oriented at
90° to precipitate 1) has planes parallel to a set of matrix planes. A Great Rule
So, these precipitates have an orientation relationship with the
matrix phase.
Make life as simple as possible while you’re at the TEM!
To begin, there are some points you should keep in mind. When Tilt the specimen so that at least one grain is at a low-in-
characterizing a microstructure containing a second phase, it is dex zone axis. 9
best if you start by orienting your specimen to obtain low-in-
dex, zone-axis patterns (such as <001>, <110>, <111> for cubic
structures or Œ0001; Œ1011; N for hexagonal structures).
N Œ1120
Generally, you’ll find that the matrix and second phase will
share a rational plane, and this will be most obvious if you ex-
amine low-index zones in the matrix phase. In particular, if the
second phase has multiple crystallographic variants, examining
the microstructure using higher-index zones may give you the
impression that the second phase is randomly distributed, which
is not the case.

Licensed to ([email protected])
 3.2 Spinodal Alloys 83

The low-index matrix patterns will guide you in your analysis and then age the alloys within this gap they will spinodally de-
of the second phase. Also, for the analysis of interface planes or compose into a triaxially modulated two-phase structure via the
boundary planes, you may be able to tilt your specimen so that development of composition waves/fluctuations along <100>

Part I
the interface or boundary of interest is ‘edge-on’, i.e., parallel directions. The two phases that compose this triaxially modu-
to the incident beam. Then, by taking DPs from each adjacent lated microstructure are disordered Fe-rich α (A2 structure) and
region, and one from the boundary or interface region, you may the coherent B2-ordered Be-enriched phase (Miller et al. 1985).
be able to identify the crystallographic direction common to both What is most interesting about these alloys is that spinodal de-
regions, and determine the plane of the phase boundary. The composition is accompanied by very significant increases in
habit plane (or facet plane) between two phases is the crystallo- hardness. If we carry out prolonged aging near the top of the
graphic plane that is the interface between two phases, such as metastable miscibility gap, another metastable phase forms. This
the long straight interface of a plate-like precipitate in a matrix. phase was identified by TEM and atom probe field-ion micros-
With this information, you are well on your way to identifying copy as B32-ordered FeBe (Miller et al. 1984).
the OR between those regions.
In this section, we will be dealing with the disordered A2 (bcc)
In this chapter, we emphasize ORs in alloys, why they are im- structure and two ordered structures that are bcc-derivatives:
portant, and how you can use TEM and electron diffraction to B2 and B32. Both the B2- and B32-ordered phases are coherent
identify these relationships. In particular, we will use examples with the bcc ferrite. The OR of interest for these cubic structures
of phase transformations and precipitation to illustrate selected is:
ORs. We will include ORs between the parent and ordered
phases in Fe–base and Ni–base alloys, matrix-precipitate ORs
h001imatrix == h001ippt and f100gmatrix == f100gppt (3.1)
in Ni–base alloys and steels, and a commonly encountered situ-
ation, if you are working in ferrous alloys, namely, the presence This is also known as the ‘cube-on-cube’ OR. You’ll see that
of Fe3O4 on the surface of low-alloy steel foils. Although this all ORs are given in a similar form whereby specific low-index

Chapter 3
chapter concentrates on alloys, it is important to remember that crystallographic directions and crystal planes in the one phase
other materials, such as ceramics and semiconductors, can also are parallel (or close to parallel) to directions and planes in the
exhibit orientation relationships between phases. other phase.

3.2 Spinodal Alloys 3.2.1 Example: Ordered FeBe Phases

and A2 Matrix
What is a spinodal alloy? It is one in which the strengthening
phase microstructure is created by uphill diffusion, without the To illustrate the benefit of utilizing the OR to evaluate the mi-
need for undercooling to drive a specific nucleation process. The crostructure, take a look the bright-field TEM image of the
microstructures are very fine scale (down to nanometers) and are spinodally decomposed Fe-25 at.% Be alloy in Fig. 3.2a. This
remarkably uniform, as any book on phase transformations will image provides us with little information as to the nature of the
describe (e.g., Porter et al. 2009). We often find that the spinodal aged microstructure. However, if we take an SAD pattern from
process results in an ordered second phase. We are interested in this specimen (Fig. 3.2b) we get significant information. Firstly,
this phase transformation because such spinodal alloys are the you can see faint B2 superlattice reflections in the {½ 00} lo-
simplest for examining ORs since, by definition, the new phase cations in this matrix [001] zone axis pattern. The B2 structure
shares the crystallographic orientation with the original parent is developed by the atomic rearrangement of the Be atoms so
phase. After much research (mainly via TEM) we know that the that they are located at the center of what can be described as
resultant two-phase microstructure is formed by local atomic a bcc unit cell. This rearrangement results in a change in the
ordering and chemical (elemental) separation. Such phase sepa- structure factor of the unit cell so that you can now see the
ration develops via long-range, low-amplitude composition fluc- (originally forbidden) reflections in the SAD pattern. If we carry
tuations as opposed to classical, highly localized (short-range) out a detailed examination of the fundamental reflections we
nucleation and growth. Over the years, fundamental research in can identify the presence of satellite reflections with streaking
the area of phase transformations has given us information on along the [001] directions, as you can see in Fig. 3.2c. We can
the mechanism of formation of these spinodally decomposed and measure the separation of the satellites from the fundamental
ordered structures. Importantly, the ordered phases can dramat- reflection and thus determine the wavelength of the composition
ically affect many useful properties of the alloy. fluctuations according to the equation developed by Daniel and
Lipson (1943). If we use one of the B2 superlattice reflections to
We’ll focus on the Fe–Be alloy system because we know there form a centered dark-field image, the B2-ordered regions in the
are several ordered structures in the Fe–Be phase diagram. The microstructure will image brightly. With this dark-field image,
Fe–Be alloy system also contains a metastable miscibility gap we can clearly show the triaxial modulations in the microstruc-
in the composition range ~ 15–50 at.% Be. If we first anneal ture as shown in Fig. 3.2e.
 84 3 Electron Diffraction and Phase Identification

a b
Part I
Chapter 3

c e

022 012s 002 012s 022

021s 011 001s 011 021s

020 010s 000 010s 020

021s 011 001s 011 021s

022 012s 002 012s 022

d 50 nm

Fig. 3.2 a Bright-field TEM image of the spinodally decomposed Fe-25 at.%
Be alloy. b [100] selected area electron diffraction pattern (SADP) showing
B2 superlattice reflections and sidebands from a spinodally decomposed Fe-25
Be alloy. c Corresponding schematic SADP. d (110) reflections with sidebands
associated with <100> modulations. e (100) dark-field TEM image of the
B2-ordered phase in the spinodally decomposed Fe-25 Be alloy; [100]-oriented
grain
 3.3 Superalloys with Ordered Precipitates 85

Another zone axis pattern (Fig. 3.3) obtained from this speci- a
men gives us further information on the presence of the FeBe
B2-ordered second phase. In this [011] pattern, you can see faint

Part I
reflections located at the {100} and {1, ½, ½} locations. The
B2-ordered structure coexists with the disordered body-centered
cubic A2 phase. The complete superposition of the B2 reflections
over the matrix A2 reflections further tells us that the B2 phase
is fully coherent with the parent phase, i.e. the atomic planes are
continuous across the interface. Thus, the OR between the parent
and ordered phase is intuitively obvious:
f100gA2 matrix == f100gB2 ; and h001iA2 matrix == h001iB2 :
 (3.2)

Although we can obtain centered dark-field TEM images using

B2 reflections present in any zone axis pattern obtained from the
A2 + B2 structures, a <001>-oriented specimen is ideal. This is
because dark-field images obtained using a B2 superlattice re- b
flection in <001> patterns very clearly show the crystallographic
nature of the FeBe phase, since the spinodal decomposition oc- 112 111s 110 111s 112
curs along <001> directions in this alloy. Recalling the image in
Fig. 3.2 obtained using the (100) superlattice reflection, you can
see this oriented structure in the [001]-oriented dark-field TEM

Chapter 3
image. You should note that the brighter phase in the dark-field
image is the ordered FeBe phase.
002 001s 000 001s 002

The Fe–Be system is also interesting because an additional

metastable phase, B32-ordered FeBe, can form in Fe–Be alloys
during prolonged aging near the top of the miscibility gap. This
particular phase was first identified using both SAD and micro-
diffraction techniques. Due to its coexistence with the disordered
A2- and B2-ordered FeBe phases, this phase was first identified 112 111s 110 111s 112
as a D03-ordered phase, Fe3Be by Davies and Richman (1964)
using XRD, and subsequently by Higgins et al. (1974) based on
electron diffraction. Subsequent research in this alloy system,
Fig. 3.3 a [011] SADP containing superlattice reflections confirming the pres-
however, revealed that the zone axis patterns for other orienta-
ence of the FeBe B2-ordered second phase. Note the shadow of the objective
tions were inconsistent with the D03 structure. Figure 3.4 pro-
aperture around the (001) superlattice reflection in this SADP. This double ex-
vides three zone axis patterns showing both the A2 matrix with
posed SADP is a convenient way to document the specific reflection used to
the second phase. In Fig. 3.4a, the [001] zone axis pattern con-
form the centered dark-field TEM image. b Schematic [011] SADP with the B2
tains no superlattice reflections, whereas, in Fig. 3.4b, the [011] superlattice reflections
zone axis pattern, the superlattice reflections are visible at the ¼
and ¾ {222}, {226}, and {662} locations (Miller et al. 1984). If
you look at Fig. 3.4c, you’ll see that the [111] zone axis pattern Use Several Zone Axes
shows no evidence of superlattice reflections. From all of this we
can conclude that the metastable phase is bcc-based B32-ordered It is very important to use > 1 zone axis patterns to deter-
FeBe, which exhibits a parallel relationship with the A2 matrix: mine structures and crystallographic relationships. 9

Œ100matrix == Œ100B32 and .110/matrix == .110/B32 (3.3)

Interestingly, if a [011] electron diffraction pattern is obtained

from a region containing both the B32-ordered FeBe phase and 3.3 Superalloys with Ordered
the B2-ordered FeBe phase, the resultant electron diffraction
pattern appears to be a [011] D03-ordered structure. Thus, cor- Precipitates
rect identification of the B32-ordered phase in the Fe–Be alloy
system also tells you that you must evaluate numerous zone axis Considerable research into the development/processing and
patterns in your assessment of phases and ORs to show that the properties of alloys containing ordered second phases has led
structures and crystallographic relationships are consistent. to many commercial alloys used in high-temperature applica-
 86 3 Electron Diffraction and Phase Identification

a b c
Part I

044 004 044

224 220 224 022 202

022 022
113 111 111 113

040 000 040 220 000

004 000 004 220

022 022
Chapter 3

113 111 111 113

202 022
044 004 044
224 220 224

333 331 331 333

d

Fig. 3.4 Three zone axis patterns obtained from the B32-ordered FeBe second
phase with corresponding indexed schematic SADPs: a [001] SADP; b [011]
zone axis pattern with visible superlattice reflections; c [111] SADP with no
100 nm evidence of superlattice reflections; d Centered dark-field TEM image of the
B32-ordered FeBe phase
 3.3 Superalloys with Ordered Precipitates 87

tions (e.g., a broad range of Ni–base superalloys), as well as in have a ‘globular’ morphology. More information can be obtained
areas where corrosion resistance is important (also Ni–base and in references (Sundararaman et al. 1988; Cozar and Pineau 1973;
Ni–Fe superalloys), and also for weight-specific properties, such Burke and Miller 1989).

Part I
as aerospace applications (Ti and Al alloys). As with steels, we
know that the properties of such alloys depend upon the type, We will highlight three ORs in this section to describe the vari-
size, morphology, and distribution of phases within their mi- ous phases in Alloy 718:
crostructures. So, we have to know the OR between phases if
we want to characterize and quantify the microstructure. Then 1. “Cube-on-cube” fcc matrix and L12-ordered struc-
we have a basis for rational development of heat treatments to tures:
optimize the properties of these various alloys.
h001imatrix == h001ippt and f100gmatrix == f100gppt (3.4)
Let’s start with Ni–base superalloys, which are a broad class of
alloys that generally contain Cr, Fe, Ti, Al, with low levels of C.
Some of these alloys also contain Nb, Mo, and B. These alloys 2. Fcc matrix and D022-ordered tetragonal structures:
have been developed for elevated-temperature and high-strength
applications, and account for the ever-improving performance of
h001imatrix == Œ001ppt and f001gmatrix == .001/ppt (3.5)
turbines for commercial aircraft and also for power generation.
The multielement compositions mean that we can form a variety
of phases from the fcc Ni-rich matrix during thermal treatment. 3. Fcc matrix and D0a-ordered orthorhombic structures:
As we’ll also see, this means that the diffraction patterns can be
quite complex. Furthermore, other TEM studies such as compo-
h110imatrix == Œ100ppt and f111gmatrix == .010/ppt (3.6)
sition analysis via XEDS and/or EELS is similarly complicated.
The first such alloy we’ll discuss is Alloy 718.

Chapter 3
We generally use Alloy 718, a commercial fcc Ni–base alloy, in 3.3.1 Example: γ″ and γ′ Precipitation
a special heat-treated condition, which promotes the formation of in Alloy 718
fine precipitates throughout the structure, thereby leading to signif-
icant increases in strength (Cozar and Pinneau 1973; Oblak 1969;
Paulonis 1974). This alloy also possesses good corrosion resis- To illustrate the ORs between the ordered precipitates and the
tance. We can create two types of intragranular strengthening matrix phase, we have chosen several examples. If you exam-
precipitates in this alloy during aging at ~ 700 °C: body-centered ine the SAD patterns obtained from a group of γ″ precipitates
tetragonal D022-ordered ‘Ni3Nb-type’ γ″ (gamma double prime) within the Alloy 718 matrix (Fig. 3.5) you can see superlattice
and the face-centered cubic L12-ordered ‘Ni3Al-type’ γ′ (gamma reflections superimposed on the zone axis pattern of the disor-
prime) precipitates. (You should note that the γ″ and γ′ precipi- dered matrix. Figure 3.5a is a [100] zone axis patterns in which
tates actually contain Nb, Ti, and Al in varying proportions, as has the matrix reflections are intense, and faint additional reflec-
been confirmed using atom probe microanalysis (Burke and Miller tions are also visible. We show a schematic of the [001] and
1989)). If we perform prolonged aging at ~ 700 °C, or even higher [100] zone axis pattern in Fig. 3.5b. In Fig. 3.5c–d, the individ-
temperatures (800–900 °C), we can form another ordered phase: ual <001>variants of the γ″ precipitates are imaged, whereas
δ (Ni3Nb. This δ phase has a D0a-ordered orthorhombic structure in Fig. 3.5e multiple variants are imaged. Depending upon the
and can form at grain boundaries, at inclusions or within grains. proportion of each variant of the γ″ precipitates, your SAD pat-
With its acicular morphology, the presence of significant amounts tern may show relatively uniform reflections for all variants, or
of δ can have a deleterious effect on the mechanical performance it might only show reflections for one variant.
of Alloy 718. So we need to be able to identify this phase within
microstructures of Ni–base superalloys, in which case the alloy These superlattice reflections show some distinctive character-
can exhibit nonoptimal properties for long-term service. istics that we can relate to the shape of these fine precipitates.
If you look closely at the pattern, you can pick out three ar-
We can see in conventional TEM imaging that the D022-ordered rays of superlattice reflections based on their appearance: there
γ″ precipitates form as disks in the fcc matrix, whereas the is streaking associated with two of the three patterns. One set
L12-ordered γ′ precipitates form as cuboids. Both types of pre- of reflections is streaked along the <010> direction, another set
cipitates form coherently with the matrix, but we can see signif- is streaked along the <100> direction, and there is a third set
icant coherency strains associated with the tetragonal distortion that is not streaked in this [001] pattern. By forming a centered
of the γ″ precipitates. The morphologies of these precipitates dark-field TEM image with superlattice reflection, we can im-
derive from surface and elastic energy considerations. The coher- age the precipitates that give rise to that reflection. By forming
ency hardening associated with the γ″ precipitates are a principal a centered dark-field image using three superlattice reflections
source of strengthening in this alloy (Oblak et al. 1974). The δ positioned within the objective aperture, we can visualize the
phase can exhibit a plate-like morphology or it can appear to three variants of the γ″ precipitates, as shown in Fig. 3.6.
 88 3 Electron Diffraction and Phase Identification

a b

220 020 220 024 004 024

Part I

013 013
110 110 022 002 022
011 011
200 000 200 020 000 020

011 011
110 110 022 002 022
013 013
220 020 220 024 004 024

c d e
Chapter 3

200 nm

Fig. 3.5 a [100] SADP containing faint superlattice reflections for the three variants of γ″ precipitates. b schematic of the [001] zone axis pattern. Centered dark-field
TEM images showing the three variants of γ″ precipitates: c [100], d [010], and e [001] with [100] and [010] variants of the γ″ precipitates

Superlattice Reflections patterns reveals that the disk faces of the γ″ precipitates are par-
allel to the {100} planes in the fcc matrix. The OR between the
- γ″ and the fcc matrix is:
dustrial metals. 9 .001/ 00 == f001gmatrix and Œ100 00 == h100imatrix (3.7)

It is only the (001) planes of the γ″ precipitates that are parallel

You can clearly see the lenticular edge-on appearance of these to the (100), (010), or (001) planes of the fcc matrix. Thus, we
disk-shaped γ″ precipitates. The streaking observed in the SADP find that there are three possible crystallographic variants of γ″.
is due to the disk-shaped morphology of these precipitates. As You can clearly see that orientation relationship if you now look
was discussed in Chap. 12 of W&C, the precipitate shape changes at Fig. 3.1, which shows two γ″ precipitates and the matrix in a
the reciprocal lattice point into a feature related to the morphol- [100]-oriented sample.
ogy of the precipitate. In this case, the reciprocal lattice “point”
is now a “relrod,” which is intersected by the Ewald sphere in In addition to the lenticular disk-shaped γ″ precipitates, the
reciprocal space. The disk-shaped γ″ precipitates that lie parallel Alloy 718 diffraction patterns also provide us evidence for
to the foil surface give rise to the nonstreaked reflections. the presence of the fine coherent L12-ordered γ′ precipitates,
which we know also form during the aging treatment. The SAD
If you examine the three major zone axis patterns [110], and pattern shown in Fig. 3.8 obtained from a [001]-oriented thin
[111] shown in Fig. 3.7, you can clearly see the presence of the foil contains the presence of faint superlattice reflections in the
streaked reflections due to the γ″ precipitates. Analysis of these (100) and (110) locations. These precipitates share the same
 3.3 Superalloys with Ordered Precipitates 89

a orientation as the matrix, which can be described as a cube-

on-cube OR:

Part I
h100i 0 == h100imatrix and f100g 0 == f100gmatrix (3.8)

Because of the cubic symmetry, this can also be written as:

Œ110 0 == Œ110matrix and .111/ 0 == .111/matrix (3.9)

3.3.2 Example: D0a-Ordered δ Precipitation

in Alloy 718

The geometric configuration of the plate-like δ structures within

the fcc matrix in Fig. 3.9 tells us that it has an OR with the ma-
trix. However, depending upon the thermal processing of the ma-
terial, it is also possible to form δ with a ‘globular’ morphology.
Therefore, we need to acquire electron diffraction patterns if we
200 nm
want to identify the phase as well as assess its crystallographic
b
OR with the matrix. A detailed analysis of the OR between the
δ phase and the fcc matrix has been performed by Rong et al.

Chapter 3
(1999) and Sundararaman et al. (1988) for Alloy 718 and another
Ni–base superalloy, Alloy 625.

If you want to examine the OR between the δ precipitates and

the fcc matrix, you have to obtain several low-index zone axis
patterns like those shown in Fig. 3.10. If you orient the thin foil
specimen to the [110] zone axis of the matrix you’ll see that these
precipitates are oriented so that their long face is parallel to the
(111) planes of the matrix. Figure 3.10a is a [110] matrix SAD
pattern containing δ reflections. The zone axis of the δ precipi-
tates is [100]. A [111] matrix zone axis pattern is also shown in
Fig. 3.10b; superposed on the matrix pattern is the [100] zone
axis pattern of the δ precipitates (Fig. 3.10c). The third matrix
pattern present in Fig. 3.10d is [112], which contains the [001]
δ zone axis pattern. For this ordered orthorhombic structure, we
can conclude that the OR with the fcc matrix is
h110imatrix == Œ100ı and h111imatrix == .010/ı (3.10)

More details concerning the numerous variants of δ precipitates

Fig. 3.6 a [001]-oriented Alloy 718 centered dark-field image, obtained using can be obtained from from the work of Sundararaman et al.
three γ″superlattice reflections in the objective aperture, showing three variants (1988) and Rong et al. (1999).
of the coarse γ″ precipitates. b Corresponding [001] SADP. Note the faint image
of the objective aperture over the γ″ superlattice reflections used to form the We can find further evidence that this phase has a crystallo-
image in a graphic OR with the fcc matrix by the arrangement of the δ
precipitates formed along a grain boundary. The precipitates
appear to ‘break up’ the boundary into ‘steps’ as they grow. The
δ precipitates can lead to pronounced grain boundary migration,
as shown in Fig. 3.9a. Because these precipitates are enriched
in Nb, their formation results in a local depletion of Nb in the
matrix, and, depending upon the heat treatment of the material,
there can be a γ″ precipitate-free zone surrounding them.
 90 3 Electron Diffraction and Phase Identification

a b
Part I
Chapter 3

224 222 220 222 224

114 112 110 112 114 132 022 112 202 312

121 011 101 211

004 002 000 002 004 220 110 000 110 220

211 101 011 121

114 112 110 112 114

312 202 112 022 132

224 222 220 222 224

γ [110] γ [111]
Fig. 3.7 Two additional major zone axis patterns and corresponding indexed schematic patterns: a [110]; b [111]. You can clearly see the presence of the streaked
reflections due to the γ″ precipitates
 3.3 Superalloys with Ordered Precipitates 91

Part I
b

Chapter 3
c

022 002 022

Fig. 3.9 Diffraction-contrast images showing the geometric configuration of

020 000 020
the δ structures within the Alloy 718 fcc matrix. a Plate-like δ precipitates as-
sociated with grain boundary migration. b Discrete δ precipitates formed along
a grain boundary. Such a configuration is clear evidence for an OR between the
precipitate and matrix phases

022 002 022 Fig. 3.8 a Centered dark-field TEM image of the γ′ precipitates. b SAD pattern
obtained from this [001]-oriented thin foil with faint superlattice reflections in
the (100) and (110) locations due to fine coherent L12-ordered γ′ precipitates.
c Schematic diffraction pattern
 92 3 Electron Diffraction and Phase Identification

a b c

034 024 014 004 014 024 034

Part I

033 023 013 003 013 023 033

111 002

032 022 012 002 012 022 032

031 021 011 001 011 021 031

111 000 111

020 000 020

031 021 011 001 011 021 031

002 111 220
032 022 012 002 012 022 032

033 023 013 003 013 023 033

d e f
042 222 402 440 240 040 240 440

330 130 130 330

Chapter 3

131 111 311

420 220 020 220 420

110 110 310

310

220 000 220 000

400 200 200 400

310 110 110 310

311 111 131

420 220 020 220 420

330 130 130 330

402 222 042

440 240 040 240 440

Fig. 3.10 Low-index zone axis patterns from δ precipitates in Alloy 718 oriented so that their long face is parallel to the (111) planes of the matrix. a [110] matrix
SAD pattern containing δ reflections in a [100] orientation. b Schematic [111] matrix zone axis pattern. c Schematic [100] zone axis pattern of the δ precipitates.
d [112] zone axis pattern which contains the [001] δ zone axis pattern. e schematic [112] matrix zone axis pattern. f Schematic [001] δ zone axis pattern

Licensed to ([email protected])
 3.4 Carbide Precipitation in fcc Alloys 93

Part I
Fig. 3.11 a Bright-field and b corresponding dark-field TEM images of Cr-rich M23C6 carbides on grain boundaries in Alloy 690

3.4 Carbide Precipitation in fcc Cr content of this steel is responsible for the good corrosion
resistance of the alloy and is also important for the stability of
Alloys

Chapter 3
the austenite phase. In the Ni–base alloys, the presence of in-
tergranular M23C6 is beneficial for resistance to stress corrosion
Austenitic stainless steels and nonprecipitation hardenable Ni– cracking. Thus, we find it useful to be able to assess the presence
base alloys are also important engineering alloys used in appli- and extent of M23C6 in these materials in order to optimize pro-
cations where corrosion resistance is important, such as the pow- cessing and properties of these alloys. Once again we have to
er-generation industry. Austenitic stainless steels (the Type 300 use electron diffraction to find the ORs between the two different
series) contain elevated Cr levels (~ 18–20 wt.%) with Ni con- crystal structures.
tents in the range of ~ 8 % and higher. Corresponding Ni–base
alloys include Alloy 600 (with ~ 16 wt.% Cr and ~ 8 wt.% Fe) The OR that is of interest for these carbides is the well-known
and Alloy 690 (with ~ 30 wt.% Cr and ~ 8 wt.% Fe). Both classes ‘cube-on-cube’ OR
of alloys contain relatively low C contents (< 0.1 wt.%). When
we age these alloys in the temperature range ~ 600 to ~ 850 °C, .111/carbide == .111/matrix and Œ110carbide == Œ110matrix
copious intergranular carbide precipitation occurs along grain  (3.11)
boundaries. The fcc Cr-rich M23C6 carbide is the dominant pre-
cipitate in austenitic stainless steels and in Alloy 690. (This phase
also forms in Alloy 600, although the Cr-rich M7C3 carbide may
dominate.) This carbide is semicoherent with the fcc matrix as 3.4.1 Example: M23C6 Precipitation
its lattice parameter is approximately three times that of the ma- in a Ni–Base Alloy
trix. Another precipitate that can form in austenitic alloys (both
Fe–base and Ni–base), depending upon the alloy content, is the
Nb-rich MC carbide. Certain stainless steels contain elevated The bright and dark field TEM images in Fig. 3.11 shows Cr-
amounts of Nb, which is a very strong carbide-former. This fcc rich carbides that have formed along a grain boundary in Al-
carbide can form within the matrix or at grain boundaries. Its loy 690, a single phase Ni-30Cr-8 Fe alloy. We can identify
presence provides added strength to the alloy, and also can pin these Cr-rich M23C6 carbides from their electron diffraction pat-
grain boundaries, thus inhibiting grain growth. This carbide is terns. SAD patterns obtained from the carbide + matrix contain
incoherent with the fcc matrix. two superimposed patterns, one from the matrix and the other
from the carbides, as shown in Fig. 3.12. SAD patterns ob-
However, we generally associate the presence of intergranular tained from three different zone axes, [111], [100], and [112],
M23C6 in austenitic stainless steels with susceptibility to prefer- showed that the carbide has a d-spacing that is nearly three
ential intergranular corrosion and intergranular stress corrosion times larger than that of the fcc matrix, with carbide reflections
cracking. Depending upon the aging temperature, the precipita- located at 1/3 and 2/3 (hkl) of the matrix pattern. These car-
tion of this carbide can locally deplete the surrounding matrix bides are semicoherent with the matrix, with dislocation arrays
of Cr to levels below ~ 12 wt.%, making the steel susceptible to accommodating the misfit at the interface. Furthermore, we see
intergranular corrosion and stress corrosion cracking. The high that the patterns of the carbide and matrix phase are identical
 94 3 Electron Diffraction and Phase Identification

a b c
Part I

d e f

022 202 022 002 022

606
131 111 311

404

202
Chapter 3

220 000 220 020 000 020 220 000 220

220 440

311 111 131

202 022 022 002 022

Fig. 3.12 SAD patterns obtained from three different zone axes (a [111], b [100], c [112]) show that the M23C6 carbide has a d-spacing that is nearly three times
larger than that of the fcc matrix, with carbide reflections located at 1/3 and 2/3 (hkl) of the matrix pattern. Schematic diffraction patterns for d [111], e [100] and
f [112] zone axes

in symmetry. Analysis of the SADP patterns revealed that the the matrix reflections. Numerous faint reflections are arranged
carbides are related to the matrix according to the cube-on-cube symmetrically about the fundamental reflections. However,
OR which we just described above. you’ll see there are subtle differences as you examine the various
fundamental reflections. The intensity of the ‘many’ precipitate
reflections varies with fundamental reflections. The precipitate
reflection close to the corresponding matrix reflection is a small
3.4.2 Example: MC Carbides in a Ni–Base Alloy distance inside the matrix reflection towards (000), indicating it
has a slightly larger lattice parameter than the matrix. The addi-
tional reflections present around the fundamental reflections are
We can detect the presence of fine incoherent fcc NbC carbides the result of double diffraction as discussed in W&C Chap. 18.
in fcc Ni–base or Fe–base alloys using SAD. Figure 3.13 shows Thus, we can conclude that the carbides have a parallel orienta-
the morphology of the fine incoherent carbides in a Ni–base al- tion as the matrix, although these carbides are incoherent with
loy. Look closely at the SAD patterns in Fig. 3.13c, d. If you the matrix, unlike the M23C6 carbides. Therefore, these carbides
carefully examine Fig. 3.13d, which is an [100]-oriented zone also show a cube-on-cube OR with the matrix, just as we de-
axis pattern, you can see the MC carbide reflections adjacent to scribed for the M23C6 carbides.
 3.4 Carbide Precipitation in fcc Alloys 95

Part I
Chapter 3

Fig. 3.13 a Bright-field and b corresponding centered dark-field TEM image of NbC precipitates in a Ni–base alloy. c [111] and d [100] zone axis patterns of NbC in
Ni alloy matrix with additional reflections due to double diffraction
 96 3 Electron Diffraction and Phase Identification

3.5 Ferritic Steels 3.5.1 Relationships Between Austenite

and Ferrite, Austenite and Martensite
Part I

Despite the steel industry’s economic woes of the past several (fcc/bcc)
decades, steels continue to represent the dominant class of en-
gineering materials, with widespread use throughout the world As discussed in W&C Chap. 18, there are a few key ORs that
in a variety of industries, including automotive and white goods occur regularly because there are relatively few crystal systems.
(appliances). Over these same decades, we have made signifi- Also useful engineering alloys, such as steels, have relatively
cant advances in optimization of steels for improved properties simple crystal structures – which is why we have emphasized the
primarily through a detailed understanding of the essential re- cubic systems to date. The most important phase transformations
lationship between processing, microstructure, and properties. that occur in steels are the austenite–ferrite transformation during
Such tailoring of thermomechanical processing regimens cou- cooling from elevated temperatures, and the austenite–marten-
pled with precisely controlled alloy compositions was among site transformation that occurs during quenching the steel from
the earliest examples of nanostructurally engineered materials. the austenite phase field. Two of the most common orientation
relationships that describe the fcc-bcc crystallography are Kurd-
fcc → bcc (or bct) jumov–Sachs and Nishiyama–Wasserman.

This phase transformation is one of the most technolog- Kurdjumov–Sachs

ically important found in all of engineering materials.9 The Kurdjumov–Sachs OR is one of the most famous and cer-
tainly the oldest (Kurdjumov and Sachs 1930). It was first dis-
covered in Fe–Ni meteorites in which the Widmanstätten pattern,
Not surprisingly, TEM characterization has been able to provide clearly visible to the eye, consists of a bcc, Fe-rich phase, which
Chapter 3

the much-needed insight into the development of the complex precipitates from the fcc austenitic matrix. The close-packed (or
microstructures that we now utilize routinely in the engineering closest-packed) planes and direction in the bcc and fcc crystals
and construction fields. For example, the controlled precipita- are parallel. Specifically
tion of ultrafine (2–20 nm) microalloy carbides and carbonitrides
in low-alloy steels has tremendous impact on the properties of .111/fcc == .011/bcc (3.12a)
high strength low-alloy steels. The characterization of the pre-
cipitation process, however, is complicated because of the aus- N bcc
.101/
N fcc == .111/ (3.12b)
tenite-to-ferrite (i.e., γ → α) or austenite-to-martensite (γ → α′)
transformation. Thus, the final microstructure that we have to
characterize is either a ferritic or martensitic structure – even .12 N fcc == .21
N 1/ N bcc
N 1/ (3.12c)
though the phase we are interested in for precipitation studies
may in fact be the prior austenite. This fcc → bcc (or bct) trans- N bcc
formation is one of the most technologically important phase Œ110
N fcc == Œ111 (3.12d)
transformations in all of engineering materials. Therefore, if we
are to determine whether precipitation occurs in the austenite It is important to note that the Kurjimov–Sachs OR does not
or ferrite during processing we have to analyze the final micro- universally apply to all fcc–bcc systems; rather, there can be sit-
structure and we have to know the ORs between: i) austenite uations in which the fcc–bcc OR is described by different parallel
and ferrite/martensite; and ii) the carbides and the matrix phase. planes, such as those defined in the Nishiyama–Wasserman OR.

In this section, we will describe the major ORs that have been
identified for fcc and bcc structures (austenite, ferrite/low-carbon Nishiyama–Wasserman
martensite, respectively). We will also present you with some The Nishiyama–Wasserman OR is also between fcc and bcc
selected examples. For obvious reasons, there has been an ex- phases and therefore is closely related to the Kurdjumov–Sachs.
tensive amount of research in this area. The classic text by An- The specific details are:
drews et al. (1971) provides a wealth of information concerning
.111/ N bcc
N fcc == .110/ (3.13a)
such orientation relations. Further references dealing with the
development and application of these ORs can be found in the
literature, and a few pertinent references (but by no means all-in- N fcc == .001/bcc
.011/ (3.13b)
clusive list) are included in the bibliography. You will find the
reviews by Ohmori (1971) dealing with carbide precipitation in
steels to be particularly useful. .211/fcc == .110/bcc (3.13c)

We’ll now illustrate some of the important ORs that you may
Œ110fcc == Œ111bcc (3.13d)
encounter when analyzing steels or Fe–base alloys.
 3.5 Ferritic Steels 97

When written with such precision, a rotation of 5.25o about the These ORs describe the precipitation of cementite in ferrite, and
[011]bcc will transform the Kurdjumov–Sachs OR into the Nish­ the relationship between cementite and ferrite in pearlite. Be careful
iyama–Wasserman OR. In many cases, the actual rotational dif- because the Bagaryatski and Isaichev ORs can be rather difficult to

Part I
ference is less than the error in measurement, so they can be distinguish, particularly using SAD analysis, since these two ORs are
viewed as effectively the same OR. The key factor to remember related by a small (~ 3°) rotation (which is smaller than the general
is that, for both these ORs, the close-packed planes of the fcc error in using the SAD technique for determining ORs!). More re-
austenite and bcc ferrite are parallel, as are the close-packed cently, detailed studies by Zhang and Kelly (1998, 2002, 2009) and
directions. coworkers using convergent-beam electron diffraction and kikuchi
line analysis, and by Huang and Pryds (2000) have demonstrated that
the Isaichev OR is generally observed for precipitation of cementite
in martensite and ferrite rather than the Bagaryatski OR.
3.5.2 Relationship Between Cementite
(Orthorhombic Fe3C or M3C) and Ferrite/ The use of the convergent-beam electron diffraction and kikuchi
line analysis as described by Zhang and Kelly (1998) is more
Tempered Martensite accurate than conventional SADP analysis, thereby reducing the
error in determination of ORs. They have noted, as have previous
Not only are there the basic phase transformations from austenite researchers, that the accuracy of conventional SADP analysis is
to ferrite or austenite to martensite, but there can be complex not sufficient to conclusively distinguish between the Bagaryatski
carbide, nitride, or carbonitride precipitation reactions in the and Isaichev ORs. However, with the improved accuracy of the
austenite phase or in the ferrite phase, as well as the eutectoid convergent beam kikuchi line diffraction technique, Zhang and
transformation reaction (austenite to ferrite plus lamellar Fe3C, Kelly (2002)have successfully identified the Isaichev but not the
also termed ‘pearlite’). As these precipitation reactions involve Bagaryatshi OR to describe cementite precipitates in ferrite or
the formation of crystalline second phases within a crystalline martensite. Therefore, if you are working with steels, take care

Chapter 3
matrix, there will be ORs that describe the crystallography of and make sure that you familiarize yourself with the Isaichev OR!
these reactions. Cementite (Fe3C), which may also be referred
to as ‘M3C’ due to the incorporation of Mn into the Fe3C phase,
is a commonly observed carbide in ferritic steels. The cementite
phase is incoherent with the matrix, and its size/morphology and 3.5.3 Relationships Between Alloy Carbides
distribution can have significant effects on the properties of the and Ferrite
steel.

Alloy carbide precipitation in steels has a significant effect on

Pitsch the properties of steels. In particular, the size, distribution, and
The Pitsch OR was proposed to describe the precipitation of morphology of the carbides formed during aging will affect the
cementite from austenite (Pitsch 1962): strength, ductility, and toughness of the alloy. Understanding
the precipitation process and the microstructural development
.001/Fe3C == .2N 15/
N bcc I .100/Fe3C == .311/
N bcc and can lead to improved processing of the alloys for the optimum
(3.14)
.010/Fe3C == .131/bcc  properties. The ORs between the ferritic (bcc) matrix and the
carbides listed below are useful for phase identification and the
characterization of steel microstructures.

Bagaryatski
This orientation relationship was developed to describe the re- Pitsch–Schrader
lationship between cementite and ferrite (pearlite formation), This orientation relationship was originally developed for the
(Bagaryatski 1950) and is described as: precipitation of epsilon carbide in martensite during low-tem-
perature tempering (Pitsch and Schrader 1958), but is also appli-
N bcc I .010/Fe3C == .11N 1/
N bcc I
.100/Fe3C == .011/ cable to the precipitation of hexagonal M2C in ferrite:
N bcc (3.15)
.001/Fe3C == .211/bcc I Œ100Fe3C == Œ011  N M2C == .011/
.0001/M2C == .011/bcc I .1100/ N bcc I
N M2C == .100/bcc (3.17)
.1120/ 
Isaichev
The Isaichev OR, similar to the Bagaryatski, also relates the
carbide crystals to the their parent ferritic or martensitic phase Rong–Dunlop
(Isaichev 1947): Mo2C (M2C) precipitates in highly alloyed high-speed steels
have been observed to exhibit this orientation relationship with
.103/Fe3C == .101/bcc I Œ010Fe3C == Œ11N 1
N bcc (3.16) the bcc matrix:
 98 3 Electron Diffraction and Phase Identification

N M2C == .012/
.0001/M2C == .021/bcc I .1100/ N bcc I characterization of this phase as formed in Type 329 duplex stain-
N M2C == .100/bcc (3.18) less steel. G-phase exhibits a cube-on-cube OR with the δ ferrite
.1120/  matrix.
Part I

This orientation relationship was observed along with the Pitsch– Figure 3.14a is a [110] zone axis pattern obtained from δ fer-
Scharader relationship in these alloyed tool steels (Rong and rite containing G-phase precipitates. Note the presence of the
Dunlop 1984). faint reflections located at the ½ (011) bcc and at ¼, ½, and
¾ (200) bcc locations. These are (111) and (200)-type G-phase
reflections. The schematic is shown in Fig. 3.14b. Similarly, the
Baker–Nutting [001] zone axis pattern for the δ ferrite, which you can see in
The Baker–Nutting OR describes the relationship between fcc Fig. 3.15a, b, provides further confirmation of the OR:
carbides (or carbonitrides or nitrides) and the bcc ferrite matrix
(Baker and Nutting 1959): Œ001G ==Œ001ı ; and Œ011G == Œ011ı (3.20)

.100/MC == .100/bcc and h100iMC == h110ibcc (3.19)

a
In summary, by evaluating the OR between fine fcc MC carbides
(where M = Nb, Ti, or V) and the matrix, you can determine whether
precipitation occurred in the fcc austenite or the bcc ferrite. The
crystallographic relationship that describes the OR between mar-
tensite or a bcc phase forming in an fcc matrix can be either the
Kurdjumov–Sachs OR, or the Nishiyama–Wasserman OR, as we
mentioned previously. The precipitation of fine microalloy (Nb, Ti)
Chapter 3

carbides in steel during thermomechanical processing can occur in

the austenite, prior to transformation. When you are examining the
microstructure of such a steel, you’ll see that the fine precipitates
will exhibit an OR with the ferrite/martensite matrix similar to that
for the retained austenite: either K-S or N-W. If the precipitation
occurs in the bcc structure (ferrite/low-carbon martensite), then the
precipitates will exhibit the Baker–Nutting OR with the matrix.
Further examples can be found in Zhang and Kelly (2009), Kelly
et al. (1990), Edington (1976), Ohmori (1971), Dahmen (1982),
and Zhang and Weatherly (2005).

b
3.5.4 Precipitation in Ferritic Structures
112 110 112

The following section provides you with some observations of

precipitation in ferritic structures and is a rough guide to some
of the ORs that you may encounter as you embark upon your
analyses of steel microstructures.
002 000 002

G-phase Precipitation in Delta (δ)Ferrite

In highly alloyed stainless steels that contain austenite and δ ferrite,
additional precipitation reactions can occur along with the forma-
tion of alloy carbides. A complex Ni silicide, termed ‘G-phase,’
can form in the bcc δ ferrite. G-phase is isostructural with the
112 110 112
intermetallic silicide Ni16Si7Ti6, but can also contain Cr, Mn, and
Mo. This complex structure is fcc and can form in bcc δ ferrite.
G-phase formation tends to occur in conjunction with the spinodal
decomposition of δ ferrite during aging in the temperature range
~ 375–500 °C. A detailed TEM study of G-phase has been per- Fig. 3.14 a [110] zone axis pattern obtained from δ ferrite containing the (111)
formed by Mateo et al. (1997) in which they provided a complete and (200)-type G-phase reflections. b Schematic of the patterns in a
 3.6 Epitaxial Oxide on Metal: Presence of Fe3O4 on Steel Foils 99

a [100] b

Part I
022 002 022

011 110

020 000 020

011 011

022 002 022

Chapter 3
c G-phase in δ ferrite
Fig. 3.15 a [001] zone axis pattern for the δ ferrite and G-phase precipitates.
b Schematic of the pattern in a. c dark-field TEM image formed using a G-phase
reflection

Note also that the lattice parameter of the G-phase is nearly four
times that of the δ ferrite. The fine G-phase precipitates are visi-
ble in Fig. 3.15c, which is a dark-field TEM image formed using
a G-phase reflection.

3.6 Epitaxial Oxide on Metal:

Presence of Fe3O4 on Steel Foils
The OR that exists for thin-film structures on substrates and for
multilayers is an epitaxial relationship. Commercial materials
development for semiconductor device applications in the elec-
tronics and magnetics industries frequently utilizes epitaxial OR
between deposition layers and substrates in order to achieve spe-
cific properties and performance. Thus, you might immediately
associate epitaxial ORs with advanced electronic devices or data
storage media. However, epitaxial ORs also come into play when
we are working with conventional TEM thin-foil specimens –
50nm
particularly those that have surface oxidation.

Metal Specimens Oxidize

Nonstainless steels and other metals can oxidize rapidly:

the thin oxide film can affect your analysis. 9
 100 3 Electron Diffraction and Phase Identification

As you know, TEM analysis of steel microstructures requires that a

you prepare electron-transparent steel thin-foil specimens (see
Chap. 10 in W&C). Steel metallurgists are rather conservative
Part I

and so the most commonly used technique for the preparation of

thin-foil specimens is electropolishing, generally with a twin-jet
electropolishing unit. Depending upon the electrolyte you use for
specimen preparation, your thin-foil specimens of nonstainless
steels may oxidize rapidly, so that a thin oxide film will form
on your foil surfaces (particularly if you used a perchloric acid
containing electrolyte); this oxide film will increase in thick-
ness as your specimens are exposed to air. The presence of such
films can complicate the analysis of your electron diffraction
patterns, particularly if you are interested in studying precipi-
tation in steels. The surface oxide, Fe3O4, can form epitaxially
on the ferrite foil surfaces. Thus, these surface oxide films will
exhibit an OR with the ferrite grains. Keown and Dyson (1966)
identified the relationship between the bcc matrix and the fcc
oxide as close to the Nishiyama–Wasserman OR. Subsequent
analysis of a vanadium-bearing steel by Davenport provided
evidence for the existence of several oxide–metal ORs: Bain, b

Kurdjumov–Sachs, and Nishiyama–Wasserman. Edington also

112 110 112
included an example from Dunlop of the Nishiyama–Wasserman
OR for epitaxial oxide on a vanadium-bearing steel, which can
Chapter 3

complicate the interpretation of the electron diffraction data for 220

second-phase precipitates within alloy.
131
An SAD pattern obtained from a [100]-oriented grain in a low-al-
002 000 002
loy steel is shown in Fig. 3.16. Some carbide reflections are also
present. However, the pattern appears quite complicated. Numer- 022
022
ous ‘diffuse’ reflections are present, which make it difficult to
assess the carbide reflections. If you take the time to do a careful
analysis of this pattern you’ll find that there are two Fe3O4 zone 113
axis patterns superimposed on the bcc matrix pattern: [110]Fe3O4 202
112 110 112
and [411]Fe3O4. Examination of the stereographic projection for
the Nishiyama–Wasserman OR between fcc and bcc structures
confirms that the epitaxial oxide exhibits the NW OR with the
bcc steel matrix.
Fig. 3.16 a SAD pattern obtained from a [100]-oriented grain in a low-alloy
steel. Numerous ‘diffuse’ reflections are present that arise from two Fe3O4 zone
axis patterns superimposed on the bcc matrix pattern. b Schematic of the pat-
terns in a

Chapter Summary

Orientation relationships between different crystals in alloy crystallographic directions and low-index planes, which
systems are most easily determined using standard SAD are parallel (or close to parallel) in both crystals. There are
patterns which contain reflections from both crystal struc- many different ORs, and we have only given some of the
tures. Only if one of the phases is too small (or there is too more common examples here, many of which have been
low a volume fraction) to give sufficient intensity in the known for decades, but all of which continue to govern the
SAD pattern might you need to resort to CBED or microdif- properties of complex engineering materials.
fraction. In general the OR is defined by pairs of low-index
 Appendix 101

Appendix

Part I
People Text-Specific Questions

Gareth Thomas was born in Wales in 1932 and died on Febru- T3.1 Identify the three arrays of superlattice reflections in
ary 6, 2014. He joined UC Berkeley in 1960 after his Ph.D. in Fig. 3.5 and suggest how to index them.
Cambridge and never left. He was key in establishing NCEM, an T3.2 Use the literature to discuss the orientation relationship
expert on steels and wrote the first Foreword to W&C. that has been observed between Alloy 718 and another
Ni–base superalloy, Alloy 625.
T3.3 Discuss the precipitates that can form in austenitic alloys
(both Fe–base and Ni–base), depending upon the alloy
Self-Assessment Questions content and illustrate your answers with examples of re-
lationships that we’ll see between the DPs.
T3.4 Using the literature for examples, describe where we see
Q3.1 Explain the difference between coherent, semi-coherent the Widmanstätten relationship.
and incoherent and how you expect the difference to in- T3.5 Using the literature for examples, describe where we see
fluence the DP. the Kurdjumov–Sachs relationship.
Q3.2 Does the existence of an orientation relationship between T3.6 Discuss how to distinguish between the Nishiyama–Was-

Chapter 3
two crystalline structures imply that they have matching serman OR and the Kurdjumov–Sachs OR using SAD
of lattice planes (are coherent or semicoherent?)? patterns.
Q3.3 What do we mean by the notation D0a? T3.7 Summarize by sketching the DPs how you can recognize
Q3.4 Sketch and label the [111] and [0001] DPs. the Pitsch, Bagaryatski, and Isaichev ORs.
Q3.5 Explain in more detail what we mean by the phrase T3.8 Summarize by sketching the DPs how you can recognize
‘spinodal alloy.’ the Pitsch–Schrader, Rong–Dunlop, and Baker–Nutting
Q3.6 What is a triaxially modulated two-phase structure? ORs.
Q3.7 How are the A2, B2, and B32 structures related? T3.9 Delve into the literature to explain the DPs you’d see
Q3.8 We comment on ‘broad range of Ni–base superalloys.’ from Ni16Si7Ti6.
What do we mean by broad? T3.10 Construct the most important DPs for cementite and com-
Q3.9 What is an acicular morphology? pare your results with reported observations. (Always
Q3.10 Look closely at Fig. 3.14, and explain where and why we remember to cite the observations.)
see the MC carbide reflections. T3.11 Construct the 001 and 111 patterns you expect to see as
Q3.11 How are the austenite-to-ferrite and austenite-to-marten- Cu starts to oxidize.
site transformations different? T3.12 Discuss the extra complications in DPs you’ll expect to
Q3.12 What complications can you produce in the DP when you see if precipitates form along the grain boundary.
electropolish a metal specimen and let it sit for a while T3.13 The δ phase can exhibit a plate-like morphology or it can
before putting it in the TEM? appear to have a ‘globular’ morphology. Use the litera-
Q3.13 Explain the stereographic projection for Fig. 3.15. ture to discuss when these shapes occur.
Q3.14 What does the streaking in Fig. 3.5 tell you? T3.14 What is an atom probe anyway, and why do we prefer
Q3.15 What is cementite? TEM?
Q3.16 You might form a centered dark-field TEM image with T3.15 In Fig. 3.4 superlattice reflections are visible at the ¼ and
superlattice reflection. Is this a good idea or not? ¾ {222}, {226}, and {662} locations. Explain why this
Q3.17 What is a B2 superlattice reflection and why are they is so.
originally forbidden in Fig. 3.2b?
Q3.18 Could you perform prolonged aging at ~ 700 °C of an
alloy and if so could you see the DP change?
Q3.19 How does the lenticular disk-shaped γ″ precipitates affect
the DP?
Q3.20 Think of a ceramic and a semiconductor system that
would give a pair of related DPs like any of those shown
in this chapter.
 102 3 Electron Diffraction and Phase Identification

References Miller MK, Burke MG, Brenner SS, Soffa WA, Laughlin DE,
Alexander KB (1984) Identification of a B32 metastable pre-
Part I

cipitate in the Fe-Be system. Scripta Metall 18:285–290

Oblak JM, Paulonis DF, Duvall DS (1974) Precipitation of γ in
the γ′ of nickel-base superalloys. Met Trans 5:143–153
General References Ohmori Y (1971) The crystallography of the lower bainite trans-
formation in a plain carbon steel. Trans ISIJ 11:95–101
Andrews KW, Dyson DJ, Keown SR (1971) Interpretation of Ohmori Y, Ohtani H, Kunitake T (1971) The Bainite in Low
Electron Diffraction Patterns. Plenum Press, New York Carbon Alloy High Strength Steels. Trans ISIJ 11:250
Dahmen U (1982) Orientation Relationships in Precipitation Paulonis DF, Oblak JM, Duvall DS (1969), Trans. ASM, 62:611–
Systems. Acta Metall 30:63–73 622
Edington JW (1976) Practical Electron Microscopy in Materials Pitsch W (1962) Die Orientierungszusammenhang zwishen Ze-
Science. N.V. Philips (Eindhoven), Macmillan Philips Tech- mentie und Ausenit. Acta Metall 10:897–900
nical Library Pitsch W, Schrader A (1958) Arch Eisenh. Die Ausscheidungs-
Porter DA, Easterling KE, Sherif MY (2009) Phase Transforma- form des Zementits im Ferrit 29:485–488
tions in Metals and Alloys, 3rd edn. CRC Press, Boca Raton, Rong W, Dunlop GL (1984) The crystallography of second-
FL ary carbide precipitation in high speed steel. Acta Metall
32:1591–1599
Rong W, Dunlop GL, Kuo KH (1986) An O-lattice interpretation
Specific References of orientation relationship between M2C precipitation and fer-
rite. Acta Metall 34:681–690
Bagaryatski YA (1950) Likely mechanism for the tempering of Rong Y, Chen S, Hu G, Gao M, Wei RP (1999) Prediction and
martensite. Dokl Akad Nauk SSSR 73:1161–1164 Characterization of Variant Electron Diffraction Patterns for
Chapter 3

Baker RG, Nutting J (1959) Precipitation Processes in Steels. gamma double prime (γ″) and delta (δ) Precipitates in an IN-
Iron and Steel Institute Special Report No. (64):1–22 CONEL 718 Alloy. Metall Mater Trans 30A:2297–2303
Burke MG, Miller MK (1989) A combined TEM / APFIM ap- Shi YN, Kelly PM (2002) The crystallography and morphology
proach to the study of phase transformations: Phase identi- of Mo2C in ferrite. J Mater Sci 37:2077–2085
fication in the Fe-Be system. Ultramicroscopy 30:1992–205 Sundararaman M, Mukhopadhyay P, Banerjee S (1988) Precip-
Cozar R, Pineau A (1973) Metall. Trans., 4:47–59 itation of the γ′-Ni3Nb phase in two nickel base superalloys.
Daniel V, Lipson H (1943) An X-Ray Study of the Dissocia- Metall Trans 19A:453–465
tion of an Alloy of Copper, Iron and Nickel. Proc Roy Soc Verbeken K, Barbe L, Raabe D (2009) Evaluation of the Crystal-
181(987):368–378 lographic Orientation Relationships between FCC and BCC
Davies RG, Richman RH (1964) Superlattices in Iron-Base Al- phases in TRIP Steels. ISIJ International 49:1601–1609
loys (Structures and Mechanical Properties). Trans Met Soc Zhang MX, Kelly PM (1998) Crystallography of spheroidite and
AIME 236:1551–1557 tempered martensite. Acta Mater 46:4081–4091
Higgins J, Nicholson RB, Wilkes P (1974) A new paradigm for Zhang MX, Kelly PM (2002) Accurate Orientation Relationship
metallic alloys in Materials Science. Acta Metall 22:201–217 between Ferrite and Austenite in Low Carbon Martensite and
Huang X, Pryds NH (2000) Crystallography and morpholgy of Granular Bainite. Scripta Mater 47:749–755
cementite precipitates formed during rapid solidification of a Zhang MX, Kelly PM (2009) Crystallographic features of phase
ferritic stainless steel. Acta Mater 48:4073–4082 transformations in solids. Progr Mater Sci 54:1101–1170
Isaichev IV (1947) Orientation of Cementite in Tempered Car- Zhang WZ, Weatherly GC (2005) On the Crystallography of
bon Steel. Zhur Tekhn Fiziki 17:835–838 Precipitation. Progr Mater Sci 50:181–292
Kelly PM, Jostsons A, Blake RG (1990) The orientation relation-
ship between lath martensite and austenite in low carbon, low
alloy steels. Acta Metall Mater 38:1075–1081 Software used for Generation of Selected-Area
Keown SR, Dyson DJ (1966) J Iron Steel Institute (London)
204:832–836 Diffraction Patterns
Kurdjumov GV, Sachs GZ (1930) Über den Mechanismus der
Stahlhärtung. Physik 64:325–343 JEMS by Pierre Stadelmann (to produce the schematics of
Mateo A, Llanes L, Anglada M, Redjaimia A, Metauer G (1997) SADPs)
Characterization of the Intermetallic G-Phase in an AISI-329 Zuo JM, Mabon JC (2004) Web-based Electron Microscopy
Duplex Stainless Steel. J Mater Sci 32:4533–4540 Application Software: Web-EMAPS. Microsc Microanal
Miller MK, Burke MG, Brenner SS (1985) Measurement of 10(Suppl 2) (http://emaps.mrl.uiuc.edu/ (to produce the sche-
characteristic wavelengths in modulated microstructures by matics of SADPs))
field-ion microscopy. J Microsc 139:41–47 Staley JT, DIFPAT Software (Scientific Software Services) 1988

Licensed to ([email protected])
 Convergent-Beam Electron
4

Part I
Diffraction: Symmetry
and Large-Angle Patterns

Chapter Preview

This chapter falls into two separate but related parts. The Chapter 20 of W&C presents details of how to obtain con-
first considers the use of diffraction to determine the sym- vergent-beam patterns. That chapter includes a brief dis-
metry of the crystal. The second elaborates on a set of tech- cussion of large-angle convergent-beam electron diffrac-
niques, known generically as Tanaka methods. tion (LACBED) patterns, also known as Tanaka patterns.
The second section of this chapter gives more detail on
The use of convergent-beam electron diffraction (CBED) LACBED and other members of the family of Tanaka meth-
patterns to determine crystal symmetry has been previewed ods. We give a step-by-step recipe for obtaining the patterns,
in Chap. 21 of W&C. In this chapter we take those ideas and the principal applications are described.
and develop them further. A worked example is included to
show how the process is carried out in detail. We explain the
strengths and pitfalls of the method at some length.

Chapter 4

© Springer International Publishing Switzerland 2016

C. B. Carter, D. B. Williams (Eds.), Transmission Electron Microscopy, DOI 10.1007/978-3-319-26651-0_4 103
 104 4 Convergent-Beam Electron Diffraction: Symmetry and Large-Angle Patterns

4.1 Symmetry Crystal Symmetry

Part I

The determination of the symmetry of crystals is of great impor- The measured symmetry of a crystal may depend on the
tance for several reasons: symmetry assists in crystal identifica- method of measurement. It depends on how sensitively
tion by assigning the structure to one of the 230 space groups; it the method can detect deviations from symmetry. 9
determines the possible sites for atoms within the unit cell and
so facilitates the determination of the crystal structure; and it
has important repercussions for the physical properties of the The actual determination of symmetry by CBED proceeds in
material. For example, only crystals that do not have a center two main steps: we first determine the point group and then we
of symmetry can show piezoelectric properties. Another exam- determine the space group. We assume that we already know the
ple arises with regard to the constraints on the anisotropy of unit cell of the crystal, and in particular whether the unit cell is
properties, which are governed by crystal symmetry. Books by centered (see W&C Chap. 21). In this chapter we assume some
Nye (1957) and Lovett (1999) give systematic descriptions of familiarity with the ideas of crystal symmetry. If you need help,
this field. there are several good introductory books on this subject, for
example Hammond (2009).
However, symmetry is not a simple matter. As we consider the
symmetry of a crystal, we have to ask: “what do we mean by
symmetry and how perfect does the symmetry have to be?” We
could argue that no real crystal has symmetry. It is not plausible 4.2 Point-Group Determination
that a grain of salt consists of an exact cube, down to the position
of every single atom, yet, if it does not, the object is not a cube
and cannot have cubic symmetry. So what exactly do we mean The method of determining the point group given here is based
when we say that a crystal has a specific symmetry (in this case on the examination of CBED patterns taken centered on a zone
cubic symmetry)? We will take it that what we mean is this: the axis. This is not the only way to do it. For example, Tanaka
real crystal behaves (in any experiment that we perform) as if has proposed that it is more economical if we take patterns that
it has the symmetry of the hypothetical, perfect (and infinite) are slightly off the zone axis, while Morniroli has developed a
crystal; that we are unable to detect any difference between the method using precession. While there are advantages to both
properties of the crystal and a set of properties with the stated these methods, we feel that the method we describe is, in general,
symmetry. the best one to begin with. We’ll make further comments on these
alternative methods later in Sect. 4.5. The first formulation of the
The conclusion to be drawn from this assumption is that the ideas of this section was in Buxton et al. (1976); the practical
symmetry of an object depends on how we measure it. Conse- procedure we describe here was given in Eades (1988a).
quently, as measuring techniques become more refined, we may
Chapter 4

need to assign a crystal to a lower symmetry. For example, while A crucial point to understand is that in a diffraction experiment,
an earlier method may have been unable to detect any deviation the symmetry of the experiment itself (that is the symmetry of the
from the symmetry assigned to the crystal, the newer method crystal and the electron beam taken together) is not the same thing
can. This explains why it is that CBED does not always give the as the symmetry of the crystal. (This is true for all diffraction ex-
same answer to a symmetry determination as earlier methods. periments, though we shall consider its relevance only to CBED.)
Most of the symmetry assignments you will find in the literature The symmetry of the experiment is the symmetry of the crystal
were made by X-ray diffraction. However, X-ray diffraction is but with a certain direction (the direction of the beam which you
not always good at determining symmetry. Unless special meth- decide by tilting your specimen) picked out as a special direction.
ods are employed, X-ray methods will not detect the absence of For example, the crystal may have a mirror at 45 degrees to the
a center of symmetry. And materials that cannot be made without direction of the beam, but the experiment (for this chosen beam
a high density of twins will often be given incorrect symmetry direction) does not have such a mirror. The operation of the mir-
assignments by X-ray methods. ror would change the direction of the beam from vertical to hori-
zontal. Since a symmetry operation leaves the object in question
Convergent-beam diffraction, in contrast to X-ray diffraction, is unchanged, it cannot be a symmetry operation of the experiment.
very sensitive to small deviations from symmetry and, in many Taken together, these different symmetries mean that the same
cases, is the most sensitive method available (so far) for identify- crystal gives different symmetries to the experiment when it is at
ing the symmetry of a crystal. It is also important that CBED can different orientations with respect to the beam.
determine the symmetry of a grain down to micrometer size. This
higher spatial resolution increases the value of CBED because Point-Group Determination.
it can give a good result for the symmetry of a structure which
is not available in grains large enough to make other techniques
possible. The symmetry of the pattern leads to the symmetry of the
experiment which leads to the symmetry of the crystal.9
 4.2 Point-Group Determination 105

Perhaps surprisingly, a mirror plane perpendicular to the beam between reflections that are in the zero-order Laue zone. If
can be a symmetry operation of the experiment. Although the these effects are so much stronger than diffraction effects in-
operation of such a mirror would send the beam upwards in- volving reflections in other Laue zones that the latter may be

Part I
stead of downwards, the use of the reciprocity theorem (Pogany neglected, then the symmetry of the diffraction experiment is
and Turner 1968) allows us to reverse the direction of the beam the symmetry of an experiment using, not the crystal, but a pro-
without changing the experiment. jection of the crystal. In that case, the symmetry is the symme-
try as if the crystal were a two-dimensional object, rather than
When we take all of these factors into account, we can show a three-dimensional object. The symmetry of the experiment,
that there are 31 different possible symmetries for a diffraction in this case, is called the ‘projection diffraction group’ (instead
experiment on a crystal. We call these the ‘diffraction groups’ of the diffraction group). There are ten projection diffraction
(Buxton et al. 1976). groups. If, in the CBED patterns we obtain, there are no visible
three-dimensional effects, we are unable to determine the dif-
As we indicated above, each different crystal symmetry is as- fraction group – we can determine only the projection diffrac-
sociated with several different diffraction groups, depending on tion group. The three-dimensional diffraction effects are those
the orientation of the crystal with respect to the beam. The point effects that involve diffraction out of the zero-order Laue zone.
symmetries of crystals are the 32 point groups. There is no spe- Within the patterns in the zero-order Laue zone, they appear as
cial significance of which we are aware to the fact that there are diffraction effects, which we recognize as HOLZ lines.
31 diffraction groups and 32 point groups. That the numbers are
so close is just a coincidence. Pattern Symmetries

At any zone axis then, the CBED experiment will have a sym-
metry which corresponds to one of the 31 diffraction groups. We can identify more than one symmetry in a single
This determines the symmetry of the patterns observed. The CBED pattern. 9
symmetry of the patterns is related to the symmetry of the
experiment, which is, in turn, related to the symmetry of the
crystal. Thus, in looking at a single CBED pattern, we can recognize as
many as four distinct symmetries (each of them one of the ten
Now, the CBED patterns themselves can have only one of the two-dimensional point groups):
two-dimensional symmetries compatible with crystallographic
symmetry. There are ten of these: 1, 2, 3, 4, 6, m, 2mm, 3m, 1. The symmetry in the bright-field disk including HOLZ lines
4mm, 6mm. They are the two-dimensional point groups. So, or the effects of HOLZ diffraction in general.
since the patterns we observe can have only one of ten sym- 2. The symmetry in the bright-field disk when HOLZ lines are
metries, it would seem that we cannot expect to get very far absent or when the HOLZ lines are ignored.
in determining which of the 31 diffraction groups applies to 3. The symmetry of the whole pattern, including diffraction out

Chapter 4
our experiment. However, the patterns we observe have more as far as a higher-order Laue zone. In determining this sym-
complexity than this would suggest and, often, we can learn a metry the symmetry of Kikuchi lines can (and should) be
lot about the symmetry of the experiment from a single CBED included.
pattern. This is because the symmetry in the bright-field CBED 4. The symmetry of the reflections in the zero-order Laue zone
disk is not necessarily the same as the symmetry in the pattern when there are no visible effects from HOLZ diffraction or
as a whole. Thus two, two-dimensional point-group symme- when such effects are discounted.
tries may be available to us from one experimental pattern.
There is another aspect of complexity in CBED patterns. In You’ll find it helpful to use a little table, Table 4.1. For any
CBED, the strongest effects come from dynamical diffraction zone axis pattern, we fill in as many of the four spaces in the

Table 4.1 Matrix to show the symmetries in a zone axis pattern

Symmetry of the Symmetry of the Use this row for..

bright-field disk whole pattern

Three-dimensional symmetry – including HOLZ diffraction Diffraction group

Two-dimensional symmetry – excluding HOLZ diffraction see note for this box Projection diffraction
group

Note: in this box, the ‘whole pattern’ is restricted to the zero-order Laue zone. In the box above, it refers to the whole pattern extending out as far as the
pattern extends (beyond the zero-order Laue zone).
 106 4 Convergent-Beam Electron Diffraction: Symmetry and Large-Angle Patterns

table as possible. Depending on the zone axis, we may be able axis. Part of the reason for this is that CBED patterns will
to fill in all four slots, or – more likely – only some of them, be needed at several zone axes anyway, in order to carry out
because the information is not always present to allow us to the second part of the symmetry determination: finding the
Part I

assign all four symmetries. Almost any combination of entries space group.
is possible depending on the crystal and the experimental con-
ditions. However, first we have to complete the determination of the
point group. As indicated above, our aim is to establish the dif-
We take the entries (or entry, as the case may be) in the upper fraction group at each of several zone axes, or at least to get,
row in Table 4.1 and compare them with the entries for each of for each zone axis, a small set of possible diffraction groups.
the 31 diffraction groups in Table 4.2. This will give some set In general this information can be used to assign a unique
of possible diffraction groups. We take the entries (or entry) on point group for the crystal. This is done using Tables 4.4 and
the lower row of Table 4.1 and compare them with the entries in 4.5.
Table 4.3. This will give some set of possible projection diffrac-
tion groups. This may further restrict the set of possible diffrac- Table 4.4 is a matrix. Along the horizontal axis are listed the
tion groups. This is because each diffraction group is linked to 32 point groups; up the side are listed the 31 diffraction groups.
a particular projection diffraction group, as indicated by column Squares are shaded only if the point group can (at some orien-
four in Table 4.2. tation or orientations) give rise to that diffraction group. Con-
versely, if we know the diffraction group, it tells us which point
groups could have given rise to that experimental symmetry.
Diffraction Groups and Projection Diffraction Groups
Table 4.5 repeats the same information but adds to it. It tells us
the zone axes at which each kind of diffraction group is found.
In an experiment with a given diffraction group, there is Since this table tells us which crystal orientations give each of
only one possible projection diffraction group. 9 the possible diffraction groups, it can be used to calculate the
angles between zone axes with designated diffraction groups.
Table 4.5 may also be helpful in determining the simplest way
We should point out that the results given in Tables 4.2 and to resolve ambiguities between point groups. (A note of caution
4.3 are based on the assumption that the sample has parallel when using Table 4.5: there are some point groups in which
surfaces and is oriented perpendicular to the electron beam. the indexing depends on which of the space groups is applica-
In practice, samples are frequently wedge-shaped and are ble. For example, space groups P6N m2 and P6N 2m are both space
generally tilted. As a result, the observed symmetries may be groups with point group 6N m2 but their entries, in Table 4.5, for
reduced with respect to the symmetries expected. However, N and h1100i
h1120i N would be interchanged, as would those for
except for samples very steeply tilted or in the form of a wedge [uu.w] and [uNu:w]. Similar cautions apply to the following point
with a sharp angle, these effects are often small and may be groups: 3Nm, 32, 4N 2m, and mm2.)
neglected. These surface-orientation effects are more promi-
Chapter 4

nent when the surfaces of the sample are atomically smooth Here is an example of how this works. If we found a zone axis
and caution should be used in interpreting symmetries of such with diffraction group 6mRmR, Table 4.4 would tell us that only
samples. point group 622 is possible for the crystal. However, if we had
found a symmetry of 2mRmR, there would be six possible point
If all four boxes in Table 4.1 were filled in, then we have all groups and we would need information from other zone axes to
the information that can be obtained about the crystal symme- get a unique assignment. Clearly (from Table 4.4) it can be seen
try at this zone axis (but see the discussion of the symmet- that we get more valuable information if the symmetry of the
rical many-beam excitation method in Sect. 4.5.1). It is pos- pattern is high.
sible that this is enough to determine the point group of the
crystal, though it does not happen often. It is more likely that If we had found 6mRmR at one zone axis and 2mRmR at another,
several diffraction groups remain possible after this analysis, we can see from Table 4.5 that these two zone axes must have
either because we could not assign symmetries to some of the been 90 degrees apart. (If this is not clear, note that 6mRmR can
boxes in Table 4.1, or because there are diffraction groups that come only from point group 622 at the [0001] zone axis and
have the same symmetries in all four boxes (2 and 21R, 2mm that 2mRmR could be at h1120i
N or h1100i
N , both of which are 90o
and 2mm1R, 4 and 41R, 4mm and 4mm1R, 6 and 61R, 6mm away.) These ideas make it clear that it will be to our advantage
and 6mm1R). Further work at the microscope can help reduce to record the goniometer settings each time we take a zone axis
this ambiguity. There are two approaches: do more work at pattern, so that we can compare the experimental angles between
the same zone axis or tilt to a completely different zone axis. zone axes with the angles given by the table.
We recommend the latter as a general rule. It is more effec-
tive to leave the analysis incomplete at one zone axis but to In summary, the information we get on the diffraction groups at
get information at several other zone axes than it is to work several zone axes is used in conjunction with Tables 4.4 and 4.5 to
harder to squeeze that last bit of information at the first zone eliminate all possible point groups but one. As will be seen later, it
 4.2 Point-Group Determination 107

Table 4.2 Diffraction group symmetries Table 4.3 Projection diffraction group symmetries

Diffrac- Symmetry in Symmetry of Corresponding Projection Symmetry in the Symmetry of the

Part I
tion the bright- the whole of projection diffrac- diffraction group bright-field disk zero-order Laue
group field disk the pattern tion group when 3D effects zone, 3D effects
1 1 1 1R are absent/ignored absent/ignored

1R 2 1 1R 1R 2 1

2 2 2 21R
2R 1 1 21R
21R 2 2 21R 21R 2 2

mR m 1 m1R
m m m m1R
m1R 2mm m m1R m1R 2mm m

2mRmR 2mm 2 2mm1R

2mm 2mm 2mm 2mm1R
2RmmR m m 2mm1R
2mm1R 2mm 2mm 2mm1R 2mm1R 2mm 2mm

4 4 4 41R
4R 4 2 41R
41R 4 4 41R 41R 4 4

4mRmR 4mm 4 4mm1R

4mm 4mm 4mm 4mm1R
4RmmR 4mm 2mm 4mm1R
4mm1R 4mm 4mm 4mm1R 4mm1R 4mm 4mm

Chapter 4
3 3 3 31R
31R 6 3 31R 31R 6 3

3mR 3m 3 3m1R
3m 3m 3m 3m1R
3m1R 6mm 3m 3m1R 3m1R 6mm 3m

6 6 6 61R
6R 3 3 61R

61R 6 6 61R 61R 6 6

6mRmR 6mm 6 6mm1R
6mm 6mm 6mm 6mm1R
6RmmR 3m 3m 6mm1R
6mm1R 6mm 6mm 6mm1R 6mm1R 6mm 6mm
Symmetries of on-axis zone-axis convergent-beam patterns for exper-
iments with the given diffraction group. The fourth column gives the
projection diffraction group for the experiment if three-dimensional
effects are absent or can be ignored in the experiment.
 Chapter 4 Part I

108
Table 4.4 Diffraction group symmetries for each point group

6mm1R

3m1R
6mm
6mRmR
61R
31R
6
6RmmR
3m
3mR
6R
3
4mm1R
4RmmR
4mm
4mRmR
41R
4R
4
2mm1R
2RmmR
2mm
2mRmR
m1R
4 Convergent-Beam Electron Diffraction: Symmetry and Large-Angle Patterns

m
mR
21R
2R

2
1R
1

1N
3N
6N

4N

1
2
4
3
6

m
32
23

N
m3N
N

3m

3m

222
422
622
432

2/m
4/m
6/m
N

N
N

6m2

42m
43m
m3m

mm2
4mm
6mm

mmm
4/mmm
6/mmm

The diffraction groups for the experiments are given up the left-hand side. Point groups of the crystal are given along the bottom. If a square at the intersection of a particular point group with
a particular diffraction group is shaded, then a crystal with that point group can give rise to an experiment in which that diffraction group occurs. If the square is not shaded, then that diffraction
group cannot occur in an experiment with a crystal with that point group.
 4.3 Space-Group Determination 109

is very good to have more than the minimum information to deter- like the 200 reflection in silicon – which we will not discuss
mine the point group. If there is redundant information, we can, by further in this chapter.) By definition, all points in a lattice are
showing consistency, be more confident in our result. equivalent. It is therefore always possible to select, for any

Part I
crystal lattice, a primitive unit cell with only one lattice point
per cell. However, in many cases we do not use the primitive
unit cell. For example, in the case of a face-centered cubic
4.3 Space-Group Determination structure, we use a cubic unit cell with a volume four times
the volume of the primitive cell. Thus, when we calculate the
reciprocal lattice from this unit cell, we end up with a unit cell
There are 32 point groups and 230 space groups. There are up (in reciprocal space) which is four times smaller than the unit
to 28 space groups associated with each point group. So deter- cell (in reciprocal space) calculated from the primitive unit
mining the point group is only the first step towards determining cell (in direct space). Since Bragg’s law tells us that we get a
the space group. The possibility of several different crystal sym- diffracted beam for every scattering vector equal to a recipro-
metries (space groups) having the same point-group symmetry cal lattice vector, this would appear to tell us that, if we use
arises from the possibility that the crystal may have a screw the conventional cubic unit cell, there will be four times more
axis instead of a rotation axis and/or may have a glide plane reflections than if we use the primitive cell. Clearly this cannot
instead of a mirror. In the point-group operations, a screw is not be so. We cannot change the intensity in a beam by changing
distinguished from a rotation axis, nor a glide from a mirror. the way we think about the crystal. If we use a conventional
The problem then is to detect and identify screw axes and glide unit cell that is larger than the primitive unit cell for the crys-
planes. This is done through the study of ‘forbidden reflections.’ tal, there will be extra points in the reciprocal lattice but there
can be no diffraction associated with those points. The famous
Space-Group Determination rule – for a face-centered cubic crystal, reflections which do
not have hkl all even or all odd will not be present – takes away
the three extra ‘reflections’ in each unit cell of reciprocal space
The identification of a space group consists in finding that we artificially introduced by making the unit cell (in direct
which of the mirrors in the point group are glide planes space) larger than it need be. It should be clear from this that
in the crystal and which of the rotation axes in the point these reflections never appear. They cannot appear in kinematic
group are screw axes in the crystal. 9 diffraction but they also cannot appear by double diffraction.
Changing the way we choose to think about the unit cell cannot
give intensity at a place where it would otherwise be absent;
In a crystal which contains a screw axis (not parallel to a cor- not by double diffraction, nor by triple diffraction, nor by any
responding rotation axis) or which contains a glide plane (not other trick.
parallel to a corresponding mirror), some reflections will have
a structure factor which is identically zero. This is a result

Chapter 4
Unconditional Forbidden Reflections
which arises from the symmetry. This means that there will be
zero intensity in that diffracted beam as long as the diffraction
is kinematic. In X-ray diffraction, the intensities will be zero The forbidden reflections that arise from the choice of a
and the absence of the reflection is taken as an indication of non-primitive unit cell can never be seen in any diffrac-
the presence of the appropriate screw axis or glide plane. In tion experiment. 9
electron diffraction, where the diffraction is almost always dy-
namical, these kinematically forbidden reflections often have
substantial intensity because of double diffraction. However, The rules that govern which points in the reciprocal lattice give
as will be explained in Sect. 4.3.2, in CBED, we have another, rise to diffraction and which do not, are rules that remove a com-
more powerful, way of detecting and analyzing the forbidden plete three-dimensionally periodic set of points. That is why dou-
reflections. First though, we need to make clear that there are ble diffraction cannot add two allowed vectors to give a vector
two kinds of forbidden reflection, and explain the properties that is not allowed.
of each.
By contrast, the kinematically forbidden reflections that arise
from the presence of screw axes and glide planes do not consti-
tute a three-dimensionally periodic set. For example, if the crys-
4.3.1 Forbidden Reflections tal contains a two-fold screw axis, along the line of reflections
that runs through the origin and along the direction of the screw
axis, every second reflection is kinematically forbidden (has a
There are two main kinds of forbidden reflections: those which structure factor of zero). All the other reflections are allowed.
arise from the use of non-primitive unit cells, and those which Figure 4.1 contrasts these two situations. Figure 4.1a shows the
arise from screw and glide symmetries. (There is a third kind diffraction pattern down a 〈211〉 zone axis for a face-centered
that arises from atoms in special positions in the unit cell – cubic crystal. The x’s are forbidden reflections while the solid
 110 4 Convergent-Beam Electron Diffraction: Symmetry and Large-Angle Patterns

Table 4.5 Diffraction groups at each zone axis of each point group

〈111〉 〈100〉 〈110〉 〈uv0〉 〈uuw〉 [uvw]

Part I

m3m
N 6RmmR 4mm1R 2mm1R 2RmmR 2RmmR 2R

43m
N 3m 4RmmR m1R mR m 1

432 3mR 4mRmR 2mRmR mR mR 1

〈111〉 〈100〉 〈uv0〉 [uvw]

m3N 6R 2mm1R 2RmmR 2R

23 3 2mRmR mR 1

[0001] N
h1120i N
h1100i [uv.0] [uu.w] [uNu:w] [uv.w]

6/mmm 6mm1R 2mm1R 2mm1R 2RmmR 2RmmR 2RmmR 2R

6m2
N 3m1R m1R 2mm m mR m 1

6mm 6mm m1R m1R mR m m 1

622 6mRmR 2mRmR 2mRmR mR mR mR 1

[0001] [uv.0] [uv.w]

6/m 61R 2RmmR 2R

6N 31R m 1

6 6 mR 1

[0001] N
h1120i ŒuNu:w [uv.w]

3m
N 6RmmR 21R 2RmmR 2R

3m 3m 1R m 1
Chapter 4

32 3mR 2 mR 1

[0001] [uv.w]

3N 6R 2R

3 3 1

[001] 〈100〉 〈110〉 [u0w] [uv0] [uuw] [uvw]

4/mmm 4mm1R 2mm1R 2mm1R 2RmmR 2RmmR 2RmmR 2R

42m
N 4RmmR 2mRmR m1R mR mR m 1

4mm 4mm m1R m1R m mR m 1

422 4mRmR 2mRmR 2mRmR mR mR mR 1

[001] [uv0] [uvw]

4/m 41R 2RmmR 2R

4N 4R mR 1

4 4 mR 1
 4.3 Space-Group Determination 111

Table 4.5 (Continued)

Part I
[001] 〈100〉 〈u0w〉 [uv0] [uvw]

mmm 2mm1R 2mm1R 2RmmR 2RmmR 2R

mm2 2mm m1R m mR 1

222 2mRmR 2mRmR mR mR 1

[010] [u0w] [uvw]

2/m 21R 2RmmR 2R

m 1R m 1

2 2 mR 1

[uvw]

1N 2R

1 1

Fig. 4.1 Schematic representations a b

of forbidden reflections in diffrac-
tion patterns. a 〈211〉 zone axis for a
face-centered cubic crystal. b Pattern
from a crystal with a screw axis. The
reciprocal lattice points marked by x’s
are kinematically forbidden 000 000

Chapter 4
circles are allowed. The periodic forbidden reflections are peri- 4.3.2 Black Crosses
odic in three dimensions, not just two. It is not possible to find
a combination of allowed vectors that will reach a forbidden po-
sition. In Fig. 4.1b, the diffraction pattern is shown for a crystal Although double diffraction can produce intensity in reflec-
having a two-fold screw axis along the direction marked by an tions that are kinematically forbidden, it is not without effects
arrow with only half an arrowhead (this is the conventional sym- from symmetry. It turns out that for each double diffraction
bol for a screw axis). The forbidden reflections are marked with route like that shown in Fig. 4.1b, there is another double dif-
x’s. They lie along only one line. The allowed vectors shown on fraction route that will have equal amplitude but opposite phase
the diagram can be added to give a forbidden reflection. – provided that the beam orientation is at an exact symmetry
condition. The result of this opposite phase is that the double­
The conclusion to be drawn from this analysis is that, in elec- diffraction routes cancel and a zero of intensity is obtained
tron diffraction, there will generally be diffracted intensity at the (but only at that special symmetry condition). This effect is
positions of ‘forbidden reflections’, provided that the forbidden not clear in a selected-area diffraction pattern but in a CBED
reflections arise from screw axes and glide planes and not from pattern it becomes obvious. There is intensity in the CBED
the choice of a nonprimitive unit cell. There will be intensity disk from double diffraction but, at certain positions within the
in the positions of the reflections we wish to use to identify the disk – along the symmetry lines – there is no intensity. These
space group. So it is not the absence of the reflection we will features are known by several names: dynamical extinctions;
use; rather, it is a special feature of this double-diffracted inten- black crosses; dark bars; Gjønnes–Moodie lines, GM lines, and
sity that can be seen in CBED patterns. For a more complete GS lines. They are called dynamical extinctions because they
classification of all of the subtleties of forbidden reflections see are zeros of intensity even when multiple diffraction occurs;
Ciston et al. (2008). they are extinctions even when diffraction is dynamical. They
 112 4 Convergent-Beam Electron Diffraction: Symmetry and Large-Angle Patterns

Fig. 4.2 Convergent-beam patterns

at the [110] zone axis of FeS2. a Cen-
tered on the zone axis, showing GM
Part I

lines in each alternate reflection

along the row. b The sample has been
tilted slightly off the zone axis to
bring the reflection to the left of the
bright-field disk to the exact Bragg
condition. Now both extinction lines,
making a black cross, can be seen

are called Gjønnes–Moodie lines or GM lines for the people a dynamical extinction. Just as the extinction is independent
who first codified their occurrence correctly (there were earlier of the sample thickness, the extinction is independent of beam
papers that got part of the story) (Gjønnes and Moodie 1965). voltage. If there is a suggestion that the dark feature seen is not
They are called black crosses and dark bars colloquially be- a dynamical extinction, then another way to check is to change
cause that is what they look like. Goodman (2001) calls them the operating voltage of the microscope and see whether the
GS lines for ‘glide and screw’ lines. We will refer to these fea- feature is retained. This is somewhat inconvenient and is not
tures as dynamical extinctions or GM lines. Figure 4.2 shows usually necessary.
two CBED patterns. Figure 4.2a is at the zone axis orientation.
Every second disk along the row through the origin has a radial GM Lines and Not GM Lines
dark line through it. Figure 4.2b is tilted so that one of the re-
flections is at the Bragg orientation. It has two black lines, one
radial and one tangential at the locus of the Bragg condition, Dynamical extinctions will be present for all thicknesses
forming a black cross. The conditions for the occurrence of of sample and all voltages of the microscope.
these dynamical-extinction features are somewhat complex. If the zero of intensity (the extinction) is not present
In some cases the lines of extinction are truly dark. In other across every second disk along the row and across the
cases, intensity may be present as a result of double diffraction whole width of each those convergent-beam disks, it is
via a HOLZ reflection rather than double diffraction within the not a GM line. 9
zero-order Laue zone. There are many circumstances when one
arm of the cross is truly dark, and the other has intensity from
Chapter 4

HOLZ double diffraction. However, the intensity introduced by Dynamical extinctions (GM lines) occur in three cases:
double diffraction via a HOLZ reflection is generally so weak
that both arms of the cross look dark. It is also worth noting 1. When there is a two-fold screw axis along the direction
that the diffuse scatter that is always present in diffraction pat- of the row of GM lines.
terns (especially if they have not been energy filtered) will be 2. When there is a glide plane oriented so that the plane
present in the regions of extinction. contains both the direction of the row of GM lines and
the direction of the zone axis (the direction of the beam).
This raises the question: how do we recognize a dynamical 3. When both a screw axis in the orientation of 1, and a glide
extinction? It is not as difficult as it might seem from the issues plane at the orientation of 2 are present.
raised in the previous paragraph. Two things help us: the extinc­
tion features will be present and in the same places regardless In short, when you see a line of black crosses, it implies either a
of the sample thickness; and the features must occur in every horizontal screw axis or a vertical glide plane or both.
second reflection along a row which includes the direct beam,
of a zone axis CBED pattern. If there is something that looks If you want to use this information to determine the space group
like a dynamical extinction but which disappears when the pat- of the sample, you must first determine which of the three cases
tern is obtained from a different area of the sample (a different applies. This you can do in a straightforward way, provided that
thickness) it is not a GM line. However, the appearance may the point group of the crystal has already been determined. If
change. The width of the line is a function of thickness: the you know the point group, then the diffraction group of the zone
thicker the specimen, the narrower the line of extinction. If axis diffraction pattern is known. Table 4.6 gives the meaning
there is something that looks like a dynamical extinction in of the dynamical extinction for each diffraction group. The way
one disk but not in other alternating disks along the row, it is in which these ideas are presented here was developed in Eades
not a dynamical extinction. There is a third check that you can (1988b); Tanaka et al. (1983b) and Tanaka and Terauchi (1985a)
use to confirm that a dark feature in a CBED disk is, in reality, have given a more thorough, and somewhat more detailed, char-

Licensed to ([email protected])
 4.3 Space-Group Determination 113

Table 4.6 The meaning of dynamical extinctions

Single row of Perpendicular rows Deduction Tanaka Goodman

Part I
GM lines of GM lines symbol symbol

For the diffraction group given … … the deduction is:

mR 2mRmR Screw axes, parallel to each row of GM lines A 2B 2 S

2mRmR 4mRmR B3

m 2mm Glide planes, parallel to each row of GM lines (and to the zone axis) A 2B 2 G
2m 4mm A3

2RmmR 4RmmR Glide plane, if parallel to the whole-pattern mirror A 2B 2 G

A3
or or or
Screw axis, if perpendicular to the whole-pattern mirror A 2B 2 S
B3

2RmmR Glide plane, parallel to the whole-pattern mirror (and to zone axis) A 2B 2 G
A3
and and and
Screw axis, perpendicular to the whole-pattern mirror A 2B 2 S
B3

m1R 2mm1R Glide plane, parallel to each row of GM lines (and to zone axis) A 2B 2 GS
2mm1R 4mm1R and A 3B 3
Screw axis, parallel to each row of GM lines
See note

Note: these rules apply when the dynamical extinctions are features of diffraction in the zero-order Laue zone. If there is a glide plane perpendicular to
the zone axis, then every second reflection in the zero-order Laue zone has zero intensity from diffraction in the zero-order Laue zone (see Sect. 4.6.2).
Such reflections can have intensity only due to diffraction via HOLZ reflections. The table does not apply to extinctions in these latter cases.

acterization of dynamical extinctions. Symbols in Table 4.6 refer each pattern which has GM lines, use the table to deduce the
to a classification of dynamical extinctions given in these publi- presence of two-fold screw axes and/or glide planes.
cations of the Tanaka group. A more technical and definitive ac-

Chapter 4
count is given by Goodman (2001), who used different symbols, From the information you accumulate from all the zone axes,
also given in the table. prepare a summary showing which of the point-group symme-
try operators are symmorphic and which are nonsymmorphic.
The table considers two cases: the CBED pattern contains a sin- Symmorphic means that a translation is not involved in the
gle row of GM lines; or the pattern contains two perpendicular symmetry operation: mirror, rotation axes, inversion, and inver-
rows of GM lines. These are the only two possibilities. We use sion axes. Nonsymmorphic means that a translation is involved:
the table as follows. From knowledge of the crystal point group glide planes and screw axes. This summary can be presented
and the indexing of the zone axis, you can determine the dif- in different ways but we have found it convenient to use a ste-
fraction group corresponding to the experimental pattern. Find reographic projection representation of the symmetry elements.
that diffraction group in the first or second column of the table
(depending on whether it has one or two rows of GM lines) and Then you can compare this summary with the listing of the space
read the result in the ‘Deduction’ column. groups associated with the known point group. The easy way
to do this is to turn the pages of the International Tables and
compare the information you have with the symmetry of each
possible space group. If all is well, you will find that only one
4.3.3 Complete Procedure for Space-Group space group will match.
Determination
There are many international tables, but – in our field – the term
“International Tables” refers always to The International Tables
With these preliminaries established we can now summarize the of Crystallography or the earlier version called The International
overall procedure. At the microscope, take on-axis CBED pat- Tables for X-ray Crystallography. This is a set of several vol-
terns at as many different zone axes as possible. Study each of umes. The one you need for this work is the first volume in the
the resulting patterns to see which of them have GM lines. For set (Hahn 2002).
 114 4 Convergent-Beam Electron Diffraction: Symmetry and Large-Angle Patterns

Fig. 4.3 [010] zone axis pattern from a b

Ni3Mo. This pattern and all the pat-
terns from Ni3Mo in this chapter were
Part I

taken at 120 kV. In Figs. 4.3–4.8, pat-

terns a and b were obtained under
the same conditions except that the
figure a was taken at a longer camera
length to reveal more detail near the
center of the pattern. These patterns
were obtained on film. As a result, the m
central region of the short-camera-
length patterns (b) is overexposed
and may give a false impression. For
example, there is a contrast reversal
in the center of the short-camera-
length pattern of Fig. 4.4

m B = [010] m

When doing symmetry analyses, it will not always be the case

4.4 Ni3Mo – A Worked Example that the patterns will have been indexed in advance. In this ex-
ample the patterns are indexed. This will not affect our symmetry
analysis, but will make the discussion convenient by allowing us
We will use the example of Ni3Mo to illustrate the determination to refer to features in the patterns.
of symmetry. The reason for this choice is that it incorporates
most of the problems that can arise. Ni3Mo is a phase which
occurs in nickel-based superalloys in turbine blades.
4.4.1 Ni3Mo – a Worked Example, Part I:
This example, which we have taken from a paper by Kaufman Point Group
et al. (1983), is based on patterns obtained from six zone axes in
Ni3Mo. These patterns are shown in Figs. 4.3–4.8. Two patterns
are given for each zone axis. One is of a longer camera length to Consider first the pattern at the [010] zone axis Fig. 4.3. (The
Chapter 4

show detail in the zero-order Laue zone (ZOLZ), while the other order in which we discuss the patterns has no influence on the
was taken with a shorter camera length so that the pattern is visible result.) The long-camera-length pattern on the left has a vertical
to a wider diffraction angle including the first-order Laue zone. mirror and a horizontal mirror, in both the bright-field disk and in
the pattern as a whole. Both have 2mm symmetry. Next we have
Not all of these patterns were obtained from the same grain. to decide which symmetries these are. Inspection of the patterns
The specimen cannot be tilted enough for that. The specimen shows that no fine lines produced by interactions with HOLZ re-
was poly­crystalline and different patterns were obtained from flections are visible. Therefore these are projection symmetries.
different grains. You must do this with caution because errors This gives us the result shown in Table 4.7.
would be introduced if the different grains used were of different
phases. Therefore when using more than one grain, it is essen- Now consider the short-camera-length pattern to the right of
tial to take patterns at several zone axes in each grain and thus the pair. We count all the information in the image both within
establish – through the overlap – that all grains are of the same the Bragg reflections and in the diffuse scatter between them.
structure. XEDS can also be helpful in establishing that different This symmetry is also 2mm, giving the result shown in Ta-
grains are of the same phase, but again you should use caution ble 4.8.
since different phases can have very similar compositions.
There is no way that this pattern can give us the three-dimen-
It is also worth remarking that, as a general rule, it is highly sional symmetry for the bright-field disk, so this is all the infor-
desirable to take more patterns than the necessary minimum. mation we have. Let us see what we can do with it.
Frequently you’ll find that the symmetry in some patterns will
be uncertain. Then having more than the minimum number of We have both bright-field projection symmetry and whole-pattern
patterns will allow you to use a self-consistency argument to projection symmetry: they are both 2mm. Only one projection
resolve doubts. diffraction group has this combination (see Table 4.3). The pro-
 4.4 Ni3Mo – A Worked Example 115

Fig. 4.4 [101] zone axis pattern from a b

Ni3Mo

Part I
m

B = [101]

Fig. 4.5 [100] zone axis pattern from a b

Ni3Mo. c is an enlargement of the
bright-field disk

Chapter 4
m m
c

B = [100]
 116 4 Convergent-Beam Electron Diffraction: Symmetry and Large-Angle Patterns

Fig. 4.6 [011] zone axis pattern from b

Ni3Mo a
Part I

B = [011]

Fig. 4.7 [001] zone axis pattern from b

Ni3Mo
a

m
Chapter 4

B = [001]
m

Fig. 4.8 [110] zone axis pattern from b

Ni3Mo a

B = [110]
 4.4 Ni3Mo – A Worked Example 117

Table 4.7 Symmetries from the left-hand pattern of Fig. 4.3

Symmetry of the Symmetry of the Use this row for..

Part I
bright-field disk whole pattern

Three-dimensional symmetry – including HOLZ diffraction Diffraction group

Two-dimensional symmetry – excluding HOLZ diffraction 2mm 2mm Projection diffraction group

Table 4.8 Symmetries for both patterns of Fig. 4.3

Symmetry of the Symmetry of the Use this row for..

bright-field disk whole pattern

Three-dimensional symmetry – including HOLZ diffraction 2mm Diffraction group

Two-dimensional symmetry – excluding HOLZ diffraction 2mm 2mm Projection diffraction group

jection diffraction group is 2mm1R. Table 4.2 tells us that there details look fine enough to be HOLZ effects but it is not clear.
are four diffraction groups associated with projection diffrac- We may not be sure whether to put 2mm for bright field into
tion group 2mm1R. They are 2mRmR, 2mm, 2RmmR and 2mm1R. the upper or lower box in the matrix. However, we can be
Of these, both 2mm and 2mm1R have 2mm three-dimensional sure that the 2mm for the whole pattern must be a projection
whole-pattern symmetry (the one piece of information we had symmetry because we already know that the 3D symmetry is
not yet used) – see column 3 of Table 4.2. Therefore we learn m, see Table 4.10.
that, for the [010] zone axis, two diffraction groups are possible.
In this case, we cannot get further than this. Sometimes it may be This is enough to determine that the projection diffraction group
worthwhile taking more pictures at the same zone axis to fill in is 2mm1R. Therefore the diffraction group is 2RmmR. 2RmmR can
the empty slots in the matrix. Or it may be valuable to tilt off the be the result of a diffraction experiment for a crystal with any
zone axis (see Sect. 4.5, below). However, in general, it is more one of nine point groups (Table 4.4).
cost effective to get information at other zone axes. In this case,
in particular, no new information would be obtained by getting For this specimen, we now know that, at one zone axis we have
the 3D bright-field symmetry – as you can check from the table. 2RmmR (nine possible point groups), and at another zone axis
we have 2mm1R (five possible point groups) or 2mm (two point
We can now make a list of all of the crystal point groups that can groups). Clearly, we need to see which point groups can give

Chapter 4
give rise to projection groups 2mm and 2mm1R – there are seven these symmetries at different orientations. From Table 4.4 we
of them, see Table 4.4. However, the number is still large and, in can see that only five point groups give the correct combination.
this case, we do better to go to a new zone axis. Each of the five point groups that can give rise to 2mm1R can
also give rise to 2RmmR. These are the possibilities that remain:
Consider the [101] zone axis Fig. 4.4. Let us start this time with mmm, 4/mmm, 6/mmm, m3N, and m3m N .
the short-camera-length pattern to the right. This has a horizontal
mirror but clearly does not have a vertical mirror. We could see At this point we would, in most cases, go on to look at other
this looking at the detail of the reflections in the HOLZ ring, but zone axes to reduce the options further. In this case, however, we
we do not need to look so closely. If there is symmetry, it will would find that the other four zone axes have the same diffraction
be in the diffuse scatter and in the Kikuchi lines as well as in group as one of the two we have already treated (see Table 4.11),
the Bragg reflections. The left-hand side of the pattern is very so no new information is obtained that way. However, we can
different from the right, so the whole pattern symmetry is m – instead look at a different kind of information. If the crystal had
and this is the three-dimensional symmetry (since it includes the one of these three point groups – 6/mmm, m3N, and m3m N – then, as
HOLZ data), Table 4.9. can be seen from Table 4.4 or 4.5, at some orientation there would
be a six-fold symmetry. And if the point group were 4/mmm at
We can go immediately to Table 4.2 and see, with that infor- some orientation there would be a four-fold axis. If the specimen
mation alone, that only three diffraction groups are possible: has been extensively tilted to (as near as possible) all orientations,
m, m1R, or 2RmmR. Only those three have m in column three then it is very hard to believe that such a symmetry would have
of the table. If we now take into account the left-hand pattern, been missed. Table 4.5 can also be used to show that, for example,
at a longer camera length, we can see that both the bright-field if the point group were 6/mmm, then the 2mm1R zone axes would
disk and the whole pattern have symmetry 2mm. However, be spaced 30 degrees apart. If you had taken goniometer readings
there is some uncertainty here. Are we looking at projection at each orientation at which the zone axis patterns were obtained,
symmetries or three-dimensional symmetries? Some of the you could check this. In this case, the patterns were spaced not
 118 4 Convergent-Beam Electron Diffraction: Symmetry and Large-Angle Patterns

Table 4.9 Symmetry from the right-hand pattern at the [101] zone axis

Symmetry of the Symmetry of the Use this row for..

Part I

bright-field disk whole pattern

Three-dimensional symmetry – including HOLZ diffraction m Diffraction group

Two-dimensional symmetry – excluding HOLZ diffraction Projection diffraction group

Table 4.10   Symmetries at the [101] zone axis

Symmetry of the Symmetry of the Use this row for..

bright-field disk whole pattern

Three-dimensional symmetry – including HOLZ diffraction m Diffraction group

Two-dimensional symmetry – excluding HOLZ diffraction 2mm Projection diffraction group

at 30 degrees but were, in fact, oriented at 90 degrees from each improbable. If the pattern has an ‘almost symmetry,’ as is the case
other (within experimental error – microscope goniometer stages for the [100] zone axis mentioned above, there are two possible
are not so very accurate). So for these reasons, we can exclude explanations: i) the crystal is close to having a symmetry but does
point groups 4/mmm, 6/mmm, m3N, and m3m N . not have it; or ii) the crystal does have the symmetry, but the ex-
periment in the microscope did not have the symmetry (normally
Putting all of this information together, we can conclude that because of the imperfect sample). So the question is, how do we
Ni3Mo has the orthorhombic point group mmm. To increase our distinguish between these two possibilities?
confidence in this result, we should now check the diffraction
group of each of the other four patterns and that they are separated There are two approaches. The first is experimental. At the zone
by angles consistent with Table 4.5. With this level of over-de- axis in question, you have to take lots of patterns (especially on
termination, if it all fits together, we can be very confident of the a microscope with a CCD camera when cost and convenience
result. are not an issue) at different places on the specimen. Also take
patterns at different focus settings. (The symmetry of the pattern
can also be broken if the beam is not focused in the specimen
plane.) It requires only one pattern to show an exact symmetry
4.4.2 Qualifications for the demonstration to be convincing. Even if all the others do
not show the symmetry, we can dismiss them as resulting from
Chapter 4

imperfect regions in the specimen.

Before going on to discuss the space group it is appropriate to
add a few further comments. If you look closely at several of the The second approach is to look at the nature of the symmetry
patterns in this set, you will see some anomalies. Let us take for breaking in the experimental pattern. Compare, for example, pat-
example the [100] zone axis Fig. 4.5. In the left-hand pattern look terns [100] and [011] Figs. 4.5 and 4.6. In the [100] pattern (as
at the 020 and 0 2N 0 disks. Clearly the one to the left, 02N 0, has a previously noted) you’ll see that the vertical mirror is broken by
white patch that is brighter than the corresponding patch in 020. the difference in brightness between the 020 and 02N 0 disks. On
Does this mean that there is no vertical mirror and that we must the other hand, in the [011] pattern, within the bright-field disk,
change our assessment of the symmetry? This question raises an the vertical mirror is broken by two vertical HOLZ lines, which
issue that is both a strength and a weakness of CBED as a method are not symmetrically placed. Why, in the first case, do we say
of symmetry determination. For the symmetry in the conver- that this is likely to be the result of an imperfect sample and we
gent-beam pattern to be perfect, the crystal must be perfect, free are prepared to believe that the vertical mirror is real, whereas, in
from strain and with parallel faces perpendicular to the beam. As the second case, we assert that the vertical mirror cannot be pres-
we’ve already noted, this is rarely the case in TEM specimens. If ent in the crystal? In some sense, this comes down to experience,
your specimen is tilted, wedge shaped, strained, or otherwise less which is unfortunate if you are the one who is lacking in experi-
than perfect, the convergent-beam patterns will tend to show bro- ence. However, perhaps you can see that, in the [100] case, the
ken symmetries. How then do we judge the symmetries of the pat- kind of distortion of the pattern is the kind of change that occurs,
terns, and hence the symmetry of the crystal? Let us start with the for example, with a change of thickness of the specimen. And if
converse. If the symmetry in the experimental pattern is perfect, the convergent beam were slightly under- or overfocused, that
we can assume that the crystal has the corresponding symmetry. would be enough to produce the effect. Conversely, in the [011]
Symmetries in convergent-beam patterns do not arise by accident. pattern, the position of the HOLZ lines would not be shifted by
We justify this statement by noting that the dynamical diffraction a poor experiment (they could change in contrast or sharpness,
at zone axes is so complex that an accidental symmetry is highly but hardly in position).
 4.4 Ni3Mo – A Worked Example 119

Table 4.11 Summary of the GM line analysis for Ni3Mo. The diffraction groups as determined from the convergent-beam patterns for the first three zone axes could
be 2mm 1R or 2mm. However, 2mm can be excluded because it is not compatible with 2RmmR (see text)

Part I
Zone axis Figure Diffraction group GM line Meaning Conclusion
from Table 4.6

[100] 4.5 2mm1R 010 Glide plane Glide plane on (001)

and screw axis Screw axis along [010]

[010] 4.3 2mm1R 100 Glide plane Glide plane on (001)

and screw axis Screw axis along [100]

[001] 4.7 2mm1R – –

[110] 4.8 2RmmR – –

[011] 4.6 2RmmR ??

[101] 4.4 2RmmR 010 Screw axis Screw axis along [010]

In addition to the two approaches to assessing which symme- 4.4.3 Ni3Mo – a Worked Example, Part II:
try to assign to a given convergent-beam pattern, you should Space Group
also note that a good experimentalist will always take patterns
at more zone axes than the minimum. Then the need for the
symmetries at each zone axis to be consistent with the other As discussed in Sect. 4.4.1, we have established that the point
zone axes, also helps us get it right. Overdetermination is a group of the phase is mmm. This has two-fold axes along three
virtue. orthogonal directions and mirrors perpendicular to each of those
directions. To complete the determination of the space group
Judging the Symmetry we have to establish which of those two-fold axes (in the point
group) is a screw axis (in the crystal) and which of the mirrors (in
the point group) is really a glide plane. To do this we go looking
The determination of symmetry by the methods given for GM lines in the CBED patterns.
here cannot be done on autopilot. You will have to use
your judgment and keep thinking as you analyze your We have six zone axes, so we look for GM lines in all of them.
results. 9 Again we would like overdetermination to gain confidence in our
result. Let us take them one at a time. (We will take them out

Chapter 4
of order to discuss the easy ones first.) At the [100] zone axis
In assigning the symmetries to enter into the matrix, it may also (Fig. 4.5) there is a very clear GM line in the reflections along the
be helpful to recognize that the symmetry in the bright-field disk horizontal row. It can be seen in reflections 010 and 01N0. You can
cannot be lower than the symmetry in the whole pattern. also see, by close examination of the short-camera-length pattern
to the right, that every second reflection along the row has a line
Finally, how do we find something good to say about this need through it. Note, however, that the vertical dark bar through the
to use our judgment instead of a clear-cut decision? The benefit 010 reflection is not a dynamical extinction (it is not at the Bragg
comes from the absolute need to recognize what we are seeing. If condition), it is just a coincidence that the contrast has that feature.
the pattern is ‘perfect’ we can assert with great confidence that the
crystal has the corresponding symmetry. If we see some symmetry At the [010] zone axis (Fig. 4.3), you can also see an obvious
breaking but make a judgment that the symmetry exists, we cannot GM line in the 100 and 1N00 reflections. In the [001] pattern
help but be aware of the doubtfulness of the process. And when we (Fig. 4.7) there are no GM lines (though there is a relevant fea-
write our papers, we can choose language that matches our level ture of interest – see Sect. 4.6.2). The [110] zone axis (Fig. 4.8)
of certainty. In the case of symmetry determination by CBED, is interesting. We might, at first, be tempted to say that there is
it is nearly impossible to claim more certainty than truly exists. dynamical extinction in the 001 reflection. However, look out to
the 003 reflection and you will see quite clearly that there is in-
tensity along the radial region. Therefore the lack of intensity in
the 001 reflections is not related to dynamical extinction. At the
[011] zone axis (Fig. 4.6), we have a problem. The 011N reflection
looks as though it might have a GM line but it is impossible to
be sure since the disks overlap and, in the overlap regions, we do
not know if the intensity comes from the 011N reflection or from
the reflections adjacent to it.
 120 4 Convergent-Beam Electron Diffraction: Symmetry and Large-Angle Patterns

Finally, we look at the [101] diffraction pattern (Fig. 4.4). At first a b

sight, we might think that there are two GM lines one horizontal
and one vertical. However, if you look more closely, you can see
Part I

that, in the 1N01 reflection, although there is no intensity along

most of the radial line, there is a small region at the inner edge
of the disk where, very clearly and precisely on the symmetry
line, there is intensity. So we can conclude that there is a GM
line only in the 010 reflection (and every second reflection along
that line).

How Many Patterns Do You Need? Fig. 4.9 a Conventional stereographic projection to represent the symmetry
elements of a crystal with point group mmm. b Symbols replaced with symbols
It should be clear from way this example is turning out showing the actual symmetry elements for Ni3Mo
that the more patterns you have at different zone axes, the
better. 9
space group that meets these requirements. It is the space group
Pmmn (number 59).
Let us turn all of this into a table to see where we stand. To com-
plete the table we will need to note that, for the [101] zone axis The final step of this process is to go back and check that all the
(Fig. 4.4), the line of GM lines is vertical but that the whole pattern experimental data are consistent with this conclusion. One such
mirror is horizontal. Thus, when we use Table 4.6 to interpret the check is to compare each zone axis pattern with the tables given
data, we assign the 010 GM lines as due to a screw axis along 010. by Tanaka and Terauchi (1985a, p. 162) or Goodman (2001).

The three zone axes where things are clear tell us that we have
two screw axes and one glide plane, and are consistent in doing
so. We have not used the other three zone axes for the moment. 4.5 Additional and Alternative
They are also consistent with these results but we will delay dis-
cussing that (see Sect. 4.6) until we have completed the exercise. Symmetry Methods
We can represent the result schematically (Fig. 4.9), replacing
the stereographic projection for the symmetry elements of the We recommend the method given above as the standard method
point group by one in which we show the space-group symmetry of crystal symmetry determination in the TEM. However, you
elements. should note that there are two additional methods that may be
valuable. Both have the potential advantage that they may reduce
Chapter 4

There are 28 space groups (numbered 47 to 74 in the Interna- the number of zone axes that you need to study. These methods
tional Tables – Hahn 2002) belonging to the point group mmm. are discussed in the following two sections.

To make further progress it is desirable to know whether the unit

cell is primitive or centered. This can easily be determined from
the way the planes in the reciprocal lattice project onto each 4.5.1 Symmetry Determination from Off-Axis
other (Ayer 1989). Details are given in W&C Chap. 21.3.C. In Patterns
this case we can see that the lattice is primitive by inspection,
for example, of the short-camera-length patterns at two zone
axes. In both the [010] zone axis, Fig. 4.3, and the [001] zone As we have discussed above, there are 31 possible symmetries
axis, Fig. 4.7, you can see that the reflections in the first-order of the diffraction experiments when an electron beam is inci-
Laue zone lie on a net that is vertically above the net formed by dent on a crystal in the transmission electron microscope, and we
the zero-order zone reflections. If the unit cell were centered the call these the diffraction groups. However, even in ideal circum-
net would be the same but be laterally displaced. Thus, we can stances when all the possible symmetry information is available
exclude centered space groups. This leaves us with 16 primitive in the pattern, it is not possible to uniquely identify the diffraction
space groups (numbered 47 to 62) among which the space group group from an on-axis diffraction pattern in every case. As you
of the sample is to be found. We next turn the pages of the In- can see from Table 4.2, there will always be some undecided
ternational Tables looking for the space group that has two of cases. This is because a CBED pattern taken at the exact zone axis
its axes as screw axes (with no two-fold axes parallel to them) orientation does not display all the symmetry of the experiment.
and one glide plane (with no mirror parallel to it). The other axis
will be two-fold and the other two planes will be mirrors. You Included among the symmetries of each diffraction group are
can see the results of this study in Table 4.12. There is only one symmetries within the diffracted beams. You cannot see these
 4.5 Additional and Alternative Symmetry Methods 121

Table 4.12 Screw axes and glide planes in the primitive space groups of the
mmm point group.

Part I
Space group Non-symmorphic ele-
ments

Number Symbol Full symbol

47 Pmmm P2/m2/m2/m no screws or glides Specimen

48 Pnnn P2/n2/n2/n three glides

49 Pccm P2/c2/c2/m two glides

50 Pban P2/b2/a2/n three glides

51 Pmma P21/m2/m2/a one screw, one glide

52 Pnna P2/n21/n2/a one screw, three glides

Fig. 4.10 If the semiangle, θ, of the convergent beam is equal to the Bragg
53 Pmna P2/m2/n21/a one screw, two glides
angle, θB, then – since the Bragg diffracted beam is deviated by 2θB – the disks
54 Pcca P21/c2/c2/a one screw, three glides just touch. For the diffracted beam to show both sides of the Bragg condition,
55 Pbam P21/b21/a2/m two screws, two glides the disks would have to overlap

56 Pccn P21/c21/c2/n two screws, three glides

57 Pbcm P2/b21/c21/m two screws, two glides all the beams that contribute to the overlap region so you cannot
58 Pnnm P21/n21/n2/m two screws, two glides see what the intensity for a particular beam is. Therefore you
cannot determine the symmetry this way.
59 Pmmn P21/m21/m2/n two screws, one glide

60 Pbcn P21/b2/c21/n two screws, three glides If you tilt the beam off the zone axis, you can arrange that the
61 Pbca P21/b21/c21/a three screws, three glides Bragg condition runs through the center of a diffracted disk
(see for example, Fig. 4.2b) and the internal symmetry for that
62 Pnma P21/n21/m21/a three screws, two glides
reflection can be assessed. If you do this, typically only one
reflection will be set up with the Bragg condition so that its
symmetry can be determined. The other reflections (or at least
symmetries in an on-axis pattern for the following reason. The most of them) will be tilted further away from the Bragg con-
symmetries in the diffracted beams are symmetries about the dition. This would appear to make it possible to determine the

Chapter 4
exact Bragg condition. To see these internal symmetries, you internal dark-field symmetries only one reflection at a time, and
need to see the diffraction on both sides of the Bragg condition. that would make the determination of the diffraction group a
However, this is not possible in an on-axis pattern. cumbersome business. It turns out not to be so difficult. Tanaka
and his group have shown that, by tilting to a specific off-axis
Figure 4.10 is a schematic diagram showing CBED patterns orientation, which they call a ‘symmetrical many-beam ex-
when the convergence angle has been increased as much as pos- citation’, all the relevant symmetry information for the zone
sible without overlap of the disks. It is easy to see that the semi- axis can be obtained (Tanaka et al. 1983a; Tanaka and Terauchi
angle of convergence, θ, is equal to the Bragg angle in this case 1985a). The idea is to tilt the beam to give an orientation where
(compare Fig. 20.3 in W&C). Therefore, the Bragg condition for several disks have the Bragg condition through the middle of
the diffracted beam is found at the inner edge of the correspond- the disk at the same time.
ing disk. The range of angles displayed in the disk all lie to one
side of the Bragg condition. There are four relevant nets (two-dimensional lattices) for the
reflections in the zero-order Laue zone: oblique, rectangular,
In order to see both sides of the Bragg condition, there are two square, and hexagonal. In the case of an oblique lattice the in-
possibilities: you can increase the convergence angle so that the formation you need can be obtained by setting up a two-beam
disks overlap, or tilt off the zone axis. (Although, in an on-axis condition for any reflection. That will reveal whether or not the
zone axis pattern, ZOLZ reflections do not include the Bragg diffraction group includes the operation 1R (which is the same as
condition, reflections in higher-order Laue zones do include the asking whether or not the crystal contains a mirror perpendicular
Bragg condition. Unfortunately this does not help us with sym- to the zone axis). We will not discuss oblique lattices further.
metry determination, as you can read in Eades 2001.) The for-
mer possibility does not work for symmetry determination. If the Figure 4.11 shows how you can tilt the beam to get the most
disks overlap, the intensity is a combination of the intensities of practical many-beam condition for complete symmetry determi-
 122 4 Convergent-Beam Electron Diffraction: Symmetry and Large-Angle Patterns

Fig. 4.11 Illustrating how to tilt to a b c

the symmetric many-beam condition
for rectangular (a), square (b), and
Part I

hexagonal (c) diffraction nets

o o o

nation. When the beam is at a zone axis orientation, the Ewald zone axis (because the scattering is less strongly dynamical). In
sphere is tangent to a plane in the reciprocal lattice. (Go back on-axis patterns, if HOLZ effects are not visible in the ZOLZ re-
and read W&C Chap. 12 if this statement isn’t obvious.) If the flections, you can usually see the three-dimensional symmetry in
beam is tilted slightly off the zone axis, the Ewald sphere cuts the HOLZ itself in short-camera-length patterns (see above). You
that plane in a circle, called the Laue circle. In Fig. 4.11, you can won’t find this recourse available when you tilt your specimen
see that the center of the Laue circle (which is also the projection off axis to the symmetrical many-beam condition.
of the center of the Ewald sphere onto the plane) is marked with
a cross. In the case of rectangular and square lattices, the beam Finally, in this section, it is worth adding a comment on ± g
is tilted so that the center of the Laue circle falls at the center of determinations. By this we mean taking one pattern tilted off
the rectangular or square array of reflections. The disk marked axis to the Bragg condition for reflection g, and then tilting off
o is the direct beam. For a hexagonal lattice, the beam is tilted axis in the reverse direction to bring reflection – g to the Bragg
until the center of the Laue circle lies on top of one of the closest condition. This was discussed, and examples were given, in the
reflections. The lines across the disks show the position of the paper by Buxton et al. (1976) and also in the work by Tanaka
Bragg condition for each reflection. and Terauchi (1985a). It might be said that this idea arose from a
historical accident. Because all the people involved in this work
Dark-field CBED disks have the center of the pattern (not neces- were brought up on defect analysis (g.b analysis, where pairs of
sarily a center of symmetry) at the point where the Bragg condi- images are taken under these conditions: W&C Chaps. 25 and
tion intersects the radial line through the zone axis. In Fig. 4.11 26), it was natural for them to look to see how symmetry was dis-
you can see that only the reflection to the right of each diagram played in pairs of diffraction patterns of this kind. However, after
includes the center of the pattern – only for that disk does the you have tilted the specimen (or the beam), it is very hard to be
radial line through the zone axis (the cross in the center of the sure that the beam is hitting exactly the same place on the spec-
pattern) go through the disk. The diffracted beams can have only imen as before. And, if it is not, then there you cannot guarantee
the following symmetry: a two-fold axis (which is the same as that the symmetry will be correctly displayed. For example, if the
Chapter 4

an inversion center in two dimensions), a radial mirror, a mirror patterns are taken from areas of different thicknesses, symmetries
along the Bragg condition, combinations of these, or no symme- may be broken. For this reason, we would not now recommend
try at all. Tanaka’s group has tabulated how these symmetries are the use of ±g patterns for symmetry determination, and details of
revealed in the four or six reflections shown in Fig. 4.11 (Tanaka these symmetries (which are given in earlier publications) have
et al. 1983a; Tanaka and Terauchi 1985a). In many cases you’ll been omitted here.
find that a single pattern is enough to determine the diffraction
group. In other cases you’ll need a combination of the symmet-
ric many-beam pattern with an on-axis pattern to determine the
diffraction group uniquely. Therefore, you may find it advan- 4.5.2 Symmetry from Precession Patterns
tageous to take a symmetric many-beam pattern as well as the
on-axis pattern. You should always take the on-axis pattern since
that is needed for the space-group determination. You’ll find it’s Precession diffraction patterns are a kind of diffraction pattern
not usually much extra work to take the symmetric many-beam introduced to transmission electron microscopy by Vincent and
pattern, and so it would seem to be a worthwhile thing to do. Midgely (1994). (W&C Chap. 18.8 and Spence Chap. 5, this
The drawback with this symmetric many-beam method is that volume) To acquire these patterns you have to make hardware
it relies on HOLZ effects being visible in all of the four or six additions to your standard TEM. For the patterns we discuss here
relevant reflections. (Otherwise, of course, the information is in connection with symmetry determination, a zone axis of the
reduced to the projection diffraction group.) Even in on-axis pat- crystal is aligned with the optical axis of the microscope, but the
terns HOLZ effects are not always visible in the ZOLZ. (Go back incident beam is tilted, typically by a couple of degrees. Keep-
and read W&C Sect. 20.8.B to remind yourself how to increase ing that angle constant, the beam azimuth is then rotated about
the chances of seeing HOLZ lines.) When you tilt the specimen the axis to form a conical scan. However, using deflection coils
off axis, the contrast from HOLZ effects is lower than at the around the beam after it has come through the specimen, the con-

Licensed to ([email protected])
 4.6 More on Glide Planes 123

ical scan is cancelled to keep the diffraction pattern stationary on 1. Use the positions of the reflections in the precession pat-
the screen and the camera. In a time-averaged pattern, obtained terns to determine the crystal class, and the centering of
in this way, we find that the intensity of the diffraction spot is the unit cell.

Part I
an average across the bend contour (or equivalently, an average 2. Use the extinctions (determined from the precession pat-
along the relrod) for that reflection. If you want to carry out such terns) to limit the number of possible space groups. There
an experiment, you’ll require at the very least additional software will be only a few of them.
to control your scan coils before and after the specimen, and in 3. Use the symmetry of convergent-beam patterns to distin-
many cases you’ll need additional hardware to make such control guish between the possibilities identified in step 2.
practical. The number of microscopes equipped to acquire such
patterns is rapidly growing, but is still rather small. You can find details of these methods in Morniroli and Steeds
(1992) and Morniroli et al. (2012). Morniroli (2011) has made
The initial interest in precession patterns in the TEM arose from available comprehensive tabulations of the use of precession
the fact that, because of the way the intensities are averaged, the diffraction for the determination of the unit cell and of crystal
intensity of each diffraction spot is rather closely related to the symmetry on his web site.
structure factor for that reflection. This is not true for other kinds
of diffraction pattern. The result is that precession patterns can
be used to solve crystal structures. (Solving a crystal structure
means finding the coordinates of each atom in the unit cell.) 4.6 More on Glide Planes
Again, because of the way the intensities of the reflections are
averaged through the relrods, we find that the reflections have When a crystal contains a glide plane (with no mirror parallel
intensities that are a better representation of the symmetry of to it) a set of reflections is kinematically forbidden. This is
the crystal than in conventional selected-area diffraction (which why glides can produce GM lines. The kinematically forbidden
for several reasons does a poor job of displaying symmetry). reflections for a glide plane are half of the reflections in the
Precession patterns are frequently obtained as ‘spot patterns,’ reciprocal lattice plane which passes through the origin and
that is to say that the convergence angle of the beam at the spec- is parallel to the glide plane. This is indicated schematically
imen is rather small, and the diffracted beams appear as spots in Fig. 4.12, which shows kinematically forbidden reflections
or very small disks. Thus, you’ll find that there is generally no in red. If the electron beam is in the direction A, then the ze-
visible intensity variation across the (very small) disks. The re- ro-order Laue zone contains the line of reflections a-aa, and
sult is that the patterns reveal the whole-pattern symmetry but the red reflections will have GM lines. However, if the incident
not the bright-field symmetry. Provided that the HOLZ effects beam were in the direction B, the reflections in the zero-order
are small enough, you’ll find that the patterns will also reveal the Laue zone would be those along the line b-bb. There are no
whole-pattern projection symmetry. kinematically forbidden reflections along this line and there
will be no GM lines at this zone axis. Therefore, at a zone axis

Chapter 4
From the symmetry point of view, a particular strength of the which is parallel to a glide plane, GM lines may appear, but
precession method is that it provides a very good way of identi- they may not appear if the choice of zone axis does not include
fying screw axes and glide planes from the forbidden reflections. the forbidden reflections in the diffraction pattern. This is one
Particularly if you have tilted the beam through a large angle (3o, reason why, in determining the space group, it is very important
for example), forbidden reflections will be absent from preces- to take patterns at many zone axes. In the case of Ni3Mo given
sion patterns. (See Fig. 18.16 in W&C.) This approach provides earlier, we see an example of this. The [110] zone axis (Fig. 4.8)
us with an alternative to the use of dynamical extinctions in the is parallel to the glide plane that has been found to be present,
determination of space groups. Another strength of precession but there are no GM lines for this reason. The [100] and [010]
diffraction is that it gives good symmetry information even zone axes which are parallel to the same glide plane do give
from thin specimens, when CBED may be more limited. A third patterns with GM lines.
strength is that it identifies all dynamical extinctions, not just
those that give GM lines. Indeed, adding the use of precession GM Lines from Glides
diffraction to the use of CBED provides an alternative method
of determining both space-group and point-group symmetries.
Not every zone axis that is parallel to a glide plane pro-
Because precession patterns allow you to identify all forbidden duces patterns with dynamical extinctions. 9
reflections, not only those that give GM lines, it proves valuable
to approach symmetry determination by a different route from
the standard convergent-beam method we have just described
above. When precession patterns are available, you can pursue
symmetry determination in three steps:
 124 4 Convergent-Beam Electron Diffraction: Symmetry and Large-Angle Patterns

B A
Part I

bb

a aa

Fig. 4.12 Schematic diagram of a diffraction pattern when there is a glide plane Fig. 4.13 Enlargement of part of Fig. 4.8 to show more clearly the presence
parallel to the page. Reflections in red are kinematically forbidden. See text for of a GM line in one of the HOLZ reflections, the one at the head of the arrow
details

4.6.1 GM Lines in HOLZ Reflections be so weak it will normally be overlooked.) So there will be no
intensity at all in any of the reflections that are kinematically
forbidden.
For a screw axis the kinematically forbidden reflections lie
along a single row in the reciprocal lattice, and GM lines occur Something to Look Out For
in reflections in the zero-order Laue zone (if the beam is inci-
dent perpendicular to the screw axis) or not at all. However,
for a glide plane parallel to the beam, there are kinematically You should always check the patterns you take at short
forbidden reflections throughout the reciprocal lattice plane camera lengths (which show higher-order Laue zones)
that is parallel to the glide. This means that kinematically to see that the spacings between the reflections in the
forbidden reflections can, and do, appear in reflections in zero-order zone are the same as in the higher-order zones.
higher-order Laue zones. There is an example in the Ni3Mo If they are different, it is telling you something. 9
patterns discussed. At the [110] zone axis (Fig. 4.8), the usual
set of GM lines is missing, as discussed in the previous para-
graph. However, close inspection of the figure shows that one The effect of this is that the spacing between reflections in the
Chapter 4

of the reflections in the first-order Laue zone is divided by a higher-order Laue zones is smaller than the spacing in the ze-
GM line (an enlargement of this region is given in Fig. 4.13 so ro-order zone. If no horizontal glide plane is present, the spac-
that it can be seen more clearly). This confirms the presence ings are the same in all zones. We can see an example of a glide
of the glide even though the GM lines are not present in the plane perpendicular to the beam in the [001] zone axis pattern
zero-order zone. (Fig. 4.7). Along the vertical direction, the spacing between re-
flections in the first-order Laue zone is half of that in the zero
layer. This is a clear signature of a glide plane perpendicular to
the zone axis. A systematic treatment of this topic is given in
4.6.2 Glide Planes Normal to the Beam Morniroli and Steeds (1992).

As discussed above, glide planes parallel to the incident beam

produce GM lines in kinematically forbidden reflections. This 4.7 Beyond Symmetry
means that there is no intensity along the symmetry directions.
That there is intensity in other directions within the conver-
gent-beam disk comes from double diffraction. If we now We’ve stated several times that there are 230 space groups. That
consider a zone axis with the beam normal to the glide plane, is, there are 230 different possible symmetries for a crystal and
the situation is different. In this case, Fig. 4.12 represents the the methods described in the previous sections of this chap-
zero-order Laue zone diffraction pattern and we can see that ter can uniquely identify most of them – but not all. Among
no double diffraction routes in the zero-order Laue zone can the 230 space groups there are 53 space groups that cannot be
give intensity in any of the kinematically forbidden reflections. uniquely identified by GM-line analysis, as shown in Table 4.13.
(Double diffraction routes into the forbidden reflections can (Note that a number of earlier publications gave incomplete lists.
occur via reflections in other Laue zones but the intensity will The first complete list was given by Tsuda et al. 2000.) Although
 4.7 Beyond Symmetry 125

Table 4.13 Space groups that may not be uniquely identified by GM analysis alone. A ‘C’ in the penultimate column indicates sets of space groups that can be dis-
tinguished using coherent convergent-beam diffraction (see text). A ‘P’ in the last column indicates sets of space groups some at least of which can be distinguished
using precession diffraction (see text)

Part I
Space group Space group
number symbol

A Pairs of space groups that cannot be distinguished by GM lines because 23 24 I222 I212121 C
they have the same symmetry elements 197 199 I23 I213 C

B Pairs of space groups that cannot be distinguished from each other 75 77 P4 P42 P
because one has a rotation axis while the other has a screw axis that does 79 80 I4 I41 P
not give GM lines 83 84 P4/m P42/m P
85 86 P4/n P42/n P
89 93 P422 P4222 C P
90 94 P4212 P42212 C P
97 98 I422 I4122 C P
207 208 P432 P4232 C P
209 210 F432 F4132 C P
211 214 I432 I4132 C P

C Pairs of space groups that cannot be distinguished because they are 76 78 P41 P43
enantiomorphous pairs 91 95 P4122 P4322
92 96 P41212 P43212
212 213 P4332 P4132

D Sets of three space groups that cannot be distinguished. In each set of 143 144 145 P3 P31 P32 P
three, the second and third space groups form an enantiomorphous pair 149 151 153 P312 P3112 P3212 C P
while the first space group of the set has a rotation axis 150 152 154 P321 P3121 P3221 C P
168 171 172 P6 P62 P64 P
177 180 181 P622 P6222 P6422 C P

E In each set of three, the second and third space groups form an enantio- 173 169 170 P63 P61 P65 P
morphous pair while the first space group has a screw axis that is not 182 178 179 P6322 P6122 P6522 C P
handed

these space groups may not be distinguished by GM-line anal- tified in one of three ways. First, if the specimen is very thin,
ysis, they may be distinguished by other methods: for example, kinematically forbidden reflections will be very weak or ab-
by finding kinematically forbidden reflections as suggested in sent because their intensity is dependent on double diffraction.

Chapter 4
the discussion of category (B) below, and by methods mentioned Therefore, you may find it possible to identify the forbidden
in Sect. 4.7.3. reflections by taking a sequence of diffraction patterns for in-
creasingly thin areas of your specimen and noting that this set
The table organizes these space groups according to the reason of reflections disappears as you choose thinner regions. Search-
that they are not uniquely identified. In the first group (A) are two ing for forbidden reflections in this way is analogous to the way
pairs where the symmetry elements are the same in each member it is done in X-ray diffraction. Second, if your specimen is tilted
of the pair but where they are arranged in different three-dimen- away from the zone axis, double diffraction routes may become
sional configurations. unavailable. Thus, especially in LACBED (see sections below)
instead of seeing a more or less uniform line of intensity along
The next category of space groups (B) that are not uniquely the Bragg condition, you’ll see that the intensity in a forbidden
identified by GM line analysis arises from the presence of one reflection will come and go as double diffraction routes become
of the following screw axes: 42, 31, 32, 62, or 64. These screw available. Third, you can easily find kinematically forbidden
axes do not contain a 21 screw axis and hence do not give GM reflections in precession patterns (see Sect. 4.5.2 above). The
lines. Therefore, as far as GM lines are concerned, we may not cases where precession patterns (or other techniques for identi-
distinguish these space groups (the right-hand member of the fying those forbidden reflections that do not give GM lines) can
pair) from the space group that has a rotation axis instead of the distinguish between space groups, which are not distinguished
screw axis (the left-hand member of the pair). (Note that screw by CBED, are marked with a P in Table 4.13.
axes 41, 43, 61, 63, and 65 do contain 21 screws and so give GM
lines which we can use as a means of identification.) There is another group (C) of sets of space groups that can-
not be distinguished by GM methods alone. These are cases
Although the space groups with these screw axes do not give involving enantiomorphous pairs. Enantiomorphous pairs are
GM lines, they do give kinematically forbidden reflections. In pairs of space groups where one is left-handed and the other is
such cases, the kinematically forbidden reflections can be iden- right-handed, but they are otherwise the same (see next section).
 126 4 Convergent-Beam Electron Diffraction: Symmetry and Large-Angle Patterns

They have screw axes of opposite sense. For the four pairs in Certain symmetry elements are not compatible with enantiomor-
C, there are GM lines but they are the same for both right- and phism. In particular, any crystal having a point group containing
left-handed space groups. an inversion center, a mirror, or a four-fold inversion axis can-
Part I

not be enantiomorphic (see International Tables Sect. 10.2.1.1

The next group of space groups that we cannot distinguish by – Hahn 2002). This means that enantiomorphic crystals must
GM analysis (D) also contains enantiomorphous pairs, but, in belong to one of the following point groups: 1, 2, 222, 4, 422,
these cases, the enantiomorphous pairs do not give GM lines, 3, 32, 6, 622, 23, and 432 (Table 10.2.1.1 International Tables
therefore we cannot distinguish them from a related space – Hahn 2002).
group with a rotation axis rather than a screw axis. In these sets
of three, the first space group of each set can be distinguished In order to determine the handedness correctly, you must match
from the enantiomorphous pair if we can identify kinematically a calculated pattern with your experimental pattern. This is not
forbidden reflections by one of the methods mentioned above always easy because there is often ambiguity in the indexing
in the discussion of type B. If we use precession, only the first and, if the indexing is wrong, then the handedness will be wrong.
of each set of three can be uniquely identified. For example, Johnson and Preston (1994) and Johnson (2007) have discussed
P321 can be distinguished from P3121 and P3221, but P3121 how valuable it is to tilt your specimen to an orientation such
cannot be distinguished from P3221. However, if we use coher- that the ZOLZ has a mirror to facilitate indexing. Of the 11 point
ent diffraction (see Sect. 4.7.3), all three can be distinguished groups listed above, which allow enantiomorphism, only crystals
(in principle). in point groups 1 and 3 have no orientation that will show a mir-
ror in the pattern (the mirror will be in the bright-field disk and
The final group (E) consists of two sets of three space groups in the projection pattern of the zero-order Laue zone).
where the second and third members of the set form an enan-
tiomorphous pair, while the first member of the set has a screw In developing the list of space groups, it was decided that a right-
axis which is not handed. All three space groups in each set give handed coordinate system would always be used. This means
the same GM lines. that a pair of enantiomorphic crystals with oppositely handed
screw axes will be assigned to different space groups. There are
11 such “pairs of enantiomorphic space-group types” (Inter-
national Tables Sect. 8.2.2 – Hahn 2002). You can find these
4.7.1 Enantiomorphous Pairs: Handedness 11 pairs in sections C, D, and E of Table 4.13.

Incidentally, turning the specimen upside down is not gener-

In most cases we do not care which of the two enantiomor- ally helpful or even relevant. What you get when you turn the
phic forms of a crystal we are studying: the right-handed or specimen upside down depends on the symmetry of the crystal
the left-handed. However, partly for completeness and partly (so ignore this suggestion if you still have W&C’s first edition
for intellectual satisfaction, as well as for those cases where it (1996)).
Chapter 4

really matters, you’ll find it interesting to know which of two

enantiomorphs you have in your microscope. You won’t find it
difficult to see the difference between the diffraction patterns
from two enantiomorphs. What is difficult is to say which is 4.7.2 Polarity
which.

When we compare diffraction patterns, taken with the beam par- For crystals with a center of symmetry, along any given direction
allel to the same direction for two specimens which are enantio- the positive sense looks the same as the negative sense. This is
morphs, we see that they are mirror images of each other. So we not true for crystals which lack a center of symmetry. Take a
should be able to tell the right-handed form from the left-handed simple example: GaAs. Along the [111] direction of GaAs, the
form – if only we knew which pattern corresponds to which atoms lie in nearest-neighbor pairs. Depending on the orientation
form. Unfortunately, there is no way to know which is which of the crystal, in the positive sense of the [111] direction the pairs
unless we know the structure of the material. If we have the will be either Ga–As or conversely, As–Ga. Sometimes it may
atomic coordinates of the atoms in the unit cell – in short, if we be important for you to know which polarity the specimen has.
know everything about the crystal except its handedness – then In particular there are many cases in the semiconductor industry
we can perform a dynamical-diffraction calculation to simulate (for compound semiconductors which lack centers of symmetry)
the diffraction pattern, and hence identify the handedness. Pro- when it is vital to know the polarity of the material.
grams are now readily available to do this and so it is an entirely
practical proposition, albeit one that is a bit more complicated This is another case where the symmetry of the patterns cannot
than simple symmetry determination. An example of the deter- tell us the answer but the intensities in the patterns can. If we
mination of handedness is given by Johnson and Preston (1994), know the crystal structure, we can calculate a dynamical dif-
and the book by Spence and Zuo (1992) includes a discussion of fraction simulation of the CBED pattern and compare it to ex-
both handedness and polarity. periment.
 4.9 LACBED 127

There are two complications to this procedure. First, it will gen- the lattice spacing. Under these conditions, and if the beam is
erally be the case that we need to determine the polarity of the focused slightly above or below the specimen, you’ll see that the
specimen in a bulk sample outside the microscope, rather than overlap region has a pattern of interference fringes. These fringes

Part I
the polarity of the specimen as we see it in the microscope. This can be used to distinguish between some of the space groups that
means that, in preparing your TEM specimen, you have to keep cannot be distinguished by the methods of Sects. 4.2, 4.3, and
very careful track of the relationship between the original bulk 4.4, as shown by Saitoh et al. (2001). In Table 4.13, the space-
specimen and the specimen that goes into the microscope. You’ll group ambiguities which can be resolved by coherent CBED are
find this can be quite difficult since a specimen is typically turned marked with a C in the final column.
upside down many times in the course of preparing it to go into
the microscope. And then, in most cases, it will be turned upside
down once again in going through the airlock. Alternatively, you
could place a marker of some kind onto the specimen. 4.8 Tanaka Methods
The second complication is that the differences between the
zone axis patterns for the two polarities are often quite small When we were taught the basics of TEM, we were told that
and depend on specimen thickness. This is particularly true for there are two ways to operate the microscope: to have an in-­
GaAs since the scattering power of these two atoms is so similar. focus image of the specimen, or to have an in-focus view of the
Thus a single simulated pattern is unlikely to provide adequate diffraction plane. This is equivalent to saying that the picture
discrimination. You can be more certain of the result if both the we acquire (onto film or onto a CCD) should be an image con-
simulation and the experiment are done for several thicknesses. jugate to the specimen or an image that is a Fourier transform
of the specimen plane. In about 1980, Michyoshi Tanaka had
Careful the inspiration to see that, in many cases, we can gain more
information by operating away from these two standard condi-
tions. As a result of this inspiration we have become liberated to
Polarity and handedness are tricky to determine. There take images in which there is both diffraction information and
are many opportunities to get the sign wrong. You will image information (though they cannot both be in focus). There
have to be very careful and check each step to get it right. is a whole family of variations on this basic approach. They
 9 give different kinds of information and have a variety of names.
The most widely used of the Tanaka methods is the one also
known as large-angle CBED or LACBED (Tanaka et al. 1980;
The practical determination of polarity was pioneered by Tafto Tanaka and Terauchi 1985a), which is discussed briefly in W&C
and Spence (1982) who found that, for GaAs (and for sphalerite Sect. 20.4. Here we will discuss this first. We will also give a
structures in general), simulation of the patterns is not necessary. brief discussion of some of the other methods later (Sect. 4.13).
In specific off-axis orientations, characteristic dark and bright Morniroli (2002) has written a fine book on this topic, with a lot

Chapter 4
crosses in (200)-type reflections fix the polarity. This behavior is of detail, which you’ll find most helpful if you are a beginner.
independent of thickness because of the very large extinction dis-
tances of the weak reflections involved. Jager et al. (2002) pres- Be Adventurous
ent a more recent example of polarity determination. Nicholls
et al. (2007) have shown that, by combining Tanaka methods
(see following sections) with dynamical simulations, it is pos- Experimenting with operating modes of the microscope
sible to determine the polarity of nanorods that have a diameter is good and can lead to new information. 9
as small as 19 nm.

4.7.3 Coherent Convergent-Beam Diffraction 4.9 LACBED

We know that when the convergence angle is increased suffi- LACBED originated as a solution to the problem of getting in-
ciently, the disks in CBED overlap (W&C Fig. 20.3). The in- formation from crystals with large unit cells. If a crystal has a
tensity in the overlap region has contributions from both over- large unit cell, the spacing of points in the reciprocal lattice is
lapping disks. However, there are two distinct cases. For the small. This means that disks in CBED patterns must be small or
most part, the intensity in the overlap regions is just the sum they will overlap. Normal CBED patterns contain rather limited
of the intensities from the two disks. However, if the illumina- information as a result. The intensity variation within the disks
tion is coherent enough, it is not the intensities that add, but the can be mapped over only a small range of angles, and there is
amplitudes. So you need a coherent (field-emission) source and not much to see. The idea was to find a way to display the vari-
illumination conditions that produce a spot which is smaller than ation of diffracted intensity with angle of incidence over a large
 128 4 Convergent-Beam Electron Diffraction: Symmetry and Large-Angle Patterns

angular range – even when the unit cell is large. This is what of the specimen is small and so we can make drawings to show
LACBED does. how things work without considering the lens and we just let the
electrons travel far enough from the specimen.
Part I

A Source of Confusion
Since the spot diffraction pattern is in a plane very close to the
specimen, this plane is conjugate with the selected-area aperture.
When doing LACBED, you will need to keep in mind We can insert the area-selecting aperture to select a diffracted
that, although you get the LACBED pattern in diffraction beam. (Again this is the reverse of normal operation. Normally
mode, if you see a spot diffraction pattern, you are in we select a diffracted beam with the objective aperture.) Now
image mode. 9 when we go to diffraction mode, we will obtain a CBED disk
from the selected diffraction spot alone because the other beams
have been excluded by the selected-area aperture. Since there is
The principle of the method is as follows. We focus the beam not no overlap, we can make the convergence angle as large as we
onto the specimen, as in normal CBED, but just a little above like and see a CBED pattern for a very wide range of angles – but
or below it – typically just a few micrometers away from the for only one beam at a time.
specimen. Thus, if we were looking at a focused image of the
specimen, we would see, not a sharp spot, but a slightly spread Figure 4.15 shows a conventional CBED pattern and LACBED
beam, illuminating a small area. However, we do not focus on patterns from the same specimen, and at the same angular scale.
the specimen, we focus the microscope so that we image the As you can see, the disks of the conventional pattern show a
plane where the beam is focused. In this plane we observe a spot very small angular range and very little detail. By contrast, the
diffraction pattern. The paradox is that the microscope is being LACBED patterns cover a much larger angular range and show
operated in imaging mode but we see a diffraction pattern. This complex diffraction patterns.
pattern is located not in the back-focal plane of the microscope
but just a few micrometers from the specimen itself. The conventional CBED pattern of Fig. 4.15a was taken at the
same zone axis as Fig. 4.5. The aperture was a little smaller and
Figure 4.14 illustrates the principle of the LACBED method. the specimen thickness was different but you can clearly see the
We can remind ourselves here that a diffraction pattern shows relation between the two figures. Figure 4.15b is the LACBED
how the electrons are scattered as a function of angle. In TEM, pattern from the direct beam (the bright-field disk). You can see
we normally view this pattern in the back-focal plane of the a wealth of detail extending very far beyond the region visi-
objective lens (see Sect. 9.3 in W&C). However, we would get ble in the bright-field disk of Fig. 4.15a. Figure 4.15c shows
the same pattern with no lenses if we just view the pattern very a dark-field LACBED pattern taken in the 010 reflection. As
far from the specimen. (This is how X-ray diffraction is done.) we discussed in Sect. 4.4.3, this reflection contains GM lines.
Very far from the specimen, in this context, means at a distance The figure here allows us to see the GM lines in all their glory.
very large compared with the size of the region of the specimen Down the center of the pattern in both the horizontal and vertical
Chapter 4

that is illuminated. In considering selected-area diffraction, we directions there are lines where there is no diffracted intensity.
find it convenient to draw ray diagrams showing the operation Figure 4.4d is a dark-field LACBED pattern from a reflection
of the lens to give the diffraction pattern. This is partly because that is not kinematically extinct. As you might expect, the in-
it is what really happens and partly because the illuminated area tensity is concentrated along the position corresponding to the
is large. In CBED methods (even LACBED) the illuminated area Bragg reflection.

Fig. 4.14 Schematic diagram to a b c

illustrate LACBED. a Beam focused
on the sample as for normal conver-
gent-beam diffraction. b Beam fo-
cused below the sample – the effects
of diffraction by the sample are not
included. c Same as b, but now the
Specimen Specimen Specimen
effects of diffraction by the sample
are included to show how a spot dif- Spot diffraction
fraction pattern is formed in the plane pattern
where the beam is focused
 4.9 LACBED 129

lens. Other microscopists prefer the height-change method. The

detailed instructions which we give below can be easily modified
for that method.

Part I
4.9.1 The Nature of LACBED Patterns

As we’ve just seen, LACBED patterns contain both diffrac-

tion information and image information. In order to illustrate
this point we will initially limit the discussion to a bright-field
LACBED pattern. Each point in the pattern corresponds to elec-
trons that have gone through the specimen at a different angle
of the beam with respect to the crystal structure. Therefore, the
pattern maps the change of transmitted intensity as a function of
angle, just as the bright-field disk of a CBED pattern does. At the
same time, each point in the image corresponds to electrons that
have gone through the specimen in a different place. Therefore
the pattern contains an image of your specimen. This image is
not in focus (see Fig. 4.16).

Image and Diffraction Pattern Mixed

In LACBED and other Tanaka methods (with the excep-

tion of some rastering techniques), the patterns have a
mix of diffraction information and image information.9

In Fig. 4.17 is shown a LACBED pattern where we put both the

image and diffraction information to good use. The specimen is a
cross-section through GaAs that had a layer of Al deposited onto
the surface. We can see, first, that the top of the image is bright

Chapter 4
Fig. 4.15 a Conventional convergent-beam pattern from Ni3Mo at the [100]
zone axis. It is therefore (apart from details of the way it was obtained) the same because there we are looking above the surface of the original
as Fig. 4.5. b Bright-field LACBED pattern from the same sample at the same specimen (the Al layer is at the top but is not visible in this im-
zone axis, and taken from the same area. c Dark-field LACBED pattern taken in age). Notice that some of the Bragg lines split as they approach
the 010 reflection (the reflection to the right of the bright-field disk). The GM the surface. We’ll discuss more about this kind of splitting later,
extinctions are clearly visible as dark lines which run through the pattern along but it is an indication that the specimen is strained in those areas.
lines that are nearly vertical and horizontal. The horizontal line, in particular, is In particular it tells us that the planes that are responsible for the
not straight because of the buckling of the foil. d Dark-field LACBED pattern tak- Bragg lines are bent through the thickness of the foil. Some of
en in the 001 reflection. d has been rotated 90 degrees with respect to the other the Bragg lines are not split, so we know those planes are not dis-
parts of this figure so as to fit conveniently into the space. It is characterized by torted by the strain. For the planes that are bent, in this one image
brightness along the locus of Bragg reflection we can see how the strain varies with depth below the surface.

The uses of LACBED patterns can be (loosely) divided into two

As you can see from the discussion above, in order to make this categories. On the one hand, there are applications where the
work, the focused beam and the specimen are at different heights, aim of the experiment is to get the same kind of information as
and you need to look at where the beam is focused, not where we would obtain from conventional CBED, for example, sym-
the specimen is focused. There are two ways of doing this. One metry information. In such cases, the image of the specimen
is you start with the beam focused on the specimen and then superimposed on the diffraction information is an undesirable
you change the position of the specimen by using the height price that we have to pay in order to see the pattern over a wide
control on the specimen stage. This works well but means that range of angles. On the other hand, as we will see later, there are
your specimen is now moved off the eucentric height and you’ll applications where we make positive use of both the image and
have to readjust the specimen height later. The second method diffraction information in the pattern. In these cases, we may be
(which we prefer) leaves the specimen at the eucentric height; we able to obtain information regarding the specimen that could be
achieve the required configuration by defocusing the objective obtained in no other way. The reason we make the distinction
 130 4 Convergent-Beam Electron Diffraction: Symmetry and Large-Angle Patterns
Part I

Fig. 4.17 A LACBED image of a cross-section through a sample of GaAs with

Fig. 4.16 A LACBED pattern, which shows, in addition to the usual diffraction a layer of Al deposited onto the original surface. The white space at the top
detail from the sample, a shadow image of amorphous dust particles on the sur- of the image is a “shadow image” of empty space above the sample. The Al
face. It illustrates the way in which LACBED patterns mix diffraction and image layer is barely visible. The arrows indicate Bragg lines that do (red) and do not
information (blue) show splitting due to the strain field near the surface, produced by the
Al overlayer

between these two approaches is that they give rise to rather 4.9.2 Obtaining LACBED Patterns in Practice
different operating conditions, at least in the details. In particu-
lar, if the image information is a nuisance rather than a help for
your experiment, you should aim to obtain the LACBED pattern Let us now see how this works in practice.
from the smallest possible area of the specimen, whereas, in the
Chapter 4

other case, you will need to adjust the area of the specimen that Step 1: tilt the specimen to the desired orientation and set it at
you image on the pattern to get the area of interest into the field the eucentric height.
of view. However, you’ll find that the general procedure is the
same in both cases. Step 2: start with a normal, in-focus image of the specimen. Fo-
cus the illumination to a small spot, just as you would to obtain
There is a benefit to LACBED, which is independent of the use a CBED pattern (but do not go to diffraction at this time).
to be made of the patterns. In normal CBED, the intensity of the
focused electron beam is very high. Therefore, CBED may not be Step 3: defocus the objective lens by a rather large amount. The
practical if your specimen is beam-sensitive. Since we defocus question is “by how much?” We will discuss that later. If your
the beam when we form the patterns, LACBED may be preferred microscope allows it, it is convenient to defocus by a known
when working with such materials. amount (‘four clicks at step size four,’ for example) so that you
can return to a focused condition by simply reversing the steps.
In the diagram drawn above (Fig. 4.14), the beam was shown as
focused below the specimen. In fact you can focus the beam above Step 4: using the ‘second-condenser-lens control’ (the manufac-
or below your specimen: the result is the same. We usually draw turers may call this the intensity control, the illumination control
the diagrams with the beam focused below the specimen because or the brightness control), refocus the illumination to get a sharp
it is easier to see what is going on. In that case the spot diffraction spot on the screen. It will not be a single spot but an array of
pattern is real. If you focus the beam above your specimen, the spots making a diffraction pattern.
diffraction pattern is a virtual pattern (see Fig. 4.18), but otherwise
things are the same and you will not notice the difference. Step 5: introduce the selected-area aperture, and center it around
one of the spots of the diffraction pattern. (We suggest that you
use the direct beam or bright-field spot, the first time you try it.)
The selected-area aperture may need to be very small.
 4.9 LACBED 131

Fig. 4.18 Diagram to show the for-

mation of a virtual spot diffraction
pattern when doing LACBED with the

Part I
beam focused above the sample

Virtual spot
diffraction pattern

Specimen Specimen Specimen

Step 6: go to diffraction. You should see a single CBED disk. The most important parameter is the defocus distance, that is to
say, how far from the specimen you focus the electron beam. If
Step 7: change to a much larger condenser aperture or (if your you make this distance small, the spots in the diffraction pattern
TEM permits it) remove the condenser aperture entirely. You will be very close together and you will need a very small se-
should now see a LACBED pattern. lected-area aperture. However, the area of your specimen that
contributes to the image in the pattern will also be small. Con-
Step 8: to get the best pattern, do a little fine adjustment with versely, if you make the defocus distance large, the separation
the condenser lens (illumination, intensity, brightness) setting. of the spots in the diffraction pattern will also be large and you
can use a larger selected-area aperture. But the image on the
Step 9: return to normal operation. First, go back to a standard pattern will correspond to a large area of your specimen. Your
condenser aperture. This is most important: if you do not do this choice of defocus distance, therefore, will depend on the reason
first, you may get intensities that are much too high, and risk for undertaking the experiment.
damaging either the screen or the camera. Then: go to image
mode; remove the selected-area aperture; spread the illumina- You should choose the size of the SAD aperture so that the re-
tion; obtain a focused image (either by undoing the defocus of quired diffraction spot is cleanly separated from all the other
step 3 above, or by focusing the image). spots. This means that it needs to have a diameter about equal

Chapter 4
to the spot separation for the chosen defocus (but see additional
comments below in Sect. 4.10). If your aim is to minimize the
area of your specimen contributing to the pattern (so as to min-
4.9.3 Choosing the Parameters imize the effects of buckling and thickness variation), then you
need a very small aperture. An aperture size of 5 μm or less is
suitable.
The method described is simple and should be clear. However,
you’ll have noted that there are several parameters to be set: the Your spot size (the setting of the first condenser lens) will be
spot size, the degree of defocus, the size of the selected-area ap- selected based on a number of factors. The spot size is, in this
erture. Rather than propose specific values for these parameters, case, just the size of the diffraction spots. So a smaller spot size
we suggest that appropriate values be found by trial and error will (up to a point – see below) make it easier for you to separate
at the microscope, since each experiment will be best served by the diffraction spots cleanly. The spot size is also (as always)
different values. We gave you similar advice for forming CBED related to the intensity in the beam. As a general rule you’ll find
patterns in W&C Chap. 20. there is plenty of intensity in LACBED patterns so that a small
spot may be used without difficulties from this point of view, but
Don’t Burn Up the Screen it may not always be so.

If your spot size is small, the image detail superimposed on the

LACBED methods use a very large condenser aperture diffraction pattern will be sharper. As we discussed above, the
(or none at all). Be careful to do things in the right order image detail in a LACBED pattern cannot be in focus. Specifi-
or you can cause expensive damage to the screen or the cally, the image is like a shadow image of your specimen made
camera. 9 with a source of illumination equal to the spot size. So your spot
size limits the sharpness of detail that you can see.
 132 4 Convergent-Beam Electron Diffraction: Symmetry and Large-Angle Patterns

from the same specimen. They all show bright-field diffraction.

4.10 Spherical Aberration and LACBED We are going to be discussing the lines in these patterns, and we
need a name for them. Some authors refer to them as Kikuchi
Part I

lines because they occur geometrically in the same places as

LACBED patterns are obtained with a beam of a large conver- Kikuchi lines. However, it is an essential part of the formation
gence angle. This is implicit in step 6 of the procedure given of Kikuchi lines that the diffraction that produces them follows
in Sect. 4.9.2. Using a large condenser aperture, or no aperture inelastic scattering of the electrons. In these images, inelastic
at all, in the second condenser means that the convergence of scattering may not occur and is certainly not an integral part of
the beam is many times larger than it is for regular CBED, for the process. Therefore, we shall follow the example of Morni-
example. As a result, spherical aberration is an important factor roli and call them Bragg lines, because they occur where planes
in the procedure. You can see this effect very clearly by carrying are oriented so as to give Bragg reflection. The first image is
out step 7 of the procedure before step 5. The diffraction pattern a conventional CBED pattern from a defect-free region. The
that you see at step 4 will generally consist of clear sharp spots second image is taken the same way but with the beam sitting
(which can easily be separated by the selected-area aperture). on a dislocation. As you can see, the Bragg lines in the pattern
However, if you remove the condenser aperture at this point become fuzzy and split because of the strain field of the dislo-
(note that this is not recommended as a general rule and should cation. The third image is taken in LACBED mode, although
be done with caution because of the very high intensities that can the image shows only that part of the pattern corresponding to
be produced), the spots in the diffraction pattern become much the same angular range as the first two images. In this case, the
larger, due to the spherical aberration. Indeed, spots that seemed dislocation stretches across the image (between the arrowheads)
(with the condenser aperture in place) to be well separated may and you can see that the Bragg lines are split where they cross
(with the aperture removed) have a strong overlap. This would the dislocation but are sharp where they are remote from the
mean that your selected-area aperture will no longer be able to dislocation.
separate, say, the bright-field pattern from a nearby dark-field
reflection. To solve this you should use a larger defocus to get It turns out that there is a rather simple relation between the
the spots further apart. splitting of the Bragg lines produced by a dislocation and the
conditions of the experiment. In a LACBED pattern, if a dislo-
Step 8 in the sequence given above is also related to spherical cation of Burgers vector b crosses the Bragg line resulting from
aberration. When you insert the selected-area aperture into the diffraction to a reciprocal lattice vector g, then the Bragg line
microscope it is set so that it is in the plane of the image on the will be split so that
screen of the microscope. As we described it, that would put it in
the plane of the paraxial focus for the electron probe. However, g.b = n
once we are in LACBED mode, it is better to have the select-
ed-area aperture in the plane of the circle of least confusion (see where n is the number of nodes in the split line. That is, n is equal
W&C Sect. 6.5 if you are confused by this) in the formation of to the number of splittings: the number of times the intensity
Chapter 4

the electron probe. You can do this, looking at the LACBED goes to the background intensity.
pattern, by making fine adjustments to your condenser lens as
we just described in step 8 of the procedure. This idea was developed by Cherns and Preston (1989) and refer-
ences therein, although they initially counted subsidiary fringes
Clearly, there are great advantages to doing LACBED in a mi- rather than nodes and therefore used g.b = n + 1. The result in the
croscope that has an aberration corrector for STEM operation (as form given here (counting nodes and using g.b = n) is now the
long as you don’t make the convergence angle too large). standard way to do it.

If you have an interest in quantum mechanics, you may be in-

terested to know that this result can be seen as a manifestation
4.11 Crystal Defects in LACBED of something called the Berry phase (Bird and Preston 1988).
As the electron wave field propagates through the specimen,
Patterns: Dislocations it is subject to a phase shift of 2πn due to the distortion of the
lattice planes, and this is what produces the n nodes in the im-
As a general rule, when we take CBED patterns, our aim is to age. The dislocation displaces or offsets the diffracting planes,
get the best possible pattern, which means taking the pattern and for each offset of one plane there is a phase shift of 2π in
in a region of the specimen that is free from strain and free the wave field. Since g.b is the number of offsets, it gives the
from defects. We want the specimen to be uniform through number of nodes.
its thickness in the region where the beam interacts with it.
The symmetry of the patterns is broken when there is distor- In Fig. 4.19c, you can see that several Bragg lines cross the po-
tion of the specimen. However, it turns out that, in the case sition of the dislocation and each of them has a different number
of LACBED patterns, the effects produced by defects can be of nodes. Thus if we index each of the Bragg lines we get a set
used to characterize the defects. Figure 4.19 shows three images of equations: one gi.b = ni for each Bragg line. If we have at least

Licensed to ([email protected])
 4.11 Crystal Defects in LACBED Patterns: Dislocations 133

Fig. 4.19 a Bright-field conver-

gent-beam pattern at 100 kV from an
Fe-30Ni-19Cr alloy, at the [114] zone

Part I
axis. b As a, but with the beam sitting
on a dislocation. c LACBED pattern
from an area including the same dis-
location as in b. The dislocation lies
across the image between the two ar-
rowheads. The splitting of the Bragg
lines where they are crossed by the
dislocation is clearly visible

three such equations, the equations can be solved for the com- another reflection and take another picture of the same area.
ponents of b. (There are conditions on the gs that are used. We You repeat this several times. In each picture of the set, you
will obtain a unique solution only if no g is a multiple of another have to determine whether a particular dislocation is visible or
and if the three gs are not coplanar.) invisible. If it is visible, g.b > 1/3, and if it is invisible, g.b < 1/3
for the corresponding g. The disadvantages of this traditional
There is a little additional complexity here. The values of n are approach are:
not necessarily positive. In order to get the right answer we must
determine the sign of n (Morniroli 2002). The sign of n is re- zz It is inconvenient to have to tilt to a series of different, specific
vealed by the orientation of the splitting. If the Bragg lines bend orientations.
to the left, n has one sign, and, if they bend to the right, it has zz The criterion for invisibility is not always clear and there is
the opposite sign. In order to say which sign is which, we must an additional complication in that contrast can also be related
assign the definition of the direction of the dislocation consis- to a second term: g.b×u, where u is the direction of the dis-
tently and also index the Bragg lines correctly. Finally the sign location.
of the deviation parameter must be known. This will be easy if zz The relation used for invisibility, g.b < 1/3, is not an equality
both Bragg lines (for ± g) can be seen, but will require some care but a comparison; this tends to leave some uncertainty in the
if they are not. result.

Signs The LACBED method avoids all of these problems. Almost

any orientation of the specimen will work. In many cases you

Chapter 4
can get an answer from a single orientation. The number of
To get the Burgers vector correctly, you must do the in- nodes is uniquely determined and you can solve the equations
dexing with great care so that you get the signs right. 9 to give a unique answer for the Burgers vector. In principle,
the LACBED method could be used to determine the Burgers
vectors of dislocations on both sides of a grain boundary from
Let’s examine an example of the complete Burgers vector de- the same LACBED pattern. The LACBED method also makes
termination using Fig. 4.20, from the work of Tanaka’s group it possible for you to determine the Burgers vectors in some
(Tanaka et al. 1988). The pattern was taken using a modified cases that are particularly difficult for the conventional method:
LACBED method (see Sect. 4.13.1), which shows several dif- when the Burgers vector is very small, or when your material is
fracted beams. The number of nodes can be most clearly seen in highly anisotropic.
the dark-field diffraction. We find
There are disadvantages of the LACBED method, however. The
N N 1/:b
N
g.440/:b = −2I g.113/:b = 4I and g.57 =2 most important of these is that it works only if the dislocations are
1 N so far apart that the splitting of Bragg lines from one dislocation is
This leads to b = Œ101: separate from the splitting due to other dislocations. The LACBED
2
method will not work when the dislocation density is high or for a
dislocation pile-up, for example. You have to take care when using
This method of determining the Burgers vector of a dislo- the Cherns–Preston rules for Burgers vector determination. Chou
cation has many advantages over the traditional method of et al. (1992) have shown that, in cases where the atomic number
using the invisibility criterion in a set of two-beam images of the elements in the specimen is high and when the diffraction is
(see W&C Chap. 26). In this classic method, which dates back strong, many-beam effects may modify the splitting of the Bragg
to the 1950s, you tilt the specimen to a two-beam condition lines. They also give a nice argument, based on amplitude-phase
and take a picture. You then tilt to a two-beam condition for diagrams, which makes it easy to see why g.b = n.
 134 4 Convergent-Beam Electron Diffraction: Symmetry and Large-Angle Patterns

Fig. 4.20 A LACBED pattern from

silicon at 200 kV. A dislocation lies
across the bright-field disk in the
Part I

position marked by arrows. From the

number of nodes in the several re-
flections (more easily counted in the
dark-field patterns) the Burgers vec-
tor of the dislocation can be uniquely
identified

Burgers Vector Determination whereas for stacking faults the splitting will occur in all reflections
that introduce a phase change into the electron wave field; that is to
say, for all reflections except those where g.R is an integer, where
When it can be used, LACBED is better than the ‘two- R is the displacement across the fault. Many beautiful examples of
beam method.’ 9 patterns showing the effects of these defects are given in the books
from the Tanaka group (Tanaka et al. 1988, 1994).

Despite these limitations and despite some difficulties that can

rise experimentally, it is generally the case that the LACBED
Chapter 4

method is better than the standard method for Burgers vector 4.13 Other Tanaka Methods
determination.

The insight that Tanaka had to operate the microscope away from
standard conditions has led to several techniques, of which the
4.12 Crystal Defects in LACBED LACBED method is the most widely used. As we have seen, it
is both straightforward and powerful. But we would encourage
Patterns: Stacking Faults you to explore other variations on this idea. There may still be
and Antiphase Boundaries new wrinkles to find.

Here we will briefly discuss three of these varieties of Tanaka

As we described above, LACBED provides a powerful and sim- methods. They are ‘bright- and dark-field LACBED’, ‘conver-
ple way of determining the Burgers vectors of dislocations. Con- gent-beam imaging (CBIM)’, and rastering methods.
trast in LACBED patterns is also produced by stacking faults and
antiphase boundaries, as well as other defects in the specimen.
Typically, such defects produce a splitting of the Bragg lines but
a splitting that has a different character from that in the case of a 4.13.1 Bright- and Dark-Field LACBED
dislocation. In these cases the Bragg line separates into two and
then rejoins, so that there is no nodal line, as shown in Fig. 4.21.
‘Bright- and dark-field LACBED’ is hardly a good or conve-
The general appearance of this splitting is similar for stacking nient name, but we are not aware that any other name has been
faults and antiphase boundaries except that, in the case of antiphase coined. LACBED allows us to overcome the limitation due to
boundaries, the splitting occurs only in the superlattice reflections, disk overlap in conventional CBED diffraction, and to see the
 4.13 Other Tanaka Methods 135

Fig. 4.21 Effect of a stacking fault

on a LACBED pattern. A stacking
fault lies across the LACBED pattern

Part I
between the dashed lines. To the left,
two Bragg lines of less importance
show characteristic division of the
line into two and then rejoining. To
the right, a strong Bragg line shows
a more complex pattern

diffraction over a large angular range. Its limitation is that we a b

can look at only one diffracted beam at a time. In the bright- and
dark-field LACBED technique we can image many beams at the
same time – all of them over an extended angular range (Tanaka
and Terauchi 1985a, b).
Specimen

Bright- and dark-field LACBED starts in the same way as the

normal LACBED method. You focus the beam in a plane that is
fairly close to the specimen but not on the specimen. However,
you don’t introduce an aperture in this case. Even if the con-
vergence angle of the beam (remember, this is selected by your
choice of condenser aperture) is such that the disks overlap in
a CBED pattern, close to the plane where the beam is focused
there will be a region where the disks do not overlap. You can
see this in Fig. 4.22. Figure 4.22a shows normal CBED operation

Chapter 4
with an illumination angle such that the disks just about touch
(in the far field or in the back-focal plane of the objective lens).
c d
Figure 4.22b shows how the disks overlap if we increase the il-
lumination angle. In the standard LACBED method (Fig. 4.22c)
an aperture removes all but one of the beams and so, in the far
field, produces an extended angular range without overlap. In
the bright- and dark-field LACBED method, there is no aper-
ture, so in the far field the disks overlap, but we do not look in
the far field. If instead we look near the place where the beam
is focused, for example at the plane marked by a dashed line in
Fig. 4.22d, we note that we get all the disks displayed simulta-
neously without overlap.

Bright- and Dark-Field LACBED

This is a valuable technique. You should work out how

to do it on your microscope. 9 Fig. 4.22 How it is possible to produce Tanaka patterns in which many beams
are seen over a wide angular range. a Out-of-focus convergent-beam pattern.
b Out-of-focus pattern obtained with a convergence angle so large as to make
the disks overlap considerably. Such a pattern is called a Kossel pattern. c Same
An example of such a pattern is given in Fig. 4.23. It should be as b, but with an aperture inserted so as to give a LACBED pattern. d To get the
clear to you that such images are very powerful and are espe- many-beam large-angle pattern the aperture is left out, but we observe a plane
cially useful for symmetry determination and defect analysis. like the one indicated by a dashed line instead of looking in the diffraction plane
 136 4 Convergent-Beam Electron Diffraction: Symmetry and Large-Angle Patterns

Fig. 4.23 Pattern taken in the ma-

ny-beam mode of large-angle diffrac-
tion. The sample is silicon at the [111]
Part I

zone axis and the beam voltage was

60 kV
Chapter 4

In symmetry determination they reveal the internal symmetries croscope has that facility. The other reason that the technique is
of all of the reflections at a zone axis and hence give you more little used is that it does not have a good name. We all know that a
complete information on the diffraction group. Although, as you catchy name is important. How can you talk to your colleagues if
should recall, for all methods where the beam is not focused on you have to say “I did bright- and dark-field LACBED”? Would
the specimen, the presence of an image superimposed on the the technique begin to be used as much as it should, if we started
diffraction information can give apparent breaking of real sym- to call it ABLAD, for All-Beam Large-Angle Diffraction?
metries. In the case of Burgers vector determination the advan-
tage of the method is that it allows you to count the number of
nodes in dark-field diffraction, where it is often more clear than
in the bright field. 4.13.2 Convergent-Beam Imaging (CBIM)
What is surprising then is that this technique is not more fre-
quently used. Two things affect this, perhaps. First, to be able In Tanaka methods, image information and diffraction information
to set up the microscope so as to focus on the dashed plane in are both displayed in the image we see. As we have already noted,
Fig. 4.22d is not always easy. If you want to try this technique it is not possible for the diffraction pattern and the image to be
(and we recommend that you do), you can try to set it up (as both in focus at the same time. It may sometimes seem that both
for normal LACBED) using only the condenser and objective are in focus because we take advantage of our ability to make a
lenses. However, since these lenses both change the plane in very small probe and so the blurring can be quite small. If you
which the beam is focused (in modern microscopes with immer- look at a LACBED pattern in the back focal plane, you’ll see that
sion lenses), the setting up may be tricky. The alternative is to the diffraction information is focused but the image is a blurred
use ‘free lens control’ to adjust the intermediate lens, if the mi- shadow image with a resolution about that of the spot size. In the
 4.13 Other Tanaka Methods 137

4.13.3 Rastering Techniques

Part I
In the Tanaka techniques described up to this point (and in
Chaps. 20 and 21 of W&C) the diffraction patterns are obtained
‘in parallel,’ that is a single exposure gets the whole pattern
at once. There are methods we use to obtain the information
sequentially: by scanning or rastering. If a parallel beam is
used, a diffraction pattern is obtained for a single angle of inci-
dence of the beam on the specimen. If your specimen were then
rocked, the kind of pattern obtained by the bright- and dark-field
LACBED method would be built up one pixel at a time. This
has been done only once (it seems that only one group was crazy
enough to do it, since it requires a mechanical stage that can be
tilted with very great accuracy without moving the specimen),
see Brunner et al. (1981). It seemed unlikely to be done again,
but new computer-controlled piezoelectric stages might make
it plausible to repeat their method. The alternative is to rock
the beam, leaving the specimen stationary. This is easier since
we can now tilt (or rock) the beam under electronic control.
However, tilting the beam would cause the diffraction pattern
to move on the screen (or camera). To compensate for that, we
unrock the beam after it emerges from the specimen, so that
Fig. 4.24 CBIM pattern. The sample is from a silicon multilayer structure with the pattern remains stationary on the screen, even as the beam
a boundary between layers down the middle of the pattern (arrows). Attention is tilting at the specimen. In fact, the same set-up of deflection
is drawn to a shift of an intersection point between Bragg lines. This reflects a coils that is rapidly becoming of more widespread application in
change in lattice parameter across the junction. The Bragg lines can be seen to order to perform precession experiments (see Sect. 4.5.2 above)
be misaligned. This is the result of a rotation of the crystal across the interface can be used for mapping diffraction as a function of angle. Such
a beam-rocking technique was developed simultaneously and
independently by Tanaka et al. (1980) and by Eades (1980). At
bright- and dark-field LACBED method both the image and dif- one time it was proposed to call it ‘extended angular domain
fraction information are blurred. The aim of CBIM is to obtain a electron scattering’ (with a simple acronym), but – happily – that
pattern which mixes image and diffraction information but where never caught on. The advantage of these techniques (though the

Chapter 4
it is the image that is in focus (Humphreys et al. 1988). This may extra work involved may not thereby be justified) is that all the
be helpful when we are looking for changes in crystallographic diffraction information is obtained from the same area of the
properties on a very local scale (see, for example, Fig. 4.24). specimen so that there are no distortions of the patterns from
buckling, defects, or thickness variation, see Fig. 4.25.
Since we want to have the image in focus, we must use the ob-
jective lens to focus on the specimen plane. Then we use the con- Rastering
denser lens to focus the beam at some plane above or below the
specimen. Then we see an illuminated disk crossed by diffraction
features, which do not look noticeably out of focus if the spot If you have computer-controlled rastering on your micro-
size is small enough. The technique has mostly been used with scope, these methods can give you large-angle patterns
an objective aperture to remove all beams but one (generally from a local area so that there is no image information.9
the bright-field beam). Thus this technique is limited in angular
range to that of a nonoverlapping CBED disk. It would seem that,
using free lens controls, you could increase the angular range
by using the area-selecting aperture to remove the other beams. Koch (2011) has recently taken advantage of the improvements
in cameras and computer power to develop three innovations
that promise to make rastering methods more valuable and more
widely used. First, he has shown how the effects of aberrations
can be corrected. When the aim is to rock a parallel beam about
a point on the specimen, the beam moves because of the lens
aberrations, especially the spherical aberration of the objective
prefield. Koch has shown how to use computer control of the
 138 4 Convergent-Beam Electron Diffraction: Symmetry and Large-Angle Patterns
Part I

Fig. 4.25 Bright-field LACBED pattern produced by the rastering technique. The sample is GaAs at 120 kV. Because the beam is rocked and thus all of the pattern is
obtained from the same area of sample, the pattern does not have the distortion that is characteristic of regular LACBED patterns
Chapter 4

scan coils to correct for this and hence greatly reduce the size Koch has also invented an ingenious scanning method that
of the area of the specimen that contributes to the pattern. Next, allows us to obtain what looks like a bright- and dark-field
he has shown that, if a rastering method is used, it is advan- LACBED pattern (see Sect. 4.13.1) in a single exposure. A par-
tageous to acquire the patterns from the camera separately for allel beam is rocked about a point on the specimen (as in earlier
each beam orientation. Each pattern from the camera looks like techniques), but the de-scanning below the specimen does not
a selected-area spot pattern, but is obtained for a different angle compensate completely for the rocking above the specimen.
of incidence of the beam on the specimen. The intensities of For example, the de-scanning might cancel only 90 % of the
the beams in these spot patterns are then assembled to produce incident-beam rocking. Then the spots on the screen (or on the
large-angle patterns, one for every beam. The particular advan- camera) move, but move a distance smaller than they would
tage of this method is that each spot pattern from the camera can with no descanning. In this way, CBED disks are built up but
be processed to remove the diffuse scatter: that is, the inelasti- with the paradoxical condition that the angular scale within the
cally scattered electrons. Thus, even a TEM without an energy disks is different from the angular scale from one disk to another.
filter can be used to obtain what is, in effect, an energy-filtered Thus many CBED disks can be imaged over a wide angular
diffraction pattern. An example is shown in Fig. 4.26. range without overlap.
 4.13 Other Tanaka Methods 139

Part I
Fig. 4.26 Large-angle patterns from SiTiO3 at the 001 zone axis, obtained by rastering. The bright-field and two dark-field patterns are shown. The patterns were
obtained one pixel at a time and treated to remove the diffuse scatter.

Chapter 4
 140 4 Convergent-Beam Electron Diffraction: Symmetry and Large-Angle Patterns

Chapter Summary
Part I

Two topics have been covered: symmetry and Tanaka meth- Tanaka methods:
ods.
By operating the microscope away from standard condi-
Symmetry: tions, it is possible to make images that contain both dif-
fraction and image information.
Convergent-beam diffraction is often the most sensitive and
most accurate method of determining crystal symmetry. – These patterns can be made by several methods. The most
widely used is usually called LACBED for large-angle
– We carry out the procedure in two steps. First the point CBED.
group is determined from the symmetry of zone-axis – We gave you a step-by-step account of how to obtain
CBED patterns. Then we determine the space group from LACBED patterns. This should make it relatively easy
the presence of dynamical (and sometimes kinematical) for you to learn to obtain these patterns on your micro-
extinctions. scope.
– A worked example was given to help you understand the – The method of acquiring LACBED patterns was devised,
full process. originally, to make it possible to get better symmetry
– Alternative methods for symmetry determination were information from crystals with large unit cells. They are
also presented. There may be circumstances when you now used much more widely: for mapping strain across
want to consider them. the sample and for characterizing defects.
– Sometimes we need to go beyond the determination of
the space group. We may need to know the polarity of a In summary, Tanaka techniques deserve to be more widely
sample, or we may need to know its handedness. These used since they are valuable both for symmetry studies and
topics were also discussed briefly. for defect analysis. You should make sure to include them
in the repertoire of things you can do at the microscope.
Chapter 4
 Appendix 141

Appendix

Part I
People Q4.16 Why is the density of reflections in the zero-order Laue
zone generally the same as the density in other Laue
zones?
This is one of those fields where the pioneers are still, thankfully, Q4.17 If the density of reflections in the zero-order Laue zone
active. Check out your library for the early works by Michiyoshi is half that in other Laue zones, what does it imply?
Tanaka, John Steeds, and Roger Vincent and then try to find the Q4.18 What is the difference between handedness and polarity?
early work in Germany. Q4.19 In Tanaka methods, the image information and the dif-
fraction information can never both be truly in focus at
the same time. Why?
Q4.20 In Tanaka methods (under suitable conditions), the image
Self-Assessment Questions information and the diffraction information can seem to
be in focus at the same time. Why?
Q4.21 The way in which LACBED patterns are made involves
Q4.1 If you found, in the literature, the same crystal given a the use of what might seem to be the ‘wrong’ aperture.
different symmetry determined by X-ray diffraction from Explain.
the symmetry determined by CBED, which would you Q4.22 LACBED patterns are of limited use for the determina-
trust and why? tion of point groups. Why?
Q4.2 In answering the previous question, could there be cir- Q4.23 Dark-field LACBED patterns tend to be very helpful for
cumstances that would reverse your answer? the determination of space groups. Why?
Q4.3 What is a point group? A space group? Q4.24 Using LACBED can, if improperly carried out, put the
Q4.4 How does the symmetry of a diffraction experiment (in camera or the screen of the microscope at risk. Explain.
the TEM) differ from the crystal symmetry? What is the situation to avoid?
Q4.5 How is it possible to observe four different symmetries Q4.25 What determines the optimum amount of defocus to be
in a single CBED pattern? used in LACBED?
Q4.6 Why are high-symmetry patterns more valuable than Q4.26 LACBED and bright- and dark-field LACBED can be
low? used very effectively to determine the Burgers vectors of
Q4.7 The structure-factor rules given, for example, in W&C dislocations. The procedure is straightforward, but there
Chap. 16, are related to which of the two kinds of forbid- are two points that require care. One is correctly indexing

Chapter 4
den reflections described in this chapter? the reflections. What is the other?
Q4.8 How would you recognize a dynamical extinction when Q4.27 LACBED and bright- and dark-field LACBED can be
you see one? How would you check if you are in doubt? used very effectively to determine the Burgers vectors
Q4.9 Does every experiment, in which a glide plane is parallel of dislocations. Why does the method not succeed when
to the beam, give rise to dynamical extinctions? If not, the dislocations are located close to each other?
why not? Q4.28 LACBED (including bright- and dark-field LACBED)
Q4.10 What is the logic behind the use of the ‘symmetrical ma- and the rastering methods described in Sect. 4.13.3 were
ny-beam excitation’ method? What is the limitation of all developed to address the same problem. Which prob-
this method? lem is that?
Q4.11 Earlier publications tabulate symmetries for ± g patterns. Q4.29 How does the way the beam is deflected in the rastering
Why do we not recommend this approach? techniques of Sect. 4.13.3 differ from the way the beam
Q4.12 Draw a diagram to show how a precession diffraction is deflected in precession diffraction?
pattern is obtained.
Q4.13 To obtain a precession pattern, the scan above the sample
is compensated by a ‘descan’ below the sample. The de-
flection of the beam in the descanning is not at the same Text-Specific Questions
azimuthal angle as the deflection of the scanning. Why?
Q4.14 Why are the intensities in precession patterns more
closely related to the structure factors than the intensities T4.1 Fig. 20.2.B in W&C shows a CBED pattern from silicon.
(for example) in a selected-area diffraction pattern? For this pattern, fill out as much as you can of the matrix
Q4.15 Dynamical extinctions can be seen in reflections in the shown in Table 4.1. Use this result to say as much as you
first-order Laue zone when there is a glide plane, but not can about: the projection diffraction group; the diffrac-
when there is a screw axis. Why? tion group; the point group of this crystal. Use Table 4.5
 142 4 Convergent-Beam Electron Diffraction: Symmetry and Large-Angle Patterns

Fig. 4.27 Two zone-axis convergent-beam

patterns from a metastable alloy of aluminum
and germanium. No patterns with higher sym-
Part I

metry were found at any other orientation.

From Kaufman and Fraser (1985)
Chapter 4

to determine whether your results are consistent with the

information given in the figure caption of Fig. 20.2.B. References
T4.2 Fig. 4.27 shows CBED patterns from two zone axes
from a metastable phase of an aluminum-germanium al-
loy. The figures are from a paper by Kaufman and Fraser General References
(1985). No patterns of higher symmetry were found for
this phase. Deduce the diffraction group for each pattern. Cowley JM (1992) Electron Diffraction Techniques vol. 1. In-
From this deduce the point group. Then use dynamical ternational Union of Crystallography and Oxford University
extinctions to deduce the space group. Press, Oxford
T4.3 Look at each of Figs. 4.3–4.8 in turn. Determine which Cowley JM (1993) Electron Diffraction Techniques vol. 2. In-
of them show HOLZ lines in the bright-field disk. In ternational Union of Crystallography and Oxford University
each case, say what that tells you about the symmetry. Press, Oxford
As mentioned in the chapter, this particular compound is Eades JA (1989) Convergent-Beam Diffraction. J Electron Mi-
quite complicated to analyze. There are features in some croscopy Techn 13(1):2
patterns that may look like HOLZ lines but are not. You Mansfield JF, Lin YP, Graham RJ (1986) The Library of Con-
will need to take great care. vergent Beam Electron Diffraction Update: No 1. Norelco
T4.4 Work out in detail how you would operate your micro- Reporter: Electron Optics 33:54–66
scope to obtain a bright- and dark-field LACBED pattern Morniroli JP (2002) Large-Angle Convergent-Beam Electron
(as described in Sect. 4.13.1 and illustrated in Figs. 4.22 Diffraction (LACBED). Societe Francaise des Microscopies,
and 4.23). Paris
T4.5 Find out whether your microscope has the hardware and Spence JCH, Zuo JM (1992) Electron Microdiffraction. Plenum,
software needed to do precession diffraction and the ras- New York
tering techniques described in Sect. 4.13.3. If it does, Tanaka M, Terauchi M (1985) Convergent-Beam Electron Dif-
work out how each of the techniques can be performed fraction. JEOL, Tokyo
with your instrument.

Licensed to ([email protected])
 Appendix References 143

Tanaka M, Terauchi M, Kaneyama T (1988) Convergent-Beam Hammond C (2009) The Basics of Crystallography and Diffrac-
Electron Diffraction II. JEOL, Tokyo tion, 3rd edn. IUCr Texts on Crystallography. Oxford Uni-
Tanaka M, Terauchi M, Tsuda K (1994) Convergent-Beam Elec- versity Press, Oxford (A good introduction to the concepts

Part I
tron Diffraction III. JEOL, Tokyo needed in this text)
Tanaka M, Terauchi M, Tsuda K, Saitoh K (2002) Conver- Humphreys CJ, Maher DM, Fraser HL, Eaglesham DJ (1988)
gent-Beam Electron Diffraction IV. JEOL, Tokyo Convergent-beam imaging – a transmission electron micros-
The Bristol Group (Compiled by J Mansfield) (1984) Convergent copy technique for investigating small localized distortions
Beam Electron Diffraction of Alloy Phases. Adam Hilger, in crystals. Philos Mag A58:787–798
Bristol (See also the supplement, which includes an erratum Jager Ch, Spiecker E, Morniroli JP, Jager W (2002) Polarity de-
list for the book, and which is given in the following entry) termination of III–V compound semiconductors by large-an-
Note: The four books listed above with Tanaka as first author are gle convergent beam electron diffraction. Ultramicroscopy
now available as free pdf downloads from http://www.tagen. 92:273–283
tohoku.ac.jp/labo/terauchi/activity/cbedbook.html Johnson AWS (2007) Chiral Determination. Direct Interpreta-
tion of Convergent Beam Electron Diffraction Patterns us-
ing the Series Expansion of Cowley and Moodie. Acta Cryst
Specific References B63:511–520
Johnson AWS, Preston AR (1994) Some notes on the selection of
Ayer R (1989) Determination of Unit Cell. J Electron Microsc structural chirality by CBED. Ultramicroscopy 55:348–355
Tech 13:16–26 Kaufman MJ, Eades JA, Loretto MH, Fraser HL (1983) A Study
Bird DM, Preston AR (1988) Observation of Berry’s Geometri- of a Cellular Phase Transformation in the Ternary Ni-Al-Mo
cal Phase in Electron Diffraction from a Screw Dislocation. Alloy System. Met Trans A14:1561–1571
Phys Rev Lett 61:2863–2866 Kaufman MJ, Fraser HL (1985) Characterization of metastable
Brunner M, Kohl HJ, Niedrig H (1981) Electron Diffraction crystalline phases in the Al-Ge alloy system. Acta Metall
with Rocking Beam and Rocking Crystal (Analogies: CTEM- 33:191–203
STEM). Optik 58:37–55 Koch CT (2011) Aberration-compensated large-angle rock-
Buxton BF, Eades JA, Steeds JW, Rackham GM (1976) The ing-beam electron diffraction. Ultramicroscopy 111:828–840
Symmetry of Electron Diffraction Zone Axis Patterns. Phil Lovett DR (1999) Tensor Properties of Crystals, 2nd edn. Adam
Trans A281:171–194 Hilger, Bristol
Cherns D, Preston AR (1989) Convergent Beam Diffraction Morniroli JP (2002) Large-Angle Convergent-Beam Electron
Studies of Interfaces, Defects and Multilayers. J Electron Diffraction. Société Française des Microscopies, Paris
Microsc Tech 13:111–122 Morniroli JP (2011) (and specifically http://www.electron-dif-
Chou CT, Preston AR, Steeds JW (1992) Dislocation Contrast in fraction.fr/atlas_download_121.htm) http://www.elec-
Large Angle Convergent-Beam Electron Diffraction Patterns. tron-diffraction.fr/
Philos Mag 65:863–888 Morniroli JP, Steeds JW (1992) Microdiffraction as a tool for

Chapter 4
Ciston J, Deng B, Marks LD, Own CS, Sinkler W (2008) A quan- crystal structure identification and determination. Ultrami-
titative analysis of the cone-angle dependence in precession croscopy 45:219–239
electron diffraction. Ultramicroscopy 108:514–522 Morniroli JP, Ji G, Jacob D (2012) A Systematic Method to Iden-
Eades JA (1980) Zone-Axis Patterns formed by a new Dou- tify the Space Group from PED and CBED patterns. Part I
ble-Rocking Technique. Ultramicroscopy 5:71–74 – Theory. Ultramicroscopy 121:42–60
Eades JA (1988a) Symmetry Determination by Convergent-beam Nicholls DP, Vincent R, Cherns D, Sun Y, Ashfold MNR (2007)
Diffraction. EUREM 88; IOP Conf Series 93(1):3–12 Polarity determination of zinc oxide nanorods by defocused
Eades JA (1988b) Glide Planes and Screw Axes in Conver- convergent-beam electron diffraction. Philos Mag Lett
gent-beam Diffraction: The Standard Procedure, Microbeam 87:417–421
Analysis. Ed. Newbury, pp 75–80 Nye JF (1957) Physical Properties of Crystals. Oxford Univer-
Eades JA (2001) Electron diffraction symmetries in reflections sity Press, Oxford
in Higher-Order Laue Zones. Acta Microscopica 10:23–27 Pogany AP, Turner PS (1968) Reciprocity in electron diffraction
Gjønnes J, Moodie AF (1965) Extinction Conditions in the Dy- and microscopy. Acta Cryst A24:103–109
namical Theory of Electron Diffraction. Acta Cryst 19:65–67 Saitoh K, Tsuda K, Terauchi M, Tanaka M (2001) Distinction
Goodman P (2001) Space-group determination by conver- between space groups having principal rotation and screw
gent-beam electron diffraction. In: Shmueli U (ed) Interna- axes, which are combined with twofold rotation axes, using
tional Tables for Crystallography, vol B. International Union the coherent convergent-beam electron diffraction method.
of Crystallography (There is also a 2006 edition accessible Acta Cryst A57:219–230
online), pp 285–306 Spence JCH, Zuo JM (1992) Electron Microdiffraction. Plenum,
Hahn T (ed) (2002) International Tables for Crystallography Vol- New York
ume A: Space-group symmetry, 5th edn. International Union Tafto J, Spence JCH (1982) A Simple Method for the Determi-
of Crystallography (There is also a 2006 edition accessible nation of Structure-Factor Phase Relationships and Crystal
online.) Polarity Using Electron Diffraction. J Appl Cryst 15:60–64
 144 4 Convergent-Beam Electron Diffraction: Symmetry and Large-Angle Patterns

Tanaka M, Sekii H, Nagasawa T (1983b) Space-Group Determi- Tanaka M, Terauchi M, Tsuda K (1994) Convergent-Beam Elec-
nation by Dynamical Extinction in Convergent-Beam Elec- tron Diffraction III. JEOL, Tokyo
tron Diffraction. Acta Cryst A39:825–837 Tanaka M, Terauchi M (1985b) Simultaneous Observation of
Part I

Tanaka M, Terauchi M (1985a) Convergent-Beam Electron Dif- Bright- and Dark-Field Large-Angle Convergent-Beam Elec-
fraction. JEOL, Tokyo tron Diffraction Patterns. J Electron Microsc 34:128–135
Tanaka M, Saito R, Sekii H (1983a) Point-Group Determina- Tsuda K, Sitoh K, Terauchi M, Tanaka M, Goodman P (2000)
tion by Convergent-Beam Electron Diffraction. Acta Cryst Distinction of space groups (I23 and I213) and (I222 and
A39:357–368 I212121) using coherent convergent-beam electron diffrac-
Tanaka M, Saito R, Ueno K, Harada Y (1980) Large-Angle tion. Acta Cryst A56:359–369
Convergent-Beam Electron Diffraction. J Electron Microsc Vincent R, Midgley PA (1994) Double conical beam-rocking
29:408–412 system for measurement of integrated electron diffraction
Tanaka M, Terauchi M, Kaneyama T (1988) Convergent-Beam intensities. Ultramicroscopy 53:271–282
Electron Diffraction II. JEOL, Tokyo
Chapter 4
 Electron Crystallography,
5

Part I
Charge-Density Mapping,
and Nanodiffraction

Chapter Preview

This chapter does contain rather a lot of physics but don’t really is concerned with pushing the limits of the TEM and
let that put you off. You can look up Dirac–Fock calcula- our ability to spell.
tions, the Mott formula, and what the Debye–Waller factor
is later if you want to. Ditto for the Simplex algorithm, Strain mapping and the precession method are two really
the bootstrap method, Poisson statistics and the Fienup al- hot topics in TEM; we’re sure you will see them soon if
gorithm, but if you don’t immediately do so, you should you haven’t already. Some TEMs are essentially purchased
have no problem in following the discussion. We’ve left the primarily for one or other of these two applications. You’ll
word electron in the title of the chapter to emphasize that want to look at Chap. 6 for information on the DM plug-in
this is not X-ray crystallography. The electron version has that allows the strain to be plotted as an image. Diffuse scat-
many features to recommend it – the interactions are stron- tering is a topic that we usually try to gloss over. You’ll see
ger and most researchers now have access to a TEM with a here why that is so and just how much information might
field-emission gun. With the FEG you can examine a very be available if you have the right TEM and take the time.
small volume. One feature that comes up again is the peren- Ronchigrams feature extensively in Chaps. 7, 9, and 11.
nial problem that we need thin specimens, which means we Once you can say the word correctly, you can practice pty-
may have relaxation due to the proximity of a surface. Then chography. Chap. 8 is devoted to holography; you’ll want
we discuss using QCBED to learn about bonding. Clearly to reread this chapter after you’ve read through Chap. 8.
this will be an incredibly important tool when it is more
routine. We reference W&C quite often because this chapter

Chapter 5
Chapter 5

© Springer International Publishing Switzerland 2016

C. B. Carter, D. B. Williams (Eds.), Transmission Electron Microscopy, DOI 10.1007/978-3-319-26651-0_5 145
 146 5 Electron Crystallography, Charge-Density Mapping, and Nanodiffraction

5.1 Can We Quantify Electron In this chapter we will describe how the information in TED
patterns may indeed be extracted quantitatively for comparison
Diffraction Data? with models. This information includes the mapping of local
Part I

strain fields (so important now in controlling the speed of the

For almost a century, X-ray crystallography has been the method current generation of semiconductor devices), the mapping of
of choice for solving crystal structures, from proteins to tur- chemical bond charge densities, and the accurate measurement
bine-blade alloys. Given the intense interest in nanostructures, of the structure factors of noncentrosymmetric crystals, the de-
and the inability of X-ray methods to solve the structures of such termination of defect structure from diffuse scattering, and many
small crystals (except perhaps in powder form, with few atoms new possibilities such as diffractive imaging, where images of
per cell), we are led to ask why is there no general electron crys- isolated particles are reconstructed from the diffraction pattern
tallography method, based on electron microdiffraction? This alone, using a computer to replace the imaging lens. We also
omission is especially puzzling when we consider that electron summarize current efforts to solve crystal structures by electron
nanodiffraction produces the strongest signal from the smallest diffraction, explain why it has proven so difficult, and briefly
volume of matter in all of science. This chapter is concerned describe the precession (see W&C p 295) and other approaches
with the problems of quantifying electron diffraction data – we to the problem of minimizing multiple scattering.
show why it has been so difficult, and describe some successful
approaches, with suggestions for more to inspire future genera- We may summarize the differences between electron and X-ray
tions of students to develop this underused method which is so diffraction as follows: electron Bragg intensities are normally
full of potential. not angle-integrated, our rocking-curve widths are a much larger
fraction of the Bragg angle, our Ewald sphere is much larger
In X-ray diffraction, the thickness of the crystal being studied is relative to lattice spacings (so that it is normal to have many
never considered, yet the experimental X-ray Bragg intensities reflections simultaneously close to the Bragg condition, rather
agree so well with calculated values that they can be used for than one at a time as in small-molecule, X-ray work), our ex-
solving crystal structures once the phase problem is solved. In tinction distances are much shorter, our electrons are diffracted
transmission electron diffraction (TED), a glance at any dark- by the total electrostatic potential (including the contribution
field image of a wedge-shaped sample shows how sensitive the from the nucleus) rather than the charge density due to electrons
Bragg beam intensities are to thickness, which must therefore alone, and multiple scattering (rather than single scattering) is the
be accurately measured in order to obtain good agreement be- normal condition. Finally, electrons have the great advantages
tween calculations and experiment. Since it is very difficult to of allowing the use of lenses, and, for field-emission microdif-
measure the thickness of TEM samples with the nanometer pre- fraction, of providing the strongest continuous signal from the
cision needed, TED data has historically been considered un- smallest volume of matter in all of physics. A unique property of
quantifiable, and useful only for qualitative purposes. It is this electrons is their sensitivity to the charge state of ions in a crystal,
problem, together with the strength of multiple scattering (which which is the basis of the quantitative CBED (QCBED) method,
prevents solution of the phase problem) which has prevented reviewed in Zuo (2004).
electron diffraction from becoming a routine method for solv-
ing crystal structures, despite the urgent need for such a method We saw in Chap. 21 of W&C that thickness can be rather accu-
to apply to individual nanocrystals. The difficulty of collecting rately measured by CBED. In addition, if the atomic structure of
3D microdiffraction data without moving the sample with respect a crystal is known, modern computer software can quickly simu-
to the electron probe (and so changing thickness) is an important late the CBED pattern including all multiple-scattering eff­ects, so
additional difficulty not present in X-ray work. that their presence is no longer a disadvantage, and may actually
amplify sensitivity to, for example, the ionic (charge) state of
There have been three important exceptions to this generally atoms in a crystal. Finally, we know that CBED patterns can be
pessimistic trend. In biology, cryoelectron microscopy uses mea- obtained from regions of crystal so small that they are highly
surements of the Bragg beam intensities from protein monolay- perfect and not bent.
Chapter 5

ers (combined with images) to solve their structure. This works

because monolayer organic crystals of light elements provide For all these reasons, it is not surprising that the greatest suc-
weak scattering from a film of constant, known thickness. In sur- cess in matching electron diffraction patterns to simulations has
face science, following the famous solution of the silicon (111) come from QCBED, where we have a large number of pixels
7 × 7 surface reconstruction (arguably the most important prob- within each Bragg disk to match to calculations. This method
lem in surface science at that time) by quantitative TED, many has the remarkable property that the smaller the probe, the more
more surface structures have since been solved by this method. information we get! (A larger beam-divergence, producing larger
Finally, the quantitative convergent-beam method (QCBED) has diffraction disks and more sampled data, is needed to reduce
been highly successful for mapping out the bond-charge distri- the size of a diffraction-limited probe.) In addition, for known
bution between atoms in solids in crystals of known structure. At crystal structures the orientation of the beam may be determined
present, the most promising direction of research for a general very precisely with respect to the crystal as discussed in the next
method of solving inorganic microcrystals appears to be the pre- section. The orientation can be determined after recording the
cession electron diffraction method. pattern, and does not depend on the operator’s skill in setting
Chapter 5
 5.2 Quantitative CBED for Charge-Density Mapping 147

which was closed down around the point P, the intensity at P′

would not change. Opening up the iris just gives a larger angu-
lar view of the crystal ‘rocking curve,’ but does not change the

Part I
intensity within the iris. If we close the iris almost to a point at
P, we have an SAD pattern! It’s similar for an iris closed around
Q. Thus we have the remarkable result that a CBED pattern
is nothing but a set of independent SAD patterns laid side by
side, each one for a slightly different incident beam direction.
Closing the iris does, however spread the illumination over a
much wider area, since it reduces the numerical aperture of the
probe-forming lens. Yet the CBED pattern gives us this same
information from a tiny, defect-free region of crystal that is not
bent. (We are assuming Bragg scattering only, from a perfectly
crystalline parallel-sided slab, so that ‘more is the same’; then
broader illumination does not introduce any new information.
We’ll consider what happens when the probe becomes smaller
than the unit cell in Sect. 5.7).

CBED and SAD

CBED patterns are just a bunch of SAD-like point-dif-

fraction patterns laid beside each other, each one for a
slightly different incident beam direction. 9

5.2 Quantitative CBED for Charge-

Fig. 5.1 Simplified ray diagram for convergent-beam electron diffraction. If only
elastic Bragg scattering is allowed, electrons from P can only arrive at points
Density Mapping
such as P′, one in each disk; similarly for Q and Q′. Each of these families of
points such as P′ defines one selected-area (SAD) diffraction pattern, yet they Electrons are the glue that hold atoms together and give mate-
come from a very small region of crystal. The camera length is L. g is a reciprocal rials their strength and many other bulk properties. However, it
lattice vector, λ the electron beam wavelength has proven remarkably difficult to ‘see’ chemical bonds, mainly
because they result from such tiny perturbations of the atomic
up precise diffraction conditions, as is the case for other dif- charge density. This bond-formation affects mainly the low-order
fraction modes, since QCBED patterns always span a range of structure factors, modifying them by less than 1 % from their
orientations. neutral-atom values. So the basic idea of CBED refinement is to
adjust the structure factors in a CBED simulation until the cal-
In order to get good agreement between simulated and experi- culated pattern fits the experimental one. The initial simulation
mental patterns, it is necessary to use an elastic-energy filter (GIF uses structure factors for neutral atoms, not ions, for a crystal of
or Omega filter, and generally just called an ‘energy filter’), a known structure. The refined electron structure factors are con-
well-calibrated CCD detector and preferably a cold stage or cool- verted to X-ray structure factors, which are Fourier coefficients
ing holder. With the ideal neutral atom positions known, other of charge density, and a density difference map is formed to

Chapter 5
parameters may then be adjusted for best fit, such as sample thick- reveal the chemical bonds between atoms.
ness, local strain, temperature factors, the accelerating voltage,
and the ionic and covalent bond charges associated with the atoms. It is remarkable that CBED has finally given us the tool we need
to see chemical bonds in inorganic crystals, for reasons given
Before discussing the QCBED and related methods, it is im- below. Later, we’ll see (in Fig. 5.6) the bond-orbital charge den-
portant to understand the simple relationship between select- sity for copper oxide (cuprite) obtained using a combination of
ed-area diffraction (SAD) ‘point’ patterns and CBED patterns. CBED refinement (for the low orders) and X-ray measurements
This is shown in Fig. 5.1. The important thing to understand (for the high orders and temperature factors). Because of the
is that points P and Q in the illumination aperture are ‘imaged’ errors introduced by ‘extinction’ (a multiple-scattering effect
independently onto the detector at P′and Q′ by the objective for X-rays within and between different mosaic blocks), it is
Chapter 5

lens, which is not shown. (We know this because dirt on the not possible to obtain images of this kind using X-ray crystal-
condenser aperture is seen ‘in-focus’ at the CBED disk edges.) lography alone. The QCBED method has been successful for
This means that if this illumination aperture C2 were an iris two reasons:
 148 5 Electron Crystallography, Charge-Density Mapping, and Nanodiffraction

QCBED Can Be a Big Project! QCBED Beats Other Methods: Three Reasons
Part I

Make sure you need the bonding information badly The probe is so small it avoids defects, thickness can be
enough to justify the time and effort before you begin! accurately determined, and electron scattering from an
 9 ion rises steeply at low angles (unlike that from an atom,
or X-ray scattering). 9

zz Electron scattering at small angles is far more sensitive to CBED is an extraordinarily efficient technique. The wide range
the charge state of atoms in a crystal than either X-rays or of illumination angles used provides a large amount of informa-
neutrons. tion (in the form of rocking curves) from many diffracted orders,
zz The small CBED probe is so small that, unlike X-rays, it sees presented simultaneously, from the smallest possible region of
only perfectly crystalline material. crystal. Different regions of the pattern are sensitive to different
crystal parameters; the following is a summary of all the infor-
This means that the ‘perfect crystal’ or ‘multiple scattering’ the- mation that can be extracted from incoherent CBED patterns:
ory will be an excellent fit to the data. (The probe is smaller than
one mosaic block, in X-ray terms). Thus QCBED now allows zz The outer reflections and HOLZ line intensities are sensi-
us to make ‘extinction-free’ measurements on real materials, tive to atomic position parameters and temperature factors,
using the multiple-scattering computer simulations appropri- the inner reflections to interatomic bonding effects, while the
ate to a perfect crystal. But the measurements are difficult and inner HOLZ lines crossing the central disk may be used to
time-consuming – a typical density map project, using available determine lattice parameters, to find the Bravais lattice and
software, might occupy a good postdoc for six months. to measure local strains.
zz The whole pattern may be used to determine the crystal space
Two types of CBED patterns may be distinguished – coherent group, as described in Chap. 4 in this volume.
and incoherent. To a good approximation, field-emission instru- zz Two-beam patterns can be used to make quick estimates of
ments give coherent patterns, while LaB6 and tungsten sources crystal thickness in crystals with small primitive unit cells.
give incoherent patterns. Unless the diffraction disks overlap,
the two kinds of elastic-scattering patterns should be identical In this section we discuss the relevant theory for extracting this
for parallel-sided slab samples of perfect crystal, since, as dis- information.
cussed above, each plane-wave component of the incident cone
of illumination acts independently, and scattering is only possible X-rays are diffracted by the electron charge-density distribution
through discrete Bragg angles. In this section we are concerned in a crystal, while electrons are diffracted by the electrostatic po-
with the information that can be extracted from incoherent tential, including the nuclear contribution. (X-rays don’t see the
CBED patterns, for which the probe diameter, for reasons of in- nucleus, electron beams do.) These quantities are related by Pois-
tensity, usually exceeds 50 nm. Strictly speaking, the distinction son’s equation. If the charge density is parameterized in terms
between coherent and incoherent CBED is made according to of atomic coordinates, multipole expansions, Debye–Waller
whether the coherence width of the electron beam in the plane factors, strains, or bond charges, then a limited number of these
of the illumination aperture is larger than (coherent) or smaller parameters may be refined from QCBED patterns, depending on
than (incoherent) the illumination aperture. In the former case the size and power of available computers. Because of this limit,
the probe itself may be considered to be completely coherent. no general method for solving unknown crystal structures by
In the later, it is partially coherent. In general, the incoherent electron diffraction has been devised; however, few-parameter
CBED patterns discussed in this section will be obtained from structures may now be solved in many cases.
rather thicker crystals. Thicknesses of 0.1–0.8 µm are common
Chapter 5

for incoherent CBED, depending on the kV used. Probe widths What we mean by the term ‘refinement’ is a quantitative com-
are also generally larger for incoherent work (in the range 10– parison of experimental, energy-filtered microdiffraction patterns
100 nm), being determined by the product of the total magnifi- with calculated patterns, using methods that are closely similar
cation of the probe-forming lens system and the virtual source to the Rietveld method of X-ray and neutron diffraction, and that
size, rather than the aberrations of the lenses. Nevertheless, for include all multiple-scattering effects. The comparison is based
these thicker crystals, all the concepts described in this and the on a goodness-of-fit index, using standard least-squares optimi-
following section can also be applied to coherent CBED patterns zation methods. With a few important exceptions, the atomic
if the diffraction orders do not overlap. positions have been known in all published QCBED work, so
that refinement consists of adjusting the other parameters listed
above for best fit. Since global minima are most likely to be
found if the starting parameters are close to their final values,
the crystal structures which have been solved by CBED have so
far been simple, few-parameter structures.
Chapter 5
 5.2 Quantitative CBED for Charge-Density Mapping 149

Part I
Fig. 5.3 Silicon (111) systematics at 196.15 kV. The bright disk is (000), with
(111) disks on either side. Fine HOLZ lines can be seen running parallel to the
straight line drawn across all the disks. The intensity along the red scan line is
shown in Fig. 5.5

a set of pixels, one in each Bragg disk, but now just to the side
of the previous family. In this way, our assignment of Kt values
to every pixel in the (000) disk is sufficient to define the beam
Fig. 5.2 Two Ewald sphere orientations differing by α, just off (continuous lines) orientation for every pixel in the entire pattern. We can then
and on (dashed lines) the Bragg condition. Note the direction of Kt and how it compare computed intensity distributions with the experimental
appears in Fig. 5.4. The excitation error is Sg – this (or Kt) tells the computer pro- pattern – these are generated as functions of sg (or Kt). We are
gram the incident beam direction for every pixel in the detector. The continuous assuming that the pattern comes from a known crystal structure
and dashed diffracted rays would go to different detector pixels, since they travel and has been indexed.
in slightly different directions
Figure 5.2 shows how an excitation error can be assigned to
A quantitative analysis of CBED patterns is only possible if a every pixel in the CCD camera used to record the CBED pattern,
unique excitation error sg has been assigned to every point on using the relation
the experimental pattern. For 2D patterns it is more convenient
sg  2B =  g˛ 5.1
to use the value of Kt, the tangential component of the incident
electron beam wave vector, as discussed further below. The ex-
citation error is a measure of the deviation from the Bragg con- Here the excitation error sg is defined in terms of the deviation
dition, and measures the distance from a reciprocal space point angle α from the Bragg angle. The component of the incident
to the Ewald sphere, measured in the direction of the surface wave vector Ko that lies in the plane of the crystal slab is Kt.

Chapter 5
normal. (See also Sect. 12.6 of W&C.) Figure 5.3 shows an experimental energy-filtered CBED pat-
tern for the (111) systematics reflections in Si. The Omega filter
It is very important to understand, from Figs. 5.1 and 5.2, how was set to accept all electrons falling within a range of about
Kt defines the incident beam direction. For a beam along a zone 8 eV around the elastic zero-loss peak, in order to reduce in-
axis, Kt = 0, corresponding to the pixel at the center of the (000) elastic background. The accelerating voltage was measured (as
disk. Centered on this origin we can lay down the 3D reciprocal described later) as 196.15 kV, corresponding to a nominal setting
lattice, which will define one pixel in every other CBED disk. of 200 kV. Excitation errors must be assigned along the red line
(The experimental distance to a neighboring disk will be X = L g shown, so that we can obtain a plot of diffracted intensity against
λ, as in Fig. 5.1). But the convergent-beam illumination provides excitation error (as shown later in Fig. 5.5).
rays filling a cone of orientations onto the sample. So now we
Chapter 5

take a different value of Kt, corresponding to a beam coming in at Although the background of inelastic phonon scattering is greater
a small angle to the zone axis, and arriving next to the pixel at the in the zone-axis orientation, it may be simpler to match the entire
center of the (000) disk. Again we can erect a lattice and define zone-axis pattern to calculations. For this more general case of
 150 5 Electron Crystallography, Charge-Density Mapping, and Nanodiffraction
Part I

Fig. 5.5 Comparison of experimental (dots) and calculated intensity (arbitrary

units) along the line drawn in Fig. 5.3 for the elastically filtered Si (111) sys-
tematic Bragg reflections. With this agreement between theory and experiment
it is possible to measure structure factors to an accuracy of 0.1 % in the most
favorable cases, which is sufficient to directly observe the bonds in the resulting
charge-density map of a crystal

Debye–Waller factors, unit-cell parameters, accelerating volt-

age, sample thickness, and possibly trial values of the absorption
parameters (e.g., those given by Bird and King (1990)). These
numbers take account of the depletion of the Bragg scattering,
which we detect, by inelastic processes such as plasmons, which
cause some electrons to lose energy, so that they cannot pass
through the filter. The direction of the incident beam is defined
by the two components Kx, Ky of Kt, which also defines a family
of ‘conjugate’ points (such as all the P′ in Fig. 5.1), differing by
reciprocal lattice points, one in each CBED disk g. For a given
value of Kt, the program will give the intensity Ig(Kt) for one
Fig. 5.4 The Laue circle in the ZOLZ. The definition of Kt is indicated, with the pixel in every Bragg disk. The variation with Kt gives all the
electron beam wave vectors K and Ko other pixels. The program should include the effect of HOLZ
lines (3D diffraction), absorption, and should be applicable to
a 2D pattern, the excitation errors can be assigned to each point noncentrosymmetric crystals. A more advanced program is de-
in the experimental pattern using scribed by Zuo (1993, 2000); here many-beam code is combined
with a least-squares optimization routine in order to automate
2Ksg = −2K t  g − g2 5.2 matching experimental and computed patterns. This program
automatically adjusts the low-order structure-factors and thick-
where Kt = 0 in the axial orientation. Otherwise, we must deter- ness for best fit to the diffraction data, starting with the values
mine the value of Kt, which defines the center of the Laue circle for neutral unbonded atoms. A preliminary refinement will also
for the central point of the CBED pattern in the (000) disk, as be needed to optimize the beam direction values for each pixel.
shown in Fig. 5.4. This circle is formed by the intersection of the Figure 5.5 shows the result of using this program to fit the ex-
Chapter 5

Ewald sphere with the zero-order Laue zone (ZOLZ), or with a perimental data in Fig. 5.3.
plane of reflections approximately normal to the beam. Kt is a
2D reciprocal-space vector in the ZOLZ, which runs from the The fitting is achieved by minimizing a goodness-of-fit parame-
center of the circle to the origin of reciprocal space. The circle ter χ2, using the method of maximum likelihood:
can often be seen as a bright ring of reflections passing through
the origin, allowing an approximate assignment of Kt by inspec- X fi  .dtheory − I exp /2
2 = i i
5.3
tion. Different values of Kt can then be assigned to every pixel i
i2
of the experimental pattern within the central (000) disk. The
Bragg angle and the electron wavelength must be accurately Here the experimental CBED intensities are given by Iiexp, and
known. the calculated points are given by Iitheory. The fi is a weight coeffi-
cient, which can be adjusted to increase the importance of certain
Many software packages are now available for simulating CBED contributions to χ2 that are sensitive to particular parameters.
patterns. As input, they require the atom positions and types, the Here σi is the variance of the ith point, which can be measured
Chapter 5
 5.2 Quantitative CBED for Charge-Density Mapping 151

from successive experiments or by using i2 = Ii , assuming brational amplitude of the atom. We look up the electron atomic
exp

Poisson statistics. Further, c is a normalization coefficient, which scattering factor, fie(s), in a Gaussian expansion table by Doyle
can be found by either normalizing the theory and experiment and Turner (1968); Ω is the volume of the unit cell. This equation

Part I
at a particular point, or by taking the first-order derivative of χ2. expresses the great advantage of electron diffraction compared
We have found the simplex algorithm (Press et al. 1986) to be the to other probes of condensed matter, because it appears to show
most robust (if not the fastest) method for finding minima in χ2, that, for zero scattering angle (s = 0), whereas the X-ray scat-
but no algorithm can guarantee finding a global minimum. You tered intensity is equal to the number of electrons in the atom
will have to repeat the entire refinement using different starting or ion, the electron scattering from an ion is infinite (but finite
parameters to confirm that a minimum is not local. from an atom)! The limit of this expression for zero angle has
been discussed in many papers in the literature (e.g., O’Keeffe
Noncentrosymmetric Crystals and Spence 1994); when screening is properly considered in a
crystal it remains true that the electron scattering from an ion
rises dramatically (but fortunately not to infinity) at small angles
These are much more difficult for QCBED, since you (relative to the neutral atom), whereas the X-ray scattering does
must measure both the magnitude and phase of all the not. Thus electron diffraction becomes extremely sensitive to
low-order structure factors! 9 ionicity at low scattering angles, and can provide entirely new
information in just the angular range where X-ray diffraction
has a serious problem with errors due to extinction (multiple
The two- and three-beam expressions for the intensity of Bragg scattering). The mean value of the electrostatic potential Vo (i.e.,
beams can be very useful in understanding the general form of the electron structure factor for s = 0) can be measured by elec-
the intensity oscillations in CBED disks and these were widely tron holography, as you’ll see in Chap. 8; it is probably more
studied in the days before fast computers. See Zuo et al. (1989) to sensitive to these tiny but crucial bonding effects that control
understand how the absence of inversion symmetry affects these materials properties than any other measureable quantity. Un-
patterns, and how they can be used to measure structure factor fortunately this quantity also depends weakly on the shape of
phases in acentric crystals with an accuracy of better than one the sample, but is usefully sensitive to the surface dipole layers
degree (this is not possible by other methods). These accurate of a thin slab.
phase measurements are needed to map out charge densities in
acentric crystals, which has been done in very few cases. (For
centrosymmetric crystals, with a suitable choice of origin, the As discussed in the last section, the characterization of
phases are all zero or 180 degrees.) the detector and the measurement of accelerating voltage
are most important for QCBED. 9

How You’ll Spend Your Time Some of the materials whose bonding has been studied by
QCBED include MgO, Si, GaAs, YBa2Cu3O7-δ, the transition
Most of the time involved in a QCBED project will prob- metal oxides (conductivity is due to charge hopping in these
ably be spent calibrating and characterizing the detector, ‘strongly correlated’ materials), GaN, TiO2, MgB2, AlN, γ-TiAl,
then measuring the kV accurately, using Si HOLZ lines. Cu2O , Cu, SiC, and BeO (for accurate phase measurement in a
 9 noncentrosymmetric crystal). Papers on these and other materi-
als can be found by web search with keyword CBED. Perhaps
the most accurate measurement of structure factors to date was
The relationship between the dimensionless X-ray structure fac- intended to answer the question “How ionic is MgO?” and made
tors fiX(s) and electron structure factors Vg (in volts) is important, use of DFT calculations for comparison. An important analysis is
since we must start with an approximate electrostatic potential the measurement of the slight rotation of oxygen octahedra with

Chapter 5
for the sample. Following Mott and Bethe we can write the fol- temperature in SrTiO3 by QCBED (Tsuda and Tanaka 1995). It
lowing for atoms of atomic number Zi in a unit cell: is important to emphasize that the result of this work is a quan-
X  Zi − f x .s//  titative density map, in which, for example, the absolute number
Vg =
jej i
exp.−Bi s2 / exp.−2ig:/ri of electrons in the unit cell per unit volume (surplus to a known
16 2 "0 ˝ i s2 neutral atom) is mapped out over the cell volume between con-
47:8780 X e tours.
= fi .s/ exp.−2igri /
˝ i The result of this analysis is a charge-density difference map,
5.4 which is the difference between the measured total charge den-
sity and that of a set of calculated atoms or ions placed on the
Chapter 5

Here the sum is over the atom positions ri in the unit cell, s = sin lattice sites. This procedure is needed because the changes in
θB/λ = |g|/2 and the Debye–Waller factor Bi = 8 2 hu2 i has been total charge density due to bonding are so small, involving per-
introduced for atomic species i, with hu2 i the mean-square vi- haps 0.01 % of the total density.
 152 5 Electron Crystallography, Charge-Density Mapping, and Nanodiffraction

calculations, are seen to arise predominantly from states of one

It is important to understand that practically all of the particular orbital character, can we assign features in a difference
bond charge between two atoms is already accounted for map to an ‘orbital.’ In simple terms, we must assume a radial
Part I

within the overlapping charge spheres from neutral (e.g., form for the density. Then the monopole contribution defines the
gas-phase) atoms, even without any additional charge re- ‘occupancy,’ or number of electrons in the ‘orbital.’
laxation (and energy reduction) to the crystalline state.9
Ionic and Covalent Parts

It is this additional relaxation that we seek to map out by

QCBED; it’s a tiny effect! It requires measurement of X-ray A best-fit spherical charge density laid down on each
structure factors (the Fourier coefficients of charge density) to an atom site is used to define the ionic part of the bonding
accuracy of better than 1 %. It is equally important to understand – everything else which is nonspherical is counted as
that ionicity is not strictly an observable quantity: it depends on covalent. 9
how charge is partitioned between different atoms, i.e., it de-
pends on ionic volume. Nevertheless the concept is widely used
and forms the basis of the ‘book-keeping’ that ceramists and The nonspherical part will be fitted with higher-order angu-
chemists use in keeping track of electron transport in crystals. lar-dependent terms. Hence in the cuprite difference map shown
in Fig. 5.6, the spherical ion part has been subtracted, leaving
We must now discuss how we decide which part of the charge only the nonspherical part. If cuprite were entirely ionic, this
density is ionic and which is covalent. There have been many map would be blank! In the multipole model, the crystal charge
approaches to the partitioning problem (see the online refer- density is expanded as a sum of nonspherical pseudoatomic den-
ences). We’ll use the example of cuprite, Cu2O. The results are sities,
shown in Fig. 5.6. Energy-filtered, 2D, CBED patterns were
recorded through the Omega energy filter on the Leo 912 using
X
crystal .r/ = ia .r − R − ri /
a liquid-N2-cooled, double-tilt holder, and the resulting inten- R;i
sity along lines in the patterns was compared with Bloch-wave
calculations. Structure factors were automatically adjusted in with
the calculations by simplex optimization for best fit to the data, X
using the known atomic coordinates and approximate tempera- a .r/ = S .r/ + PIm˙ RI .r/yIm˙ .; / 5.5
ture factors, which were later refined further. The Mott formula Im
(Eq. 5.4) then allowed us to obtain X-ray structure factors,
which are the Fourier coefficients of charge density, from the Here S is the spherical atomic charge density, yIm˙ .; / is the
Vg values. We could then map out the valence electron charge-­ real spherical harmonic function and RI .r/ = rnI exp.−˛r/ is the
density distribution by subtracting the charge density for a set of radial function with population coefficient P. The basis functions
neutral atoms (or ions), calculated by various methods, from the are normalized with NIm˙. In the model,
measured total charge density. However, this difference between a a a a
Fourier maps (the ‘deformation density’) is a very inefficient Cu = Cu + + .1 − q/.Cu − Cu+ /

way to present data, since not all the plane-wave Fourier coef-
ficients are independent. and
Oa = Oa + 2q.Oa 2− − Oa /
Ionicity

with q as the charge transfer from Cu to O. We obtain the spher-

Not strictly an observable quantity, ionicity depends on ical, atomic-charge density from multiconfiguration Dirac–Fock
Chapter 5

how we partition charge between atoms. It is the total (MCDF) calculations. For the nonspherical part, we can use
charge density in the unit cell which really matters. 9 local symmetry-adapted spherical harmonic functions. For ox-
ygen, we can include cubic harmonics of octupole and hexade-
capole character using a local coordinate with the z-axis along
A far more efficient method (basis set), which converges more <111>.
rapidly with many fewer terms, uses spherical harmonics, or
solutions for the wave function for the hydrogen atom. Here we Measurements using electrons and X-rays are frequently done
use this ‘multipole expansion’ to represent the charge density, at different temperatures. A combined refinement can some-
not the wavefunction. In this way we obtain an approximation to times therefore be based on the almost linear dependence of
the charge density of different ‘orbitals.’ However, it is important temperature factors between 100 K and room temperature, with
to realize that individual orbitals do not actually exist (since the the exact temperature of the electron measurement treated as a
choice of basis functions is arbitrary). Only if the difference map parameter (there may be uncertainties regarding beam heating).
shows localized charges which, when compared with theoretical In the case of Cu2O, anharmonic temperature factors lead to a
Chapter 5

Licensed to ([email protected])
 5.3 Strain Mapping, High Voltage, Lattice Parameters Measured by QCBED 153

Fig. 5.6 a Measured surfaces of a b

constant charge-density difference in
Cu2O (cuprite). Blue shows less charge

Part I
than MCDF ions (producing ‘holes’ in
the shape of dz2 orbitals), red is more.
Note metal–metal bonds at center of
copper tetrahedra. (The coppers form
an FCC lattice.) If cuprite were purely
ionic (Cu + O--) this picture would be
blank (Zuo et al. 1999). b Measured
surfaces of constant charge-density
difference in Cu2O along the (vertical)
copper–oxygen bond. Blue shows Copper
less charge than MCDF ions (produc-
ing a dz2 ‘hole’), red is more

Copper

Oxygen

significantly better fit. Temperature factors enter in two places 5.3 Strain Mapping, High Voltage,
in the dynamical theory of electron diffraction. By multiplying
the electron-scattering factors, the temperature factor controls Lattice Parameters Measured
the width of the atomic potential, which becomes wider at high by QCBED
temperatures. (Hot atoms are fatter!) Secondly, the temperature
factor also occurs as an exponential damping on the thickness
dependence of the beams, thus limiting penetration. In 2005, the International Technology Roadmap for Semiconduc-
tors (ITRS) announced that, whereas the intentional introduction
It is essential that you quote errors for the electron densities. You of stress to increase computer speed was well established, ‘direct
can analyze the errors in this analysis using the bootstrap method measurement of stress in a nanosized buried area such as the
(real statistics) by introducing random noise into simulated Fou- stressed channel (of a device) is not presently possible.’ In this
rier coefficients. The experimental measurements have Gaussian section we continue the account given in Chap. 4 in this volume
noise with a standard deviation. We then fit this noise-added data to see how this may done using QCBED. Stress and strain map-
set. We repeat this process several times, and so obtain the stan- ping at the nanoscale is needed in many other areas of materials
dard deviations in the density map. We also compare band-struc- science; the results may often be related to impurity atom con-
ture calculations for a Cu2O crystal with the experimental results centrations or stoichiometry changes.
using the Wein 2000 DFT code. In summary, there were two
main results from this work on Cu2O: Using the fine HOLZ lines which cross the central disk of a
CBED pattern (see Fig. 5.3 and 5.7), it is possible to measure

Chapter 5
zz It gave direct evidence for d-s hybridization on the Cu I site, lattice strains from regions of crystal whose dimensions are
resulting in d-orbital holes. somewhat larger than the probe width and averaged through the
zz It provided the first clear experimental evidence for metal– thickness of the sample. The uncertainty principle relates the
metal bonding, in the form of covalent charge concentration width of a HOLZ line to the width of crystal, which contributes
between Cu atoms. to it. Thus, for a HOLZ line, one tenth of the width of the central
disk, the contributing width of crystal is ten times the width of
We found that about 0.22 electrons are removed from dz 2 states an ideal diffraction-limited probe. This ‘width’ of crystal must
(creating holes), and moved to from the metal–metal bond be measured normal to the HOLZ planes. Strains cause the fine
charges. In the figure, the blue dumbbells are the dz 2 orbital HOLZ lines to shift laterally relative to the ZOLZ intensity dis-
holes (with less charge than simple spherical ions on the lat- tribution, and this effect can be quantified. We can also relate
Chapter 5

tice sites), while the red blobs are the surplus charge between these strains to compositional variations using Vegard’s law. The
metal atoms. smallest strain that can be measured is about 0.0001 (10−4); how-
ever, there are some complications:
 154 5 Electron Crystallography, Charge-Density Mapping, and Nanodiffraction

1. The strains may have resulted from the process of thin- As mentioned in Chap. 4 in this volume, it is possible
ning the TEM specimen. This problem may be addressed to superimpose the HOLZ lines on a shadow image of
by using thicker specimens and a higher kV, combined the sample by using an out-of-focus electron probe; we
Part I

with finite-element modeling of the thin-film elastic re- can then combine, in one image, strain measurement
laxation. and real-space imaging at limited resolution. You’ll find
2. The HOLZ line positions are equally sensitive to the kV the history and applications of this ‘convergent-beam
used (see Fig. 5.7) and to strains. The strain ∆a/a due imaging’ (CBIM) approach traced in Humphreys et al.
to an angular shift ∆θ in a HOLZ reflection g may be (1988). The method is closely related to the formation
estimated from of Ronchigrams and large-angle CBED patterns that we
discuss below in more detail. If a small time-dependent
2K0 
5.6 variation in one of the lattice constants (or cell angles) is
a  EO g
− = = = 
a  2EO g g2 added before displaying the pattern, it will immediately
become clear from the resulting wobble in some line
From Eq. 5.6 we see that the sensitivity of HOLZ line positions positions, which lines are most sensitive to which cell
to strain increases with accelerating voltage and angle, so that constants or angles. In this way you can make the best
we should always use the highest-order HOLZ lines. HOLZ lines choice of orientation.
become sharper in thicker crystals. If these are faint, a cooling
holder and energy filter will increase their contrast. In an instru- We may conclude that high accuracy in the measurement of
ment fitted with continuous accelerating voltage controls, we can lattice parameters or accelerating voltage can only be achieved
measure strain directly at the microscope in cubic crystals by only at the cost of considerable computational effort. The aver-
restoring a HOLZ line pattern to its reference shape for the un- age composition and structure factors must be known (because
strained crystal and noting the voltage change ∆Eo needed. Since changes in composition affect any coupled low-order reflections,
a fractional change ∆ in lattice spacing produces a fractional and hence the ‘dynamical shifts’), and thin-film relaxation effects
change ∆/2 in accelerating voltage, it is clear that HOLZ lines may pose the greatest problem for many materials, depending on
may provide a rather accurate method of relative strain mapping; elastic constants, specimen geometry, and chemical inhomogene-
however, we must consider the ‘dynamical shifts’ discussed be- ities. (Spinodal decomposition in a thin film, for example, causes
low. In the general case involving noncubic crystals, it would a characteristic elastic relaxation.)
be necessary to measure all six independent cell parameters to
characterize the strain fully. (This number may be reduced by We can draw some rather rough conclusions. At about 100 kV,
using symmetry arguments; the expansion is often assumed to and ignoring thin-film relaxation and atomic number and density
be isotropic so that no symmetry change is involved.) effects, strains and accelerating voltages may be determined to
about one part in 100 (at best) using a straightforward application
3. Multiple scattering can shift the positions of HOLZ of the Bragg law; they may be determined to about one part in
lines from that predicted by Bragg’s law. The simplest 1,000 (at best) using the perturbation corrections (Bethe poten-
demonstration of this effect occurs if Bragg’s law is used tials), and they may be determined to perhaps one part in 10,000
to determine accelerating voltage using the HOLZ lines (at the very best) using a full dynamical refinement.
seen at three different major zone axes of the same Si
crystal. The results vary by several kV due to ‘dynami- One of the most critical industrial applications of QCBED has
cal shifts.’ The effect has been analyzed in detail using been to determine the strain map around individual semiconduc-
three-beam theory, and may be avoided by minimizing tor devices. Here strain is intentionally introduced to increase
the excitation of strong beams. This suggests that we carrier mobility and speed. An example is based on a comparison
should use orientations near sparse zone-axes for great- of multiple-scattering simulations of the entire central CBED
est accuracy. Zuo (1992) gives an example of a detailed disk with experiment for a shallow trench isolation structure.
analysis: variations in the a/b cell constant ratio were For sparse zone axes, the dynamical corrections may be small, so
Chapter 5

mapped out for YBa2Cu3O7-δ, in order to measure the that we can use the much faster kinematic simulations; however,
local oxygen deficiency δ. Using digitized patterns and a dynamical simulations allow us to use a wider range of zones.
Rietveld-type analysis, with a goodness of fit index, the We can even map the strains around a 65-nm node (2005) met-
accelerating voltage could be measured to an accuracy al-oxide field-effect transistor with SiGe source and drain. Here
of 14 V, and the cell constants to accuracies of 100 pm we used a FEG Libra TEM fitted with an Omega elastic-energy
and 60 pm. As an example, it was found that an oxygen filter at 200 kV. Elastic relaxation during sample thinning is al-
deficiency of 0.1 was responsible for the strain in one ways a concern; however, the observation of sharp rather than
local region. The measurement of local lattice spacings blurred HOLZ lines indicates uniform strain. The extraction of
using QCBED and full dynamical pattern-matching cell parameters from HOLZ line distributions in the central (000)
is used to make local maps of lattice parameters. This CBED disk has also been automated in the kinematic ASAC
method should be used when you need the greatest ac­ software package.
curacy. It may also be used with coherent CBED pat-
terns, as in the strain measurements in Si/Si–Ge layers.
Chapter 5
 5.4 Spot Patterns – Solving Crystal Structures 155

Part I
119.4 kV 119.6 kV 119.8 kV

117.6 kV 117.8 kV 118.0 kV

Fig. 5.7 Showing changes in the appearance of HOLZ lines in a silicon CBED patterns with small changes in accelerating voltage. Note changes in small triangles

5.4 Spot Patterns – Solving Crystal Since then, the biggest successes of TED have been in biology,
which is relevant to the possibility of solving the organic struc-
Structures tures of interest to materials science (such as organic semicon-
ductors, block copolymers, and structures at the organic–inor-
In the early days of electron diffraction, researchers in Russia put ganic interface). For organic membranes which form 2D crystals
a lot of effort into solving crystal structures. Much of this work exactly one molecule thick (cf. bacteriorhodopsin, the purple
was devoted to clay minerals, layer structures, and the intriguing membrane) the problem of measuring thickness is solved, and

Chapter 5
possibility of detecting hydrogen. Electrons, unlike X-rays, are the light C, O, N, and H atoms in proteins will not produce much
scattered by the nucleus, and because ionization increases elec- multiple scattering. Then the standard X-ray software could be
tron scattering, images of hydrogen atoms in crystals have been used to solve the phase problem. In practice, because proteins
reported by this method. contain so many atoms, and because we can’t use the anom-
alous-absorption methods of X-ray phasing, HRTEM images
Poor Reproducibility and “Quality” taken over a wide range of tilts are used to help phase the data.
The Fourier transform of a weak-phase-object image-intensity
distribution gives the structure factor phases directly, within the
These are the main limitations of electron diffraction data limited resolution of the image-forming lenses.
for solving crystal structures. Can you collect the same
data twice; e.g., from different areas of the same crystal-
Chapter 5

There have been many attempts to extend this approach to inor-

line sample. If not, does it matter? 9 ganic materials. But your own experience should show that this
type of electron diffraction data is rarely reproducible. Could
 156 5 Electron Crystallography, Charge-Density Mapping, and Nanodiffraction

you record spot diffraction patterns from the same sample that diffraction intensities are symmetric. (They must obey Friedel’s
showed the same relative intensities on two different occasions? law.) This alignment requires considerable skill; it also maxi-
mizes the phonon background, which unfortunately cannot be
Part I

Fortunately, great accuracy in Bragg-intensity measurement is removed by the energy filter. The aim is to obtain a trial structure
not needed to solve structures (correct phases are much more for further refinement. Variations in thickness under the illumi-
important), so the key to success is the intelligent use of ad­ nated region, the perturbing effects of multiple scattering, and
ditional information. We may have a situation where there are sample bending are the main difficulties. The curvature of the
many more measured Bragg reflections than wanted atomic co- Ewald sphere can be taken into account in a kinematic analy-
ordinates. Data quality has improved recently with the use of sis. Various approximations have been made even in the single-­
elastic filtering, low-temperature stages, and area detectors that scattering theory, such as neglect of excitation errors, or choice
are more linear than film. (Keep this last point in mind when of the power of the structure factor used to model extinction or
we talk about image plates.) For a completely unknown inor- multiple-scattering effects:
ganic structure, much additional information can be obtained by
CBED, including the space group and the cell dimensions and Testing
angles, as described in Chap. 4 in this volume. A simple method
of indexing an unknown spot pattern of known space group is
then to display the indexed patterns (e.g., using ‘WebEMAPS’) Are there any experimental tests for the absence of multi-
of any crystal that has that space group. You can then view this ple scattering which can be applied to an unknown struc-
pattern on a laptop at the TEM, and use the mouse to vary the di- ture? 9
rection of the simulated beam direction, until the indexed laptop
pattern resembles the ones (at symmetry axes) seen on the TEM
screen. You can obtain an estimate of the stoichiometry and of
the type of atoms present in an unknown material using XEDS 1. Since electron diffraction patterns with multiple scatter-
or EELS. Tables of bond lengths, and high-resolution images, ing do not obey Friedel’s law (i.e., I(g) ≠ I(-g)), whereas
can also provide additional information, and powder X-ray data single-scattering patterns do, a crystal that is known
may be available. a-priori to be noncentrosymmetric in some projection,
but which produces a symmetric diffraction pattern, must
For many oxides, the local structural unit, such as an octahedron, be diffracting under single-scattering conditions. If it was
may be known, and these arrangements can be seen directly in thick enough to produce multiple scattering, the pattern
HREM images. (Again, a short stacking sequence in the beam would lack the inversion symmetry which the crystal also
direction is needed for interpretability). Powder electron diffrac- lacks, and so reflect the true symmetry of the crystal.
tion patterns can be used to identify phases, from their char- 2. The angular width ∆θ between minima of a CBED
acteristic sequence of d-spacings, using the powder diffraction rocking curve from planes with spacing d is d/t for sin-
files. For example, ‘NIST Standard Reference Data Base 15’ is gle scattering, or approximately d/ξ if there is multiple
a powder diffraction file for electron diffraction. (Will we some- scattering. Here t is the sample thickness and ξg is the
day be able to enter a d-spacing sequence in Google?) Errors in two-beam extinction distance (see W&C Chapter 23).If
d-spacing measurements from electron diffraction patterns may, CBED disks of constant intensity are found in thin areas
however, lead to many candidate structures, but will exclude a of a light-element crystal, so that ∆θ is large, we must
far greater number. either have t small or ξg small. Since ξg is in fact large near
the Bragg condition for light elements, it is reasonable to
It has been argued that the weakly excited HOLZ reflections in conclude that we have single-scattering conditions.
the outer rings of the diffraction patterns avoid multiple scatter- 3. The essential requirement for single scattering is that
ing, and so will give a better estimate of structure factors. This the (000) beam is much stronger than any other; we can
is found to be true only for sparse zone axes, where it may be a check this directly from CBED patterns taken from nm-
Chapter 5

useful method. In general, the refinement for atomic positions sized regions near a sample edge. The small CBED probe
concentrates on these high-order reflections, whereas bonding reduces thickness variations.
studies are concerned with the lowest orders. 4. The observation of space-group-forbidden reflections is
not a good test for single scattering, since, as shown by
The technique of solving structures using spot patterns works Gjønnes and Moodie, they remain forbidden under multi-
best for layer-structure crystals with a short axis along the beam ple-scattering conditions at certain diffraction conditions
direction. These are more likely to provide extended regions of – including the Bragg condition. In SAD patterns from
constant thickness (a few nm thick at most), and the resulting large areas of slightly bent crystal, however, it may be a
2D density map will usually be sufficient to solve the structure, useful test, since these cover a range of orientations.
without requiring you to collect 3D data sets. Detailed calcula-
tions, showing the maximum thickness possible to avoid multi- Figure 5.8 shows the nature of the multiple-scattering problem.
ple scattering in oxides, have been published. For these very thin In the single-scattering (kinematical) approximation used to solve
samples the crystals must be aligned with the beam so that the crystal structures, the intensity of a Bragg beam g is given by
Chapter 5
 5.5 The Precession Method 157

sin2 .sg t/
Ig =  2 Vg2 t2
.sg t/2 5.7

Part I
I0 = 1

Here σ = 2πm|e|λ/h2 is a constant, t is the sample thickness, sg

is the excitation error, Io is the direct (000) beam, and Vg is the
Fourier coefficient of crystal potential (in volts) that we want;
from these we can obtain a potential map V(r) showing the
atom positions. (Using Poisson’s equation, we can convert this
into a charge-density map.) In the absence of absorption, for
noncentrosymmetric crystals, Vg = V-g* so that V(r) is real. For Fig. 5.8 Single-scattering intensity variation with thickness of the (111) Bragg
centrosymmetric crystals the Vg values are real, and the phase beam scattered by a thin silicon crystal at 100 kV in the (110) orientation. The
problem consists of determining their sign. actual oscillatory variation produced by multiple scattering is shown for compar-
ison. Solution of crystal structures requires single scattering conditions, in this
Based on this expression, the intensity of the (111) Bragg case samples less than about 4 nm thick
beam from a thin Si crystal in the [110] zone-axis orientation
is shown as a function of thickness at 100 kV. However, if all (Zeolites have enormous importance as catalysts, e.g., in man-
multiple-scattering effects are accurately included, the intensity ufacturing new clean fuels.) These radiation-sensitive, large-
oscillates as shown in the second curve. (Si here is a special unit-cell, silicate, cage-and-tunnel structures can be imaged us-
case – in general the multiply scattered beams show an irreg- ing low-dose techniques, and TED patterns obtained from the
ular intensity variation with thickness that is not periodic). To same small region as the images. The TED spot patterns are
the extent that the curves shown agree at very small thickness often streaked due to planar faults (or doubled by twinning), and
(t < 4 nm?), we can use the resulting Bragg intensities to solve the explanation of features in the spot patterns in terms of the
the phase problem and hence the crystal structure. Other beams HREM images makes a fascinating exercise. A straight-channel
will oscillate with a different period, so that their intensity ratio zeolite consisting of 12-membered rings was solved by Liu and
soon deviates seriously from the kinematic value. others, using a combination of HREM images and TED spot
patterns in this way, and provides a fine example for students to
There is no reason not to use the CBED mode under kinematic study. The addition of X-ray powder diffraction data to the elec-
conditions, thereby giving us data from very small regions of tron-diffraction and high-resolution-imaging data can greatly
defect-free, not-bent crystal, and flexible orientation determina- assist this work – e.g., indexed TED spot patterns from many
tion. In the zone-axis orientation, for example, one pixel could single-crystal grains might be used to determine the degeneracy
be selected from the center of each CBED disk to provide a kine- of the powder diffraction rings. In 1969, Allpress et al. discov-
matic spot pattern where the excitation error will be known (by ered that planar faults in transition metal oxides accommodate
geometry, Eq. 5.2 with Kt = 0) at each of these pixels or ‘beams.’ nonstoichiometry.
(Recall that CBED patterns are equivalent to a set of ‘point’ SAD
patterns, laid side by side.) These ‘oxygen-sponge’ crystals have responded to changes to
the oxygen pressure on our planet since life began. We agree
Despite the problems created by multiple scattering, radiation that, since 1969, HREM has proven much more powerful for the
damage, and crystalline imperfections, there have now been study of defects and interfaces in crystals than for solving crystal
many successes for electron crystallography, often combined structures, which awaits the development of techniques for the
with imaging. Because interest has often centered on radia- automated acquisition of 3D data.
tion-sensitive or disordered materials taken from a family of
related known structures, each case tends to be different, and no

Chapter 5
general, automated method has emerged. Going beyond the layer
structures discussed above (whose structure can often be inter- 5.5 The Precession Method
preted from a single projection) will require automated methods
for the collection of diffraction data in 3D; these have yet to be
developed – you have to bring the electron probe back to the Another important difference between X-ray diffraction and TED
same position after tilting the sample. This could be done either is that X-ray Bragg intensities are normally integrated in angle
by defocusing the probe and using the shadow image, by using – the sample is rocked through the Bragg condition during data
the STEM mode, or by adapting the software currently used for collection. In electron diffraction we have such a large width to
cryo-EM tomographic imaging. (Note that numerical phasing the rocking curve that in, e.g., SAD patterns we normally pick
methods generally work much better in 3D than in 2D.) off just one point within the rocking curve.
Chapter 5

Perhaps the most important example of electron crystallogra- It turns out that this angle integration can help reduce multi-
phy has been its application to the family of zeolite structures. ple scattering; this effect is used in the precession method of
 158 5 Electron Crystallography, Charge-Density Mapping, and Nanodiffraction

TED (W&C p. 295), based on the X-ray method of the same The last decade has seen rapid growth in applications of the
name. In 1939, Blackman showed that if the results of the two- precession method for solving nanocrystal structures, with the
beam intensity were integrated over angle or excitation error sg development of commercial systems and software packages –
Part I

(Chap. 23, W&C), then the result was independent of thickness interested researchers should start with the review of Midgley
if the thickness was greater than two extinction distances, so and Eggeman (2015).
that multiple-scattering effects were cancelled out. Specifically,
he found
W=1
Z
ˇg .t; W/ˇ2 dW
5.6 Diffuse Scattering, Defects,
INg .t/ =
ˇ ˇ

W=−1
Phonons, and Phase Transitions
ˇ ˇ2
 ˇUg ˇ t 5.8
= for t < 2g Since the cross-section for electron scattering is at least
sK 2 g 1,000 times greater than that for X-rays, while the source bright-
ness for field-emission guns and fourth-generation synchrotron
ˇ ˇ
 ˇUg ˇ
= for t > 2g
sKg undulators are comparable, it would seem that electron diffrac-
tion is the ideal tool for the study of point defects in crystals.
Here Ug = 1/(λξg) and w = sgξg. This seems too good to be true, (The instrumentation is a bit cheaper too.) Indeed this is one
but remember it assumes that only two Bragg beams are excited area where electron microscopy can readily compete with syn-
in the crystal. The precession method combines this useful ap- chrotron-based research. Leaving aside line and planar defects,
proximation with another – that high-order beams, the ones that the main contributions to the continuous scattering seen between
are most sensitive to the atom positions we seek, also tend to Bragg reflections arises from thermal-diffuse (phonon) scatter-
be weak and so less subject to multiple scattering. In the CBED ing, from substitutional disorder, and from interstitial defects or
method, the integration over sg could be accomplished digitally clusters, all of which is referred to generally as ‘diffuse scatter-
after reading out the CBED pattern, just by adding together all ing.’ These forms of scattering can be distinguished by the fact
the pixels in each CBED disk. that thermal-diffuse scattering falls to zero at zero angle (unlike
scattering from substitutional disorder, which is a maximum),
However, not all CBED disks in a single zone-axis pattern and by the temperature dependence of the scattering.
span the Bragg condition, as required by these equations (in
Fig. 5.4, those on the Laue circle do). The solution is to use Again, the ideal instrument is a field-emission TEM fitted with
beam-deflectors under software control to scan the incident an image-plate area-detector and an Omega filter. The filter will
beam around the surface of a cone, and to de-scan the beam remove plasmon and inner-shell background from the patterns,
after the sample on a similar cone (Vincent and Midgely 1994). but cannot separate the phonon scattering from the continuous
Thus the beam swings across the Bragg condition for each elastic scattering from the defects. (The use of low temperatures
high-order spot in succession, but all the intensity generated will help, but remember that the zero-point vibrational motion
at each reflection is summed into one point at the detector. makes a surprisingly large contribution to thermal scattering.) A
The patterns look rather like normal SAD patterns, but the cooling or heating stage is also useful, since diffuse scattering is
intensities are less affected by multiple scattering and extend to the ideal tool for the study of phase transitions and the defects
higher resolution. The use of image plates also has advantages associated with them. Diffuse scattering can quickly be simulated
here in view of their large number of pixels, dynamic range, for several models of the point defect of interest, for comparison
and linearity. These plates can conveniently be used with an with experimental data. The most likely difficulty is the necessity
Omega elastic filter and cold stage (to reduce the thermal dif- to average over equivalent orientations of the defect in the crystal.
fuse-scattering background).
A striking case is the occurrence of intense planes of diffuse
Chapter 5

How well does it work? In one application of the method (Wei- elastic scattering seen at the metal-insulator ‘Verwey’ transition
rich et al. 2006) all 11 heavy atoms in the asymmetric unit of a in magnetite iron oxide at 123 K (Zuo et al. 2000). The diffuse
(Cs, Nb, W)5O14 oxide framework structure were found correctly scattering has been analyzed to produce a model for the mecha-
after phasing electron precession data with the SIR97 program, nism of the phase transition in terms of spiral chains of octahe-
followed by refinement with SHELEX-97 (both popular X-ray dra. It provides a good example of what can be done, since the
programs). These atom positions from TED agreed within 9 pm diffuse electron-scattering measurements have been correlated
with those obtained by powder X-ray diffraction. But the posi- with bulk resistivity measurements against temperature. These
tions of neither the oxygen nor the cesium atoms could be estab- measurements can be used to reveal the character of nanoscale
lished, due to their weaker scattering. This was all done using a loops (interstitial or vacancy).
single zone-axis precession pattern, not full 3D data. The large
final crystallographic R-factor of 39 % suggests that consider- It is also actually possible to see the electron diffraction streaks
able multiple scattering remains, which limits the detectability around Bragg peaks due to individual phonon modes (as in neu-
of lighter atoms. tron diffraction) under the right conditions of sample thickness
Chapter 5
 5.7 Diffractive Imaging, Ptychography, STEM Holography, Ronchigrams, and All That 159

and beam divergence, using an image-plate detector and an ener- 2. Bragg diffraction if the probe is smaller than the unit cell.
gy-imaging filter. This shows directly the failure of the Einstein What happens when the probe is smaller than the crystal unit
model of independent vibrating atoms. It is the transverse acous- cell? Can we still have Bragg scattering? Figure 5.9a shows

Part I
tic modes that are mainly seen; in 3D we can plot out both lines the effect experimentally. In order to make such a small probe,
and sheets of this thermal-diffuse scattering, whose temperature it is essential that the diffraction disks overlap, in which case
dependence can suggest mechanisms for phase transitions. they interfere. This is explained in Fig. 5.9b. The idea is that
the probe is itself just an image of the point-like field-emitter.
Given the success of earlier work analyzing diffuse X-ray scat- (It’s similar to how a telescope forms an image of a star.) If
tering from defects and phase transitions (e.g., in alloys), this we ignore spherical aberration for the moment, the telescope
seems a very profitable direction for new research, now that we (Rayleigh) formula for the (diffraction-limited) size (FWHM)
have the imaging energy (or energy-imaging) filters, area detec- of the probe would then be
tors, and high-brightness sources that make quantitative electron
diffraction possible.
dp = 0:61=C 5.9

where θC is the ‘beam divergence’ angle shown in Fig. 5.1. Now

Bragg’s law gives the centers of the CBED disks at angles 2θB
5.7 Diffractive Imaging, in Fig. 5.9 according to
Ptychography, STEM Holography, dhkl = =2B 5.10
Ronchigrams, and All That
We can now combine these two expressions for the case where
the full-width of the probe, dp, is just equal to the d-spacing dhkl
Now we’ll introduce you to some of the ideas at the frontier of for the first-order reflection. The result is
the subject of electron microdiffraction – ideas that are suitable
for research topics. This section is actually particularly important R = 1:2B 5.11
if you want to learn how to align a sub-Å STEM.
So a 20 % overlap of the disks is needed to make the probe as
1. Defects and very small probes. We have repeatedly said that, small as the d-spacing. (Of course it will be larger if there is
on a field-emission instrument, electron nanodiffraction pro- spherical aberration, but the interference will still occur. Aberra-
vides the strongest signal from the smallest volume of matter tions cannot reduce the size of the probe, so we have lower limit.)
known to science. It follows that this should be the ideal mode Since, in this model, all the radiation comes from the same point,
for the study the structure of defects in thin crystals, provided we see coherent interference within the region of overlap. This
that the effects of radiation damage (such as hole-drilling and interference is the origin of contrast in the Ronchigrams that
contamination) are not severe. The use of conventional CBED we use to align sub-Å STEM instruments, for ptychography, for
patterns to study defects is described in Chap. 4. For smaller shadow imaging, and for in-line holography in STEM. So we’ll
probes, we will see in Sect. 5.8 that once the probe becomes now briefly discuss this. It is also fundamental to an understand-
smaller than the largest d-spacing, the microdiffraction inten- ing of the ‘Tanaka’ wide-angle CBED method and of the CBIM
sity distribution depends both on the probe location, which (convergent-beam imaging method). Remember first, that for a
must be known, and on defocus and aberration constants. (You normal CBED pattern from a parallel-sided, unbent slab of crys-
can sometimes determine the location by positioning the probe tal, if we ignore defects and inelastic scattering, then the intensity
over a symmetry element in the structure.) Because of this, the distribution within the CBED disks does not depend on the probe
analysis of these coherent nanodiffraction patterns becomes location. The probe spans many unit cells. A full analysis of the
very like that of high-resolution images – the structural and case where orders do overlap (Spence and Cowley 1978) shows
optical parameters of a model are adjusted for best fit to the the following three points:

Chapter 5
experimental pattern.
zz If Eqs. 5.11 and 5.12 are hard to accept, Fig. 5.9b explains
The Intensity Distribution Changes this (and more) using a ray diagram. The probe is focused a
distance ∆f above the sample. Cones of radiation are kicked
off to the side through angles 2θB as they pass through the
Only the intensity distribution in the overlapping region sample. You can see that these cones will overlap if the beam
of coherent microdiffraction disks changes with probe divergence is greater than the Bragg angle. The rearward ex-
position, aberrations, and probe focus. If the illumination tension of these Bragg cones toward the source defines ‘vir-
aperture is further enlarged, eventually the centers of the tual sources S,’ which are images of S and so are coherent
disks can no longer be identified, and all crystallographic with it. The rays from these virtual sources therefore interfere
information is lost, replaced by diffraction from the single
Chapter 5

where they overlap at the detector.

column of atoms under the probe. 9 zz If the sample is not crystalline but a small isolated object,
the source S would produce a shadow image of the sample
 160 5 Electron Crystallography, Charge-Density Mapping, and Nanodiffraction

within the central disk, and indeed this is what we see when your way out, to phase all the Bragg reflections (if you
we defocus the probe in CBED mode. (The image is inverted knew the phase of the first one). This is ptychography,
if the probe is underfocused below the sample.) first proposed in visible-light optics and later for electron
Part I

zz If the sample is crystalline, do we get a shadow image of microscopy by Hoppe (1969). Spence (1978) investigated
the crystal lattice? The answer is: most certainly – as seen these multiple scattering artifacts and Nellist et al. (1995)
experimentally Fig. 5.9a. You can also understand this re- demonstrated the method experimentally in STEM as a
sult from the ray diagram. By geometry, the virtual sources S means to ‘super-resolution.’ A review of the method is
must lie on the points of the reciprocal lattice for the crystal, given in Spence (2013), and the method has now been
and the two shown will produce Young’s fringes on a distant extended to nonperiodic samples and used in other fields
screen. These fringes then become just a magnified lattice (for example X-ray and optical imaging) as probably the
image of one d-spacing. Let’s check this. The magnification best general approach to aberration-free ‘lensless imag-
of the shadow image is just M = (L + ∆f)/∆f ~ L/∆f. The theory ing’ with super-resolution. John Rodenburg has led much
of Young’s pin-hole experiment (treating S as the pinholes, of this development with many remarkable results, such
separated by ds) would predict fringes at the detector with as lensless three-dimensional optical microscopy (God-
angular period α = λ/ds. Now ds = 2θB∆f, so the period of the den et al. 2014) and subnanometer ptychography using
fringes on the screen is P = Lα = M (λ / 2θB). But λ / 2θB = dhkl, a field-emission SEM (Humphry et al. 2012) – his re-
the first-order lattice spacing. Hence the fringes in Fig. 5.9a view gives an excellent pedagogical account of the entire
have the period of the crystal lattice, but are magnified by the field (Rodenburg 2008). The truly remarkable and unique
shadow-image magnification, which increases to infinity as feature of this method is that, for thin crystals, the step-
the defocus is reduced to zero. ping-out process gives you the phase of Bragg reflections
corresponding to d-spacings that are much finer than the
Why don’t we use these lattice images instead of HREM ‘lattice probe width, and which, by Eqs. 5.9 and 5.10, need only
images’? Are they less affected by aberrations? In fact they are be as small as the large first-order spacing (allowing a
affected (in a different way) by aberrations, but we can show that relatively large, electron-source size). Interference only
the resolution is ultimately limited by the size of the electron ever need occur between neighboring reflections, not
emitter. These images have many interesting properties. Those between the central beam and the highest order, as in
aberrations, which are an even function of scattering angle (like normal HREM imaging; it gives a systematic method
defocus and spherical aberration), cancel on ‘achromatic circles’ that involves recording these patterns as a function of
passing through the midpoints of the Bragg conditions. So de- probe position. This geometry is also a promising one for
tectors placed at those positions can be used to form an aberra- schemes directed at the ‘dynamical inversion problem,’
tion-free STEM image (like tilted-illumination HREM). What where the aim is to read in such a pattern, analyze it, and,
happens if we place a detector at such a position and then scan in real time at the TEM, produce a charge-density map of
the probe in a raster? the sample crystal structure showing all the atom types
and their positions.
Form a DF STEM Image 4. In-line holography. In Fig. 5.9b, if we imagine a com-
pact, isolated, noncrystalline sample, we have exactly the
geometry that Gabor proposed for holography in 1948.
The intensity within the region of overlapping discs g This arrangement has been used to reconstruct electron
varies sinusoidally with probe translation along g, and so holograms using a STEM (Lin and Cowley 1986). Again,
can be used to form a dark-field STEM lattice image. 9 ultimate resolution is limited by the size of the electron
source (i.e., by partial spatial coherence). For this to qual-
ify as an interpretable hologram, we require single-scat-
The theory of STEM imaging is treated in Chap. 11 – for our tering conditions.
purposes just note that we can form three kinds (at least) of 5. Ronchigrams. Ronchigrams are out-of-focus coherent
Chapter 5

STEM images: i) The Z-contrast image, formed from the (very) microdiffraction patterns formed with the illumination
high-angle scattering mostly beyond the range of Bragg scatter- aperture removed (see Fig. 5.9a). They are routinely used
ing where d-spacings are smaller than the vibrational amplitude to measure the aberrations of astronomical telescope mir-
of the crystal planes; ii) the dark-field image, formed with an an- rors. For STEM, this illumination aperture is the objec-
nular detector collecting all the overlapping, interfering regions tive aperture. Ronchigrams can be used to perform the
shown in Fig. 5.9a; iii) a bright-field image using some of the electron-optical alignment of aberration-corrected STEM
scattering in the central disk. instruments with the necessary very high precision. The
Ronchigram is an aberration figure, which can be read
3. Ptychography (from the Greek ‘to fold’). In Fig. 5.9a, into a computer and analyzed to give values of the aber-
the fringes within the overlap between two reflections ration constants. These values are then fed back to the ab-
could clearly be used to determine the phase difference erration corrector, which is then calibrated, and therefore
between them. By an obvious ‘stepping out’ process us- can be supplied with new desired aberration constants by
ing linked pairs of reflections, you could then bootstrap the operator, including zero.
Chapter 5
 5.7 Diffractive Imaging, Ptychography, STEM Holography, Ronchigrams, and All That 161

The (very simplified) principle of Ronchigram analysis is as

follows. Consider a noncrystalline sample in Fig. 5.9a. In the
presence of spherical aberration, the outer rays are too strongly

Part I
focused relative to the unaberrated paraxial rays. Hence the
magnification of the shadow image M = L/∆f varies across the
shadow image or Ronchigram. If these aberrated rays from the
outer zones of the STEM objective lens happen to be focused on
the sample (while the paraxial rays are focused beyond it), then
the magnification will be infinite for the shadow image formed
by the aberrated rays. The effect is seen as a bright ring of infinite
magnification in the Ronchigram, whose radius is approximately
θo = 2 (∆f / 4Cs)1/2. Hence CS can be found if ∆f is known. For a
crystalline sample in the two-beam condition, two large ‘eyes’
are seen at the stationary phase condition, where defocus and
spherical aberration effects tend to cancel and the result varies
slowly with angle. The ellipses seen in Ronchigrams can be used
to find CS , ∆f, astigmatism, and coma. A much more sophis-
ticated analysis is used in commercial autoalignment systems
for aberration-corrected STEM instruments, and this general
method, based on multiparameter optimization, is described in
detail in Pennycook and Nellist (2011).

6. Diffractive (lensless) imaging. With its origins in work

of David Sayre in the early 1950s and independently that
of Gerchberg and Saxton in 1971, researchers in the sig-
nal-processing field developed an algorithm in the early
1980s which will nearly always solve the phase problem,
given the far-field intensity scattered by an isolated, com-
pact, “real object.” (This means a sample with negligible
absorption, so that the refractive index is specified by
one real number.) So we have the possibility of replac-
ing our lenses with computers, and so obtaining aber-
ration-free images using radiations for which no lenses
exist! In principle, the resolution then becomes limited
only by the wavelength. The Fienup (1982) algorithm
iterates between real and reciprocal space, putting in
known information at each step. Three pieces of infor-
mation are found to be needed – the approximate shape
of the boundary of the object (mathematicians call this
the ‘support’), the single-scattered Fraunhofer diffraction
pattern intensity, and the sign of the scattering potential.
The more demanding case where the scattering potential
or refractive index is complex requires a very precise
knowledge of the sample boundary.

Chapter 5
This promising method has now been used to reconstruct aber-
ration-free images of samples from X-ray, neutron, and electron
diffraction patterns. For TEM, the big problem is the need for

Fig. 5.9 a Field-emission nanodiffraction pattern from thin FeS2 crystal. The dif-
fraction disks interfere where they overlap. Within the overlap region only, the
Chapter 5

intensity distribution depends on the probe position and aberation coefficients.

(Tanaka et al. 2003.) b Ray diagram to explain interference between overlapping
orders, in-line holography, Ronchigrams, and the Tanaka method
 162 5 Electron Crystallography, Charge-Density Mapping, and Nanodiffraction

an isolated sample, since any supporting substrate (such as thin age plate as an electronic film that requires a specialized im-
carbon) contributes too much scattering. Tricky! One spectacular age-plate reader. Image plates are the best system for recording
success in reconstructing an atomic-resolution image of a carbon weak diffuse scattering from defects or phonons; the latter also
Part I

nanotube from its coherent transmission diffraction pattern alone requires the use of an in-column filter tuned to the elastic scat-
can be found in Zuo et al. (2003). For this reason, approaches tering and a cold stage to minimize the thermal background
based on ptychography in STEM, which works for a laterally to scattering from defects. The ‘direct-injection’ solid-state
extended thin slab of sample, have become more popular, despite single-electron detection systems used for cryo-electron mi-
the increased complexity and the problem of exactly locating croscopy have yet to be evaluated for recording diffuse scatter­
the probe position, which could limit resolution. (The far-field ing; however, these ‘low-dose’ detectors are not suitable for
scattering, with plane-wave illumination, does not depend on the recording the associated Bragg reflections. CCD systems,
beam location.) These ideas are all further discussed and related while invaluable for a quick look at the data, introduce many
in Spence (2013). imperfections into the data, so that each CCD camera must
be individually characterized if CCD data is to be used for
QCBED analysis. (Guess how many image plate systems there
are near your lab.) The modulation transfer function (MTF),
5.8 Equipment for Quantitative the detective quantum efficiency (DQE), and the sensitivity of
the detector must all be carefully measured. This information
Electron Diffraction is then used to correct the detector response by deconvolution.
Deconvolution of the modulation transfer function of the de-
The accuracies cited above in QCBED work have only become tector system is possible without noise amplification, because
possible since the development of commercial imaging energy it is the low, rather than the high, spatial frequencies that must
filters for electron diffraction. These filters allow us to exclude be corrected.
electrons that lose more than a few eV when traversing the sample
from the measurements. A typical, modern, QCBED system might The accelerating voltage must be measured (e.g., using Si as a
be based on a field-emission TEM/STEM instrument fitted with standard sample) by refinement of HOLZ line positions. This
either an in-column or post-column imaging energy filter, and calibration of the microscope will probably have to be repeated
a liquid-N2-cooled, double-tilt goniometer with a large angular every few months. The MTF functions for similar CCD cameras
range. The cooled holder is essential to reduce contamination; you from the same manufacturer may differ significantly. Incomplete
hope that the post-column filter and its software were carefully deconvolution of the detector response function has a large effect
tested in diffraction mode before purchase. The largest angle that on the minimum value of χ2 obtainable, since ‘blooming’ in the
electrons can be passed through an in-column filter is an important detector smooths out the sharp minima at the HOLZ lines. For
parameter; it will limit recording of outer HOLZ lines. quantitative analysis of diffuse elastic scattering from defects, a
cold stage and image plates, together with an in-column elastic
How the size of the energy-filtering window varies with scatter- Omega filter are essential. Apart from the use of cold stages,
ing angle is also important; it depends on the aberrations of the two other methods have been proposed for minimizing thermal
filter. The use of a charge-coupled-device (CCD) area detector diffuse scattering:
using a single-crystal-YAG scintillator in combination with an
in-column imaging filter allows efficient parallel detection of the 1. The ‘Tanaka’ method of large-angle CBED has the effect of
entire filtered CBED pattern in a single exposure; this is probably eliminating most thermal scattering. A very small aperture is
the best arrangement (better than a GIF) for QCBED. placed in a plane conjugate to the electron source.
2. Off-axis electron holograms have the effect of elastic energy
The in-column filter also allows us to use an image plate de- filtering with an extremely small energy window, sufficient
tector (not possible with a post-column filter); think of the im- to exclude phonon scattering.
Chapter 5

Chapter Summary

This intent of this chapter is not to introduce new ideas so extract some of it. Most of all, the aim is to ask you to think
much as give you an idea of where you can go with what about how you can do the experiment in the TEM instead
you already know. It gives you an idea of how much infor- of the XRD system or a synchrotron, and perhaps get better
mation is hidden in ‘diffuse scattering’ and how you might data or data from a smaller volume of material.
Chapter 5

Licensed to ([email protected])
 Appendix 163

Appendix

Part I
People Q5.14 Why are the ‘trial values’ of the absorption parameters
‘trial values?
Q5.15 Why are crystals more difficult in CBED if they lack a
John Cowley F.R.S. was born in Adelaide on February 18, 1923 center of symmetry?
and died on May 18, 2004. With Alex Moodie he was responsi- Q5.16 What is charge hopping and what are strongly correlated
ble for the multislice method of image simulation; we now used materials?
it for simulation for nearly everything to do with electron dif- Q5.17 How ionic is MgO?
fraction in the TEM. He combined diffraction with imaging and Q5.18 Why is there a slight rotation of oxygen octahedra in
spectroscopy for REM, SREM, holography, and much more. some perovskites?
Q5.19 Why is a charge-density difference map not just called a
Walter Hoppe was not only the Director of the Max Planck In- charge-density map?
stitute at Martinsried from 1964 until 1985, he also invented Q5.20 What is Friedel’s law?
ptychography in the TEM. He lived from March 21, 1917 until
November 3, 1986.

Text-Specific Questions
Self-Assessment Questions
T5.1 Download Zuo’s many-beam code combined with a least-
squares optimization routine and apply to a problem of
Q5.1 Define the term ‘electron crystallography.’ your choosing.
Q5.2 What is the phase problem? T5.2 Show how you could use the two- or three-beam expres-
Q5.3 Why is collecting 3D microdiffraction data more difficult sions for the intensity of Bragg beams to understand in-
for TEM than XRD? tensity oscillations in a CBED disk.
Q5.4 What is the difference between quantitative TED and T5.3 Why is the word ‘Gaussian’ in the expression ‘Gaussian
QCBED? expansion table by Doyle and Turner’?
Q5.5 In your own words, summarize the differences between T5.4 What is DFT and is it suitable for MgO calculations?
TED and XRD. T5.5 Discuss the ways we can do strain mapping. How does
Q5.6 Discuss this statement: ‘Strongest continuous signal from CBED compare?
the smallest volume of matter in all of physics.’ T5.6 Look at other parts of the book and explain what a Ron-
Q5.7 If the probe is smaller we get the more information. Ex- chigram is.
plain why this is so, or not so. T5.7 Will we someday be able to enter a d-spacing sequence
Q5.8 What is the relation between ‘opening the iris’ and a rock- in Google? Discuss the implications.
ing curve? Explain your answer. T5.8 We say that ‘the structure of layer materials can often be
Q5.9 If the probe becomes smaller than the unit cell, how can interpreted from a single projection.’ Discuss this state-
we interpret a diffraction pattern? ment and when it might not be valid.
Q5.10 Are coherent and incoherent CBED patterns different? T5.9 We say ‘numerical phasing methods generally work
Explain. much better in 3D than in 2D.’ Discuss this statement.

Chapter 5
Q5.11 Why do we use the whole pattern to determine space T5.10 Look at the original paper by Vincent and Midgely (1994)
groups and what is this obtained from? on precession diffraction and explore the internet to see
Q5.12 What is the Rietveld method of X-ray and neutron dif- if it is being used today.
fraction? T5.11 Can we observe the Verwey transition in magnetite and,
Q5.13 Why is the background of inelastic phonon scattering if so, what do we see?
greater in the zone-axis orientation? T5.12 What is ptychography and is it ever used? If so, explain
when.
 164 5 Electron Crystallography, Charge-Density Mapping, and Nanodiffraction

structures. Acta Cryst B25:1156–1164 (The first useful obser-

References vations of atomic structure by electron microscopy)
Armigliato A, Balboni R, Frabboni S (2005) Improving spatial
Part I

resolution of convergent beam electron diffraction strain map-

ping in silicon microstructures. Appl Phys Lett 86 (063508-
General References 1-3. The ASAC software package)
Blackman M (1939) On the intensities of electron diffraction
Bird D, King Q (1990) Absorptive form factors for high-energy rings. Proc Roy Soc A173:68–82 (The beginning of the pre-
electron diffraction. Acta Cryst A46:202 cession technique)
Champness PE (2001) Electron diffraction in the transmission Doyle P, Turner P (1968) Relativistic Hartree-Fock X-ray and
electron microscope. BIOS Scientific/ Royal Micros Society, electron scattering factors. Acta Cryst A24:390–397 (For the
Oxford (A very useful introduction to electron diffraction) electron atomic scattering factors – with over 2,600 citations)
Cowley JM (1992) Techniques of transmission electron diffrac- Fienup J (1982) Phase retrieval algorithms – a comparison. Appl
tion. Oxford Univ. Press, New York. (Many useful chapters Optics 21:2758–2769 (Signal processing field algorithm –
on electron diffraction by our teacher. This early work met with over 1,800 citations)
with mixed success, and was often not accepted by the X-ray Humphreys CJ (1979) The scattering of fast electrons by crystals.
community) Rep Prog Phys 42:1825–1887 (How CBED software works)
Cowley JM (1995) ‘Introductory’ text-book on the physics of Humphry MJ, Kraus B, Hurst AC, Maiden AM, Rodenburg JM
diffraction. Diffraction Physics, 3rd edn. North-Holland, New (2012) Ptychographic electron microscopy using high-angle
York dark-field scattering for sub-nanometre resolution imaging.
Cowley JM (1953) Structure analysis of single crystals by elec- Nature Commun 3:730
tron diffraction. I. techniques. Acta Cryst 6:516 Nellist P, McCallum B, Rodenburg J (1995) Resolution beyond
Dorset D (1995) Covers the organic-crystal structure field, with the ‘information limit’ in transmission electron microscopy.
emphasis on non-biological organics such as waxes, lipids Nature 374:630–632 (Ptychography, as a method of su-
and polymers. Structural Electron Crystallography. Plenum per-resolution in STEM)
Press, New York O’Keeffe M, Spence JCH (1994) On the average coulomb poten-
Peng LM, Dudarev SL, Whelan MJ (2004) High Energy Electron tial (Φ 0 ) and constraints on the electron-density in crystals.
diffraction and Microscopy. Oxford University Press Acta Cryst A50:33–44 (On the surface dipole layers of a thin
Pennycook SJ, Nellist (eds) (2011) Scanning transmission elec- slab)
tron microscopy: imaging and analysis. Springer Science and Press W, Flannery B, Teukolsky S, Vetterling W (1986) Numeri-
Business Media cal Recipes. Cambridge University Press, New York (For the
Rodenburg JM (2008) Ptychography and related diffractive im- simplex algorithm)
aging methods. Advances in Imaging and Electron Physics Spence J, Zuo JM (1992) Electron Microdiffraction. Plenum,
150:87–184 New York (Gives a Fortran source-code to calculate CBED
Spence JCH (2013) High-resolution electron microscopy, intensity distributions Ig(Kt) with full documentation. Also
4th edn. Oxford University Press plot out the kinematic positions of HOLZ lines. Includes dis-
Weirich T, Labar J, Zou X (2004) Electron Crystallography. cussion of the ‘Tanaka’ method of large-angle CBED. This
NATO Science Series 211 book has been extensively revised and extended for publica-
Midgley PA, Eggeman AS (2015) Precession electron diffraction tion in 2016 as “Electron diffraction and Imaging”. Zuo JM,
– a topical review. IUCrJ 2(1):126–136 Spence JCH, Springer (2016).)
Zuo J (2004) Measurements of electron densities in solids: a Spence JCH, Cowley JM (1978) Lattice imaging in STEM.
real-space view of electronic structure and bonding in inor- Optik 50:129–142 (The probe spans many unit cells. A full
ganic crystals. Rep Prog Phys 67:2053–2103 (The basic idea analysis of the case where orders do overlap)
of CBED refinement reviewed with many applications – the Tsuda K, Tanaka M (1995) Refinement of crystal structure pa-
Chapter 5

best overall review of quantitative CBED) rameters using convergent-beam electron diffraction: the
The book on electron crystallography edited by Weirich et al. low-temperature phase of SrTiO 3. Acta Cryst A51:7–19
(2004) contains many useful case studies, and the books by (Rotation of Oxygen octahedra with temperature in SrTiO3
Cowley (1992), Dorset (1995), Peng et al. (2004), and Spence by QCBED)
and Zuo (1992) (see new 2016 edition) are essential reading Zuo J (1992) Automated lattice-parameter measurement from
for students interested primarily in electron diffraction rather HOLZ lines and their use for the measurement of oxy-
than imaging. gen-content in YBa 2 Cu 3 O 7-δ from nanometer-sized
region. Ultramicroscopy 41:211–223 (More on how CBED
software works)
Zuo J, O’Keeffe M, Rez P, Spence JCH (1997) Charge Density
Specific References of MgO: Implications of Precise New Measurements for The-
ory. Phys Rev Lett 78:4777–4780 (How ionic is MgO? Full
Allpress JG, Sanders JV, Wadsley AD (1969) Multiple phase dynamical refinement)
formation in the binary system Nb 2 O 5 -WO 3 . VI. Electron
microscopic observation and evaluation of non-periodic shear
Chapter 5
 5.8 References 165

Zuo J, Kim M, O’Keeffe M, Spence JCH (1999) Direct observa- Humphreys CJ, Maher DM, Fraser HL, Eaglesham DJ (1988)
tion of d-orbital holes and Cu-Cu bonding in Cu 2 O. Nature Convergent-beam imaging – a transmission electron-micros-
401:49–52 (The example of Cu2O) copy technique for investigating small localized distortions

Part I
Zuo J, Hoier R, Spence JCH (1989) 3-beam and many-beam in crystals. Philos Mag A58:787–798
theory in electron-diffraction and its use for structure-factor Kim M, Zuo JM, Park GS (2004) High-resolution strain mea-
phase determination in non-centrosymmetric crystal-struc- surement in shallow trench isolation structures using dynamic
tures. Acta Cryst A45:839–851 (For references to the work electron diffraction. Appl Phys Lett 84:2181–2183
of Kambe, Gjonnes, Moodie, etc) Kirk MA, Davidson R, Jenkins M, Twesten R (2005) Measure-
ment of diffuse electron scattering by single nanometre-sized
http://emaps.mrl.uiuc.edu/ – Go there for a free web-based pro- defects in gold. Philos Mag 85:497–507 (See Kirk et al.
gram for these simulations. Input data is entered on the web (2005))
and results returned as a pdf email attachment. The job runs Lin JA, Cowley JM (1986) Reconstruction from in-line elec-
on computers at the University of Illinois. Chapter 16 de- tron holograms by digital processing. Ultramicroscopy
scribes use of the JEMS program for similar purposes. 19(31):179–190 (Development of the Ronchigram as an
aberration figure to give values of the aberration constants.
Electron holograms using an STEM)
Background References Lin YP, Bird DM, Vincent R (1989) Errors and correction term
for holz line simulations. Ultramicroscopy 27:233–240
Allen LJ, Faulkner HML, Leeb H (2000) Inversion or dynami- (Complications when measuring strain)
cal electron diffraction data including absorption. Acta Cryst Liu Z, Ohsuna T, Terasaki O, Camblor M, Diaz-Cabanas M, Hi-
A56:119–126 (Ptychography giving a charge-density map of raga K (2001) The first zeolite with three-dimensional inter-
the sample crystal structure showing all the atom types and secting straight-channel system of 12-membered rings. J Am
their positions. (See also Spence 1998.)) Chem Soc 123:5370–5371 (Solving the structure of a zeolite)
Bader R (1990) Atoms in molecules: A Quantum Theory. Oxford Loretto M (1984) Electron beam analysis of materials. Chapman
University Press, New York (Partitioning problem and the and Hall, London
multipole expansion) Lynch DF, Moodie AF, O’Keefe MA (1975) N-beam lattice im-
Coppens P (1997) X-ray charge densities and chemical bonding. ages. V. The use of the charge-density approximation in the
Partitioning problem and the multipole expansion. Oxford interpretation of lattice images. Acta Cryst A31:300–307 (De-
Univ. Press, New York tailed calculations, showing the maximum thickness we can
Liu Y, Eades J, Mazumder J (1994) Solving the crystal-struc- use for oxides – see other papers in the series too)
ture of a submicrometer phase in TEM. Ultramicroscopy Matsumura S, Tomokiyo Y, Oki K (1989) Study of temperature
56:253–268 factors in cubic-crystals by high-voltage electron-diffraction.
Eades JA (1988) Microbeam Analysis 75. San Francisco Press, J Electron Microsc Techn 12(3):262–271
San Francisco Moodie AF, Etheridge J, Humphreys CJ (1996) The symmetry
Ecob RC (1986) Comments on the measurement of foil thick- of three-beam scattering equations: Inversion of three-beam
ness by convergent beam electron-diffraction. Scripta Metall diffraction patterns from centrosymmetric crystals. Acta Cryst
20(7):1001–1006 A52:596–605
Godden TM, Suman R, Humphry MJ, Rodenburg JM, Maiden Morniroli JP Large-angle convergent-beam electron diffraction
AM (2014) Ptychographic microscope for three-dimensional Application to crystal defects. (in english). Monograph of the
imaging. Opt Express 22(10):12513–12523 French Society of Microscopies. ISBN 2-901483-05-4. Gives
Gramm F, Baerlocher C, McCusker LB, Warrender SJ, Wright method of determining Burgers vectors for dislocations from
PA, Han B, Hong SB, Liu Z, Ohsuna T, Terasaki O (2006) CBED patterns.
Solving a complex zeolite by a combining powder diffraction Ohsuna T, Liu Z, Terasaki O (2002) Framework determination
and electron microscopy. Nature 444(7115):79–81 of a polytype of zeolite beta by using electron crystallogra-

Chapter 5
Hoier R, Kim M, Zuo J, Spence J, Shindo D (1995) Sample phy. J Phys Chem B106:5673–5678 (Practical details of the
thickness and beam divergence, using an image plate detector method of combining images and diffraction patterns to solve
and elastic imaging filter. Proc MSA. Springer, New York the sensitive zeolite structures)
Holmestad R, Zuo JM, Spence JCH, Hoier R, Horita Z (1995) Own CS, Marks LD, Sinkler W (2005) Electron precession:
Effect of Mn doping on charge-density in gamma-TiAl by A guide for implementation. Rev Sci Instr 76(3) (article #
quantitative convergent-beam electron-diffraction. Philos 033703. Practical details of one system for the precession
Mag 72:579–601 method)
Honjo G, Kodera S, Kitamura N (1964) Diffuse streak diffrac- Pike WT, Brown LM, Kubiak RAA, Newstead SM, Pow-
tion patterns from single crystals .i. General discussion + ell AR, Parker EHC, Whall TE (1991) The determination
aspects of electron diffraction diffuse streak patterns. J Phys of strain in Si-Ge superlattices by electron-diffraction in a
Chapter 5

Soc 19:351 (Japan. Plot outlines and sheets of this thermal scanning-transmission electron-microscope. J Cryst Growth
diffuse scattering) 111(1–4):925–930 (Strain measurements in Si/Si-Ge layers)
Hoppe W (1969) First proposed Ptychography for electron mi- Plamann T, Rodenburg J (1998) Electron ptychography. II.
croscopy. Acta Cryst A25:495 Theory of three-dimensional propagation effects. Acta Cryst
 166 5 Electron Crystallography, Charge-Density Mapping, and Nanodiffraction

A54:61–73 (The idea ptychography elaborated extensively, should read, in which atomic defects seen in the high-resolu-
a systematic method is given which involves recording these tion are related to the diffuse scattering)
patterns as a function of probe position) Zhang P, Istratov A, Weber E, Kisielowski C, He H, Nelson C,
Part I

Spence J (1978) Phase determination by STEM. SEM 1978. Ed. Spence JCH (2006) Direct strain measurement in a 65 nm
Johari, O. IITRI, Chicago node strained silicon transistor by convergent-beam elec-
Spence JCH, Taftø J (1983) ALCHEMI: a new technique for tron diffraction. Appl Phys Lett 89 (161907-1-3. Strains are
locating atoms in small crystals. J Microsc 130:147–154 mapped in FET with SiGe source and drain)
Spence JCH, Zuo JM, O’Keeffe M, Marthinsen K, Hoier R Zhu J, Cowley JM (1983) Microdiffraction from stacking faults
(1994) On the minimum number of beams needed to distin- and twin boundaries in F.c.c. crystals. J Appl Cryst 16:171–
guish enantiomorphs in x-ray and electron-diffraction. Acta 175 (Fault vector of planar faults in CuAu alloys are deter-
Cryst A50:647–650 mined by applying a kind of ‘g.b’ analysis in microdiffraction
Spence JCH (2006) Diffractive (lensless) imaging. Science of patterns)
Microscopy. In: Hawkes P, Spence JCH (eds) Aberration-free Zuo JM, McCartney MR, Spence JCH (1996) Performance
images using radiations for which no lenses exist. Springer, of imaging plates for electron recording. Ultramicroscopy
New York 66:35–47
Spence JCH (1998) Direct inversion of dynamical electron dif- Zuo JM, Spence JCH (1993) Coherent electron nanodif-
fraction patterns to structure factors. Acta Cryst A54:7–18 fraction from perfect and imperfect crystals. Philos Mag
Tanaka M, Terauchi M, Tsuda K (2003) Convergent-beam elec- A68(5):1055–1078
tron diffraction III. JEOL, Tokyo Zuo JM (1994) New method of Bravais lattice determination.
Terasaki O (2000) The structure of MCM-48 determined by elec- Ultramicroscopy 52:459–464
tron crystallography. J Electron Microsc 48:795–798 (see also Zuo J, Pacuad J, Hoier R, Spence J (2000) Diffuse electron scat-
Nature 408, 449 Applications of electron crystallography to tering measurements have been correlated with bulk resistiv-
block-copolymers and liquid crystals) ity measurements against temperature. Micron 31:527–532
Treacy M, Gibson J, Howie A (1985) On elastic relaxation and Zuo JM, Spence JCH, Hoier R (1989) Accurate structure-factor
long wavelength microstructures in spinodally decomposed phase determination by electron-diffraction in noncentrosym-
In x Ga 1-x As y P 1-y epitaxial layers. Philos Mag A51:389– metric crystals. Phys Rev Lett 62:547–550
417 (Finite-element modeling of the thin-film elastic relax- Tafto J, Gjonnes J (1985) The intersecting kikuchi line technique
ation) – critical voltage at any voltage. Ultramicroscopy 17:329–334
Vainshtein B (1964) Structure analysis by electron diffraction. Zuo JM, Spence JCH, Downs J, Mayer J (1993) Measurement
Pergamon, New York (The early days of electron diffraction of individual structure-factor phases with tenth-degree accu-
in Russia) racy: the 00.2 reflection in BeO studied by electron and X-ray
Vincent R, Midgley P (1994) Precession method of electron dif- diffraction. Acta Cryst A49:422–429
fraction. Ultramicroscopy 53(3):271–282 (Precession method Zuo JM, O’Keeffe M, Rez P, Spence JCH (1997) Charge density
of electron diffraction) of MgO: Implications of precise new measurements for the-
Weirich TE, Portillo J, Cox G, Hibst H, Nicolopoulos S (2006) ory. Phys Rev Lett 78:4777–4780
Ab initio determination of the framework structure of the Zuo JM (1993) Automated structure-factor refinement from
heavy-metal oxide CsxNb2.54W2.46O14 from 100 kV precession convergent-beam electron diffraction patterns. Acta Cryst
electron diffraction data. Ultramicroscopy 106(3):164–175 A49:429–435
(Electron crystallography applied to transition metal oxides Zuo JM (2000) Electron detection characteristics of a slow-scan
and to the mesoporous framework silicates; asymmetric unit CCD camera, imaging plates and film, and electron image
of a (Cs,Nb,W)5O14 oxide framework) restoration. Microsc Res Techniq 49:245–268
Woolfson M, Fan HF (1995) Physical and non-physical methods Zuo JM (1996) Electron detection characteristics of slow-scan
of solving crystal structures. Cambridge University Press (A CCD camera. Ultramicroscopy 66:21–33 (Comparison of
good introduction to the numerical methods used to solve the film, CCD, Image Plate detectors and in-column filter. See
Chapter 5

phase problem) also later papers by this author)

Wu J, Spence J (2003) Low-dose, low-temperature conver- Zuo JM, Vartanyants I, Gao M, Zhang R, Nagahara LA (2003)
gent-beam electron diffraction and multiwavelength anal- Atomic resolution imaging of a carbon nanotube from dif-
ysis of hydrocarbon films by electron diffraction. Microsc fraction intensities. Science 300:1419–1421 (Demonstrates
Microanal 9:428–441 (Blank Disks. Compare four different reconstruction of the image of a double-walled carbon nano-
kinematic approximations with experimental TED data from tube at atomic resolution (stretched across a hole) from an
the known structure tetracontane) electron microdiffraction pattern. True “Lensless imaging.”)
Yagi K, Cowley JM (1978) Electron microscopy study of order- Zvyagin B (1967) Electron diffraction analysis of clay minerals.
ing of potassium ions in cubic KSbO 3. Acta Cryst A34:625– Plenum, New York (More on the early days of electron dif-
634 (Gives a superb example of using electron crystallogra- fraction in Russia)
phy applied to the family of zeolite structures, which students
Chapter 5
DigitalMicrograph
6

Chapter 6
Chapter Preview

In this chapter we will showcase and discuss DigitalMicro- chapter free of charge which is obviously highly recom-
graph, a software package widely used for TEM data acqui- mended for the read.
sition and analysis. Although the software package is pro-
prietary, it is so widely used that it is regarded as a de facto Using DigitalMicrograph as a platform we will discuss
standard for TEM data processing. The program is fit for issues of scientific image understanding and processing
general data-processing purpose, but it is custom-tailored which sometimes are software specific, but more often gen-
to analytical microscopy data providing convenient tools erally apply to scientific data analysis with any software.
and techniques to analyze spectra, images, and spectrum We will then explore some common image transformations
images up to five-dimensional data. The software is highly and filters and discuss their impact on the validity of data.
extendable by both commercial and free plug-ins, and an Throughout this chapter we will repeatedly use the Digi-
in-built easy-to-learn scripting language gives you the tools talMicrograph Scripting Language as a complementary tool
to develop your own acquisition or analysis methods. You so that you can pick up the valuable scripting skill as we
can download and install the basic version described in this go along.

Chapter 6

© Springer International Publishing Switzerland 2016

C. B. Carter, D. B. Williams (Eds.), Transmission Electron Microscopy, DOI 10.1007/978-3-319-26651-0_6 167
 168 6 DigitalMicrograph

6.1 Introduction the GMS version 2.3.2 and install it on an English Windows 7
64-bit operating system. Because software development is such
a fast-paced business compared to book publishing, more recent
6.1.1 What Is DigitalMicrograph? versions may now be available and possibly differ in some as-
pects. In particular the visual style of the user interface heavily
depends on user choices and the operating system, so don’t get
DigitalMicrograph is an all-in-one application developed for distracted by this.
Chapter 6

analytical TEM. It is both acquisition software for all Gatan-in-

terfaced hardware at the microscope, and data visualization and The installation – after download and extracting the archive files
analysis tool. It was originally developed for digital image pro- – is a two-step process. First, the license has to be installed using
cessing and the analysis of both images and electron energy-loss the license installer application which comes with the down-
spectra acquired with Gatan cameras, but it has since developed loaded free-license archive. After that, the GMS installer is used
into a more general tool, providing means to deal with the multi- and guides you through the installation – you can simply accept
tude of currently available signals in a fully equipped TEM such all predefined settings.
as EELS, EDS, CL, STEM BF/DF/ HAADF, and more. Contrary
to most other commercial software in the field of microscopy,
DigitalMicrograph is not tied exclusively to a specific micro-
scope or detector vendor. Instead it interfaces to most commer- 6.1.3 A (Very) Quick Overview
cially available equipment and unifies acquisition and evaluation
methods across the whole range of microscope installations.
When you start DigitalMicrograph for the first time you might
Using the software you can extract, visualize, measure, and com- feel overwhelmed by the amount of tools, windows, and palettes
pare data from different detectors, often acquired simultaneously presented to you, but don’t get scared. You will quickly learn
in a single experiment. The software fully supports spectrum-im- to value them all – and to hide or simply ignore those things
aging techniques which acquire analytical signals such as EELS you don’t need. The work space of the application, as shown in
spectra or diffraction patterns as a function of spatial position, Fig. 6.1, generally consists of a tool area at the top, a central area
time, specimen tilt – or a combination thereof – by handling up in which data is presented within individual, named windows for
to five-dimensional data sets. Software versions since GMS 2.0 each data set, a tabbed text output area, and several individual
take full advantage of 64-bit technology such that large data sets tool palettes. These items can be freely arranged within the appli-
of hundredths of megabytes can be handled without restrictions. cation or ‘docked’ at the border. Palettes can be collapsed or ex-
panded with the little triangular icon, or closed completely with
DigitalMicrograph also offers many specialized evaluation rou- the little X icon – you can display them again from the Window
tines for the different signal types some of which have become menu. Favorite working layouts can be saved and restored with
general standard techniques in research. These routines are usu- a name. To learn more about individual items you can always
ally commercially available as extension plug-ins either from access the help documentation shown in Fig. 6.2 by pressing F1
Gatan Inc. or from third-party software developers. However, or via the Help menu command.
even the freely available basic version without these extra rou-
tines is already a powerful data analysis platform, especially for One important feature of DigitalMicrograph is that all data sets,
image analysis. In addition you can easily extend this function- while held in memory, are automatically assigned to a unique
ality using the in-built scripting language and implement your label consisting of one to three letters. This label is shown in the
own evaluation routines. window title of the data separated with a colon from the data’s
filename. Throughout the application – and also in script code
– the data can be referred to by using this label, but it is only
valid as long as the data is held in memory. If it is closed and
6.1.2 Installing DigitalMicrograph Offline reopened, a new label will be assigned. We will use the same
labeling throughout the chapter so that ‘image ABC’ will refer
to the data set labeled with ABC in the application.
DigitalMicrograph is the central application of the single
all-inclusive software package called Gatan Microscopy Suite DigitalMicrograph Images
(GMS) which is typically installed on computers controlling
Gatan hardware such as cameras, STEM detectors or scan
generators. Licenses for installations without hardware drivers The term ‘image’ is often used as a synonym for data of
and without special analysis plug-ins can be requested free of any dimensionality in DigitalMicrograph – and also in
charge via the Gatan website from which one can also down- this chapter. 9
Chapter 6

load the latest installer file via FTP (see Online References). In
this chapter we are using the 64-bit variant of the installer of
 6.1 Introduction 169

Fig. 6.1 The typical DigitalMicro-

graph application work space after
first install on a Windows 7 operating

Chapter 6
system

Fig. 6.2 The embedded help docu-

mentation is accessible by pressing
F1 at any time

A second important workflow concept in DigitalMicrograph is ‘top’ or ‘front-most’ image is generally the active one, and menu
the ‘stacking’ of windows. Whenever more than one data set is commands or tools often act on this image only.
open, the according windows have an ordering from ‘front’ to
‘back’ even if they are visible next to each other. One window is
always ‘at the front’ and all others are arranged behind. Selecting
Chapter 6

a window by clicking on it brings this window to the front like

selecting a card from a stack of cards and putting it on top. This
 170 6 DigitalMicrograph

6.2 Understanding Data

6.2.1 What is an Image?

Before we learn more about DigitalMicrograph we have to

Chapter 6

understand some important basics of digital image processing

and analysis of scientific data in general. You may have some
experience with graphic software like Paint, Photoshop, GIMP
or similar programs – if so then don’t confuse this with image
analysis! The purpose of graphic programs is to create visually
pleasing (or interesting) images as perceived by the human eye,
but image analysis is all about numbers and information derived
therefrom – the visual representation of data comes second at
best and most often an old saying comes true: “There is more
to it than meets the eye.” Always remember: your images are
Fig. 6.3 DigitalMicrograph defines the pixel origin in the corner. A pixel with
measurements and experimental data, even if they sometimes the indices (1;2) is therefore anchored with its top-left corner at the position
may also just look ‘pretty’. x = 1 and y = 2 as indicated by the black circle. The value at this point is used to
determine the color of the whole pixel area
In DigitalMicrograph – and in most other scientific image pro-
cessing software – an image is not more, not less, than an in-
dexed array of numeric values. Proper images are per definition
two-dimensional and thus represented by a double-indexed set
of numeric values Ii,j with i and j being the indices of individ-
ual pixels running from zero to the number of pixels the image
is wide, respectively high. ‘Images’ of different dimensionality
have the according number of indices. Spectra, for example, are
one-dimensional and therefore have a single index. You should
also note that we have not yet defined what the actual value I
should be like. We will return to that a bit later.

First, we want to note another difference between ‘images’ as

intuitively understood in every day’s physical context and ‘im-
ages’ in digital data processing: physical images are continuous
and have a measurable area. One may magnify an image and
split it into ever smaller subareas, but – down to the physical
limits – one always deals with areas. Not so in digital imaging:
arrays of numbers don’t have a physical area. Each number is a
dimensionless point!

To close this gap, one usually defines an area assigned to a single

value and then calls this area a pixel (which is a term coined after
‘picture element’). As with most definitions, there is space for
certain flexibility, and one has to specify how the single value is Fig. 6.4 Information can be lost when continuous data is represented by a dis-
assigned to the area. The two most common variants are to either crete number of (sample) points
assign a pixel area to the point-value in the center of the area,
or to assign it to the point-value in the corner of the area. Digi-
talMicrograph uses the corner definition, as shown in Fig. 6.3. of microscopy, because it can easily be confused with the im-
aging resolution of the microscope. It is therefore safer to use a
The discrete nature of digital images also introduces a new com- different term. You will sometimes come across ‘pixel resolution’
plication: sampling. Whenever continuous information is digi- in text books, but we will stick with ‘sampling’ in this chapter.
talized into discrete data, information can be lost, as illustrated in Higher sampling then means more points per length unit, more
Fig. 6.4. Mathematically, this effect is well-described in the so- information, and better but larger data.
Chapter 6

called sampling theorem. The number of data points per length

unit is often referred to as image resolution in the context of A final word on nomenclature: the term pixel is strictly speaking
digital imaging. However, this is very misleading in the context correct for a single value of two-dimensional data only, but you
 6.2 Understanding Data 171

will see it often more generally used for data of different dimen- below a peak in the histogram gives you the exact color those
sionality as well. Other common terms are voxel (from ‘volume pixels have in the Image Display.
element’) in three- or multidimensional data, and channel for

Chapter 6
one-dimensional data. Now notice the thin black line right through the histogram. This
line describes how the available colors of the vertical bar are
projected onto the value range depicted in the horizontal bar.
Simply pick a color on the vertical bar, imagine a horizontal
6.2.2 Image Display line to the right until you intersect with the diagonal, and project
vertically downwards onto the horizontal bar. In Fig. 6.5a this
45-degree line performs a 1:1 mapping of the colors onto the
Let us now return to the more natural aspect of images – how value range, but in Fig. 6.5b one of the lines (ignore the other
you can see them. We have already established that scientific for now) is much steeper, and you can see how the dark values
images are just a collection of numbers, which leaves the ques- are stretched all to the first peak and how the contrast in the im-
tion how we represent them to the human eye. The most com- age has changed accordingly. Indeed, the sliders of the Display
mon convention is to assign a gray-scale value to the numbers, Control tool window control the steepness – the Contrast – and
usually with lower values getting darker shades than higher the left-right offset – the Brightness – of the mapping line. Try
values (Fig. 6.5a). The problem here is that the human eye can it out. You can quickly reset the values to the default ‘central’
only differentiate between a limited number of gray shades, and value by clicking on the text just above a slider. The third slider,
it is often necessary to adjust the image to see certain details in Gamma, controls the second black line in the histogram. On top
stronger contrast. But, if numeric values are assigned to certain of the color projection we have just explored, we can stretch or
gray-levels, isn’t then changing this gray-level changing the compress the color range in a nonlinear manner.
numbers? For most graphic programs this is indeed the case and
the reason why you should not use them for image evaluation. The sliders in the Display Control tool are not the only way to
You really don’t want to change your measured experiment alter your Image Display. Very often one is only interested in a
data! DigitalMicrograph, on the other hand, keeps the way it subsection of the whole image, for example the bright or the dark
displays data – the Image Display – completely separate from half of Fig. 6.5. Instead of changing contrast and brightness set-
the way the data is stored – the image data. You can change the tings you can then decide to use the whole available color scale
Image Display however you like, for example by altering gray- over a specific range of pixel values, and set all values outside
level contrast or using colors, but the data – the stored numeric this range to the highest and lowest color representation. The
values – always stay the same. Because your data is always effect is very similar to adjusting contrast and brightness, but
‘safe’ you can freely play with different display settings while it is often more comfortable and gives you more flexibility for
investigating your data to make those things you are interested nonlinear contrasts. There are different ways to set the display
in most visible. limits, as they are called, but the quickest is to use the Histogram
tool. You can select a value range directly by clicking into the
It is useful to fully understand how DigitalMicrograph converts histogram and dragging the mouse with the mouse button still
numbers to pixel colors, and how the display can be altered to pressed and releasing it when you are satisfied with the selected
your needs. Looking at Fig. 6.5 you should first notice the little range. The histogram then shows only this range spread out over
Histogram tool window. It always shows the data of the currently the whole range as illustrated with the left (respectively right)
selected image. A histogram is a counting-diagram of all values peak of the example image in Fig. 6.5c and 6.5d. To reset the
in the image. The horizontal axis covers a range of values, and display limits to the default values simply click on the horizontal
the vertical axis counts how many pixels in the image are within bar below the histogram.
a certain value range. Seeing two peaks like in Fig. 6.5a means
that the majority of pixels in this specific image are of one of You should also be aware that Image Displays can be set to sur-
two gray-levels – the brighter left half or the darker right half vey the image and to automatically set the display limits. This
of the image. is typically done for live-images during acquisition but is a gen-
erally available feature for any Image Display. When the image
Next, notice the two gray gradient bars attached to the histogram. data – the stored values – change, new display limits are found
The vertical bar to the left shows the available color scale used to and set and the gray levels in the image are updated accordingly.
represent the data. Most often it is simply a continuous gradient
from black to white, but you can select different color schemes There are more options to optimize an image display in Digi-
like the ‘temperature’ scale in Fig. 6.5e. Tip: if you left-click talMicrograph: you can assign colors and color gradients, invert
and drag-select a range in the vertical bar, a color-picker dialog the contrast, display or hide scale labeling at image borders,
appears and creates a color gradient over the selected range. switch autosurveying the display limits on and off, choose the
way the optimum display limits are determined, and set non-
Chapter 6

The horizontal bar below the histogram shows the colors used to linear contrast modes like the Pseudo-Contours or Equalized
represent pixels of a certain value range and is directly linked to (Fig. 6.5f). All these and more options are accessible via the
the range of the histogram. In other words: the color represented Image Display dialog which you can open by right-clicking on
 172 6 DigitalMicrograph
Chapter 6
Chapter 6

Fig. 6.5 DigitalMicrograph strictly separates data from display. The same data is shown in six different display settings. a Standard linear gray-scale distribution.
b Adjusted brightness and contrast levels. c Restricting gray scale to lower values. d Restricting gray scale to higher values. e Using a false color scheme. f Using
false color and “equalized” contrast

Licensed to ([email protected])
 6.2 Understanding Data 173

the Image Display and selecting Image Display from the Graphic Files
appearing context menu.

Chapter 6
So far we have changed the appearance of images in Image Dis- Never store or share your scientific data images (only) in
plays, but remember that images are just number arrays. We can a graphic image file format or you will lose data. 9
– and now will – display them completely differently as well. By
right-clicking on the Image Display and selecting an item
from the Display Type context menu the image can be represented Of course graphic file formats are sometimes exactly what you
as 3D surface plot with pixel values being represented as height want. If you are preparing a publication or presentation you will
(Fig. 6.6a) or as plain array of numbers in a spreadsheet (Fig. 6.6b). need standard graphic formats and DigitalMicrograph is capa-
ble of storing in most common formats. However, to make it
We finish this section with advice which by now is hopefully absolutely clear that you are not storing your data, the according
obvious to you. command in the File menu is called Save Display As …

Graphic file formats like BMP, GIF, or JPG always store an im- Tip: for quick copy-and-paste transfer of images from DigitalMi-
age as it is displayed but not the underlying numeric data. You crograph into publishing software it is sometimes useful to first
are losing valuable information if you don’t keep the original create an ‘as-displayed’ copy of your image. You can do this
data files. Some file formats like TIFF are capable of storing with the Create Image From Display command in the
more than the pure display, but even they cannot always keep the Display menu. Copy-and-paste of the actual image can, on the
full data, as we are going to discuss in the next section. other hand, be used to transfer the numeric data into a text file or
a spreadsheet application.

6.2.3 Number Formats

We are now returning to the beginning of the section where we

defined images – or, more generally, defined any ‘data’ in Digi-
talMicrograph – as a regular and indexed array of values. What
we haven’t discussed so far is what these values are, and in what
format they are stored. This topic roots deeply in informatics and
general computing, and you will find complete and extensive
explanations in any standard programming textbook. Here we
are only covering the practical consequences.

Computers can store and handle numbers in different formats.

Because the physical memory to store a single number is limited,
so is the value that can be stored. There are two main ways of
storing numbers, one for integer numbers and the other for real
numbers – and DigitalMicrograph supports both.

Integer numbers are well-defined and can be ordered and indexed

– the index often being the value itself! As a result, computers
can handle them very well and calculate with them absolutely
precisely. The only limitation is the amount of memory used to
store a single number. With a digital memory of 1 byte – 8 bits
which can be either 0 or 1 – it is possible to create 256 differ-
ent combinations. A 1-byte integer number can therefore store
a range of 256 values, by definition either the values from 0 to
255 (1-byte unsigned integer), or from −128 to +127 (1-byte
signed integer). If you compute with two such numbers and the
result lies outside this range, then it depends on the program what
the result will be. Some programs will ‘warp’ the number range
Chapter 6

such that 253 + 9 = 6, but most – including DigitalMicrograph

– will clip the value at the range limit such that 255 + 5 = 255.
Fig. 6.6 DigitalMicrograph can display images also as surface plots (a) or Computers can increase the range of integer numbers by using
spreadsheets (b) more memory per number. Four-byte unsigned integer numbers,
 174 6 DigitalMicrograph

for example, use 32 bits and can therefore store values between 0
and (232 − 1) or 4,294,967,295. The largest handicap of integer
numbers is that one cannot perform all types of calculations with
them. Most programs – including DigitalMicrograph – will cal-
culate a simple division like 8/3 = 2 because any number gets
truncated to the integer value.
Chapter 6

Real numbers are more flexible but they might be inaccurate.

Computers are not capable of storing real numbers precisely –
but neither are you! Or have you ever succeeded in writing down
all decimals of π? Or even 1 divided by 3? To store real numbers
with a finite amount of memory, computers use a floating-point
representation of the number, and the numbers are therefore
often called ‘float’ numbers. Using the DigitalMicrograph ter-
minology we will continue calling them ‘real’ numbers. Real
numbers use a certain amount of memory to store a significant
number of digits, while the rest is used to store a scaling expo-
nent. The amount of memory to store the scaling factor limits the
available range (which is huge and generally big enough), but
the amount of memory to store the significant digits limits the
precision, which sometimes can become a problem. But unless
you are interested in very, very precise values, or intend to write
numeric algorithms – in which case you really should consult
an informatics textbook – it is fairly safe to say: real numbers
should be your general choice for any computation and also in
data analysis within DigitalMicrograph. There are far more ‘un-
expected’ results using integer number computation than using
real-number computation.

Let us go through a practical example. You can find out about

the data type of any image in DigitalMicrograph by looking at
the Image Status tool (Fig. 6.7). The window tells you for the
selected image its label (the letter combination identifying it
within the application), its type (the number format), and its
size (the number of pixels in each dimension). If the mouse cur-
sor is within the image window, you can also read the value of
the pixel under the cursor. You can see that the iron jump-ratio Fig. 6.7 The Image Status tool shows you type and size of the active image
and the pixel value under the mouse cursor. All values are shown as ‘raw’ or
image in Fig. 6.7b, which is a ratio of two EFTEM images as
‘calibrated’ values depending on the display mode which can be changed by the
described in Chap. 13, is of type “Real 4,” but that the STEM
little ruler icon in the Target tool
bright-field image in Fig. 6.7a is of type “Integer 4 Unsigned.”
Computed images in DigitalMicrograph are typically stored in
‘real’ format, whereas acquired, unprocessed images are often CTRL + ENTER. We use a similar script – with the according
stored in ‘integer’ format because many detectors are counting image letter – to normalize the jump-ratio image as well. Com-
devices. Let us now do some example computation with both paring the images before (Fig. 6.7) and after (Fig. 6.8) normal-
images. ization, we see not much difference for the jump-ratio image,
but the bright-field image looks considerably worse. It has lost
We want to normalize the image and store intensity as percent- a lot of detail and appears flat in contrast. What has happened?
age of the maximum intensity. Such computations are most All intensity values of the STEM image were originally be-
conveniently done using the easy scripting language of Dig- tween 2,000,000 and 2,200,000 and the Image Display showed
italMicrograph. You can open a script window – which is a the variations over the difference of 200,000 in a scale from
simple text editor – from the File menu using New Script … black to white in 256 shades of gray. After normalization, the
In the editor, we write the equation we want to calculate using values where all between 90 and 100 and the images display
an image’s label whenever we refer to the image. To scale an showed the variation over the difference of 10 in a scale from
image ‘B’ between 0 and 100 we simply divide all values by black to white again in 256 shades of gray. However, because
Chapter 6

the maximum value in the image and then multiply by 100 as the data is stored as integer, there are only 10 different values
shown in the script in Fig. 6.8. The script is executed either and consequently 10 different shades of gray. Even worse, most
using the Execute button in the script window, or by pressing pixels now have the identical value of exactly 98 and all finer
 6.2 Understanding Data 175

Fig. 6.8 Normalized images from

Fig. 6.7 using simple one-liner scripts.
The integer-valued image (a) loses

Chapter 6
contrast detail, but the real-valued
image (b) doesn’t

variations are irretrievably lost. For the jump-ratio image stored zero, the Image Display uses 256 gray shades to show variations
in real numbers there is no problem. The scale of the number between plus and minus 0.1e−7 so that even this spurious contrast
has changed, but their precision has not. is strongly visible in the display.

We have now seen a typical problem of integer values: loss of So are all images wrong then? No. The error in real-number
information. The same loss happens when image data is stored computations is most often negligible and not even visible, but
in a graphic file format, because only the displayed gray shades what you should take home from the example is the fact that you
– usually 256 – are stored. Try to avoid them, but if you abso- should not always expect analytical exact results from numerical
lutely need to transfer data between different programs and are computations and that these inaccuracies become particularly
restricted to such formats, always choose the one which gives visible for small numbers when the relative error – the deviation
you the largest range of values and scale your data into this range of the true value divided by the true value – is particularly large.
– using a similar script as we have just used – before saving. For practical image analysis in DigitalMicrograph this leads to
the following advice.
Next we create a similarly problematic case for real-value im-
ages as shown in Fig. 6.9. The script is mathematically easy. In Misleading Image Contrast
the first line we create a new image D as a copy of the image B.
Note that a script can use an image label which is not yet used
for an image to automatically create that image for you. In the If you see some strange contrast in an analyzed image,
following lines we multiply, add, divide, and subtract from our always check the actual numeric values at the pixels and
original image in a way which should not change the original don’t trust the visible contrast alone. 9
values in the end. Subtracting our original copy should therefore
give an image full of exact zero values – but that is not what
we get in Fig. 6.9! Instead there is some spurious contrast, and So far we have discussed integer and real values in DigitalMicro-
the sum of all pixel values is not zero either. This is an effect graph images and data, but there are two more data types: RGB
Chapter 6

of the limited precision to which real numbers can be stored. values and complex numbers. What both types have in common
Individual pixel values do not differ more than about 0.0000001 is the fact that they are composites which store more than one
from exactly zero, but because all image values are that close to value at each pixel position – and the Image Display is again
 176 6 DigitalMicrograph

Fig. 6.9 Real-valued images may

show spurious contrast due to round-
ing issues after computation
Chapter 6

a completely separated entity which presents these composite script line, combining three images with their value ranges scaled
values in a specific way. between 0 and 255 (Fig. 6.11).

RGB stands for red, green, and blue and the triplet of numbers – The second composite number type in DigitalMicrograph is the
each of which is stored as a 1-byte unsigned integer number – is complex value. Complex numbers are a powerful mathematical
interpreted as a single color following the additive color mixing concept which builds on the idea that real numbers represent all
standard of most graphic programs. An RGB triplet has values possible points along a one-dimensional line, the number line.
between 0 and 255 for each component, higher values repre- Complex numbers, as extension to that, represent all points in
senting a ‘brighter’ contribution of that prime color. In image a two-dimensional plane, the complex plane. They have many
analysis, in particular for analytical mapping like EFTEM, it is applications in physics – including electron microscopy – and
sometimes useful to represent the ‘combination’ of up to three they also are required for some involved image processing like
maps as an RGB color image to visualize relationships. Because that we are going to see in the section on image filtering. Com-
each color represents one element (or property) in the image, mon graphic software can only rarely handle complex-valued
mixed colors can be easily interpreted as proportional combina- images, but they play an extremely useful part in scientific image
tion of the elements (or properties). Figure 6.10 shows such an analysis software, which is why we describe them in a bit more
RGB color-mix image of three elemental distribution maps. Red detail here.
Chapter 6

and green represent the single elements, but the purple areas in-
dicate a 1:1 mixture of ‘red’ and ‘blue’. DigitalMicrograph offers In mathematics, a complex number c can be written in the form
the interactive ColorMix tool to combine any three images into c = a + i ∙ b, with a and b being the real and imaginary part of the
RGB images, but the same task can also be achieved in a simple number, respectively. a and b can then also be understood as
 6.2 Understanding Data 177

Fig. 6.10 RGB images combine three

quantitative gray scale EFTEM maps
in one image. Mix-colors represent a

Chapter 6
mixture of the according elements

the two coordinates of a point in the complex plane where the shade to one aspect of the complex number only, and the user
x-axis is the real-number line, and the y-axis is called the imag- can choose which type of display is wanted in the Image Display
inary axis (Fig. 6.12). In DigitalMicrograph, a complex number setting (Fig. 6.13). The available choices are: Real, Imaginary,
is simply a pair of two real numbers, which are then interpreted Modulus, Phase, and Log of Modulus. The first two choices dis-
as real part a and the imaginary part b of the complex number. A play either a or b like any other real-value image and ignore the
complex value image therefore contains two real values at each other. The next two choices consider the polar coordinate repre-
pixel, which brings up the question of how such a pixel should sentation of the complex number in the complex plane. Instead
be displayed. DigitalMicrograph solves this by assigning a gray of describing the point by the two coordinates, it is described by

Chapter 6

Fig. 6.12 Definition of the

complex number c as a com-
Fig. 6.11 Any three images scaled between 0 and 255 can be added to an RGB posite of a real part a and an
image with a simple script imaginary part b
 178 6 DigitalMicrograph
Chapter 6

Fig. 6.13 Different ways of displaying a complex images in DigitalMicrograph. In the Image Display one can choose to represent the real part, the imaginary part,
the modulus, or the phase of the complex number as gray value

the line from the origin to the point, which has a length and an composite types RGB and complex into any other type require
angle with respect to the x-axis. The Modulus display assigns some additional choice, which you can pick from, prompted with
the gray shade according to this length (ignoring the angle), the a dialog.
Phase display according to the angle (ignoring the length). The
fifth display setting, Log of Modulus, also assigns gray shades
according to the length, but uses logarithmic scaling. This set-
ting is particularly useful for Fourier-transformed images, as we 6.2.4 Image Calibration and Image Tags
will soon see, and it is therefore the default display of complex
images in DigitalMicrograph.
We have now a good understanding of ‘images’ as far as the main
Having discussed all the types of data DigitalMicrograph sup- data is concerned, but isn’t there more to microscope images than
ports, we take a quick look at changing those data types in ex- pure intensity values at a certain pixel? Of course there is. For
isting data. The Change Data Type command in the Edit proper analysis of your experimental data you usually need to
menu changes the data type of the front-most data to the type know at least the magnification of the image or rather the ‘size’
you select from a dialog. By now you should be aware that some of a single image pixel in physical units, otherwise you will not
conversions are safe while others may lead to information loss. be able to measure any distances in your sample.
For example, you can always store a 1-byte integer value as a
4-byte integer number or as a 4-byte real number, but the conver- It is important to realize that the physical size of a pixel is not a
Chapter 6

sion in the other direction can be problematic. DigitalMicrograph property of the pure data, which is only an indexed array of num-
will present a warning dialog in such a situation because data bers. Measuring distances in images from one point to the other
type changes can generally not be undone. Conversions from the as number of pixels is as useless as measuring the geographic
 6.2 Understanding Data 179

distance of two cities on a map in centimeters on the map – we

need a scale in both cases!

Chapter 6
The act of assigning a physical meaning to a pixel dimension is
called image calibration and there are many different ways to
do it, but in the end it always results in a certain scale expressed
as ‘units per pixel’. In some cases – for example in diffraction
patterns – it is also important to specify the center of the diffrac-
tion pattern in the image so that a general calibration defines
three things per dimension in the data: a unit, a scale, and an
origin. DigitalMicrograph stores and displays this calibration
information in the Image Display Info in a separate Calibration
section (Fig. 6.14a). Note that each data dimension has its own
calibration, even if images will generally have identical calibra-
tion in the x- and y-direction. There is also a calibration for the
actual data values. It is called intensity calibration because it
most often describes the image intensity. It is here that a certain
numeric value obtained from the detector can be calibrated to
gain physical meaning. If you know your beam current and de-
tector properties you can, for example, calibrate image intensity
so that you can read the number of detected electrons per pixel,
or you may want to calibrate any computed map in the physical
units of the mapped property. The important thing to remember
is: the separation of data and calibration is similar to the sepa-
ration of data and data display. It is very useful as it allows easy
recalibration, but it can also be a possible pitfall for the unwary.
It is important to remember that all image processing is done
with the stored (not calibrated) values.

Image Calibrations

DigitalMicrograph stores calibrations – both the dimen-

sional and the intensity calibration – separately from the Fig. 6.14 Additional image information like calibrations (a) and tags (b) are
data. Changing calibrations does not alter data values9 shown in the Image Display Info dialog of an image

An example: with beam-sensitive samples one is often inter- You can display and change the tags of any image in the Im-
ested in electron dosages. Summed intensities from image areas age Display Info of the image under the section Image and
can be directly interpreted as detected electrons in that area, the subsection Tags (Fig. 6.14b). More importantly, you can
but depending on the detector, a single detected electron may access, read, and set such tags using the scripting language so
not necessarily increase the pixel value by exactly one. It is that you can, for example, store important processing infor-
therefore necessary to do an intensity calibration and to find a mation with the result images, as shown in the example script
scale in ‘electrons per unit’. Writing a script to get the mean in Fig. 6.16.
intensity from a selected area in an image is extremely easy,
but it will compute the mean of the stored values and therefore Raw vs Calibrated View
not tell you the mean electron count directly. Instead, a script
has to explicitly take the calibration into account, as shown in
Fig. 6.15. The Target palette allows quick switching between raw
and calibrated view. Click on the little line/ruler icon left
Calibrations are important, but they are usually not the only thing of the image label to switch. A line indicates raw, while
one needs to know about an image acquired during an experi- the ruler icon indicates calibrated view. 9
ment. Data analysis often requires additional information like
microscope parameters or what other data treatment steps have
Chapter 6

already been done to the data in its current form. DigitalMicro- Many data analysis routines in DigitalMicrograph write and read
graph stores all such additional information as so-called tags in additional information in the image tags, and they will not work
a hierarchical structure called Tag Group. if certain tags are missing. EFTEM elemental mapping, for ex-
 180 6 DigitalMicrograph

Fig. 6.15 Scripts use raw values

when doing math, but calibration
values can be accessed with specific
commands
Chapter 6

Fig. 6.16 Image tags contain all

additional information of acquired
data in a hierarchical structure. This
information can be read and written
using very simple script commands.
Simple colons specify branches in the
structure

ample, requires the information stored during acquisition to be ROI Control

able to compute the elemental map.

Both image calibrations and image tags are automatically saved The Control palette shows detail information for the cur-
together with the data in the files, but only if you save in the DM rently selected ROI. A little icon on the left allows selec-
file format! Saving images in a graphic format – even if it can tion of a specific control point of the ROI which is then
store all the data – will ignore this supplementary information marked with a slightly bigger square. Position information
and may lead to ugly surprises. in the palette refers to this point, and it can be shifted in
single-pixel steps using the little arrow buttons. Rectan-
gle-ROIs additionally show the width and the height of the
rectangle. Line-ROIs additionally show the length of the
6.2.5 Some Simple Tools line and its angle with respect to the x-axis. The little arrow
buttons can resize and rotate the line around the selected
control point. Loop- and Curve-ROIs show both the width
Equipped with a full understanding of images in DigitalMicro- and height of the smallest containing rectangle, and the
graph it is now time to showcase some of the most frequently total length of the ROI. All information is given either in
used tools of the software which you will likely use in your raw or calibrated units, depending on the chosen view. 9
everyday data analysis.

We have already seen that the Image Status palette tells us image The switch between raw and calibrated view is also important
coordinates and values of the pixel under the mouse pointer. In when you measure distances or positions using the ROI (short for
Fig. 6.7 the shown values are the raw values – the pixel indices region-of-interest) markers. ROIs can be placed with the mouse
in x and y, and the stored number value at the position – but the using one of the five tool icons (Fig. 6.17): Rectangle, Line,
same window can show the calibrated values (for x, y, and in- Point, Loop, and Curve.
tensity) instead. To switch between raw and calibrated view, you
can check or uncheck the Display Calibration Units checkbox ROIs are by default drawn as red–white dashed lines. They can
in the Image Display Info (Fig. 6.14a), but there is also a more be selected with the standard selection tool (the arrow), and are
convenient shortcut in the Target palette (Fig. 6.17). Do you no- then highlighted by green squares.
Chapter 6

tice the little ruler icon right of the filled circle?

 6.2 Understanding Data 181

Cropping Images and Spectra

Chapter 6
Rectangular-ROIs are great for quick image crop-
ping. Draw the ROI to the areas you want to have,
press CTRL + C to copy the selected area, then press
CTRL + ALT + V to paste this area as a new image win-
dow. All calibrations and image tags are copied as well.
You can also drag an ROI from one image into another
with the mouse to create a copy of same size in the sec-
ond image, which is useful if you want to crop multiple
images to the same size. The same trick also works for
Range-ROIs in one-dimensional data. 9

Any image display can have multiple ROIs, but the dashed
drawing style indicates that a ROI is ‘volatileʻ. Placing any new
ROI automatically deletes all other volatile ROIs on the display.
To add an ROI without removing other volatile ROIs from the
Fig. 6.17 Different Region-of-Interest (ROI) tools can be set to measure image
display you have to press the SHIFT key while you are adding
features with the Control palette. Use the little ruler icon on the Target palette to
a new ROI. Also: if you press SHIFT while you are resizing
switch between calibrated and not calibrated view
a line-ROI the ROI is restricted to 45-degree angles, while a
rectangle-ROI is restricted to be square. Pressing the ALT while
resizing a rectangle-ROI ensures that the each pixel dimension
of the rectangle is a power of 2 (i.e. 1, 2, 4, 8, 16 pixels etc.). selection on the displayed profile is also reflected on the marker,
and double-clicking the marker opens a little dialog allowing
ROIs are a very powerful tool both as visual markers and to mea- numeric entry of profile position and integration width in pixels.
sure features. They can be set and adjusted with greater flexibility
using the scripting language, as shown in the example of Fig. 6.18. Measuring Layer Widths
This script adds a single ROI marker simultaneously to two differ-
ent image displays, which has the very nice effect that moving the
line in one display simultaneously moves the other line too. You Extract a line-profile across the layer with the profile tool.
can use scripts like this to compare the same area in both images. With the profile marker selected, expand profile averaging
width with the + key. Draw a range-ROI in the extracted
profile image to roughly select the layer width. Adjust the
Option Keys
extraction profile marker in the image to be exactly per-
pendicular to the interface using the selection indicators
DigitalMicrograph often uses the concept of simulta- as a visual guide. Adjust the range-ROI in the extracted
neously pressed keys (usually SHIFT, CTRL, ALT, or a profile – it should reach from one point of inflexion in graph
combination thereof) during an action to get modified be- intensity to the other – to measure the layer width. 9
havior. For example: if you close an image window with
the little X icon, you are usually asked to save the data if
it is unsaved. Closing it with ALT does not ask, closing it
with SHIFT closes all open image windows, and closing
it with ALT and SHIFT clears all images – without asking 6.2.6 Extracting Subsets of Data
to save, so be careful! 9

In everyday work with digital images you will regularly be faced

The next tool to be introduced is probably the most useful for with the task of extracting ‘some’ part of the data from a larger
image measurements: the profile extraction tool. The tool works data set. We have already seen two examples in the previous sec-
similar to the line-ROI tool, but in addition to the ROI marker, tion: using copy-and-paste together with a rectangle- or range-
data is extracted along the drawn line and presented as a sepa- ROI to crop out some data, and using the profile tool to extract
rate, one-dimensional line profile data (Fig. 6.19). The profile a one-dimensional profile from an image. These two are also
is linked to the marker so that moving or resizing the marker representatives of two different ways of extracting information:
Chapter 6

immediately updates the profile. The extracted line can also be taking a subset of the stored data values, and computing a subset
‘widened up’ perpendicular to the line direction using the +/– from the stored data values using interpolation and extrapolation
keys. Intensity is then averaged over this width, which is partic- (Fig. 6.20). We will come back to data interpolation in the next
ularly useful to measure interfaces or features in noisy data. Any section and concentrate on real subsets of data here.
 182 6 DigitalMicrograph

Fig. 6.18 ROIs can be changed

in appearance and behavior using
scripting. Command names are of-
ten long and one often needs nested
commands. To improve code readabil-
ity one can put the first parameter of
a command in front of the command
Chapter 6

name, separated by a dot. This turns

nested commands into easily readable
command-chains

Fig. 6.19 The profile extraction tool extracts and optionally averages data along a line. The averaging box allows easy alignment of the extracted data across an
interface, and interface width can be conveniently measured in the extracted and averaged line plot
Chapter 6

Licensed to ([email protected])
 6.3 Digital Image Processing 183

Because the extracted subset can have multiple dimensions as

well, this triplet (direction, length, step) is specified for each
of the dimensions of the subset. The command also needs to

Chapter 6
know the number of dimensions of the source data and the
wanted data subset. Figure 6.21 gives you some extraction ex-
amples.

Spectrum-Image Extraction

DigitalMicrograph offers a convenient extraction tool for

three- or multidimensional data, but it is included in the
‘spectrum-imaging’ extension, which is not part of the
free offline version. However, you can still extract data
with some very simple scripting. 9

6.3 Digital Image Processing

One of the main advantages of digital images over photographs

is the possibility to easily alter the information to highlight cer-
tain data aspects. In the simplest case this is done by altering the
display of the data only, as we have done in previous sections,
but one can also apply mathematical rules which alter the data
values themselves.

Save It!

Make it your personal rule to always back-up acquired

raw data separately from your processed data. You may
need to return to it later, and all data processing can de-
stroy information. 9

Nowadays even the ‘slow’ computers perform so fast that dig-

ital image processing has become an integral part of everyday
scientific image analysis. However, it is historically interesting
that not so long ago some of these image transformations were
carried out physically with real photographs and optical benches
(see Chap. 31 of W&C).

We Want to Know

Rigorous scientific work requires exact knowledge of

what was done to raw data to produce the results from
which conclusions can be drawn. You can ruin your sci-
Fig. 6.20 Extracting data with the profile extraction tool interpolates the values entific reputation fast and permanently by being careless
of the image. Only for horizontal or vertical profiles is the extracted data a per- in your publications. Always specify all your data treat-
fect match to the individual pixel values ment steps in enough detail to enable readers to repro-
duce the results from the raw data and keep the raw data
Chapter 6

With the versatile ‘SliceN’ command one can specify a starting available for requests. 9
point in the data (by the pixel indices) as well as a direction, a
number of steps, and a step-size for subsequently taken points.
 184 6 DigitalMicrograph

Fig. 6.21 A couple of script exam-

ples to extract data subsets using the
versatile SliceN command
Chapter 6
Chapter 6
 6.3 Digital Image Processing 185

Fig. 6.22 Scripts can easily mask an

image based on pixel value or position

Chapter 6
DigitalMicrograph offers a multitude of convenient routines of tools for masking images, but they are custom-tailored to dif-
from its Process menu and from the Image Filtering fraction patterns and Fourier transforms, both of which require
tool palette, but the field of image processing is so vast that you point-symmetric masks with respect to an origin. We will re-
should always be prepared to look outside the box as well. Keep turn to these tools a bit later and instead showcase two simple
an eye on actual literature, and if you discover algorithms which ‘masking scripts’ in Fig. 6.22. Masks are typically created either
suit your needs, you can create your own image transformations based on image intensity (‘thresholding’) or pixel coordinates,
easily with the scripting language. In all image processing you and they set a pixel’s value to a specified value (most often zero)
should stick to the advice … Save it! if a condition isn’t met.

Masking by Script

6.3.1 Image ‘Filters’

Use the script A = condition ? B : c to create an image A
with values of image B at pixels which fulfill the condi-
The term ‘image filter’ roots back to the time of physical image tion, and values of image C at pixels which don’t fulfill
processing, when specific image information was enhanced by the condition. Images A, B, C have to be of same size.
removing obscuring information with a filter or mask as part of The condition can use both the pixel value or the pixel
an optical bench. In digital image processing the meaning of position. For the latter, the intrinsic variables icol and
‘image filtering’ has been expanded to all processing steps which irow are evaluated as the pixel indices. 9
alter the image content to emphasize a specific aspect. The pro-
cess is in most cases irreversible, and image filtering is often only
a necessary prerequisite for further mathematical image analysis. A second group of image filters are ‘local’ filters which alter
pixel values depending on the values of their neighbor pixels. A
The Best Filter simple example is the directional derivation filter, where a pixel
value is replaced by the difference of the value and the value
of the neighbor to the left or bottom (Fig. 6.23). The filter sup-
The exact result of a filter on an arbitrary image is often presses gradients and emphasizes edges in one direction.
hard to predict and can depend on noise level and ab-
solute values of the image. Therefore there is often no Local filters can often be described by convolution kernels, as
‘generally best’ filter for any specific task and you should discussed in Sect. 31.5 of W&C. The kernel describes the local
play with different filters and filter parameters to optimize response of a pixel’s neighborhood and is applied to each pixel of
your results. 9 the image, as illustrated in Fig. 6.24. DigitalMicrograph offers a
convenient way of defining and using kernels via the Process-
ing/Spatial Filter menu. The most common kernels
The simplest example of an image filter is masking – setting used for smoothing (noise suppression) and edge detection are
specific pixel values to zero. DigitalMicrograph provides a set already predefined (Fig. 6.25).
Chapter 6
 186 6 DigitalMicrograph
Chapter 6

Fig. 6.23 The derivation filter suppresses directional gradients and emphasizes edges

Fig. 6.24 Image convolution with

a kernel: at each pixel the ‘kernel’
is overlaid and multiplied. The pixel
value is replaced by the mean-value
of the multiplied kernel. All pixels are
evaluated simultaneously
Chapter 6

Fig. 6.25 The Laplace filter is a convolution filter to emphasize sharp transitions in contrast
 6.3 Digital Image Processing 187

Chapter 6
Fig. 6.26 Script example of a custom-defined filter detecting local maxima

Local filters may also compute a ‘new’ pixel value by analyzing a Spatial Frequencies
neighborhood and reporting the result. DigitalMicrograph offers
some such filters in the Process/Non-linear Filter
menu. The Range Filter for example replaces a pixel value Spatial frequency – or often only ‘frequency’ – in images
by the value range of all pixels of the local neighborhood and denotes the reciprocal value of some pixel-periodicity. If
therefore visualizes the local ‘dynamic range’ of an image. a function has periodic behavior and produces the same
intensities every 10 pixels, the function has a spatial fre-
You can also define your own image filters by scripting. Fig- quency of 1/10 pixel−1. 9
ure 6.26 gives the script example of a local maximum filter. Only
pixels with value higher than their neighborhood remain, while
others are set to zero. The Fourier transform of an image describes the image as a sum
of weighted sine and cosine functions of form
Image filters can also be based on global image properties. Nor-
malization of image values to the minimum and maximum value a cos.!t/ + b sin.!t/ (6.1)
in the image would be a trivial example. A very popular and
extremely important class of ‘global’ image filters is the Fourier Here the mathematics can be handled more conveniently if com-
or frequency filter, which we will discuss after the next section. plex numbers and Euler’s formula are used instead of the sine–

6.3.2 Fourier Transformation in Images

Fourier transformations (FTs) are often used when data needs

to be represented as frequency distribution rather than intensity
distribution. To illustrate this, consider the schematic forest in
Fig. 6.27. One could describe it by describing the trees at each
position: ‘At (0;0) there is an ash tree, at (1;0) there is an ash
tree, at (2;0) there is a fir tree, etc.’ Alternatively, one could
also describe it with the frequency the firs are appearing: ‘It is
a forest of ash trees, but the third tree in every direction is a fir.’
In a similar manner, one can describe an image as the image
intensity at each pixel, or as the sum of many weighted periodic
functions of same shape but different frequency and direction. A
Fourier transform of an image does exactly that.

Fourier Transformation

Fourier transformations (FTs) are a powerful and central

mathematical concept used in many places. If you are not
Chapter 6

familiar with the basics of FTs it is possibly a good idea

to read up on it. 9 Fig. 6.27 How to best describe this forest? You can either specify all 81 trees
as ‘spatial information’ or just the periodic variations as ‘frequency information’
 188 6 DigitalMicrograph
Chapter 6

Fig. 6.28 Images with a sine-distribution of intensity (top row) and their Fourier transform images. A single spot-pair in the FT represents an according sine-dis-
tribution. Spots further at the border represent higher frequencies. Any image can be understood as sum of individual sine-distributions as shown in the last image
which is the sum of the first four (both in real space and in the Fourier transform)

cosine functions of the equation above. The function weight is

then given by a single complex-valued coefficient c It is therefore not surprising that the FT of a micrograph image
from a sample location has a very strong connection to the dif-
a cos.!t/ + i b sin.!t/ = c exp.i!t/ (6.2) fraction pattern of the same sample location. It can be used for
diffraction-related analysis such as indexing crystal orientations.
It turns out that an image of N pixels can be represented by N
such weighted functions which can again be nicely arranged as Diffractogram
regular grid, as shown in Fig. 6.28. The Fourier transformation
of an image is represented as an image of same size with the
intensities at a pixel representing the complex weighting c of a The Fourier transformation of a high-resolution micro-
specific function of form given in Eq. 6.2. The center of this FT graph is often called diffractogram because it resembles
image represents the zero-frequency – the constant value of the a recorded diffraction pattern from the same sample lo-
image – and towards the borders the spatial frequency of associ- cation. 9
ated functions increases.

Point Symmetry Historically, diffraction patterns were mainly recorded on pho-

tographic films which had a logarithmic response to electron
dose. To have a visual ‘like-for-like’ representation of digitally
It is a mathematical property of Fourier transforms that computed diffractograms, they are best displayed on a logarith-
any transformation of a purely real-valued image yields mic scale as well, and ‘log-of-modulus’ is therefore the default
a point-symmetric FT image. 9 display mode for complex images in DigitalMicrograph.

Diffractogram vs Diffraction Pattern

To compute the FT of any image in DigitalMicrograph you can
either select Process/FFT from the menu or simply press
CTRL + F. The result of the transformation is a complex image, Do not confuse diffractograms with recorded diffraction
and what you are seeing is a gray-scale representation of the patterns. They are similar and share some properties, but
modulus of the complex values on a logarithmic scale. This is the they are not identical, and grave errors can be made by
default representation of complex images in DigitalMicrograph neglecting the subtle differences. 9
and it will soon become clear why.
Chapter 6

Fourier transformation mathematics can also be used to describe As described in the main course book, TEM imaging and diffrac-
the imaging process of round lenses between image plane and tion can be understood in terms of electron wave functions. An
focal plane. In fact, what you get in the focal plane of a lens is initial wave function generated by the electron source interacts
the Fourier transform of the object you have in the image plane. with a sample and is modulated to become the exit wave func-
 6.3 Digital Image Processing 189

tion, which is then imaged using the objective lens. Images and and are a valuable tool while sitting at the microscope. One can
diffraction patterns are recorded in the image and the focal plane use the menu command Process/Live/FFT to compute
of the objective lens, respectively. Wave functions are described the FT of an image which automatically updates whenever the

Chapter 6
by complex numbers, but recording them on film or digital de- data in the source image changes, for example to observe a live
tector can only capture the modulus. The phase information is camera view.
lost. As a result, a diffraction pattern is the (real-valued) modulus
of the FT of the (complex-valued) exit wave function, whereas
a diffractogram is the (complex valued) FT of the (real-valued)
modulus of the exit wave function. 6.3.3 Fourier Filtering
Two important implications of this difference are that it is not
possible to create a sample image from a recorded diffraction pat- One main application of FTs is image filtering in the frequency
tern alone, and that spots which do not appear in diffraction pat- domain, in particular to reduce image noise. ‘Image noise’,
terns because of destructive interference (‘forbidden reflexes’) the random fluctuation in intensity values, can have specific
can be seen in diffractograms computed from high-resolution statistical properties depending on the detected signal and the
TEM images. In addition, there can also be computational arti- detection methods, but in a simplified model it represents ‘ran-
facts in FTs, as we will discuss next. domness’ of values in all pixels and, consequently, all spatial
frequencies. Image contrast, on the other hand, is generally a
The mathematics of Fourier transformations requires the data systematic change of intensity values over a larger area and
to be of infinite size. The algorithm to compute FTs therefore thus represented by specific frequencies predominantly in the
assumes that images are periodically continued outside their de- lower frequency range. Therefore, one can reduce noise levels
fined size, which can lead to artifacts. Consider a sine function in an image by taking the FT of the image, attenuating all fre-
with a periodicity of 20 pixels. The FT of this sine has exactly quencies which do not contain signal, and then taking the in-
one nonzero value at the spatial frequency of 0.05 pixel−1. Fig- verse FT. For example: a low-pass filter removes contributions
ure 6.29 plots this sine function in two images of different size. of higher spatial frequencies – the intensity variations from one
The first is 100 pixels wide, the second only 95 pixels. While pixel to the next – which are predominantly noise. However,
both images plot the identical sine function, their respective FTs sharp, step-like features like interfaces get affected as well, and
look very different. You can understand this by periodically con- one has to find a compromise between noise reduction and im-
tinuing the images. While the first will give you a perfect sine, age blurring.
the second will give you a step every 95 pixels. It is this step
which is represented by the additional frequencies in the FT. DigitalMicrograph has a convenient tool for image filtering. Us-
The same effects can be seen in FTs of two-dimensional im- ing the Image Filtering palette one can interactively adjust
ages whenever the intensities at the left and right – respectively the filtering parameters while observing the results. For a better
top and bottom – borders do not match. These computational understanding it is useful to additionally compute the FT of the
artifacts can be suppressed by smoothly attenuating intensities result using the Process/Live/FFT command (Fig. 6.31).
to a constant value towards the border (Fig. 6.30). The menu
command Process/Diffractogram does exactly that. In In atomic resolution images the signal is usually concentrated
addition it shifts image values by the mean intensity, which re- in very specific frequencies – the diffractogram spots. It is pos-
duces the intensity of the central pixel in the FT and optimizes sible to emphasize this signal by filtering out all other signals.
the image for direct comparison with diffraction patterns. First compute the FT, then use the masking tools to select all
areas to be included or excluded, then use the menu command
FTs at the Microscope Process/Apply Mask …, and finally inverse transform the
image. Figure 6.32a shows you the example of such filtering,
and the image contrast has clearly improved. But look closely:
In TEM, the FTs of the image of amorphous material at there are now atom columns visible in the border where the
high magnification can be used to observe objective lens original image didn’t show any! Surprised? You shouldn’t be.
astigmatism and focus. A slight defocus will produce rings If you filter out all signals but those of a specific periodicity, it
in the FT which become more elliptical with astigmatism. is only natural that the image will show you this periodicity. To
In bright-field STEM, the same technique can be used strengthen this point we do the following experiment: starting
to visualize condenser lens astigmatism and focus. Fur- out with an image of pure random noise we take the FT and then
thermore, scan distortions and instabilities also produce apply exactly the same masks as for Fig. 6.32a. The inverse
distinct features in FTs which can be used to detect and transformed result is shown in Fig. 6.32b. As you can see, we
quantify them. 9 have actually created an atomic resolution image out of pure
noise. Unfortunately, such artifacts have also been published as
Chapter 6

real results in the past – both knowingly and unknowingly. Do

We finish this section by mentioning that FTs can also be a great not make the same mistake – it will likely ruin your reputation
tool to visualize and analyze artifacts of the imaging process forever.
 190 6 DigitalMicrograph

Fig. 6.29 Fourier transforms assume

periodic continuation. If the period-
icity of a signal does not match the
image size, the ‘step’ is accounted for
by (many) additional frequencies. a
Periodic signal. b FT of signal. c Illus-
tration of periodic continuation
Chapter 6
Chapter 6
 6.3 Digital Image Processing 191

Fig. 6.30 Atomic resolution STEM image (a) and its Fourier transform (c). The
cross-like artifacts in the FT are artifacts caused by the (incorrect) periodic con-
tinuation of the structure. Simple Butterworth filter (b) suppresses this artifact

Chapter 6
in the FT (d)

Fig. 6.31 Fourier low-pass filtering of images. Reducing the cut-off radius reduces noise but blurs sharp interfaces. DigitalMicrograph offers an interactive Image
Chapter 6

Filtering tool to adjust the filter parameter while observing the result
 192 6 DigitalMicrograph
Chapter 6

Fig. 6.32 Fourier filtering of an atomic structure (a) and of pure noise (b) using the same frequency mask: left is the image, center is the FT with the mask, right
is the back-transformed image with only the masked intensities. While the atomic resolution image looks ‘cleaner’ it also shows ‘atom columns’ in the amorphous
region. The ‘random’ image also shows ‘atomic structure’ stressing the danger of introducing artifacts by filtering

You Get What You Filter For 6.3.4 Coordinate Transformations

Take extreme caution when using Fourier filtering on A standard image represents intensity as a function of spatial
high resolution images as it introduces artifacts which coordinates X and Y in a Cartesian coordinate system. However,
can easily be confused with the signal one is looking sometimes it is beneficial to represent the same information in a
for. Be adamant to never draw conclusions from filtered different coordinate system to either visualize a different aspect
images alone. If the signal is not visible in the raw image, or to simplify further data treatments.
it is really not there. If you use filtered images in publica-
tions, make sure you specifically state this together with Take diffraction ring patterns as an example. Without additional
the filtering parameters. 9 plugins there is currently no convenient tool in DigitalMicro-
graph to extract average radial intensity. However, if you repre-
sent an image in polar coordinates, i.e. as intensity as a function
of radius and angle with respect of a defined center point, you
Chapter 6

can use the default line profile tool to do exactly that (Fig. 6.33).
Some other examples of useful coordinate transformations are

Licensed to ([email protected])
 6.4 Scripting and Plugins 193

Chapter 6
Fig. 6.33 Coordinate transformation by script. Presenting the diffraction pattern rings in a polar-coordinate transform allows the use of the profile extraction tool
to get radial averages

simple image rotation or scaling, subpixel translations or the 6.4 Scripting and Plugins
Hough transformation used in Kikuchi-band analysis.

Using the scripting language you can define any kind of image One of the most versatile features of DigitalMicrograph is its
coordinate transformation using the Warp command. The com- scripting language. Throughout the previous sections we have
mand takes three input parameters: the source image, a transfor- used it to perform image manipulations and computations to give
mation equation for X, and a transformation equation for Y. For a quick taste of how easy it can be used as a simple tool. In most
each point of the destination image the two equations calculate cases there is no need to export data out of DigitalMicrograph
the according coordinate in the source image. The value at this for evaluation, and you can run scripts directly while acquiring
point is then stored in the destination. If the coordinate lies in the data at the microscope. You can even have scripts analyze a
between pixels, the four neighboring pixels are used for bilinear live-stream of data to guide you to the more interesting sections
interpolation of the value. of your sample or help you tune the equipment.

Interpolation Is Not Reversible You don’t need much programming skill to use simple scripting,
but the language is surprisingly sophisticated on a closer look.
It supports structured object-oriented coding, and it is perfectly
All coordinate transformations except shifts by integer possible to create complete applications with it. People have
numbers of pixels require data to be interpolated. In the used it to create their own tools within DigitalMicrograph, in-
general case it therefore not possible to fully reverse the cluding custom-made floating palettes and dialogs. The scripts
transformation without loss. 9 can be installed as regular menu commands or tool icons, and
they can be saved as separate plugin files which can be easily
Chapter 6

shared, often even across different program versions of Digi-

talMicrograph. Even better, scripting is an integral part of all
 194 6 DigitalMicrograph

installations, including the free offline-license version, which mercial, some free of charge – which offer complete solutions
makes DigitalMicrograph a solid platform for free open source for many problems. You can get plugins for special techniques
or open share data analysis tools similar to ImageJ, but with the like holography or focus series reconstruction, and more gen-
benefit of being a de-facto standard application at most electron eral plugins like for principal component analysis or geometric
microscope installations. phase analysis. Authors of scientific papers also often share their
plugins on request or as scripts via the scripting database. You
It is also possible to access microscope hardware through script- will find a short overview of some resource sites at the end of
Chapter 6

ing commands and create customized acquisition routines and this chapter, but it is generally a good idea to spend at least a bit
experiments. As DigitalMicrograph is a central application of time looking for already made solutions on the web before
which uses hardware interfaces to different hardware vendors, jumping into large coding projects.
such scripts can run cross-platform on different microscope in-
stallations and with different third-party detectors without any Finally, it should also be mentioned that there are two alterna-
need of adaptation. Scripting is also often the simplest way to tive options to extend the functionality of DigitalMicrograph.
combine various signal sources in a simple application. The first is to request a software development kit (SDK) from
Gatan which is basically the code framework of DigitalMicro-
Over the years the more enthusiastic script-writing microscopists graph in the programming language C++. Using the SDK you
have formed a loose community, sharing their experience over can create your own C++ application and link it into DigitalMi-
the internet and helping newcomers with their problems. There crograph. In particular for computational expensive algorithms
are various resources on the internet, most notably a database of this can give substantially better performance than a scripted
example scripts and a dedicated [dm-script]section on the Q&A solution. If the external application is completely independent
site StackOverflow. Searching for ‘DigitalMicrograph scripting’ of DigitalMicrograph – for example a data evaluation algorithm
will immediately give you some good starting points. The pri- which processes a data set but doesn’t require any user interface
mary source of information on scripting, however, is the online – the second option might be even easier: the scripting language
help documentation of DigitalMicrograph itself, which is acces- contains a command which allows calling external applications
sible via the Help/Search … menu command. and waiting for them to return. One can therefore create an
independent application which can be called as stand-alone
While it perfectly possible and usually a good learning process executable and write a script which first stores necessary data
to create your own tools with the scripting language of Digi- to the disk, then calls the application with some parameters,
talMicrograph, you should not forget that you not always have and finally reads in the processed results once the application
to reinvent the wheel. There are various sources – some com- has finished.

Chapter Summary

We have given you a first overview of the widespread data ing which is by no means complete. We have sensitized
acquisition and analysis application DigitalMicrograph, you to look out for possibilities and potential problems of
and you are now capable of viewing and analyzing micro- scientific image processing. You now have a basic under-
scopic images with its standard tools. You have a good un- standing of the most common image filters and know how
derstanding of what a 'digital image' is and how it should to compute, understand and utilize Fourier transformations
be treated in a scientific way. We have explained the sepa- of images.
ration between ‘data’ and ‘data display’ as a main concept
of DigitalMicrograph and shown how the display can be Spread throughout the whole chapter you have seen short
optimized for specific information. You also learnt the dif- examples of the powerful scripting language which is part
ferent numeric formats in which data might be stored and of DigitalMicrograph, and you can write similar tool-scripts
the implications for data evaluation and export. to handle simple image manipulations and evaluations. You
are aware that such scripts can be packaged as ‘plugins’ and
In the second part of the chapter we have given a very basic that various plugins for specific microscopic techniques and
introduction into the vast field of digital image process- evaluation schemes do already exist.
Chapter 6
 Appendix 195

Appendix

Part I
People T6.3 What are the advantages of keeping image ‘display’ set-
tings separate from image ‘data’?
T6.4 Which different settings or tools can you use to visualize
Peter Roland Swann (see page 75) founded the company Gatan faint intensity differences in an image?
with his brother Rex in 1964. T6.5 Explain how you can find the exposure time of an ac-
quired micrograph in a stored DigitalMicrograph image.
T6.6 Using Fig. 6.5, explain the main features of the Histo-
gram tool and how it displays the mapping of values to
Self-Assessment Questions colors in the image.
T6.7 Assume you have the image of a perfect layer structure of
exactly known layer thickness. Describe how you would
Q6.1 How would you go about calibrating image intensity in use this information and existing tools in DigitalMicro-
camera images in electrons? graph to calibrate the image.
Q6.2 Why are ‘unprocessed’ TEM camera images of type in- T6.8 In the Fe jump-ratio image of Fig. 6.7, how would you
teger but ‘gain normalized’ images of type real? extract and compare intensity values from the matrix
Q6.3 Most EFTEM data analysis requires multiple images to across the smaller particle at the bottom of the image?
be spatially registered with each other. Which image fil- T6.9 Give three examples for digital image filters and their
ters are typically applied for automated registration tech- application.
niques and why? T6.10 Give three examples for the use of Fourier transforma-
Q6.4 What happens to the (back-transformed) image if in its tions in microscopy-related image processing.
Fourier transform all complex values get their phase set T6.11 Explain why the “masking tools” in DigitalMicrograph
to zero (but the modulus remains)? What if all the mod- are all point-symmetric.
ulus values get set to zero (but the phase stays)? T6.12 Using Fig. 6.10 discuss the merits and pitfalls of present-
Q6.5 What would be needed to be able to reconstruct a ing chemical information in color-mix RGB images.
high-resolution image from a recorded diffraction pat-
tern?
Q6.6 How can you calibrate a high-resolution TEM image of
Si in [110] zone axis using Fourier transformations and References
standard DigitalMicrograph tools?
Q6.7 What is the highest spatial frequency a Fourier transform
image contains?
Q6.8 Cross-correlations compute the relative shift of two im- General References
ages using Fourier transformations. Why is it often ben-
eficial to filter one of the images with a filter to reduce Some references to the general principles of image processing
image contrast at the borders to a constant value? Why are given in W&C. Although the present chapter is intended
is it contra-productive to apply the filter to both images? to be self-contained, you may find it interesting to see how
Q6.9. How can Fourier transforms of STEM images be used large the topic really is.
to diangose scan instabilities caused by instabilities in
electric circuits or main power lines?
Specific References
The following publications highlight the use of DigitalMicro-
Text-Specific Questions graph scripting in technique development for both data ac-
quisition and evaluation.
Mitchell DRG, Schaffer B (2005) Scripting-customised mi-
T6.1 What is the main difference between graphic software croscopy tools for Digital Micrograph™. Ultramicroscopy
and scientific image-processing software? 103(4):319–332
T6.2 Explain all the information which might be lost if an im- Kimoto K, Kothleitner G, Grogger W, Matsui Y, Hofer F (2005)
Chapter 6

age recorded at the microscope (with DigitalMicrograph Advantages of a monochromator for bandgap measurements
and a camera) is stored as JPEG image. using electron energy-loss spectroscopy. Micron 36(2):185–
189
 196 6 DigitalMicrograph

Lozano-Perez S, Titchmarsh JM (2007) EFTEM assistant: A tool Online References

to understand the limitations of EFTEM. Ultramicroscopy
107(4):313–321 http://www.gatan.com/ Company page of Gatan Inc., the source
Mitchell DRG (2008) DiffTools: Electron diffraction soft- of DigitalMicrograph. You can request the free license from
ware tools for DigitalMicrograph™. Microsc Res Techniq this page and download the latest version.
71(8):588–593 http://ftp.gatan.com/ Official public FTP server of Gatan Inc.
Mitchell DRG (2008) Circular Hough transform diffraction Software updates and installers can be downloaded from here.
Chapter 6

analysis: A software tool for automated measurement of se- http://portal.tugraz.at/portal/page/portal/felmi/DM-Script Ex-

lected area electron diffraction patterns within Digital Micro- haustive database of DigitalMicrograph scripts. Highly rec-
graph™. Ultramicroscopy 108(4):367–374 ommended.
Shi J, Williams DR, Stewart PL (2008) A Script-Assisted Mi- http://stackoverflow.com/questions/tagged/dm-script Commu-
croscopy (SAM) package to improve data acquisition rates on nity driven Q&A for programmers containing a dedicated
FEI Tecnai electron microscopes equipped with Gatan CCD DigitalMicrograph Scripting tag. Highly recommended.
cameras. J Struct Biol 164(1):166–169 http://lists.asu.edu/ Email list server of Arizona State University
Sader K, Schaffer B, Vaughan G, Brown BRA, Bleloch A (2010) hosting a DigitalMicrograph Scripting User Group called
Smart Acquisition EELS. Ultramicroscopy 110(8):998–1003 DMSUG.
Braidy N, Le Bouar Y, Lazar S, Ricolleau C (2012) Correcting http://dmscript.tavernmaker.de/ Private page hosting plugins and
scanning instabilities from images of periodic structures. Ul- general scripting tutorials in an e-book.
tramicroscopy 118:67–76 http://www.dmscripting.com/ Private page hosting example
Kim KH, Zuo JM (2013) Symmetry quantification and mapping scripts, tutorials and general information on scripting.
using convergent beam electron diffraction. Ultramicroscopy http://www.hremresearch.com/ Company page selling plugin
124:71–76 solutions for DigitalMicrograph. Also contains free plugins.
Lucas G, Burdet P, Cantoni M, Hébert C (2013) Multivariate sta- http://temdm.com/ Private page hosting free plugins for Digi-
tistical analysis as a tool for the segmentation of 3Dspectral talMicrograph.
data. Micron 52(53):49–56
Chapter 6
 
Electron Waves, Interference,
7

Chapter 7
and Coherence

Chapter Preview

Chapter 7
The mysterious electron waves are indispensable for tar- To try to keep the purpose in sight we’ll summarize each
get-oriented operation of a TEM and hence for under- part of the discussion (in the boxed summaries) as we go
standing the achieved results: we simply cannot make any through the chapter and then summarize the chapter at
meaningful TEM investigation without considering electron the end.
waves. Therefore we have compiled in this chapter an over-
view of the basics of waves and the peculiarities of electron You’ve read the word in W&C (e.g., Chap. 5, also in
waves. This chapter is meant as an orientation guide for the Chaps. 28–30) and you’ll read about the relevance of coher-
domain of waves, and is not intended to be at all compre- ence in Chaps. 8, 9, and 11. Different authors have slightly
hensive, but it will help you when finding and understanding different ways of expressing the concept so we encourage
the literature on electrons. you to read and compare – and think! The concept is essen-
tial if you want to understand interference, which is itself
The nature of this topic means that there is a lot of math, the critical concept in holography; we’ll cover that topic in
and that the math is unavoidable, but we’ll try to point out Chap. 8. If you have time to play with the math, concentrate
the physical interpretation of the equations as we proceed. on the Kirchhoff diffraction integral.

© Springer International Publishing Switzerland 2016

C. B. Carter, D. B. Williams (Eds.), Transmission Electron Microscopy, DOI 10.1007/978-3-319-26651-0_7 197
 198 7 Electron Waves, Interference, and Coherence

7.1 Introduction This equation links together space r and time t. As a second-or-
der differential equation, it has a 2D space of solutions: all linear
combinations of solutions, such as created by interference, are
Chapter 7

Electrons were the first detected elementary particles. They ex- again a possible solution.  is the phase velocity of the wave.
hibit typical particle properties such as mass and momentum trans-
fer. However, electrons are not classical particles behaving like an As you may easily show by calculating the respective deriva-
apple that falls from a tree: macroscopic bodies are considered as tives, a solution for the wave function is given as
behaving fully deterministically according to the laws of classic
 .r; t/ = a.r/ exp.iı.r; t//
physics. If all relevant parameters such as starting point coordi-
nates, momentum, and forces are perfectly known, then the space– = a.r/ exp.i2kr + .r/ − !t//
time behavior of the apple can be predicted to whatever accuracy
we want. In contrast, electrons are quantum particles governed by with an amplitude a.r/ and a phase ı.r; t/ given by the phase
quantum mechanics laws and are described as so-called electron shift .r/, time-frequency ! and time t, as well as by kr; the
waves. For quantum particles, the spatial coordinates and mo- wave vector k with modulus k = jkj = 1= points into the local
menta cannot be perfectly known simultaneously (the ‘Heisenberg propagation direction along gradr .ı.r; t//.
Uncertainty Principle’), and hence the space–time behavior cannot
be predicted accurately; instead, only probabilities for finding an The phase describes the spatial structure of the wave in the fol-
Chapter 7

electron in a certain volume can be given determined by wave lowing sense: the surface that connects all the points of equal
properties. Depending on the experiment, we think of electrons phases for a fixed time defines a wave front, for the example of
either as particles or as waves. For example, in the photoelectric a plane wave or a spherical wave shown later in Fig. 7.1. In any
effect, they convincingly show up as particles. Mostly in TEM, case, a similar wave front appears at a corresponding distance
but always in HRTEM and holography, electrons, equally con- (‘wavelength λ’), if the phase is increased by 2π. The phase ve-
vincingly, appear as waves, and hence both their amplitude and locity mentioned above is given by  = !=.2k/.
phase have to be considered for understanding the experimental
findings: we can only understand TEM by using wave optics. In general, the wave vector k also varies with r. Therefore, as
it propagates, the wave front usually changes its appearance.
Electrons Are Waves Propagation here does not mean a motion in time but the struc-
ture development in space with r at a given instant. It represents
a wave field distributed in r and oscillating in time t. After time
Every HRTEM experiment is a direct proof that electrons T = 2=!, we find the same wave structure in space. The phys-
are waves. 9 ical meaning of the wave function  .r; t/ is that, as they prop-
agate and oscillate in time, waves transport energy. The energy
flow is given by the so-called Poynting vector, which is oriented
Consider electrons emitted from the electron source. Each elec- along the wave vector k; the modulus of the Poynting vector is
tron is a wave. Each electron wave propagates down the micro- proportional to j j2 =   , which is called the intensity of the
scope column according to the laws of wave optics: it is focused wave; the complex conjugate   results by inverting the signs of
by the lenses, it is deflected by the deflectors, it interacts with the all imaginary units i. In electron microscopes the energy flow is
object hence forms the object exit wave; finally, by interference equivalent to the current density, i.e., the intensity.
of the diffracted partial waves, it builds up the image wave and
hence the intensity image found on the detector. For a better For each space–time point .r; t/ the wave may be visualized as a
understanding of the wave behavior, first of all, we’ll summarize point in the complex plane (Fig. 7.2). Other representations are
the general properties of waves. also possible, e.g., by means of cosine and sine functions; how-
ever, apart from very simple cases, these are much more difficult
to handle mathematically than the exponential function. Because
of the Euler-relation exp.i˛/ = cos.˛/ + i sin.˛/, which is a lin-
7.2 Description of Waves ear combination of cos and sin functions, all these representa-
tions are equivalent.

You’re familiar with how we use waves and wave functions in Usually, the wave in the point r0 at time t0 is given in polar coordi-
TEM from W&C; now we are going to become a little more nates by an arrow of length a at the angle ı counted counter-clock-
mathematical and more rigorous. We’ll also change the notation wise, which may represent the expression 2kr + '.r/ − !t. With
slightly, but you’ll be able to follow without problems. A wave increasing time, the arrow moves clockwise on the dotted circle
 .r; t/ is a phenomenon periodic both in space, r; and time, t. It as shown in Fig. 7.1, but with increasing r counter-clockwise.
obeys the so-called wave equation (Helmholtz equation) Moving at a fixed time in the stationary wave field from point r0
to r1 shifts the arrow on some trajectory to the new position given
1 @2  .r; t/ by amplitude and phase in r1. Here, r1 is assumed more advanced
r 2  .r; t/ = (7.1)
 2 @t2 in propagation direction and hence ı increases.
 7.2 Description of Waves 199

Property of Waves

Chapter 7
A fundamental property of waves is that they are ex-
tended; hence they cannot be localized in sharp spot-like
particles. This is why we have a problem conceptualizing
electrons as particles and as waves. Waves tend to spread
everywhere and fill any available space, whereas particles
are confined in a sharp point. 9

In Fig. 7.1a, a plane wave is inclined at an angle ; the wave

front stays plane everywhere. Figure 7.1b shows a hemispherical

Chapter 7
Fig. 7.1 Simple forms of propagating waves. a Plane wave inclined at an angle
wave; the radius of the sphere increases with distance of wave
γ. b Hemispherical wave. c Modulated wave front front to center. At large radial distances, it may locally be approx-
imated as a plane wave. Finally, in Fig. 7.1c, the modulated wave
front changes rapidly due to locally varying phase gradients and
Fig. 7.2 Representing a wave in
curvatures. The wave front changing under propagation can only
the complex plane
be determined by means of the Kirchhoff diffraction integral,
which we’ll discuss later.

7.2.2 Spherical Wave

Here, the wave emerging from a point forms centered spherical

shells of increasing radius r. Because of isotropy, we can sim-
ply use the modulus r = jrj and the wave vector given by the
In the following examples, the waves are considered stationary in respective radius vector of modulus k. The wave is described by
space in that they do not change their spatial structure with time; exp.i2kr/=r. Because of energy conservation, it must be nor-
after a time period 2=! the wave pattern looks the same. Then malized by the factor 1=r damping the amplitude with increasing
the time-dependence !t merely describes the periodic oscillation distance, and hence the intensity by 1=r2. The curvature of a
of the spatially fixed wave pattern; for many considerations, the spherical wave decreases with the distance; hence it approaches
time-dependence needs not be considered. Then the wave equa- a plane wave for very large radii.
tion results as the Helmholtz equation only in r
r 2  .r/ = −.2k/2  .r/ (7.2)
7.2.3 Modulated Wave
7.2.1 Plane Wave
In general, a wave looks much more complicate than either of the
above cases. It is modulated, e.g., by an object which gives rise
Plane waves have a wave vector k that is independent of r. A to phase shifts; i.e., we have local retardation or advancement,
plane wave is described in the x–z plane by the phase factor hence bending and changing the direction of the wave vector and
exp.i2k.x sin  + z cos  // where the angle between propaga- the wave front. As it propagates, the wave may rapidly change its
tion direction and z-axis is . In a strict sense, a wave can only shape, which may only be determined by means of the Kirchhoff
be a plane wave if it is infinitely extended with a constant am- diffraction integral giving the more general solution of the wave
plitude a.r/; the reason is that a finite wave front always shows equation.
diffraction phenomena at the rims, i.e., a finite wave front bends
at its rim. Since, experimentally, everything is finite (!), waves
are always confined. So, they will always be slightly curved, in
particular around the edges of the confinement. Furthermore, the
assumption of infinite extension gives theoretical problems with
the normalization in that the energy content in such a wave would
also be infinite! Nevertheless, a plane wave is a very interesting
model wave for understanding the principles of wave optics.
 200 7 Electron Waves, Interference, and Coherence

Summary of Waves oidal interference pattern with spatial frequency qC  kˇ. If the
waves would be curved, the angle of superposition and hence
also the period of the fringes would change locally. The fringe
Chapter 7

– Waves are periodic both in space and time. contrast is defined by C = IImax
max Imin

– Waves are described by a wave function  given by

+Imin

the quantities amplitude and phase, following the wave Using the Euler-relation exp.i#/ = cos.#/ + i sin.#/, we find the
equation. intensity’
– Waves cannot be localized; instead they will fill any
available space: the wave excited by a stone falling in I.x/ =    = a21 + a22 + 2a1 a2 cos.2qC x + '/ (7.4)
a pond will shortly fill the whole pond.
– Waves transport energy. The energy flow per unit area with ' = '1 − '2 . Remarkably, I.x/ does not depend on time be-
and unit time is proportional to the intensity. cause the two waves oscillate with the same frequency, and hence
– The intensity is found by    = j j2. 9 time dependency drops out. In space, I.x/ oscillates between the
intensities Imax = .a1 + a2 /2 and Imin = .a1 − a2 /2 in the x-direc-
tion with the so-called spatial frequency qC = 2kx = 2k sin.ˇ=2/
, hence qC  kˇ for small angles ˇ. Similarly to the frequency
of time-periodic events, the spatial frequency qC is the reciprocal
7.3 Interference
Chapter 7

spatial period (‘reciprocal fringe spacing’) of the cosinusoidal

pattern.

Interference of two plane waves is the simplest hence elementary With the intensities I1;2 = a21;2 of the two waves, we can describe
process in wave optics. It allows, e.g., building up more compli- the interference pattern as an equation:
cate waves and understanding diffraction. Mathematically speak-
ing, summing up many plane waves is Fourier synthesis; accord- I.x/ =    = .I1 + I2 /.1 + C cos.2qC x + '// (7.5)
ingly, decomposing a given wave into a number of plane waves p
is Fourier analysis. Therefore, having understood the properties C = 2I1 +II12I2 means the contrast of the cosinusoidal interference
of plane waves helps understanding more complicate ones. fringes. The fringe contrast 0  C = IImax −Imin
max +Imin
 1, it only equals 1,
if a1 = a2 = a. In this case of optimum contrast at equally tuned
Assume, for simplicity, that we have two plane waves with con- amplitudes, the cosinusoidal interference pattern reads
stant amplitudes a1;2 and phases '1;2,
I.x/ = 2a2 .1 + cos.2qC x + '// (7.6)
1 = a1 exp.i'1 / exp.+i2kx x/ exp.+i2kz z/ expŒ−i!t
(7.3)
2 = a2 exp.i'2 / exp.+i2kx x/ exp.+i2kz z/ expŒ−i!t The position of the fringe with respect to x = 0 is determined
by the phase difference, ', between the respectively superim-
tilted about the y-axis by the respective angle ˙ (Fig. 7.3). Then posed wave points. Consequently, the interference pattern allows
the wave vectors read as k1;2 = .˙kx ; 0; kz / with kx = k sin  and measuring phase differences, which makes interferometry and
kz = cos  they superimpose at the angle ˇ = 2. Superposition holography so attractive, because phases are lost in the usual
yields the wave intensity pattern    = j j2 of a single wave.

Ψ = Ψ1 + Ψ2 Remarkably, the waves, after leaving again the area of interfer-

ence without being detected, behave as if no interference had
This  is again a solution of the wave equation; it produces an occurred, i.e., as if they had propagated through free space quite
intensity pattern given by I.x/ =   . alone according to the wave equation.

Two plane waves inclined by an angle ˙ superimpose at some

distance downstream in Fig. 7.3. They form a regular cosinus­

Fig. 7.3 Interference of plane waves

with '1 = '2
 7.5 Propagation of Waves 201

Summary of Interference

Chapter 7
– Waves propagating at the same time through the same
area interfere, i.e., they form a sum-wave with locally
enhanced or decreased amplitude and with a position
(phase) given by the local phase difference of both in-
terfering waves.
– For example, two plane waves form a regular cosinusoi-
dal pattern with the spatial frequency qC = 2k sin.ˇ=2/  kˇ.
This is a very basic relation for interfering waves with
the same wave number k superimposed at an angle ˇ.
– Interference allows measuring the phase difference be-

Chapter 7
tween the superimposed waves. This is the basis for all
kinds of interferometry and holography.
– The waves leave the interference area completely un-
Fig. 7.4 Modulation of a wave propagating through an object
disturbed, i.e., they appear subsequently, as if they
would not have experienced any interference at all. 9
agation in free space. This is described by the Kirchhoff diffrac-
tion integral.

An object may modulate both the amplitude and phase of the in-
coming wave 0. The amplitude component of the object changes
7.4 Modulation of a Wave the wave amplitude, e.g., by absorption or dynamical diffraction
effects such as interference within the object; the phase compo-
by an Object nent of the object shifts the phase of the wave, hence bends and
shifts the wave front. At the object exit plane we find the object
Now, let us consider a more specific situation. Assume we have wave obj(x,y).
an illuminating wave 0 .r/, not necessarily a plane wave, propa-
gating through an object, and leaving the object from the exit face Summary of Modulation of a Wave by an Object
described by the 2D-coordinate robj = .x; y/ (Fig. 7.4). By interac-
tion with the object it may be modulated in amplitude (‘amplitude
object’) and phase (‘phase object’) such that the ‘object exit wave’ – A wave transmitted through an object may be modu-
lated in amplitude and phase.
obj.robj / = a.robj / exp.i'.robj //  0 .robj / (7.7) – The modulation is given by the object transmission
function a.robj / exp.i'.robj //.
results. You will note that the object exit wave is, formally and – The relation between the object transmission function
simply, just given by the illuminating wave 0 .r/ in the object and the underlying object structure needs a thorough un-
exit face without the object, multiplied with the ‘object trans- derstanding of interaction of the waves with the object.
mission function’, a.robj / exp.i'.robj // provided by the object. In – Without understanding the interaction, no findings can
general, how the amplitude and phase of the wave are changed be interpreted. 9
by the object depends on robj; the phase '.robj / indicates the de-
viations of the wave front from the wave found when there is
no object. For an illuminating plane wave 0 .robj /  1 assumed
in the following, the object exit wave obj.robj / agrees with the
object transmission function; therefore, these two terms are 7.5 Propagation of Waves
often used synonymously. In any case, severe difficulties may
arise for understanding how the object transmission function
a.robj / exp.i'.robj // results from a given object structure. This The Kirchhoff diffraction integral in the most general form may
depends on the interaction processes occurring in the object, be found in math textbooks, where it’s best kept. After some
which may be very complicated. small rearrangements it may be written as

Propagating away from the object in the near-field the wave still
 ˇ ˇ
exp i2k ˇrP − robj ˇ 2
Z
looks similar to the object exit wave; however, with increasing  .rP / = C1 obj.robj / ˇ ˇ d robj (7.8)
ˇrP − robj ˇ
distance it will develop a very complicate appearance, because object
both amplitude and phase will change substantially under prop-
 202 7 Electron Waves, Interference, and Coherence

Fig. 7.5 Propagation of a wave Fig. 7.6 Geometry for calculating

Fresnel approximation in 1D
Chapter 7

It describes the elementary Huygens waves released from the be considered constant in the exponential function, because k is
object points robj weighted with the object transmission function a large number and therefore even small variations in x give rise
obj.robj / = a.robj / exp.i'.robj //, superimposing in the receiving to large phase variations. With all the constants collected in the
point P at rP, as shown in Fig. 7.5; C1 is a complex constant. prefactor C2, what remains is
Chapter 7

In general, the integral is hard to solve analytically. There are,

however, two very interesting approximations relevant in near- .x − /2
Z  
field (Fresnel approximation) or in far-field (Fraunhofer approx-  ./ = C2 obj.x/ exp i2k dx (7.10)
2z
imation). object

We can calculate the dotted wave front at some distance from the This equation says that the wave at the ‘receiving point’ P is
object exit wave using the Kirchhoff diffraction integral. given by the Fresnel-integral of the object exit wave. Often,
the exponential function is called ‘Fresnel propagator.’ Con-
The Kirchhoff diffraction integral is the mathematical formu- sequently, the propagated wave  ./ results from the object
lation of Huygens principle: every point in the object wave wave simply by convolution with the Fresnel propagator. In
releases a spherical wave weighted with the local wave value microscopy, we can see the effect of the Fresnel-integral by
a.robj / exp.i'.robj //. looking at the Fresnel fringes that appear at edges under slight
defocus. In a usual intensity image, Fresnel fringes only present
These spherical waves interfere creating a wave value, e.g., the squared amplitude modulation of the wave, whereas the
 .rP / = a.rP / exp.i'.rP // in the receiving point P. The Kirch- corresponding phase modulation of the wave is only visible
hoff diffraction integral is the mathematical description of this in a hologram.
process.
Unfortunately, we can’t solve the Fresnel-integral analytically
All the Huygens waves released from the object are summed up even for very simple object structures such as an opaque edge.
at the receiving point P. This is because of the quadratic exponent, i.e., the curvature of
the Fresnel propagator, which describes the primordially spher-
ical Huygens waves in parabolic approximation.

7.5.1 Fresnel Approximation in the Near-Field

of the Object
7.5.2 Fraunhofer Approximation
in the Far-Field of the Object
For simplicity, consider a 1D object with coordinate x as illus-
trated in Fig. 7.6; the generalization to 2D is straightforward.
At an axial distance z assume a parallel line containing the Fortunately, at large distances, the solution of the Kirchhoff
considered receiving point P in rP with the x-coordinate . Then, diffraction integral is much more interesting and also easier to
Pythagoras yields find. This is because at a very large distance from the object
s (where z ! 1) the spherical Huygens waves become almost
2
ˇrP − robj ˇ W= r = z 1 + .x − /  z + .x − /
2 plane waves, at least locally. Then in the Fresnel propagator, for
(7.9)
ˇ ˇ
z2 2z comparably small objects, the quadratic term x2 =.2z/ may be
neglected;  2 =.2z/ may be shifted into the complex prefactor
C3, and the Fraunhofer approximation follows:
This approximation is inserted into the Kirchhoff diffraction in-
tegral. The denominator r may be considered nearly constant for
Z

small variations of x, meaning small objects; however, it cannot

 .#/ = C3 obj.x/ exp.i2kx#/dx (7.11)
object

Licensed to ([email protected])
 7.6 Imaging: Formation of the Image Wave 203

This equation is essentially an expansion of the object exit

wave in plane waves, each propagating into a respective angle
# = =z. Besides the prefactor C3, it is the Fourier transform of

Chapter 7
the object transmission function; the Fourier transform describes
far-field propagation – and this is the diffracted wave forming;
e.g., it could be a Laue diffraction pattern of a perfect crystal.
There, z is called the diffraction length.

In usual diffraction experiments only the intensity of the dif-

fracted wave is detectable, whereas its phase is not accessible.
This is the famous phase problem of diffraction. The phase loss
means a considerable loss of information about the object struc-
ture, since the phases contain the positions in real space as ‘geo-

Chapter 7
metrical phases;’ furthermore, they answer in the ‘Zernike phase’
the question whether the object is an amplitude or phase object, or
a mixture of both. Consequently, without the phase information Fig. 7.7 Abbe theory of microscopic imaging
of the diffracted waves, the object cannot be determined uniquely
from a diffraction pattern; you would need a-priori knowledge
or further assumptions about the object structure. Nevertheless,
diffraction is very worthwhile: it allows analyzing atomic details 7.6.1 Fourier Transform of the Object Exit Wave
such as lattice spacings, unit cells, and many more; however,
this information does not present atomic resolution, because it is
averaged about the diffracting object area of many nanometers. The objective lens focuses the above far-field wave, virtually
propagated to infinity, into the back focal plane; this is because,
Summary of the Propagation of Waves by definition, the back focal plane is the image plane of infinity
hence ideally fulfills the far-field condition of Fraunhofer diffrac-
tion. Consequently, in the back focal plane we find the Fourier
– The object exit wave propagates in space. transform of the object exit wave (‘Fourier spectrum’)
– During propagation it changes its appearance in ampli-
tude and phase.
Z
spec.q/ = obj.x/ exp.−i2qx/dx
– This is described as the Kirchhoff diffraction integral, (7.12)
which, in general, is difficult to solve. object

– There are two very important approximations: Fresnel Each point in the back focal plane represents a plane wave dif-
approximation in the near-field, Fraunhofer approxi- fracted at the object into a corresponding angle ϑ, related to the
mation in the far-field. spatial frequency q = kϑ commonly used as coordinate in Fourier
– The Fraunhofer approximation is essential. It is given space.
by the Fourier transform of the object exit wave. It
allows us to understand diffraction and, as shown in The Fourier transform is perfect also for the phases in the spec-
the following, image formation in a microscope. 9 trum, if the object exit face is placed in the front focal plane.
Since for microscopy the objective lens is meant to yield a high
magnification, the object exit face is in fact positioned very
closely in front of the front focal plane, and hence we can assume
that in the back focal plane we indeed find the perfect Fourier
7.6 Imaging: Formation transform of the object exit wave.
of the Image Wave But, unless we are interested in a diffraction analysis of the ob-
ject structure, we should not record the Fourier transform, be-
Because of the phase problem, the object wave cannot be deter- cause then the phase problem would come up.
mined from a diffraction pattern, i.e., from the intensity distribu-
tion in Fourier space. Instead, we need a focusing device such as
a lens, which, utilizing these phases, brings back the propagated
wave to real space. According to Abbe’s theory of the micro- 7.6.2 Building the Image Wave by Inverse
scope, lens imaging works in two steps, as shown in Fig. 7.7. Fourier Transform of the Fourier Spectrum
The following description is for 1D objects; generalization to 2D
using vectors r in real space and q in Fourier space is straight­ Without recording the diffraction pattern, the waves in Fourier
forward. space propagate further. From each point q, a spherical wave
 204 7 Electron Waves, Interference, and Coherence

spreads into image space, carrying the complete information, with the formal result
i.e., the amplitude and phase of the wave diffracted at the object,
into the Fourier point q. Since the desirable magnification M is ima.Ox/ = A.Ox/ exp.i.Ox// (7.19)
Chapter 7

high ( 50 or more), the image plane is comparably far away;

the travel distance is given by M*f with focal length f, and hence Evidently, with a real lens, amplitude A.Ox/ and phase ˚.Ox/ of the
the primordially spherical waves flatten out. Consequently, the image wave, besides xO = Mx, are different from amplitude a.x/
image plane can be considered to lie in the far-field of the back and phase '.x/ of the object exit wave, unless we have aberration
focal plane. This shows that, in the image plane, the image wave correction so that the PSF really is a point, i.e., PSF.Ox/ = ı.Ox/.
is built up by interference of all the (plane) waves arriving from Then the convolution results in ima.Ox/  obj.Ox/ .
the spots q in the back focal plane. Again, the result is simply
described as a Fourier transform: Smearing by the PSF
Z
ima.Ox/ = spec.q/ exp.i2qOx/dq
(7.13) The image wave is given by the object exit wave “smeared
backfocalplane
out” by (i.e., convoluted with) the point spread function
generated by the objective lens. 9
The integral over q indicates that it is now leading from Fourier
Chapter 7

space back to real space. Combining steps I and II, one obtains
We’ll look at the details and some applications in, e.g., Chap. 8
ima.Ox/ = FT−1 ŒFT Œobj.x/ (7.14) on electron holography.

resulting in Summary of the Formation of the Image Wave

ima.Ox/  obj.Ox/ (7.15)
The microscopic imaging is best described by Abbe’s
with xO = Mx. theory:
– In the back focal plane of the objective lens, the Fraun-
Consequently, the image wave arising by an ideal lens perfectly hofer far-field wave is formed. It is the Fourier trans-
agrees with the object exit wave. ‘Ideal’ means that the lens is form of the object exit wave, called object exit wave
infinitely extended, hence accepts all diffraction angles ϑ out to Fourier spectrum.
90°, and that it does not suffer from imaging aberrations hence – Due to shortcomings of a real lens, the object spec-
focuses perfectly. Unfortunately, real lenses cannot be ideal. To trum is altered by multiplication with the wave trans-
cover real imaging, the above procedure has to be modified by fer function (WTF), resulting in the image wave spec-
means of a so-called wave transfer function (WTF), trum.
– The inverse Fourier transform of the image wave spec-
WTF.q/ = B.q/ exp.−i.q// (7.16) trum is the image wave in the first intermediate image
plane.
with a real aperture function B(q) grasping the restricted open- – In case of lens aberrations, image wave and object exit
ing of the lens or a purposely inserted mask, and a phase factor wave do not agree. For a meaningful interpretation of
given by the wave aberration χ(q) describing aberrations of the the image, this has to be understood in detail.
lens. Before the inverse Fourier transformation in step II, this Remember: the amplitude contrast transfer function
WTF has to be multiplied on the Fourier spectrum of the object, (ACTF) and the phase contrast transfer function (PCTF)
yielding used in transfer theory are related by WTF = ACTF +
iPCTF. 9
ima.Ox/ = FT−1 ŒWTF.q/FT Œobj.x/ (7.17)

With Nthe convolution theorem saying that FT Œg  f  = FT Œg

N
FT Œf 
FT Œg  f  = FT Œg FT Œf , and with the
7.7 Electron Wave Function
complex point spread function (PSF)
PSF.Ox/ = FT−1 ŒB.q/ exp.−i.q// Up to now, we have described wave properties in a very general
manner applicable, e.g., to sound waves and electromagnetic
which gives the image wave of a point in the object, the image waves such as light. Fortunately, all these concepts can also be
wave can most concisely be summarized as applied to matter waves such as electron waves. There are tiny
O differences, such as that photons travel with the speed of light,
ima.Ox/ = obj.Ox/ PSF (7.18) as electrons are always slower than the speed of light, as de-
 7.8 Electron Interference 205

termined by the acceleration voltage Ua. For matter waves in In the following, if not otherwise stated, we assume perfect
general hence also for electrons, their basic nonrelativistic wave coherence, i.e., a well-defined electron wave. Coherence is ad-
equation, equivalent to the Helmholtz equation for the electron dressed again in Chaps. 8, 9, and 11.

Chapter 7
wave function  , is the Schrödinger equation
 
1
.−i„r + eA/2 − eV  = E (7.20)
2m0 7.8 Electron Interference
with electron rest mass m0, magnetic vector potential A, and
electric potential V; the total energy E is the sum of kinetic Electrons were detected as the first elementary particles, and
and potential energy, and h is Planck’s constant. For relativistic many experiments verify electrons appearing as particles.
electrons accelerated by the voltage Ua, the wave vector in an Therefore, electron waves are per se very astonishing. Electron
electric potential V and magnetic potential A, both assumed interference, not understandable in terms of particles, allows

Chapter 7
constant in time, follows from the (relativistic) Klein–Gordon revealing the evidence and basic properties of electron waves
equation as experimentally without the need of an elaborate theoretical ap-
p paratus going beyond the one sketched so far. Experimental elec-
k=
2em0 .Ua + V/
eP − A
e
(7.21) tron interference is most impressively realized by means of the
h h Möllenstedt biprism in a comparably simple setup (Fig. 7.8): the
positively charged biprism filament splits the wave issued from
with eP the unit vector in momentum direction. For A = O, the the source in two partial waves, and deflects them such that they
corresponding wavelength  = 1= jkj is the well-known de Brog- overlap in the detection plane downstream. There we find the
lie wavelength cosinusoidal interference pattern shown in Fig. 7.9.
h
= (7.22) In Fig. 7.8, the wave leaving the source is split in two partial
p waves, which, by action of the positively charged biprism fil-
ament, are superimposed downstream, forming a cosinusoidal
at the momentum interference pattern. Angle α determines the degree of spatial
p coherence discussed later, angle β the spatial frequency qc.
p = 2em0 .Ua + V/ (7.23)
At UF = 0 V, we see the two partial waves at right and left in
with the relativistically corrected acceleration voltage Fig. 7.9, separated by the shadow thrown by the biprism fil-
  ament. Fresnel fringes arising from diffraction at the filament
eUa edges decorate them. With increasing voltage, the partial waves
Ua = Ua 1 + (7.24)
ma c2 are shifted towards each other such that they overlap at about
UF = 20 V . The fine fringes are the cosinusoidal interference
For example, for electrons accelerated by Ua = 100 kV , the fringes showing fringe spacing decreasing with increasing fila-
wavelength amounts to  = 3:7 pm, which is much smaller than ment voltage; they are modulated by the coarse Fresnel fringes,
atomic structures in solids. This is one of the main reasons why as clearly seen in the line scans on the right.
electron microscopy is so successful.

Wavelengths

The electron wavelength is about five orders of magni-

tude smaller than the wavelength of visible light. 9

For the beginning, assume field-free space, i.e., V = 0 and A = 0.

Then the simplest solution of the wave equation is again a plane
wave of amplitude a, described by means of
 .r; t/ = a exp.i.2kr + ' − !t//

with the wave vector k of modulus k = jkj = 1=, pointing in the

direction of the electron momentum p = hk, and a phase '.The
frequency ! is given by the total energy by means of the Planck
relation E = „! with „ = h=2. Fig. 7.8 Möllenstedt electron biprism interferometer
 206 7 Electron Waves, Interference, and Coherence
Chapter 7
Chapter 7

Fig. 7.9 Electron interference pattern with increasing biprism voltage UF

The biprism consists of a thin conductive filament perpendicular 7.9 Findings

to the drawing plane, which is positioned between two parallel,
grounded electrodes. A biprism voltage UF can be applied to the
filament with respect to them. The spherical wave issued from If we can detect the electrons with sufficient time resolution, we
the source is split up by the filament in two partial waves; by would observe the build-up of the interference pattern by single
means of the filament voltage, they are deflected towards each electrons dropping one-by-one on the detector (Fig. 7.10): each
other by the angle ˙, respectively, hence overlap downstream. electron impinges in a well-localized impact spot, i.e., it does not
On the detector, a cosinusoidal interference pattern is found spread over the whole interference field, hence, at first glance,
made up by fringes of spacing 1=qC with qC = kˇ. The larger it does not show the characteristic delocalization of a wave; in­
the biprism voltage UF, hence also the deflection angle  and the stead, it shows up as a localized particle. However, in contrast to
angle of superposition ˇ, the finer is the fringe spacing. classical particles, for these quantum particles we cannot predict
the trajectory, hence not the impact position beforehand: most
probably, the other electrons impinge in different impact spots.
In principle, this is true even more, if the starting conditions
for all electrons are made the same. (Be sure you understand
the Heisenberg Uncertainty Principle.) This becomes clear by
realizing that, though the impacts are well-localized, they are
distributed across the whole interference field.
 7.11 Coherence 207

Since the final intensity arises from averaging over a large num-
ber of mutually independent electrons, this pattern must be a
property of each single electron.

Chapter 7
7.10 Interpretation

In the terminology of modern physics, electrons are considered

Fermions. This means that no more than two electrons (with
opposite spin, ‘Pauli principle’) occupy the same quantum state
corresponding to one wave. In contrast, Bosons such as photons

Chapter 7
can occupy a same quantum state in a large number, as realized
in a laser. Instead, for electrons the occupancy number of a state
is usually even far less than 2 and, depending on the brightness of
the electron source, only about some 10−6 of the available states
are really occupied with an electron. Therefore, accommodation
of the large number of detected electrons in different states needs
a correspondingly large number of different waves. These differ-
ent waves are not related statistically in that, for example, their
phases do not show any fixed relation; quantum mechanics calls
these different states ‘orthogonal states.’ This means that, at su-
Fig. 7.10 Build-up of an electron interference pattern by single electrons
perposition of different waves, no mutual interference phenomena
show up between them, because the respective cosinusoidal term
cancels out under averaging over the random phase differences
Interference Is Good
between them; the resulting intensity is then simply given by the
sum of the intensities of the single waves. In this sense, different
Electron interference allows us to study very basic phe- electron waves are called mutually incoherent. Interference can
nomena of quantum physics. It’s the basis for image for- only be found if parts of the same wave, which evidently carry
mation in HRTEM and holography. 9 the same phase, are superimposed (‘self-interference’). This is
realized by the biprism: the wave of each electron is split into
two parts, which have a fixed phase relation; these parts show in-
At very small exposure times, only very few electrons are re- terference. In fact, we have single-electron interference, as Dirac
corded, which do not reveal any wave character. With increasing put it already in 1932. Then the final questions remain: how can
exposure time, the accumulated electrons built up the interfer- we see electron interference at all, if only self-interference within
ence pattern, as you see in Fig. 7.10. Since all electrons are mu- one wave is possible and the respective electron produces only
tually independent, the pattern must be a common property of a localized impact rather than an extended interference pattern?
each of them! How can all the different incoherent waves behave so that they
accumulate a well-defined interference pattern with very many
At first glance, with only a few electrons recorded, the impact mutually incoherent electrons?
positions seem randomly distributed. This does not surprise
us since the electrons move independently through the micro- Food for Thought
scope column without mutual interaction. While an electron
is hitting the detector, the next to come is still in the emitter
more than 1 m away; therefore we assume that interaction be- Every electron is only interfering with itself! 9
tween these electrons may virtually be excluded: the electrons
are quite alone in the microscope. Nevertheless, collecting a
large number of electrons with sufficiently identical, so-called
‘coherent’ waves, we find a nonhomogeneous distribution of
impacts oscillating periodically in x. With increasing exposure 7.11 Coherence
time, the impact density approaches the intensity distribution
I.x/ = 2a2 .1 + cos.2qC x + '//, which represents the proba-
bility distribution of detecting an electron at a given position. In spite of mere ‘self-interference,’ observation of electron inter-
This illuminates the interpretation of the electron wave as a prob- ference with many electrons is possible, if all the different waves
ability wave. In no case can the extended wave pattern be derived are assumed identically shaped, i.e., they propagate with the same
from a sharply localized single electron impact. wavelength from the same point of origin. Nevertheless, their
phases are mutually random, because the electrons are in different
 208 7 Electron Waves, Interference, and Coherence

In Fig. 7.12a, identical items are averaged, yielding identity;

this corresponds to electron interference repeated with the same
electron – which, alas, cannot be realized experimentally. In
Chapter 7

Fig. 7.12b, slightly differing representatives of the same spe-

cies (members of the Triebenberg-Lab) are averaged, revealing
their common properties; this corresponds to a partially coherent
ensemble of different but very similar electrons disclosing an
interference pattern as common property, however, at reduced
contrast. In Fig. 7.12c, averaging about different species renders
Fig. 7.11 Coherent vs incoherent averaging no common ensemble properties, and accordingly no interfer-
ence can be seen with very different electrons.
states. Fortunately, at interference of different parts of each single
wave, the random phases drop out, and hence the respective inter- An extended source is visualized being composed of incoherent
ference intensities produced by the different waves look the same: point sources with respective heights (Fig. 7.13). Independently
the average taken over these very similar interference patterns ap- from the rest, each point source gives rise to an interference
proximately agrees with all single ones from ‘self-interference.’ pattern with contrast 1; according to the emitting source point
Consequently the different electrons, though mutually incoherent, position rS, the respective fringes are shifted aside in the detector.
Chapter 7

fill up a well-visible interference pattern. This would be the case Incoherently summing up the interference patterns for all source
if the electron source would be an ideal point source emitting a points results again in the cosine pattern recorded by the detector.
well-defined single wavelength. Then, all waves would look the It is a fringe pattern with the same spacing, but shifted aside
same, however, with random phases. However, such a fictitious and with damped contrast, as described by the degree of spatial
source, although theoretically possible, cannot be realized. All real coherence in the following.
sources emit from an extended area, and they emit an extended
spectrum of different wavelengths. Waves emitted from different The question whether a source consists really of incoherent points
source points or with different wavelengths are assumed ‘incoher- or whether the waves at the source already have some extension
ent,’ meaning that they do not have a mutually fixed phase relation. of, say, some nanometers, is still open. This is a fascinating prob-
In this sense, a real source is named an incoherent source in that we lem not yet solved. Of course, such an extension would modify
do not see any interference phenomenon between waves emitted the description of coherence. Since we don’t know better, we
from different source points. This means that, on the detector plane, assume in the following that the source is built up of incoherent
we cannot average the waves (‘coherent averaging’) and obtain points.
the intensity from the averaged wave: an averaged wave does not
exist. Instead, we have to average the intensities of the individual
waves to obtain the intensity distribution produced by the wave-as-
sembly of all collected electrons. Averaging by intensities is called 7.11.1 Spatial Coherence
incoherent averaging.

Note in Fig. 7.11 that under incoherent averaging the mutual Assume an extended source given by the normalized intensity
interference terms with mixed indices are missing, because the distribution i.rS / such as shown in Fig. 7.13 where each of the
respective waves are mutually incoherent. Only self-interference source points produces an interference pattern of contrast 1, how-
survives. ever, specifically given by intensity and position of the source
point. All these mutually incoherent interference patterns have
The Pattern to be summed up incoherently, to obtain the one found on the
detector.

The detected interference pattern is the (incoherently) Assume further that this source emits only one wavelength. Then
averaged intensity of many electrons behaving in almost each point of the source contributes to the interference pattern
the same manner. 9 with the intensity

dI.x/ = 2i.rS /.1 + cos .2qC x +  + ı.rS /// drS

An important question remains. ‘Under which circumstances
does the total intensity, that results from averaging the inten- Depending on the source coordinate rS, the fringes are shifted
sities of the single waves over the extended source, still show aside by the phase ı.rS / = 2qC ab rS I a and b are the distances
wave phenomena such as interference fringes?’ This question explained in Fig. 7.13. Integration over the source yields the
is answered by the degree of coherence, which is derived in the averaged intensity
following.
(7.25)
ˇ ˇ
I.x/ = 2.1 + ˇSC .˛/ˇ cos 2qC x +  + SC .˛//
 
 7.11 Coherence 209

Chapter 7
Chapter 7

Fig. 7.12 Visualization of averaging effects for: a identical items; b slightly different items; c very different items
 210 7 Electron Waves, Interference, and Coherence

Fig. 7.13 Electron interference with Fig. 7.14 Distribu-

an extended electron source tion of spatial co-
herence in a plane
Chapter 7

illuminated by a
given source

Here we have introduced the ‘complex degree of spatial coher-

ence’ 7.11.2 Coherent Current
ˇ ˇ
SC .˛/ W= ˇSC .˛/ˇ exp.iSC .˛// (7.26)
For wave optical experiments, we have to make the degree of co-
with ˇSC .˛/ˇ the resulting contrast and SC .˛/ the resulting po- herence sufficiently large. Therefore, by means of the condenser
ˇ ˇ
Chapter 7

sition of the interference fringes. optics, the radius rG of the illuminating source has to be made
correspondingly small. The crucial question is how much elec-
The argument ˛ W= ab ˇ is the angular distance, as seen from the tron current is available at a given degree of spatial coherence.
source, of two points in the detector plane subsequently super- This is derived in the following.
imposed by the biprism (see Fig. 7.8). ˛ is the same for all su-
perimposed pairs of points in an interference pattern;ˇ however,
ˇ it We are starting out from the axial brightness of electron radiation
changes with the biprism voltage. The modulus 0 < ˇSC .˛/ˇ < 1 defined as B = A˝ I
with the electron current I emitted from the
damps the interference contrast C independently from the coordi- area A into the solid angle ˝. According to Abbe sine condi-
nate x of the interference pattern. The phase SC .˛/ shifts the po- tion, brightness, given by the electron source and proportional
sition of the fringes, compared with a point source on optical axis. to the accelerating voltage, is invariant under optical imaging,
i.e., it cannot increase in the optical column. Inserting the data
For a source i.rS /, the degree of coherence is given as of the Gaussian source, i.e., A = rG2 and ˝ =  ˛ 2, we find the
Z brightness as B = I=.rG˛/2 . With the degree of coherence
SC
 .˛/ = i.rS / exp.i2k˛rS /drS SC .˛/ = exp −.k˛rG /2 derived above for a Gaussian-shaped
(7.27) electron source we can write for the coherent current emitted into
source
the cone illuminated at a degree of coherence SC
which is the Fourier transform of the source intensity distribu-
tion (‘Van Cittert–Zernike theorem’) emitting the wave num- ˇ ˇ B
Icoh ˇSC .˛/ˇ = − 2 log ˇSC .˛ ˇ
ˇ ˇ
(7.28)
ber k . Please note that for a symmetric source, SC .˛/ = 0 k
or . As an example, let us consider a Gaussian source
   
2 It decreases with increasingˇ degree of coherence (Fig. 7.15). At
i.rS / = exp − rrGS = rG2 . The degree of coherence fol- perfect coherence, i.e., at ˇSC ˇ = 1, no current is available. Since
  ˇ

  both B and k2 are proportional to the accelerating voltage, kB2 is

lows as SC .˛/ = exp − .k˛rG /2 with SC = 0. This is illus- independent from it, and hence Icoh is a mere property of the
trated in Fig. 7.14. emitter. There is no way to make it larger if the microscope is
already properly aligned.
The source in Fig. 7.14 is assumed Gaussian, monochromatic,
and isoplanatic. Isoplanatic means that each point in the illumi- For a degree of coherence ˇSC ˇ approaching zero, the current
ˇ ˇ

nated plane ‘sees’ the same source. Then the degree of spatial co- would mathematically tend to infinity; physically it is limited by
herence SC .˛/ is also Gaussian. It answers the question: ‘What the total current emitted by the source. This suggests that there
is the coherence between two points at a mutual angular distance is always a tiny degree of coherence, even if the whole emitted
˛ in the illuminated plane?’ Consider a point at x0; it is always current is used.
fully coherent with itself, because the angular distance with itself
is ˛ = 0, and hence ˇSC .0/ˇ  1. Consider the point atˇ x and aˇ The plot in Fig. 7.15 answers the question: “How many electrons
ˇ ˇ

neighboring one at x0 + c; since with c also ˛ increases, ˇSC .˛/ˇ are available for doing wave optics atˇ a given degree of spatial
decays with c. This means that, superimposing these two points coherence?” The curve shows ˇ SCthat
ˇ SC ˇ = 1 we have full co-
ˇ at 
ˇ

by an optical system such as mirrors, lenses ˇor a biprism, we herence but no electrons; at ˇ ˇ # 0 we have ˇ theˇ totally emitted
would find interference fringes with a contrast ˇSC .˛/ˇ < 1 ap- electron current, but nearly no coherence. At ˇSC ˇ = 0:6, we have
ˇ

proaching zero for large c, hence large ˛. So we can summarize Icoh = 108 e=s available, which allows optimal wave optics. Illu-
this idea as follows. In the illuminated plane, the intensity is the minating an area of 100  100 nm2, we obtain 100 e=.0:1 nm/2 =s.
same everywhere, but coherence between points is limited to Numbers are valid for B = 2  108 A=.cm2 sterad/ at Ua = 100 kV.
small patches around each point.
 7.11 Coherence 211

7.11.5 A Generalization

Chapter 7
The above description of coherence properties in an electron
interferogram is the simplest case. When imaging an object in a
TEM – conventionally or holographically – the situation is more
complicate in that each electron, according to its wavelength
and illumination direction, experiences a slightly different in-
teraction with the object as well as slightly different focusing
action of the optics, in particular when aberrations are present.
Then the simple application of the degree of coherence may not
be sufficient.
Fig. 7.15 Coherent current vs degree of spatial coherence

Chapter 7
Instead, for describing wave optics in a TEM most accurately, we
should follow this four-step procedure (but we don’t):
7.11.3 Temporal Coherence
zz Run the wave for a single electron through the object and
the optics. For the illumination direction you’re using and
In a similar way, the influence of the spectrum s./ of emitted the chosen electron energy take into account all respective
wave numbers can be described,
p where  = k − k0 means the dif- specific effects occurring at interaction and focusing.
ference with respect to k0 = 2em0 Ua =h; again, e,m0, Ua, and zz Determine the intensity of the wave in the detector.
h mean charge and rest mass of the electron, the relativistically zz Do the same for all contributing electrons – their waves differ
corrected accelerating voltage and Planck’s constant, respec- in wavelength, energy, and source position, i.e., illumination
tively. The degree of temporal coherence results as direction.
zz Obtain the recorded signal by summing over all intensities
of the different waves in the detector plane. This is the final
Z
SC .n/ = s./ exp.i2n=k0 /d (7.29)
image.

with n the order of interference, which is the ordinal number of

a fringe counted from the center of the fringe pattern. Again,
jtc .n/j describes the contrast reduction of the interference pat- 7.11.6 Coherence at Inelastic Interaction
tern, now depending on the order of interference, i.e., on the
coordinate x in the interference pattern. Evaluation shows that
for electron sources with a usual energy spread of about 1 eV at The coherence of the illuminating ensemble is preserved when it
FWHM (‘full width at half maximum’), the contrast reduction is interacts with an object, if the interaction is perfectly elastic, i.e.,
negligible for n < 1000. That is why in interferometry or holog- if there is no residual energy transfer involved. At inelastic in-
raphy, temporal coherence can be set to 1. Nevertheless, energy teraction such as excitation of plasmons, phonons, or inner-shell
spread plays a very important role also in HRTEM and electron transitions, however, coherence with the elastically scattered or
holography, since in combination with chromatic aberrations of unscattered waves is strongly affected. The main reason is that in
the optics it can strongly affect the imaging quality. an interference pattern an energy transfer δE produces a time-de-
pendent phase shift with the frequency ı! = ıE=„ which blurs
the interference fringes during the exposure time . For the usual
  1 s, an energy transfer of ıE  4:135  10−15 eV destroys co-
7.11.4 Total Degree of Coherence herence. This is confirmed experimentally. Therefore, at any of
the usual inelastic processes, the respective electron wave field
becomes incoherent with the elastic and unscattered ones. We
For the mostly realistic case that the emitted spectrum does not can interpret it such that its origin electron wave collapses in the
depend on the source coordinate rS, the total degree of coherence region of an inelastic interaction.
can simply be factorized as
After inelastic interaction, the electron propagates again as a
tot .˛; n/ = sc .˛/tc .n/ (7.30) wave, however, newborn with a random phase and hence in-
coherent with the remainder. Interestingly, coherence measure-
Since in electron holography the number of fringes rarely ex- ments show that the newborn wave has, already at birth, a cer-
ceeds 1,000, jtc .n/j  1 is reasonable and tot = sc .˛/ can be tain extension, which depends on the process involved. There
assumed. is coherence inside the inelastic process. At bulk plasmons in
aluminum it amounts to about 30 nm diameter. This brings a
novel facet of coherence, namely the role of the object state. The
highly interesting topic of inelastic coherence is just evolving.
 212 7 Electron Waves, Interference, and Coherence

Summary of Coherence
Chapter 7

– Coherence is a bilateral relation: “A is coherent with B”

or “A is coherent with B, but A is incoherent with C.”
– It does not make sense to say “A is coherent.”
– Different electrons are in different quantum states with
random phases hence mutually incoherent.
– Dealing with electrons, we are talking about the statis-
tical coherence in the ensemble of electrons detected.
By means of the degree of coherence it describes the
visibility of wave phenomena observable with many
mutually incoherent electrons accumulated during ex-
posure time.
– Ensemble coherence is a property of the detected en-
semble of electrons. It does not say anything about
whether electrons are waves or particles: each electron
is a wave.
Chapter 7

– Ensemble coherence is given by spatial coherence and

by temporal coherence.
– Spatial coherence is given by the Fourier transform of
the intensity distribution of the source.
– Temporal coherence is given by the Fourier transform
of the spectrum of wave numbers emitted by the elec-
tron source.
– Inelastic interaction destroys coherence with the elastic
waves. 9

Chapter Summary

Many of the essential phenomena in an electron microscope We offer some access to the peculiarities of electron waves,
such as at interaction of electrons with the object, propaga- and to the question, how can we bring together the ideas of
tion, focusing and defocusing, action of aberrations, diffrac- electron waves and particles.
tion and interference, and finally the image formation on
the detector can definitely not be understood with electrons The term “coherence” is often used, visualized, and de-
considered as particles running on trajectories. All these scribed. Coherence does not say anything about whether
steps need waves and wave optics. electrons are waves or not: they are waves. The difference
between coherence and incoherence is whether we see the
For operating an electron microscope and for understanding waves or not. Coherence is a property of the ensemble of
the findings you have to develop a thorough familiarity with the electrons collected on the detector.
waves, wave properties, and wave behavior. We offer you
some help. As you see, the mathematical description of many wave opti-
cal phenomena leads to Fourier transforms. Is this frightening
In this chapter you find the basics for understanding inter- you? In case you are not acquainted with Fourier transforms
ference, modulation of a wave by an object, propagation and their properties, this is an excellent opportunity to ac-
of a wave in space leading to the well-known diffraction quire some familiarity: Fourier transforms make life so easy,
phenomena, and forming the image wave in a microscope. not only in optics, imaging, diffraction, and spectroscopy, but
also in many other fields of natural sciences.

Licensed to ([email protected])
 Appendix 213

Appendix

Chapter 7
People space’? What is the relation between diffraction angle
and spatial frequency?
Q7.4 What is the meaning of “point spread function”?
Joseph Fourier lived from March 21, 1768 to May 16, 1830. He Q7.5 Describe the convolution theorem of Fourier transforms
was a mathematician and physicist, not only known for his series in your own words.
and his transform but also for discovering the greenhouse effect! Q7.6 Explain how we can visualize convolution.
He traveled to Egypt with Bonaparte in 1798. Q7.7 What is an incoherent source?
Q7.8 Using an incoherent source, how can you reach coherent
Hermann Ludwig Ferdinand von Helmholtz was a polymath illumination of your object?
born in Potsdam on August 31, 1821, who died on September 8, Q7.9 Why does it not make sense to say “A is coherent”?
1894, when 73 years old. If all he had done was to invent the Q7.10 How many electrons are available for doing wave optics
ophthalmoscope, he would have made an enormous contribution. at a given degree of spatial coherence?
He also made major contributions to understanding heat, color,
and electricity, and much more, and not through unconscious
inference. His Ph.D. students included Albert Michelson (the
thesis advisor of Robert A. Millikan) and Friedrich Schottky, the Text-Specific Questions
father of Walter H. Schottky!

Walther Kossel (born January 4, 1888 in Berlin; died May 22, T7.1 What is the difference between coherent summation and
1956, in Tübingen) was the Ph.D. advisor of Gottfried Möllen- incoherent summation?
stedt who worked on interference phenomena in convergent T7.2 What is the meaning of ‘self-interference’?
beams (Kossel–Möllenstedt interference). He is also known for T7.3 Can different electrons interfere? Explain your reasoning.
the Kossel–Stranski–Volmer ledge-growth of thin films, which T7.4 Explain in your own words the phase problem of diffrac-
dates from his 1938 paper; his father won the Nobel Prize for tion.
medicine in 1910. T7.5 Sketch the Abbe theory of imaging in your own words.
Write it down as concisely as possible – possibly in one
Friedrich Lenz was born on March 21, 1922, and died in 2014, equation. Insert the influence of aberrations.
aged 92 years. He worked with Bodo von Borries, the first Pres- T7.6 Explain the Huygens principle.
ident of IFSM, and himself served for 8 years on the IFSM Ex- T7.7 What is the mathematical description of Huygens princi-
ecutive Council. He moved to Tübingen to work with Gottfried ple?
Möllenstedt, and soon became Professor of theoretical electron T7.8 Why does the illumination intensity decrease when you
optics, where he really did work on lens equations. increase the degree of coherence?
T7.9 Which figure of merit of the source determines the ob-
John Henry Poynting was born in Monton, England on Septem- tainable coherent current?
ber 9, 1851 and died 90 miles south in Birmingham aged 61 on T7.10 What is the meaning of the angle alpha in the degree of
March 30, 1914, He was advised by Maxwell at Cambridge and spatial coherence?
wrote a well-used physics text with J. J. Thomson.

References
Self-Assessment Questions

Q7.1 You want to confine a plane wave with a slit. What hap- General References
pens to the wave?
Q7.2 A plane light wave is running through a water droplet. Born M, Wolf E (1980) Principles of Optics, 6th edn. Pergamon
Sketch the object exit wave. Is it still a plane wave? Ex- Press, London
plain the meaning of ‘phase shift.’ Hecht E (1989) Optics. Addison Wesley, Boston
Q7.3 Meaning of Fourier transform: write down explicitly the Champaney DC (1973) Fourier Transforms and their Physical
FT of an object exit wave. What is the coordinate in Fou- Applications. Academic Press, London and New York
rier space? Why is Fourier space also called ‘reciprocal
 214 7 Electron Waves, Interference, and Coherence

Davis EA, Falconer I (1997) J J. Thompson And The Discovery Ehrenberg W, Siday RE (1949) The refractive index in electron
Of The Electron. CRC Press, Boca Raton-Florida (For some- optics and the principles of dynamics. Proc Phys Soc B62:8–
thing different) 21 (London)
Chapter 7

Brigham EO The Fast Fourier Transform. Prentice Hall, New Aharonov Y, Bohm D (1959) Significance of electromagnetic
Jersey potentials in quantum theory. Phys Rev 115:485
Lichte H, Lehmann M (2008) Electron holography – basics and Scherzer O (1949) The theoretical resolution limit of the electron
applications. Rep Prog Phys 71:016102 (Online at stacks.iop. microscope. J Appl Phys 20:20
org/RoPP/71/016102) Anaskin F, Stoyanova IG (1965) The Phase of an Electron
Reimer L (1989) Transmission Electron Microscopy, 2nd edn. Wave Scattered Over a Semiplane. Radiat Engin Electr Phys
Springer-Verlag, Berlin 13:1104–1110
Peng L, Dudarev SL, Whelan MJ (2010) High-Energy Electron Möllenstedt G, Lenz F (1962) Some Electron Interference Ex-
Diffraction and Microscopy. Monographs on the physics and periments and Their Theoretical Interpretatio. J Phys Soc Jap
Chemistry of Materials. Oxford University Press, Oxford 17BII:183–186
Tonomura A (1998) The Quantum World Unveiled by Electron Zernike F (1938) The concept of degree of coherence and its
Waves. World Scientific, Singapore application to optical problems. Physica 5:785–795
Röder F, Lichte H (2011) Inelastic Electron holography – First
results with surface plasmons. Eur Phys J Appl Phys 54:33504
Chapter 7

Specific References doi:10.1051/epjap/2010100378

Möllenstedt G, Düker H (1956) Beobachtungen und Messungen

an Biprisma-Interferenzen mit Elektronenwellen. Z Physik
145:377–397
Electron Holography
8

Part I
Chapter Preview

Chapter 8
This chapter is essentially a tutorial on the subject of elec- in Chap. 29 of W&C. Those pages give you the flavor, and
tron holography. Although holography is widely known an idea of the possible formation we can extract by studying
from the 3D images on key-rings and postcards or Lloyd this phenomenon. In this chapter, we examine the details.
Cross’s artwork, the concept was originally proposed as a
way to improve the resolution of the electron microscope, Holography is pure wave optics! For the following, you
and led to Dennis Gabor’s Nobel Prize in 1971. The term have to understand the basic wave properties such as inter-
“hologram” was formed by combining the Greek word ho- ference and coherence. Even if you are familiar with wave
los, meaning “whole,” with the Greek word gramma, mean- optics, we recommend that you first carefully study Chap. 7
ing “message.” You’ll have read two pages on holography on Electron waves, interference, and coherence.

Chapter 8

© Springer International Publishing Switzerland 2016

C. B. Carter, D. B. Williams (Eds.), Transmission Electron Microscopy, DOI 10.1007/978-3-319-26651-0_8 215
 216 8 Electron Holography

8.1 Big Problem with TEM: Phase with its object amplitude a and object phase φ by means of con-
volution with the point spread function PSF, which, as shown
Contrast later, is the Fourier transform of the wave transfer function
Part I

WTF = B exp.−imic / . Look at Chap. 28 of W&C for an in-

As you know, TEM is unrivaled in imaging nanoscopic struc- troduction to the PSF. By this convolution, the amplitudes and
tures. Because of the extremely short wavelength of only a few phases of the object and image waves are intermixed with the ef-
picometers, it allows us to determine atomic-scale details of ma- fect that the object phase is partially transferred into the intensity
terials. The best resolution obtained to date is 0.05 nm, which has and hence is only partially visible; plus, alas, it is mixed up with
become possible with correction of both spherical and chromatic the amplitude information. In the end, this is described by the
aberration. Because of aberration correction (and the better un- famous phase contrast transfer function PCTF and the amplitude
derstanding of contrast mechanisms), the findings can really be contrast transfer function ACTF, which are the imaginary and
interpreted quantitatively. So, the TEM is in principle a measur- real part of the WTF and are related by WTF = ACTF + iPCTF.
ing device at atomic dimensions. However, to achieve 50-pm This shows that if we have perfect correction of the spherical
Chapter 8

resolution, the contrast transfer function (CTF) of the imaging aberration and defocus, i.e., mic  0, the PCTF equals zero for
system has to be tuned very accurately based on a high-precision all spatial frequencies and hence the intensity shows perfect am-
measurement of aberration coefficients. Today’s long-term sta- plitude contrast but no phase contrast at all. You have to really
bility of correctors and microscopes, however, is unfortunately understand these topics in some detail before entering hologra-
not nearly good enough, which means we always have experi- phy. So, please read the next section about the wave modulation
mental uncertainties due to imprecise imaging conditions. This and imaging in a conventional TEM very carefully!
means that, when recording a micrograph, a while after you’ve
corrected the aberrations, residual aberrations will be back again.

You can correct the residual coherent aberrations after recording 8.2 Wave Modulation
the data (a posteriori) if you can record both the amplitude A and
the phase Φ of the image wave: and Conventional Imaging
img .x; y/ = A.x; y/ exp.i˚.x; y// (8.1) In the following writing, even if it is not otherwise stated, we
assume perfect coherence, i.e., a well-defined electron wave.
However, there we have a big problem with conventional TEM: Hence we neglect all incoherent aberrations, e.g., chromatic ab-
only the intensity of the final image wave is ever recorded on erration, and assume perfect parallel illumination.
the detector – the phase is lost when we record the intensity of
the image wave: A wave modulation is determined by local change of object ampli-
tude a and phase φ. Consequently, by interacting with an object,
I.x; y/ = img .x; y/ img .x; y/ (8.2) both the amplitude and phase of the object exit-wave obj may be


altered. In general, the interaction of the electron wave with the

The loss of phase information means that we lose a lot of infor- object is very difficult to describe because the object properties
mation about the object structure. In particular, residual aberra- and the interaction processes have to be modeled very accurately.
tions cannot be corrected a posteriori, leaving an uncertainty in In particular, for atomic details, this is only possible numerically.
the quantitative interpretation of micrographs that show atomi- We can model the object by giving scattering potentials or, in a
cally resolved detail. more sophisticated way, by considering the electron density dis-
tribution using DFT (density functional theory); then, we simulate
Furthermore, long-ranging phase modulations on the scale of the interaction with the help of multislice calculations such as
several tens of nanometers caused by electric potentials, e.g., EMS. In the following, only the basic phenomena are outlined.
in p–n junctions in semiconductors or by magnetic domains in
magnetic materials, are difficult to image with a conventional
TEM. Surprisingly, the situation is even worse with modern
Cs-corrected TEMs, since the hardware correction of aberrations 8.2.1 Amplitude Modulation
results in a high band-pass filtering of the phase information so
that atomic details can be imaged with high contrast and high
precision; long-ranging phase modulations are, however, sup- The amplitude a may be locally modulated, e.g., by the following
pressed, hence not contributing to image contrast anymore. So mechanisms:
we can say that the (Cs-corrected) TEM is blind for coarse phase
modulations from electric potentials and magnetic fields! zz Scattering absorption contrast: electrons are scattered into
large angles and are intercepted by the aperture of the objec-
Considering only coherent aberrations, the image wave img is tive lens, hence are missing in the image wave.
related to the object exit-wave zz Interference within the object: waves are diffracted in a crys-
tal interfere and hence give rise to enhancement or reduction
obj .x; y/ = a .x; y/ exp .i' .x; y// (8.3) of the amplitude.
 8.2 Wave Modulation and Conventional Imaging 217

zz Inelastic scattering: electrons suffer an energy transfer of

more than about 4  10−15 eV and lose coherence with the
coherent (unscattered or elastically scattered) wave field.

Part I
Consequently, they drop out of the coherent wave we’re
considering and contribute to the background, which is in-
coherent with it.

By these mechanisms, we find the 2D amplitude modulation

a .x; y/ of the object exit-wave.

8.2.2 Phase Modulation

Chapter 8
A plane wave running through an object with electric potential Fig. 8.1 Due to an electric potential V (a) and a magnetic domain structure B
(b), there is finally a phase modulation j(x) of the imaging electron wave, which
V and/or a magnetic potential A is modulated in phase, i.e., its
propagates from top to bottom. l0 = 1/|k0| is the wavelength in vacuum of the
wave front is locally bent, or shifted backwards or forwards. In
plane electron wave. The flux of the B-field enclosed by the two green electron
the following, we compare this modulated wave front with the
paths gives the magnetic phase shift in x1
corresponding plane wave without an object. For this, we assume
a trajectory s orthogonal to the wave front leading through the
specimen. To obtain the phase at the object exit plane, we have ' .x; y/ = 'obj .x; y/ − 'vac (8.8)
to evaluate the so-called eikonal equation along all possible tra-

Chapter 8
jectories (Fig. 8.1) Let us make the reasonable assumption that V << Ua and A is so
weak that the scalar product can be understood as
Eikonal Equation
k ds  k dz (8.9)

The German word Eikonal is from the Greek meaning meaning that the trajectories essentially run parallel to the optic
“image” and translates as “eikonal equation.” 9 axis z. This is true for kinematic interaction, i.e., there is no
scattering at high angles due to dynamical interactions, which is
fulfilled when the object is tilted outside a low-indexed zone-axis
orientation. Then the phase shift by the object follows as
Z
'obj = 2 kobj ds (8.4)
Z (p )
object .Ua + V.x; y; z//
'.x; y/ =2k0 p  − 1 dz
Ua
taken through the object, with the wave vector obj
I (8.10)
e
− 2 A.x; y; z/ ds
p
2em0 .Ua + V/ e*
kobj = ep − A (8.5) h
h h
Performing the somewhat tricky integration, we finally show that
and ep the unit vector along the trajectory. The asterisk means we
take account of the relativistic correction. e
'.x; y/ =  Vproj .x; y/ − 2 ˚.x; y/ (8.11)
h
When there is no object present, the wave running through field-
free vacuum to the same plane will experience a phase given by as illustrated in Fig. 8.1. We have introduced the abbreviations
Z e
'vac = 2 k0 ds
(8.6)
 = 2
h (8.12)
vac

with the vacuum wave vector the interaction constant with  the relativistic electron velocity;
p we find   0:0073= .V nm/ for 200-keV electrons.
2em0 Ua 
k0 = ep (8.7) Z
h Vproj .x; y/ = V.x; y; z/dz
(8.13)
obj
The phase shift introduced by the object relative to the unmod-
ulated wave front is then given by
 218 8 Electron Holography

is the projected potential induced by the object. For example, There is only one way around this ‘missing-phase’ problem: to
assuming a sample that is uniform along the electron trajectory become recordable, the phases have to be translated into intensi-
z with mean inner potential VMIP, the projected potential is simply ties. The phase-contrast transfer in high-resolution TEM imaging
Part I

the mean inner potential times the specimen thickness t: partially achieves this, but not at all satisfactorily, as shown by
the phase-contrast transfer function (PCTF) sketched in Fig. 8.2.
Vproj .x; y/ = VMIP t (8.14) In the final image, a fraction of the image contrast, namely the
‘phase contrast,’ stems from the phase structure of the object
Furthermore, we have defined the magnetic flux exit-wave. However, this fraction is indistinguishably scrambled
I with the amplitude contrast that stems from amplitude modula-
˚ .x; y/ = A .x; y; z/ ds (8.15) tions. In particular, slowly varying ‘large area’ phase structures
are not visible, because the PCTF approaches zero for small spa-
tial frequencies, i.e., for large periods.
as enclosed by two trajectories running through the object and
Chapter 8

vacuum, respectively (Fig. 8.1). This results from the magnetic Propagation of the object wave in the z-direction leads to near-
induction B = curl
R A and Stokes’ Htheorem saying that the mag- field, ‘Fresnel’ diffraction, which is seen when we defocus; this
netic flux ˚ = B dF equals Ads taken along the closed is described in Chap. 7. Far away from the object, the far-field
area F loop (‘Fraunhofer’), diffraction wave is built up as described by the
path surrounding the area F. ˇFourier
˚ spectrum ˇ2FT obj .x; y/ of the object wave; its intensity
˚
ˇFT obj .x; y/ ˇ is found, e.g., in the back focal plane of the
Amplitude and Phase Information objective lens representing the well-known electron diffraction
pattern; again, the phases are invisible.
The phase is much easier in interpretation, because it is
Diffraction Pattern and Diffractogram
generally a more direct measure of object properties. 9

The diffraction pattern is the modulus squared of the Fou-

rier transform of the electron wave exhibiting all asym-
metries as observed in the TEM in diffraction mode. In
8.2.3 What Do We See in an Electron Image? contrast, the diffractogram is the modulus squared of the
Fourier transform of an HRTEM image intensity showing
the lattice spacings but no asymmetries, e.g., due to resid-
We can always describe the object exit-wave leaving the object as ual crystal tilt slightly out of zone axis orientation. 9

obj .x; y/ = a .x; y/ exp .i' .x; y// (8.16)

The lens aberrations alter FT obj .x; y/ by multiplication with
˚ 
It contains the information about the accessible object properties. the coherent WTF
In all the applications discussed so far in this book, however,
images and diffraction patterns are only recording intensities. WTFcoh .q/ = B .q/ exp .−imic .q// (8.17)
The respective phases are not recorded, because by the basic
laws of physics, we don’t have detectors to do so! This means with the aperture function B.q/, the coherent-wave aberration
that we lose a substantial amount of information about the object, mic .q/, and the 2D spatial frequency vector q = qx ; qy . For the
 
because the interaction with the object, in the first place, mod- correct sign convention it is important to note that we use
ulates the phases much more than it changes the amplitudes of
the electron waves. In particular, when phases are missing com- O obj qx ; qy = FT obj .x; y/
  ˚ 
pletely it means that we have a loss of accessibility to electric Z1
and magnetic structures in the object.   
= obj .x; y/ exp −2i qx x + qy y dx dy

Zernike Phase Plate

−1 (8.18)

In light optics, the problem of switching amplitude and as Fourier transformation from real to reciprocal space. As al-
phase is solved by application of the so-called Zernike ready introduced in Chap. 7, the coherent wave aberration de-
phase plate in the back focal plane of the objective lens. scribes axial aberrations like spherical aberration, defocus, astig-
This phase plate shifts the phases of diffracted beams matisms, and so on, like a phase plate in the back-focal plane of
by π/2, whereas the zero-beam remains unchanged by the objective lens. In the image plane, we find the image wave
propagating through a tiny hole. Comparable setups for
electron microscopy are currently under development.9 img .x; y/ = A .x; y/ exp .i' .x; y// (8.19)
 8.3 Principle of Image-Plane Off-Axis Holography 219

Part I
Chapter 8
Chapter 8
Fig. 8.2 a Contrast transfer from the object exit-wave into the image wave and its recording on the detector as intensity within the weak object approximation.
Note that in conventional TEM the phase channel is blocked, hence the phase information is lost. b PCTF and ACTF for an image Cs-corrected TEM under negative
Cs imaging (NCSI) conditions (spherical aberration Cs = −18 µm, defocus Dz = 6.5 nm for Ua = 300 kV), which pronounces the phase contrast transfer in particular for
high spatial frequencies as required for HRTEM. However, object exit-wave phases with low spatial frequencies, e.g. stemming from locally slowly varying potential
distributions as known from p–n junctions, are not imaged onto the detector. For demonstration of realistic imaging conditions, both PCTF and ACTF are damped by
the envelope function Etc for partial temporal coherence (chromatic aberration Cc = 1.3 mm, energy spread ΔE = 1 eV)

given by inverse Fourier transformation of the image Fourier 8.3 Principle of Image-Plane Off-
spectrum, i.e., by
Axis Holography
img .x; y/ = FT−1 FT obj .x; y/ WTFcoh .q/
˚ ˚  
(8.20)
Electron holography corrects the phase loss that occurs in con-
which, because of the convolution theorem, can be written as ventional imaging, because it translates the phases into inten-
sity variations in a hologram that is recordable by a detector.
img .x; y/ = obj .x; y/ ˝ FT−1 fWTFcoh .q/g (8.21) Comparable to Gabor’s proposal, image-plane off-axis electron
holography is a two-step procedure (Fig. 8.3): the actual electron
where ˝ means ‘convolution.’ This convolution with the PSF hologram is recorded in a slightly modified TEM, whereas the
given by PSF .x; y/ = FT−1 fWTFcoh .q/g is the reason that, usu- reconstruction is done numerically applying the laws of Fourier
ally, the image wave img does not agree with the object wave optics.
obj; the image amplitude A and phase  may differ from the
object amplitude a and phase ', because of the objective-lens
aberrations mic .q/ of the electron microscope and an aperture
B .q/ in the back focal plane. Finally, the image intensity reads as 8.3.1 Recording a Hologram
I.x; y/ = img .x; y/ img

.x; y/ = A2 .x; y/ (8.22)
Translation of the complete wave into a recordable intensity is
and hence the phase distribution ' .x; y/ is lost. For the sake achieved by interference of the image wave img with another
of simplicity, incoherent aberrations (such as chromatic aberra- coherent wave called the reference wave ref (Fig. 8.4). Now,
tions) are not taken into account in the above discussion since an interference pattern (‘hologram’) results with the intensity
this would require the application of the transmission cross-co- distribution
efficient (TCC).    
Ihol .x; y/ = img .x; y/ + ref .x; y/ img .x; y/ + ref .x; y/
(8.23)
 220 8 Electron Holography
Part I
Chapter 8

Fig. 8.3 Schematic workflow of image-plane off-axis electron holography

as a two-step procedure. The electron hologram as built up by means of the
Fig. 8.4 Sketch of the electron optical setup for image-plane off-axis electron
Möllenstedt biprism is recorded by the CCD camera and fed into the computer
holography. The object plane is illuminated by a plane electron wave. The object
as two-dimensional data array. By means of Fourier optics, the image wave is
area of interest is close to a hole in the sample so that behind the sample both
reconstructed. Also, the (residual) coherent wave aberration can be corrected
the object-modulated object exit-wave and an unmodulated reference wave are
by applying a phase plate in Fourier space yielding the object exit-wave. For
found. Both waves are imaged by the objective lens where (residual) wave aber-
finally deducing object properties like its atomic structure, the object exit-wave
rations alter the object exit-wave into the image wave; the reference wave is not
has to be determined by simulation and quantitatively compared with the ex-
changed by aberrations. By means of a positive bias at the Möllenstedt biprism,
perimentally gained object exit-wave; alas, the inverse scattering problem for
both waves are deflected towards each other resulting in two partial coherent
determining the object structure directly from the object exit-wave is not yet
virtual electron sources in the back-focal plane of the objective lens. In the image
solved. In any case, the database given by amplitude and phase is much better
plane, both waves are brought to an overlap under the superposition angle β
than available from the mere intensity image
forming the off-axis electron hologram on the detector

In general, if the reference wave is also modulated, this may

result in very complicated interference patterns hard to interpret. Ihol .x; y/ = 1 + A2 .x; y/ + 2C A .x; y/ cos .2qc x + ˚ .x; y//
To avoid this, it is better if the reference wave passes the object (8.25)
plane through a field-free area, e.g., a hole adjacent to the object
area of interest. Then a plane reference wave ref .x; y/ = 1 re- The term qc = kˇ is called the carrier spatial frequency of the
sults, which is neither modulated in amplitude nor in phase, and interference fringes, because it ‘carries’ the information of the
has the normalized amplitude of the illuminating wave. image wave. Again, C = IImax −Imin
max +Imin
is the fringe contrast or fringe
visibility mainly given by the degree of coherence, instabilities,
So far, the most successful method for electron holography and the modulation transfer function (MTF) of the CCD camera.
is image-plane off-axis holography, where the hologram is In the hologram, both the amplitude A and phase ϕ are encoded
recorded in the image plane (Fig. 8.4). ‘Off-axis’ means that in the intensity distribution, namely in contrast and position, re-
image wave and reference wave are superimposed at an an- spectively, of the cosinusoidal interference term; hence they can
gle β. This is best achieved by means of a Möllenstedt elec- be recorded by a detector (Fig. 8.5). So, holography solves the
tron biprism inserted between the back focal plane and the missing phase problem!
image plane by means of a modified selected-area aper-
ture holder (see Chaps. 13.III, as well as 23.7.A and 29.11
of W&C). The biprism tilts the two waves towards each
other, resulting in img .x; y/ ! img .x; y/ exp .ikˇx/ and 8.3.2 Reconstructing the Object Exit-Wave
ref ! ref exp .−ikˇx/ , respectively. The hologram arising
from overlapping both tilted waves (with ref .x; y/ = 1 ) in the
image plane under the angle of superposition ˇ (related to the In the next stage, we have to recover the image wave from the
object exit-plane) reads as hologram. To do this, we perform numerical image processing
according to wave optical laws by means of the following steps.
 
Ihol .x; y/ = img .x; y/ exp.ikˇx/ + exp .−ikˇx/
Step 1 Fourier Transform the Hologram
h i
 (8.24)

img .x; y/ exp .−ikˇx/ + exp .ikˇx/
The extraction of amplitude and phase of the image wave from
the modulated interference pattern is best done in Fourier space.
resulting in The Fourier transform of the hologram (Fig. 8.6) reads
 8.3 Principle of Image-Plane Off-Axis Holography 221

Part I
Chapter 8
Chapter 8
Fig. 8.5 Electron hologram of GaAs in [110] orientation as recorded in an image Cs-corrected TEM at 300 kV acceleration voltage with 2 s exposure time. The speci-
men thickness increases from the lower right corner, where amorphous material most probably from preparation is observed, to the upper left corner. The magnified
area clearly shows the fine interference fringes with a spacing of s = 37 pm. When you take a close look at the fringes, you will find that they are bending. This is the
phase information of the image wave we would like to retrieve in the reconstruction process. To demonstrate the atomic details, both the reconstructed image wave
as well as object exit-wave are only shown from the area enclosed by the dashed quadrate

FT fIhol g =ı .q/ + FT A2
˚ 
˚ 
centerband Step 3 Inversely Fourier Transform the +1 Sideband
+ C FT img ˝ ı .q − qc / + 1 sideband
(8.26) Back to Real Space
n o
+ C FT img ˝ ı .q + qc / − 1 sideband This yields the reconstructed image wave


img;rec .x; y/ = C Arec .x; y/ exp .i'rec .x; y// (8.27)

Note that the centerband represents the Fourier spectrum only
of the sum of the intensities of the image and reference waves, It is given as a complex array of data, from which ampli-
whereas each of the sidebands represents the Fourier spectrum of q ˚
the complete, complex image wave and its conjugate complex, tude Arec .x; y/ = Re img;rec + Im img;rec and phase
2 ˚ 2
respectively. Besides the sign in the phase function, both side-
bands carry exactly the same information; this is due to the fact 'rec .x; y/ = arg img;rec are extracted as separate data sets and
˚ 
that a Fourier transformation of a real-valued function (here the displayed separately as amplitude and phase image. Here we dis-
hologram intensity Ihol) is always point-symmetric relative to the tinguish between the image wave img .x; y/ used for recording the
origin in reciprocal space. hologram and the image wave img;rec .x; y/reconstructed from the
hologram, because they may differ due to the holographic process.

Step 2 Mask Out the +1 Sideband

We can use a numerical aperture to cut out the +1 sideband and Step 4 Correction for Distortion-Induced Phase
center it on the origin of Fourier space. This subimage˚ represents Modulations and Amplitude Normalization
the Fourier spectrum of the recorded image wave FT img , just

weighted with the fringe contrast C. Within the holographic reconstruction process, the bending of
interference fringes is intrinsically interpreted as phase modula-
 222 8 Electron Holography
Part I
Chapter 8

Fig. 8.7 Reconstructed image wave displayed in amplitude and phase using
a mask with 8 nm−1 radius for sideband isolation and the reconstructed empty
wave as retrieved from an empty hologram for correction of distortion-induced
phase modulations and for normalization of the amplitude image. Due to resid-
ual aberrations, the typical dumbbell structure of a zincblende structure in [110]
Fig. 8.6 Inner part of the Fourier spectrum from the GaAs hologram shown in orientation cannot be observed
Fig. 8.5. The high carrier frequency of qc = 1 / s = 1 / (37 pm) = 27 nm−1 allows a
clear separation of centerband and the two sidebands. The location of the hole
in the sample and hence the reference wave, as well as the sign convention in
the used Fourier transformation determines which of the two sidebands is the
+1 sideband. In the further reconstruction process, the +1 sideband is isolated
by a numerical mask and centered in Fourier space. Inverse Fourier transforma-
tion retrieves the image wave in real space

tion of the electron wave. However, the bending of interference

fringes may also stem from distortions induced by inhomoge-
neous charging of the Möllenstedt biprism filament, distortions
of the projector lens system of the TEM, or even from distortions
in the glass-fiber optics used for coupling the scintillator with
the CCD-chip in the camera. All these effects are summarized
Fig. 8.8 Reconstructed object exit-wave displayed in amplitude and phase after
under the notion of distortion-induced phase modulations, which correction of residual aberration of 8 nm defocus and (3 nm, 155 grad) two-fold
in a first approximation adds to phase of the reconstructed image astigmatism. Now, the dumbbell contrast can clearly be observed having a spac-
wave. ing of 1.41 nm. However, due to the short exposure time and hence the low
number of registered electrons, the retrieved object exit-wave is rather noisy,
Since holography is extremely sensitive to any bending of in- not allowing unambiguously distinguishing between Ga and As atomic columns
terference fringes, these effects have to be taken into account
within the reconstruction process. This can easily be achieved,
when immediately after recording of the object hologram Ihol a Residual Aberrations
second hologram is recorded where we leave all the imaging pa-
rameters unchanged except we shift the specimen out of the field
of view. This so-called empty hologram Iempty is reconstructed Because of corrector instabilities and higher-order aber-
with the same parameters yielding the empty wave in ampli- rations, the correction of residual aberrations is almost
tude and phase. Then, the image wave is divided by the empty always mandatory for quantitative HRTEM work with
wave in the complex domain. (Essentially it’s like a flat-field- spatial resolution around or better than 0.1 nm. 9
ing correction.) From this division, the reconstruction process
benefits in two ways. First, the amplitude is normalized, if the
contrast C of both holograms Ihol and Iempty is equivalent. Second,
the phase is corrected for the artificial phase modulations due to Step 5 Correction for (Residual) Coherent Aberrations
distortions. In the case where everything works perfectly (e.g., The original idea from Dennis Gabor was to use holography to
ignoring time averaging during recording, see later discussion), correct the TEM’s wave aberrations. In fact, arbitrary coherent
img;rec .x; y/  img .x; y/ holds, and the results may be inter- wave aberrations of the objective lens can be modeled by means
preted directly and quantitatively, as shown in Fig. 8.7. of a phase plate χnum, which is applied to the Fourier spectrum of
the image wave. If num  mic, the imaging process of the ob-
jective lens is inverted and aberrations are correspondingly cor-
rected, yielding the object exit-wave obj .x; y/ shown in Fig. 8.8.

Licensed to ([email protected])
 8.4 Properties of the Reconstructed Object Exit-Wave 223

8.3.3 What Have We Achieved so Far? transformation of tiny patches of the reconstructed object
exit-wave; this process allows us to display residual crystal
tilt relative to the low-index zone-axis orientation. Another

Part I
By the comparably complicated holographic method we have example is the isolation and reconstruction of partial waves
now available amplitude and phase of the image wave, or, after and hence imaging the object in the light of single reflections.
correction of aberrations, the object exit-wave. Both waves are This process is comparable with dark-field imaging in the
complex-valued quantities. Therefore, in addition to the conven- TEM but with the big advantage that we are also retrieving
tional intensity image, we have gained the amplitude and phase the (Fourier-) phase of these reflections locally varying over
information, which means that now, in addition to the crystallo- the field of view in real space. Consequently, all aspects of the
graphic structure, we see in the phase and thus the electric and object structure can be extracted and evaluated quantitatively.
magnetic fields produced by the object. These fields are an es-
sential part of the structure-properties relation we are looking for. For advanced interpretation as holography specialist, you have
to be aware about the following fundamental difference be-

Chapter 8
Inelastic Electrons Contribute tween conventional imaging and holography. In conventional
TEM, the final image represents the time average over the in-
tensities of all
E detected electron waves, which can be written
Because inelastically scattered electrons can also signifi-
D
as img img

. In holography, the reconstructed wave represents
cantly contribute to the conventional image intensity for- the time average of all detected electron waves written as img .
˝ ˛
mation (and this is not the only reason), the reconstructed This wave has never existed before, because the recorded elec-
amplitude is normally not the square-root of the usual trons have never been in one wave: each electron had its own
image intensity. 9 wave (see Chap. 7 on electron interference). Consequently, be-
sides the inelastic scattering contributions, the conventional im-
Dage intensity
E is˝ not ˛the
D squared reconstructed image wave, i.e.,

Chapter 8
 .
E

8.4 Properties of the Reconstructed
img img ¤ img img

This conclusion is particularly critical if the object changes

Object Exit-Wave during exposure time. As a simple example, consider a light
atom as a pure phase object: there will be no contrast in the
The interpretation of a wave in terms of the object is easier than conventional image. The wave reconstructed from the hologram
that of an intensity image, because the information read out of exhibits no amplitude modulation but a phase modulation. If,
the hologram is intrinsically quantitative, linear, comprehensive, however, after half of the exposure time, the atom is hopping out
and with zero electron energy loss. of the field of view, the conventional image still would show no
contrast, whereas the reconstructed averaged image wave will
zz Quantitative: the hologram is a 2D-array of measurement exhibit both amplitude and phase modulation. Without going
data. The data is the amplitude from the fringe contrast, and through the formulas, you can understand this from the fact that
of the phase from the fringe position. Amplitude normaliza- phase modulations are encoded in the hologram as bending of
tion is achieved automatically by use of an empty hologram interference fringes. If the atom hops during exposure, the CCD
recorded with the same parameters, however, without object. will record the average hologram with and without bending of
zz Linear: transfer of waves is linear, whereas intensity may con- interference fringes at the location of the hopping atom. There-
tain nonlinearities, which give rise to artifacts that are often fore, the hologram contrast is reduced at this location, leading
hard to interpret. Choosing a zincblende structure in the [110] to a modulation also in the amplitude of the reconstructed image
orientation at the right specimen thickness, you can obtain wave. In the real world, the interaction process between imaging
dumbbell-like HRTEM-contrast even in a TEM that does not electron wave and atom(s) normally modulates both amplitude
have enough spatial resolution to resolve the dumbbells. This and phase of the object exit-wave. Therefore if the specimen
contrast is obtained purely from strong nonlinear contribu- is changing under electron beam irradiation during exposure,
tions. Due to the linear nature of reconstructed waves, this contrast is produced in HRTEM and holography, which cannot
artifact cannot occur in holography. be interpreted quantitatively.
zz Zero loss: the reconstructed wave displays only interactions
that have suffered less than 4  10−15 eV energy transfer, hence The Beam Changes the Specimen
are nearly perfectly zero-loss filtered. The findings can almost
uniquely be interpreted in terms of elastic scattering.
zz Comprehensive: the object exit-wave is revived numeri- Always be aware of altering specimens as, e.g., obser-
cally and hence can be comprehensively evaluated by all the vations of moving adatoms on surfaces during one ex-
wave optical tools you can think of, both in real space and in posure or a series of exposures! Independent of the HR-
Fourier space. For example, diffraction patterns comparable TEM-method used, the quantitative interpretation of the
to selected-area diffraction in the TEM but only having the micrograph may significantly be hampered. 9
width of a couple of unit cells can be determined by Fourier
 224 8 Electron Holography

8.5 Requirements of Holography Fourier transforms (FFTs), extracting amplitude and phase
from the wave, tilting the wave, adding phases, and unwrap-
ping the phases. Unwrapping means the following: computing
Part I

These advantages of holography have to be paid for with con- by means of arg f g = arctan .Im f g = Re f g/ (in computer
siderably higher demands on the electron microscope used and languages, please use the atan2-function with two arguments
capabilities of image processing. for computation of the phase because this function consid-
ers the appropriate quadrant in the complex number space),
zz Coherence: for conventional imaging, the coherence require- the phase values appear ‘wrapped’ in the range .0; 2/. This
ments say that at least the PSF with diameter δ psf produced means that, when a phase value would exceed 2, it will start
by the aberrations has to be illuminated coherently. Since at again at 0. Often it is essential that we obtain the full phase
Scherzer-focus, δ psf  5ıscherz can be estimated, we need a range by ‘unwrapping,’ i.e., by adding or subtracting 2 cor-
coherently illuminated diameter of about only 1 nm. In holog- respondingly.
raphy, at least in the direction perpendicular to the biprism, we
Chapter 8

need much better coherence in that the width of the hologram The computational resources are not demanding so that the re-
has to be illuminated coherently. In principle, we can realize construction can be performed nowadays on any PC. The re-
any degree of coherence jsc j; however, we are dose-lim- construction software is commercially available, but you can
ited by the coherent current Icoh = − kB2 log.jsc j/ (derived in easily program the basic steps. In any case, then you will really
Sect. 7.9 on coherence). In any case, an electron source with understand the operations being performed on the holographic
high brightness B, such as Schottky emitter or FEG, is indis- data sets.
pensable. Since we do not need isotropic coherence, elliptical
illumination by means of the condenser stigmator consider-
ably enhances the dose rate in the object plane.
zz Field of view: the reconstructable field of view is given by the 8.6 Quality Criteria
hologram width. From the point of view of the investigated
object area, it should be often as wide as possible; from the
point of view of collectable electron dose, it should be al- The quality criteria for the reconstructed wave are the field of
ways as narrow as possible. Aiming at object details larger view, lateral resolution, and signal resolution. The role of the
than about 10ıscherz  2 nm , we are free to select any field first two is evident. Signal resolution describes the discernibility
of view up to several micrometers. Unfortunately, at high of amplitude or phase differences well above noise. These three
resolution, because of the aberrations, the field of view has criteria are strongly related to each other, but a full discussion
to be broader than at least 4 psf (psf is the diameter of PSF) will be beyond the scope of this introductory chapter. Therefore,
to grasp the intended information about fine object details. we will restrict to signal resolution in particular in the phase im-
In an image-Cs-corrected TEM, psf is significantly reduced, age, since the phase is the most important source of information.
hence smaller holograms with improved fringe contrast can
be recorded. The resulting reconstructed object exit-wave is The smallest phase difference detectable from hologram fringes
consequently exhibiting a higher signal-to-noise ratio import- (‘phase detection limit’) in two adjacent reconstructed pixels is
ant for further quantitative evaluation. generally given by
zz Fringe spacing: the spacing of the hologram fringes deter- p
mines the reconstructable lateral resolution qrec. The reason 2
ı'lim = p snr (8.28)
is the following: to mask out the sideband completely from C N
the centerband, in general the carrier frequency must meet
qc  3qrec for a crystalline samples in a zone-axis orientation; with fringe contrast C and N the electron number collected in
this means that in real space at least three fringes sample one each of them. The quantity snr is the signal/noise ratio at which
reconstructed pixel (see also Fig. 8.6). Due to instabilities that we want to measure the phase difference; usually, snr = 3 is a
result in an image spread, this is an upper approximation. For good choice.
all other samples, in particular in medium-resolution hologra-
phy, imaging electric potentials, and imaging magnetic fields, The contrast C is determined by the degree of spatial coherence
qc  2qrec is sufficient. jsc j, by instabilities given by Cinst and by the MTF (called CMTF
zz Pixel number of detector: each fringe has to be sampled by here) of the CCD camera. It may be further reduced by the fac-
more than four pixels of the detector. Therefore, finally the tor Cinel = exp.−d=inel / which stems from inelastic interactions
number npix of detector pixels limits the number of recon- with a specimen of thickness d at a mean free path inel for in-
structable pixels as nrec = npix =12 and nrec = npix =8, respec- elastic interaction. Finally, we have C = jsc j Cinst CMTF Cinel for
tively, depending on the carrier frequency qc as discussed in the fringe contrast.
the topic above.
zz Image processing: in holography, image processing means do- Consequently, when you’re setting up a holographic experiment,
ing wave optics in the computer. Therefore you need the ap- you must pay the greatest attention to the optimization of inter-
propriate software tools. You need the software, e.g., for fast ference fringe contrast C. For high-quality holographic recon-
 8.7 Application to Electric Potentials on Nanometer Scale 225

as a snapshot of the experiment and subsequent advanced eval-

uation giving unique analysis possibilities of the object under
investigation. Whereas holography on the atomic scale is applied

Part I
only by a few microscopy groups world-wide, holography on
the mesoscopic scale with fields of view of a couple of hundred
nanometers up to several micrometers has found its way into
a reasonable number of laboratories. This wide application is
due to the fact that in the mesoscopic scale, electron holography
in terms of lateral resolution is the most advanced technique
for measuring electric potentials and magnetic fields as phase
shift of the electron wave (cf. Sect. 8.2.2). In the following, we
are going to concentrate on applications of electron holography
measuring electric potentials.

Chapter 8
Fig. 8.9 Reconstructed object exit-wave displayed in amplitude and phase as
retrieved from a series of 20 holograms and 20 corresponding empty holograms,
each of which has an exposure time of 2 s. Due to the improved SNR by a fac-
tor of (20)1/2 ≈ 4.5, Ga (blue) and As (red) atomic columns can clearly be distin- 8.7.1 Phase Shift Due to Electrostatic
guished allowing the determination of the structural polarity
Potentials
structions, the fringe contrast C should not be less than 10 % in
the vacuum area of the field of view. Because of both inelastic We showed in Sect. 8.2.2 that the phase ' .x; y/ of the object
interactions and strong amplitude modulations (in particular in exit-wave is modulated by the projected potential Vproj
atomic resolution), the contrast may vanish within the object
(8.29)

Chapter 8
' .x; y/ =  Vproj .x; y/
area. This has definitely to be prevented by going to thinner ob-
ject areas, and at atomic resolution by choosing a defocus such
that amplitude modulation is smaller. The interaction constant is σ. This can be understood by remem-
bering that the electron wave propagates through the sample in
Fringe Contrast z-direction and collects information on the electric potential that
is dependent on position x and y according to
Have a fringe contrast as high as possible! Only then,
Z
Vproj .x; y/ = V .x; y; z/ dz
consider improving the number of registered electrons (8.30)
by longer exposure times or higher extraction voltage at obj

FEG. 9 In real-world samples, there are many sources for electric poten-
tials V(x,y,z) in materials; there are often several contributions
to the electrostatic potential and hence to the phase shift of the
The number of registered electrons should be increased but as electron wave. In the following, we are listing the most prom-
a secondary consideration. We can do this by optimizing the inent contributions, which can in principle be measured with
illumination and using longer exposure times. Longer exposure electron holography:
times, however, have the disadvantage of allowing instabilities
to reduce both the resolution and fringe contrast. A clever way Crystal potentials. matter consists of atoms, and atoms consist of
of working is to record a series of holograms such that each a positively charged nucleus and a negatively charged electron
hologram has a short exposure time. Then, the reconstruction cloud. From the point of view of a high-energy imaging electron
process is a bit more demanding because of correlating all data traveling at a relativistic velocity, and for the first approximation
sets; however, the results in term of signal-to-noise ratio pay off assuming an isolated atom model, the Coulomb potential of the
all our efforts, as shown in Fig. 8.9. nucleus is only partially screened by the electron cloud. The cor-
responding electrostatic potential of a crystal can be developed
into a Fourier series according to

8.7 Application to Electric

X
V cryst .x; y; z/ = VMIP + Vg exp .2igr/
(8.31)
Potentials on Nanometer Scale
g¤0

with Vg being the potential as Fourier component for the lattice

The previous sections have introduced electron holography with vector g in the crystal’s unit cell. VMIP is the so-called mean inner
examples of recording, reconstruction, and aberration correction potential as the zero component of the Fourier series, which is al-
on the atomic scale. They clearly showed the benefits of storing ways positive and represents the atomic potential averaged over
all available object (and microscope) properties in a hologram a unit cell. It depends strongly on the elements present in the unit
 226 8 Electron Holography

cell and hence it can be used for obtaining a compositional con- thicker the object, the better the signal. However we have to keep
trast in the phase of the object exit-wave. However, at a low-in- in mind the following restrictions:
dex zone-axis orientation, the phase of the object exit-wave is
Part I

modulated by all significant Fourier contributions as exploited zz The objects should be flat and thinner than the extinction
in HRTEM and atomic-resolution electron holography according length to avoid strong dynamic interaction. Otherwise phases
to dynamical interaction theory. In a kinematical approximation from dynamic interaction, which are sensitive to local vari-
achieved by tilting the sample out of the low-indexed zone-axis ation of thickness and tilt, will hamper the subsequent inter-
orientation, contributions to image formation from potentials Vg pretation.
with g ¤ 0 are strongly reduced. Therefore, the phase shift φ zz The objects should be much thinner than the inelastic mean
is almost only given by the mean inner potential VMIP and the free path. Otherwise damping of fringe contrast due to loss of
thickness t of the sample: coherence will impair the signal/noise properties of the recon-
structed wave and hence the discernibility of atomic details.
' .x; y/ =  VMIP .x; y/ t .x; y/ (8.32) zz In any case we have to be aware that the thinner the object the
Chapter 8

more the result is dominated by the surface structure, e.g., re-

If the local thickness t .x; y/ is known, the mean inner potential arrangement of surface atoms and effect of surface potentials.
VMIP can be determined as a measure for elemental composition. zz The field of view should be as small as possible, just sufficient
to cover the area of interest, e.g., across an interface or a de-
Potentials due to inner electric fields. The macroscopic proper- fect. This saves the precious coherent electrons.
ties, and correspondingly the function of many modern emerging zz The region of interest should be located nearby a hole, be-
materials that are often structured on the nanometer scale, may cause we need the reference wave. As a rule of thumb, this
arise from intrinsic electric fields due to charge redistributions. region can be at a distance from the object edge that is equiv-
alent to the width of the interference field.
A nice example of this can be found in InGaN/GaN multiquan-
tum wells (MQW), where the InGaN unit cell in the quantum At medium resolution for details larger than about 1 nm, a dimen-
well is tetragonally distorted and exhibits not only a spontaneous sion used for imaging and measuring phases due to slowly locally
polarization but also a piezoelectric polarization as electric field varying potentials, the requirements are different. Potential varia-
E which gives rise to an electrostatic potential Vpol according to tions in structured materials are very small, and hence the phase
E = −r Vpol. modulation may be buried in noise. Therefore, the precision of
measuring the tiny variations of intrinsic fields well above noise
A second example that has generated a lot of interest, in particu- has to be improved by integration over a higher sample thickness.
lar in semiconductor research and industries, is doping. Doping Contrary to HRTEM, the sample thickness should be rather high;
a semiconductor hardly changes the scattering potential even at for optimization of the signal/noise ratio, as a rule of thumb, the
high doping levels, so that the distribution of dopants cannot be thickness should be about two times the inelastic mean free path
imaged by TEM. Also the mean inner potential is only slightly for inelastic scattering, resulting in sample thickness of about
modified, too small a change for holographic imaging. However, 200–300 nm of most semiconductors at 200 kV acceleration volt-
in p–n junctions, due to redistribution of free charge carriers, the age. However, we have to be aware that the phase represents
resulting built-in potential Vpn, which is roughly in the order of the projected structure, i.e., it averages over all variations along
Vpn  VMIP =10 depending on the semiconductor and its doping propagation direction. This hampers interpretation if structure
level, can itself be imaged (and measured with support of theo- changes occur on a scale smaller than object thickness. In this
retical modeling) with a lateral resolution of a few nanometers. case, the only remedy is holographic tomography.
In Fig. 8.7c, an example of imaging the p–n distribution in a
semiconductor is shown.
Mesoscopy

For holography at mesoscopic dimensions, use a thick

8.7.2 Experimental Considerations sample and set the defocus of the lens to minimum image
contrast such that you can hardly see the object details
anymore in the TEM. 9
The demands on the objects (specimens) depend on the kind of
information we want to gain. They are different for atomic and
for medium resolution.
zz By examining also the amplitude of the reconstructed wave
At atomic resolution we want to measure the atom species and we can make sure that the phase is not influenced by the ef-
atomic fields as well as atomic structures, both perfect materials fects of dynamic interaction.
and materials with defects. Since, depending on the atomic num- zz Because of the structure size, often between 10 and 100 nm
ber, the phase shift of single atoms may be too weak for detecting and more, a large field of view with a width of several 100 nm
single atoms, a few monolayers are needed. At first glance, the up to 1 µm is desirable for imaging the complete structure.
 8.8 Further Derivatives of Electron Holography 227

This can be accomplished by using a first imaging lens that

has a large focal length; this is often achieved by having a
Lorentz lens as an additional lens in the lower pole-piece of

Part I
the objective lens. We switch off the objective lens that we
normally use, and then use the Lorentz lens for imaging the
object exit-wave. In image-Cs-corrected TEM, the so-called
Lorentz-mode is implemented by using one of the first trans-
fer lenses of the corrector as an objective lens with a high
focal length.
zz You want to choose the wave aberration χmic such that the
ACTF is maximized and the cross-talk between amplitude and
phase as given by the PCTF is minimized. In the experiment,
this imaging condition is gained by going to Gabor defocus

Chapter 8
p
DzGabor = −0:67 Cs  (Cs is the (uncorrected) spherical ab-
erration of the Lorentz lens) and minimizing the contrast in
the image intensity (assuming a pure-phase object).

Fig. 8.10 Phase shift of the electron wave at a sample with p–n junction. The
8.7.3 Application Example: p–n Junctions largest contribution to the phase shift stems from the mean inner potential,
whereas the built-in potential of the p–n junction only produces an additional
slight phase difference between n- and p-doped area. Please note that, due to
The basic building blocks of all electric semiconductor devices outdiffusion of the excess charge carriers, n-doping produces a positive potential

Chapter 8
are diodes. They consist of p–n junctions, which are produced and p-doping a negative potential. Correspondingly, the additional phase shift in
by selective doping of the semiconductors. The lateral distribu- n and p is positive and negative, respectively
tion of dopants cannot be imaged by electron holography, be-
cause the change of the mean inner potential that occurs because
of their different atomic numbers is only in the order of millivolt. ing this task with nanometer resolution, giving a high impact in
The built-in potential Vpn arising from the doping effect, how- semiconductors research and industry.
ever, is in the order of volts, hence it is certainly measurable. In
fact, the potential distribution is what is really relevant and is
actually more interesting than the dopant distribution, because
it is the potential that controls the semiconductor properties. 8.8 Further Derivatives of Electron
For the phase calculation, Vpn adds to the mean inner potential
VMIP following Holography
 
' .x; y/ =  VMIP t + Vpn .t − 2t0 / (8.33) An introductory chapter on electron holography would not be
complete without giving a look at emerging holographic meth-
The total sample thickness is denoted by t; due to (amorphous) ods, namely electron holographic tomography and dark-field
surface layers, pinning of Fermi levels at surfaces etc., the p–n electron holography.
junction is fully active only in the core with a ‘reduced sam-
ple thickness’ of t − 2t0 , where t0 is known as the dead layer
(Fig. 8.10). Note that this simple dead-layer model is only a very
first approximation to the actual surface layer of the specimen; 8.8.1 Holographic Tomography
it is often insufficient for describing the experimental outcome
because there is a mismatch between measured built-in potential
and the one that is expected from band diagram calculations. With the terminus technicus ‘holography,’ often the idea of
This is currently a hot topic of research that aims at providing a 3D imaging is known from visible-light optics. In visible-light
full quantitative evaluation of the measured phase distribution optics, the angular width of scattering is so large that the object
of p–n junctions. may be visualized from different angles, hence creating 3D for
an observer in real life as the hologram captures the wide an-
For the electric performance of a field-effect transistor (FET) as gular width. In electron optics, however, the angular width of
shown in Fig. 8.11, it is not the actual distribution of dopant at- scattering is so small that 3D is nearly impossible, even with
oms (as measured by secondary ion mass spectroscopy (SIMS)) holography. Consequently, a micrograph or a holographic re-
that is of importance, but only the electrically active ones that construction represents only object properties that are projected
produce the electrostatic potential distribution and determine the along the electron beam. This finding actually already shows a
position of the p–n junction. Electron holography is accomplish- way out of our dilemma: we simply widen the angular width arti-
 228 8 Electron Holography
Part I
Chapter 8

Fig. 8.11 Holographically reconstructed object exit-wave of an n-MOSFET (a) and p-MOSFET (b). The phase images clearly show the potential distribution of the
highly doped source and drain areas. Due to charge separation in the p–n junction, the n+-doped area exhibits a positive potential shown as a stronger phase shift
in respect to the substrate. With inverse sign, this is also true for the n-MOSFET. Note please that the amplitude images do not show any contrast due to doping

ficially by recording a set of holograms under different tilts of the 8.8.2 Dark-Field Holography
sample. The corresponding set of 2D waves forms a 3D data set
in both amplitude and phase; after tomographic reconstruction
of the phase, we see the 3D-potential distribution, e.g., in core– The switching speed of modern electrical circuits can signifi-
shell structures. This sophisticated method is known as electron cantly be improved by straining the substrate under the gate of
holographic tomography. a field-effect transistor. Furthermore, the optical properties of
 8.8 Further Derivatives of Electron Holography 229

advanced, quantum-dot structures and multiquantum wells are Finally, the strain tensor "Q is defined by its components
also influenced by strain. Strain is a measure of how strongly "ij = @ui .x; y/ =@j with i = x; y and j = x; y, respectively. Experi-
the crystalline atomic lattice of a strained region of interest is mentally, the geometric phase 'gG .x; y/ is measured by dark-field

Part I
distorted with respect to an unstrained reference region. Conse- holography. For this, the crystalline sample is tilted such that
quently, strain is a parameter that can strongly affect the prop- the diffracted wave g is strongly excited. To minimize aberra-
erties of materials, so strain must be determined along with the tions, we also tilt the illumination so that the diffracted wave g
other variables such as composition. is centered on the optical axis. An objective diffraction aperture
eliminates all other diffracted waves from the dark-field image.
The shift of atoms with respect to undistorted lattice planes is In the dark-field image plane, a biprism brings the selected dif-
described by the displacement field u .x; y/. This shift gives rise fracted wave of the strained region of interest to overlap with the
to an extra phase term stamped on the diffracted waves in Fourier correspondingly diffracted wave of an unstrained region as the
space, which is also denoted as geometric phase. It must not be reference wave and thus forms the so-called dark-field electron
confused with the phase shift of the whole object exit-wave in hologram. The reconstructed phase reveals the real-space distri-

Chapter 8
real space as reconstructed by electron holography. The strain-in- bution of the strain-induced phase in Fourier space. Assuming
duced phase is given by that the object is homogeneous (except for the strain) and that
it has a constant thickness, the phase reconstructed out of the
'gG .x; y/ = −2gu .x; y/ dark-field electron hologram can directly be interpreted as the re-
al-space distribution of the geometric phase 'gG .x; y/ of reflection
in the diffracted wave g in Fourier space conjugate to the lattice g. For more complicated samples, we leave you the challenge of
vector a in real space. The actual displacement field is deter- consulting the literature.
mined by two measurements of geometric phases 'g1 G
.x; y/ and
G
'g2 .x; y/ of two noncollinear diffracted waves g1 and g2 yielding

Chapter 8
1  G G

u .x; y/ = − a1 'g1 .x; y/ + a2 'g2 .x; y/
2

Chapter Summary

Nowadays TEMs are more than just microscopes with an The phase of the electron wave contains essential informa-
extremely high magnification such that they are capable of tion about the object such as the presence of electric poten-
imaging atoms; rather, they are instruments which allow tials and magnetic fields.
measurements with a high precision, utilizing sophisticated
methods. So, whenever you like to answer precisely ques- Electron holography can reveal both amplitude and phase.
tions like “which atoms are where?” and “how strong are
the magnetic fields and electrostatic potentials in my sam- For recording a hologram, a reference wave, coherently su-
ple?,” you should try using off-axis electron holography, as perimposed to the image wave, produces an interference
introduced in this chapter. pattern.

In this chapter you have learned a few things that you may The reconstruction is easily performed by means of Fourier
not have fully appreciated before. transforms.

The interaction of electrons with an object can only be un- Since the complete complex-valued wave is retrieved, most
derstood wave optically. sophisticated wave-optical analysis tools like aberration
correction, nanodiffraction, and imaging in the light of sin-
Waves comprise amplitude and phase. gle reflections can be performed.

The image formation is wave optics. Electron holographic tomography opens the window to
3D measurements of electric potentials.
Conventional TEM produces intensity images that have lost
the phase of the wave. Dark-field electron holography is very sensitive for measur-
ing displacement fields due to strain.
 230 Appendix

Appendix
Part I

People
Text-Specific Questions
Dennis Gábor, born June 5, 1900 in Budapest, died February 8,
1979 in London; inventor of holography, 1947; received Nobel
Prize, 1971. T9.1 Calculate the imaging of an object exit-wave into the
image wave and its recording as intensity on the detector
Gottfried Möllenstedt, born September 14, 1912 in Versmold, within the weak object approximation.
died September 11, 1997 in Tübingen; fundamental experiments T9.2 Perform the tricky integration to show that the phase shift
on electron interferometry; inventor of the Möllenstedt biprism of the electron wave is caused by the projected potential
(together with Heinrich Düker in 1954), initially using a met- and the enclosed magnetic flux.
al-coated spider’s web. T9.3 Look up the mathematical formulation of the wave aber-
ration χ. Which aberrations are included and which are
Chapter 8

Akira Tonomura, born April 25, 1942, died May 2, 2012; known not? Why?
for his work on electron holography and in particular for the ex- T9.4 Show that two coherent inclined waves brought to an
perimental verification of the Aharonov–Bohm effect. His family overlap produce a cosinusoidal interference pattern on
moved away from Hiroshima two months before August 1945. the detector.
He worked in Tübingen 1973–4. T9.5 Show by calculation that distortion-induced phase mod-
ulations can indeed be corrected by an empty hologram.
T9.6 How are aberrations corrected numerically?
T9.7 What are dead layers in the context of imaging p–n junc-
Self-Assessment Questions tions, and why is this model too simple?
T9.8 Make a sketch of the ray paths for dark-field electron
holography.
Q9.1 Why should you consider using electron holography in
an image Cs-corrected TEM for atomic imaging?
Q9.2 Why it is not possible to image the potential distribution
of a p–n junction in a fully image Cs-corrected TEM? References
Q9.3 Why it is so important to image the phase of the object
exit-wave? How this is done at least partially in conven- General References
tional TEM?
Q9.4 What is the difference between a diffractogram and a dif- Voelkl E, Allard LF, Joy DC (eds) (1999) Introduction to Elec-
fraction pattern? tron Holography. Springer Science+Business Media New
Q9.5 Name the contributions to shifting the phase of an elec- York
tron wave. Tonomura A (2010) Electron Holography. Series in Optical Sci-
Q9.6 Why is there a difference between an object exit-wave ences. Springer, Berlin Heidelberg
and an image wave? Lichte H, Lehmann M (2008) Electron holography – basics and
Q9.7 What are experimental prerequisites for imaging electric applications. Annu Rev Modern Phys 71:016102
potentials on the mesoscopic scale? Lehmann M, Lichte H (2002) Tutorial on Off-axis Electron Ho-
Q9.8 Why does electron holography use an elliptically shaped lography. Microsc Microanal 8:447–466
illumination?
Q9.9 Why it is so important to first maximize the interference
fringe contrast and only second the number of registered
electrons? Historic Papers
Q9.10 What are the possible applications of electron holography
in material sciences? Gabor D (1948) A New Microscopic Principle. Nature 161:563–
Q9.11 What is the difference between the phase of the electron 564
wave and the geometric phase? Gabor D (1949) Microscopy by reconstructed wave fronts. Proc
Q9.12 What are the reasons that reconstructions from electron R Soc. A197:454–487
holograms are not 3D? Möllenstedt G, Düker H (1956) Beobachtungen und Messun-
Q9.13 What is the difference between strain and displacement gen an Biprisma-Interferenzen mit Elektronenwellen. Z Phys
field? 145:377
 Appendix 231

Leith EH, Upatnieks J (1962) Reconstructed wave fronts and Atomic Resolution Electron Holography
communication theory. J Opt Soc Am 52:1123–1130
Möllenstedt G, Wahl H (1968) Elektronenholographie und Re- Lichte H (1986) Electron holography approaching atomic reso-

Part I
konstruktion mit Laserlicht. Naturwissenschaften 55:340–341 lution. Ultramicroscopy 20(3):293–304
Tonomura A, Fukuhara A, Watanabe H, Komoda T (1968) Op- Linck M, Freitag B, Kujawa S, Lehmann M, Niermann T (2012)
tical Reconstruction of Image from Fraunhofer Electron-ho- State of the art in atomic resolution off-axis electron hologra-
logram. Jpn J Appl Phys 7:295 phy. Ultramicroscopy 116:13–23
Wahl H (1975). Bildebenenholographie mit Elektronen. Thesis, Linck M (2013) Optimum aberration coefficients for recording
University of Tübingen. high-resolution off-axis holograms in a Cs-corrected TEM.
Ultramicroscopy 124:77–87
Lubk RFA, Lichte H, Bredow T, Yu W, Mader W (2010) Long-
Possibilities and Limits of Hardware Aberration range correlations in In2O3 (ZnO)7 investigated by DFT cal-
Correction culations and electron holography. Ultramicroscopy 110:400

Chapter 8
Ji CL, Houben L, Thust A, Barthel J (2010) On the benefit of the
negative spherical-aberration imaging technique for quantita- Imaging and Measurements of Electric Potentials
tive HRTEM. Ultramicroscopy 110:500
Rose H (2010) Theoretical aspects of image formation in the Frabboni S, Matteucci G, Pozzi G, Vanzi M (1985) Electron
aberration-corrected electron microscope. Ultramicroscopy Holographic Observations of the Electrostatic Field Associ-
110:488 ated with Thin Reverse-Biased p–n Junctions. Phys Rev Lett
Uhlemann S, Müller H, Hartel P, Zach J, Haider M (2013) Ther- 55:2196
mal Magnetic Field Noise Limits Resolution in Transmission McCartney MR, Smith DJ, Hull R, Bean JC, Voelkl E, Frost B
Electron Microscopy. Phys Rev Lett 111:046101 (1994) Direct observation of potential distribution across Si/

Chapter 8
Barthel J, Thust A (2013) On the optical stability of high-reso- Si pn junctions using off-axis electron holography. Appl Phys
lution transmission electron microscopes. Ultramicroscopy Lett 65:2603
134:6–17 Rau WD, Schwander P, Baumann FH, Höppner W, Ourmazd A
(1999) Two-Dimensional Mapping of the Electrostatic Po-
tential in Transistors by Electron Holography. Phys Rev Lett
Performance and Limits of Holography 82:2614–2617
Twitchett AC, Dunin-Borkowski RE, Midgley PA (2002) Quan-
Harscher A, Lichte H (1996) Experimental study of amplitude titative Electron Holography of Biased Semiconductor De-
and phase detection limits in electron holography. Ultrami- vices. Phys Rev Lett 88:238302
croscopy 64:57–66 Cooper D, Ailliot C, Barnes JP, Hartmann JM, Salles P, Benas-
Lichte H (2008) Performance limits of electron holography. Ul- sayag G, Dunin-Borkowski RE (2010) Dopant profiling of fo-
tramicroscopy 108:256 cused ion beam milled semiconductors using off-axis electron
Voelkl E, Tang D (2010) Approaching routine 2π/1000 phase holography; reducing artifacts, extending detection limits and
resolution for off-axis type holography. Ultramicroscopy reducing the effect of gallium implantation. Ultramicroscopy
110:447 110:383
Niermann T, Lehmann M (2013) Averaging scheme for atomic Kruse P, Schowalter M, Lamoen D, Rosenauer A, Gerthsen D
resolution off-axis electron holograms. Micron 63:28–34 (2006) Determination of the mean inner potential in III–V
semiconductors, Si and Ge by density functional theory and
electron holography. Ultramicroscopy 106:105
Ponce FA (2011) Electrostatic energy profiles at nanometer-scale
Advanced Holographic Setup in group III nitride semiconductors using electron hologra-
phy. Ann Phys, Berlin 52:75–86
Lichte H (1996) Electron holography: Optimum position of the
Zhou L, Gonschorek M, Giraud E, Feltin E, Carlin JF, Grandjean
biprism in the electron microscope. Ultramicroscopy 64:79–
N, Smith DJ, McCartney MR (2012) Measurement of polar-
86
ization-induced electric fields in GaN/AlInN quantum wells.
Harada K, Tonomura A, Togawa Y, Akashi T, Matsuda T (2004)
Appl Phys Lett 101:251902
Double-biprism electron interferometry. Appl Phys Lett
Simon P, Huhle R, Lehmann M, Lichte H, Mönter D, Bieber Th,
84:3229
Reschetilowski W, Adhikari R, Michler GH (2002) Electron
Tanigaki T, Aizawa S, Park HS, Matsuda T, Harada K, Shindo
holography on beam sensitive materials: organic polymers
D (2014) Advanced split-illumination electron holography
and mesoporous silica. Chem Mater 14:1505–1514
without Fresnel fringes. Ultramicroscopy 137:7–11
Tanji T, Urata K, Ishizuka K, Ru Q, Tonomura A (1993) Obser-
vation of atomic surface potential by electron holography.
Ultramicroscopy 49:259–264
 232 8 Electron Holography

Wanner M, Bach D, Gerthsen D, Werner R, Tesche B (2006) Wolf D, Lubk A, Röder F, Lichte H (2013) Electron holographic
Electron holography of thin amorphous carbon films: mea- tomography. Current Opinion Solid State Mater Sci 17:126–
surement of the mean inner potential and a thickness-inde- 134
Part I

pendent phase shift. Ultramicroscopy 106:341 Hÿtch MJ, Houdellier F, Hüe F, Snoeck E (2008) Nanoscale
holographic interferometry for strain measurements in elec-
tronic devices. Nature 453:1086
Electron Holographic Tomography and Dark-Field Hÿtch MJ, Houdellier F, Hüe F, Snoeck E (2011) Dark-field elec-
tron holography for the measurement of geometric phase. Ul-
Electron Holography tramicroscopy 111:1328–1337
Twitchett-Harrison AC, Yates TJV, Dunin-Borkowski RE, Midg-
ley PA (2008) Quantitative electron holographic tomography
for the 3D characterisation of semiconductor device struc-
tures. Ultramicroscopy 108:1401–1407
Chapter 8

Licensed to ([email protected])
Focal-Series Reconstruction
9

Part I
Chapter Preview

It is exciting to obtain atomically sharp images of materials modifications but needs a series of images taken at different
by high-resolution electron microscopy (HRTEM), but it defocus values of the objective lens; this series then serves
can be frustrating, or at least time consuming, to interpret as input for a numerical inversion of the imaging process.
such images after you’ve acquired them. It’s like a kalei-
doscope! You can obtain an amazing variety of different Since it is not possible to understand the inversion of im-
images from one and the same object or structure, and the aging without having understood the nature of the imaging
question becomes “which is now the right image?” The rea- process itself, in the following we start completely from
son for this difficulty is the fact that the HRTEM contrast scratch and revisit the imaging theory step by step in a more
depends simultaneously on the properties of the object and fundamental way than was done in W&C. The drawback to
on the imaging properties of the microscope, which together this approach is that there is a rather steep learning curve;

Chapter 9
create an uncomfortably large number of combinatorial pos- on the other hand, the big reward is that new light is shed
sibilities for obtaining different output. on what we thought we already understood! So, we gain
a different and perhaps deeper perspective on the imaging
A successful strategy to cope with this unsatisfactory situa- process. This new perspective of the theory can be nicely vi-
tion is to determine the states of the electrons just after they sualized in a graphical way. Then the fundamental principle
have passed the object and before they enter the imaging behind FSR becomes almost intuitively clear!
system of the microscope. The state of each electron, which
is ideally free of any microscope influences, is called the ob- Of course, all the important, practical requirements of FSR
ject wave function and will be the ‘object of our desire’ in this are addressed in detail so that you, the reader, should have
chapter. The wavefunction can be retrieved either by electron no difficulties in performing a successful focal-series ex-
holography or by focal-series reconstruction (FSR). Holog- periment after reading this chapter. Finally, the power of
raphy requires a dedicated experimental setup and allows the FSR method is demonstrated with a few experimental
thereby an almost direct recording and retrieval of the wave- examples. You will find terms like ‘coherent’ used in this
function, as we saw in Chap. 8. FSR requires no hardware chapter so don’t forget Chap. 7!

© Springer International Publishing Switzerland 2016

C. B. Carter, D. B. Williams (Eds.), Transmission Electron Microscopy, DOI 10.1007/978-3-319-26651-0_9 233
 234 9 Focal-Series Reconstruction

9.1 Motivation: Why the Effort? Summarizing the above considerations we have a rather undesir-
able situation! What we see in an image depends on the setting
of the objective lens defocus, and – in fact, we will see later on
Part I

You’ve gone through Chaps. 6, 28 and 30 of W&C so you know – on many optical properties of the instrument. Because of the
that high-resolution TEM is essentially an atomic-scale imaging two mechanisms of information loss and information encoding,
technique composed of two fundamental physical processes the resulting image contains a mixture of information about i)
the object and ii) our TEM. Regarding the image contrast, we
zz Electron diffraction inside the object are therefore observing an entanglement of object properties and
zz The electron optical imaging process – due to the electromag- instrumental properties. The two classical remedies to this entan-
netic lenses of the microscope. glement problem are either to use instrumental and experimental
optimization, or to use image simulation techniques. While such
Electron diffraction inside the object is governed by quite com- solutions are possible, there are also fundamental limitations that
plex quantum mechanical principles. In the present discussion, we’ll now discuss.
we’ll treat the electron diffraction part only in passing and then
just use approximations; we can choose experimental conditions One very successful way to reduce the entanglement problem is
that satisfy these approximations quite well and thereby mini- to optimize the experimental setup! Optimized imaging condi-
mize complications. tions, and even special optical hardware elements, are employed
to obtain an optimum image, without specifying at this stage
The core of the FSR technique is, however, intrinsically linked exactly what ‘optimum’ means. While the optimization of the
with the second physical process, the optical imaging. In a wider imaging conditions and the hardware are almost mandatory in
sense, FSR can be seen as a numerical inversion of this opti- many situations, there are still noticeable limits observed with
cal-imaging process, and we will thus focus strongly on the im- this approach. It is not always possible to assure that we have
aging aspect throughout the chapter. such an optimized imaging state, because of fundamental phys-
Chapter 9

ical limitations, because of limits in hardware construction, and

The technique of FSR itself consists of two parts: an experi- because of instrumentation instabilities. The phenomenon of
mental one and a computational one. Within the experimental information loss introduced above cannot be overcome using a
part, a series of atomically resolved images is acquired with the single image, whether it is optimized or not.
CCD (charge-coupled device) camera of the TEM. The images
of this series are taken from exactly the same object area with Another approach to cope with the ‘entanglement’ problem is to
the help of an automated computer procedure – each image has use the image-simulation technique. We use a guessed atomic
a different value of the objective lens defocus. The images in model of the object and frequently enough also a guessed setting
the series, which can consist of as few as 2 or as many as 20 or of the microscope optics! Then, we simulate the images using a
30 images, exhibit a different image contrast; the imaging con- computer. Depending on the match between simulated and ex-
ditions are changed by changing the defocus from one image to perimental images, we say we have verified the guessed atomic
the next. In the second, purely computational step of the FSR model or we discard it and try again. From a practical point of
technique, all the acquired images are finally processed by a view this approach is limited to simple atomic configurations,
computer and brought to a common representation in the form where we have a-priori knowledge. When the structure of the
of a quantum-mechanical wavefunction. This wavefunction is object is complicated, with perhaps hundreds of unknown atomic
more general and more complete than each of the single images. positions, we may not be able to develop a sufficiently reasonable
The general principle of the FSR technique is symbolically out- initial guess to the real object structure.
lined in Fig. 9.1.

What is the reason for doing so much work? A clue to the answer Apart from the practical limitations, there is also an
can be already obtained by analyzing the possible reasons for unfortunate conceptual weakness associated with the
such differences in the images. In principle, two images of the simulation approach. The conceptual dissatisfaction
same object will be different for two reasons: with image simulation is due to the fact that we have to
employ a lot of theoretical and computational skills just
zz Either we are capturing genuinely different information about for modeling a certain specific microscope, thus produc-
the object in each image, meaning that the potentially avail- ing instrument and situation-dependent results that lack
able object information cannot be acquired by a single image any generality. It is not in the best interest of a materials
alone. scientist to have images where the object information is
zz Or the contained object information is the same in all images, scrambled with information about the microscope and
but is encoded each time in a different way! whatever instrument settings were being used at the mo-
ment of data acquisition! 9
The first reason would be considered as information loss; the
second one is information encoding. As we will see later on, both
phenomena – information loss and information encoding – occur From the above considerations it is clear that we need to com-
whenever we form images: both hamper image interpretation. bine the information contained in different single images and to
 9.2 Quick Walk Through Electron Diffraction 235

the electron wavefunctions that you might encounter. Therefore

the next section is a short ‘focused’ walk through electron dif-
fraction as we’ll use it.

Part I
9.2 Quick Walk Through Electron
Diffraction
The object wavefunction is the result of electron diffraction
Fig. 9.1 General principle of the FSR technique. A series of high-resolution im- inside the object. The object wavefunction plays a central role
ages is recorded by an automated procedure from the same object area using as an intermediate state in ‘forward’ high-resolution imaging
different settings of the objective lens defocus. The quantum mechanical ex- on the one hand and is also the outcome of FSR as a kind of
it-plane wavefunction, which consists of amplitude and phase, can be retrieved reverse imaging technique on the other hand. Therefore it is
from the series by means of numerical procedures useful to become familiar with this quantum mechanical con-
cept. Unfortunately, electron diffraction is a quite extensive and
display this combined information in a way that does not depend mathematically demanding topic, and it would not be appro-
on the particular experimental circumstances. In FSR we thus priate to treat it exhaustively within the framework of FSR.
seek a representation of the information contained by a complete Electron diffraction is a very fundamental topic in its own right.
focal series, which is as independent as possible from the obser- In this chapter we’ll treat electron diffraction with a hands-on
vation instrument. We can achieve the desired disentanglement approach mainly helped by approximations that are valid only
between object and instrument properties by a kind of inversion for very thin objects.

Chapter 9
of the imaging process, since this imaging process is the reason
we obtain different images from the same object area. As we The interaction of the electrons with the atoms inside the scat-
said earlier, the diffraction process, which takes place before tering object is very strong compared to X-ray or neutron scat-
the actual electron optical imaging comes into play, is guided tering. (Remember Chap. 5.) Because of this strong interaction
exclusively by the electron–object interaction. It is therefore not we usually can’t describe the interaction as a single scattering
surprising that the outcome of the electron diffraction, which process. An electron that has undergone an elastic scattering
is the electron wavefunction at the exit plane of the object, is event will in general not immediately leave the object, but will
a suitable representation for the data collected within a focal be re-scattered by many other atoms before it leaves the ob-
series. The electron wavefunction at the exit plane of the object ject, leading to a multiple scattering phenomenon. Moreover,
(the object wavefunction or exit-plane wavefunction) carries all a classical particle-optical explanation is not really appropriate
information about the electron–object interaction and is ideally for describing elastic electron scattering. We have to switch to
free from instrumental artifacts. a wave-mechanical formalism in order to describe things cor-
rectly, starting with an incoming plane wave illuminating the
The Object Wavefunction object. Instead of using the expression ‘scattering’, which is a
particle-optical view, we prefer the expression ‘diffraction’: it is
better suited to describe the interaction of the incoming electron
The object wavefunction represents uniquely the com- wave with the atoms inside the object. As you know from W&C,
plete object-related information behind any image. Pro- the phenomenon of electron diffraction is correctly described by
ducing different images by varying optical properties of solving the relativistically-corrected Schrödinger equation for an
the microscope does nothing else than to reveal different electron wavefunction  .r/ in a periodic crystal potential V.r/.
portions of the object wavefunction in an always-differ- This analytical approach is called the Bethe–Bloch formalism.
ent manner. 9 The application of this formalism is only viable for small peri-
odic crystal patches and is not well suited for larger non-peri-
odic objects. In practice, a numerical approximation called the
So our aim is to retrieve the object wavefunction from a series of ‘multislice algorithm’ is used instead, which converges to the
images, since it contains the information that actually generated analytical Bethe–Bloch solution.
all the possible images and is thus the most complete and most
general representation of a high-resolution experiment. Warning

As you know from W&C, the object wavefunction is a mathe-

matically complex-valued function and has both amplitude and The view of electron diffraction developed here is very
phase; alternatively we can say it has a real part and an imaginary useful for intuitive comprehension, yet shortened and
part. In order for you to interpret the outcome of FSR and also simplified from a scientific point of view. Caveat emptor.
to understand the principles behind the reconstruction technique,  9
you’ll have to be somewhat familiar with the basic properties of
 236 9 Focal-Series Reconstruction

electron wave is still small when the object thickness is small;

i.e., ˚.r; t/  1. This assumption leads directly to the so-called
weak-phase-object approximation (WPOA):
Part I

 .r; t/  1 + i˚.r; t/ = 1 + iVP .r/t (9.2)

The popularity of the WPOA arises from the fact that a straight­
forward interpretation of the wavefunction is still possible. While
the real part represents the undiffracted incoming beam, the imag-
inary part is directly proportional to the projected crystal potential
and thus exhibits sharp maxima at the position of atomic columns.
(By ‘undiffracted’ we mean non-diffracted, but no one says that!)
If we can in some way image the imaginary part of the wave-
function we would have a direct observation of the structure of
the atomic object. We will return to the WPOA in the following
sections, when we discuss imaging the wavefunction.

Fig. 9.2 Symbolic illustration of the phase-object approximation. a The wave- Having derived such a convenient expression for electron scat-
length of an electron wave is shortened when traversing an atom. b Lines of tering it is crucially important that we are aware of the short-
equal phase belonging to the electron wavefunction. Be careful: contrary to in- comings and the limits of the POA/WPOA formalisms. Unfor-
tuition, the phase is at maximum at the atomic position, while the lines of equal tunately, these approximations are almost never applicable in a
phase drawn here fall behind the vacuum reference strict sense for realistic materials science specimens! You can
already predict the breakdown of these approximations within
Chapter 9

the simple theoretical framework set up here. From the WPOA

In this section we start by deriving an approximation for the of Eq. 9.2 we can see that the electron intensity    would
wavefunction  , even though the approximation only holds for grow in an unlimited way with the object thickness t; this is a
very thin objects. We start with an incoming electron plane-wave clear violation of the conservation of particles! Also, the POA of
at the entrance surface of the object; so the traversed object thick- Eq. 9.1, which is the better approximation of the two discussed,
ness t is still zero. Thus at the entrance surface, we have for the reveals its weakness directly. The electron intensity    is ex-
electron wavefunction  .r; t = 0/ = 1. Since the atoms inside the actly 1 and remains constant for all specimen thickness values,
object constitute an optically thicker medium for the electrons thus showing no observable modulation. So, what went wrong?
than the surrounding vacuum,Diffraction,
the wavelengthImaging, and
of the incoming The answer is that the propagation of the electron wave through
electron wave is shortened at the atom location. This behavior is the vacuum between the atomic planes has been neglected. The
shown symbolically in Fig. 9.2a. The phase of the electron wave symbolic drawing of Fig. 9.2b is wrong because the wave fronts
thus proceeds faster inside the atom, leading finally to a mod- behind the atom at the bottom of the figure have been drawn
ulation of the phase of the electron wave. The resulting phase statically. In reality, each point of the wave front is an emitter
modulation is also symbolically shown in Fig. 9.2b, where lines of secondary spherical waves, which have to be superimposed
of equal phase are drawn. onto each other, leading additionally to a lateral propagation of
the wave front on its way to the next atomic plane; this is the
Translating the phenomenological view of Fig. 9.2 into a math- well-known Huygens–Fresnel principle.
ematical expression, we obtain the so-called phase-object ap-
proximation (POA), for the electron wavefunction  .r; t/ at the The propagation of the electron wave in the vacuum, which is
object thickness, t, missing in our formalism, is included in the multislice algorithm
that we mentioned at the beginning of this section. Within the
 .r; t/  exp Œi˚.r; t/ = exp Œi VP .r/t (9.1) framework of the multislice algorithm the object is subdivided
into slices, corresponding to atomic planes perpendicular to the
Here Φ is the phase of the wavefunction, σ is the interaction incident electron beam. While the diffraction of the electrons
constant, and VP is the projected crystal potential per unit length. by a single slice is still described using the POA formalism pre-
In the POA formalism the distinct succession of individual atoms sented here, an additional propagation step is executed by the
separated by vacuum along the incident beam direction has been multislice algorithm between any two successive slices. This
abandoned and is replaced by a projected potential that is contin- additional propagation step allows for a lateral redistribution of
uous and constant along the direction of the incident beam. The electrons, leading to an intensity modulation in a plane perpen-
projected potential is strongly peaked at the position of atomic dicular to the incident beam, which cannot be reproduced by the
columns and is defined in the units of Volt/nm. POA formalism.

Continuing with the approximate description of electron diffrac- Finally we will roughly outline what the electron wavefunction
tion, we can further assume that the phase modulation of the looks like for realistic specimen thickness values. This outline
 9.3 From the Wavefunction to the Image 237

is made using a phenomenological description of the results ob-

tained by the so-called channeling theory. The channeling theory
is again an approximation to the full Bethe–Bloch formalism. In

Part I
channeling, the Schrödinger equation is solved for each single
atomic column individually, so it neglects multiple scattering
between neighboring columns.

According to channeling theory the electrons move with a sort

of pendulum motion along atomic columns. So this special phe-
nomenon of electron diffraction is called Pendellösung behavior
(Pendellösung means ‘pendulum solution’) and is depicted in
Fig. 9.3. The repetition period of the electron motion along the
thickness axis is called the extinction length ξ, as you know from
Fig. 9.3 Electron trajectories according to the channeling theory. The electrons
W&C. At a thickness ξ/2 the electron intensity is focused on the
move in a pendulum-like path along the atomic columns. Two types of atomic
column position, while at a thickness ξ the electron intensity columns are drawn: type A consists of light atoms, type B of heavy atoms. The
corresponds again to that of the incoming plane wave, i.e., the repetition period of the electron motion along the thickness axis is called extinc-
atomic column is extinct again. The extinction length depends tion length ξ. At a thickness ξ/2 the electron intensity is focused on the column
sensitively on the charge number of the atoms inside a column. position, while at a thickness ξ the electron intensity corresponds again to the
The higher the atomic charge number (i.e., the heavier the atom) uniform intensity of the incoming wave. In this example note the special thick-
is, the smaller is the extinction length and the faster the pendu- ness value ts where the type A columns are visible, whereas the type B columns
lum swings along the incident beam direction. It is nicely shown are completely invisible
in Fig. 9.3 that thickness values may exist where a column of an
atomic species A can show up in the electron intensity, whereas

Chapter 9
a column of a different species B can be totally extinct. tential ‘string’ of the atomic columns. Typical extinction lengths,
ξ, are on the order of 10–50 nm, depending on the atom species
and on the interatomic distance along the column direction. Since
Practical hint: high-resolution experiments for FSR are the POA of Eq. 9.1 does not yield any intensity modulation along
best performed where the object is thin, around or be- the direction of the atomic columns, it is clear that the validity
low 5 nm, or more precisely below ξ/4 of the heaviest of the POA/WPOA formalism holds only up to a thickness of at
involved atom column. Advantages are: maximum ξ/4, which corresponds in most cases to only a few nm.

(1) The POA/WPOA help at least to understand intui-

tively the observed wavefunction features. Phase maxima
occur at atomic column positions; the phase then quite 9.3 From the Wavefunction
closely relates to the underlying atomic structure accord-
ing to a simple projection principle. to the Image
(2) When columns of different atomic species are pres- In order to understand FSR, which we can think of as a sort of
ent, confusion due to extinction and non-extinction of inversion of the imaging process, it quite important to be clear on
the different columns can be avoided. Since all columns what we mean by ‘imaging’. Otherwise it is hard to understand
are in the same common Pendellösung regime where the what precisely the inversion procedure achieves and how this
phase increases monotonously with thickness and atomic is done. Some properties of imaging are determined already by
number, a distinction between atomic species is possi- the fundamentals of nature and are completely independent of
ble. The higher the phase peak, the heavier the respective the apparatus used for observation. Other properties are more
atom species. instrument specific. Some textbooks put much emphasis on the
latter aspects, i.e., on the more instrument-specific details. The
(3) Object misorientation is a big practical issue in HR- instrument can be shown, an analog to light optics can be drawn,
TEM, whether or not you intend to use image reconstruc- ray paths can be displayed, lens groups can be discussed, and
tion. The accuracy requirement for object orientation in- performance parameters, such as resolution, can be presented.
creases approximately linearly with sample thickness. A
thinner object area offers thus offers a better chance for The traditional approach of starting with the more evident and
controlled HRTEM experiments. 9 obvious is certainly wise and also necessary. On the other hand,
there is also a danger that fundamentals are incorrectly taken
for granted, because we know similar things from everyday ex-
The Pendellösung behavior for thicker objects reflects the strong perience. As a consequence, such fundamentals might be over-
Coulomb interaction between the positively charged atom cores looked and wrongly mixed with instrumental properties. Since
and the negatively charged electrons, which are trapped by the po- the core idea of FSR is very general and almost independent of
 238 9 Focal-Series Reconstruction

particular instrumental properties, and since we already have First, explanations are facilitated by talking about wave vectors
lots of traditional approaches to explain HRTEM imaging, we in terms of ‘beams’ reminding us of crystal reflections, and sec-
follow here a different route. In order to highlight the basic idea ond, the final numerical calculations will actually be performed
Part I

behind FSR, or more generally behind phase retrieval methods in a discrete form. (That’s what computers do.)
using parameter variation, we highlight first some general fun-
damentals of coherent imaging. The advantage of discussing On the Nomenclature for Reciprocal Vectors
the forward-imaging procedure in some detail first is that, once
this process is understood, its inversion process is almost self-­
evident! Whereas in the W&C textbook a distinction is made
between reciprocal vectors inside the crystal (symbol g
used for diffraction) and reciprocal vectors in the optical
system (symbol u used for imaging), we do not make
9.3.1 Imaging with a ‘Neutral’ Microscope a distinction here. We use the symbol g continuously
throughout diffraction and imaging for the following
reasons. First, a sudden artificial nomenclature change
We assume now a ‘neutral’ microscope, which does nothing applied to the wavefunction at the exit surface, which
other than magnify and image the object wavefunction (OWF) marks the transition between diffraction and imaging,
 .r/, without any further alteration. We can be even more radical might be confusing! Second, notations using g and k are
and abandon the idea of using a microscope at all. Neglecting a widely accepted standard in the optical literature, which
the need for magnification, which is not at all important from might facilitate further reading. 9
a theoretical point of view (☺), we could insert a hypotheti-
cal observation screen directly at the exit surface of the object.
The image intensity I.r/ observed there is given by the modulus The transformation of the real-space image intensity I.r/ of
Chapter 9

square of the OWF. Eq. 9.3 to Fourier space yields the somewhat complicated look-
ing expression for the discrete Fourier transform I.g/, with
I.r/ =  .r/  .r/ (9.3)
X
I.g/ =  .g + k/  .k/: (9.4)
By writing the OWF in the amplitude-phase notation k
 .r/ = A.r/ exp.i˚.r//, an intrinsic shortcoming of the imaging
process becomes obvious. Because I.r/ = A2 .r/, the real-space This equation represents an autocorrelation in Fourier space, as
phase information represented by ˚.r/ is completely lost. Taking we’ll now explain. Eq. 9.4 describes a summation over the pair-
the POA wavefunction from Eq. 9.1, we find I.r/ = 1, meaning wise interferences between two beams with wave vectors .g + k/
that the image contains absolutely no information about the ob- and k. The clue to understanding Eq. 9.4 is the special sorting
ject! In order to be able to observe any useful signal from a thin scheme defined by the argument vectors. By always choosing
object, the EPW must therefore be modified in some way before two beams with the special wave vector combination g + k and
taking the modulus square. This modification of the EPW is more k, we are assured that the resulting interference term contributes
or less intentionally achieved by the special transfer properties to the desired image Fourier coefficient I(g). This is because the
of a real microscope and will be treated later on. complex conjugate of one of the two factors of the sum leads
to a subtraction of the two respective argument vectors g + k
The Magnification and k which yields the image vector g. In order to collect all
possible two-beam interference terms contributing to a fixed
image Fourier coefficient I(g), we have to sum over all possible
Note that the magnification is set to 1 throughout the text vectors k. With N diffracted beams in the object wavefunction,
 9 the full image calculation for all image coefficients I(g) requires,
in total, the calculation and sorting of all possible N 2 pairwise
interference terms between any two beams of the object wave-
It is particularly important that we also discuss the consequences function.
of the collapse of the wavefunction as described by Eq. 9.3 in
Fourier space. In the following we choose the wave vector (or The Fourier-space formulation of the image intensity given by
diffraction vector) g as the conjugate coordinate to the real-space Eq. 9.4 offers far-reaching insights, which are hidden in the re-
vector r; as usual, the modulus g = jgj is a spatial frequency. al-space formulation of Eq. 9.3. Let’s assume a 1D object spec-
Furthermore we have   g, where θ is the diffraction angle trum limited by a hypothetical aperture at a spatial frequency
and λ the electron wavelength. Thus the larger the diffraction gmax. Now we choose the two opposite beams right at the border
angle θ, the larger the respective spatial frequency g, and the of the aperture with the respective wave vectors gmax and −gmax
smaller the size d = 1/g of the related object detail. Although the and let them interfere. According to Eq. 9.4 we assign the first
‘continuous’ notation of the Fourier transform is actually correct, beam gmax to g + k and the second beam −gmax to k. According to
here we deliberately choose a discrete notation for two reasons. Eq. 9.4 the interference of these two beams leads, by subtraction
 9.3 From the Wavefunction to the Image 239

of the arguments gmax and −gmax, to an image frequency of g = 2

gmax. This is very surprising, since we obtain in the image spec-
trum a spatial frequency of 2 gmax, which does not belong to the

Part I
initial object spectrum limited by an aperture at gmax. The image
spectrum thus extends twice as far in Fourier space than does the
object spectrum, resulting in contrast features half as small as the
smallest object detail! This phenomenon is called ‘half-spacing
contrast’ and is a consequence of the non-linearity of the imaging
process. It is therefore, in general, not correct to determine the
resolution of a microscope just by measuring the extent of an
image Fourier transform. Keep in mind that this ‘fake’ resolution
has nothing to do with specific properties of the microscope, but
Fig. 9.4 Experiment for the demonstration of nonlinear contrast in HRTEM.
follows from fundamental principles.
a bright-field image of BaTiO3 taken along the [110] zone axis with a CS-correct-
ed Tecnai F20 microscope. b Dark-field image from the same object area taken
In order to clarify the sorting scheme of Eq. 9.4 in more detail with a special objective aperture blanking out the central beam. All observed
we’ll discuss now a concrete example. Let’s assume a discrete contrast features are thus due to nonlinear interference. The diffraction pattern
periodic object spectrum composed of a set of beams with wave with the special aperture is shown in the inset. No nonlinear contrast is seen in
vectors {−3g, −2g, −g, 0, g, 2g, 3g}. In the resulting image spec- the vacuum (no diffraction!) and at the amorphous sample edge. Weak nonlinear
trum we will find coefficients with argument vectors {−6g, −5g, contrast features appear in the thin crystalline area next to the amorphous edge,
−4g, … 0, … 4g, 5g, 6g}. From these 13 coefficients of the image while they are prominently visible in the thicker sample regions
Fourier spectrum we take out, e.g., the coefficient belonging to
the frequency 4g. The image Fourier coefficient at 4g carries
contributions from the following set of beam interference pairs: formation contained in the EPW is lost when we form the image.

Chapter 9
{(3g,−g), (2g,−2g), (g,−3g)}. Note that the occurrence of the After Fourier transformation and linearization of the image in-
frequency 4g in the image spectrum is a consequence of non- tensity, we are actually expressing the loss of phase information
linearity; this frequency does NOT actually belong to the object in a different way by Eq. 9.5. Since the image intensity must be
spectrum! a real-value quantity, the Fourier transform of the image intensity
must exhibit the so-called Friedel symmetry; i.e., I.−g/ = I  .g/,
At the present stage, the nonlinear theory is left behind as soon where the asterisk denotes complex conjugation. This symmetry
as possible by picking out from Eq. 9.4 only such pairwise in- is implemented in Eq. 9.5 by the superposition of the information
terference terms, where the undiffracted beam is involved and carried by a beam with wave vector g with that carried by a beam
neglecting all interference terms between two diffracted beams. with wave vector −g. The information carried by two opposite
This restriction to the so-called linear-interference terms, which beams becomes indiscernible as a consequence of the collapse
include the ‘undiffracted’ beam, is justified for thin objects obey- of the wavefunction.
ing the WPOA. We assumed in deriving Eq. 9.3 that the con-
stant real part of the EPW, which corresponds to the undiffracted Linear and Nonlinear
beam, is much larger than the imaginary part, which is formed by
the diffracted beams. Pairwise interferences between diffracted
beams are thus significantly smaller (of second order) compared We just are just about to leave nonlinear theory. The ex-
to the first-order interferences involving the ‘undiffracted’ beam. periment of Fig. 9.4 gives direct evidence of the impor-
When neglecting nonlinear interferences, which play a role in tance of nonlinear interference terms. However, keeping
thicker object areas, we obtain a simple expression for the linear track of them at this point does not add very much to our
image intensity IL basic comprehension. 9

IL .g ¤ 0/ =   .0/ .g/ +  .0/  .−g/ (9.5)

This degeneracy in Fourier space has its direct counterpart in
From a combinatorial point of view, linearizing the image in- real space. Setting the undiffracted wave  .0/ = 1 in Eq. 9.5
tensity means an enormous reduction of the number of pairwise and using a property of the Fourier transform (FT), according
interference terms contributing to the image, from initially N 2 to to which FT Œ  .r/ =   .−g/, we find that the real-space lin-
only 2.N − 1/ involving the ‘undiffracted’ beam. The manifold ear image intensity corresponds to a superposition of the wave-
superposition of interference terms per image Fourier coefficient function  .r/ and its complex conjugate   .r/, which yields
according to Eq. 9.4 is reduced in the linear formalism to a just twice the real part of the wavefunction. In real space, the col-
twofold superposition, involving pairwise opposite beams in lapse of the wavefunction thus corresponds to a projection of
Fourier space. the complex numbers representing the wavefunction onto the
real axis of the complex-number plane.
We have already stated, by the use of the nonlinear real-space
formulation of the image intensity of Eq. 9.3, that the phase in-
 240 9 Focal-Series Reconstruction

The Collapsing Wavefunction Since a previous phase shift does not change the fundamental prop-
erties of a wavefunction upon its collapse, all basic mechanisms
discussed in the context of our ‘neutral’ microscope remain valid.
Part I

There are two equivalent views of the collapse of the The linear image intensity is thus twice the real part of the relevant
wavefunction within the linear formalism. In real space wavefunction, which is now the image-plane wavefunction, thus
linear image formation means a superposition of a wave-
function Ψ(r) and its complex conjugate Ψ*(r). Since one IL .r/ = 2 Re.I .r// = 2 Re.O .r// cos  + 2 Im.O .r// sin 
observes as a result only the real part of Ψ(r), the imagi- (9.7)
nary part, and thus the real-space phase, is lost.
In Fourier space linear image formation means a super- Historically, HRTEM theory has concentrated on acquiring the
position of a wavefunction Ψ(g) and its spatially point-in- imaginary part of the object wavefunction ΨO, since, at least
verted conjugate Ψ*(–g). Thereby the ability to discern when following the WPOA of Eq. 9.2, all the object information
between any two opposite diffraction directions g and –g is resident in the imaginary part of the object wavefunction ΨO.
is lost. 9 Such a strategy would require us to set the phase angle χ to an
odd multiple of π/2. Unfortunately, there is so far no such device
available in high-resolution TEM, which allows for an ideal rigid
The real-space result can, of course, be obtained with- phase shift of the complete diffracted part of object wavefunction
out the complicated detour to Fourier space. Composing with satisfactory quality. Such a device exists in light optics in
a wavefunction 0 .r/ of its mean value 1 (the ‘dc’ part) and the form of a Zernike phase plate. Attempts to construct such
its modulated part  .r/, i.e., 0 .r/ = 1 +  .r/ , we obtain a device for electron microscopy are still ongoing. Besides the
I.r/ = 0 .r/0 .r/ = 1 + 2ReΠ.r/ +  .r/0 .r/ for the image obvious advantage for directly imaging the imaginary part of the
intensity. When we neglect the second-order term  .r/0 .r/ on object wavefunction, such a device would also be beneficial for
the right-hand side we find that the resulting linear image inten- the reconstruction of the object wavefunction in a linear frame-
Chapter 9

sity is given by IL .r/ = I.r/ − 1 = 2 ReŒ .r/, which conforms work. Setting χ to 0 would allow us to record the real part of the
to Eq. 9.5 and is again twice the real part of the wavefunction. object wave function, setting χ to π/2 the imaginary part, and then
the full wavefunction would be almost directly available, at least
from the point of view of linear theory.

9.3.2 Linear Imaging with a Constant-Phase- A Constant-Phase-Shift TEM

Shift Microscope
Within linear theory, a constant-phase-shift microscope
From linear imaging theory it follows thus that a ‘neutral’ mi- would allow for a tomographic recording of the object
croscope is perfect for capturing the real part of a wavefunc- wavefunction in the 2D complex-number plane. 9
tion in the final image plane. In order to re-establish the full
wavefunction we would also need a second projection within
the complex number plane, which would give us the missing We can interpret Eq. 9.7 in two different ways. So far we have
imaginary part. Having captured real and imaginary parts by two usually stated that the diffracted part of the object wavefunction
separate exposures, the full wavefunction could then be re-estab- is first rotated in the complex number plane according to Eq. 9.6
lished. However, from the above considerations it is clear that by an angle −χ to become the image plane wavefunction. The ac-
once the wavefunction is formed in the image plane, the collapse tual imaging process was described in a second step as a projec-
of the wavefunction will always yield its real part, regardless tion of the resulting image plane wavefunction onto the real axis
of the wavefunction’s special properties or previous history. A of the complex number plane. Alternatively, we can describe the
straightforward way to obtain a second projection would be to linear imaging process directly in one single step as the projec-
exchange the real and imaginary parts in some way before the tion of the initial object-plane wavefunction onto a rotated axis
collapse takes place in the image plane. It is important to note in the complex number plane, which is no longer the real axis,
that such an exchange concerns only the diffracted part of the but is now an axis rotated by an angle χ away from the real axis.
wavefunction, whereas the undiffracted part (central beam, dc-
value) still remains real and equal to 1. Such an exchange be- This alternative view is possible since it does not matter, whether
tween real and imaginary part of the initial object wavefunction the diffracted part of the wavefunction is rotated towards the
can be mathematically described as a phase shift over an angle −χ projection axis by an angle −χ, or the projection axis is rotated
applied to the object wavefunction ΨO, which leads to an image towards the wavefunction by an opposite angle χ. Only the rel-
plane wavefunction ΨI, which is finally imaged. The image plane ative rotation counts and in both cases the projection gives the
wavefunction is then given by same result. Following the alternative description, taking linear
images using different projection axes defined by χ, produces
I .r/ = O .r/ exp.−i/ (9.6) a direct tomographic recording of the complex-valued object
wavefunction in the complex number plane.
 9.3 From the Wavefunction to the Image 241

This is like traditional tomography techniques, where an object limit. The information limit is a very general criterion and is
is recorded from different viewing directions: the object wave- defined as the size or length di of the smallest object detail that
function is now recorded from different viewing angles χ in the can be transferred in a linear way by a microscope. In other

Part I
complex-number plane. In both cases, restoring the initial object words, the information limit defines the highest spatial frequency
means re-assembling the different views in order to obtain the gmax = 1/di of the object spectrum that can be linearly imaged. Just
full object, which has one dimension more than the input views. be aware that the contrast transfer will not actually end abruptly
In traditional tomography, this additional dimension is the third at gmax. It is actually described in the full temporal coherence
spatial dimension; in the present context this is the second di- theory by a continuous damping function, which is equal to 1 at
mension of the 2D complex-number plane. low spatial frequencies g (no dampening for large object details)
and falls below a defined threshold value of 1/e2 = 0.13 at a spa-
Unfortunately the imaging properties of a real objective lens al- tial frequency gmax (strong dampening for small object details).
low for neither a straightforward setting of a constant phase shift So we arbitrarily assume that object information represented by
of 0, nor for one of π/2, as we will see in the following section. In spatial frequencies larger than gmax, which are damped by a factor
the following section we will deal with the particular properties smaller than 1/e2, can no longer be discriminated from noise. The
of real objective lenses and the resulting phase shifts that are information limit of modern intermediate-voltage electron mi-
actually available. Or expressed in the alternative tomographic croscopes with typical accelerating voltages of 200 or 300 kV is
view, we will show what projection directions are implied physi- roughly 0.1 nm. Such microscopes are thus able to image object
cally by a real objective lens. After clarifying the intrinsic nature details in the Å range or even below the 100 pm range.
of those projection directions, we will continue the tomographic
view gained here and apply it to a real microscope in the next- After reducing our discussion of the partial coherence theory to
but-one section. the above paragraph, we can say that the remaining fundamental
imaging properties of a TEM are linked to its objective lens.
For Clarity Unlike in the ‘neutral’ and constant-phase-shift microscopes

Chapter 9
discussed in the previous sections, a real objective lens can be
regarded as a device for modifying a dispersive wavefront. Alter-
Partial spatial and partial temporal coherence phenom- natively, since beams run perpendicular to the wavefront, we can
ena, which are important for quantitative real-world re- also regard the objective lens as a device for displacing beams
constructions, are not considered in this chapter. 9 relative to each other. The first view describing the wavefront is
closely related and also nicely links to the quantum mechanical
treatment of diffraction inside the object. The alternative view
describing beam paths is more related to classical geometrical
9.3.3 Linear Imaging with a Real Microscope optics, where a deterministic view of isolated light or particle
trajectories is considered sufficient to treat the imaging process.
Although the classical approach can never reproduce mutual
We are now going to deal with a real microscope that has distinct interference phenomena between possible beam paths, it does
physical properties. Apart from the accelerating voltage, which nevertheless offer some intuitively useful insights at a stage be-
already entered our previous considerations via the wavelength fore interference takes actually place. We will thus keep both
λ, the most important contrast transfer properties are linked to complementary views in mind, being always aware that only the
the objective lens. Moreover, coherence phenomena, which de- quantum-mechanical view is really correct and complete.
fine the resolution that is finally obtainable with the instrument,
play an important role. Understanding coherence phenomena, Figure 9.5 illustrates how a wavefront emerging from a point
which are described by the formalisms of partial spatial and source is modified by the primary properties of a lens. An ideal
partial temporal coherence, is quite demanding! (You saw this lens would convert an incoming, diverging, spherical wavefront
in Chap. 7.) The treatment of partial coherence within the frame- into an outgoing, converging, spherical wavefront, resulting in a
work of standard forward-imaging formalism would go beyond perfect, point-like image of a point-like object. Neglecting mag-
reasonable space limits in this chapter; a treatment of this topic nification, the ideal lens would behave thus like a perfect ‘wave­
in the reverse direction is even more exhausting! However, the front mirror’. However, as we learned in the previous section, a
basic ideas behind reconstruction are very general and luckily ‘neutral’ microscope equipped with such an ideal lens is of very
do not depend on coherence phenomena, except for the fact limited use for imaging a thin object.
that they have to be technically incorporated in order to obtain
quantitatively correct results! Since the treatment of coherence A fundamental property of a real objective lens is its spherical
phenomena would thus unnecessarily obscure the insight to the aberration. Due to spherical aberration, which cannot be pre-
main topic we will treat FSR mainly within a fully coherent vented for a single, electromagnetic, round lens, the wavefront
framework. behind the lens will be more strongly curved than it would be
behind an ideal lens. The actual wavefront, deformed due to
One of the most significant outcomes of the partial temporal spherical aberration, moves ahead of the ideal wavefront by
coherence theory is a resolution criterion called the information an amount s, which is proportional to the fourth power of the
 242 9 Focal-Series Reconstruction

The Nomenclature for Defocus

Part I

Deviating from W&C for the defocus we use the symbol

Z. We need a single symbol instead of the ∆f used in
W&C, since here the latter could be incorrectly read as a
product between ∆ and f! And don’t let this new ‘s’ (path
difference) confuse you. 9

A second fundamental property of an electromagnetic lens is

its focusing power. Unlike the spherical aberration, we can eas-
ily change the focusing power of a lens by altering the current
through the lens coils. The stronger the current, the shorter is the
focal length; the weaker the lens current, the longer is the focal
length. As you can see in Fig. 9.5, if we change the focusing
power (we defocus the lens), we change the shape of the wave-
front compared to the in-focus reference wave. When the lens is
underfocused by an amount Z, the point-like object would be im-
aged correctly in a plane located at a distance Z below the actual
image plane. The corresponding wavefront is thus less curved
than the ideal reference wave; we then say that the relative path
Fig. 9.5 Illustration of wavefront deformation and beam displacements occur- difference is negative for underfocusing. When the lens is over-
Chapter 9

ring upon spherical aberration, underfocusing and overfocusing with respect to focused by an amount Z, the point-like object would be imaged
the situation of an ideal lens. Neglecting magnification, an ideal lens acts as a correctly in a plane located at a distance Z above the actual im-
mirror for the spherical wavefront emitted by a point-like object. In the presence age plane, resulting in a stronger curved wavefront compared
of spherical aberration the resulting wavefront is more strongly curved than to the reference wave, and we say the relative path difference is
the spherical reference wave. For underfocusing, the wavefront is less strongly positive. The optical path difference of a diffracted beam with
curved, for overfocusing the wavefront is more strongly curved than the ideal respect to the undiffracted beam is given by s = 12 Z  2 = 12 Z 2 g2
reference wave. Diffraction angles are denoted by θ, the spherical aberration , which is again frequency dependent, and which  can bealso ex-
constant by CS , and the defocus value by Z. Optical path differences are denoted pressed as a phase difference Z with Z = 2  12 Z g2 . In this
by s, the sign of s is surrounded by circles. The lateral image-plane displacement formulation, the defocus value Z is negative for underfocusing
of beam paths, which run perpendicular to the wavefronts, is indicated by r. The and positive for overfocusing.
displacement direction is indicated by arrows. The ensemble of individual beam
displacements r leads to a broadening of the image of the ideal point-like object, The total phase shift caused by spherical aberration and defocusing
resulting in a so-called error disk. The minimum radius of the error disk caused together is given by the aberration function  = CS + Z, with
by spherical aberration is indicated by rmin  
1 1
.g/ = 2 Cs 3 g4 + Zg2 (9.8)
diffraction angle θ. For the path length difference s we can show 4 2
that s = 14 Cs  4 = 12 Cs 4 g4, where Cs is the spherical aberration
constant, λ is the relativistically-corrected, electron wavelength, The phase shift is no longer a constant for all spatial frequencies,
and g is the spatial frequency corresponding to a diffraction but shows a dispersive behavior with respect to the spatial fre-
angle θ. The path difference s (which is not the s you are used quency g. Due to the g4 and the g2 dependencies of the contribut-
to) can then be expressed in units of the wavelength λ, or by ing terms, a constant phase shift χ for all spatial frequencies is not
a further multiplication with 2π, even more conveniently in possible. Referring back to the previous section, this means that
terms of a phase difference χ, with χ = (2π/λ)s. In the presence of the projections describing the imaging process no longer occur
spherical aberration we therefore obtain a frequency-dependent on a flat plane in the complex number plane, but rather on a plane
phase difference
 between  the diffracted and undiffracted beams that is curved along the spatial frequency direction. We’ll return to
of Cs = 2 14 Cs 3 g4 . Note that we choose the sign conven- this full 2D view of the projection process in the complex number
tion such that the phase difference is positive for a wavefront, plane in the following section; here we’ll just introduce the main
which is more strongly curved than the ideal spherical reference technical aspects of contrast transfer in the traditional 1D view.
wave.
Since the phase shifts induced by the objective lens now depend
on the spatial frequency g, Eq. 9.6, which describes the transition
from the object wavefunction ΨO to the image-plane wavefunc-
tion ΨI, must now be re-formulated in Fourier space, leading to

I .g/ = O .g/ exp.−i.g// = O .g/t.g/ (9.9)

Licensed to ([email protected])
 9.3 From the Wavefunction to the Image 243

Fig. 9.6 Coherent contrast transfer

function (CTF) of the electron micro-
scope JEOL 4000EX for the defocus

Part I
values: a +25 nm, b 0 nm, c −25 nm,
d −50 nm. Parameters: CS = 1 mm,
accelerating voltage U = 400 kV. The
defocus setting (d) corresponds to
the Scherzer defocus. The abscissa
represents reciprocal distances, i.e.
object distances become smaller in
real space from left to right

Chapter 9
The aberration function is χ(g), and t(g) denotes the complex-val- azimuthally isotropic in the present context, it is sufficient to
ued phase-transfer function. Historically, the contrast transfer continue the discussion with spatial frequencies g instead of
has been calculated in 1D based on a weak-phase-object (WPO) wavevectors (or diffraction vectors) g. When the CTF is equal
wavefunction, where the real part does not show any modulation to 1 at a certain spatial frequency g, the real-space imaginary
and is thus no longer taken into account. The WPO wavefunction part of the wavefunction is transferred unaltered for this fre-
given by Eq. 9.2 is now consequently stated in Fourier space, quency. In the case that the CTF is equal to −1, the real-space
yielding imaginary part is inverted at this frequency, and when the CTF
amounts to zero, the real-space imaginary part related to the
O .g; t/  ı.g/ + iVP .g/t (9.10) frequency g is not imaged at all. In the latter case we say we
have a transfer gap.
Now we insert Eqs. 9.9 and 9.10 into the Fourier-space formu-
lation of the linear image intensity of Eq. 9.5 to show that, for a Figure 9.6 shows the CTFs for a 400 kV microscope (λ = 1.644 pm)
weak-phase object, the linear image intensity is given by with a spherical aberration constant of CS = 1 mm for the four
defocus values of +25, 0, −25 and −50 nm. You can see that for
IL .g ¤ 0/ = 2 VP .g/t sin..g// = 2 FT ŒIm.O .r// sin..g// low spatial frequencies below approximately 3/nm the defocus
(9.11) term is dominating, leading to a positive value of the CTF for the
overfocus value of Z = +25 nm and to a negative value of the CTF
FT denotes a Fourier transform, and Im denotes the imaginary for the underfocus values of Z = −25 and Z = −50 nm. For higher
part. Note that taking the Fourier transform and taking the imag- spatial frequencies the CTF generally exhibits a strong oscilla-
inary part are not interchangeable, and that the imaginary part tory behavior. The oscillations with the many contrast reversals
is taken in real space. A direct comparison of Eq. 9.11 with the and transfer gaps are very undesirable from the viewpoint of a
constant-phase-shift formulation of Eq. 9.7 makes the mecha- clean imaginary-part transfer. As we discussed in the context
nism of the frequency dispersion introduced by the aberration of an ideal constant-phase-shift microscope, an optimum imag-
function clear. Moreover, we can see from that comparison that inary-part transfer would require a CTF that is either +1 or −1
the real-part transfer has been abandoned because we assume for all spatial frequencies.
a WPO.
We can extend the region that exhibits no oscillatory behavior
The term sin(χ(g)) is called the coherent-contrast transfer func- outwards in Fourier space by balancing the effect of a constant
tion (CTF) (one ‘C’ because we usually forget to say ‘coher- and positive spherical aberration with an appropriate underfo-
ent’). The CTF indicates to what extent a Fourier coefficient cus value. It is then possible to keep the value of the aberration
with wavevector g of the real-space imaginary part of the object function χ(g) close to − π/2 for a relatively wide range of spatial
wavefunction is transferred into the image. Since the CTF is frequencies, and thus keep the value of the CTF close to −1. This
 244 9 Focal-Series Reconstruction

special defocus value is called Scherzer defocus and the corre- The PSF p(r) is the inverse Fourier transform of the phase trans-
sponding CTF is displayed in Fig. 9.6d. The spatial frequency fer function t(g) introduced in Eq. 9.9, thus p(r) = FT−1[t(g)].
gS where the Scherzer CTF changes its sign defines the point The transition of the object wavefunction to the image-plane
Part I

resolution dS = 1/gS of a microscope. Object details down to a wavefunction described by Eq. 9.9 in Fourier space can thus be
size of dS can thus be transferred in the imaginary-part mode, expressed alternatively in real space via the Fourier theorem as
as is required for a WPO. Since the Scherzer CTF is negative a convolution of the object wavefunction with the PSF, yielding
in the frequency interval [0, gS] atomic positions will appear as
dark dots in the final image because the crystal potential and the I .r/ = O .r/ ˝ p.r/ (9.15)
imaginary part of the wavefunction are always positive. Without
further derivation, the definitions of the Scherzer defocus ZS and The symbol ⊗ means convolution. The convolution of the ob-
the point resolution are given in the following ject wavefunction with the PSF describes in real space a lateral
re-distribution or a delocalization of information, resulting in a
blurring of the object wavefunction.
 1=2  −1=4
4 3
ZS = − Cs  I gS = Cs 3
3 16 (9.12)
For overfocus values in general, and also for moderate under-
A problem with contrast transfer arises in cases where the highest focus values, the largest beam displacement ∆r within the beam
transferred spatial frequency gmax, which defines the information bundle occurs at the maximum frequency gmax. In this case the
limit of a microscope, is larger than the Scherzer frequency gS. displacement is positive and directed towards the beam incli-
This is the case for most high-resolution instruments installed nation with respect to the optical axis. For large underfocus
today: they are equipped with highly coherent field-emission values, the largest modulus of beam displacements ∆r within
electron sources, allowing for contrast transfer significantly be- the beam bundle occurs at a beam with intermediate spatial fre-
yond the Scherzer frequency gS. For this purpose Fig. 9.7 does quency gint=(Z/(3Csλ2))1/2 < gmax. In this case the displacement ∆r
not only display the ‘coherent’ CTF, which extends in principle is negative and directed against the respective beam inclination
Chapter 9

to infinity in Fourier space, but also the ‘partially coherent’ CTF, with respect to the optical axis. Going from overfocus towards
which reflects the gradual termination of the contrast transfer underfocus, the radius R of the PSF adopts a minimum value at
due to partial coherence at the frequency gmax. From Fig. 9.7a, b the defocus where the two beams with frequencies gmax and gint
it is obvious that the cost of the flattening of the CTF below the exchange their role as outermost beams within the beam bundle.
Scherzer frequency gS is a strong oscillatory behavior of the CTF The explicit evaluation of this condition yields an optimum de-
in the frequency band above the Scherzer frequency between gS focus value Zopt, with
and gmax.
3
Zopt = − Cs 2 g2max (9.16)
Oscillations of the CTF are not only undesirable from the view­ 4
point of an optimum imaginary-part of the contrast transfer, but
have further consequences. Strong oscillations of the CTF reflect This Z is frequently called the Lichte defocus or the defocus
a strong change of the aberration function χ(g) with variation of of least confusion, a wonderful expression! Choosing the de-
the spatial frequency g. What this means mathematically is that focus value Zopt minimizes the radius R of the PSF and thereby
the strong oscillations originate from a significant gradient of minimizes the blurring of the object wavefunction and of the
the aberration function. The gradient of the aberration function final image. With this defocus the minimum diameter of a beam
∇χ(g) is linked to the beam displacements indicated by arrows bundle affected by spherical aberration alone (rmin in Fig. 9.5)
in Fig. 9.5. Except for a constant of 2π, the beam displacement, is transferred from a point above the image plane exactly onto
∆rCs, due just to spherical aberration is given by ∇χCs, and the the image plane. By choosing the defocus Zopt, the radius of the
beam displacement ∆rz due to just defocusing is given by ∇χZ; point-spread function adopts a minimum value of Rmin, with
you can verify these statements using the proportionalities in-
dicated in Fig. 9.5. The combined displacement of a beam with Rmin =
1
Cs 3 g3max (9.17)
spatial frequency g is then given by 4
r.g/ = rCS .g/ + rZ .g/
A typical non-Cs-corrected microscope equipped with a
(9.13)
= 1=.2/r.g/ = Cs 3 g3 + Zg field-emission gun suffers from strong contrast delocalization
when we try to make use of the extra resolution in the band
The actual correct vector notation has been omitted in Eq. 9.13 between point resolution and information limit. Unless this fre-
due to the fact that only azimuthally isotropic functions are in- quency band is sacrificed by inserting an objective aperture, the
volved. The maximum modulus of beam displacements ∆r(g) minimum achievable point spread Rmin is on the order of more
occurring within the complete beam bundle with frequency range than 1 nm, which is approximately ten times larger than the in-
[0,gmax] defines the radius R of the point-spread function (PSF), formation limit. A point spread of about 1 nm is disastrous with
with respect to obtaining a ‘sharp’ image: an acceptable image delo-
ˇ ˇ calization should not be larger than the value of the information
R = jr.g/jmax = ˇCs 3 g3 + Zgˇmax (9.14) limit.
 9.3 From the Wavefunction to the Image 245

Fig. 9.7 Contrast transfer functions

(CTFs) for different 200 kV microscopes
at characteristic defocus values. Left

Part I
column: coherent CTFs; right column:
corresponding partially coherent CTFs
including partial temporal and partial
spatial coherence envelopes. a,b CTF
at Scherzer defocus Z = −67 nm for
a microscope with a spherical ab-
erration of CS = 1.2 mm. c,d CTF for
same microscope at the defocus of
least confusion Z = −238 nm. e,f CTF
at Scherzer defocus Z = −12 nm for a
CS-corrected microscope with a spher-
ical aberration CS = 40 μm. The point
resolution is denoted by gS, the infor-
mation limit by gmax

Chapter 9
Obtaining a point spread of the order of 1 nm is only possible, When attempting to make use of the information limit and ac-
without CS correction, with the optimized defocus Zopt, which cessing the resolution regime above the point resolution, we
lies typically in the very-far-underfocus region of approximately can’t avoid oscillations of the CTF. Then the optimum defocus
−200 nm. Scherzer defocus, which is typically on the order of Zopt is a good compromise. It tends to distribute the frequency
−50 nm, would be a completely wrong choice for a contrast of CTF oscillations evenly inside the frequency band extended
transfer up to information limit, since then the resulting point- up to the information limit. Oscillations of the CTF that are too
spread can easily exceed 5 nm, making any direct image inter- rapid, and thus correspond to a strong gradient of the aberration
pretation impossible. function, are avoided, so we minimize the image delocalization.
We can see a comparison between the CTFs corresponding to
these two defocus settings in Fig. 9.7 for a 200 kV microscope
Scherzer defocus is thus only optimum for imaging up to (λ = 2.508 pm) with a spherical aberration of CS = 1.2 mm.
the point resolution, because then a wide, flat CTF is used
inside the relevant frequency band and all oscillations of We can completely remove the conflict between optimum imag-
the CTF are moved outside. 9 inary-part transfer and minimum contrast delocalization, which
is typical for field-emission microscopes, if we first correct the
spherical aberration. With such a corrector we can tune the value
 246 9 Focal-Series Reconstruction

of the spherical aberration, CS, so that the optimum defocus Zopt is the choice of the defocus values for the series. We’ll postpone
and Scherzer defocus ZS coincide. Then we also achieve an the question of the optimum focal distance between the images
approximate coincidence between point resolution and the in- of the series, but this needs to be clarified: where should the
Part I

formation limit. Using Eqs. 9.12 and 9.16 we can immediately series as a whole be placed along the focal axis?
show that a coincidence of the two  basic defoci is reached for
−1
an optimum value of Cs = 64=27 3 g4max . The correspond- Since a series of many images is recorded and we want to exploit
ing values for the optimum defocus and the minimum radius
−1 of the total information contained in the series anyhow, it doesn’t
the point-spread function are then Zopt = −16=9 g2max and really matter exactly what defocus values are used for the single

Rmin = 16=27g−1max for a Cs-corrected microscope. (Cs-corrected images of the series. This situation is completely different from
does not mean that Cs is zero.) single-shot imaging, where it is crucial that we hit exactly the
right defocus value. The only concern for the acquisition of a
The very large point spread of an uncorrected field-emission TEM, focal series is not to waste information unnecessarily. We can
which can be an order of magnitude larger than the information waste information in two ways – both are fundamentally linked
limit 1=gmax, can thus be decreased to a satisfactory level of about to the gradient of the aberration function.
one half of the information limit. For a 200 kV microscope with an
information limit of 0.125 nm the optimum value for the spherical Firstly, we can waste information in real space by choosing a
aberration is CS = 0.04 mm and that for the optimum defocus is defocus which causes an unnecessary large radius R of the PSF,
Zopt = −11 nm. A CTF corresponding to such an optimized con- and which is essentially larger than Rmin. Due to the convolution
dition is shown at the bottom of Fig. 9.7. For completeness we with the PSF, object points that are closer than a distance R from
should mention that we usually deliberately overcompensate the the edge of the recorded object area, will ‘bleed out’ informa-
spherical aberration when a corrector is available so that we have tion beyond the recorded object area. Since this information is
a negative spherical aberration. All that we said before remains outside the recorded area, it is lost and cannot be used later to
valid, except for the fact that the respective signs of the spherical reconstruct the wavefunction that belongs to such a point inside
Chapter 9

aberration, of the defocus and of the CTF, have to be inverted. the recorded area. Compared to the recorded area, the area that
As a consequence of the sign reversal of the CTF we now ‘see’ we can actually reconstruct is thus always smaller in the x- and
bright-atom contrast instead of the previous dark-atom contrast: y-directions by the radius R of the PSF in use. In order to maxi-
the atomic positions are imaged as bright dots. mize this area, we should make sure that the radius R of the PSF,
and thus ∇χ, are kept small, resulting in defocus values near the
An important feature of the CS-corrected CTF for FSR is high- optimum defocus Zopt.
lighted in the following. As we can see in Fig. 9.7, the predom-
inant feature of the CS-corrected CTF is that we don’t have any The Midpoint of a Focal Series
unwanted oscillations because we have a very low gradient of the
aberration function, which leads to the desired low point spread.
However, because we have a very low gradient of the aberration Place the midpoint roughly near the optimum defocus Zopt.
function, the CTF reaches values reasonably close to –1, which – For a non-corrected microscope Zopt lies in the deep
gives an optimum imaginary-part contrast, but we don’t transfer underfocus region. A few tens of nanometers less un-
before g = ~3 nm−1. So, details in the object that are larger than derfocus than Zopt work best in practice.
about 0.3 nm cannot be transferred correctly according to the – For a CS-corrected microscope Zopt lies close to zero
requirements of a WPO. Thus by extending the optimum trans- defocus. Avoid possibly featureless images between
fer properties up to the information limit, we unintentionally Zopt and zero defocus. 9
caused a transfer gap at low frequencies. For microscopes with
an even better information limit of 0.1 nm or less, the situation
gets worse! Even small crystal unit cells are then too large to Secondly, we can waste information in Fourier space because of
be imaged in the desired imaginary-part mode. Taking images the partially coherent, dampening envelopes, which we haven’t
with different focal values away from the optimum defocus for discussed here to keep the discussion short! However, the enve-
CS-corrected microscopes allows us to introduce a moderately lope of the partial spatial coherence, which is linked to the finite
strong gradient of the aberration function and thereby to capture size of the electron source, has an impact here. This dampening
more of the critical low frequencies. envelope also contains the gradient of the aberration function,
and has the form exp(−c(∇χ)2) = exp(−cR2), where c is a constant
and is of no particular importance here. In order to keep this
Reflection time dampening function close to 1 and to prevent a severe contrast
At the end of this section, when we have worked through the loss it is again important to keep the gradient ∇χ of the aberration
linear transfer theory of a real microscope in some detail, it is function as small as possible, so we prefer defocus values close
important to reflect the consequences arising for FSR. Since the to the optimum defocus Zopt.
wavelength and the spherical aberration (the main constituents
of the aberration function), plus the information limit, are usually The choice of Zopt has different consequences for the absolute
fixed for a given microscope, the only remaining free parameter defocus values, depending on whether a CS-corrected microscope
 9.3 From the Wavefunction to the Image 247

is used or not. For a non-corrected microscope, Zopt lies in the a dispersive behavior to the phase-shift microscope, or we can
very-deep-underfocus range of approximately −200 nm, depend- start from Eq. 9.11 and add a cosine term to the traditional way
ing on the particular microscope in use. Since the derivation of describing contrast transfer by the sine term of the CTF. Both

Part I
of Zopt is a purely optical one – it doesn’t reflect the fall-off of alternatives lead to the same complete description of the linear
the atomic scattering amplitudes for higher scattering angles as image intensity (if we get it right)!
an object property – the theoretically predicted point spread on
the overfocus side of Zopt is an overestimation of the actually For the sake of compactness, we make the following assignments
observed one. While the delocalization effect might be large for for the FT of the real and imaginary parts of the wavefunction
such high-frequency details, according to the optics laws we Ψ(r), be it the object wavefunction or the image-plane wave-
derived here, their contribution to the image contrast can never- function, with
theless be quite small due to the object properties. It turned out
that the advice from theory, which is to place the center image X.g/ = 2 FT ŒRe. .r// =  .g/ +   .−g/
(9.18)
of a focal series exactly at Zopt, is not always optimum for an Y.g/ = 2 FT ŒIm. .r// = −i. .g/ −   .−g//
experiment with a non-corrected microscope. Going some tens
of nanometers less than Zopt into the underfocus region, where Note that X(g) itself is, in general, a complex-valued Fourier
the delocalization effect appears to be smallest, turns out to be coefficient, and that the Friedel symmetry, i.e., X(−g) = X*(g),
more favorable. holds, since the real part of Ψ(r) is a real-valued function. The
same is of course true for Y(g) as the Fourier transform of the
For a corrected microscope with an optimally-adjusted CS value real-valued imaginary part of Ψ(r). By defining the Friedel-sym-
of a few tens of µm, the situation is totally different because metric quantities X(g) and Y(g) a-priori, we remove from viewing
all the relevant defoci are located within about ±10 nm around the explicit superposition of Ψ(g) and Ψ*(−g) on the right-hand
zero defocus and ∇χ is low anyhow. Here too, however, there is side of Eq. 9.18. It will therefore be sufficient to regard only
a reason why it is not always optimum to place the midpoint of the +g frequency axis, because the negative −g frequency axis

Chapter 9
a series exactly at the value of Zopt. The images of a focal series carries only the redundant complex-conjugate information. For
need to be laterally aligned with sub-pixel accuracy in order the sake of clarity, and to avoid confusion (☺), we’ll postpone
to compensate a possible drift of the object during the experi- until Sect. 9.4 a second, alternative way of treating explicitly the
ment. However, when using a CS-corrected TEM for imaging superposition of Ψ(g) and Ψ*(−g), which makes a distinction
a very thin object, the image contrast may almost completely between the +g and the −g directions.
vanish between Zopt and zero defocus and image alignment can
become very difficult, or even impossible, due to a lack of sig- The extension of the constant-phase formalism of Eq. 9.7 using
nificant contrast features in this defocus regime. So if we want a frequency-dispersive behavior, or the extension of the exclu-
to avoid this low-contrast regime we should start a series at sive imaginary-part transfer of Eq. 9.11, by adding the missing
the first defocus value exhibiting significant contrast and then real-part transfer, respectively, then yields for the linear image
move further away from zero defocus towards larger defocus intensity the expression
values in the course of the series acquisition. In this way, the
series is placed completely in the positive defocus regime for IL .g ¤ 0/ = XO .g/ cos..g// + YO .g/ sin..g// (9.19)
a microscope corrected to a small negative CS value. Likewise,
the series is then placed in the negative defocus regime for a Here the subscript ‘O’ refers to the object wavefunction, and the
microscope corrected to a small positive CS value. In both cases subscript ‘I’ refers to the image-plane wavefunction, as before.
an image close to Scherzer defocus is automatically included It is now very instructive for the comprehension of the mix-
in the series. ing of real-part and imaginary-part transfer to re-formulate the
above expression in an extended way, which reflects, in detail,
the physical processes involved. For this purpose we construct
a complex-valued vector, which carries X(g) as the Fourier co-
9.3.4 From Oscillations to Windings: efficient of the real part of Ψ(r) in its first dimension, and Y(g)
an Integral View on Linear Imaging as the Fourier coefficient of the imaginary part of Ψ(r) in its
second dimension. We then obtain the following very instructive
re-formulation of the linear image intensity
In this section our goal is to summarize the insights into linear
imaging that we’ve just gained and move towards an overview of
" ! !# !
cos..g// sin..g// XO .g/ 1
this topic. The constant-phase-shift microscope of Sect. 9.3.2 of- IL .g ¤ 0/ =
− sin..g// cos..g// YO .g/ 0
fered an interesting tomographic aspect of linear imaging, while
missing a model for the optical properties of a real objective lens.
! !
XI .g/ 1 (9.20)
Section 9.3.3 dealt with a real objective lens and its dispersive =
YI .g/ 0
phase shifting properties, while missing the tomographic aspect.
Since the missing aspects of the previous two sections are per-
fectly complementary, we can either start from Eq. 9.7 and add
 248 9 Focal-Series Reconstruction
Part I
Chapter 9

Fig. 9.8 3D visualization of the frequency-dispersive linear contrast transfer by an electron microscope. The axis denoted by ‘g’ is the spatial frequency axis. Pro-
jections of the displayed complex-valued functions onto the two axes denoted by ‘Re’ and ‘Im’ yield their respective real- and imaginary part contributions in real
space. a The real part of the object wavefunction is displayed by an exemplary Gaussian function in the Re direction. The additionally displayed helical surface is a
3D representation of the phase transfer function t = exp(−iχ(g)) and defines in the present context an angle −χ(g) by which the real part of the object wavefunction
has to be rotated around the g-axis in order to obtain its contribution to the image-plane wavefunction. b Image-plane wavefunction originating from the real-part
of the object wavefunction after the rotation has taken place. c Projection of the image-plane wavefunction onto the real axis, yielding the contribution of the real
part of the object wavefunction to the linear image intensity. This projection reflects the so-called cosine transfer function. d–f Analogous transfer of the imaginary
part of the object wavefunction into the linear image intensity. Here the Gaussian function representing the imaginary part of the object wavefunction is along the Im
direction in (d), while the helix defining the dispersive rotation remains unaltered. The final projection reflects the so-called sine transfer function, which is identical
to the CTF treated previously. (Scherzer defocus is chosen here for easy recognition)
 9.4 From the Images to the Wavefunction 249

Two fundamental physical processes are reflected by the above tails larger than approximately 1 nm are thus always imaged in
expression. HRTEM via this cosine transfer mechanism, and therefore only
the real part of the corresponding wavefunction is captured in

Part I
zz First, the term in the square bracket describes a frequency-dis- the image. For higher spatial frequencies we generally have a
persive rotation of the real and imaginary parts belonging to mixture of sine and cosine transfer; the cosine transfer is at max-
the initial object wavefunction given by the vector (XO(g), imum when the traditional CTF exhibits its transfer gaps.
YO(g)) by means of a rotation matrix to result in the final
image-plane wavefunction given by the vector (XI(g), YI(g)). The aberration helix winds in a particular sense, which is
zz Second, the product of the resulting image-plane wavefunc- linked to the sign of the gradient ∇χ(g). You can see in the
tion (XI(g), YI(g)) with the vector (1, 0) describes a projection example of Fig. 9.8 that this winding sense changes in the mid-
of the image-plane wavefunction onto the real-axis of the dle of the Scherzer plateau, where ∇χ(g) is zero and the pos-
complex number plane in a sense that only the real-part of itive ∇χCs(g) (due to spherical aberration) starts to outweigh
the image-plane wavefunction contributes to the linear image the negative ∇χZ(g) due to a negative defocus. The angular
intensity. winding frequency of the aberration helix along the spatial fre-
quency axis is proportional to the modulus |∇χ(g)|; the strong
Referring back to Sect. 9.1 where we described imaging in its beam displacements ∆r(g) are thus seen in the high winding
widest sense as a combination of information encoding and in- frequency at the respective frequency g. The variability of the
formation loss, these two processes actually find their mathe- winding frequency is also the reason why the object we are
matical description in Eq. 9.20. The application of the rotation discussing is not a helix in a strict mathematical sense, since
matrix can be seen as the information encoding process, which the latter would have a constant slope when progressing along
is then followed by the projection operation as an information the winding axis.
loss mechanism.

Chapter 9
It is very important that we state precisely what we mean by
complex-number plane in this context, since there are actually 9.4 From the Images
two involved! One complex-number plane refers in Fourier
space to the single vector coefficients X, and Y, respectively, to the Wavefunction
which are each by itself complex valued, albeit Friedel-sym-
metric. The other complex-number plane refers in real space to In the previous sections we were concerned with the linear con-
a projection axis, and thus determines the ratio between real and trast-transfer theory, which explains the mechanisms determin-
imaginary part resulting from (or surviving) a projection in real ing the contrast of single, separate images. In the following we
space. In the following discussion, we always mean the latter will first explain how the information contained by the object
view of projection directions in the real-space (Re, Im) plane. wavefunction is spread into an ensemble of images taken at dif-
This view is explicitly reflected in the construction of the (X, Y) ferent defocus values. Based on this insight, we will then derive
vector, where the first component is the real-space real part, and the actual inversion process, which gathers all the information of
the second component is the real-space imaginary part of the a focal series to re-establish the object wavefunction.
wavefunction.

The two processes described by Eq. 9.20 are visualized by the

3D graphs of Fig. 9.8. The central element of this visualization 9.4.1 Tomographic Interpretation of Focal
in the (Re, Im, g)-space is a helical surface winding around the Series
spatial-frequency axis. This surface is a 3D representation of
the complex-valued phase-transfer function t(g) of Eq. 9.9. It
elucidates the action of the rotation matrix of Eq. 9.20 as a fre- Before we actually begin the inversion of the imaging process, it
quency-dispersive intermixing of the real and imaginary parts is very instructive to change the point-of-view within the 3D sce-
originating from the initial object wavefunction. nario of linear imaging developed in the previous section. We
have actually already developed the idea of changing the final
A projection of the helical aberration surface onto the real axis projection axis within the complex-number plane within the
gives us the so-called cosine transfer-function cos(χ(g)); the pro- framework of the constant-phase-shift microscope. According
jection onto the imaginary axis gives the so-called sine trans- to this idea it does not make a difference whether the real and
fer-function sin(χ(g)); we discussed the latter exhaustively as the imaginary parts of the object wavefunction are rotated by the
CTF in the previous section. aberration function before the final projection onto the real axis
in the image plane takes place, or if the object wavefunction
It should now be evident that the cosine transfer, which is tra- is left unchanged and the final projection axis is rotated in the
ditionally omitted, is actually the dominant contrast-transfer opposite direction. In contrast to the constant-phase-shift micro-
mechanism especially at low frequencies, since there cos(χ) ≈ 1 scope, the projection axis cannot now be shifted rigidly for all
and sin(χ) ≈ 0. Assuming reasonable defocus values, object de- spatial frequencies together, but in a dispersive way, leading to
 250 9 Focal-Series Reconstruction

a winding of the projection direction. For this purpose we can 9.4.2 Fundamental Properties of Focal Series
rearrange Eq. 9.20 to give an alternative expression for the linear
image intensity, with
Part I

Within the linear formalism it is possible to deduce fundamen-

When we Record a Focal Series tal properties of FSR that are based on the tomographic view
developed previously. These properties do not depend on a par-
ticular mathematical or computational implementation of the
We are actually recording a frequency-dispersive to- reconstruction scheme. The intention behind this strategy is that
mogram of the object wavefunction within the complex we can demonstrate that the properties discussed above depend
number plane! 9 solely on the laws of physics and mathematics; we don’t want
you to attach seemingly mystical capabilities to this or that par-
ticular recon­struction scheme, or to a special numerical imple-
!" ! !# mentation thereof, even though you may frequently suspect such
XO .g/ cos..g// − sin..g// 1 a tendency in others!
IL .g ¤ 0/ =
YO .g/ sin..g// cos..g// 0
! !
=
XO .g/ cos..g//
(9.21) There is, in general, an analytical solution to the linear
YO .g/ sin..g// reconstruction problem.
We have N different images available with different defocus val-
ues Zn, and thus N different aberration functions χn(g), with
The expression in square brackets now describes a rotation of  
the projection axis (1, 0) in the image plane to give a new pro- 1 1
n .g/ = 2 Cs 3 g4 + Zn g2 (9.22)
jection axis (cos(χ), sin(χ)), while the object wavefunction given 4 2
Chapter 9

by (XO,YO) is left unchanged.

Then we find two new equations by multiplying Eq. 9.19 with
Since the right-hand side of Eq. 9.21 describes a projection of cos(χ(g)), and sin(χ(g)), respectively. A subsequent summation
the object wavefunction onto a certain direction of the com- of these equations over all N images then yields:
plex-number plane, taking a focal series with different values of
the defocus, and thus different values of the aberration function χ, N
X N
X
means nothing other than making a tomographic recording of the IL;n .g/ cos.n .g// = XO .g/ cos2 .n .g//
object wavefunction within the complex number plane, albeit in n=1 n=1
a literally, very twisted form. The projection axes in the 3D (Re, N
X
Im, g)-space belonging to different defocus values are given by + YO .g/ cos.n .g// sin.n .g//;
a helical function with opposite rotation sense to the aberration n=1 (9.23)
helix introduced in Sect. 9.3. N
X N
X
IL;n .g/ sin.n .g// = XO .g/ cos.n .g// sin.n .g//
A display of the frequency-dispersive projection axes is given in n=1 n=1
Fig. 9.9 for four different defocus values. In order to highlight N
the defocus-variation principle alone, we’ve excluded all other
X
+ YO .g/ sin2 .n .g//:
constant contributions to the aberration function χ (a contribution n=1
from the spherical aberration and a contribution from a focal 
offset common to all images); the result is a constant real-axis The two complex-valued linear equations can be solved for
projection for the first image of Fig. 9.9. The enormous simpli- each wave vector g separately with respect to the two unknowns
fication of the linear-imaging approximation compared to the XO(g) and YO(g). Thus, in general, the real and imaginary parts
full nonlinear formalism becomes obvious when you look at, of the object wavefunction can be restored; note the emphasis
e.g., the spatial frequencies of 1, 3, 5 and 7 nm−1. For each of placed on the term in general. This is because we haven’t made
those frequencies we draw a separate (Re, Im)-plane and can a statement with respect to the coefficients of the linear equation
solve the reconstruction problem by following the rotation of system, which are given by the sums on the right-hand side. The
the projection axis on that particular (Re, Im) plane, independent values of the above coefficients depend on the particular choice
of all other frequencies. It is important to note that the rotation of the focal values Zn, which leads to the problem of missing
frequency of the projection axis at a spatial frequency g is given projections; we’ll address this problem next.
by .1=2/@=@Z = .1=2/g2. One consequence of this projec-
tion is that the complete projection helix drawn in Fig. 9.9 does
not rotate as a rigid object when we vary the defocus; instead it There is a lowest reconstructible frequency.
rotates slowly for low spatial frequencies and more quickly for From the tomographic view developed in Sect. 9.4.1 it will be
high spatial frequencies. For this reason a focal series contains quite obvious that at least two substantially different projections
at each spatial frequency a different set of projections of the of the object wavefunction are necessary in order to restore the
object wavefunction. object wavefunction. We still have a problem with low spatial
 9.4 From the Images to the Wavefunction 251

Part I
Chapter 9

Fig. 9.9 Tomographic view of the focus-variation principle in the 3D (Re, Im, g) space. The displayed surfaces can be interpreted as frequency-dependent directions
in the complex (Re, Im) plane onto which a wavefunction is projected upon imaging. Only the defocus-dependent variable change of the projection directions is
displayed for the defocus values (a) 0 nm, (b) 8 nm, (c) 16 nm, (d) 32 nm and for an accelerating voltage of 200 kV. The projection directions are highlighted for the
exemplary frequencies 1, 3, 5, 7 nm−1 as the intersection of the planes g = const with the helical surfaces, which are defined by the complex conjugate of the corre-
sponding phase transfer functions
 252 9 Focal-Series Reconstruction

It is commonly agreed that two substantially different projections

will be available when the respective projection directions differ
from each other at least by π/4. Applying this condition to a fo-
Part I

cal series, where the defocus is changed continuously from the

first image to the last one, requires that the aberration functions
belonging to the first and the last image should differ from each
other by at least π/4 in order to be able to reconstruct a certain
spatial frequency. If we denote the total focal length covered by
a series as L, then this condition yields an estimate for the lowest
reconstructible spatial frequency gmin, with
1p
gmin  1=.L/ (9.24)
2

Focal series are frequently recorded with an equidistant focal

spacing δ between two images; we then have a total focal length L
covered by a series of N images L = (N−1) δ. In such a case an in-
crease in the number of recorded images helps to lower the value
of the lowest reconstructible frequency gmin. If we are using a
200 kV microscope (λ = 2.508 pm), with a typical focal step width
of δ = 4 nm and a typical number of N = 15 images, we find that
the lowest reconstructible frequency has a value of gmin ≈ 1.3 nm−1.
Chapter 9

There can be ‘blind spots’ in the reconstructed

frequency spectrum
The following argument strictly applies only to equidistant focal
series, but addresses implicitly analogous problems, which can
also occur (but be hidden) when we have a randomly-spaced
focal series. Apart from the low-frequency gap discussed above,
there is a second situation that can lead to a lack of different pro-
jections. Such a situation occurs for equidistant series at spatial
frequencies g, where the projection direction changes from one
image to the next exactly by multiples of π. Without restricting
Fig. 9.10 Tomographic view of the focus-variation principle in the 3D (Re, Im, the generality of our treatment, we could, e.g., always capture
g) space. The surfaces display the defocus-dependent and variable part of the the real part belonging to a certain frequency g with alternating
phase transfer function belonging to eight different images of an equidistant fo- sign, without ever recording the imaginary part. This situation is
cal series. With opposite winding direction, these surfaces can also be interpret- comparable to the strobe-light effect, where a periodically mov-
ed as frequency-dependent directions in the complex (Re, Im) plane onto which ing object comes seemingly to rest. For a given focal step-width
a certain spatial frequency g of the wavefunction is projected upon imaging. δ, so-called resonance frequencies appear at spatial frequencies
Projections are acquired approximately every π/4 for highest spatial frequency in g, where the aberration function changes from image to image
the foreground. When going backwards along the frequency axis towards g = 0,
exactly by multiples of π, i.e., when πλδg2 = nπ. A gapless recon-
the variety of projections decreases due to the g2 dispersion of the aberration
struction without such ‘blind spots’ is only possible up to the
function. At g = 0, only one and the same projection is acquired by all the dif-
first resonance frequency gres with n = 1, thus
ferent images
p
gres = 1=ı (9.25)
frequencies, since the focal rotation speed of the projection spi-
ral tends to zero when approaching zero spatial frequency. As a The focal step size δ of equidistant series can always be chosen
consequence, we essentially obtain only cosine transfer, and the in such a way that the first resonance frequency falls beyond the
low-frequency content of the recorded images yields again and information limit and does not pose any problem in practice.
again the same projection of the real part of the object wave- For a 200 kV microscope and a focal step size of 4 nm, the first
function, while the imaginary part hardly influences the image resonance frequency is gres = 10 nm−1.
intensity. You can see the resulting mathematical ill-posedness
immediately from Eq. 9.23 by setting all the sine terms to zero. There is an optimum focal step size for equidistant
In this case, only the real part is accessible, which is identical to series
the recorded low-frequency image content itself! This problem is The optimum focal step size δ should be a compromise between
analogous to the well-known missing-cone problem in traditional two competing requirements. On the one hand, we want it to be
tomography and is visualized in Fig. 9.10. as large as possible in order to maximize the total focal length

Licensed to ([email protected])
 9.4 From the Images to the Wavefunction 253

of the series and thereby to extend the reconstructible bandwidth is more useful for an intuitive comprehension of the processes,
at the low-frequency end. On the other hand, we want the size the disentangled representation yields a direct expression for
of the focal step to be small enough to avoid any resonances the inversion of the imaging process, albeit in a less instructive

Part I
below the frequency gmax, which defines the information limit. manner!
Resonances occur when the change of the aberration function
between two successive images amounts to π at gmax. A safe way We can write down the linear solution to the inverse imaging
to stay away from the latter condition is to apply smaller phase problem for a focal series of N images using a back-projection
changes of approximately ¾ π at gmax, which leads to an estimate expression in the form
for an optimum step size δopt, with N
X
ıopt  3=.4g2max / (9.26)
 .g ¤ 0/ = fn .g/IL;n .g/ (9.27)
n=1

In this way we safely avoid the first resonance frequency at a The Fourier coefficient Ψ(g) of the object wavefunction is a sum
(tolerable) loss of low frequencies, and the frequency disper- over all N Fourier coefficients IL,n(g) of the linear image intensity,
sion of the focal aberration function stays well balanced (see each of them weighted by a complex-valued filter function fn(g).
Fig. 9.10). For a 200 kV microscope with an information limit Our goal now is to find an analytical expression for the filter
of gmax = 8 nm−1 we obtain δopt ≈ 4.7 nm. Since we cannot adjust function fn(g).
the focal-step size on a real microscope continuously, we simply
take the next possible step size smaller than δopt. We start from the expression of the linear intensity of image n, in-
cluding a phase-transfer function tn(g), with tn .g/ = exp.−in .g//.
Equidistantly spaced focal series are preferable Thus we have
over randomly spaced focal series
The argument in favor of equidistant series will be held on a IL;n .g ¤ 0/ =  .g/tn .g/ +   .−g/tn .−g/ (9.28)

Chapter 9
heuristic basis, nevertheless a statement is already possible at this
stage. The problems occurring near the minimum reconstructible By making use of tn .g/ = tn .−g/ and by multiplying Eq. 9.28
frequency and the resonance frequency discussed above result one time with tn(g) and another time with tn*(g), we obtain the
from an unbalanced distribution of projection directions – lead- following two equations after summation over all images
ing in the latter cases to the most extreme extent – to completely
missing projections. With an irregular focal-step size we cannot
N
X N
X
guarantee that a well-balanced distribution of projections occurs IL;n .g/tn .g/ =  .g/ tn2 .g/ + N  .−g/;
continuously for all relevant spatial frequencies. So, some fre- n=1 n=1
(9.29)
quencies will exhibit a more favorable variety of collected pro- N
X N
X
jections, which will lead to a more favorable set of coefficients IL;n .g/tn .g/ = N .g/ +  .−g/
tn2 .g/:
in Eq. 9.23; other frequencies will be less favored by the choice n=1 n=1

of irregular focal spacings (focal-step sizes). Equidistant series

give us the widest possible balanced collection of projections for Multiplication of the upper equation with 1=N tn2 .g/ and sub-
P
a continuous frequency band. sequent subtraction of the result from the lower equation yields,
after rearrangement for Ψ(g), an expression for the coherent,
linear, filter function fn(g) in its most simple form

9.4.3 An Explicit Solution to the Linear

 P 
tn .g/ − tn .g/ N1 Nn=1 tn2 .g/
Inversion Problem fn .g/ = P  P  (9.30)
N N
N − N1 2
n=1 tn .g/ n=1 tn .g/
2

In this section we’ll derive an explicit analytical solution to the

linear reconstruction problem. Apart from wanting to be com- The above expression is instructive for many reasons. We
plete in our analysis, the availability of an analytical solution will recognize that for very large image numbers N we have
also help us answer the important question that we have not yet fn .g/ ! .1=N/tn .g/, and thus the first term in the enumerator is
treated quantitatively, namely “how many images are required the actual back-projection function, which builds up the object
for a reconstruction?” wavefunction Ψ(g). However, this function alone is not suffi-
cient when we have a small number of images, and the task
While the view we developed on imaging and reconstruction of the second term in the enumerator is to diffuse, or destroy,
was mainly based on an entangled representation of Ψ(g) and remainders of the complex conjugate part Ψ*(−g). Together, the
Ψ(−g) – we followed the physical processes acting on the real two terms in the enumerator thus break-up the Friedel symme-
and the imaginary part of Ψ(r) in real space, the following for- try of the images in Fourier space, which was caused by the
malism is aimed directly at a disentangled Fourier-space rep- collapse of the wavefunction upon imaging, as we described at
resentation of Ψ(g). So, whereas the entangled representation the end of Sect. 9.3.1.
 254 9 Focal-Series Reconstruction
Part I
Chapter 9

Fig. 9.11 Denominator of the coherent linear filter function after normalization to one for four equidistant focal series consisting of (a) 2, (b) 5, (c) 10, and (d) 20 im-
ages. Defocus step size is δ = 4 nm, accelerating voltage is 200 kV. The first resonance frequency is located at 10 nm−1, the second at 10 √2 nm−1. The resonance gaps
as well as the low-frequency gap become narrower as the number of images increases. Note that the first maximum of the two-image filter function in (a) occurs at
p
10= 2 nm−1, where the phase shift between the two images amounts to π/2 and thus two independent projections are acquired

The denominator of the linear filter function (DLF) also offers can see in Fig. 9.11, a reconstruction from only two images pro-
valuable quantitative insights. The DLF is displayed in Fig. 9.11 duces a single maximum of the DLF at a frequency where ex-
for an equidistantly acquired, focal series with different total num- actly two perpendicular projection directions occur; i.e., where
bers N = 2, 5, 10 and 20 of the contributing images. We can ob- the aberration function changes exactly by π/2 between the two
serve two general features from any of the DLFs displayed there. successive exposures. In the two-image case, the quasi-‘blind’-
The DLF is zero at zero spatial frequency and at the respective spots near zero frequency and near the first resonance frequency
resonance frequencies we derived earlier. This means that a com- extend over a wide range, meaning that a high signal blow-up is
plete reconstruction at those critical frequencies would require a necessary, and thus a poor signal-to-noise ratio occurs for such
division by zero, which is of course not recommended! Also in the frequencies. Increasing the number of images to N = 5 narrows
vicinity of those critical frequencies a division by a small number the quasi-‘blind’-spot regions considerably and the DLF is then
would be necessary in order to restore the respective coefficient of already flat for a wider range of frequencies. Image numbers
the object wavefunction. For example, for low spatial frequencies N > 5 are still very useful to narrow the low-frequency recon-
a division by a small number would be required in order to blow struction gap and to extend the retrievable bandwidth of frequen-
up the very small signal obtained from the imaginary part of the cies. However, thepgain in low frequencies is only proportional
wavefunction, since all the images contain essentially the real part to approximately N for equidistant series so it is not very ef-
obtained by the cosine transfer mechanism. At a resonance fre- ficient. In practice, a value for N of N = 10 to 20 images turned
quency, instead of the real part, the signal can originate from any out to be most suitable, especially when considering the effect
projection direction in the complex number plane, but the perpen- of increasing radiation damage.
dicular projection is almost missing, requiring again an excessive
blow-up of the almost-missing, perpendicular projection. Since a In summary, the DLF indicates quantitatively how well the re-
division by such small numbers is not favorable with respect to construction problem is posed with respect to the number of input
noise amplification, we need more elaborate expressions than the images. Be aware that the well-posedness or ill-posedness of the
simplest one presented in Eq. 9.30, which was derived for the first reconstruction problem is a matter of the available input data
time by Schiske (1968); more elaborate expressions including a and does not at all depend on the capabilities of the numerical
noise treatment were developed later. reconstruction scheme actually used, at least in the linear case.
Proud proclamations of two-image reconstructions should there-
The evaluation of the DLF gives also a quantitative answer for fore always be questioned carefully with respect to the actually
the number of images to be included in a focal series. As we reconstructed frequency bandwidth: conjuring is not magic! This
 9.4 From the Images to the Wavefunction 255

statement can also be generalized towards nonlinear reconstruc- zz A second well-known procedure is generally called the Ger-
tion, with the exception of a few very brilliant, but more esoteric chberg–Saxton algorithm, although it should be more pre-
situations, which we won’t discuss here because we’re tired. cisely called the Misell algorithm in this special context.

Part I
The Misell algorithm is based on the assumption that the
Warning amplitude of images is known if you take the square root
of the image intensity. By assigning constant, or even ran-
dom, phase values in a first step, you obtain an initial guessed
Be cautious when you hear about reconstruction schemes wavefunction, which consequently has the correct amplitude
that “need only 2 or 3 images”. 9 but a completely wrong phase. Then you optimize the wave-
function by its propagation between any two image planes of
the recorded series, always picking up the respective image
plane amplitude. Since the amplitude information of a certain
plane contains phase information from all other planes, the
9.4.4 Nonlinear Reconstruction phase information can be finally retrieved.
zz A third approach towards the solution of the nonlinear inver-
sion problem is based on the so-called transport of intensity
The majority of considerations made so far within this chapter equation (TIE). In this case, the intensity change between two
have been based on the linear imaging formalism. The reason images taken at different defocus is described in real space
behind this strategy is the fact that the linear formalism gives by a Poisson-type formalism. We can find an approximate
the widest possible insight into the core mechanism of the fo- analytical solution of the TIE equation, and a nearly exact
cus-variation principle. Moreover, the linear formalism is correct solution by applying multi-grid methods.
in the limit of very thin objects and in this special case even of- zz A fourth approach towards the solution of the nonlinear in-
fers an analytical solution to the inversion problem. Fortunately, version problem is the use of stochastic algorithms. This

Chapter 9
in HRTEM we usually try to image objects that are as thin as approach, which employs either a simulated annealing algo-
possible and therefore in many cases we stay in a near-to-linear rithm or a genetic algorithm, is extremely powerful in the case
imaging regime. In this near-to-linear regime, the linear for- of periodic objects, but cannot be applied efficiently to general
malism will no longer produce results, which are correct on a non-periodic objects.
quantitative basis, but they are still interpretable and hopefully
comprehensible on the basis of our previous discussion. Apart from the stochastic approach, implementation of the three
other mentioned non-periodic approaches may exhibit signifi-
In a strict theoretical sense, we have to use nonlinear reconstruc- cant differences with respect to the reconstructible bandwidth
tion schemes whenever we make the transition from the WPOA or to the correct treatment of partial coherence, which we’ll not
of Eq. 9.2 to the POA of Eq. 9.1, which is the case in almost discuss further.
any realistic experiment. In contrast to the linear formalism, the
field of nonlinear reconstruction is an intense mathematical/com-
putational one. On the one hand it is intellectually much more MAL
challenging and brilliant, but on the other hand it offers very little The numerical reconstructions shown in the next section were
additional insights into the underlying physical mechanisms! It is made with a tandem combination of a noise-improved version
therefore way beyond the scope of this textbook, and the Editors, of the linear inversion scheme outlined in the previous section,
for us to try explain the different approaches to the nonlinear followed by an iterative nonlinear procedure called the maxi-
inversion problem in satisfactory depth. We won’t even judge the mum-likelihood (MAL) scheme. Although the iterative MAL
performance or compare them with each other. For this reason, scheme is capable of retrieving the object wavefunction from
we’ll only give a very simple classification of nonlinear inver- scratch, it turns out that that we can speed up the calculations
sion schemes. We’ll then follow this overview with a coarse de- significantly when we use the linear solution as input for a start-
scription of the particular inversion scheme, which will be used ing wavefunction for further nonlinear MAL iterations. In fact,
for the examples of reconstructions shown in Sect. 9.5. it can be shown that the outcome from the first iteration step of
a slightly different MAL variant (which we won’t discuss here)
In contrast to the linear inversion problem, there is no direct is mathematically identical to the analytical linear solution.
analytical solution to the nonlinear inversion problem. Nonlin- Because of this very close mathematical relationship, the tan-
ear inversion schemes are therefore mostly iterative numerical dem combination of a linear scheme followed by the nonlinear
procedures, which progress stepwise towards a solution for the MAL scheme is often referred to simply as ‘MAL’. Amongst
object wavefunction. the above-mentioned nonlinear reconstruction procedures, the
MAL scheme, which dates back to a Brite–Euram project in the
zz A typical procedure of this kind is the wonderfully named mid-1990s, is the best documented, and most thoroughly tested
maximum-likelihood algorithm, which approaches iteratively one and has produced the highest number of results relevant
towards the correct object wavefunction on the basis of a gen- to materials science. A combination of the original MAL soft-
eral least-squares formalism. ware with an up-to-date user interface, which can be handled
 256 9 Focal-Series Reconstruction

inverse imaging problem; but you have to remember that the

MAL procedure is an indirect procedure. If we use a typical set
of 15 input images of size 1,024 × 1,024 pixels, we can reach
Part I

full convergence after ~10 iteration steps, which takes less than
a minute of CPU time on a modern PC.

To conclude this section on nonlinear reconstruction we should

mention that reconstructed wavefunctions are usually not dis-
played via their real and imaginary parts, as has been done so
far in the context of linear reconstruction. The display of real
and imaginary parts adheres closely to the WPOA and already
loses its physical meaning when we make the transition to the
POA, where a display in the amplitude-phase notation is more
appropriate. Nevertheless, we can deduce a connection between
the two display modes for the case of weakly modulated wave-
functions in the close-to-linear regime.

When denoting the real part as Re. / = 1 + ˛ and the imagi-

Fig. 9.12 Schematic display of the working principle of the iterative nonlinear nary part as Im. / = ˇ, with a; ˇ  1, we obtain the follow-
MAL reconstruction scheme. Simulated images from a trial wavefunction Ψ are ing approximate relations for the amplitude A and the phase
compared with an experimental focal series. Based on this image comparison, ϕ: A  1 + ˛, and   ˇ. Thus the spatial modulation of the
an update dΨ to the wavefunction Ψ is calculated with the goal that images amplitude follows the modulation of the real part, whereas
simulated from the updated wavefunction Ψ+dΨ fit increasingly better to the the spatial modulation of the phase follows that of the imagi-
Chapter 9

experimental focal series. This cycle is multiply repeated until the difference mea- nary part. Following the derivation of the POA of Eq. 9.1, the
sure S2 converges towards a minimum value. It is then assumed that the final phase is peaked at atomic-column positions and thus gives us a
trial wavefunction, denoted here by amplitude and phase (A,Φ), represents the straightforward interpretation in terms of the object structure,
actual experimental wavefunction
whereas the amplitude information is often more complicated.
Therefore, usually only the phase of a reconstructed wavefunc-
also by non-experts, is commercially available under the name tion is displayed.
TrueImageTM.

As is outlined in Fig. 9.12, the MAL scheme starts with an initial

guess for the object wavefunction. This initial guess can be either 9.4.5 Numerical Correction of Residual
a plane wave when starting from scratch, or better, a previously Aberrations
calculated, linear solution of the object wavefunction. On the ba-
sis of this initial guess all images belonging to the focal series are
simulated and compared to the experimentally acquired images. Phase-retrieval methods, like FSR or electron holography, show
A difference measure, the so-called S2-value, is calculated by their full power when dealing with residual aberrations of the
accumulating the squared intensity difference between simulated objective lens. So far, only two contributions to the aberration
and experimental images over all pixels within an image and function χ(g) have been treated in this chapter. These are the
over all images of the series. An update dΨ to the initial guess Ψ spherical aberration CS and the defocus Z. These are the ‘clas-
of the wavefunction is made, based on an analytical functional, sical’ aberrations in TEM. Unfortunately, there are many more
which is the derivative of S2 with respect to the current wave- unwanted aberrations that can affect our wavefunctions, and con-
function Ψ. sequently also the resulting images.

In a next step, images are calculated on the basis of the newly up- The Wavefunction is a Dynamic Object
dated wavefunction Ψ + dΨ, and the whole iteration cycle starts
again and is repeated several times until the S2-value reaches a
satisfactorily small level determined by the experimental noise. Aberrations can be removed numerically at any time,
Except for noise, a full convergence of the S2-value towards zero whereas an image is static and cannot be corrected. 9
would indicate that the images calculated from the trial wave-
function reproduce exactly the experimental images, and we as-
sume in such an instance that the trial wavefunction and actual These aberrations can be problematic for two reasons. First, not
wavefunction are identical. all of them can be corrected well enough using appropriate hard-
ware, at least not as successfully as in the case of spherical ab-
If we have enough images included in the series, this assumption erration. Second, not all of the residual aberrations are stable in
is justified in practice because it gives a unique solution to the time so they can degrade the image contrast again a few minutes
 9.5 Application Examples 257

after correction; you have to permanently monitor and re-adjust there are two other nomenclature systems which originate
the hardware. Especially when we want resolution near or be- from the ray-optical tradition.
low the Å limit, the influence of residual aberrations becomes zz The older ray-optical nomenclature uses a historically grown

Part I
increasingly important. Practical experience with FSR has shown multitude of characters for the various aberrations in com-
that reconstructed wavefunctions are significantly affected by re- bination with a single index and individual pre-factors that
sidual aberrations; the effect is more general than the exception. are not as systematic and different from the pre-factors 1/m
Unlike an image, which is affected by an aberration, a wave- defined in Eq. 9.32.
function allows us immediately to remove or compensate for zz The newer ray-optical nomenclature uses double indexed co-
this aberration numerically as long as the aberration is known. efficients cm,n with similar systematic pre-factors as we do.
In some cases, aberrations can even be reliably determined from
the wavefunction itself, without any further measurement. The Note that there is one important difference between the dou-
possibility of eliminating residual wave aberrations is thus a crit- ble-indexed ray-optical nomenclature and the double-indexed
ical strength of any phase-retrieval method. wave-optical nomenclature! The second azimuthal index n in
the newer ray optical nomenclature is identical to ours, BUT the
We’ll now denote an initially reconstructed wavefunction by first index m is by 1 lower than in our case, thus cm−1,n = amn. The
ΨR, which is still affected by residual aberrations. Furthermore, difference by 1 between the ray-optical and the wave-optical
we denote the aberration function describing these unwanted indexing schemes is due to the fact that the exponent character-
residual aberrations by χR, and the ‘true’ aberration-free object izing the ray deviation ∆r is lower by 1 than the exponent of the
wavefunction by ΨO. The initially reconstructed wavefunction respective phase contribution to the wave aberration function χ,
ΨR, which is ‘contaminated’ by residual aberrations, is then since ∆r ~ ∇χ.
given by
Notation for Aberration Coefficients
R .g/ = O .g/ exp.−iR .g// (9.31)

Chapter 9
The ‘true’ object wavefunction ΨO can then be immediately re- Apart from the distinction between wave optical and ray
covered by a multiplication of the initially reconstructed wave- optical nomenclatures, there are two ray optical notations
function ΨR with the complex-conjugate phase-transfer function used for aberration coefficients: one uses a single index
exp(iχR(g)). Thus once the residual aberrations are known, they and one uses a double index. 9
can be corrected by a so-called numerical phase plate defined
by χR(g).
It is, of course, essential to know that the order of aberrations is
In contrast to the classical aberrations denoted by CS and Z, commonly addressed in the system of the ray-optical tradition,
which are azimuthally constant, most of the additionally possible thus the defocus is generally regarded as a first-order aberration,
residual aberrations are not azimuthally isotropic, and therefore and the spherical aberration as a third-order aberration.
the vector notation of the wavevector g, or an alternative polar
notation (g,φ), are really necessary in this context. Contributions
to the residual aberration function χR(g,φ) can be subdivided into
basic wave aberrations χmn(g, φ), with 9.5 Application Examples
 
1
mn .g; '/ = 2 amn gm cos Œn.' − 'mn /
m (9.32) From the many successful applications in the field of materials
n  m; m + n even science that have been produced so far using the MAL imple-
mentation of the FSR technique, we’ll only show two exam-
Here amn denotes the respective wave aberration coefficients. ples here but will include references to others on the website. In
Now we can identify a40 with the spherical aberration coefficient most of these applications it would hardly have been possible to
CS that we introduced previously, and a20 with the defocus Z. deduce all the relevant materials science from the raw images
Those two classical aberrations are thus only two wave aberra- alone without reconstruction. Applications include, e.g., the first
tions amongst many possible that can affect a wavefunction. A detection of light atoms such as C, N, and O at sub-Å resolution,
visual overview of the elementary wave aberrations is given in near-pm-precision measurements of interatomic distances across
Fig. 9.13 in the form of so-called phase plates. interfaces, the investigation of dislocation cores, the determina-
tion of composition profiles in semiconductor heterostructures,
You should be aware of the fact that three different nomencla- the determination of the crystallographic structure of catalyst
ture systems are used for the aberration coefficients and there’s materials, the investigation of graphene sheets, and many more.
nothing we can do to change that. The two examples presented here are more educationally moti-
vated, and illustrate the possibilities using either a conventional
zz We are coming from the wave optical side and use the wave field-emission microscope, or a CS-corrected field-emission mi-
aberration coefficients amn as we defined above. Additionally, croscope.
 258 9 Focal-Series Reconstruction
Part I
Chapter 9

Fig. 9.13 Basic wave aberrations displayed as phase distributions in Fourier space up to the sixth order of the aberration function. Positive phase values are
displayed from yellow to red, negative phase values from yellow to green. Discontinuities in the displayed color scheme occur every π/2. The wave aberration coeffi-
cient a20 denotes the defocus, the wave aberration coefficient a40 denotes the spherical aberration. Furthermore, a22 denotes twofold astigmatism, a31 coma, and a33
threefold astigmatism. In general, ann denotes an n-fold astigmatism

9.5.1 Twin Boundaries in BaTiO3 cation of a numerical phase plate for the correction of residual
aberrations, is shown in Fig. 9.15b.

Figure 9.14 shows two HRTEM images of an arrangement of After correction for residual aberrations, in this case mainly
twin boundaries in a thin film of BaTiO3. These two images were coma (Fig. 9.17), the phase of the wavefunction yields an al-
taken along the [110] zone axis of BaTiO3 and belong to a focal most direct structure projection. Ti-columns are revealed by the
series of 20 images made with a CM20FEG electron microscope more prominent phase maxima, whereas BaO columns are lo-
operated at 200 kV. Note the strong effect of contrast delocaliza- cated at the less pronounced phase maxima. This behavior is no
tion, which is typical for a field-emission microscope without a longer consistent with the WPOA, which predicts higher phase
corrector for the spherical aberration. Even though the contrast maxima for the heavier atomic columns, i.e., in the present case
delocalization has no apparently visible effect in periodic crystal for the BaO columns. The object thickness was determined by
areas, it exists there too; the blurring effect becomes more evi- image simulations to be 8 nm, which is clearly already beyond
dent at the central twin junction. The contrast delocalization is so the validity region of the WPOA. According to the Pendellösung
strong that any direct image interpretation, or even an attempt to behavior, the BaO columns are approaching their extinction
set up a structure model for further image simulations, is hardly length at this object thickness, whereas the Ti columns are still
possible. Such images are typical for non-corrected field emis- prominently visible in the phase. A magnified view of the phase
sion microscopes. image from the center of the twin junction, together with the
corresponding structure model is shown in Fig. 9.16. Contrary
Figure 9.15a displays the phase of the wavefunction initially to their behavior in the bulk surrounding, two BaO columns in
reconstructed from the series of 20 images. This wavefunction the center of the polygonal twin junctions (marked by arrows in
is strongly affected by residual aberrations of the objective lens the structure model) exhibit a phase peak that is even higher than
and is still not suitable for a direct structure interpretation. The that of the Ti columns. This effect can be explained by a partial
phase of the final wavefunction, which was obtained after appli- occupancy of these particular BaO columns.
 9.5 Application Examples 259

Fig. 9.14 Two HRTEM images (a) Z =

−123 nm, (b) Z = −195 nm of Σ3 twin
boundaries in BaTiO3 taken with a

Part I
CM20FEG electron microscope. These
images belong to a focal series of
20 images. The apparent contrast
delocalization is indicated at the ver-
tical boundary by arrows

Fig. 9.15 Phase of the reconstructed

object wavefunction from the same
area as shown in Fig. 9.14. a dis-
plays the phase of the object wave-
function before numerical aberration

Chapter 9
correction and b displays the phase
after correction for residual two- and
threefold astgmatism, and for coma.
The phase plate used for aberration
correction is shown in Fig. 9.17.
Atomic columns are revealed by
sharp phase peaks in the corrected
wavefunction. This example demon-
strates the importance of numerical
aberration correction
 260 9 Focal-Series Reconstruction

Fig. 9.16 Phase of the reconstruct-

ed object wavefunction and structure
model of the same area (barium:
Part I

green, oxygen: blue, titanium: red).

The arrows mark half-occupied BaO
columns
Chapter 9

9.5.2 Stacking Fault in GaAs

Figure 9.18 summarizes a HRTEM experiment made with the

first-ever, 200 kV CS-corrected CM200FEG microscope installed
in Jülich; CS was set to an optimized negative value of −0.04 mm.
The observed object is a specimen containing GaAs/InGasAs
heterointerfaces. A focal series of misfit dislocations and stack-
ing faults, which occur at the interface regions, was recorded
along the common [110] crystal axis.

The combined application of CS-corrected imaging with FSR

can be very fruitful. Figure 9.18c displays the CS-corrected im-
age taken close to Scherzer defocus, which exhibits white-atom
contrast according to the theory. Figure 9.18d displays the re-
constructed and numerically aberration-corrected phase image.
The difference between the Scherzer image of Fig. 9.18c and the
reconstructed phase image of Fig. 9.18d is striking and only the
reconstructed phase image allows further quantitative structure
analysis.

We conclude with three frequent similar experiences that you

should watch for.
Fig. 9.17 Numerical phase plate used to correct the wavefunction shown in
Fig. 9.15. The radius of the displayed Fourier-space disk is 6.5 nm−1, correspond-
zz Residual aberrations can deteriorate the imaging quality of a
ing to a resolution of 0.15 nm. Positive phase values are displayed from yellow to
CS-corrected microscope, as is the case for any other micro-
red, negative phase values from yellow to green. Discontinuities in the used color
scheme occur every π/2. The black line denotes a value of the residual aberration
scope. Such residual aberrations can be removed from the
function of π/4. The displayed phase distribution is mainly determined by the
retrieved wavefunction by numerical a-posteriori correction.
axial coma of the objective lens zz The low-frequency sine-transfer gap explained in Sect. 9.3.3,
which is an unwanted side effect of CS-correction, can lead
to significant imaging artefacts. Such artefacts appear in
 9.5 Application Examples 261

Part I
Chapter 9
Fig. 9.18 a Small patches extracted from a focal series of 20 images of GaAs taken with the spherical-aberration corrected Philips CM200ST FEG microscope in-
stalled at Jülich Research Centre. The images were taken along the [110] zone axis of GaAs using a negative value of the spherical aberration of CS = −0.04 mm. The
displayed image patches contain 2 × 1 projected GaAs unit cells and were extracted from exactly the same object area. b Plot of relative focal distances between the
images as obtained automatically and self-consistently by means of FSR without injection of any a-priori knowledge about the object. Moderate focal instabilities,
which tend to vary periodically in time, can be recognized. c Larger patch extracted from the first image of the focal series containing a stacking fault. The stacking
fault is imaged close to Scherzer defocus. Owing to the choice of a negative spherical aberration, atomic columns appear bright on dark background. d Phase of the
reconstructed wave function belonging to exactly the same area as the image patch shown in sub-figure (c). The reconstructed phase yields a direct insight into the
projected structure of the stacking fault (marked by ‘SF’ and a black arrow), where each atomic column is revealed by a unique phase maximum

Fig. 9.18c as white ‘ghost’ dots at tunnel positions, where no So the three experiences noted above, i.e., 1) residual aberra-
atom columns are located. The danger of an ad-hoc misinter- tions, 2) low-frequency sine-transfer gap, and 3) temporal av-
pretation of such dots as atomic positions is clear, especially eraging effect, lead to the drastic difference between the single
when observing unknown lattice defects. CS-corrected image and the reconstructed phase of the GaAs
zz A mobile amorphous surface layer on the specimen will fre- stacking fault, as is shown in Fig. 9.18.
quently deteriorate the image contrast.

In the latter case, averaging over many images of a focal series,

which are recorded over a time interval of approximately 1 min,
reduces the contrast of such mobile cover layers and enhances
the signal from the underlying stable crystal. Moreover, we can
show analytically using linear theory that a wavefunction recon-
structed fromp N images exhibits a signal-to-noise ratio, which is
roughly by N=2 better than that of a single image. To be fair,
we should also mention that, for the same reason, FSR is not a
suitable tool for studying weakly bound, amorphous objects, be-
cause these tend to change under electron irradiation. However,
in the case of stable crystalline samples, the averaging effect is
always an advantage.
 262 9 Focal-Series Reconstruction

Chapter Summary
Part I

This chapter reviews many of the concepts that are the basis tion of a focal series corresponds to a tomographic recording
of the HRTEM chapters in W&C but takes the discussion to of an electron wavefunction on twisted surfaces. With the
a much deeper level; it also gives you hints as to how you help of this tomographic interpretation, which is the central
can go even further. As a rule, we can regard FSR as a nu- idea of this chapter, the inversion of the imaging process can
merical inversion of the HRTEM imaging process based on be readily understood and also mathematically derived. The
a series of typically 10–20 input images. The output of the fascinating possibility that we can correct the resulting elec-
reconstruction is an electron wavefunction, which combines tron wavefunction numerically for parasitic lens aberrations
all the information contained by the single input images, and was briefly outlined and experimental examples were shown.
is ideally free from instrumental imaging artifacts.
Concerning experiments, there are only three specific pa-
A real understanding of the inversion principle requires a rameters that we need to adjust to be able to acquire a suc-
profound understanding of the imaging process itself. We cessful series on a given microscope. These are:
acquire this understanding step-by-step in a modular way by – The number of images in a series.
studying the properties of three different microscope types, – The absolute defocus value of the series midpoint.
which are – The relative focal step-size between images.
– A neutral microscope
– A constant-phase-shift microscope Simple formulas to guide you in choosing these parame-
– A real microscope. ters are given so that you can start the right experiment
immediately, regardless of whether the microscope is aber-
Chapter 9

By combining the characteristic properties of these three ration-corrected or not. Finally, we remind you of the funda-
microscope types in this chapter we were able to develop mental principle of science: you must give sufficient detail
mathematically and graphically a comprehensive description in your publications to allow others to repeat, and thus test,
of the linear imaging process. In this description, the acquisi- your analysis (by experiment and computation).

Licensed to ([email protected])
 Appendix Appendix Application Examples 263

Appendix

Part I
People Q9.8 How do the projection surfaces in the 3D (Re, Im, g)
space of a constant-phase-shift microscope differ from
those of a real microscope?
The development of the FSR method in electron microscopy Q9.9 How are the projection surfaces in the 3D (Re, Im, g)
began roughly in the 1970s. Over time, generations of scientists space related to the well-known contrast transfer function
contributed to the evolvement of various mathematical variants, (CTF)?
numerical implementations, and actual applications of FSR. For Q9.10 How large roughly is the minimum contrast delocaliza-
this reason, developing an only rudimentarily fair genealog- tion obtainable with a non-corrected field emission mi-
ical table of important contributors would simply require too croscope? What minimum contrast delocalization can be
much space. Instead, we name here only one person, who was obtained with a CS-corrected version?
frequently ignored in the scientific literature and who is barely Q9.11 By which feature of the projection surfaces in the 3D (Re,
known: Peter Schiske. He is the guy who invented the FSR prin- Im, g) space can one observe the function ∇χ(g)?
ciple! Peter Schiske (1924–2012) was an Austrian physicist, Q9.12 Is there an analytical solution to the linear reconstruction
mathematician and early computer expert. In the early 1950s he problem? Is there an analytical solution to the nonlinear
collaborated with the well-known Austrian physicist and elec- reconstruction problem?
tron optician Walter Glaser, who was a strong proponent of the Q9.13 At which defocus value should one approximately place
wave optical view of image formation based on the Schrödinger the midpoint of a focal series?

Chapter 9
equation. Not surprisingly, we also made strong use of the wave Q9.14 Is there a substantial difference in the defocus value Zopt
optical view throughout this chapter. It was finally Peter Schiske between an uncorrected microscope and a CS-corrected
who published in 1968 for the first time the idea of focal-series microscope? State typical values.
reconstruction in electron microscopy. The original paper was a Q9.15 Does the choice of the classical Scherzer defocus instead
short conference abstract written in German. Fortunately, due to of Zopt make any sense for FSR with an uncorrected field
its fundamental importance, Schiske’s original paper has mean- emission microscope?
while been translated to English and re-printed (see References). Q9.16 Which single parameter apart from the wavelength de-
termines the lowest reconstructible frequency gmin?
Q9.17 Which single parameter apart from the wavelength deter-
mines the resonance frequency gres in the high-frequency
Self-Assessment Questions regime?
Q9.18 Which single parameter apart from the wavelength de-
termines the optimum focal step size δopt?
Q9.1 Why is it favorable to have the object wavefunction Q9.19 Would it be efficient from a mathematical point of view
available compared to a single HRTEM image? to decrease the lowest reconstructible frequency gmin by
Q9.2 Are you aware of a TEM technique, where the object increasing the number of recorded images? Additionally,
wavefunction can be obtained without requiring a multi- what physical effect related to the object hampers such
tude of images? an attempt?
Q9.3 State at least two reasons why it is advisable to perform Q9.20 Is it a good idea to try a reconstruction using only a small
HRTEM experiments in thin-object regions. How many number of images (e.g., less than five)? What exactly is
nanometers are roughly considered as ‘thin’? sacrificed when reducing the number of images?
Q9.4 How is the thickness dependence of a wavefunction re- Q9.21 Is there an advantage of recording an equidistant focal
lated to the atomic charge number? series over recording an irregularly spaced series?
Q9.5 What kind of artifacts are introduced even by a ‘neutral’ Q9.22 State three reasons why the application of FSR can yield
microscope, which just produces a magnified image of an advantage over taking just one CS-corrected image at
the object wavefunction? the optimum defocus?
Q9.6 What would you expect to see in images of a thin object Q9.23 Would it be a good idea to apply FSR to amorphous ma-
made with a ‘neutral’ microscope? terials or to nanoparticles lying loosely on a support film?
Q9.7 If a constant-phase-shift microscope was available Q9.24 Who is the third Austrian named in the “People” section
(Zernike phase plate), how many images would be re- at the end of the chapter? Which view in physics/optics
quired to restore the object wavefunction? do they share?
 264 9 Focal-Series Reconstruction Appendix

Text-Specific Questions Reimer L (1984) Transmission Electron Microscopy. Springer

Series in Optical Sciences, vol. 36. Springer, Berlin Heidel-
berg New York Tokio (Useful as background to section 9.2.)
Part I

T9.1 Derive the fundamental Eq. 9.12, which is one possible Tillmann K, Thust A, Urban K (2004) Spherical aberration cor-
definition of the Scherzer defocus ZS, by requiring that rection in tandem with exit-plane wave function reconstruc-
χ(g*, ZS) = −2 π/3, where g* is the position of the local tion: interlocking tools for the atomic scale imaging of lattice
minimum of χ. defects in GaAs. Microsc Microanal 10:185–198 (Example
T9.2 The local minimum of the aberration function g* can be of application)
easily recognized as a dip in the plateau of the Scherzer
CTF displayed in Fig. 9.6d. What property of the respec-
tive helical surface of Fig. 9.8a, d changes at the specific Specific References
frequency g*?
T9.3 Derive the fundamental Eq. 9.16, which defines the op- Bethe H (1928) Theorie der Beugung von Elektronen an Kristal-
timum defocus Zopt, by setting ∇χ(gint, Zopt) = −∇χ(gmax, len. Ann Phys 87:55–129 (For the Bethe–Bloch formalism)
Zopt), and by solving the resulting third-order equation. Gerchberg RW, Saxton WO (1972) Practical algorithm for deter-
Take care with the signs! It might be helpful to use |Z| in mination of phase from image and diffraction plane pictures.
combination with compensating signs. Optik 35:237
T9.4 Derive the optimum value of the spherical aberration CS Kübel C, Thust A (2006) TrueImage – A software package for
and the thereby resulting minimal achievable radius Rmin focal-series reconstruction in HRTEM. NATO Science Series,
of the point-spread function for a CS-corrected microscope Series II. Math Phys Chem 211:373–392
by demanding that ZS of Eq. 9.12 equals Zopt of Eq. 9.16. Misell DL (1973) Examination of an iterative method for solu-
T9.5 What is the purpose of the first and the second terms in tion of phase problem in optics and electron optics. 1. Test
the enumerator of the linear filter function of Eq. 9.30? calculations. J Phys D 6:2200–2216
Chapter 9

T9.6 What important property is reflected by the denominator Schiske P (2002) Image reconstruction by means of focus series.
of the linear filter function of Eq. 9.30? J Microsc 207:154–154 (Reprint of the original 1968 paper)
T9.7 Explain why the titanium atom columns produce stronger
phase peaks in the image of Fig. 9.15 (right-hand side)
than the barium atom columns, although titanium is the Diffraction, Imaging, and Cs-correction (i. e.
lighter and therefore also the weaker scattering element.
T9.8 In the reconstructed phase of Fig. 9.16 it can be observed ‘Forward’ Theory)
that the two polygonal structure units stacked above each
other are absolutely identical. However, in both input Bethe H (1928) Theorie der Beugung von Elektronen an Kristal-
images of Fig. 9.14 one can observe clear differences len. Ann Phys 87:55–129 (The Bethe–Bloch formalism)
between the upper and the lower units. How can one and Cowley JM, Moodie AF (1957) The scattering of electrons by
the same image exhibit different contrast for identical atoms and crystals. 1. A new theoretical approach. Acta Cryst
object features? 10:609–619 (The Multislice Approach)
T9.9 When comparing the number of intensity maxima in Goodman P, Moodie AF (1974) Numerical evaluation of n-beam
the CS-corrected Scherzer image of Fig. 9.18c with the wave-functions in electron-scattering by multi-slice method.
number phase maxima in Fig. 9.18d one finds that there Acta Cryst A30:280–290 (More Multislice)
are many more intensity maxima than phase maxima. Haider M, Rose H, Uhlemann S, Schwan E, Kabius B, Urban
Whereas the phase maxima are all located at Ga or As (1998a) A spherical-aberration-corrected 200 kV transmis-
atomic column positions as is expected, some of the in- sion electron microscope. Ultramicroscopy 75:53–60 (Using
tensity maxima are located at empty crystal positions and the corrector in Sect. 9.3.3.)
are thus artifacts. Which mechanism can be responsible Haider M, Uhlemann S, Schwan E, Rose H, Kabius B, Urban
for the ‘ghost’ atoms in the CS-corrected Scherzer image? K (1998b) Electron microscopy image enhanced. Nature
392:768–769 (Using the corrector in Sect. 9.3.3.)
Humphreys CJ (1979) Scattering of fast electrons by crystals.
References Rep Prog Phys 42:1864 (For Bloch waves)
Jia CL, Lentzen M, Urban K (2003) Atomic-resolution imag-
ing of oxygen in perovskite ceramics. Science 299:870–873
(Negative CS)
Jia CL, Lentzen M, Urban K (2004) High-resolution transmis-
General References sion electron microscopy using negative spherical aberration.
Microsc Microanal 10(2):174–184 (Negative CS)
Humphreys CJ (1979) Scattering of fast electrons by crystals. Lentzen M, Jahnen B, Jia CL, Thust A, Tillmann K, Urban K
Rep Prog Phys 42:1825 (Useful as background to section (2002) High-resolution imaging with an aberration-corrected
9.2.)
 Appendix References 265

transmission electron microscope. Ultramicroscopy 92:233– Kübel C, Thust A (2006) TrueImage – A software package for
242 (First applications of CS correction) focal-series reconstruction in HRTEM. NATO Science Series,
Lichte H (1991) Optimum focus for taking electron holograms. Series II. Math Phys Chem 211:373–392

Part I
Ultramicroscopy 38(1):13–22 (For his defocus Zopt) Misell DL (1973) Examination of an iterative method for solu-
Reimer L (1984) Transmission Electron Microscopy. Springer tion of phase problem in optics and electron optics.1. Test
Series in Optical Sciences, vol. 36. Springer, Berlin Heidel- calculations. J Phys D 6:2200–2216 (The Misell algorithm)
berg New York Tokio Thust A, Lentzen M, Urban K (1994) Nonlinear reconstruction
Rose H (1990) Outline of a spherically corrected semiaplanatic of the exit plane-wave function from periodic high-resolution
medium-voltage transmission electron-microscope. Optik electron-microscopy images. Ultramicroscopy 53:101–120
85:19–24 (Using the corrector in Sect. 9.3.3) (Use of stochastic algorithms)
Scherzer O (1949) The theoretical resolution limit of the electron Thust A, Coene WMJ, de Beeck MO, Van Dyck D (1996) Fo-
microscope. J Appl Phys 20:20–29 (For his defocus) cal-series reconstruction in HRTEM: Simulation studies on
non-periodic objects. Ultramicroscopy 64:211–230 (Using
general least-squares)
Linear FSR Thust A, Lentzen M, Urban K (2000) Scanning Microscopy
Suppl 1:435 (Use of stochastic algorithms)
Saxton WO in Advances in Electronics and Electron Physics.
(1978). Academic Press, London, Suppl. 10, Ch. 9.7. When
you are thinking about Eq. 9.30. Numerical Aberration Correction
Saxton WO (1980) Correction of artifacts in linear and non-linear
high-resolution electron-micrographs. J Microsc Spectrosc Thust A, Overwijk MHF, Coene WMJ, Lentzen M (1996) Nu-
Electron 5:665–674 (When you are thinking about Eq. 9.30) merical correction of lens aberrations in phase-retrieval HR-
Saxton WO (1986). Proc. 11th Int. Congr. for Electron Micros- TEM. Ultramicroscopy 64:249–264 (Numerical aberration

Chapter 9
copy, Kyoto, post deadline paper 1 (unpublished). When you correction; dislocation core)
are thinking about Eq. 9.30.
Saxton WO (1994) What is the focus variation method – is it
new – is it direct. Ultramicroscopy 55:171–181 (When you Materials Science Applications of FSR
are thinking about Eq. 9.30)
Schiske P (1968). Proc 4th Reg Congr Electron Microscopy, Barthel J, Weirich TE, Cox G, Hibst H, Thust A (2010) Struc-
Rome, 1, 145. If you can find it when you are thinking about ture of Cs-0.5[Nb2.5W2.5O14] analysed by focal-series recon-
Eq. 9.30. struction and crystallographic image processing. Acta Mater
Schiske P (2002) Image reconstruction by means of focus series. 58:3764–3772 (Catalyst)
J Microscopy 207:154–154 (When you are thinking about Van Dyck D, Jinschek JR, Chen FR (2012) ‘Big Bang’ tomogra-
Eq. 9.30. Reprint of Schiske P (1968)) phy as a new route to atomic-resolution electron tomography.
Nature 486:243–246 (and 489, 460. Graphene)
Houben L, Thust A, Urban K (2006) Atomic-precision determi-
Nonlinear FSR nation of the reconstruction of a 90 degrees tilt boundary in
YBa2CU3O7-delta by aberration corrected HRTEM. Ultrami-
Allen LJ, Oxley MP (2001) Phase retrieval from series of images croscopy 106:200–214 (Near-picometre-precision)
obtained by defocus variation. Optics Commun 199:65–75 Jia CL, Thust A (1999) Investigation of atomic displacements
(The transport-of-intensity formalism) at a Sigma 3 {111} twin boundary in BaTiO3 by means
Coene WMJ, Thust A, de Beeck MO, Van Dyck D (1996) Max- of phase-retrieval electron microscopy. Phys Rev Lett
imum-likelihood method for focus-variation image recon- 82(25):5052–5055 (Near-picometre-precision)
struction in high resolution transmission electron microscopy. Jia CL, Thust A, Urban K (2005) Atomic-scale analysis of the
Ultramicroscopy 64:109–135 (Using general least-squares) oxygen configuration at a SrTiO3 dislocation core. Phys Rev
Gerchberg RW, Saxton WO (1972) Practical algorithm for deter- Lett 95:article #225506 (Dislocation core)
mination of phase from image and diffraction plane pictures. Jia CL, Rosenfeld R, Thust A, Urban K (1999) Atomic structure
Optik 35(2):237 (The Gerchberg-Saxton algorithm) of a Sigma=3, {111} twin-boundary junction in a BaTiO3 thin
Kirkland EJ (1984) Improved high-resolution image-processing film. Phil Mag Lett 79(3):99 (Twin arrangement in barium
of bright field electron-micrographs. 1. Theory. Ultramicros- titanate)
copy 15:151–172 (Using general least-squares) Jinschek JR, Yucelen E, Calderon H, Freitag B (2011) Quanti-
Kirkland EJ, Siegel BM, Uyeda N, Fujiyoshi Y (1985) Improved tative atomic 3-D imaging of single/double sheet graphene
high-resolution image-processing of bright field electron-mi- structure. Carbon 49:556–562 (Graphene)
crographs. 2. Experiment. Ultramicroscopy 17:87–103 (Us- Kisielowski C, Hetherington CJD, Wang YC, Kilaas R, O’Keefe
ing general least-squares) MA, Thust A (2001) Imaging columns of the light elements
carbon, nitrogen and oxygen with sub Angstrom resolution.
 266 9 Focal-Series Reconstruction

Ultramicroscopy 89:243–263 (Applications include e.g. the

first reconstruction of light atoms)
Mahalingam K, Eyink KG, Brown GJ, Dorsey DL, Kisielowski
Part I

CF, Thust A (2006) Quantifying stoichiometry of mixed-cat-

ion-anion III–V semiconductor interfaces at atomic resolu-
tion. Appl Phys Lett 88(9):article #091904 (Semiconductor
heterostructures)
Mahalingam K, Eyink KG, Brown GJ, Dorsey DL, Kisielowski
CF, Thust A (2008) Compositional analysis of mixed-cat-
ion-anion III–V semiconductor interfaces using phase re-
trieval high-resolution transmission electron microscopy. J
Microscopy 230:372–381 (Semiconductor heterostructures)
Thust A, Rosenfeld R (1998) State of the art of focal-series re-
construction in HRTEM Proc ICEM 14, Cancun, Mexico. vol.
1., pp 119–120 (Use N images to reduce noise)
Tillmann A, Thust K, Urban K (2005) Spherical aberration cor-
rection in tandem with exit-plane wave function reconstruc-
tion: interlocking tools for the atomic scale imaging of lattice
defects in GaAs. Microsc Microanal 10(2):185–198 (A struc-
tural analysis of the stacking fault in GaAs)
Chapter 9
 
Direct Methods for Image
10

Part I
Interpretation

Chapter Preview

The ultimate goal of TEM is not to obtain nice images but ture, which is then ‘refined’ in a second step by quantitative
to advance materials science. This means that TEM has to fitting with the experimental intensities. In TEM a direct
evolve from describing to understanding materials prop- method has to ‘undo’ the blurring effect of the imaging pro-
erties. Understanding means matching observations with cess and the scattering of the electrons by the object.
ab-initio calculations. Since all the structure-property rela-
tions are encoded in the positions of the atoms, they form
the ultimate language between theory and experiment. TEM
can thus be considered as a communication channel between
object and observer, and the images as data planes from
which the 3D atom positions have to be extracted quantita-
tively. The resolution of the newest generation of TEMs is
sufficient to determine the atom positions quantitatively. The
key to successful quantitative refinement is the availability
of a pertinent parametric model of the images. The model
includes all the ingredients needed to perform a computer
simulation of the image. It is parametric in the quantities of
Fig. 10.1 Scheme of the principle of model-based fitting
interest, such as the atom positions. Quantitative refinement
is then done by fitting the model to the image with respect to
the unknown quantities which are the atom positions. How- The term ‘direct methods’ actually has a special meaning
ever, due to the aberrations of the TEM this information in crystallography. They are actually a ‘family’ of methods
is entangled in the images. Hence a quantitative structure that are used to estimate the phases of the Fourier transform

Chapter 10
refinement of atom positions without prior knowledge is an (the FT used in previous chapters) to determine the scatter-
optimization problem in a high dimensional space. ing ‘density’ starting with the corresponding magnitudes.
Such methods are widely used to determine the structure of
In such a problem, convergence to the global optimum proteins (so-called protein crystallography). In this chapter
solution cannot be guaranteed unless the starting point is we start as in Chap. 9 with focal-series reconstruction (FSR)
sufficiently close to the global optimum. This means that and then delve into other methods. So the FSR is an exam-
we should use a ‘direct method’ that is able to derive a suf- ple of a ‘direct method’. It may take at least two lectures
ficiently accurate structure model directly from the images to work through this chapter and a lot longer to understand
from which, in a second stage, we can start the quantitative it from scratch. This Chapter builds on the knowledge you
refinement. The first step is called ‘resolving’ and the sec- have from W&C and the preceding chapters; you’ll find
ond step is called ‘refining’ the atomic structure. The whole Chap. 9 especially helpful since we start with the FSR but
process is called model-based fitting and is summarized in with different words and a somewhat different motivation.
Fig. 10.1. These terms are borrowed from X-ray crystal- This may seem like a difficult subject for some people, espe-
lography where direct methods are used to determine the cially if they are not physicists, but it is worth persevering!
phases of the diffracted beams to ‘resolve’ the atomic struc-

© Springer International Publishing Switzerland 2016

C. B. Carter, D. B. Williams (Eds.), Transmission Electron Microscopy, DOI 10.1007/978-3-319-26651-0_10 267
 268 10 Direct Methods for Image Interpretation

10.1 Introduction translation invariant, so that the shape of the PSF is independent
of the position of the pixel.
Part I

A major problem in TEM is that the object information may be

X
i.r/ = f .rn /p.r − rn / (10.1)
strongly delocalized by the image transfer in the TEM, which n
makes the interpretation of experimental images in terms of the
structure very difficult. Over the years, different so-called direct Thus the final image is the weighted sum of the point-spread func-
methods have been developed to overcome this problem. These tions. If we now take the limit at which the points are taken infini-
are methods that use some a priori knowledge, which is gener- tesimally close together, the sum in Eq. 10.1 becomes an integral
ally valid, irrespective of the (unknown) structure of the object. Z
At present, existing possibilities are, e.g., high-voltage TEM, i.r/ = f .r0 /p.r − r0 /dr0 (10.2)
correction of the spherical aberration in the TEM, high-angle
annular dark-field STEM, focal-series reconstruction, and off-
axis holography. In the present chapter, we will focus on the which in mathematics is called a convolution product.
focal-series reconstruction method. As you know from Chap. 9,
this method makes use of a series of images collected at different i.r/ = f .r/  P.r/ (10.3)
defocus values in order to retrieve the complex exit wave, which
is formed at the exit plane of the object under study. Next, we’ll This result is valid in 1D, 2D and even in 3D (tomography)
explain the basics of the so-called channeling theory in order to imaging.
show how the reconstructed exit wave could be interpreted rel-
atively easily in terms of the structure of the object under study. The blurring limits the resolution of the imaging device. Indeed,
Finally, we’ll show you how to extract quantitative structure in- when two points are imaged with a distance smaller than the
formation using statistical parameter estimation theory. We thus ‘width’ of the point-spread function, their images will overlap
combine a direct method with a theoretical model describing so that they become indistinguishable. The resolution, defined as
the image intensity distribution to refine the unknown structure the smallest distance that can be resolved, is then related to the
parameters quantitatively. In this way, we can determine atom width of the point spread function.
column positions with picometer precision.
It is very informative to describe the imaging process in Fourier
space. Let us call the Fourier transforms of f(r), p(r) and i(r)
and F(g), P(g) and I(g). The convolution theorem states that the
10.2 Basics of Image Formation Fourier transform of a convolution product is a normal product.
If we thus Fourier transform we obtain

An explanation of the focal-series reconstruction method re- I.g/ = F.g/:P.g/ (10.4)

Chapter 10

quires a basic understanding of the image formation in the TEM.

The principles of image formation are well established and are F(g) represents the content of the object in the spatial-frequency
usually derived from the theory of Fourier optics. For complete- domain (which is also known as the Fourier domain) and P(g)
ness, and because we sometimes use slightly different nomencla- is the transfer function. Every imaging device can thus also be
ture, we’ll begin this section by reviewing the basics. characterized by its transfer function (which is also known as
the band filter) P(g). P(g) describes the magnitude with which a
spatial frequency g is transferred through the device. This result
is shown graphically in Fig. 10.3. The noise, N, is also indicated.
10.2.1 Real imaging The transfer function P(g) affects both the amplitude and the
phase of the spatial frequency component F(g).

The quality of an imaging device is characterized by the ‘blurred’ The resolution of the instrument ρ is defined from the cut-off 1/g
image p(r) it makes of a sharp point with r the position vector in between signal and noise beyond which no spatial information is
the image plane. p(r) is called the point-spread function (PSF). transferred. This is the type of resolution in the sense as defined
In the field of signal processing this is called the ‘impulse-re- by Rayleigh. It is inversely related to the ‘width’ of the point-
sponse function’. Let us now consider a very simple case of real spread function. If the transfer function would be constant (i.e.,
imaging of a real object such as the picture in Fig. 10.2. (You’ll perfectly flat) in the whole spatial frequency range, the impulse
recognize a link to Fig. 31.27 of W&C.) The object is described response function would be a delta function so that i(r) = f(r).
by the function f(r). Now every pixel rn in an object can be con-
sidered as an independent point. Thus the pixel rn with value If the point-spread function is known, the original image can be
f(rn) is imaged as a blurred disk around rn given by f(rn)P(r−rn). restored up to the resolution ρ by the following steps: Fourier
If we now assume that the imaging is linear, the image of a set transform i(r) to I(g), multiply I(g) by 1/P(g) and Fourier trans-
of pixels is the same as the sum of the images of the pixels. form back to i(r). This is called image restoration or deblurring.
Furthermore we can assume that the imaging characteristics are 1/P(g) is called a deconvolution filter. However, a problem oc-
 10.2 Basics of Image Formation 269

Part I
Fig. 10.3 Sketch of a transfer function

can be blurred by the atmosphere, by the telescope, and by the

photoplate or camera. Let us denote the respective point spread
functions of the successive steps by p1(r), p2(r), p3(r) … then the
final point spread function is given by

p.r/ = p1 .r/  p2 .r/  p3 .r/ (10.5)

and its Fourier transform

P.g/ = P1 .g/:P2 .g/:P3 .g/ (10.6)

The total transfer function is thus the product of the respective

transfer functions. The resolution is then mainly limited by the
weakest (i.e., poorest) step in the imaging chain.

10.2.3 Coherent Imaging

Chapter 10
In case of coherent imaging, as in high-resolution TEM, the ob-
ject function Φ(r) (which is also called the object or exit wave)
and the point-spread function p(r) are both complex functions
with an amplitude and a phase part. Coherent imaging is linear in
Fig. 10.2 Principle of linear image formation original image (a); Gaussian point- terms of these complex functions. Therefore, the image function
spread function with different sizes (b); blurred image (c), and deblurred image(d) ψ(r), also called an image wave, is a complex function as well,
which is given by:

curs for those values of g for which the transfer function is zero  .r/ = ˚.r/  p.r/ (10.7)
since dividing by zero will yield unreliable results. A modified
or in Fourier space
type of a deconvolution operator that takes care of this problem
is the so-called Wiener filter. The attainable resolution after de-  .g/ = ˚.g/:P.g/ (10.8)
blurring depends on the resolution of the imaging device. Fig-
ure 10.2d shows an example of image restoration. However, the detection of the image on a photoplate or a CCD
camera is nonlinear. In this step of the image formation, the in-
tensity, i.e., the square of the amplitude of the image wave is
recorded and the phase is lost:
10.2.2 Successive Imaging Steps
I.r/ = j .r/j2 (10.9)

In many cases an image is formed through many imaging steps If the camera also causes blurring, then the image I(r) will be
or devices. Each of these steps (if linear) has its own point-spread further convoluted with the point-spread function of the camera,
function. For instance, the image of a star through a telescope as explained in Sect. 10.2.2.
 270 10 Direct Methods for Image Interpretation
Part I

Fig. 10.4 Imaginary parts of the point spread function (a) and the transfer function (b) of a TEM. Dotted line: without incoherent damping. Solid line: with incoher-
ent damping. Axes are in Glaser units G1 = CS1=4 3=4

If we could retrieve the image phase by ‘holographic’ methods, be included by averaging the image intensity over the different
it would be possible to deconvolute the complex point-spread focus values and beam directions.
function p(r) of the TEM and then reconstruct the object wave
and enhance the resolution. In the case of a weak object, the averaging over the incoherent
fluctuations of the microscope results in a damping of the phase
transfer function.

10.2.4 High-Resolution Imaging in the TEM P.g/ = A.g/ exp Œ−ig D.˛; ; g/ (10.12)
Chapter 10

This is called the coherent approximation. More on the damping

As explained in Chap. 28 of W&C the transfer function P(g) of factor D is given in Chap. 28 of W&C.
the TEM is given by
The imaginary parts of the point-spread function and the trans-
P.g/ = A.g/ exp Œ−i.g/ (10.10) fer function including the damping of an HREM are shown in
Fig. 10.4.
A(g) represents the physical aperture with radius gA selecting the
imaging beams; thus 1; for jgj  gA The resolution as defined from the first zero of the transfer function
A.g/ =
0; for jgj > gA 1 1=4
S =  0:65CS 3=4 = 0:65G1 (10.13)
The total phase shift due to spherical aberration and defocus g
combined is
is generally accepted as the standard definition of the structural
1
.g/ = CS 3 g4 + "g2 (10.11) resolution of a TEM. (Note the unit we are using.) It is often also
2 called the point resolution. It is defined as the smallest detail that
can be interpreted directly. It is also equal to the width of the
with Cs being the spherical aberration coefficient, ε the defocus, point-spread function. The information beyond the intersection
and λ the wavelength. S is transferred with a phase that is not constant and, as a con-
sequence, is redistributed over a larger image area.
For simplicity we neglect higher-order aberrations. The TEM
is also subject to incoherent aberrations such as the temporal The point where the transfer function reaches the noise level is
incoherence (fluctuations of the defocus) and spatial incoherence called the information limit. It is possible to retrieve the phase of
(the incident beam comes in along different angles). This can the image wave, and since the complex PSF of the TEM is known
it can be deconvoluted from Eq. 10.7 so that we can reconstruct
 10.4 Interpretation of the Reconstructed Exit Wave 271

the electron wave at the exit face of the object up to the informa-
tion limit. This process is then called exit wave reconstruction.

Part I
Recently it has become possible to construct TEMs with an aber-
ration corrector, which allows to choose and reduce the spherical
aberration constant Cs [Rose (2008), Erni (2010), Schiske (1973)].
However, the shape of the point-spread function and transfer func-
tion in Fig. 10.4 remains unchanged since the axes are scaled in
universal Glaser units, which scale with the ¼ power of Cs. (See
Sect. 28.7 of W&C for a definition of the Glaser unit.) This means
that the resolution can be improved by a factor 2 if the Cs value is
reduced by a factor 16. For a TEM without spherical aberration
corrector operating at an accelerating voltage of about 300 kV, the
point resolution is of the order of 2 Å. For an aberration-corrected
TEM the point resolution is well below 1 Å. Another advantage of
a Cs-corrected microscope is that the information in the image is
much more localized. However, for a quantitative interpretation
of the images, we still benefit from using wave-reconstruction
methods combined with quantitative, model-based refinement.

10.3 Focal-Series Reconstruction

of the Exit Wave
Fig. 10.5 Experimentally reconstructed exit wave for YBa2CuO8. a Reconstruct-
The amplitude of the high-resolution image can be directly de- ed phase; b structure model; c experimental image
rived as the square-root of the intensity. However, in order to
reconstruct the whole wavefunction one has to recover the phase. ries reconstruction method is explained in much more detail in
We can recover the phase using holographic methods, as we saw Chap. 9 – what we want to do here is use it for our direct methods.
in Chap. 8. Two methods exist for this purpose: off-axis hologra-
phy or focus-series reconstruction, which is a kind of in-line ho- You should note that, once the exit wave is reconstructed, it is
lography. Off-axis holography is based on the original idea of Ga- in principle possible to recalculate all the images of the focal
bor, that is, to retrieve the phase from the interference between the series within the noise level. Hence, the reconstructed exit wave

Chapter 10
image wave and a reference wave [Gabor (1948), Lichte (1991)] contains all object information that is experimentally attainable.
. In off-axis holography, the beam is split by an electrostatic bi- Thus, in practice, we do not have to store the original images,
prism into a reference beam, and a beam that traverses the object. only the reconstructed wave.
Interference of both beams in the image plane then yields fringes,
the positions of which yield the phase information. Ideally, the reconstructed exit wave will be free from any imag-
ing artifacts, which means that the visual interpretability of the
In order to retrieve this information we need a really good reconstruction is enhanced considerably for thin objects when
high-resolution camera (CCD), a powerful image processor, and a compared to the original experimental images. In practice, the di-
field-emission source to provide the necessary spatial coherence. rectly interpretable resolution is pushed from the so-called point
In the focal-series reconstruction method, the focus is used as a resolution down to the information limit.
controllable parameter so as to yield focus values from which
we can extract both amplitude and phase information [Thust et Figure 10.5 shows the exit wave of an object of YBa2Cu4O8 (the
al. (1996), Schiske (1973), Saxton (1978), Kirkland (1984), Van high-TC superconductor), which was historically the first exper-
Dyck and Coene (1987), Van Dyck et al. (1993), Coene et al. imental result obtained with focal-series reconstruction.
(1996), Thust et al. (1996)]. Images are captured at very close
focus values so as to collect all information in the 3D image
space. A first estimate of the complex image wave is obtained
by a weighted sum of these focal images, which is then refined 10.4 Interpretation
using a maximum-likelihood fitting procedure. Once the image
wave is reconstructed, we can eliminate the TEM aberrations by of the Reconstructed Exit Wave
deconvolution with the complex transfer function of Eq. 10.12. In
this way the whole imaging process that takes place in the TEM is In this section, we’ll explain how the reconstructed exit wave can
reversed (undone), and the final result is the electron wave at the relatively easily be interpreted in terms of the structure of the ob-
exit face of the object, which we call the exit wave. The focal-se- ject we are studying. In particular, we’ll make use of the so-called
 272 10 Direct Methods for Image Interpretation

Fig. 10.6 Schematic drawing of elec-

tron channeling. a The atoms of the
columns are indicated by colored cir-
Part I

cles. b Lens action of the atoms

channeling theory. Electron channeling occurs when the incident consider each atom as a thin lens as shown in the figure. Each
electron beam is parallel to the atom columns of an object, as ‘atom’ lens is a perfect lens even if the shape of the electrostatic
occurs in a crystal and may occur close to a particular crystal de- potential is parabolic. By passing ‘through’ successive lenses,
fect. Then, the electrons are trapped in the electrostatic potential the electron wave is focused at periodic distances. This periodic
of the atomic column and they scatter dynamically along that distance is called an extinction distance [Van Dyck and Chen
column. This picture provides physical insight and explains why (1999)]. It is a function of the average mass density of the col-
a one-to-one correspondence is maintained between the exit wave umns, which in turn is a function of the atomic number of the
and the projected structure, even in the case of strong dynamical atoms and the repeat distance of the atoms in the column. Note
scattering. Moreover, the theory is very useful when we want to that when the crystal thickness is equal to the extinction distance,
invert the dynamical scattering, that is, to derive the projected the exit wave is equal to the entrance wave we would have if the
structure from the exit wave. Finally, we can use it to determine column had disappeared.
the composition of an atom column with single atom sensitivity.
This effect is called dynamical extinction. Note also that for cer-
tain thicknesses, the focusing effect of a light column at the exit
Chapter 10

face can be stronger than that of a heavy column. Therefore, it is

10.4.1 Electron Channeling possible to visualize light columns in the presence of a column
of heavy atoms. This is a big advantage of electron diffraction
when compared to X-ray diffraction (XRD). You can find more
It is well known that exit waves and even high-resolution TEM on channeling theory in references [Van Dyck and Op de Beeck
images often show a one-to-one correspondence with the con- (1996), Van Dyck (2002), Geuens and Van Dyck (2002), Geuens
figuration of projected atom columns when a crystal is viewed and Van Dyck (2005), Van Aert, et al. (2007), Van Dyck and
along a zone axis, i.e., parallel to these columns. In other words, Chen. (1999), Cowley and Moodie (1957)].
the image looks like the projected structure! This correspondence
can occur if the distance between adjacent columns is not too You may find the following analogy helpful. If we assume that the
small and the resolution of the TEM is sufficient. From this ob- speed of the electron is constant, then the depth z in the column is
servation, it has been suggested that for crystals viewed along proportional to the time travelled. So now we can say that the path
a zone axis with sufficient separation between the columns, the of the electron along the column is the same as its motion would
complex wave function at the exit face of the crystal mainly de- be through a 2D assembly of potential wells as a function of time:
pends on the projected structure, i.e., on the type and position of the 2D wells correspond to the potentials of the different projected
atom columns. The physical reason for local dynamical diffrac- atom columns. Thus the equation describing the motion of the
tion is the channeling of the electrons along the atom columns electron is the same as the time-dependent Schrödinger equation
parallel to the beam direction. Due to the positive electrostatic in which time t is replaced by depth, z, divided by the velocity.
potential of the atoms, a column of atoms can act as a guide, or
channel, for the incident electron: the electron can scatter dynam- The basic result of the channeling theory is that the wavefunction
ically without leaving the column. at the exit face of a column is expressed as

This channeling effect is schematically represented in Fig. 10.6.

   
E
 .r; z/ = 1 + exp −i kz − 1 .r/ (10.14)
A simple intuitive way to understand electron channeling is to E0

Licensed to ([email protected])
 10.4 Interpretation of the Reconstructed Exit Wave 273

Part I
Fig. 10.7 a Experimentally reconstructed exit wave of an Al sample with a Σ5 Fig. 10.8 Reconstructed exit wave of STO with two defocus images a, b Two
interface. b The amplitude of ψ(r) −1 is peaked at the positions of the columns typical images of the focal series. c Phase of the reconstructed exit wave (repro-
and c the phase of ψ(r) −1 yields the ‘mass’ of the columns. These results en- duced from Jia and Thust 1999)
abled us to determine the positions of the Cu atoms which play a role in the
mechanical properties of the defect (courtesy of C. Kisielowski)
10.4.2 Argand Plot
This result holds for each isolated column. In a sense, the whole
wavefunction is uniquely determined by the eigenstate ϕ(r) of the A convenient way to visualize the effect of electrons passing
Hamiltonian of the projected columns and its energy E, which through a column is by plotting each pixel of the complex exit

Chapter 10
are both functions of the ‘density’ of the column and the crystal wave of a column in an Argand plot. (This discussion may re-
thickness z. mind you of the phasor diagrams in Chap. 27 of W&C – which is
not a coincidence.) This is a representation in which each pixel is
It is clear from Eq. 10.14 that the amplitude of ψ(r) −1 is peaked plotted as a point in a complex plane with its x-coordinate corre-
at the center of the column and the phase of ψ(r) −1 increases sponding to the real pixel value and the y-coordinate correspond-
linearly with depth. The phase is inversely related to the ‘den- ing to the imaginary pixel value. This is shown in Fig. 10.9.
sity’ of the column. In this way the exit wave still retains a
one-to-one correspondence with the projected structure. You This representation of electron channeling was first introduced by
can see the effect of this in Fig. 10.7 for the experimentally re- Sinkler and Marks [Sinkler and Marks (1999)]. From Eq. 10.14
constructed exit waves of a Cu sample with a Σ = 5 grain bound- it follows that an Argand plot of the pixels of the atom columns
ary. Furthermore, it is possible to parameterize the exit wave in of the same type but at different thicknesses is expected to be a
terms of the atomic number Z and the interatomic distance d of circular locus which starts at the origin and centers on the x-axis.
the atoms constituting the column. This procedure allows us to The point of the origin (0,0) represents the entrance wave (zero
retrieve the projected structure of the object from matching it thickness). The radius of the circle should be identical for all
with the exit wave. atom types. The constant angular increment corresponds to an
extra atom in the column which is different for each atom type.
In practice, it is possible to retrieve the positions of the columns
with high precision (10 pm) and to obtain a rough estimate of the This so-called channeling map is shown in Fig. 10.10; the
density of the columns. scattering waves are simulated for different column types in
[100] orientation [Geuens and Van Dyck (2002), Geuens and
Figure 10.8 shows a reconstructed exit wave from STO together Van Dyck (2005)]. Finding a universal radius for all atom types
with two defocus images. From this exit wave the atom positions may come as a surprise, but it can be explained [Op de Beeck
were determined with a precision of 2 pm in full agreement with et al. (1995)]. It is possible that the reconstructed exit wave is
theory (DFT calculations). not exactly positioned at the exit face of a column. This is the
 274 10 Direct Methods for Image Interpretation
Part I

Fig. 10.11 Schematic representation of surface roughness due to different ver-

tical atom positions compared to neighboring atom columns. This will result in
different defocus values for each atom column

Fig. 10.9 Principle of the representation in Argand space. The complex value of
a pixel is plotted with the real value along the horizontal axis and the imaginary 10.5 Quantitative Structure
value at the vertical axis. This representation is particularly informative to visual-
ize the effect of electron channeling Refinement
In the previous sections we saw how we can reconstruct the
exit wave and how we interpret this reconstruction in terms of
the structure by means of the channeling theory. Such a direct
method may be used to derive the structure relatively easily in a
qualitative manner. Furthermore, with this method we can now
obtain sub-Å resolution, allowing us to visualize the individual
projected atom columns of materials with columnar structures,
such as perfect crystals or crystals containing defects in the struc-
ture, viewed along a main zone axis. Although, direct methods
are very useful when we want to derive an approximate structure
of the object under study, you may wonder if qualitative structure
determination will be sufficient for the materials science of the
future.

In materials science, the last decades are characterized by an

Chapter 10

evolution from macro- to micro- and, more recently, to nano-

technology. In nanotechnology, nanomaterials play a critical
role, of course. In saying nanomaterials, we include nanoparti-
cles, nanotubes, and single atomic layers, and they may be lay-
ered magnetic materials, superconducting materials, etc. Many
Fig. 10.10 Argand plots for columns of different type in [100] zone axis ori-
entation. The radius of the circle is independent of the atom type. The angular
of the interesting properties of these materials are related to
increment corresponds with adding a single atom to the column. The angular
their structure. Therefore, one of the central issues in materials
increment of a single Au atom is about eight times that of a single Si atom which science is to understand the relations between the properties of
corresponds with the ratio of the respective atomic numbers Z a given material on the one hand and its structure on the other. A
complete understanding of this relation, combined with recent
progress in building nanomaterials atom by atom, will enable
case when the exit surface of an object is not flat as shown in materials science to evolve into materials design, that is, from
Fig. 10.11. describing and understanding towards predicting materials
with interesting properties. In order to understand the proper-
Then the exit wave is propagated further over a certain distance, ties–structure relation, experimental and theoretical studies are
which is called the defocus distance, or simply the defocus. Be- needed. Essentially, theoretical studies allow us to calculate
cause of this propagation, the pixel in the Argand plot moves the properties of materials with known structure, whereas ex-
along a different circle. We call this the defocus circle. The angle perimental studies allow us to characterize materials in terms
along this circle is proportional to the defocus. This plot is shown of structure. In practice, however, the combination of the two
in Fig. 10.12. approaches is not yet generally doable. One of the reasons is
that present experimental characterization methods, of which
Figure 10.13 then shows how the surface profile can be deter- high-resolution TEM is the most appropriate one, cannot yet
mined from the defocus circle for an Au column in [110] zone always determine atom positions with picometer precision
axis orientation. [Muller (1998), Muller (1999), Kisielowski, et al. (2001)].
 10.5 Quantitative Structure Refinement 275

Part I
Fig. 10.12 a Simulation of the effect of defocus (from −150 Å to +150 Å) for an atom column containing 5 Au atoms leading to an extra defocus circle. b Defocus
circles shown for different numbers of atoms in a column. All circles go through a common reference point on the Argand plot

Chapter 10

Fig. 10.13 a Structure model in which two different types of columns are indicated together with the number of atoms. b Scheme of the wedge shape. c By deter-
mining the defocus for every column one can determine the defocus map which then reveals the surface profile (d)
 276 10 Direct Methods for Image Interpretation

Fig. 10.14 Illustra- rive such ‘expectation’ models. For example, the ‘expectation’
tion of the differ- model of a high-resolution TEM image is closely related to the
ence between reso- expression given by Eq. 10.9 and the expectation model of a
Part I

lution and precision reconstructed exit wave is closely related to the expression given
by Eq. 10.7 using Eqs. 10.8, 10.11 and 10.12. For both models,
an expression for the object function ψ(r) or φ(r) can be derived
from the channeling theory using, for example, Eq. 10.15. The
parameters of the expectation models include both structure
parameters and microscope settings, characterizing the object
under study and the microscope, respectively.

In the next step, the unknown parameters of the model are esti-
mated using statistical parameter estimation theory. We do this
10.5.1 Precision Versus Resolution by optimizing a criterion known as the ‘goodness of fit’, such
as the likelihood function or the least-squares sum. This crite-
rion quantifies the similarity between the observations and the
A common misunderstanding is that one is inclined to think that expectation model. The estimated parameters we want are those
a precision of the order of 1 pm requires a further improvement for which this criterion is optimal. In essence we are looking
of the spatial resolution to 1 pm, respectively, which is far be- for the optimal value of a criterion in a parameter space whose
yond the present possibilities. The reason for this misunderstand- dimension is equal to the number of parameters to be estimated.
ing is that structure determination is, up to now, mainly based This search for the global optimum of the criterion of goodness
on visual inspection of high-resolution images or reconstructed of fit is an iterative, numerical, optimization method. Generally,
exit waves. From the point of view of visual inspection, good the dimension of the parameter space is high. Consequently, it
resolution is important. Indeed, resolution allows us to visually is quite possible that the optimization procedure ends up at a
distinguish neighboring components in the observations. local optimum instead of at the global optimum of the criterion
of goodness of fit, so that the wrong structure is suggested. To
Precision, on the other hand, is a fundamentally different notion. solve this dimensionality problem, i.e., to find a pathway to the
In order to measure parameters precisely, we must have a physi- global optimum in the parameter space, we have to choose a
cal model of the observations; this model may be adapted to the good starting structure. Finding such a starting structure is not
observations with respect to the unknown parameters. ‘Precision’ trivial and usually requires a direct method, which may be used
then corresponds to the variance or the standard deviation of the to visually obtain an approximate structure.
measurements of the physical parameters that are thus obtained. If
the model is correct, the precision is only limited by the presence of For example, focal-series reconstruction methods in a sense in-
noise in the observations and not by the resolution. This difference vert, or equivalently, undo, the effect of lens aberrations. Conse-
Chapter 10

is summarized graphically in Fig. 10.14. In fact, an image may no quently, the exit wave that we obtain using this method can be
longer be judged by its visual quality, but should be considered as closely related to the object structure and can provide a directly
an experimental dataset from which physical parameters have to be interpretable resolution close to the information limit, which
measured as precisely as possible. In order to reach this goal, mere should just surpass the limit that allows us to discriminate the
visual interpretation of the observations is inadequate: it must be individual atom columns. Focal-series reconstruction methods
combined with quantitative, model-based methods. thus yield an approximate structure that may be used as a starting
point for a final numerical optimization procedure by adapting an
expectation model to the reconstructed exit wave.

10.5.2 Quantitative Model-Based Structure In a final step, the result we obtain in this way may be used as
Determination an improved starting point for a final numerical optimization
procedure by adapting the ‘expectation’ model to the original
set of images of the defocus series.
As we’ve just seen, the key to precise, quantitative structure
determination is the availability of an expectation model. Such The practical applicability and usefulness of quantitative mod-
a model will usually be physics-based, and will describe the ex- el-based structure determination has been demonstrated in the
pected pixel values of the image. In fact, this ‘expectation’ model study of a compound, Bi4 W2/3 Mn1/3 O8 Cl [Bals et al. (2006)].In
will include all the ingredients needed to perform a computer this example, the atom column positions have been determined
simulation of the images: it will be parametric in the quantities with a precision in the picometer range. Exit wave reconstruc-
of interest, such as the atom column positions. tion combined with statistical parameter estimation was used to
achieve this. Figure 10.15a shows a part of the phase of the re-
As we have seen in Sects. 10.2 and 10.3, the principles of im- constructed exit wave. This phase reveals the light oxygen atom
age formation are sufficiently well understood that we can de- columns in the presence of heavier atom columns. Although the
 10.5 Quantitative Structure Refinement 277

Part I
Fig. 10.15 Model-based structure determination of the compound, Bi4W2/3Mn1/3O8Cl. a Phase of the reconstructed exit wave. b Refined parameterized model.
c Model together with an overlay

phase of the exit wave is often considered to be the final re- structure using an FEI Titan G3 50–80 operated at 300 kV. Even
sult, here it has been used as a starting point for quantitative though the probe has been corrected for spherical aberration,
refinement of the atom column positions. A model has then been no visual conclusions could be drawn concerning the sequence
proposed to describe the expectations of the pixel values in the of the atomic planes at the interfaces. The refined parameter-
reconstructed phase. The unknown parameters of this model, ized model is shown in Fig. 10.16b, showing a close match with
including the atom column positions, have been estimated in the the experimental data. Figure 10.16c shows the experimental
least squares sense. Figures 10.15b and 10.15c show the model observations together with an overlay indicating the estimated
evaluated at the estimated parameters. positions of the columns and their atomic column types.

Next, interatomic distances have been computed from the es- Since the material is known, we assume that the composition of
timated atomic-column positions. From sets of equivalent dis- the columns away from the interfaces is known; the composition
tances, mean interatomic distances and their corresponding of the columns in the planes close to the interface (indicated
standard deviations have been calculated. The standard devia- by the symbol ‘X’) is unknown. Histograms of the estimated
tion, being a measure of the precision, ranges from 3 to 10 pm. peak volumes of the known columns are presented in Fig. 10.16d
Furthermore, a good agreement has been found when compar- and show the random nature of the result. The colored vertical
ing these results with X-ray powder diffraction data. It is also bands correspond to 90 % tolerance intervals. It is important to
found that projected distances beyond the information limit of note that these tolerance intervals do not overlap; this means
the microscope (1.1 Å) can be determined with picometer-range that we can clearly distinguish columns for which the difference
precision. in averaged atomic number is only 3 (TiO and MnO). Based on
this histogram, the composition of the unknown columns could

Chapter 10
Model-based, structure determination not only shows its ben- be identified, as shown on the right-hand side of Fig. 10.16c.
efits in the quantitative interpretation of exit waves but also Single-colored dots are used to indicate columns whose esti-
when analyzing HAADF STEM images. Even though HAADF mated peak volume falls inside a tolerance interval, whereas pie
STEM images are to a certain extent directly interpretable, this charts, indicating the presence of intermixing or diffusion, are
imaging technique may also greatly benefit from a quantitative used otherwise.
analysis such as statistical parameter estimation theory. This is
particularly the case when the difference in atomic number of The excellent chemical sensitivity that can be achieved by com-
distinct atomic column types is small or if the signal-to-noise bining HAADF STEM with statistical parameter-estimation
ratio becomes poor. A performance measure that is sensitive to theory is an advantage when we want to count the number of
the chemical composition is the scattering cross-section. Using atoms in a column with single atom precision. In this manner,
statistical parameter estimation theory, we can quantify the total the number of atoms in a nanosized Ag cluster embedded in an
intensity of scattered electrons atomic column by atomic column Al matrix could be determined [Van Aert, et al. (2011)]. Fig-
using an empirical parameterized incoherent imaging model. The ure 10.17a shows an aberration-corrected HAADF STEM im-
measurements thus obtained allow us to differentiate between age of such clusters viewed along the Œ101 N zone-axis. Using
atomic columns with different composition. As such, differences the model-based approach explained above, the parameters of
in averaged atomic number of only 3 can clearly be distinguished an empirical physics-based model have been estimated in the
in an experimental image – a feat that is impossible by means of least-squares sense.
visual interpretation only.
For the cluster in the white-boxed region, the refined model is
This is an important advantage when we study interfaces, as shown in Fig. 10.17b. Based on the estimated parameters, scat-
illustrated in the following example [Van Aert, et al. (2009)]. tered intensities have been computed for each atomic column.
Figure 10.16a shows an enlarged area from an experimental Since we can assume the thickness of the sample to be constant
HAADF STEM image of a La0.7Sr0.3MnO3-SrTiO3 multilayer over the particle area, substitution of an Al atom by an Ag atom
 278 10 Direct Methods for Image Interpretation
Part I
Chapter 10

Fig. 10.16 Model based quantitative analysis of an experimental HAADF STEM image of an La0.7Sr0.3MnO3-SrTiO3 multilayer structure. a Experimental image. b Re-
fined parameterized model. c Overlay indicating the estimated positions of the columns together with their atomic column types. d Histograms of estimated peak
volumes of the known columns

leads to an increase of the estimated intensity. However, due to

noise, we cannot interpret the scattered intensities directly in
terms of number of atoms in a column. Therefore, we have made
use of the so-called integrated classification-likelihood criterion
to relate scattered intensities to the number of atoms resulting
in Fig. 10.17c. This counting procedure has also been applied to
the same Ag cluster viewed along the [100] direction. When we
combine atom-counting results from different zone-axis orien-
tations we can retrieve the 3D atomic structure.

Fig. 10.17 Counting the number of atoms from STEM images of an Ag cluster
embedded in an Al matrix a Experimental HAADF STEM image. b Refined param-
eterized model. c Number of atoms for each atomic column
 10.5 Quantitative Structure Refinement 279

Chapter Summary

Part I
In this chapter, we have given yet another general introduc- Next, the basics of the channeling theory were explained
tion to the basic principles of image formation in the TEM! in order to show how the reconstructed exit wave could be
This time, these principles were derived from Fourier optics interpreted relatively easily in terms of the structure of the
theory. We have shown that they lead to rather simple ex- object under study. Finally, we have shown how you can
pressions describing the image intensity distribution. Next, extract quantitative structure information using statistical
we have shown how these expressions can be used to de- parameter estimation theory. We used a combination of a
scribe the focal-series-reconstruction method in real space. direct method and a theoretical model describing the image
Using this method we could then reconstruct the complex intensity distribution to refine the unknown structure pa-
exit wave. The purpose of the focal-series reconstruction rameters quantitatively. In doing this, we showed that atom
method is to invert, or equivalently, undo, the effect of lens column positions really can be determined with picometer
aberrations. Consequently, the exit wave that we then ob- precision. The same techniques can also be applied when
tain is much more closely related to the object structure, analyzing HAADF STEM images in a quantitative way, as
providing directly interpretable detail with a resolution that you will learn in Chap. 11. So we really can determine the
is close to the information limit, which itself just surpasses chemical composition of a material and count atoms.
the limit beyond which individual atom columns can be dis-
criminated. As such, the focal-series reconstruction method
may be regarded as a ‘direct method.’

Chapter 10
 280 Appendix

Appendix
Part I

People Text-Specific Questions

Severin Amelinckx was born on October 30, 1922 and died in T10.1 Discuss the concept of a Wiener filter, using the litera-
Antwerp on February 22, 2007. He was a pioneer in applying ture for more details.
TEM to the study of defects in crystals and led a prestigious T10.2 What do we mean by the term ‘coherent imaging’ and
group in applying not only HRTEM to crystals but particularly how would we describe this mathematically?
in using diffraction geometry to understand contrast in TEM T10.3 Is there any connection between dynamical extinction
images. described here and the extinction distance discussed in
Chaps. 15 and 25 of W&C.
Pierre Delavignette was born on July 3, 1931 and died on De- T10.4 Elaborate on the use of ‘model-based, structure deter-
cember 29, 2011, again in Belgium. He worked closely with his mination’.
Ph.D. advisor, Professor Amelinckx, throughout his life doing T10.5 Would HAADF STEM or phase-contrast TEM be more
classic work on multi-layer graphene (then known as graphite) affected by channeling?
and many other layer materials.

References
Self-Assessment Questions

Q10.1 How do we define the term ‘direct methods’? General References

Q10.2 How does the PSF relate to resolution and how does it
compare to the ‘impulse-response function’? Van Aert S (2012) Statistical parameter estimation theory – A
Q10.3 What is delocalization in relation to an image? (You tool for quantitative electron microscopy. Handbook of Na-
may find Chap. 28 of W&C helpful.) noscopy. Wiley VCH, pp 281–309
Q10.4 Where did you see channeling in W&C? Van Dyck D (2012) Atomic resolution electron microscopy.
Q10.5 How can we have an extinction distance along a single
Chapter 10

Handbook of Nanoscopy. Wiley VCH, pp 45–79

column of atoms?
Q10.6 What is an eigenstate of a Hamiltonian? (You may need
to look back at your physics books.) Specific References
Q10.7 What is a ∑ = 5 gb?
Q10.8 Where did you first encounter Argand diagrams? Van Aert S, Geuens P, Van Dyck D, Kisielowski C, Jinschek JR
Q10.9 If we go to all this trouble to obtain pm precision, what (2007) Electron channeling based crystallography. Ultrami-
is our precision usually in HRTEM? croscopy 107:551–558
Q10.10 Express the term ‘goodness of fit’ in a different way Van Aert S, Verbeeck J, Erni R, Bals S, Luysberg M, Van Dyck D,
that sounds like better grammar and explain why we Van Tendeloo G (2009) Quantitative atomic resolution map-
say ‘goodness of fit’. ping using high-angle annular dark field scanning transmis-
Q10.11 We say “invert, or equivalently, undo”. Clarify what we sion electron microscopy. Ultramicroscopy 109:1236–1244
mean by ‘equivalently’. Van Aert S, Batenburg KJ, Rossell MD, Erni R, Van Tendeloo
Q10.12 We say “the difference in averaged atomic number is G (2011) Three-dimensional atomic imaging of crystalline
only 3”. Explain. nanoparticles. Nature 470:374–377
Q10.13 So what exactly do we mean by the term ‘statistical Bals S, Van Aert S, Van Tendeloo G, Avila-Brande D (2006)
parameter estimation theory’? Statistical estimation of atomic positions from exit wave re-
Q10.14 We want to ‘determine the chemical composition … and construction with a precision in the picometer range. Phys
count atoms’. Why both? Rev Lett 96:096106
Coene W, Janssen G, de Op BM, Van Dyck D (1992) Phys Rev
Lett 29:37–43
Coene WMJ, Thust A, de Op BM, Van Dyck D (1996) Maximum
likelihood method for focus-variation image reconstruction
 Appendix References 281

in high resolution transmission electron microscopy. Ultra- Kisielowski C, Principe E, Freitag B, Hubert D (2001) Bene-
microscopy 64:109–135 fits of microscopy with super resolution. Physica B308–
Cowley JM, Moodie AF (1957) The scattering of electrons by 310:1090–1096

Part I
atoms and crystals. Acta Cryst 10:609–619 Lichte H (1991) Electron image plane off-axis holography of
Van Dyck D (1990) Proc. XIIth International Congress for Elec- atomic structures. In: Mulvey T, Sheppard CJR (eds) Adv
tron Microscopy (Seattle). San Francisco Press Inc, pp 26–27 Optical Electron Microscopy, vol 12. Academic Press, Lon-
Van Dyck D (2002) High-resolution electron microscopy. Adv don., pp 25–91
Imaging Electron Phys 123:105–171 Muller DA (1998) Core level shifts and grain boundary cohe-
Van Dyck D, Chen JH (1999) A simple theory for dynamical sion. In: Bailey GW, Alexander KB, Jerome WG, Bond MG,
electron diffraction in crystals. Solid State Commun 109:501– McCarthy JJ (eds) Microscopy Microanalysis Proc, vol 4(2).
505 Springer, New York., pp 766–767
Van Dyck D, Chen JH (1999) Towards an exit wave in closed Muller DA (1999) Why changes in bond lengths and cohesion
analytical form. Acta Crystallographica A 55:212–215 lead to core level shifts in metals, and consequences for the
Van Dyck D, Coene W (1987) A new procedure for wave func- spatial difference method. Ultramicroscopy 78:163–174
tion restoration in high resolution electron microscopy. Optik Op de Beeck M, Van Dyck D (1996) Direct structure reconstruc-
77:125–128 tion in HRTEM. Ultramicroscopy 64:153–165
Van Dyck D, de Op BM (1996) A simple intuitive theory for de Op BM, Van Dyck D, Coene W (1995) In: Tonomura A (ed)
electron diffraction. Ultramicroscopy 64:99–107 Electron Holography. Elsevier, North Holland, pp 307–316
Van Dyck D, Danckaert J, Coene W, Selderslaghs E, Broddin Pennycook SJ, Jesson DE (1991) High-resolution Z-contrast im-
D, Van Landuyt J, Amelinckx S (1989) The atom column aging of crystals. Ultramicroscopy 37:14–38
approximation in dynamical electron diffraction calculations. Rose H (2008) Heat loss of charged particle optics, vol. 153., pp
In: Krakow W, O’Keefe M (eds) Computer Simulation of 3–39 (Ed. Hawkes PW)
Electron Microscope Diffraction and Images. The Minerals, Rose H (2009) Geometrical charged particle optics. Springer,
Metals & Materials Society, Warrendale, pp 107–134 Berlin
Van Dyck D, de Op BM, Coene W (1993) A new approach to Saxton WO (1978) Computer Techniques for Image Processing.
object wavefunction reconstruction in electron microscopy. In: Electron Microscopy. Academic Press, New York., pp
Optik 93:103–107 236–248 (chapter 9)
Erni R (2010) Aberration corrected imaging in transmission Saxton WO (1986) Proc. XIth International Congress on Elec-
electron microscopy: an introduction. Imperial College Press tron Microscopy. Kyoto
Gabor D (1948) A new microscopic principle. Nature 161:777– Schiske P (1973) Image processing using additional statistical
778 information about the object. In: Hawkes PW (ed) Image
Geuens P, Van Dyck D (2002) The S-state model: a work horse Processing and Computer-aided Design in Electron Optics.
for HRTEM. Ultramicroscopy 93:179–198 Academic Press, London., pp 82–90
Geuens P, Van Dyck D (2005) The S-state model for electron Schiske P (1973) Image Processing of Computer-aided Design

Chapter 10
channeling in high-resolution electron microscopy. Adv Im- in Electron Optics (Ed. Hawkes P)
aging Electron Phys 136:111–226 Sinkler W, Marks LD (1999) Dynamical direct methods for ev-
Jia CL, Thust A (1999) Investigation of atomic displacements at eryone. Ultramicroscopy 7:251–268
a Σ3 {111} twin boundary in BaTiO3 by means of phase-re- Thust A, Coene WMJ, Op de Beeck M, Van Dyck D (1996a)
trieval electron microscopy. Physical Review Letters 82:5052 Focal-series reconstruction in HRTEM: simulation studies on
Jinschek JR, Kisielowski C, Van Dyck D, Geuens P (2004) Mea- non-periodic objects. Ultramicroscopy 64:211–230
surement of the indium segregation in InGaN based LEDs Thust A, Overwijk MHF, Coene WMJ, Lentzen M (1996b) Nu-
with single atom sensitivity. In: Ferguson IT, Narendran N, merical correction of lens aberrations in phase-retrieval HR-
DenBaars SP, Carrano JC (eds) Proceedings of SPIE Inter- TEM. Ultramicroscopy 64:249–264
national Symposium on Optical Science and Technology, vol
5187. SPIE, San Diego, CA, pp 54–63
Kirkland EJ (1984) Improved high resolution image processing
of bright field electron micrographs i. theory. Ultramicros-
copy 15:151–172
 
Imaging in STEM
11

Part I
Chapter Preview

Although STEM has a long history (it was first demon- hence a new mode of 3D imaging by depth sectioning or
strated in ~1938), it is only since the 2000s that it has been confocal electron microscopy.
fully integrated into commercial TEM instruments to give
levels of performance comparable to that achieved with In this chapter we present a more in-depth discussion of
dedicated machines. At the same time, new, easy to use, STEM, building on the concepts introduced in W&C to gain
dedicated STEMs have become popular for rapid screening a better feel for the nature of STEM imaging and answer
of samples much in the way a SEM is often used. One key some of the apparent paradoxes involved, such as “How
advantage of STEM is the possibility of using multiple de- can the same probe give me both a coherent and incoherent
tectors to obtain simultaneous information through different image?” “Why do I need a coherent probe for incoherent
beam–specimen interactions. imaging?” “How do I know if I have one?” We show the
key role of the detector in determining the coherent or inco-
The two most useful combinations are: 1) the HAADF (or herent nature of the image, and the role of thermal motion.
Z-contrast) image and an EELS image or line trace, allow- We present the physical arguments from real space and re-
ing correlation between atomic structure and elemental and/ ciprocal space perspectives, and also some mathematical
or electronic structure information; 2) simultaneous bright expressions for those with insatiable curiosity.
field (BF), annular bright field (ABF) and HAADF images,
with pixel-to-pixel correlation, providing very different but In W&C you’ll have read the 2 pages on Z-contrast im-
complementary views of the same region, a coherent phase aging in Chap. 22 and a bit on STEM in Chap. 9; in the
contrast image, an incoherent phase contrast image, and an present text, you’ve also just gone through the discussion
incoherent Z-contrast image. of coherence in Chap. 7 and the possibilities of Cs-corrected
phase-contrast imaging in Chap. 9. The present chapter puts
With the success of aberration correction, STEM has Z-contrast imaging on the same footing. Throughout this
gained enormously, providing a smaller probe for im- chapter we use K instead of u. We decided to do this since
proved resolution, with less illumination of neighboring the chapter is quite self-contained and this is the notation
columns, or, equally important, the ability to put more that you’ll encounter in the literature. We also have a large
current into the probe. The smaller, sharper probe gives number of scientific references but will put most of them on
both higher contrast and better signal-to-noise ratio, so that the website and update them there.
individual atoms can be seen in a Z-contrast image and
identified in EELS. Higher resolution also brings greater Z-contrast imaging does not require that we have atomic
precision in determining atomic positions, allowing, e.g., resolution, but imaging atoms is what many students want
column-by-column mapping of ferroelectric displace- to do and where you really need to understand what you are

Chapter 11
ments. In addition, we gain depth sensitivity on the na- doing, so we’ll emphasize atomic imaging; just remember,
noscale due to the great reduction in depth of field, and Z-contrast imaging has other applications too.

© Springer International Publishing Switzerland 2016

C. B. Carter, D. B. Williams (Eds.), Transmission Electron Microscopy, DOI 10.1007/978-3-319-26651-0_11 283

Licensed to ([email protected])
 284 11 Imaging in STEM

11.1 Z-Contrast STEM: an Introduction eventually becomes so small that it is comparable to the thermal
vibration amplitude of the atoms. Then there will be no well-de-
fined planes because the atoms will be constantly changing their
Part I

We first present a simple description of the underlying physical position on this scale. All phase relationships will vary randomly
principles concerning high-angle, annular dark-field (HAADF) in time as the atoms vibrate, and every electron will see the at-
imaging, which is popularly known as Z-contrast imaging. Later, oms with a different spacing and scatters in different directions.
we move on to a more in-depth description based on a quan- This is incoherent, or diffuse, scattering.
tum mechanical view of the STEM and the imaging process,
including probe formation, the benefits of aberration correction,
the role of the detector in determining the nature of the image
(coherent or incoherent) and the role of dynamical diffraction in 11.1.2 An Array of Scatterers
thicker crystals.

The Z-contrast image represents an atomic-resolution image that Now imagine passing a STEM probe across the crystalline
has a high sensitivity to atomic number (Z) and shows incoherent monolayer of scatterers shown in Fig. 11.2b. Since each spike
characteristics. It is quite different in nature from the coherent can be considered to scatter independently, each will scatter in
phase-contrast image of the TEM. The high-angle detector im- proportion to the local probe intensity it sees. This is the action
parts these two key characteristics by requiring all the electrons of a convolution, and the image is given mathematically by
used to form the image to have scattered through a large an-
gle; any electron not undergoing high angle scattering simply I.R/ = O.R/ ˝ jP.R/j2, (11.1)
passes through the hole in the detector (see Fig. 11.1). High-an-
gle scattering requires the electron to pass close to the nucleus. The object O(R) is the array of sharp scattering cross sections,
To achieve sufficiently high scattering angles the electron must and jP.R/j2 is the probe intensity profile. This is the definition
have ‘felt’ the full nuclear charge, so the scattering becomes es- of incoherent imaging: the image gives a representation of the
sentially pure Rutherford scattering with a cross-section that is object, which is blurred by the size of the probe. The sharper
proportional to Z2. In practice, smaller scattering angles are used the probe, the clearer the atoms appear. Different atoms will
to increase the number of electrons detected; then screening by have spikes with different heights, dependent on their cross-sec-
the atomic electrons reduces the exponent somewhat (so it’s not tions, which are roughly proportional to Z2. As the probe is fo-
quite the square of Z). The high-angle detector effectively re- cused and defocused the atoms come into clear view and blur
places the full atomic scattering potential with a potential for out again. The effect is not exactly symmetric in defocus, as
scattering to high angles, a (screened) nuclear scattering poten- we will see later, but when well focused the atoms are always
tial. Due to the very local nuclear potential atoms appear smaller seen bright; they do not change from black to white as they do
when ‘seen’ by the HAADF detector; they become narrow spikes in coherent phase-contrast imaging. A Z-contrast image shows
of scattering power, as shown schematically in Fig. 11.2. incoherent characteristics like those of a camera or projector
using visible light.

11.1.1 Independent Scatterers

11.1.3 As the Crystal Thickens
The second, equally important, consequence of requiring high
scattering angles is that the atoms become independent scatter- Now what will happen as we go beyond the phase-object regime
ers, which is the key requirement for incoherent imaging. Seen of a thin crystal? In a thick crystal dynamical diffraction, other-
from sufficiently high angles the atoms behave like indepen- wise known as channeling, occurs. In a phase-contrast image we
Chapter 11

dent emitters of electrons, as shown schematically in Fig. 11.3. will see contrast variations with increasing thickness having the
Ideally, all phase relationships between electrons ‘emitted’ (in period of the dynamical extinction distance, or in the case of a
reality scattered) between one atom and the next are random, diffraction-contrast image, there is no lattice image and we see
and each atom can be considered an incoherent source of dif- only thickness fringes, as shown in Fig. 11.4. Now the Z-contrast
fuse scattering. It can be compared to the incoherent emission image depends on the local intensity around each atomic site.
of X-rays by an atom, except thermal scattering is peaked in the We would expect from the simple picture just presented, that a
forward direction. If you look at the diffraction pattern from a column of n atoms would scatter n times stronger than one atom,
crystal in the TEM you can see it becomes diffuse at high angles: and so the first-order prediction is that the image would increase
coherent diffraction spots become replaced by a uniform inco- in intensity with thickness.
herent background. To form a sharp (coherent) diffraction spot
at some angle  requires that the atoms are arranged in planes This is indeed the case, but the increase is not linear for several
separated by a well-defined distance d = = , where  is the reasons. First, the thermal vibrations (phonons) have a finite co-
electron wavelength. Now at high angles the effective d spacing herence length in the beam direction of a few atomic spacings.
 11.1 Z-Contrast STEM: an Introduction 285

Scanned Probe a

Part I
Object

GaAs

Z=31 Z=33 b
Object Function
O R 

Annular I R   O R   P  R
2

Detector c

I  Z2 

Retractable Ga As
Bright Field Fig. 11.2 Schematic showing the action of the HAADF detector in the formation
Detector 1.4Å
of a Z-contrast image. a a monolayer raft of atoms. b Only the sharp parts of
the potential are effective in scattering electrons to high angles, and the at-
EELS
oms effectively become independent scatterers. c In the presence of a probe,
Spectrometer
each spike is weighted according to the probe intensity. As the probe is scanned
across the sample, it maps out the scattering power giving a direct, incoherent
Fig. 11.1 Schematic showing images available in the STEM as a probe is image showing the atom positions. Higher atomic number atoms will have high-
scanned across a crystal. A Z-contrast image is formed by mapping high angle er spikes, giving a Z-contrast image
scattered electrons as the probe is scanned, and, for a sufficiently small probe,
will show atomic resolution with strong atomic number (Z) contrast. The im-
age resolves the GaAs lattice and distinguishes the sublattice polarity, showing X ray High angle
EELS
fluorescence scattering
As columns with higher intensity than Ga columns. A simultaneously acquired
bright-field image is also available, or, alternatively, atomic-resolution electron
energy loss spectroscopy can be performed. Image obtained with an uncorrected
300 kV STEM with a probe size of ~ 1.3 Å

Therefore, the amplitude f scattered from the second monolayer

of atoms is in phase with the amplitude scattered from the first,
so the scattered amplitude from two layers is proportional to 2 f
and the scattered intensity is therefore .2 f /2, four times greater
than the intensity of a single layer, f 2. Therefore the initial thick-
ness dependence of a column of atoms goes as n2, where n is

Chapter 11
the number of atoms in the column. After a few monolayers the
phonons do destroy the coherence along the beam direction, but
now channeling becomes important. Fig. 11.3 Incoherently generated signals for imaging and analysis. a X-ray
fluorescence, spectroscopic but low intensity. b High angle electron scattering,
The atoms in a column do not contribute equally to the image. relatively high intensity dependent roughly on Z2. c Electron energy loss spec-
The strongest contribution comes from those atoms in the first troscopic imaging may also be incoherent if a large angle of collection is used
strong channeling peak, after which much of the probe is de-
pleted, scattered through both high and low angles. The scattered
electrons propagate through the remaining thickness of the crys- see Fig. 11.5a. The thicker the crystal, the higher the background,
tal, where they may be scattered again towards the detector. This but the image from the first channeling peak is always there. The
is the process of beam broadening: the beam explores regions overall form of the image is not strongly thickness dependent,
further and further from the site of the probe. The result is the as seen in Fig. 11.4 but the intensity and contrast are dependent
appearance of a relatively featureless background to the image, on thickness.
 286 11 Imaging in STEM

Fig. 11.4 Images of Si 〈110〉 tak- a b

en using a 100 kV STEM with a probe
size of ~ 2.2 Å and a probe-forming
Part I

aperture of ~ 10 mrad radius. a The

bright field image shows only thick-
ness fringes because the aperture ra-
dius is less than the scattering angle
of the lowest order Si reflections, the
{111} beams. b The Z-contrast im-
age does not show strong thickness
effects due to its incoherent nature.
c–g High-resolution Z-contrast im-
ages from thicknesses of 120, 230,
350, 470, and 610 Å showing lattice
resolution. The 2.2 Å probe is unable
c d e f g
to resolve the 1.36 Å separation of the
dumbbell, but their presence is seen
from the elongated form of each indi-
vidual dumbbell

The latter can be calculated but cannot be measured experimen-

Channeling
tally. This is also illustrated in the three right-hand columns of
Fig. 11.5a–c. In the thin crystal the probe wave function is close
Fast, negatively charged electrons tend to be attracted to to its entrance position in the unit cell, but with increasing depth
the positive atomic nuclei in the column. The effect is to is more and more affected by the crystal, becoming attracted
focus the probe more sharply in the vicinity of the stron- to the Si columns. Eventually the probe broadens substantially,
gest potential, i.e., exactly where the scattering occurs. whatever the initial probe position, which is the reason that in a
So if we place a probe over a column of atoms, aligned thick crystal a uniform background appears to the image.
with the beam direction, the intensity at each atom site
increases rapidly with thickness. A higher intensity is
scattered into the detector, so the image becomes brighter. ZN2 not ZN 2
Once an electron is scattered to high angles it is no longer
being channeled down the column, so the crystal acts as Note that because incoherent imaging is based on a sum
Chapter 11

a kind of fiber optic guide for electrons, except as they of atomic scattering cross-sections, the scattering from a
are guided down the fiber they are lost, scattered in the column goes approximately as the mean square scatter-
direction of the HAADF detector. 9 ing power, ZN2, not the square of the mean, ZN 2. Differences
in the channeling of columns are typically second or-
der compared to differences in scattering cross-section.
Channeling depends on the full potential, so it is roughly
dependent on Z, whereas the high-angle cross-section
11.1.4 Inside and Outside depends on Z2. 9

You must appreciate the difference between the Z-contrast im- None of these thickness effects change the basic incoherent na-
age, which is the detected intensity as the probe is scanned, and ture of the image. All that happens is that instead of an atomic
the intensity distribution inside the crystal for one particular scattering power, we can think of a columnar scattering power
probe position. that includes the basic thickness dependence. Equation 11.1 is
 11.1 Z-Contrast STEM: an Introduction 287

a b c d

a c

Part I
z=5Å
b

z = 25 Å

z = 50 Å

z = 75 Å

z = 100 Å

z = 125 Å

Fig. 11.5 a Simulated contributions to a Z-contrast image from slices at different depths within a Si 〈112〉 crystal. In each panel the upper half shows the inten-
sity scaled to its individual maximum, while the lower half shows the intensity scaled to the overall maximum, the first slice. The maximum image contribution comes
from near the entrance surface, with smaller contributions from deeper slices. The deeper slices also show an increased uniform background (blue) compared to the
slices near the entrance surface. b–d Intensity distributions inside the crystal for probes located as shown by arrows in (A). The dotted lines through the Si columns
provide a guide for the eye. Position a is the midpoint of the dumbbell, b is on a Si column and c is just to the right. Simulations are for a 300 kV aberration-corrected
probe with semiangle 22 mrad, coefficients of third and fifth-order spherical aberration of −0.037 mm and 100 mm, respectively, and a defocus of 20 Å

still a good approximation to the image, but now the object func- 11.1.5 The Effect of Defects
tion is thickness dependent. When we image a crystal in a zone
axis we therefore expect to see each atom column with a bright-
ness relative to other columns in accordance to its mean square The Z-contrast image also shows a simple dependence on sample
Z and a contrast that is decreasing with thickness. tilt. The strongest contrast occurs at exact zone-axis alignment,
when the zone axis lies in the center of the probe-forming ap-

Chapter 11
The cross-section always dominates, and this is a key reason erture and the channeling effect is at a maximum. The image
that Z-contrast images are usually directly interpretable, even can tolerate small misalignments (an image exists as long as the
in the presence of strong dynamical diffraction. This is a key zone axis is still contained within the probe-forming aperture)
practical advantage. In Z-contrast imaging it is not necessary but becomes increasingly streaky with increasing tilt, as shown
to choose between particular model structures, and new, unex- in Fig. 11.6.
pected structures have often been found just by interpreting the
bright spots as atom positions, the brighter the spots the higher zz Bend contours appear bright because they mark the regions
the mean square Z. of the crystal that are aligned in a planar (two-beam) or zone
axis (many-beam) geometry.
zz Defects will be visible in a HAADF image for the same rea-
son that they are visible in a diffraction contrast image: strain
fields, as occur around dislocations, will change the local
crystal orientation, change the channeling effect, and give
contrast as they do in a normal diffraction-contrast image.
 288 11 Imaging in STEM

11.1.6 Quasicrystals
Part I

One of the best examples of how the intuitive nature of the Z-con-
trast image can reveal unanticipated structures is in the field of
quasicrystals. Discovered in 1984, quasicrystals show symme-
tries that were long thought to be forbidden in nature, icosahedral
symmetry in three dimensions and in the case of the decagonal
quasicrystals, tenfold rotational symmetry in a plane. “Where are
the atoms?” was the title of an early paper. Despite many years
of diffraction studies, as well as conventional HRTEM, one key
feature was revealed only when Z-contrast imaging was brought
to bear on the problem. Images of the Al-Co-Ni decagonal phase
showed a broken symmetry that reduced the fivefold or tenfold
rotational symmetry to twofold, see Fig. 11.7a–b. However, the
resolution was insufficient to see the finer details of the atomic
arrangements, and with increasing resolution more details of the
structure came into view, as is shown in Fig. 11.7c, d. These
Z-contrast images made it possible to see, at an atomistic level,
how the quasicrystal structure could arise. The twofold sym-
metry of the basic 2-nm diameter cluster restricted the possible
overlaps from five or ten to just two, giving a precise prescription
for tiling the plane, as shown in Fig. 11.7e. Furthermore, with a
good idea of the atomic positions, total energy calculations were
performed that showed the twofold broken symmetry structure to
A B C have by far the lowest energy. Therefore, the structure is, indeed,
a thermodynamically stable phase. With aberration correction it
became possible to see the positions of the lighter Al columns
much more clearly, see Fig. 11.7f, g, allowing further refinement
of the structure.

There are many other examples of unexpected interface and

D E F grain boundary structures that have been revealed thanks to the
Fig. 11.6 A channeling pattern from Si 〈110〉 (the variation in HAADF inten- incoherent nature of the Z-contrast image, the ability to interpret
sity as a function of incident beam angle) taken using a 100 kV STEM in selected the bright spots as atomic positions, and the brightest as the at-
area mode with a near-parallel incident beam. The bright central region marks oms of highest Z.
the zone axis orientation and Kikuchi-like features are visible corresponding to
planar incidence. The images are taken at the various orientations marked on the
channeling map and show how the image contrast decreases for orientations
away from the zone axis. The angular width of the image is approximately 10˚ 11.2 An Electron’s Eye View of STEM

zz Stacking faults disturb the propagation of the electron through As you know, electrons are fundamental particles that obey the
the specimen, which changes the local illumination of atoms laws of quantum mechanics. The key concepts of quantum me-
Chapter 11

further down the column. chanics, the propagation of probability amplitudes, interference
zz Random atomic displacements, either static or time-varying, and wave-particle duality, are well illustrated by TEM, and some
can also lead to image contrast. familiarity is essential for a full understanding of STEM imaging.
To be different, we’ll first give a simple description of the elec-
We will discuss such effects later. For these reasons, although in- tron’s journey from source to detector from the perspective of the
tuitive interpretability is the hallmark of the HAADF image, we electron, a quantum-mechanical viewpoint but non-mathematical.
have to be careful because contrast effects in a HAADF image
do not always indicate Z-contrast. Electrons are indivisible quantum objects. They are emitted in-
dividually from the gun, are accelerated, travel down the column
and through the specimen until, if they reach a detector, they
make an impact, such as a flash of light in a scintillator, which
is registered as a count, a contribution to the image. This is the
particle character of the electron. Figure 11.8 shows two classic
 11.2 An Electron’s Eye View of STEM 289

a b c d

Part I
e

f g

~1.3Å

~0.95Å

e
A
Fig. 11.7 a Z-contrast image of an Al-Co-Ni decagonal quasicrystal and derived atomic structure (b) showing broken symmetry (c). Higher resolution image
showing further structural detail (d). Transition metal columns are shown in yellow, while the Al columns are seen as poorly-resolved red features. These images
revealed the atomic origins of the quasicrystalline tiling (e). f Aberration-corrected image showing more details of the light Al column positions. g Maximum entropy
processed image reduces the noise and enhances the resolution revealing many further likely Al column positions

proofs that an electron interference pattern is built up from suc- Next you pass through several lenses, which are magnetic fields
cessive flashes of light in the detector. The first shows the effect aimed at shaping your wave front into as perfect a spherical
of increasing intensity, and the second, with improved detector wave as possible so that you will converge into a point on the
development, shows the effect of increased detection time at specimen. Now with aberration correctors these lenses are more
fixed beam current. complex than the ‘round’ lenses used in the past. (We’ll colloqui-
ally call the cylindrically symmetric lenses ‘round’ lenses.) The
Imagine You Are the Electron new ‘lenses’ consist of quadrupoles, hexapoles, or octupoles, but
the principle remains the same.
So imagine you are an electron, moving along with the small
flow of current towards a thermionic source. You begin to ac- The task of the probe-forming system is to form an image of
quire thermal energy from your neighboring electrons and at- the source that is scanned across the sample. Even the cold

Chapter 11
oms until one random excitation overcomes the work function field-emission source has a spatial extent of about 5 nm. If we
of the solid and takes you outside into the vacuum of the gun, want you to converge onto a specific atomic site, the lens system
where you experience an electrostatic field that accelerates you must form a demagnified image of the source region, so that each
towards the specimen. If you are in a cold field-emission gun, successive electron will closely follow the path of the one before
the electrostatic field penetrates right into the source material and and converge near the atom or column of interest. Note that there
literally drags you out into the vacuum. As you are accelerated, is a cost involved here. We cannot have infinite demagnification
you gain the imposed momentum of the accelerator, while the because that would lead to zero probe current. Since lenses have
random thermal component remains. You can appreciate that the aberrations, there is a finite angle of acceptance leading to a good
cold field-emission source is a brighter source because there is focus, which is set by the objective (probe-forming) aperture. If
a smaller random component to your momentum, and so every we double the demagnification we halve the maximum angle
successive electron seems to come from a smaller region of the of emission from the source that will pass though the objective
source (higher spatial coherence) and has less energy difference aperture, i.e., we have ¼ of the current we had before. Aberra-
(higher temporal coherence). tion correctors have helped enormously because not only are the
 290 11 Imaging in STEM

Fig. 11.8 Two classic demonstrations a b

of the wave-particle duality of the a
a b
electron. The two series of images
Part I

both show the build-up of an interfer-

ence pattern from individual electron
events. a The effect of increasing in-
tensity, b the effect of increasing ex-
posure time. It is seen that electrons
propagate as waves but are detected b
as point particles
c d

e f

acceptance angles increased allowing higher resolution, but they exploring all possible pathways, elastic and inelastic, until you
also allow higher probe currents and therefore much improved reach a detector, at which point your wave function collapses
image quality. as you create irreversible changes in a scintillator or solid-state
detector. It is a deterministic journey in that all these probabilities
Once you have been accelerated to say 200 kV, you are traveling can be calculated and all the possible images simulated, at least
at 2/3 the speed of light. For a beam current of 1 nA the average to some approximation, but at the same time it is impossible
distance between you and the next electron is about 3 cm, so to determine exactly where your individual wave function will
you can appreciate that there is only one of you in the specimen collapse, and in this sense it is a statistical journey, as can be seen
at any one time. Furthermore, the time you spend in a sample in Fig. 11.8. Every successive electron takes its own journey and
Chapter 11

20 nm thick is only 10−16 s, which is much shorter than the period the image is built up from the sum of many such journeys with
of thermal vibrations for the atoms in the sample. Therefore, you their respective probabilities.
see a frozen snapshot of the specimen, and the electrons follow-
ing you will see a slightly different configuration. Therefore, the quantum-mechanical view of scattering is that it
remains probabilistic until you hit the detector and your wave-
From the Specimen to the Detector function collapses. While you propagate as a probability ampli-
tude, you do not know which detector you will hit or whether
Now how do you interact with the sample, and when? You have you will hit the detector on the right or the left. Only when you
many possible interactions. You can scatter elastically or inelas- reach the detector, a nanosecond or two after passing through
tically, and you can reach different detectors. Now we have to the specimen, is the path determined, and so only then does the
acknowledge the essential quantum mechanical nature of your sample recoil to the left or right as appropriate. How can this be?
journey. You may be an indivisible particle, but you propagate as This is a fine example of quantum-mechanical entanglement,
a wave of probability amplitude, a wave function, simultaneously where quanta of thermal vibration (phonons) and of electronic
 11.2 An Electron’s Eye View of STEM 291

excitations (plasmons or core loss transitions) are entangled with aperture is, therefore, properly called an objective aperture. This
your wave function until resolved by detection. It really can be terminology is typically used in connection with the dedicated
said that the scattering does not occur until the moment of de- STEM, but in TEM/STEM instruments the actual aperture used

Part I
tection. to define the angular range of the probe is the aperture that we
use as a condenser aperture in TEM mode, and therefore it is
The microscope is an excellent workbench demonstrating these often still referred to as a condenser aperture even when it is used
strange aspects of quantum mechanics. Normally we do not need as the STEM objective aperture! Here we will use the proper
to worry about such issues, but we should appreciate that we are terminology (☺), and refer to the pre-specimen probe-forming
dealing with quantum-mechanical particles and must follow the optics as the objective lens and aperture. There will be additional
laws of quantum mechanics to explain the images we see. lenses between the source and the objective aperture that are
used to demagnify the source and place it optically as far from
the objective lens as we want. These lenses still act as condenser
lenses.
11.2.1 Plane Waves and Probes
So we will assume that the condenser lens system is set for
high, effectively infinite demagnification so our source can be
It seems intuitively obvious that for atomic-resolution imaging considered to be at infinity, where it becomes an effective point
or analysis the STEM probe must be of atomic dimensions. How source. The wave front entering the objective lens therefore ap-
is such a probe formed? Let us consider an idealized system to proximates to a flat wave front, our ideal plane wave, which
begin with – a point source of infinite brightness at an infinite travels in just one direction. The wave amplitude, A.R/, is a
distance from our probe-forming lens. Don’t worry right now constant everywhere, and can be normalized to unity amplitude
that there will be zero current left by the time it reaches our lens and therefore unity intensity, as shown in Fig. 11.9a. The upper
– we will discuss the issue of current later. We can describe our diagrams in this figure represent the directions contained in the
ideal point source as emitting a spherical wave 0 .R/ = e2iKR =r beam, i.e., its reciprocal space, momentum space or k-space,
where k is a 3D wave vector that represents our high-energy elec- while the lower figures represent position space, so this is real
tron-wave propagating outward from the source at the center of space. These spaces coexist, i.e., they are different representa-
a circle towards the entrance aperture of our probe-forming lens. tions of the wave function of the electron.

Lens Terminology The Probe and Wave Packets

The terminology here can be confusing because the probe-form- A probe is formed by interfering plane waves at different an-
ing lens acts as an objective lens, since it determines the reso- gles to the optic axis. The objective lens performs this function.
lution with which we see our object, the specimen. Its entrance To see how a probe is formed, consider a 1D analog, a slit ex-

a b c d e

-Kmax Kmax

Chapter 11
1 1 1 1 1

0.8 0.8 0.8 0.8 0.8

0.6 0.6 0.6 0.6 0.6

0.4 0.4 0.4 0.4 0.4

0.2 0.2 0.2 0.2 0.2

0 0 0 0 0
-4 -3 -2 -1 0 1 2 3 4 -4 -3 -2 -1 0 1 2 3 4 -4 -3 -2 -1 0 1 2 3 4 -4 -3 -2 -1 0 1 2 3 4 -4 -3 -2 -1 0 1 2 3 4

Fig. 11.9 Schematic showing (a) a plane wave, (b–e) the formation of a localized probe from a coherent set of plane waves filling an aperture of semiangle θ.
a–b show reciprocal space, f–j real space. The objective aperture extends from −Kmax to +Kmax, and as it is progressively filled with more plane waves mutual destruc-
tive interference leads to the formation of a localized probe
 292 11 Imaging in STEM

tending from −K max to K max = max = as shown in Fig. 11.9b–e. ciprocal space is narrow in real space and vice versa. The probe
An increasing number of plane waves, 3, 5, 9 and infinity, pass intensity profile is then obtained by squaring the amplitude dis-
through this slit; when the number is infinity we have a continu- tribution,
Part I

ous distribution of amplitude in k-space. We have normalized the ˇ2

amplitude of each ‘component’ plane wave by their total number
ˇZ
(11.3)
2 2iKR
ˇ ˇ
P .R/ = ˇ A.K/e dK ˇˇ
so the total amplitude remains constant and we can see the devel-
ˇ
.
opment of the probe shape. The lower figures show the resulting
intensity distribution in real space. With three plane waves we The probe we have constructed here, in the absence of aberra-
see Young’s fringes superimposed on the uniform intensity of tions, is in fact the famous Airy disc distribution for a circular
the central beam. With five plane wave components we see the aperture as discussed in optics textbooks. It is the reference probe
strong peaks spread wider apart with three minor peaks between for the real situation in which we have aberrations, instabilities
them, and with nine components they are wider apart still with and the need for a finite source size to achieve a useable current,
seven subsidiary maxima in between. Each time we add more and it will feature frequently in subsequent sections.
component plane waves the strong peaks move further apart and
the cancellation in between them becomes greater. With an in-
finite number of plane wave components, that is a continuous
distribution of waves filling the aperture, the subsidiary peaks 11.2.2 Rayleigh, Airy and Resolution
move to infinity, i.e., we have only the one main peak in the
center. It is simple to extend this argument to 2D to form a wave
packet localized in 2D, namely our probe. Spatial resolution refers to the minimum distance at which
we can distinguish two objects in an image. Clearly this dis-
You can visualize this process somewhat differently as follows. tance will depend on the size and shape of the object, but it
Each pair of waves with equal and opposite transverse wave is convenient to define resolution as a property of the optical
vector ˙K max =n contributes Young’s fringes with amplitude instrument, independent of the nature of any particular object
cos 2Knmax x . At x = 0 every component is in phase and we see being observed. Therefore, it is usual to define resolution in
the peak amplitude, but as we move away from x = 0 each set of terms of an idealized object consisting of two point sources,
Young’s fringes becomes progressively out of phase and eventu- the so-called two-point resolution criterion. The most famous
ally there is no resultant amplitude. The larger we make K max, the such criterion is certainly the Rayleigh resolution limit, which is
quicker the components go out of phase as we move away from defined in terms of the Airy disc, the diffraction pattern arising
zero. Thus, for the smallest probe we want the largest aperture in from a uniformly illuminated circular aperture. If we have two
reciprocal space. Of course, in reality it is aberrations that limit incoherent point sources, then they contribute independently to
the maximum useable aperture, as we will discuss next. the image and resolution is defined by their individual intensity
distributions. If ˛ is the semiangle subtended by the aperture
Note that each individual plane-wave component, if it existed by in the image plane, then each of the two sources will give an
itself, would be infinite in extent, just as in Fig. 11.9a. It is be- Airy disc. Placing the peak of one over the first zero of the
cause we add them coherently that destructive interference gives other, at a radius of 0:61=˛, we see two clear peaks with a dip
us a finite wave packet. For this reason you should not think of the between them falling to 73 % of the peak intensity, as shown
probe as a collection of plane waves, because they are not allowed in Fig. 11.10a. An additional benefit of placing the peak of one
to be independent and therefore are not individual objects. The over the zero of the other is that the separation between the
probe represents the limit where there is an infinite number of two peaks in the image accurately reflects the spacing of the
plane waves at all angles within the probe-forming aperture – a two points.
continuous distribution of amplitude. The entire assembly is best
thought of as one converging spherical wave where every part is This is also the classic demonstration of the fact that incoher-
related in phase to every other – a coherent wave packet. ent imaging has better resolution than coherent imaging. If we
Chapter 11

imagine the two point sources to be coherent, and emitting in

With a continuous distribution of amplitude in reciprocal space, phase, then we must add the Airy disc amplitude distribution
the real-space distribution is just its Fourier transform (FT). In before squaring
p to get the intensity. The individual amplitudes
K-space our probe amplitude is uniform, P.K/ = A.K/, where we are roughly 2 broader than the intensity distributions, and so
now use A.K/ as a circular aperture function defining A.K/ = 1 at the separation of 0:61=˛ the two points are not resolved, as
inside the aperture and zero outside. In real space our probe am- shown in Fig. 11.10b.
plitude is the FT
Z As you might imagine, the corresponding criterion for coherent
P.R/ = A.K/e2iKR dK imaging would be to place the peak of one amplitude distribution
. (11.2)
over the first minimum of the other, which is now at a distance
of 0:81=˛, giving a dip of ~ 12 %. A dip of 25 % comes at pa
Again, in this picture, the reciprocal nature of k-space and the separation of 0:86=˛ which is approximately a factor of 2
real-space representations mean that a probe that is broad in re- greater than the resolution for incoherent imaging.
 11.3 Lens Aberrations for STEM 293

a b
12 5
Airy 1 Airy 1

Part I
10 Airy 2 4 Airy 2
Sum Sum
8
3
Intensity

Intensity
6
2

4
1
2
0
0
-1.5 -1 -0.5 0 0.5 1 1.5 -1.5 -1 -0.5 0 0.5 1 1.5
Distance ( ) Distance ( )
opt opt
c d
12 12
Airy1 Airy1
10 Airy2 10 Airy2
Sum Sum
8 8
Intensity

Intensity
6 6

4 4

2 2

0 0
-1.5 -1 -0.5 0 0.5 1 1.5 -1.5 -1 -0.5 0 0.5 1 1.5
Distance ( ) Distance ( )
opt opt

Fig. 11.10 Two-point resolution criteria are based on two point sources placed far enough apart to be distinguished as separate objects. a In the classic Rayleigh
criterion for incoherent sources, the Airy disc of one is placed on the first zero of the other. b With coherent sources emitting in phase the two sources are not resolved.
c The Sparrow criterion for incoherent sources is the separation at which the dip disappears. d An intermediate condition between the Rayleigh and Sparrow limits

An alternative definition of two-point resolution is that distance It is the function of the objective lens to convert the plane wave
at which the dip between the two points just disappears, that entering its entrance aperture into the converging spherical wave
is, they are separated by the smaller distance of 0:4886=˛, as we need for our probe. From the above discussion you can see
shown for the incoherent case in Fig. 11.10c. An intermediate that for the smallest probe we want the largest opening aperture.
situation is shown in Fig. 11.10c, with the two point sources However, the maximum useful aperture is limited because of
separated by 0:56=˛ which results in a dip of 12 % of the max- lens aberrations. It is actually very simple to include the ab-
imum intensity. errations of the lens as an additional phase factor affecting the
relative phases across the spherical wave front. In fact, the aber-
Clearly the definition of resolution is somewhat arbitrary, and a rations are defined as the error in optical path length between the
concept of resolution quality has been proposed which takes a actual wave front emerging from the lens and the ideal spherical

Chapter 11
value between zero for the Sparrow limit to one for the Rayleigh wave. The aberration coefficients have dimensions of length, and
limit. The situation in Fig. 11.10d corresponds to a resolution by convention they are expressed as a power series in angle θ.
quality of 0.5. Note however, that in this case the positions of the For example, the first few rotationally symmetric aberrations are
two peaks in intensity have moved inwards slightly so that they
1 1 1 1
no longer precisely measure the separation of the two objects. . / = f  2 + CS  4 + C5  6 + C7  8 + : : : (11.4)
2 4 6 8 ,

Where f is defocus, and Cs, C5 and C7 are the coefficients

11.3 Lens Aberrations for STEM of third, fifth, and seventh order spherical aberration, respec-
tively. For ‘round’ magnetic lenses these spherical aberration
coefficients are all positive. The aberration in radians is just
We’ll briefly review the concepts of lens aberration, and Scher-  = 2=, and the first two terms give the expression in
zer’s approach to minimizing the effect of these aberrations. W&C’s Eq. 28.8.

Licensed to ([email protected])
 294 11 Imaging in STEM

1 Now how do we choose the aperture for the best probe? If it

is too small we will have a broad, diffraction-limited probe,
whereas if it is too large we will have a broad probe due to the
Part I

Airy lens aberrations. In the uncorrected situation, the defocus is used

0.5
Cs to balance Cs, and Scherzer showed that the optimum aperture in-
Defocus troduces one wavelength of spherical aberration at its perimeter.
Sum If ˛opt denotes the semi-angle of the optimum aperture, we have
Aberration ()

1 4
CS ˛opt =
0 4 (11.7)

giving
  14
4
-0.5 ˛opt =
CS (11.8)

and defocus is set to give the opposite contribution at the aperture

-1 rim, i.e.,
0 0.2 0.4 0.6 0.8 1 1

aperture (opt)
f = − .CS / 2. (11.9)

Fig. 11.11 Aberration in units of wavelength across the optimum aperture for These conditions are referred to as the Scherzer conditions for
optimum defocus compared to the aberration free case (an Airy disc). The spher- incoherent imaging and are shown in Fig. 11.11. The spherical
ical aberration (blue curve) is exactly balanced by the defocus (green line) at the aberration term increases as the fourth power of the aperture
edge of the aperture and the maximum deviation of their sum is at (black line) angle and the (negative) defocus term increases as the square
of aperture angle, both ending up at one wavelength at the rim
of the optimum aperture. When both contributions are psummed,
zz Before ‘aberration correction’, these were the dominant aber- they result in a maximum aberration of −=4 at ˛opt = 2. This is
rations, and it was normal to compensate partially for the CS a tolerable deviation from the ideal of zero aberration, as can be
term by a small amount of negative defocus, weakening the seen by comparing the resulting probe intensity profile to an Airy
lens slightly to give less aberration at high angles. In W&C’s disc. Figure 11.12a shows that even under optimum conditions
Fig. 6.11, this corresponds to raising the specimen to a point the peak intensity of the aberrated probe is somewhat reduced,
where the rays cross closest to the axis forming a disc of least and the first minimum lies close to the first zero of the Airy disc
confusion. at a distance of 0:61=˛.
zz In a Cs-corrected STEM it is normal to run with CS slightly
negative, in order to compensate the next higher-order aberra- Although the aberrated probe does not quite reach zero, we still
tion, the positive C5 term. The optimum defocus then becomes take this as the definition of the optimum resolution for an un-
slightly positive, as we will see later. corrected system
zz Whether corrected or uncorrected, it is only possible to 1=4
achieve compensation over a restricted range of angles since dopt = 0:61=˛opt = 0:433=4 CS . (11.10)
all the rotationally symmetric aberrations have different an-
gular dependences. For the Airy disc the peak intensity is proportional to the area
of the aperture since in the absence of aberrations all waves ar-
Chapter 11

As before, we can write our STEM probe amplitude as an inte- rive in phase at the center of the pattern, as we saw above. In
gration over the probe-forming aperture, except we now include Figs. 11.12 and 11.13 we have normalized the distance by =˛
the aberration phase factor explicitly, so the aperture radius becomes 1, its area is , and the total
Z intensity passing through is therefore  2 = 9:87. With aberra-
P.R/ = A.K/e2iKR e−i.K/ dK tions, the peak intensity falls by about 20 %. The lost intensity
. (11.5)
is found in the secondary rings, which show enhanced intensity
and are pushed out to larger distances. This is more obvious if
This is just the FT of the aberrated wave, and the probe intensity we plot the integrated intensity within a certain radius, as shown
profile is in Fig. 11.12b. Whereas the ideal Airy disc has 83 % of its in-
ˇ2 tensity within a radius of 0:5=˛, this value is not reached with
the aberrated probe until more than double that radius. However,
ˇZ
P.R/ = ˇˇ A.K/e2iKR e−i.K/ dK ˇˇ (11.6)
ˇ ˇ
. this does not necessarily mean that lower resolution will be ob-
tained. If we look at the radius containing 50 % intensity, the
 11.3 Lens Aberrations for STEM 295

a b
10 1

Part I
Airy disc
Aberrated probe

8 0.8

Fractional Intensity
6 0.6
Distance ()
Intensity

4 0.4

Airy disc

Aberrated probe
2 0.2
Aberrated probe - undersize

Aberrated probe - oversize

0 0
0 0.5 1 1.5 0 0.5 1 1.5
Distance ()

Fig. 11.12 a Intensity profile for the optimum probe in an uncorrected microscope (black line) compared to the ideal Airy disc (red line). b Azimuthally integrated
probe intensity at optimum conditions in an uncorrected microscope (black line) compared to the ideal (aberration free) Airy disc (red line). Also shown are the curves
for the uncorrected microscope with an undersized (dotted line) and oversized aperture (dashed/dotted line). The oversized aperture throws a significant fraction of
the total intensity into the tails

aberrated probe has a slightly higher radius, 0:32=˛ compared larger apertures, the profiles are not symmetric either side of the
to 0:27=˛ for the Airy disc, but if we look at the HWHM for defocus that gives the highest central peak. At less than optimum
the two cases, they are both 0:25=˛. defocus the probe is broadened, while at higher than optimum
defocus, the probe develops a sharper central maximum but is
To appreciate what effect the aberrations have on the image, surrounded by a secondary ring. The change in height of this
imagine the probe as consisting of two parts, the sharp central ring may not appear that significant, but in the case of an over-
peak with no tails and the much larger diameter tail part with size aperture a substantially higher fraction of the total probe
lower peak intensity. Clearly, the sharp part will give good current ends up in the tails, as shown in Fig. 11.12b. This means
atomic resolution but the tail part will be too broad to show that there can be higher spatial frequencies in the image, i.e.,
atomic resolution and will just give an average over several col- high-resolution detail can become visible, but the extended tails
umns, i.e., a relatively uniform background image. The aberrated may illuminate adjacent columns so the overall contrast level is
probe, with reduced central peak and higher intensity in the tail, significantly reduced. Nevertheless, an oversized aperture with
will therefore mostly reduce the contrast in the image but will high underfocus can be used to demonstrate sub-Å image res-

Chapter 11
largely preserve the resolution. We will return later to a discus- olution in an uncorrected microscope, as shown in Fig. 11.14.
sion of image contrast and how to define resolution but gener-
ally, for Z-contrast imaging of a crystal lattice the full-width
half-maximum (FWHM) is a good indicator of resolution. If, on
the other hand, we are looking at a sample that is not aligned to a 11.3.1 The Benefits of Aberration Correction
zone axis, or our probe is too large to resolve the atoms, then we
are effectively imaging a continuous distribution of material. In
this case the integrated current should be a more useful measure Figure 11.13b gives probe intensity profiles for a 300 kV STEM
of resolution. after correction of CS. They have been calculated for the same
total flux passing through the aperture as in the uncorrected case
Figure 11.13a shows how probe intensity profiles in a 300 kV to highlight the effect of correcting the aberrations. This is also
STEM without CS correction vary with defocus and aperture size. realistic, since in practice we cannot have a true point source,
Comparing the profiles horizontally we see that especially for the and to see the benefit of the larger aperture we need to keep the
 296 11 Imaging in STEM

a 0.5 0.5 0.5 0.5 0.5

0.4
-50 0.4
-150 0.4
-250 0.4
-350 0.4
-450
Part I

0.3 0.3 0.3 0.3 0.3

0.75 opt
0.2 0.2 0.2 0.2 0.2

0.1 0.1 0.1 0.1 0.1

0 0 0 0 0
0 0.5 1 1.5 2 2.5 3 0 0.5 1 1.5 2 2.5 3 0 0.5 1 1.5 2 2.5 3 0 0.5 1 1.5 2 2.5 3 0 0.5 1 1.5 2 2.5 3

0.5 0.5 0.5 0.5 0.5

0.4
-250 0.4
-350 0.4
-450 0.4
-550 0.4
-650

0.3 0.3 0.3 0.3 0.3

opt
0.2 0.2 0.2 0.2 0.2

0.1 0.1 0.1 0.1 0.1

0 0 0 0 0
0 0.5 1 1.5 2 2.5 3 0 0.5 1 1.5 2 2.5 3 0 0.5 1 1.5 2 2.5 3 0 0.5 1 1.5 2 2.5 3 0 0.5 1 1.5 2 2.5 3

0.5 0.5 0.5 0.5 0.5

-250 -350 -450 -550 -650

0.4 0.4 0.4 0.4 0.4

0.3 0.3 0.3 0.3 0.3

1.25 opt
0.2 0.2 0.2 0.2 0.2

0.1 0.1 0.1 0.1 0.1

0 0 0 0 0
0 0.5 1 1.5 2 2.5 3 0 0.5 1 1.5 2 2.5 3 0 0.5 1 1.5 2 2.5 3 0 0.5 1 1.5 2 2.5 3 0 0.5 1 1.5 2 2.5 3

b 3.5 3.5 3.5 3.5 3.5

3
50 3
70 3
90 3
110 3
130
2.5 2.5 2.5 2.5 2.5

2 2 2 2 2

1.5 1.5 1.5 1.5 1.5

0.75 opt
1 1 1 1 1

0.5 0.5 0.5 0.5 0.5

0 0 0 0 0
0 0.5 1 1.5 0 0.5 1 1.5 0 0.5 1 1.5 0 0.5 1 1.5 0 0.5 1 1.5

3.5 3.5 3.5 3.5 3.5

3
50 3
70 3
90 3
110 3
130
2.5 2.5 2.5 2.5 2.5
Chapter 11

2 2 2 2 2
opt
1.5 1.5 1.5 1.5 1.5

1 1 1 1 1

0.5 0.5 0.5 0.5 0.5

0 0 0 0 0
0 0.5 1 1.5 0 0.5 1 1.5 0 0.5 1 1.5 0 0.5 1 1.5 0 0.5 1 1.5

3.5 3.5 3.5 3.5 3.5

3
50 3
70 3
90 3
110 3
130
2.5 2.5 2.5 2.5 2.5
1.25 opt
2 2 2 2 2

1.5 1.5 1.5 1.5 1.5

1 1 1 1 1

0.5 0.5 0.5 0.5 0.5

0 0 0 0 0
0 0.5 1 1.5 0 0.5 1 1.5 0 0.5 1 1.5 0 0.5 1 1.5 0 0.5 1 1.5
 11.3 Lens Aberrations for STEM 297

a b

Part I
Te Cd

222

c d

444

Chapter 11
Fig. 11.14 a A Z-contrast image of CdTe 〈112〉 taken with a 300 kV STEM under Scherzer conditions with CS of 1 mm (9.4 mrad aperture and −44 nm defocus).
The resolution of 1.36 Å is insufficient to resolve the CdTe dumbbell spacing of 0.93 Å. b Magnitude of the FT of (a) showing information transferred to the 1.86 Å
{222} spacing but not to the 0.93 Å {444} spacing. c Profile plot obtained by summing vertically over 200 pixels of an image of CdTe 〈112〉 taken with an oversized
17 mrad objective aperture and a defocus of −1250 Å showing the {444} fringes and d their corresponding spots in the magnitude of the FT

Fig. 11.13 a Variation of probe intensity profile with defocus and aperture size for an uncorrected 300 kV STEM with Cs of 1 mm. The center row is for the optimum
aperture, 9.4 mrad, where optimum defocus is −444 Å, while the upper and lower rows show the probes that result from apertures 75 % and 125 % of the optimum,
with the central defocus value close to that giving the highest peak intensity. b Probe intensity profiles after correction of Cs to −65 μm, with optimum defocus = 70 Å
and Cs = 100 mm. The center row is again the optimum aperture, 25 mrad, while the upper and lower rows again show the probes that result from apertures 75 %
and 125 % of the optimum. Curves are ideal in that they do not include non-round aberrations, source size or chromatic aberration effects
 298 11 Imaging in STEM

a b

1 1
Part I

0.8
0.5
Airy disc

Fractional intensity
0.6
Aberration ()

Corrected probe, f = 70 Å

0 Corrected probe, f = 90 Å

0.4

Airy
-0.5 f = 70 Å
Cs 0.2
C5
sum, f = 70 Å
sum, f = 90 Å

-1 0
0 0.2 0.4 0.6 0.8 1 0 0.5 1 1.5
Aperture (opt) Distance ()

Fig. 11.15 a Aberrations in units of wavelength across a 25 mrad aperture in a corrected 300 kV STEM for defocus = 70 Å (green line), CS = −65 μm (blue curve) and
C5 = 100 mm (purple line) and their sum (black line). Also shown is the sum for the same conditions but a defocus of 90 Å (dashed line). b Azimuthally integrated
probe intensity at a defocus of 70 Å (black line) compared to the ideal Airy disc (red line). Also shown is the curve for 90 Å defocus (dashed line) which results in a
slightly narrower and higher central peak, see Fig. 11.5b

source size small with respect to the probe size. The demag- 1 6
nification would therefore need to be increased for the larger 2 = C5 ˛opt (11.11)
6
aperture (as discussed later). If we decrease the source size in
proportion to the increase in aperture size we maintain the same giving
total flux passing through the aperture.   16
12
˛opt = (11.12)
zz The optimum aperture is increased from 9.4 mrad before cor- C5
rection to 25 mrad after correction, a factor of 2.7 in diameter
and 7.1 in area. Using dn to denote the limiting resolution due to nth order aber-
zz We see that the HWHM of the center (optimum) probe has rations we find that
decreased almost exactly in accord with the increase in aper- 1=6
ture diameter, and the peak intensity has increased over eight d5 = 0:455=6 C5 (11.13)
times, more than the increase in aperture area.
Chapter 11

zz The reason for this change is that there are now two terms that With our parameters this gives d5 = 0:54 Å in good agreement
can be used to balance the fifth-order aberration, the third-or- with our calculated profiles. Of course it must be remembered
der term CS, and the defocus, and it is possible to achieve a that these are ideal situations and the aberration-correction soft-
maximum deviation from the ideal wave front of less than ware may not achieve these ideals in practice in the presence of
1/8 wavelength, which is two times better than for the un- the ‘non-round’ aberrations, the chromatic aberration and the
corrected case. finite source size.
zz The closest approach to this ideal is actually at a defocus
of 70 Å as shown in Fig. 11.15a. The azimuthally integrated Comparing the probe profiles before and after aberration cor-
intensity is also then close to that of the aberration-free Airy rection highlights the two key benefits, a smaller and brighter
disc, as shown in Fig. 11.15b. probe. The first is the most obvious, the increase in resolution
from the smaller probe, but often the more important result in
We can calculate the resolution in the same manner as before. practice is not the ability to resolve finer spacings, but the gain
For example, with two wavelengths of fifth-order aberration at in contrast for a fixed lattice spacing. In Fig. 11.16 we compare
the aperture rim we have Z-contrast images of SrTiO3 taken using an uncorrected 100 kV
 11.3 Lens Aberrations for STEM 299

a b c a b

10 Å 10 Å

Part I
[001] [110]

O
5Å Ti
Sr c d

10 Å
Fig. 11.16 Z-contrast images of SrTiO3 [100] taken on (a) an uncorrected
100 kV STEM with Cs = 1.3 mm, d3 = 2.2 Å, (b) a corrected 100 kV STEM with
a probe size of ~ 1.3 Å, and (c) a corrected 300 kV STEM with a probe size of
~ 0.78 Å resolving the O columns

STEM, a corrected 100 kV STEM and a corrected 300 kV STEM.

In the last case, the probe is sufficiently small to allow the im-
aging of O columns in between the adjacent high-Z Sr and Ti Fig. 11.17 Imaging of Pt atoms on γ-alumina with a 300 kV STEM (a) before
columns. In this projection of SrTiO3 the O column is almost and (b) after aberration correction. Some Pt trimers and dimers are just visible
2 Å away from the heavier columns, and so when the probe is in (a) but individual atoms are much more visible in (b). Some faint contrast
centered over the O column there is negligible intensity at the Sr from the γ-alumina lattice is also present in (b) as it was not far from a zone
and Ti columns. Provided that the sample is thin enough we can axis. c Imaging of La atoms on the surface of a γ-alumina flake. By comparison
neglect beam broadening, meaning most of the scattering must to image simulations (d) it was established that the brightest (circled in yellow)
come from the O column itself. are on the beam exit side of the flake, the less bright (circled in blue) are on the
entrance side and the faint ones are in interstitial locations (circled in green)
This is where the second benefit of aberration correction comes
in, namely, the much higher peak intensity in the probe. In fact,
this benefit is as important as the gain in resolution since the peak The width of the image is determined primarily by the change
intensity scattered from an atom increases almost directly in pro- in intensity as the beam is scanned across the column on the
portion to the peak intensity in the probe. As indicated by Fig. 11.3, entrance surface, which is not affected by its subsequent chan-
with a corrected probe, we expect a much higher signal from in- neling along the column. We will discuss the propagation of
dividual atoms. This, in turn, gives a much better signal-to-noise the probe through thicker crystals later. Notice also that some
ratio. In effect, we receive triple benefit, improved resolution, im- La atoms are found in interstitial sites, where they show much
age contrast, and signal-to-noise ratio. Combined, these represent weaker intensity because of the absence of the contribution from
an enormous gain for the imaging of single heavy atoms on light the Al–O column itself.
supports, as shown in Fig. 11.17a, b for Pt atoms on γ-alumina.

The increased visibility of single atoms means that in favorable Today we have fifth-order aberration correctors avail-
cases they can even be observed on the surface of a crystal that able that will correct all aberrations up to the seventh

Chapter 11
is oriented to a zone axis. Figure 11.17c shows an example of order, leaving those as the limiting aberrations. The basic
the imaging of La atoms on the surface of a flake of γ-alumina. principles of balancing the aberrations remain the same,
It is apparent that the La atoms mostly occupy sites on top of except that the signs of the various contributions will
the Al–O columns, but they show two different intensity levels. change sign once more. Now the fifth-order term will be
A comparison to image simulations (Fig. 11.17d) shows that the negative to compensate the positive seventh-order term,
brightest La atoms are on the beam exit side of the alumina, the third-order term will revert to a positive value and the
when the beam intensity has been concentrated by its channeling defocus will become negative once more! 9
along the Al–O column. The intensity of the La atoms on the
beam entrance side of the specimen is distinctly weaker.
For a C7-limited system the limiting resolution due to geometric
aberrations alone becomes
Note that the channeling does not make the La atoms
appear smaller, only brighter. 9 d7 = 0:47=8 C7
1=8
(11.14)
 300 11 Imaging in STEM

a b the maximum energy transfer to a Si atom is less than its binding

energy to the cluster, therefore atoms are not detached from the
cluster by the electron beam. However, sufficient energy can be
Part I

transferred to nudge atoms into different configurations revealing

different possible conformations of the nanocluster. Figure 11.20
shows one atom jumping from left to right and back again.

Using density functional theory (DFT) calculations, we can build

3D models of the two configurations that match the images, and
then calculate the activation barrier for the two transitions. The
0.2 nm 0.2 nm barrier for moving from left to right is almost twice that for the
reverse direction, 1.44 eV compared to 0.8 eV. This demonstrates
c d a use of beam-induced motion. If thermally activated, the dif-
ference in rates for these two processes is about ten orders of
magnitude, in other words, the metastable configuration would
never be observed. For beam induced momentum transfers the
two rates differ only by a factor of two, in good agreement with
the observation of the atom on the left side twice as often as on
the right.

11.3.2 Resolution in the Third Dimension –

Depth Resolution
Fig. 11.18 Z-contrast images of Si atoms in monolayer grapheme, in (a) three-
fold coordination with an adjacent N atom, (b) fourfold coordination with adja-
cent C atoms taking two fivefold and two sixfold rings. Images obtained with a One somewhat unanticipated result of aberration correction is a
Nion UltraSTEM100 operating at 60 kV greatly reduced depth of field, z, with the larger aperture.

Aberration correction has taken us from the traditional TEM/

STEM regime where the depth of field is larger than the typical
We gain another factor of ~ 2 in optimum aperture and resolution. sample thickness, and where we view the image as a projection
Not only does this improved resolution bring increased precision of the specimen structure, to a new regime where z is less than
in the determination of atomic positions, but it also allows us to the typical specimen thickness. While it could be viewed as a
achieve atomic resolution at lower accelerating voltage, such disadvantage that all of the specimen cannot be in focus at once,
as 60 kV. We can image monolayer materials such as graphene we also have the opportunity to gain information on the speci-
below their threshold for knock-on damage, and use a Z-contrast men’s 3D structure. We can imagine obtaining a through-focal
image to identify individual impurity atoms based on their image series of STEM images; ignore beam specimen interactions for
intensity and directly identify their coordination, as shown in the time being, so each individual image represents a section
Fig. 11.18. through the specimen at a different depth. The depth of field in
incoherent optics is generally defined as the distance z over
High and Low which the focus does not change appreciably. In the absence of
aberrations it is given by
Chapter 11

No longer is Z-contrast imaging only for high-Z atoms. 

z  (11.15)
 9 ˛2

Notice how the z reduces as the square of the aperture angle,

With improved post-specimen optics, EELS efficiency is also in- much faster than the linear improvement in transverse resolution.
creased, and spectra from individual atoms can be obtained with This means that the depth of field of the uncorrected 300 kV mi-
sufficient statistics to reveal the nature of the chemical bonding croscope (with parameters corresponding to the optimum probe
from the near-edge fine structure, as shown in Fig. 11.19. shown in Fig. 11.13a) is > 220 Å, but after correction of Cs is
reduced to 32 Å, and with correction of C5 and a 50 mrad aper-
The reduced accelerating voltage reduces the energy transferred ture it would be reduced to only 8 Å! Therefore, nanometer-scale
to a given atom, but does not eliminate it, which is useful when depth sensitivity is expected in an aberration-corrected STEM.
we want to see different metastable configurations of atomic clus- The depth resolution dz will be larger than z. In fact, the con-
ters. Figure 11.20 shows sequential images of a Si6 nanocluster ventional definition in terms of the Rayleigh criterion is again
trapped in a graphene nanopore. At 60 kV accelerating voltage, obtained by placing one object above the other at the position of
 11.3 Lens Aberrations for STEM 301

Part I
0.2 nm

0.2 nm

Fig. 11.19 Si atoms in monolayer graphene in (a) 3-fold and (b) 4-fold coordination with C atoms. c EELS near edge structure from the two Si atoms showing the
threefold Si to have a strong peak near 105 eV, which is absent in the spectrum from the fourfold Si. d Spectra from a number of sp3 bonded materials, all showing
a strong peak after the threshold, due to empty d-states. The SiC spectrum is practically identical to that of the threefold Si atom in graphene, which is therefore
identified to be in an sp3 configuration. The absence of the peak in the fourfold Si shows the d-states are not empty, i.e., they are participating in the bonding, an sp2d
configuration. Identifications are confirmed by density functional theory. Spectra obtained with a Nion UltraSTEM100 operating at 60 kV

L R L R

2Å

L L R L

Chapter 11
Fig. 11.20 A sequence of images of a Si6 cluster trapped in a graphene nanopore showing one atom jumping between two positions on the left (L) and right (R)
sides of the cluster. Images obtained with a Nion UltraSTEM100 operating at 60 kV

the first zero of the intensity distribution along the z direction, This distance is a factor of 2 larger than z, and separating two
as shown in Fig. 11.21. objects vertically by this distance again results in a clear dip in
the intensity between the two objects as focus is changed.
The first zero occurs at a distance
An example of the depth sensitivity for imaging single atoms
is shown in Fig. 11.22 using an aberration-corrected 300 kV

dz = 2 (11.16)
˛2 , STEM. Several frames from a through-focal series show differ-
 302 11 Imaging in STEM

a
12
Part I

10

8
Intensity

0
-4 -3 -2 -1 0 1 2 3 4
2
Distance ( )
b
12

2 
10 1.75 



1.5 
8
Intensity

Fig. 11.22 Three frames from a through-focal series of Z-contrast images from
0
-4 -3 -2 -1 0 1 2 3 4 a Pt/Ru catalyst on a γ-alumina support, at defocus of (a) −12 nm, (b) −16 nm,
2
Distance ( ) (c) −40 nm from initial defocus setting. Arrows point to regions in focus, and in
(c) a single atom is seen in focus on the carbon support film. d Integrated inten-
Fig. 11.21 Profiles in the vertical direction along the optic axis for (a) the Airy sity of the Pt atom seen in (c) above the level of the carbon film as a function
disc, (b) two Airy discs separated by different amounts. When the first zero of of defocus, compared to a Gaussian fit. The FWHM of the fit is 12 nm but the
one coincides with the maximum of the other we see a significant dip between precision of the location of the peak intensity is 0.2 nm with 95 % confidence.
the two Results obtained with a 300 kV STEM with α = 23 mrad

ent clusters in focus at different depths. The γ-alumina support Another example of the imaging of single atoms in 3D is shown
is not on axis and gives weak background intensity. At a defo- in Fig. 11.23, a through-focal series of the thin SiO2 layer be-
cus of −40 nm a single atom comes into focus attached to the tween a HfO2 dielectric and a Si gate in a high dielectric constant
carbon support film. A movie of this focal series is available device structure. A single atom can be seen coming into and out
online (Borisevich et al. 2006a). Figure 11.22d shows the varia- of focus. Again the precision is higher than the resolution. Fur-
tion in the intensity of the Pt atom in C as a function of defocus thermore, in this case, the atoms are only visible over a range of
together with a Gaussian fit. The FWHM of the Gaussian fit is focus that is much less than the expected depth of focus. There
12 nm, which represents the depth resolution, but because of the are a number of possible reasons for this:
Chapter 11

high signal-to-noise ratio, the precision with which the peak in

the intensity profile can be determined is much greater, namely zz The most likely reason is that the specimen does not comprise
±0.2 nm with 95 % confidence. Therefore, for widely spaced, single heavy atoms on a light support, and there is a much
point objects the accuracy of depth measurements depends pri- higher background intensity. Therefore, the Hf atoms are only
marily on the signal-to-noise ratio and can be much higher than visible above the noise of the surroundings close to focus.
the depth resolution. zz Or we may have beam broadening through the specimen; this
should be a depth dependent effect and depth resolution for
Focal or Focus? objects near the beam exit surface should be poorer than that
for objects near the beam-entrance surface. This will be an
issue with thicker specimens.
Through focal series, through-focal series or through-fo-
cus series? 9 In this example, the 3D location of the stray Hf atoms was deter-
mined with relative ease, their density and spacings calculated,
 11.3 Lens Aberrations for STEM 303

f-0.5nm f±0.0nm f+0.5 nm

Part I
0.5nm

f+1.0 nm f+1.5 nm f+2.0 nm

Fig. 11.23 A sequence of frames from a through-focal series of Z-contrast images of a Si/SiO2/HfO2 high dielectric constant device structure showing an individual
Hf atom coming in and out of focus (circled). Results obtained with a 300 kV STEM with α = 23 mrad

and it was interesting to find that the Hf atoms avoided the Si- However, we have seen that the depth resolution even with
SiO2 interface. The 3D data set is analogous to that obtained next-generation correctors will be at the nanometer level, still an
in a (non-electron) confocal optical microscope, and standard order of magnitude greater than the transverse resolution. While
software can be used to form a 3D rendering of the data. Fig- deconvolution techniques could be helpful, some merging of the
ure 11.24 shows one such rendering, in which individual Hf at- two methods may be the logical path forward, combining tilt
oms are color coded according to their depth. series and focal series methods with a goal of achieving true
3D atomic resolution.
Both of these examples have involved an amorphous matrix
or support. The situation in a crystal is much more complex Another limitation of the simple depth-sectioning technique is
because of the dynamical diffraction (channeling) of the probe. that the depth resolution for extended objects is much poorer

Chapter 11
We will return to a discussion of this case later. We can also than for a point object. Physically we can understand this behav-
compare this depth-sectioning method with the established ior as follows. If the diameter of the object D is much broader
method of tilt-series tomography: the two methods represent than the probe diameter d, then with the beam placed through
opposite limits. the center of the object the change in image intensity will be
small until the defocused probe becomes comparable in size to
zz The depth-sectioning method requires no mechanical tilts as D. The depth resolution can be defined through the 3D FT of the
it is simply a through focal series which can be acquired au- probe intensity and for extended objects is given by d=˛. For the
tomatically; the entire focal series can be acquired in minutes. conditions shown in Fig. 11.21, a particle 1 nm in diameter will
The depth of field is much less than the specimen thickness. have a FWHM of around 40 nm. This limitation is overcome if
zz Typical tilt sequences take hours. Tilt series reconstruction the microscope is operated in a true confocal mode through use
requires that each view represents a direct projection of the of two aberration correctors.
specimen, i.e., the depth of focus is assumed much larger than
the object thickness.

Licensed to ([email protected])
 304 11 Imaging in STEM
Part I
Chapter 11

Fig. 11.24 A rendering of the 3D data shown in Fig. 11.23 in which individual Hf atoms are color coded according to their depth. The Si is shown in gold, the HfO2
in solid yellow, and the SiO2 layer is set transparent
 11.4 Spatial and Temporal Incoherence 305

11.4 Spatial and Temporal A Condenser lens Objective

lens
Incoherence Objective

Part I
aperture
Source

11.4.1 Spatial Incoherence

Specimen

So far we have been assuming an idealized probe limited only

by geometric aberrations, but in reality many factors will lead to
an effective broadening of the probe, including mechanical and
electrical instabilities, and the fact that the probe is an image of B
the source. We consider the source size effect first: we cannot
afford infinite demagnification since it gives zero beam current!
We have already defined the optimum aperture to give the best Source
resolution, ˛opt, which results in a resolution limit set by geo- Specimen
metric aberrations of dopt = 0:61=˛opt. In reality a source has a
finite radius, rs, which is demagnified through the probe-forming
system by a factor M, giving a geometric image size of ds = rs =M
in the absence of aberrations.
Objective
zz Successive electrons arrive on the specimen somewhere Condenser lens aperture Objective
within a distance of order ds from the center of the ideal probe. lens
zz Successive electrons are independent of each other. This
Fig. 11.25 Schematic showing how the beam current is reduced on demagnifica-
contribution to the image is a perfectly incoherent effect, a
tion of the source. a The source is focused by the condenser and objective lenses
blurring of the image, which is mathematically described as
to form a demagnified image at the specimen plane. The angular range shown in
a convolution. green is limited by the objective aperture. b Now the condenser is strengthened
to increase the demagnification by about a factor of two, and correspondingly
The distribution of source intensity is often modeled as a Gauss- the range of angles leaving the source that pass through the objective aperture is
ian, and then the image of the source is also a Gaussian distri- reduced by the same factor, giving ¼ the probe current compared to (a)
bution with a FWHM of ds. The effective resolution is therefore
estimated by adding the source size contribution and the con-
tribution due to geometric aberrations in quadrature, giving an We find the optimum condition by gradually increasing the
overall probe size of strength of the condenser lens, reducing the strength of the
objective lens to keep the specimen in focus, and retuning any
aberration correctors, until the contrast no longer improves but
q
2
d = ds2 + dopt . (11.17)
becomes limited by poor signal-to-noise ratio. The optimum con-
dition depends obviously on the task at hand: Z-contrast imaging
For ds  dopt, we achieve the ideal electron-optical resolution can tolerate a much lower beam current than spectroscopic imag-
limit that we have been discussing above, when the probe is ing; crystals scatter more than single atoms and can also tolerate
almost perfectly coherent. lower beam currents.

As we increase the demagnification we also decrease the range STEM imaging is fundamentally limited by the brightness of
of angles emitted from the source that pass through our fixed op- the source. As discussed in Sect. 5.2.A of W&C, brightness is

Chapter 11
timum aperture. The situation is illustrated in Fig. 11.25, where defined as the current per unit area per unit solid angle. If we
we see that we use the condenser lens to form an image of the consider only geometric optics, as we did for Fig. 11.25, bright-
source, which in turn becomes an object for the objective lens. ness is conserved from focus to focus along the optical system.
Increasing the strength of the condenser lens moves the image This means that we can easily relate the beam current at the spec-
of the source further from the objective lens, and the demagnifi- imen to the source size ds contribution at the specimen through
cation due to the objective lens is thereby increased; the image conservation of the brightness β, giving
of the source decreases proportionally. The cone of electrons s
emerging from the condenser lens is now more convergent and 4I
ds = (11.18)
because the objective aperture has a fixed size, much of it is 2
ˇ 2 ˛opt
.
blocked; the result is that only the much smaller angular range
shown in green passes through the objective aperture. The cur- We now plot the probe current I against the probe size d us-
rent decreases as we try to reduce the source size contribution to ing Eq. 11.17. Figure 11.26 shows the resulting plots for three
the image, and a compromise is needed. STEMs: a 300 kV uncorrected, a CS-corrected and a C5-corrected
 306 11 Imaging in STEM

105 We will see again later in this chapter that lattice imaging, both
bright-field and Z-contrast, relies on interference to form the
image. Lattice images are generated by the coherent part of the
Part I

104 probe dopt, and blurred by the incoherent part ds, which reduces
the visibility of the interference fringes or may eliminate them
Probe current (pA)

completely. To see a lattice spacing, a, with high contrast, the

1000
source contribution ds should be small compared to a, but then
Incoherent probe
the visibility of the image may well become limited by the sig-
nal-to-noise ratio. For this reason, we work with a source that
100
Coherent probe
is comparable to the size of the optimum probe defined by geo-
C5 corrected
metric aberrations. The effect is shown in Fig. 11.27a, where the
blurring
p increases the FWHM (by slightly less than the factor
10 Cs corrected
of 2 that would be expected for Gaussian shaped curves) but
Uncorrected
completely washes out the secondary maximum into a continu-
ously decreasing intensity as a function of radius. In practice it
1
is very difficult to establish whether the blurring of the probe is
0 1 2 3 4 5
really due to the source since mechanical vibrations, and electro-
Probe size (Å)
magnetic interference will also lead to an incoherent broadening
Fig. 11.26 A plot of probe current vs. probe size for uncorrected, CS corrected of the probe. These latter effects will usually have a time depen-
and C5 corrected 300 kV microscopes assuming a source brightness of 3 × 109 dence and that can assist in diagnosis of the origin of instabilities.
A sr−1 cm−2. In the green part of the plot the probe is predominantly coherent
whereas in the pink part it is predominantly incoherent, the changeover occur-
ring at the point where ds = dopt
11.4.2 Temporal Incoherence
microscope, assuming a brightness of 3  109 A sr−1 cm−2. For
ds  dopt we see small probe currents and improving resolu-
tion with increasing orders of aberration correction. This is the The other key source of incoherent probe broadening is tem-
regime of the highly coherent probe. If we increase the source poral in nature, due to the finite energy spread of the beam or
size, when ds  dopt the curves turn over and for ds  dopt we variations in the high voltage or the objective-lens current. In
have an essentially unaberrated image of the incoherent source. all cases they lead to a spread in focal length as described in
We then see the square root dependence of probe size on current. Sect. 28.13 of W&C. The effect can be modeled as an inco-
This is the regime of the incoherent probe. Note that aberration herent summation of successive electrons at different values of
correction can also be used to increase probe current at a given defocus. Figures 11.27b, c show the effect of a chromatic energy
resolution by 1 to 2 orders of magnitude, crossing from a coher- spread that is typical of a cold field-emission source, 0.3 eV, and
ent to an incoherent probe at ds = dopt, which for the brightness of a thermal field-emission (Schottky) source operating at low
we have assumed occurs at a beam current of 200 pA. emission current, 0.6 eV. We see that little change to the probe
profile occurs with the 0.3 -eV spread, while the 0.6 -eV energy
A useful expression for probe size as a function of beam current spread leads to a reduced central maximum with the lost intensity
can be obtained by rewriting Eq. 11.17 as follows appearing in the tails. However, the FWHM is unchanged for the
! 12 0.3 -eV spread and actually decreases very slightly for a 0.6 -eV
d2 spread. This reflects the fact that the high-resolution detail in
d = 1 + 2s dopt (11.19) an incoherent STEM image is insensitive to chromatic effects.
dopt , The reason will become clear when we consider imaging from a
reciprocal-space perspective later.
Chapter 11

! 12
4I
d = 1+ 2
dopt (11.20)
ˇ 2 ˛opt dopt or ,
11.4.3 “How Do I Know if I Have a Coherent
Probe?” The Ronchigram
Now, from Eq. 11.10 we use ˛opt 2 2
dopt = .0:61/2 to give an ex-
pression for the probe size as a function of beam current which
is valid for any order of aberration correction More to the point, “how do I know if I have a sufficiently coher-
ent probe to see the lattice spacings I want to see”? The best way
to see the extent of probe coherence is to look at the Ronchigram,
 1
1:1I 2
d = 1+ dopt (11.21) the distribution of intensity in far field from the specimen. The
ˇ2 .
term Ronchigram derives from its introduction by Ronchi in light
optics for measuring lens aberrations. If the probe-convergence
angle is small, then this is just the usual diffraction pattern, but
 11.4 Spatial and Temporal Incoherence 307

a b c
1 1 0.1

Part I
Ideal probe E = 0 E = 0
0.8 Source = 0.4Å 0.8 E = 0.3 0.08 E = 0.3
Blurred probe E = 0.6 E = 0.6

Relative Intensity
Relative Intensity

Relative Intensity
0.6 0.6 0.06

0.4 0.4 0.04

0.2 0.2 0.02

0 0 0
0 0.5 1 1.5 0 0.5 1 1.5 0 0.5 1 1.5
Distance (Å) Distance (Å) Distance (Å)

Fig. 11.27 a Effect of a finite source size (or effective source size due to instabilities) on the optimum probe intensity profile for a CS corrected 300 kV STEM (the
central probe in Fig. 11.13 with 90 Å defocus). The ideal probe as limited by geometric aberrations alone is shown in black, and when convoluted with a Gaussian
source size of 0.4 Å FWHM (shown in red) is broadened to the profile shown in green. b Effect of beam energy spread of 0.3 eV (red) and 0.6 eV (green). The FWHM
is little changed, but increasing intensity is thrown into the probe tails, as seen in the expanded plot (c)

the usefulness lies in observing the details of the intensity within and now inserting the expression for the probe amplitude from
the angular range that we need for our coherent probe, i.e., we Eq. 11.5, shown in red, we find
can remove the objective aperture or use one that is larger than
the optimum we are aiming for.
Z Z
iKR −i K/
Kf / = e−2 iKf R
'.R/ A.K/e−2 e dKdR (11.25)
Consider first the case for an aberration-free probe, which comes
to a precise focus position as shown in Fig. 11.28a. If we defocus
Z Z
K/ iK f R 2 KR
a distance f , then we create a so-called shadow image, as in Kf/ = A.K/e−i e−2 e '.R/dKdR
(11.26)
a point-projection microscope, where the probe focus can be
considered a point source and each ray propagates directly to the
screen (or CCD detector), leading to a bright-field image magni-
Z Z
K/
Kf/ = A.K/e−i e2 i.K −K f /R
'.R/dRdK (11.27)
fied by the ratio of the distance between sample and screen to the
defocus. If we now insert a weak-phase object (WPO) such as .
a thin amorphous carbon film at position 1, then we know from Performing the integration over R gives the FT of '.R/
W&C Sect. 28.3 that a thin specimen produces a pure-phase Z
shift, which is refraction but not absorption. This is the familiar  .K f / = A.K/e−i.K/ '.K − K f /dK
phase-object approximation (POA). We saw how the effect of the . (11.28)
specimen could be described as a multiplicative phase-shift pro-
ducing, in the case of a plane incident wave of unity amplitude, Equation 11.23 is a product of the probe amplitude in real space
an exit-face phase-shift of with the specimen transmission function, and so we find the re-

Chapter 11
ciprocal-space distribution becomes the convolution of their re-
'.R/ = ei V.R/ (11.22) spective FTs. We can rewrite Eq. 11.28 in convolution notation as
Z
The interaction constant  is defined in W&C Sect. 28.3. If in-  .K f / = A.K/e−i.K/ ˝ '.K f / = P.K f / ˝ '.K f / (11.29)
stead of unity amplitude everywhere we have our incident probe .
amplitude, then our exit face wave function is now
To find the intensity on the detector plane we must take the
 .R/ = '.R/P.R/ (11.23) square

We now take the FT to move from the specimen exit plane into I.K f / = jP.Kf / ˝  .K f /j2
the detector plane, labeling final state wave vectors by Kf,
Z Remember that (in this chapter’s notation) K f refers to position
 .K f / = e−2iKf R '.R/P.R/dR (11.24) on the detector screen, and so this is completely analogous to the
bright-field TEM image described in Eq. 28.22 of W&C. There,
 308 11 Imaging in STEM

a
what we call the probe amplitude distribution P.K/ was referred
to as the amplitude response function or point spread function
Part I

(PSF) of the imaging system, h(x,y). The transmission function is Specimen Specimen
also the same as the phase object expression used in Ch. 28. I.K f / position 2 position 1
is, therefore, just the bright-field image intensity in the STEM
detector plane exactly as we would have for the correspond-
ing defocus in a bright-field TEM image. It differs from normal
bright-field imaging in that we cannot independently change the
magnification and defocus. Note that if we move the specimen
to position 2 (or change the defocus to the opposite sign) then f
not only is the contrast changed by reversing the defocus, but the
shadow image would be inverted too.

Imagine what we would see as we reduce the defocus and go

through zero to a negative defocus. At zero defocus we have in-
Ronchigram
finite magnification but on either side of focus we have a bright-
field TEM image from a small area with positive or negative b
defocus. Exactly as with the phase-contrast imaging of a WPO
it will look bright at positive defocus (as for the specimen at
position 1 in Fig. 11.28a) and dark at negative defocus (with the
specimen at position 2).

In our ideal case with no aberrations, the entire disc of intensity

would become uniformly dark or light. This is where the sensi-
tivity of the Ronchigram method comes in; in reality the phases
are aberrated and the resulting contrast is strongly affected, as
shown in Fig. 11.29. What we see is the effect of the aberrations
superimposed on the speckle image of the carbon film, and the
aberrations can dominate the image contrast. Therefore, this is
an excellent way to assess the extent of the flat-phase region and
the symmetry of the residual aberrations.
Fig. 11.28 Schematic indicating the formation of a Ronchigram in the far field
Another striking feature of the shadow image is the existence of
of the objective lens (a) for an ideal unaberrated probe, (b) in the presence of
lines of infinite magnification! Figure 11.28b shows a ray dia-
spherical aberration, with the paraxial rays (blue) focused after the specimen,
gram in the presence of spherical aberration, where the paraxial
the high angle rays (red) focused before the specimen due to the spherical aber-
rays are focused after the specimen, but the high-angle rays are ration; the rays at intermediate angles (green) are focused at the specimen and
focused before the specimen. Clearly, at some particular angle give rise to a ring of infinite magnification in the Ronchigram
the rays will focus in the plane of the specimen, and we have a
ring of infinite magnification.
Another use for the Ronchigram is for alignment of a crystal to a
Examples of Ronchigrams for an uncorrected microscope are zone axis. In this case we no longer have a WPO but a series of
shown in Fig. 11.29. The circular pattern near optimum defocus diffracted beams, each of which is an image of the probe-forming
is a very convenient and accurate way to adjust the astigma- aperture and therefore forms a disc. With the apertures we need
Chapter 11

tism. Also, the center of the pattern unambiguously marks the for high resolution these diffraction discs overlap extensively.
optic axis of the objective lens, so the optimum aperture size Because the probe is coherent, the regions of overlap will inter-
and position can be chosen with confidence from observing the fere, and complicated contrast effects result. Contrast changes
Ronchigram. You can see this effect in Fig. 11.30, which com- as the probe is scanned. The observation of fringes in the Ron-
pares Ronchigrams in a 100 kV STEM before and after CS cor- chigram is a key indicator of the degree of coherence in the
rection. The significant increase in the diameter of the flat-phase probe. Weak fringes imply weak coherence and weak contrast
region that can be used for probe formation is clear. Images of Si in any form of STEM lattice imaging, bright-field or Z-contrast.
〈110〉 before correction cannot resolve the dumbbell spacing Figure 11.32 shows typical behavior in the case of a Si crystal
of 0.136 nm, but after correction they are resolved with a clear with an uncorrected microscope. Now the shadow image shows
dip in intensity between the two atomic columns. The FT shows a clear zone axis diffraction pattern, which allows accurate tilt
information transfer to 0.125 nm. of the exact region where the beam is in focus. As the focus is
moved through the specimen, lattice fringes are seen in the cen-
tral region of the Ronchigram inside a circle of infinite magnifi-
 11.4 Spatial and Temporal Incoherence 309

Part I
Chapter 11
Fig. 11.29 Ronchigrams from a thin specimen of amorphous carbon with increasing focus (from weak to strong) taken in a 200 kV STEM/TEM, (a) at large under-
focus where all rays focus after the specimen and a shadow image is seen. b intermediate focus corresponding to Fig. 11.2 where a circle of infinite magnification
can be seen. c Gaussian focus where now only the low angle rays are focused on the specimen and the characteristic circular disc is seen delineating the extent of
the flat phase region. d High overfocus where all rays cross the axis before the specimen and an inverted shadow image is seen
 310 11 Imaging in STEM

a b cation. At optimum focus the position of the beam with respect

to an atomic column can be easily seen by whether the center of
the Ronchigram is dark (on column) or white (in the channel).
Part I

A Warning is Required

Since the FT of the image intensity can give misleading

information, we must take care to be certain that the im-
age intensity scale includes the proper zero level, that
25 mrad contrast has not been enhanced by subtraction of a back-
ground intensity, which is sometimes done automatically
c d by the microscope or imaging software. Take a histogram
of the image intensities to reveal if any ‘clipping’ has
occurred. Clipping will increase the frequency content of
the image beyond that limited by the microscope; imag-
ine a sine wave with the tops of the peaks cut off. You’ll
have flat tops with sharp corners – it is no longer a pure
sine wave but contains higher harmonics. The same can
happen in an image, as shown in Fig. 11.31. Scan insta-
bilities can lead to ghost reflections – giving a very useful
1 nm way to diagnose instabilities by observing the changes in
the apparent positions of reflections as the scan speed and
direction is changed. So FTs of image intensities can give
e f
useful information on information transfer, but they are
not useful indicators of image resolution, which needs to
be assessed by a two-point criterion as described previ-
ously. 9

The Ronchigram can also be a useful way to search a specimen

in a dedicated STEM. Reducing the objective lens current by a
311 400 large amount leads to a large field of view. This can be a more
220 331 sensitive way to search than scanning over a very large region.
0.164 nm 0.192 nm 0.136 nm 0.125 nm
Fig. 11.30 Ronchigrams from Si 〈110〉 observed in a 100 kV STEM (a) before
and (b) after correction of CS showing the increased extent of the flat phase
region available for probe formation. c Z-contrast images before correction and
11.5 Coherent or Incoherent Imaging
(d) after correction, when the Si dumbbell at 0.136 nm is now resolved. The
increased information transfer is seen by comparing the magnitude of the FT of
the image intensity before correction, (e), with information transfer to 0.164 nm, The intensity distribution in the detector plane that we have
to after correction, (f), with information transfer to 0.125 nm calculated is for a stationary probe on the optic axis of the mi-
croscope. The STEM image contrast comes from changes in de-
tected intensity as the probe is scanned, which depends critically
Chapter 11

on the geometric form of the detector. In particular, whether the

image shows coherent or incoherent characteristics depends on
how much of the diffraction pattern is integrated by the detector.
If the detector is small in angular extent compared to features
in the diffraction pattern, then we are sensitive to those features,
i.e., we are sensitive to interference effects and we see a coher-
ent image. If the detector integrates over a large range of the
diffraction pattern, then the details of the interference pattern
are integrated, and we have an incoherent image. Here we will
see explicitly how coherent or incoherent characteristics arise in
the STEM image.

To calculate a STEM image we must allow the coordinate of the

probe to move independently of the specimen coordinate R. We
 11.5 Coherent or Incoherent Imaging 311

Fig. 11.31 An example of the gen-

eration of higher order spots in a dif-
fractogram from the effects of signal

Part I
clipping. The image is for a simulated
square lattice, with the image, the
magnitude of the FT and the corre-
sponding histograms shown for (a)
correct black level, and increasing
levels of clipping (b–c). The genera-
tion of high frequency spots is clear
in the clipped data sets

Chapter 11
do this by introducing a second real-space variable R0 to locate insert some explicit detector geometries and examine the nature
the center of the probe. It is important not to confuse these two of the image contrast. First we take the case of a point detector
coexisting coordinate spaces; we will use R0 to describe the lo- located on axis, which will give us a bright-field image.
cation of the center point of the probe intensity at the specimen,
which is the pixel location in a STEM image. The probe ampli-
tude at the specimen entrance surface becomes
Z 11.5.1 A Point Detector; Coherent Imaging
P.R − R0 / = A.K/e2iK.R−R0 / e−i.K/ dK (11.31)
.
We return to Eq. 11.24 and write the amplitude at K f in the form
Comparing this with Eq. 11.5 we see the shift of the probe by R0 appropriate for a scanning probe, with scan coordinate R0, as
in real space becomes a multiplication of all plane-wave compo-
nents by the phase factor e−2iKR0 in reciprocal space. Now let’s
 312 11 Imaging in STEM

a b c
Part I

d e

Fig. 11.32 Ronchigrams from Si 〈110〉 observed in an uncorrected 200 kV STEM with decreasing focus (from strong to weak), (a) at high overfocus showing the
zone axis pattern superimposed on a shadow image, useful for alignment purposes, (b) decreasing the lens strength towards focus increases the magnification until
(c) near focus a ring of infinite magnification appears with a lattice image in the center. At exact focus the center of the pattern changes from dark to bright depending
if the probe is on or off the atoms and the sign of the defocus (d–e)

Z
 .K f ; R/ = e−2iKf R '.R/P.R − R0 /dR (11.32) Just as in bright-field TEM imaging we recognize  .R0 / as the
. specimen transmission function and P.R0 / as the objective-lens
impulse response function from W&C’s Chap. 28. All the con-
so the intensity on the detector plane becomes trast transfer functions (CTFs) discussed and envelope functions
ˇ2 discussed in W&C’s Chap. 28 apply here also, except we have
to remember the reciprocal nature of the STEM with respect to
ˇZ
Chapter 11

I.K f ; R/ = ˇˇ e−2iKf R '.R/P.R − R0 /dRˇˇ (11.33)

ˇ ˇ
. the TEM. The beam-divergence, envelope function is now the
angular spread in the detected electrons due to a finite collec-
If we now consider a point detector on axis we obtain the expres- tor aperture. Remember that these expressions do not include
sion for a scanned bright-field image by setting K f = 0 the effect of a finite source size, which may reduce or eliminate
fringe visibility.
ˇ2
ˇZ ˇ
(11.34)
ˇ
I.R0 / = ˇ '.R/P.R − R0 /dRˇ
ˇ ˇ
.

which is again the square of a convolution 11.5.2 An Infinite Detector: Incoherent Imaging
2
I.R0 / = j'.R0 / ˝ P.R0 /j . (11.35)
Next we take the opposite extreme from a point detector, an
infinite detector, one that collects all the scattering. In the case
 11.5 Coherent or Incoherent Imaging 313

of elastic scattering, an infinite detector would give no contrast.

However, an infinite detector would be appropriate in the case of a
energy-filtered imaging with a large spectrometer acceptance-an-

Counts (arbitrary units)

Part I
1 1
gle. In this case,  .R/ now represents the amplitude of inelas- 2 2
tically scattered electrons. We work through the steps to show 3 3
how an incoherent image results. First, we expand the square in Last Co plane 4 4
Eq. 11.33, showing the amplitude in red and its complex conju- First Si plane 5 5
gate in blue (with primed variables K′ and R′ to show they are 6 6
independent of K and R) 7 7
Z 760 770 780 790 800 810
Energy (eV)
I.K f ; R0 / = e−2 iKf R '.R/P.R − R0 /dR
b
Z
0
(11.36)
e2 iKf R ' .R0 /P .R0 − R0 /dR0
.
The total detected intensity is the integral over all scattered wave
vectors K f
Z
I.R0 / = I.K f ; R0 /dK f (11.37)
.

Collecting terms in K f gives for the image intensity

Z Z
I.R0 / = '.R/P.R − R0 /dR ' .R0 /P .R0 − R0 /dR0
Z (11.38) Fig. 11.33 a Z-contrast image of a CoSi2/Si {111} interface with spectra obtained
iK f .R0 −R/
e2 dK f . plane by plane across the interface, marked with a white line. The first Si plane
shows dumbbells in a twin orientation. Spectra were recorded using a VG Micro-
scopes HB501UX STEM operating at 100 kV and a parallel EELS system designed
Now the black integral is the mathematical definition of a delta
by D. McMullen. Spectra 1–4 are background subtracted by the usual power law
function, and so we have
fit, spectra 5–6 are obtained using the spatial difference method, using a refer-
Z ence spectrum from Si far from the interface. Because of the twin, the distance
I.R0 / = '.R/P.R − R0 /dR between planes 4 and 5 is 2.7 Å. Data do not necessarily correspond with the
(11.39) precise area shown in the image, but are from the respective planes as numbered
on the plot. b Co composition profiles obtained from (a) compared to dynami-
Z
0 0 0 0
' .R /P .R − R0 /dR ı.R − R/.
cal simulations; original data points are shown as black circles with the original
± 10 percentage point error bars estimated in 1993. A reanalysis of the associated
spectra using a modern least squares fitting routine is shown as gray crosses, and
We can now eliminate the second real space variable R′ by inte- a simulation for a sample thickness of 100 Å is indicated by the black solid line.
grating over it. The delta function selects R = R0and we obtain an Both agree well with the original analysis. The simulation used a 10.3 mrad probe
expression for the image intensity in terms of a single integral. forming aperture, CS = 1.3 mm and Scherzer defocus of −693 Å. Ionization scat-
Z tering coefficients assumed a detector of 30 mrad semiangle and a 30 eV energy
I.R0 / = '.R/'  .R/P.R − R0 /P .R − R0 /dR. (11.40) window above threshold. The solid vertical gray lines indicate the planes within
the CoSi2 with the Si planes indicated by the vertical dashed gray lines

Chapter 11
I.R0 / = j'.R/j2 jP.R − R0 /j2 dR. (11.41) This is the normal way we think of elemental mapping, how-
ever, it only applies if we actually do collect all, or at least most
which is again a convolution, but now a convolution of intensi- of, the scattering (i.e., when the detector is either large enough
ties not amplitudes to detect all the scattering or at least a representative fraction
of the total scattering). It has been shown, for example, that
I.R0 / = j'.R0 /j2 ˝ jP.R0 /j2. (11.42) phase-contrast effects are seen in energy-filtered images taken
with a small collector aperture, as we might expect from the
This is the mathematical definition of an incoherent image. The discussion in the previous section where a small axial detector,
object is characterized by the square of the amplitude function, gave us a phase-contrast image. This effect is known as preser-
i.e., the intensity distribution of (in this case) inelastic scattering vation of contrast under inelastic scattering; it occurs when the
generated in the specimen, and the resolution and contrast are characteristic angle for the inelastic scattering is small. Then
determined by the probe intensity profile. the inelastic event can be thought of as just slowing down the

Licensed to ([email protected])
 314 11 Imaging in STEM

transmitted electron but not otherwise changing its direction, and It was also in 1993 that the first atomic-column EELS data was
therefore the phase contrast is unchanged. This is approximately recorded, and also the first maps of bonding configurations were
the case for plasmon scattering, where inelastic scattering angles obtained at sub-nanometer resolution based on spectral fine struc-
Part I

are relatively small, and phase contrast in the zero-loss image ture. Today, not only can we routinely obtain 2D elemental maps,
was seen to be preserved in the plasmon-loss image. but we can also perform atomic-resolution simulations for the
EELS image. A modern EELS simulation for the 1993 Co L pro-
Inelastic scattering has a Lorentzian angular distribution with a file is shown in Fig. 11.33b and is in good agreement with the
half-width given by observations. Interpretation of EELS images at atomic resolution
can be complicated by effective non-local interactions and the
extended tail of the Coulomb response function, which can lead
E
E = (11.43)
2 E. to volcano-like features or a large background in the image. For
these reasons you should always compare experiments to full
E is the energy loss and E is the beam energy.
image simulations.

zz For a typical plasmon energy of 25 eV and accelerating There is a problem in applying this theory to the case of elas-
voltage of 100 kV, the characteristic scattering angle is only tically scattered electrons: now  .R/ is again the specimen
0.125 mrad, much smaller than for typical collection aper- transmission function, and is thus a pure phase object, not an
tures. amplitude object. If all the transmitted electrons are collected
zz For a 1 keV energy loss the characteristic scattering angle is then there is no image! We must introduce some finite detector
now 10 mrad and the effective collector aperture size is at to produce contrast, so let’s consider the annular detector intro-
least 10 mrad. duced originally by Albert Crewe. The inner angle of the detec-
tor is made sufficiently large for the unscattered probe to pass
Hence, with increasing energy loss, the nature of the image con- through, and a dark-field image is formed. However, any elastic
trast becomes increasingly incoherent, regardless of the size of scattering within the angle of the probe-forming aperture is not
the collector aperture. You can see that attempting to quantify detected, so we do not collect all of the scattering.
compositions by taking the ratio of core-loss edges to the ze-
ro-loss peak will not necessarily work at atomic resolution if a This used to be referred to as the ‘hole in the detector’ problem
small collector aperture is used. The core-loss image will have and caused controversy in the early days of STEM imaging. Par-
a significant degree of incoherence, and it will therefore peak adoxically, the way to overcome this dilemma is to introduce a
for the selected atoms if optimum defocus is set for the probe. large hole in the detector! Then we recover incoherent imaging
However, the zero loss will show coherent phase contrast, with but at the cost of having less total signal, as we show next.
a quite different dependence of contrast on defocus.

1993: A Year of Firsts 11.5.3 An Annular Detector: Incoherent Dark-

Interpretable compositional maps at atomic resolution are best Field or Bright-Field Imaging
recorded under incoherent imaging conditions, using a large
spectrometer acceptance angle, so that phase-contrast effects are
eliminated as far as possible. Then the image will resemble a It is obvious that if a detector were to collect a representative
Z-contrast image for the relevant elemental species; the image in- fraction of the total scattering then the corresponding image
tensity peaks on the appropriate atomic sites. Atomic-resolution would also show incoherent characteristics, albeit with lower
spectroscopy was first demonstrated in 1993, with Co L edges overall intensity. The high-angle ADF (HAADF) detector does
taken plane-by-plane across an atomically abrupt CoSi2/Si{111} exactly this, and in addition, it forces the scattering to come from
interface, as shown in Fig. 11.33a. The spectra were recorded by near the nucleus. The scattering cross-section approaches that
Chapter 11

scanning a line parallel to the interface, averaging over many unit expected for nuclear scattering, Rutherford scattering, with a Z2
cells in the scan direction, and using the Z-contrast image to lo- dependence on atomic number Z.
cate the probe over each plane in turn. This procedure minimizes
beam damage but maintains atomic resolution perpendicular to Another way to collect a large fraction of the scattering is with
the interface. The experiment is simply an edge resolution test. Harald Rose’s annular bright-field (ABF) detector. The annulus
in this case is within the cone of the STEM probe, so it gives
In a continuum model, convolving a Gaussian probe with a step an annular bright-field (ABF) image. Typically, the collection
function gives an intensity profile that drops from 75 to 25 % angle would run from half the probe angle to the full probe an-
in a distance equal to the full-width half-maximum (FWHM) gle. The image intensity at each point in the annulus is given
of the Gaussian. The experimental profile exceeded this value, by Eq. 11.33, and the image intensity is obtained by integrating
demonstrating atomic resolution had been achieved. over the detector. It is a phase-contrast image, however, because
it is a sum over many point detectors, each with a different CTF;
the contrast is quite incoherent in nature over a large range of
 11.5 Coherent or Incoherent Imaging 315

thickness and defocus. Since the ABF image is formed from

Z
I.R0 / = jP.R − R0 / Œ'.R/ ˝ D.R/j2 dR (11.51)
low-angle scattering, it is not strongly dependent on Z, and low-Z .
and high-Z atoms can be seen clearly in the same image.

Part I
In this equation all quantities are replaced with their FTs, so the
To see the action of HAADF explicitly, we again introduce the multiplications have become convolutions and vice versa. We
notation P.K/ for the probe-amplitude distribution in Fourier again recover the form for an incoherent image
space, but now it contains the phase factor appropriate for a
scanned probe centered at probe coordinate R0, see Eq. 11.31. I.R0 / = jP.R0 /j2 ˝ j'.R0 / ˝ D.R0 /j2 (11.52)

P.K/ = A.K/e−i.K/ e2iK.R−R0 /. (11.44) but now our specimen function is no longer the square of the
full transmission function. We find the interesting result that the
We rewrite Eq. 11.28 for the amplitude in the detector plane as object depends on how we look at it! You might expect this re-
Z sult for a quantum-mechanical object. The high-angle detector
 .K f ; R0 / = P.K/'.K − Kf /dK (11.45) has filtered the object to retain only the high-angle components.
. Again we see how the scattering cross-section tends towards the
Rutherford scattering limit for a very large inner detector angle.
which for convenience we write in convolution notation as
It is interesting that historically in EM this form of image has
 .K f ; R0 / = P.K f / ˝ '.K f /. (11.46) usually been referred to as a non-linear image, because it is based
on imaging the square of the phase-contrast transmission function.
The intensity is again the square
Phase-contrast imaging was the first form of lattice imaging
I.K f ; R0 / = jP.K f / ˝ '.K f /j2. (11.47) to be developed and in the weak phase-object approximation
(WPOA) gives a linear image of the projected potential (see
We can include a detector by introducing a function D.K f /, which W&C, Chap. 28). In most branches of science, however, the
is unity over the detector and zero elsewhere. This gives a simple incoherent image is regarded as the linear image of the object,
expression for the image intensity because it gives the most direct representation of the object with-
Z out contrast reversals. A coherent image is regarded as a nonlin-
I.R0 / = jP.K f / ˝ '.K f /j2 D.K f /dK f (11.48) ear image because it does not have a unique relationship with
. the object. In conventional phase-contrast microscopy the most
obvious manifestation of this is the reversal of image contrast
Now the detector function is either zero or one, i.e., it has no by reversing the sign of the aberration function that is used to
phase, so we can insert it inside the square to give generate contrast. So which form of imaging we refer to as linear
Z depends rather on our own history!
I.R0 / = jŒP.K f / ˝ '.K f / D.K f /j2 dK f (11.49)
.
The Ideal Image

In general, this equation does not separate (in the case of ABF
imaging for example), but consider the case where we have a Should a phase-contrast image be regarded as the ideal
detector with a very large central hole, much larger than our image? 9
probe-forming aperture. Then we always have K f  K . We can
then ignore the K dependence of '.K − K f / and consider the de-
tector function as multiplying '.K f /, then
11.5.4 Atoms Are Smaller in HAADF STEM

Chapter 11
Z
I.R0 / = jP.K f / ˝ Œ'.K f /D.K f /j2 dK f (11.50)
.
The detector function multiplies the specimen transmission func- Atoms seen by the high-angle detector are intrinsically smaller
tion; therefore, it selects only the high-angle components, those than they appear to a bright-field detector, simply because the
that fall on the high-angle detector. We can consider the object detector sees only the sharp parts of the atomic potential. Far
to be filtered by the detector. In this reciprocal-space view it is from an atom there is no net charge, assuming the atom is neu-
simple to appreciate that the high-angle detector just eliminates tral. The Z electrons surrounding the atom exactly cancel the
the low-angle components of the specimen transmission func- nuclear charge Z, but the fast electron wave penetrates the atomic
tion, which pass through the hole in the detector. We only see the electrons and sees the much higher and more localized potential
sharp parts of the specimen, as in Fig. 11.2. Conveniently, atoms due to the nucleus.
do have sharp features due to the nuclear contribution to the scat-
tering potential, and this is what dominates the high-angle signal
and provides the Z2 dependence of high-angle scattering. We can
now transform back into real space using Parseval’s theorem
 316 11 Imaging in STEM

a b c
120 2.5 1.4
T=0K T=0K T=0K
Part I

100 T = 100 K T = 100 K 1.2 T = 100 K

2.0 T = 200 K
T = 200 K T = 200 K
T = 300 K T = 300 K 1.0 T = 300 K
80
1.5 0.8

V'(r)

V'(r)
60
V(r)

1.0 0.6
40
0.4
0.5
20 0.2

0 0.0 0.0

0.0 0.5 1.0 1.5 2.0 2.5 3.0 0.0 0.5 1.0 1.5 2.0 2.5 3.0 0.0 0.5 1.0 1.5 2.0 2.5 3.0

x (Å) x (Å) x (Å)

Fig. 11.34 Profiles of the projected scattering potential in Si 〈112〉 for different detectors and temperatures. a Full elastic scattering potential as applicable for bright
field imaging. The potential for scattering to high angles is narrower. b 45–200 mrad, c 90–200 mrad

Looking at Atoms 11.5.5 Transverse Coherence

Remember: atoms are different depending on how you Let us now examine the action of the detector from a real space
look at them! 9 perspective. We return to Eq. 11.38 for the image intensity and
insert a detector function
Z Z
It is this part of the wave that gets scattered to the high-angle I.R0 / = '.R/P.R − R0 /dR ' .R0 /P .R0 − R0 /dR0
detector, which is the reason that we conceptually replaced the
atoms with spikes in Fig. 11.2. Figure 11.34 shows the effect (11.53)
Z
0
e2 iKf .R −R/ D.K f /dK f
explicitly in the case of a Si 〈112〉 crystal. The full projected
potential (V(R) in Eq. 11.22) overlaps significantly between the
dumbbells, but the potential for scattering into the annular de- Now when we perform the FT with respect to K f we no longer
tectors clearly separates the two columns of the dumbbell. The get the convenient delta function in R − R0 that we had in the case
width (and also the strength) of the potential reduces with in- of an infinite detector; the expression does not collapse mathe-
creasing inner detector angle. matically into the expression for incoherent imaging. Instead of
the delta function that insisted R = R0 we must now integrate at
each R over a range of R′. Is this significant? The special feature
This is an important conclusion, because not only do of the high-angle annular detector is that it can emulate the action
we have the intrinsic resolution advantage of incoherent of the delta function by restricting the range of R′ to something
imaging that we discussed earlier, but in principle we that is small on the scale of the atomic spacings. We do not need
can reduce the atomic size just by increasing the inner the range of R′ to be infinitely small to achieve incoherence be-
detector angle. In practice, however, the reduced image tween neighboring atomic columns.
intensity will most likely mean that noise becomes the
Chapter 11

limiting factor. In addition, we may eventually be limited Look at the form of the detector function explicitly in real space.
by the finite amplitude of thermal vibrations, so that be- Due to the fall off in scattering with increasing angle, we can
yond a certain detector angle we will only lose intensity assume the outer detector angle to be infinite. This would be
without reducing the width of the potential. 9 a good approximation for an outer angle that is > 3 times the
inner angle, which can usually be achieved. The black integral
in Eq. 11.53 is the FT of the detector function, given by the dif-
You should realize that the scattering potential as calculated here ference between the FT of the infinite detector (a delta function)
is different from that used in solid-state physics to describe the and that of the central hole (an Airy disc), i.e., it represents the
interaction of an atomic core with the valence electrons. In that detector function in real space
case we are concerned with low-energy electron interactions and Z
exchange and correlation effects are all built into the interaction D.R/ = e2iKf .R −R/ D.Kf /dK f
0

potential, often via a pseudopotential.

2J1 .K i ; R − R0 / (11.54)
= ı.R − R0 / −
K i .R − R0 /
 11.5 Coherent or Incoherent Imaging 317

a b a b
Condenser or collector aperture Annular detector

Part I
2

Intensity
2

0 25 50 75 100
Inner detector angle (mrad)

Fig. 11.36 a Intensity distribution in the detector plane for two point scatterers
1.5 Å apart, illuminated by a probe centrally located between them. Inner and
outer detector angles correspond to 10.3 and 150 mrad, respectively. The circle
0.61 marks 50 mrad radius, and samples many fringes around its perimeter. b Ratio of
the detected intensity to the incoherent scattering prediction replacing the point
Fig. 11.35 Schematic illustrating the concept of a transverse coherence length scatterers by Si atoms, as a function of inner detector angle. Probe is optimum
imposed by the detector geometry. a In bright field imaging a small condenser for an uncorrected 100 kV STEM with CS = 1.3 mm
or collector aperture is used such that the coherence length in the specimen
is greater than atomic spacings, resulting in a coherent, phase contrast image. A useful criterion for the minimum detector angle i that is re-
b An incoherent image is formed with an annular detector with a hole sufficient- quired for incoherent imaging of lattice spacing a is
ly large such that the coherence length in the specimen is less than the atomic
spacings 
i = 1:22 (11.55)
Here J1 is a Bessel function, the mathematical form of the Airy a
disc for an inner angle of K f . Now we make the inner radius of
the detector large on the scale of the Bragg reflections, then the Then a corresponds to a separation in real space that is twice the
real-space width of the detector function will be correspond- distance from the peak to the first zero of the Airy disc of the
ingly much smaller than interatomic spacings. In this case one hole in the detector in Fig. 11.35. Figure 11.36b shows that in this
atom will not be able to coherently interfere with a neighboring case the deviation of the detected intensity from the incoherent
atom. So although we cannot show strict mathematical inco- signal is less than 5 %.
herence, when we look at a crystal along a major zone axis,
atomic spacings are always in the range of a few Å and we can
have independent image contributions from adjacent atoms. This
is the concept of transverse incoherence as controlled by the 11.5.6 The Origin of Contrast in the Scanned
width of the detector function in real space. Note in the case of Image
bright-field imaging, the FT of a point detector is infinite, so all
atoms contribute via their amplitudes. This is coherent imaging.
In reality we do not use an infinitely small detector, we just make We now examine what happens in the detector plane as the probe
the collector angle sufficiently small that the coherence envelope is scanned. We will see that despite the incoherent nature of the
covers several adjacent columns. The concept of a transverse image the origin of image contrast is coherent interference. Re-
coherence envelope is illustrated in Fig. 11.35. turning to Eq. 11.45, inserting P.K/ from Eq. 11.44 and taking
the square gives the intensity at any one point K f in the detector
We have not explicitly included thermal vibrations in this discus- plane
sion, and in fact, thermal vibrations are not required to destroy

Chapter 11
coherence in the transverse direction, although they do assist to
Z
K/ 2 iK R0
I.K f ; R0 / = A.K/e−i e '.K − K f /dK
a small degree. The dominant effect is due to the geometry of
(11.56a)
the detector. This can be easily appreciated with reference to a Z
K 0 / −2 iK 0 R0
test specimen consisting of two closely spaced point scatterers, A .K 0 /ei e ' .K 0 − K f /dK 0
i.e., infinitely sharp potentials. In this case the scattering factor
of each ‘atom’ is uniform in reciprocal space, and cosine fringes
are formed in the detector plane for a probe located centrally We retain the complex conjugate notation on A .K 0 / for com-
between them, see Fig. 11.36. With real atoms of finite size there pleteness, but since A.K/ is real A .K 0 / = A.K 0 / is also real.
is a strong angular decrease in scattered intensity, and the total Collecting terms gives
detected intensity becomes dominated by the region near to the
hole. As the inner hole of the detector is enlarged, there is an in-
Z Z
.K0 /− K// 2 i.K−K 0 / R0
creasing number of fringes around its perimeter, so the integrated I.K f ; R0 / = A.K/A .K 0 /ei e
intensity averages over more fringes and approaches the result (11.56b)
that would be expected for independently contributing atoms. '.K − K f /' .K 0 − K f /dKdK 0
 318 11 Imaging in STEM

a b to intersect the overlap regions at the edges of the probe-forming

aperture, for example, the ADF detector shown in Fig. 11.37b. To
obtain overlaps on the optic axis the maximum spatial frequency
Part I

is not the diameter of the aperture but the radius, and we need to
increase the probe-forming aperture to achieve overlaps on axis
and a bright-field image.

We have the same result as we have in TEM: the radius

of the objective aperture limits the bright-field resolu-
tion. This is another example of the reciprocal nature of
Bright field collector aperture Annular detector STEM and TEM (see W&C Sect. 22.6). We also see that
the resolution limit in ADF STEM is, in theory, a factor
Fig. 11.37 Schematic diffraction pattern for a cubic crystal projection with lat- of two higher than in bright-field imaging. We say ‘in
tice spacings such that neighboring discs are slightly overlapping. A detector theory’ because the small overlap regions at the limit of
placed in an overlap region will show an image as the probe is scanned. For resolution mean that there is little actual image contrast
this situation, the axial bright field detector (a) will therefore show no image, and noise may become limiting in practice. 9
whereas the ADF detector (b) shows a lattice image

We can now see how contrast arises in the STEM image. The The overlap regions in Fig. 11.37 are shaded schematically, but
intensity at this point on the detector results from the interference in reality they contain complex contrast features that depend on
of all possible pairs of beams within the incident probe with the microscope settings and lens aberrations. In Fig. 11.38 we
wave vectors K and K′ which are scattered by the sample into see some simulated Ronchigrams for a model crystal consisting
the same final direction K f . Since the probe is coherent all the again of point scatterers; the Ronchigram shows the interfer-
scattered beams are coherent and they all interfere to give the ence pattern without the angular decrease in intensity. As the
amplitude of the final state K f . Now what happens as we move size of the probe-forming aperture increases, the overlap region
the probe? Just one term contains the probe position R0: it is the increases and reaches to the optic axis, allowing us to obtain
interference term e2i.K−K /R0. This term oscillates as the probe a bright-field image. The figure also shows how the optimum
0

scans, with spatial frequency K − K0, which is the image contrast conditions lead to a large region of the overlap having a uniform
we want. intensity, i.e., being in phase. The optimum conditions chosen in
this case are those appropriate for a bright-field image, but differ
One very important point is already clear. The period of the os- from those for the dark-field image by only a small amount. Also
cillation is determined by the wave vector difference between the shown is the effect of ‘turning off’ the spherical aberration, leav-
pair of incident beams K and K′. The only possible pairs of beams ing only the defocus term which now leads to vertical fringes
that can interfere are those contained within the probe-forming because the optimum defocus for zero CS is zero. In the absence
aperture because if either A.K/ or A.K 0 / is zero then the entire ex- of lens aberrations these fringes are straight.
pression becomes zero. We arrive at the important result that the
maximum periodicity (highest spatial frequency) K − K0 that can Figure 11.39 shows the oscillation of the overlap region as the
appear in the image is given by the diameter of the probe-forming probe is moved across the unit cell from one ‘atom’ to the next.
aperture. Then beams from the edge of one disc interfere with This is the fundamental origin of STEM image contrast. If the
beams from the opposite edge of the other. Let us take the ex- conditions are changed from optimum then the overlap region
ample of a thin cubic crystal, which gives a diffraction pattern as does not show large areas of uniform phase, as seen in the case
shown in Fig. 11.37. Only those regions where the discs intersect of zero CS, in the out of focus condition shown on the right-hand
will show oscillating contrast as the probe scans, so only those side of Fig. 11.38. The fringes simply translate across the overlap
Chapter 11

regions contribute to image contrast. This is the reason that the region, integrating to negligible change in total detected inten-
enhanced transfer in Fig. 11.14d with an overfocused lens also sity, and thus resulting in negligible image contrast.
required a larger aperture. The aperture cutoff is an absolute cutoff
for image contrast. If spots are seen in the FT of the image that lie
outside the aperture cutoff then they are spurious, due to clipping
or instabilities of some sort, as discussed before. 11.5.7 Transfer Function and Damping Function
We also appreciate immediately from Fig. 11.37a that these re-
gions of overlap may not extend to the optic axis. The usual Now let us return to the basic equations for incoherent imaging,
bright-field detector is a relatively small detector to minimize the Eq. 11.42 and transform into reciprocal space. We find
damping factor due to beam divergence, and is located on the op-
tic axis, so it does not see such overlaps. We need a large detector I.Q/ = j'.Q/j2  jP.Q/j2 (11.57)
 11.5 Coherent or Incoherent Imaging 319

a Fig. 11.38 Simulated Ronchigrams for a rectangular lattice of delta function

atoms, with spacings 1.54 Å horizontally and 3.08 Å vertically, for aperture sizes
of (a) 8 mrad, (b) 12 mrad, (c) 14 mrad and (d) 18 mrad semiangle at 300 kV. Only

Part I
in (c) and (d) do the overlaps extend to the center (aperture radius exceeds the
lattice spacing). It is clear that for a good contrast bright field image the collector
aperture would need to be very small on the scale of the probe-forming aperture
to avoid averaging over the fringes in the center of the pattern. The patterns on
the left are for the uncorrected case, with CS = 1 mm, defocus – 51.5 nm, when
large patches of almost uniform contrast are present. The patterns on the right
are for the same defocus but zero CS, and a set of straight fringes is seen. Return-
ing the defocus to zero, all overlaps become uniformly bright

b
The intensity of the image spatial frequency Q is that in the
object multiplied by that same spatial frequency content of the
probe intensity. The FT of the probe-intensity profile acts as the
CTF for the incoherent image, or as it is usually referred to, the
optical transfer function or modulation transfer function. Fig-
ure 11.40 shows transfer functions corresponding to the images
shown in Fig. 11.14 for the optimum Scherzer conditions and
for the oversized aperture. You can see how the oversized aper-
ture and high defocus boost the high spatial frequencies at the
expense of the midrange and lower frequencies.

c The figure also illustrates the effect of a finite energy spread due
to fluctuations in beam energy or in lens current. In the bright-
field case, the damping envelope takes the form of an exponential
at the high spatial frequencies, as discussed in Chap. 28 of W&C.
The origin of this strong damping factor is the rapid oscillation in
the phase CTF at high angles. The oscillations express the phase
difference between the beam scattered at K and the zero-order
beam, and become progressively more rapid with increasing an-
gle. A spread of defocus shifts the peaks and troughs of the oscil-
lations so that at a high enough angle they will average to zero.
The effect on a STEM ADF image is very different. The STEM
contrast at the highest resolution comes from the interference
d of beams across the aperture diameter, +K and –K, which are
at equal angles to the optic axis. They, therefore, have exactly
the same phase changes for a fluctuation in defocus, and the
center line between overlapping discs is achromatic; therefore
the presence of fluctuations does not reduce the highest spatial
frequencies in the image. There is no loss of resolution to first
order, which is the reason that STEM ADF imaging is regarded
as being robust to chromatic aberration effects. Regions of over-

Chapter 11
lap away from the achromatic line will suffer changes in rela-
tive phase, and so it is the midrange spatial frequencies that are
damped, as seen in Fig. 11.40.

11.5.8 Longitudinal Coherence

We discussed the role of the detector geometry in imposing

transverse incoherence to the HAADF image on the scale of
atomic spacings, by ensuring neighboring columns contribute
 320 11 Imaging in STEM
Part I

Fig. 11.39 Simulated Ronchigrams as an uncorrected 300 kV probe is scanned across the artificial crystal of point atoms used for Fig. 11.38, moving in equal steps from
above one delta function atom to the next (a–e). The upper set is near optimum dark field conditions (CS = 1 mm, aperture semiangle 10 mrad and defocus −44.4 nm).
Overlap regions are bright when the probe is over an atom and dark when in between. This is the source of image contrast. The lower set is for CS = 0, but keeping the
defocus as before, which leads to vertical fringes in the overlap region. As the probe is scanned the fringes are displaced. With several fringes inside the overlap region the
integrated intensity on the detector barely changes as the probe is scanned. With optimum defocus (zero) the overlaps would be uniformly bright or dark similar to those
in the upper row

a by 2= 2 = 7.0 Å. This length along the optic axis exceeds the
typical separations of atoms along a column for a crystal in a
(0.136 nm) -1  = 0 nm major zone axis orientation; a second monolayer of atoms will
P(Q)

 = 30 nm scatter in phase with the first if separated by a distance small on

Coherent
this scale. For typical detector angles used in HAADF imaging
the transverse coherence length is much shorter than the atomic
0 1 2 3 4 5 6 7 8 9 10 11 12 13
spacing whereas the longitudinal coherence length is larger.
b spatial frequency (nm -1)
The intensity from a column of atoms will therefore initially
 = 0 nm (0.093 nm) -1 increase as the square of the number of atoms in the column, but
 = 10 nm as the length exceeds the longitudinal coherence length, scat-
P(Q)

Coherent
tering from atoms at the top and bottom of the column will go
out of phase. We can include the phase factor in the z direction
explicitly with a kinematical description of the scattering. The
0 1 2 3 4 5 6 7 8 9 10 11 12 13
kinematical model includes the thickness effect as a geometric
spatial frequency (nm -1)
phase factor, but does not include any multiple-scattering (dy-
Fig. 11.40 Incoherent transfer functions (black lines) corresponding to the im- namical) effects. Figure 11.41 shows the scattering geometry.
ages shown in Fig. 11.14 for an uncorrected 300 kV STEM with (a) Scherzer The phase difference between atoms at different height z is the
Chapter 11

optimum conditions, (b) an oversized aperture and high defocus giving extended same as that for a defocus by f , see Eq. 11.4. For this reason
transfer into the sub-Ångstrom regime. Shown in red is the transfer including a the effect of a finite column length is sometimes referred to as z
defocus spread of ∆ in each case. The value of the transfer function at zero spa- defocusing. Then the path difference for scattering at an angle 
tial frequency is proportional to the square of the objective aperture radius. The from two atoms a distance z apart is
corresponding damping envelope for phase contrast imaging under the same
conditions is shown as the dotted line z.1 − cos  /  z 2 =2 (11.58)

independently to the detected signal. For a 75 mrad inner de- in the small-angle approximation, and the total amplitude scat-
tector angle in a 300 kV STEM the transverse coherence length tered by a column of length t is
is 0:61= = 0.16 Å, which is much smaller than atomic spac-
Zt
ings. Now we consider coherence in the longitudinal direction. 2z 2

Just as in our considerations of resolution, the coherence en- Ac. / = f ei 2 dz (11.59)

velope is a 3D function, with a dimension along the optic axis 0
that is much longer than the transverse coherence length given
 11.5 Coherent or Incoherent Imaging 321

Fig. 11.41 Scattering geometry for a b

a column of atoms in (a) real space
and (b) reciprocal space showing 

Part I
the Ewald sphere and the kinematic
shape factor. c Image intensity as a z k
function of thickness for Si 〈110〉
with a 2.2 Å probe centered between
the two columns of one dumbbell,
collecting coherent scattering over
the range of 50–150 mrad. The . . . . . . . .s .
probe forming aperture is 10.3 mrad,
CS = 1.3 mm and the accelerating volt-
age is 100 kV. d The effect of phonons
is to produce sidebands shown in red.
With enough phonons, the Ewald c d
sphere effectively integrates over the k
shape factor
Intensity

qz
. . . . . . . .s .

0 40 80 120 160 200

Thickness (Å)

Here f is the atomic scattering amplitude (assumed identical for of the function shown in Fig. 11.41b. The first zero occurs for
each atom) and we have treated the column as a continuum. s = 1=t, so as the crystal thickness increases, the shape factor be-
comes narrower, and moves further away from the Ewald sphere.
The quantity s =  2 =2 is the deviation parameter from W&C; in Therefore, as the thickness increases, the number of atoms con-
reciprocal space it is the distance from the zero-layer diffraction tributing to the scattering increases, but the Ewald sphere cuts
plane to the Ewald sphere along the propagation direction. As further out from the central peak of the shape factor, so that the
drawn in Fig. 11.41b, s is negative, and the columnar amplitude is total amplitude does not increase but oscillates. Figure 11.41c
shows the detected intensity calculated for a probe located cen-
Zt
trally over a dumbbell in Si 〈110〉. The intensity rises rapidly
Ac . / = f e2isz dz (11.60) initially, when the column is short compared to the longitudinal
0 . coherence length, but then the interference becomes destructive
and the intensity goes through several oscillations, never exceed-
i  (11.61) ing the intensity scattered by the very thin crystal. This is exactly
1 − e2ist

Ac . / = f
2s . what we might expect for coherent scattering, but is not the be-
havior we find experimentally (which increases with thickness
i (11.62) for several hundred Å, see Fig. 11.4).
Ac . / = f Œ1 − cos.2st/ − i sin.2st/
2s .

Chapter 11
This discrepancy highlights the importance of thermal
sin.st/ (11.63) vibrations: they destroy the longitudinal coherence along
Ac . / = f Œcos.st/ + i sin.st/
s . the column. 9

sin.st/ ist (11.64)

Ac . / = f e
s . An illuminating way to visualize the effect of phonons on the
columnar interference is to imagine what a single phonon will
and the columnar intensity is then do to the zero-layer diffraction spots. A periodic modulation by a
wave vector qz in the z direction will produce satellite reflections
at the positions of ˙qz , see Fig. 11.41d. These satellite reflec-
 2
sin.st/
Ic . / = f 2 (11.65) tions will have the same shape factor as the main reflection. Now
s ,
at high angles, there are many phonons excited with different q,
which is the shape factor for a crystal of thickness t, as derived and furthermore, multiphonon scattering is dominant. In multi-
in W&C, Chap. 13. The general behavior is clear from the form phonon scattering many phonons are simultaneously created or
 322 11 Imaging in STEM

destroyed by the thermal scattering event with consequently a and z0 in the column. This will result in the introduction of a
large distribution of possible final wave vectors. It is at the end longitudinal coherence length. It is essential to use a phonon
point of the total momentum transfer that we place the shape model for this purpose, since the usual Einstein model approx-
Part I

factor, and so, in effect, large-angle phonon scattering places imates the situation by independently vibrating atoms, which
the shape factor at all positions in reciprocal space, in effect, by definition is an incoherent scattering model. In the Einstein
integrating the scattered intensity over the deviation parameter. model every atom is coherent with itself but incoherent with
The diffuse-scattered intensity is therefore given by all others. To see how the phonons approach this limit, we can
adopt the phonon model developed by Warren (1990) in which
sin.st/ 2 phonons are treated with a Debye dispersion relationship and
Z  
2
ITDS . / = f ds (11.66)
s . equipartition is assumed between all modes. This model is very
convenient because it reduces the time average to a modified
(11.67) Debye–Waller factor,
ITDS . / = f 2 t.
( )
Integrating the shape factor produces an intensity that is propor-

Si.qB ju − u0 j/
−2ik.u−u0 /
i = exp 2M −1 (11.71)
tional to the thickness t weighted by the atomic cross section f 2.
he
qB ju − u0 j
This is the dependence we would expect for incoherent scattering
in a kinematic model; every atom contributes equally throughout
the specimen thickness. Here M = B42 is the usual Debye–Waller factor ( is the scat-
2

tering angle, twice the Bragg angle), Si.x/ is the sine integral
The destruction of z coherence is achieved by phonons with function, qB is the Brillouin zone boundary wave vector in the
wave vectors in the z direction, which overcome the curvature z direction, and B = 8 2 hu2 i, where hu2 i is the mean-square
of the Ewald sphere. To see how phonons break up the column thermal-vibration amplitude in the transverse direction. Equa-
in a real space picture we can introduce the effect of thermal tion 11.71 represents a correlation function for atoms at different
motion into the columnar scattering amplitude. Because the heights in the column and is plotted in Fig. 11.42. Closely spaced
fast electron moves at around half the speed of light, for spec- atoms scatter coherently, but atoms far apart have a coherence
imens of the order of 15 nm thick they spend only 10−16 s in that reaches a limiting value of e−2M, which is the value from the
the specimen, which is a small time compared to the phonon Einstein model of independently vibrating atoms.
vibration period, typically 10−12 s. Therefore, the fast electron
sees a ‘frozen’ atomic configuration, and subsequent electrons
see a different configuration. This is the basis of the so-called The Warren model shows how near-neighbor atoms are
‘frozen lattice’ or ‘frozen phonon’ method for image simulation. correlated in position because the phonons have finite
Analytically, we can treat the column using a time average of wave vectors. The situation is reminiscent of the forma-
atomic displacements. Consider each atom to have an instan- tion of the STEM probe, where a sum of different trans-
taneous displacement of u where we now consider not just the verse wave vectors are in phase for some distance before
transverse component of the scattered electron wave vector, but destructive interference sets in. It is the same with the
the total scattered wave vector k by including the z component, phonon wave vectors here. Near neighbors vibrate in
i.e., k = .K f ; s/. Then the instantaneous columnar scattering am- phase, so scattering from them remains in phase as they
plitude becomes oscillate. The degree of correlation decreases rapidly with
distance, so the model shows that a vibrating column of
atoms can be thought of as a column of independently
Z
ATDS . / = f 2 e−2iku e2isz dz (11.68) vibrating packets, as shown in Fig. 11.42b. 9

and the intensity becomes

Chapter 11

For a column that is short compared to the length of the packet,

Zt Zt the scattering is coherent, but for a column of n packets, the in-
ik u ik u0 isz0 tensity is n times the intensity of a single packet, i.e., it increases
ITDS / = f2 he−2 ie2 isz
dz he2 ie−2 dz0 (11.69)
0 0 . linearly (again this is a kinematic model with no absorption or
multiple scattering effects). The intensity may not increase at
Zt Zt the same rate as in the Einstein model; however, the intensity
ik .u−u0 / is.z−z0 / from the packet can be enhanced or reduced compared to the
ITDS /=f 2
he−2 ie2 dzdz0 (11.70) incoherent value depending on whether the correlations within
0 0 the packet are constructive or destructive. The Warren model
conveniently captures the mechanism whereby thermal vibra-
The angled brackets denote the time average between suc- tions destroy long-range correlations between atoms, bridging
cessive electrons, which see different instantaneous displace- the gap between a full phonon description of the scattering and
ments. Note how this expression couples atoms at heights z the incoherent Einstein model.
 11.6 Dynamical Diffraction 323

a b Now K f is the transverse scattering vector, which is large, and s

1
is the longitudinal scattering vector that is much smaller. There-
fore the most effective phonons in the destruction of coherence

Part I
0.8
Degree of coherence

are those propagating in the z direction but with transverse dis-

placements.
0.6 0
1
2 Plots of the thickness dependence for a column of Rh atoms
0.4
3 are shown in Fig. 11.43, showing the changeover from coherent
behavior to incoherent behavior as the scattering angle increases.
0.2

0 The effect of thermal displacements depends strongly on

0 1 2 3 4 5 angle because of the appearance of  2 in the exponential.
Atom separation (n)
 9
Fig. 11.42 a The degree of coherence between one atom in a column of atoms
and its neighbors n atoms away, calculated with a phonon model (red) and in an
Einstein model (green). The phonon model shows how the degree of correlation
falls smoothly with increasing n until it reaches the Einstein value at large n. In
the Einstein model all atoms are assigned this degree of correlation except the 11.6 Dynamical Diffraction
atom at the origin, which is coherent with itself on both models. b Schematic
showing how thermal vibrations break up a column into independently vibrat-
ing packets, inside which the scattering is highly correlated. Calculations are for
We are now going to delve into Bloch theory. So if you skipped
B = 0.45, = 75 mrad at 100 kV
Chap. 14 in W&C, you’ll either want to step back to it now or
jump ahead to the discussion after Eq. 11.92.
We have already seen that it is phonons with wave vectors in the
z direction that are required to break longitudinal coherence. If The kinematical model shows us how the scattering geometry
we expand the thermal displacement into transverse and longi- coupled with thermal motion completely changes the depen-
tudinal components, as u = .U; uz /, we obtain dence of scattered intensity on crystal thickness. However, it
does not include any multiple scattering and is not realistic for
(11.72) the imaging of thick crystals in a zone-axis geometry where
D E D E
e−2ik.u−u / = e−2iKf .U−U / e−2is.uz −uz /
0 0 0

strong dynamical scattering takes place. From Fig. 11.4 we

know that this multiple scattering has no apparent effect on the

a b c
Coherent Warren Einstein
120 10 4

3.5
100
8
3
80
2.5
6
Intensity

Intensity

Intensity

60 2

Chapter 11
4
1.5
40
1
2
20
0.5

0 0 0
0 10 20 30 40 50 60 0 10 20 30 40 50 60 0 10 20 30 40 50 60
Atom Atom Atom
Fig. 11.43 Thickness dependence for a column of Rh atoms spaced 2.7 Å apart along the beam direction, for inner detector angles of (a) 15, (b) 30, and (c) 45 mrad,
calculated on a kinematic model assuming coherent scattering (black), and with thermal vibrations using a phonon model (red) and the Einstein model (green)

Licensed to ([email protected])
 324 11 Imaging in STEM

nature of the image. We can understand a simple explanation The incident probe amplitude in reciprocal space is a set of ab-
for this if we recognize that the HAADF detector is large on the errated plane waves of unity amplitude, given in Eq. 11.44 as
scale of the diffraction angles. Multiple scattering rearranges
Part I

intensity as a result of dynamical diffraction; it locally changes P.K/ = A.K/e−i.K/ e−2iKR0. (11.79)
the details of the interference pattern seen on the detector plane.
R0 is again the probe coordinate. Integrating these plane wave
components gives the total amplitude scattered to Kf from depth
Interference does not change the total scattered intensity, z inside the crystal,
it just redistributes it. 9 Z
 .K f ; R0 ; z/ = P.K/ .K f ; z/dK (11.80)
We might anticipate that the HAADF image would, therefore, be
insensitive to dynamical interference. A phase-contrast image us- Z X j
ing a point detector would naturally be sensitive to local changes  .K f ; R0 ; z/ = A.K/e−i.K/ 0 .K/
in the interference pattern. j
X j
gj .K/ı.K + g − Kf /e2ikz .K/z e−2iKR0 dK
A convenient way to see the physics of the dynamical diffraction g
is to incorporate dynamical effects into the expression for the (11.81)
probe intensity, so that it then describes the probe propagating
through the crystal. Each plane-wave component incident at wave Comparing Eq. 11.80 with Eq. 11.28 you can see how the
vector K is expanded into a set of Bloch states inside the crystal, Bloch states are acting as a kind of dynamical transmission
function for the probe. The Bloch state components contain
j j
all the information on the specimen scattering power. If we
X
 .R; z/ = "j .K/bj .K; R/e2ikz .K/z e− .K/z (11.73)
j now perform the integral over K, the delta function forces
K = K f − g and we obtain
Here "j is the excitation amplitude of Bloch state j, denoted bj, X j
with wave vector kzj in the z direction and absorption coefficient  .K f ; R0 ; z/ =A.K f − g/e−i.Kf −g/ 0 .K f − g/
j. For simplicity we will ignore absorption. The Bloch states j
have the periodicity of the lattice and can be expanded in a set X j
2ikz .K f −g/z −2i.K f −g/R0
(11.82)
gj .K f − g/e e
of plane waves as g

Now we can find the intensity at K f by squaring,

X
bj .K; R/ = gj .K/e2i.K+g/R (11.74)
g
K f −g/ j
I.K f ; R0 ; z/ =A.K f − g/e−i
X
0 .Kf − g/
The excitation amplitude is " .K/ = j
0 .K/.
We now obtain a re-
j
j
ciprocal-space expression for the image intensity. Transforming j
ikz .Kf −g/z −2 i.Kf −g/ R0
j
− g/e2
X
the Bloch states into reciprocal space gives the component at K f g .K f e
in the detector plane g (11.83)
i Kf −h/ k
X
Z A .K f − h/e 0 .K f − h/
bj .K f / = bj .K; R/e−2iK f R dR (11.75) k
ikzk .Kf −h/z 2 i.K f −h/ R0
k
.K f − h/e−2
X
h e
Z X h
bj .K f / = gj .K/e2i.K+g/R e−2i.Kf R/ dR
(11.76)
Chapter 11

g which couples pairs of Bloch states j and k. We can rearrange

Eq. 11.83 to give Eq. 11.84.
X
bj .K f / = gj .K/ı.K + g − Kf /
g
(11.77) I.K f ; R0 ; z/ =
X
A.K f − g/A .K f − h/e−i Kf −g/ i Kf −h/
e
g;h
j j
X
This delta function couples the incident and final wave vectors 0 .K f − g g .K f − g/
via the g component of the Bloch state. The reciprocal space j;k (11.84)
form of Eq. 11.73 becomes k k
0 .K f − h h .K f − h/
j
h i
j 2 i kz .Kf −g/−kzk .Kf −h/ z −2 i.h−g/ R0
X j
 .K f ; z/ = 0 .K/bj .K f /e2kz .K/z e e
(11.78)
j
 11.6 Dynamical Diffraction 325

We now take the FT with respect to R0, the probe position; R0 is Now the periodicity of the Bloch states means gk .K + h/ = g+h
k
.K/
the only real-space component remaining in the expression. We and kzk .K + g/ = kzk .K/ , so we arrive at the general reciprocal
again label the spatial frequency with respect to probe position space expression for the image intensity from the crystal at depth

Part I
as Q, the spatial frequency in our STEM image, z, in terms of a single integral over the incident probe,
Z Z
I.K f ; Q; z/ = I.K f ; R0 ; z/e−2iQR0 dR0 (11.85) A.K/A .K + Q/
X
I.Q; z/ = D.K + g/
. g
j
X
I.K f ; Q; z/ = A.K f − g/A .Kf − h/ −i K/ i K+Q/ j
(11.90)
X
e e 0 .K g .K/
g;h j;k
Kf −g/ i K f −h/
e−i e j
h i
k k 2 i kz .K/−kzk .K+Q/ z
Q .K g .K/e dK
j j
X
0 .K f − g g .K f − g/
j;k
(11.86) Notice that the only terms that depend on the detector are the
k k
0 .K f − h h .K f − h/ Bloch state components, and we can rewrite Eq. 11.90 as
j
h i
2 i kz .K f −g/−kzk .K f −h/ z
e I.Q; z/ =A.K/A .K + Q/e−i K/ ei K+Q/
Z
j
e−i2 i.h−g+Q/ R0
h i
2 i kz .K/−kzk .K+Q/ z
X j
dR0 k
0 .K Q .K/Cjk .K/e (11.91)
j;k

Performing the integral over R0 gives the delta function

ı.h − g + Q/, which allows us to perform the sum over h, which The detector sum Cjk .K/ is just a sum of Bloch wave intensity
sets h = g − Q, thus eliminating h to give contributions to the image.

D.K + g gj .K / gk .K/
X
Kf −g/ Cjk .K/ =
A.K f − g/A .K f − g + Q/e−i
X
I.K f ; Q; z/ = g (11.92)
g
j
i Kf−g+Q/ j
This expression for Cjk .K/ allows us to evaluate the contribu-
X
e 0 .K f −g g .K f − g/
j;k (11.87) tion of different Bloch state pairs to the image intensity. Now
k k we can directly appreciate the physics of the image formation.
0 .K f − g + Q g−Q .K f − g + Q/
Exactly as discussed in the context of imaging phase objects,
the HAADF detector selects Bloch states that are sharply peaked
j
h i
2 i kz .Kf −g/−kzk .Kf −g+Q/ z
e dKf
on the scale of the Airy disc of the hole in the detector. The
only such states are the 1s states. The action of the detector
We now integrate over the detector as before to obtain the image is therefore to filter the states that contribute to the dynamical
intensity diffraction.

The effect can be seen in Fig. 11.44, where the partial images
Z
A.K f − g/A .K f − g + Q/
X
I.Q; z/ = D.K f / calculated from particular Bloch states are shown for the case of
g InAs 〈110〉 at 100 kV. The images are shown for the limit of
i Kf −g/ i K f −g+Q/ j high thickness where the oscillation of the exponential term in
X
e e 0 .K f − g/
j;k Eq. 11.91 destroys all cross terms leaving only those terms with
(11.88) j = k, the independent contributions of the Bloch states. Clearly,

Chapter 11
j k k
g .K f − g/ 0 .K f − g + Q g−Q .K f − g + Q/ the 1s states on the In and As columns are responsible for al-
j
most all of the image contrast. This is despite the fact that their
h i
2 i kz .K f −g/−kzk .K f −g+Q/ z
e dK f
excitation coefficients are small (0.19 and 0.24 respectively)
Substi- compared to an excitation of 0.81 for the In 2s state. The In 2s
tuting K f − g by K results in state would contribute strongly to a bright field image, however,
Z
K/
with a 30 mrad inner detector angle the Cjj .0/ for the In 2s state
A.K/A .K + Q/e−i
X
I.Q; z/ = D.K + g/ is only 0.02, much smaller than the values of 0.16 and 0.08 for
g the 1s states of In and As. Thus, the In 2s state contributes only
ei K+Q/ j j weakly to the Z-contrast image despite its high contribution to
X
0 .K g .K/
j;k the bright-field image.
(11.89)
k k
0 .K + Q/ g−Q .K + Q/ zz Physically, the 2s state is a broader state in real space and so
j
passes mostly through the hole in the detector. The Bloch-
h i
2 i kz .K/−kzk .K+Q/ z
e dK
 326 11 Imaging in STEM

corresponding to the atomic columns. In this case, they therefore

propagate almost as plane waves. That means that each individ-
ual plane wave may not have a significant Cjj .K/ but because all
Part I

the plane waves are essentially in phase, as they are outside the
crystal, then the sum total contribution can be quite significant.

It is best to think of the probe as a sum of two components:

zz a channeling component represented by the 1s state, and

zz a non-channeling component represented by the sum of all
other states.

The non-channeling component is dominated by the plane waves

around the periphery of the aperture, as shown schematically
in Fig. 11.45a, b. This latter component will come to a focus
at a specific depth inside the crystal, much as it would outside,
whereas the 1s component tends to channel and explore more of
the column (until it is scattered out to the detector). Therefore,
as the probe-forming angle is increased we see a change over
from a channeling (projection) behavior to a depth-sectioning
behavior, even with a crystal aligned on a zone axis. The effect
is shown in Fig. 11.45c, d, where simulations are shown for Bi
atoms inside Si 〈110〉, which come into focus at the appro-
priate focus setting.

11.7 Other Sources of Image Contrast

Now we have to discuss sources of image contrast that are

not related to the atomic-scattering factor itself: we include
contrast due to changes in the thermal vibration amplitude and
contrast due to static strains. We have seen how thermal dis-
Fig. 11.44 Contributions of different Bloch states to the HAADF image of InAs. placements break the coherence of an atomic column in the
a In 1s state, b As 1s state, c both 1s states, and d all states, calculated from z direction, and as a result can greatly enhance the intensity
Eq. 11.91 using 311 beams and an inner detector angle of 30 mrad reaching the annular detector. The thermal vibration amplitude
can therefore be a significant source of image contrast. For a
simple picture of how this works we can adopt the Einstein
state filtering action of the high-angle detector is one key model and split the atomic-scattering factor into coherent and
reason why Z-contrast images tend to be intuitive in nature. incoherent parts.
1s states are located on the atom columns, and the scattering
cross section means they contribute according to their atomic The coherent scattering is reduced by the Debye–Waller factor
number. giving a cross-section for the total coherent scattering of
Chapter 11

zz The second key ingredient for incoherent imaging is again

the action of the phonons. As seen in the kinematic descrip- c = f 2 e−2M (11.93)
tion, they introduce incoherence over the thickness of the
specimen. The final result is an image that shows incoherent The cross-section for incoherent scattering is the remainder,
characteristics and Z-contrast. This explains why thickness
TDS = f 2 − c = f 2 1 − e−2M
 
fringes were seen in the bright-field image of Si, Fig. 11.4a, (11.94)
but little evidence of thickness oscillations was seen in the
ADF image, Fig. 11.4b. Remember that M = B ; then we see that with increasing an-
2
42
gle there is a rapid change over from coherent to incoherent
This picture needs some modification in the era of aberration cor- scattering. At a sufficiently high angle c tends to zero and TDS
rection because probe angles can increase to where the high-an- becomes the full atomic cross-section, f 2. However, at lower
gle components of the incident probe can be at such high angles angles, when there is a significant c. Equation 11.94 shows that
to the zone axis that they no longer see the deep potential wells the incoherent contribution from each atom will be reduced from
 11.7 Other Sources of Image Contrast 327

a b a b

Part I
c d
0

9
38.4 Fig. 11.46 Images of an Al-Co-Ni decagonal quasicrystal using an annular
Top
detector with inner angle of (a) 50 mrad, giving a Z-contrast image, and (b)
35 mrad which shows bright contrast in the centers of the quasiperiodically cor-
9 related 2 nm clusters due to thermal vibration anomalies. Images recorded using
Intensity

76.8 a 100 kV aberration-corrected STEM

9
115.2 the passage of the fast electron; random, static displacements are
equally effective as time varying ones. In this case it is usual to
define a static Debye–Waller factor contribution by defining a
9 total Debye–Waller factor as BTotal = B + Bs, where Bs = 8 2 hu2s i
153.6
0 2 4 6 8 10 and hu2s i is the mean square static atomic displacement in the
z, Å x, Å transverse direction. Clearly, random atomic displacements
comparable in scale with the thermal vibration amplitude will
Fig. 11.45 Schematic showing the contribution to a HAADF image of (a) the 1s be significant sources of image contrast, and since the atomic
states, which channel down columns until they are scattered, and (b) all other vibration amplitude is a small fraction of 1 Å (e.g., ~ 7.5 pm for
states, which, for a sufficiently large probe-forming aperture, can be focused at Si at room temperature), this is can be a sensitive measure of
different depths inside the crystal. (c) Model crystal of Bi atoms in different col- static displacements. An example of strain contrast at a Si/SiO2
umns at different depths. d HAADF intensity profiles with the probe focused at interface is shown in Fig. 11.47.
the four depths indicated in (c) show strong peaks at the corresponding Bi atoms
The random displacements that give enhanced scattering at
intermediate angles mean that the excitation of the 1s Bloch
the full f 2. This reduction is not compensated for by the coherent state is slightly lower, and the intensity scattered to high angles
component, because, as we have seen in Fig. 11.41c and 11.43, is slightly reduced. This effect is just apparent in Fig. 11.47,
once the column is longer than the longitudinal coherence length, but is much clearer in the images of an interface between a
much of the coherent scattering interferes destructively and we LaAlO3 (LAO) substrate and a La0.67Ca0.33MnO3 (LCMO) film
detect only a small fraction of c from each atom. It is in this (Fig. 11.48). Images obtained at two inner-detector-angles show
angular range that the scattered intensity becomes very sensitive a contrast reversal between the two. In both regions the image
to changes in the Debye–Waller factor. intensity is dominated by the La-containing columns (Z = 57
for La compared to 25 for Mn). For high detector-angles the
An example of this reduction is shown in Fig. 11.46; the images LCMO film shows weaker intensity than the LAO film, as
are from an Al-Co-Ni decagonal quasicrystal using two differ- would be expected for the reduced mean-square atomic num-

Chapter 11
ent detector inner-angles. With the high inner-angle a Z-contrast ber due to the Ca substitution. For lower angles, however, the
image is formed, similar to those shown in Fig. 11.7, but with LCMO shows higher intensity than LAO. The reason is that in
the lower inner-angle we see that the central region of each 2 nm LAO the La columns are straight, whereas they become zig-
ring shows bright. These are exactly the regions showing the zag columns in LCMO (viewed along the Pnma b-axis). The
broken symmetry in Fig. 11.7, and are quasiperiodically cor- transverse atomic displacements lead to an increased atomic
related. Raising the temperature can also enhance the contrast scattering cross-section.
of the central ring, proving that the origin of the contrast is an
enhanced thermal vibration amplitude from these atomic sites, We can make a simple estimate of the expected image con-
roughly by a factor of 2. trast by calculating the 1s state Bloch wave intensities in the
two structures and using Eq. 11.94 for the atomic-scattering
It is not necessary for the atoms to be actually vibrating to show cross-section. The calculated Bloch-wave amplitude at the atom
enhanced contrast, static displacements can be sufficient. As we sites reduces to 91 % of its value in a straight column, which
saw earlier, phonon configurations are effectively frozen during represents the reduced channeling effect of the zig-zag column
 328 11 Imaging in STEM

a b a b

LAO LCMO LAO LCMO

Part I

c d 2 nm

ADF Intensity
Fig. 11.47 Strain contrast at a Si/SiO2 interface is observed by comparing imag-
es with (a) a low inner detector angle of 34 mrad and (b) a higher inner angle of
43 mrad. The intensity profiles below each image are taken through the dumb-
bells outlined in white, and show an increased intensity near the interface in (a)
0 20 40 60 80
which is absent in (b). It therefore cannot be due to Z-contrast, but indicates
Inner detector angle (mrad)
some additional random atomic displacements in this region. Images recorded
with an uncorrected 300 kV STEM Fig. 11.48 Annular dark field images of a La0.67Ca0.33MnO3 film on LaAlO3 ob-
tained with an uncorrected 100 kV STEM with an inner detector angle of (a)
25 mrad (b) 75 mrad showing a contrast reversal. c Calculated intensity as a
function of inner detector angle for the La column (red) and the La0.67Ca0.33 col-
umn (blue)
Chapter 11

Fig. 11.49 a Bright field and (b) ADF

images of a Si layer containing 50 nm
B-doped layers (3.5 × 1020 B cm−3). The
B-doped layers and threading disloca-
tions are visible by strain contrast
 11.8 Image Processing 329

compared to a straight column because of the blurred projected a

potential. However, at low inner-detector-angles, this reduced
illumination of the atoms, and their reduced mean square Z, is

Part I
more than compensated by the effect of static displacements.
Figure 11.48c shows the calculated dependence of the scat-
tered intensity on the inner detector angle using these simple
approximations for the image contrast and it predicts a contrast
reversal as observed. Quantitative analysis is, of course, better
when based on accurate simulations, but it is remarkable how
the basic form of the image can be understood using simple
physical principles.

Defects that show in diffraction contrast such as dislocations

and stacking faults will also show contrast changes in the
HAADF image, although the contrast mechanism is better
termed ‘channeling contrast’ in this case; the changing exci-
tations of the various Bloch states alters the illumination of
each atom and hence the intensity scattered to high angles. 0.05 m
An example of the imaging of inclined dislocations in a su-
perlattice of Si and B-doped Si layers is shown in Fig. 11.49.
Near the beam-entrance surface the dislocations show oscilla- b
tory contrast, whereas deeper down, when the 1s states have
been absorbed, the dislocation strain field regenerates some 1s
state contribution, so the dislocations appear to be bright. The
B-doped layers are also seen as bright in the HAADF image
because the B introduces random atomic displacements in the
surrounding Si atoms, and these displacements lead to strain
contrast, as we saw in Fig. 11.47.

Another interesting example of strain contrast is seen in

Fig. 11.50a, where surface relaxations around dislocations
emerging at the surface of GaN cause sufficient transverse dis-
placements to be visible in the HAADF image. They are vis-
ible even if the dislocations, such as the screw dislocation in
Fig. 11.50b, are not visible by normal diffraction contrast. The
displacements cause a slight blurring of the atomic resolution
image. This effect is very useful for locating the dislocation cores
to allow study by EELS, for example.
0.4 nm

11.8 Image Processing Fig. 11.50 a Low magnification ADF image of GaN viewed along 〈0001〉
showing threading dislocations as bright dots due to their strain field. Approx-
imately 20 % of them are pure screw dislocations, as seen in the high magnifi-

Chapter 11
The most common form of image processing is a simple Fourier cation image (b)
filter used to reduce the noise present in the image, as shown in
the image of a CdSe/CdS/ZnS core/shell nanorod, Fig. 11.51.
The raw image shows the dumbbells clearly resolved. Applica-
tion of a band-pass Fourier filter greatly enhances the contrast Fourier filtering is a direct method in which the frequency
of the dumbbells and removes much of the uniform background content of the data is altered. As a Z-contrast image is an
present in the raw data. In both images the envelope of the peak incoherent image, essentially a convolution of the scat-
intensities reveals the nanocrystal thickness, i.e., its shape in the tering power of the object with the probe, it might be
third dimension, as seen in the schematic. expected that deconvolution of the probe would recover
the object. Unfortunately, the convolution loses all infor-
mation on the high frequencies in the object, and these
cannot be restored just by changing the frequency content
of the image. 9
 330 11 Imaging in STEM
Part I
Chapter 11
 11.8 Image Processing 331

a b c

Part I
1 nm

Fig. 11.52 a Z-contrast image of Co-doped anatase TiO2, viewed along the [ 100 ] direction, on a LaO2 substrate, seen on the left. The image is smoothed to reduce
noise, except in the inset, which shows the raw data. This region is shown in (b) after maximum entropy processing, using the HREM Research Inc. software package
and a probe of 0.071 nm FWHM. Red circles mark O columns while gray circles correspond to Ti/O columns. (c) Intensity traces along the atomic planes marked with
red arrows on (b), compared to traces from the raw image before deconvolution. The blue lines represent an average over 5 lateral pixels as indicated by the white
box on (b). The red dots represent the intensity traces from the same regions averaged in the same way on the raw data. It is seen that the bright O column actually
comprises two peaks: one (red arrow) in the expected position for O and the other (green arrow) displaced away from the O column. This peak is due to preferential
segregation of Co interstitials and is exactly at the predicted positions of Co interstitials from density functional theory, and explains the origin of ferromagnetism in
this system. Images obtained with an aberration-corrected 300 kV STEM

Fourier methods can introduce spurious details into the image, zz Principal component analysis, based on multivariate statisti-
as in the filter shown in Fig. 11.51f, g. Indirect methods are the cal analysis, has been very useful for EELS images, as we’ll
preferred alternative to Fourier deconvolution as they tend to see in Chap. 16. The entire data set is analyzed into inde-
introduce less spurious detail. They involve only forward mod- pendent eigenvalues. Spectral features that are most highly
eling, that is, the simulation of trial structures with comparison correlated form the principle components, whereas those that
to the experimental data, and are much more robust, although are uncorrelated are presumably less significant and are more
computationally more intensive. likely to represent noise. These components can be removed
from the data set, and in this way noise reduction is obtained
zz One common indirect approach is maximum entropy. Some without significant loss of spectral or spatial resolution.
assumptions are needed to recover the lost high frequency
information, because, in principle, there are an infinite num- You must always be very careful when image processing. A fea-
ber of structures that can fit any image. The maximum en- ture at an interface may be very significant but only a minor
tropy algorithm weights in the direction of the least structure, component because it appears in only a small fraction of the
i.e., maximum disorder, and will only put the least number total pixels. Remember the golden rule: you should submit your
of atomic columns necessary to account for the peaks in an unprocessed data when you submit your processed image for

Chapter 11
image. This is therefore very appropriate for Z-contrast imag- publication. In practice this is rarely done! Today, the US NSF
ing, and an example of its application to the detection of Co actually requires that data obtained with their funding remain
interstitials in TiO2 is shown in Fig. 11.52. available for others to use.
zz Another method that has been very successful is the pixon ap-
proach, which is not based on entropy but on seeking correla-
tions between adjacent pixels to distinguish structure (the ob-
ject) from noise. These patches of correlated pixels are called
pixons. An example of this method is shown in Fig. 11.53.

Fig. 11.51 Z-contrast images of a CdSe/CdS/ZnS core/shell nanocrystal viewed along the [010] direction recorded with a 300 kV aberration-corrected STEM, (a) raw
data, (b) band-pass Fourier filtered. A line trace across the raw image (c) taken across the long axis of the nanocrystal reveals the thickness profile of the core. The
sublattice polarity is also directly visible, but is more clearly seen from a line trace across the filtered image (e). This nanocrystal is oriented as shown in the ball and
stick model (d), the beam direction being vertical, and the scan direction for the thickness profile being shown by the green arrow. (f) Raw image of a CdSe nanocrys-
tal embedded in polymer recorded with an uncorrected 300 kV STEM. In this case attempts to enhance the resolution by Fourier filtering (g) introduces spurious detail
 332 11 Imaging in STEM
Part I

Fig. 11.53 a Raw Z-contrast image of a disordered intergranular film in polycrystalline β-Si3N4. The grain on the right-hand side is oriented along the [0001] direc-
tion and the interface is formed by the prismatic plane, which is seen edge on. The image reveals the attachment sites of La atoms, which are added as a dopant
to improve mechanical properties. b Pixon image processing removes the noise very effectively. Observed sites are in reasonable agreement with calculated sites
(circled) for La segregation on the bare prismatic surface

image, as seen experimentally. Removing those columns gives

11.9 Image Simulation no such streaks.

Bloch vs Multislice
So far we have been examining the physical mechanisms under-
lying image formation in the STEM, not attempting to perform
Chapter 11

an accurate image simulation. When talking about thickness As in simulation of TEM images, two distinct methods
dependence of the image and strain contrast, we have seen the are in use, the Bloch wave approach and the multislice
rapid change over from coherent to incoherent scattering with method. 9
increasing angle. None of the treatments attempted to provide a
quantitative simulation that includes both dynamical diffraction
and thermal scattering, which is necessary for a full simulation. Several methods have been developed over the years to de-
We get a rough indication of the overall form of a HAADF scribe accurately the dynamical diffraction and scattering to the
image by just modeling the HAADF image as a convolution high-angle detector. Here we will summarize the different ap-
of the incident probe intensity profile with an object function proaches to HAADF image simulation.
comprising delta functions at the atomic sites weighted by the
Z2 Rutherford scattering cross-section. Although neither dy-
namical nor thermal scattering are explicitly included, it often
works surprisingly well, as shown in Fig. 11.54. The presence
of nitrogen columns in AlN is predicted to give streaks in the
 11.9 Image Simulation 333

a b c d

Part I
aC
d2

Ca
d1

aB bAc
d1

d2

bA
0.25nm

Fig. 11.54 a Z-contrast image of an antiphase boundary in AlN from an uncorrected 300 kV STEM with derived atomic structure (b). Large circles denote Al, small
circles N, white and black denote different atomic heights, and upper and lower case letters indicate Al and N planes, respectively. c Simulation using a convolution
of the probe intensity profile with a Z2 weighting at the atomic sites. d Simulation without the N columns removes the zig-zag streaks between the Al columns. These
features, seen in the raw image, are due to the N columns, which are only 1.09 Å from the Al column and therefore unresolved. The spacing between the first Al plane
and the next, labeled d1, is measured to be approximately 0.28 ± 0.02 nm, significantly larger than the AlN (0001) planar spacing of 0.25 nm, denoted by d2. The
expansion is due to substitution of O for N in the red locations, as detected by EELS

11.9.1 Bloch Waves When you’re ready, go to Earl Kirkland’s book, which includes
the necessary computer codes.

Early analysis of the HAADF image took the view expressed The frozen-phonon method simulates the phonon scattering by
in Fig. 11.3, that high-angle scattering is diffuse, and therefore performing a number of coherent calculations over an ensemble
generated incoherently by each atom in proportion to the inten- of displaced atoms, which is a semiclassical approximation to
sity of the probe at its site. We then replace the detector by a the process of excitation and annihilation of phonons. Never-
scattering cross-section, calculated for the appropriate range of theless, unlike when we use an absorptive potential, electrons
angles, and treat the probe propagation dynamically using Bloch are not now removed from the calculation and continue to prop-
waves. We use an absorptive potential both to reduce the Bloch agate through the crystal. They can therefore undergo further
wave amplitudes with propagation through the crystal and to scattering events. Frozen-phonon simulations therefore contain
generate the high-angle incoherent scattering. This model is only Kikuchi line features in the convergent-beam diffraction pattern
appropriate at high scattering-angles where there is negligible due to elastic scattering after the phonon scattering event, and
coherent component. with enough configurations an excellent agreement with experi-
ment is obtained. The method has been extended to include pho-

Chapter 11
non-dispersion curves, although, due to the detector integration,
the changes are minimal compared to the Einstein model. The
11.9.2 Multislice disadvantage of the frozen-phonon method is the need for a large
number of beams (covering the entire range of the detector) and
use of a number of phonon configurations to achieve conver-
Early multislice simulations made the opposite assumptions! We gence in the simulation.
tracked all the coherent scattering onto the detector and found
a depth-dependent contrast, although the transverse form of the
image appeared correct. From the discussions of the previous
sections we can understand that transverse incoherence as a 11.9.3 Bloch Waves with Absorption
result of the detector integration would give the approximate
transverse form of the image, but the coherence in the longitu-
dinal direction would give thickness oscillations. Introduction of The Bloch-wave methods include a phenomenological absorp-
the frozen-phonon technique broke the longitudinal coherence. tion model by adding an imaginary part to the crystal potential,

Licensed to ([email protected])
 334 11 Imaging in STEM
Part I

Fig. 11.55 a Experimental HAADF images of SrTiO3 along 〈100〉 with intensity variations normalized to the incident beam intensity (see color scale bar (d)).
Regions of three different thicknesses are shown. The Sr columns are the brightest and the Ti-O columns are the second brightest features (see unit cell schematic
on the left). The image of the 105 nm region has been drift corrected. b frozen phonon image simulations. c Bloch wave image simulations. In each case, simulations
are shown without (left pane) and with convolution with a 0.08 nm FWHM Gaussian (right pane)

in which case the electrons are lost to the calculation. The in- 11.9.4 There Is No Stobb’s Factor in HAADF
coherent component to the image is then calculated through a
separate absorptive potential, which instead of modeling the rate
of loss of electrons, models the rate of generation of high-angle Experimental, phase-contrast, TEM-image simulations histori-
scattering cally have always significantly overestimated the image contrast
compared to experiment by the Stobbs factor. Recently, however,
it has been shown that an excellent quantitative fit can be ob-
Z
2
I.R0 / = V HA .R/ jP.R − R0 ; z/j2 dRdz (11.95)
hv tained to experimental ADF images using just a Gaussian source-
size contribution that fits the expected source size. Figure 11.55
Here VHA(R) is the potential for scattering onto the detector, as shows the results of Bloch wave and frozen-phonon simulations
shown in Fig. 11.34, and P.R − R0 ; z/ is the dynamical probe for SrTiO3 compared to experiment. Excellent quantitative agree-
wave function at depth z. This method can be very efficient as ment is seen with both methods for a thin crystal, but the Bloch
it is only necessary to include sufficient low-order beams to de- wave method fails in thicker crystals for the reasons discussed
scribe the dynamical diffraction of the probe. The high-angle above. This quantitative agreement with experiment is a very
scattering is generated kinematically; i.e., there is negligible significant result since it shows for the first time that HAADF
Chapter 11

probability for a second high-angle scattering event that would images can be well simulated using existing descriptions of elas-
return the scattered electron into the dynamical wave field. tic and thermal diffuse scattering.
Therefore, the number of beams may not need to extend even
as far as the inner angle of the detector. Of course, if coherent We now know that STEM bright field images are free of any
scattering is significant, then the beams must extend onto the Stobbs factor, clearly suggesting that our present understanding
detector to allow their interference, and the calculation becomes of imaging is sufficient. So, the Stobbs factor in TEM must be
more involved. This approach fails in thicker and/or higher Z due to something else: the detector response.
crystals when this single scattering approximation for the high
angle intensity breaks down. Quantification can also be done in an empirical manner, even
if this sounds like a contradiction in terms! Figure 11.56 shows
a compositional profile obtained from a Z-contrast image of an
InAsxP1–x wetting layer in between InP barrier layers. In this case,
the InP acts as an intensity reference, and the As composition
 11.10 Future Directions 335

profile is obtained by comparing intensity ratios measured from

group V columns in the wetting layer to those measured in In-
AsxP1–x thin films, with uniform composition that was calibrated

Part I
by high resolution X-ray diffraction. Over a moderate range
of specimen thicknesses the intensity ratios are independent
of thickness, and the extracted profile is in agreement with the
composition deduced from photoluminescence measurements.

11.10 Future Directions

The success of aberration correction has opened the door to a

whole range of new insights into the origin of materials proper-
ties using the STEM.

zz In ferroelectrics, it is possible to map local displacements on

a cell-by-cell basis and image the influence of defects and
surfaces on local polarization.
zz With EELS it is possible to track O vacancies in oxide thin
films, even to map spin states through the O K edge fine struc-
ture.
zz We have seen the development of theories for near-edge struc-
ture that properly take account of the dynamical diffraction of
the probe and solid-state bonding.
zz Combining microscopy data with DFT calculations, the ori-
gins of materials properties are being unraveled at the funda-
mental atomic level.

For example, the origin of the eight orders of magnitude en-

hanced ionic conductivity in strained epitaxial ZrO2:Y2O3/SrTiO3
heterostructures was shown to be a combination of the expansive
strain and the incommensurability of the fluorite and perovskite
O sublattices. The origin of white-light emission in sub-2 nm
CdSe nanocrystals was shown to be due to a fluxional state in-
duced by photon absorption.

Fig. 11.56 Quantification of As composition in InAsxP1-x from the Z-contrast

image. a Ratio of columnar intensities for AsxP1–x compared to pure P from exper-
iment and simulation. b Z-contrast image of a wetting layer of InAsxP1–x between

Chapter 11
InP barriers. c Deduced composition profile. Note the asymmetric form of the
profile along the growth direction
 336 11 Imaging in STEM

Chapter Summary
Part I

We have come a long way, but do we have all the resolution displacements, down to the picometer level. And depth res-
we need? Clearly, we can solve a lot of problems that were olution would improve fourfold again to the level of a few
unimaginable in 2000. But we still cannot see into amor- Ångstroms. Such gains could require the development of a
phous materials and find where every atom lies, and how monochromator to overcome chromatic aberrations; then it
it is bonded. We cannot yet make movies of point defects would be possible to probe electronic structure sub-surface,
diffusing and reacting inside materials. Enhanced resolution in 3D, in a manner similar to that done today for the surface
would bring these dreams closer to reality, closer to Feyn- by the scanning tunneling microscope. And finally we will
man’s dream in his now-famous lecture “There’s Plenty of be able to make movies of single atoms inside crystals, or on
Room at the Bottom”. In 1959, he called for resolution to be their surfaces, track their bonding and chemical reactions.
improved a hundred times, in order to ‘just look at the thing!’
One thing is quite clear – the future for imaging and spec-
Indeed, another factor of two improvements in lateral reso- troscopy in STEM is truly bright.
lution would enormously enhance our sensitivity to atomic
Chapter 11
 Appendix 337

Appendix

Part I
People Q11.18 What is a Ronchigram?
Q11.19 Explain two uses of the Ronchigram.
Q11.20 We explicitly refer to the year 1993. Why?
Albert Crewe was born in Bradford, Yorkshire, on February 18, Q11.21 ‘Atoms are smaller in HAADF STEM’ appears to be a
1927 and died on November 18, 2009 in Chicago. His thesis nonsense phrase. Explain why it is not.
advisor at the University of Liverpool was Sir James Chadwick, Q11.22 Explain the difference between ‘longitudinal’ and
who discovered the neutron. Crewe demonstrated the first field ‘transverse’ coherence.
emission STEM in 1970. Q11.23 What is the Warren model?
Q11.24 There are two methods for simulating STEM images.
Vasco Ronchi was born in Florence on December 19, 1897 and Are they both used?
died on October 31, 1988. He completed his Ph.D. in 1919 with Q11.25 What is a fluxional state?
Luigi Puccianti; Enrico Fermi started his Ph.D. in the same group
in 1920. Ronchi published ‘The History of Light’ in 1939. In
1945 he founded the Giorgi Ronchi Foundation to honor his son.
Since Ronchi was Italian, Ronchigram is pronounced Ron-kee- Text-Specific Questions
gram.

T11.1 Using the literature, discuss the current state of HAADF

imaging of quasicrystals.
Self-Assessment Questions T11.2 Why is the tail a problem in STEM. Be thorough in your
answer.
T11.3 We can observe light atoms in STEM even though we
Q11.1 What is the key requirement for incoherent imaging? have Rutherford scattering. Explain why this is such an
Q11.2 Explain concisely the difference between coherent and important result and why it is true.
incoherent scattering. T11.4 We mention ‘density functional calculations’ often. Us-
Q11.3 What do we really mean by the term ‘diffuse scattering’? ing the literature, elaborate on why we do so.
Q11.4 Explain what we mean by channeling in this chapter. T11.5 Some people say we can do confocal microscopy in
Q11.5 Explain concisely why channeling is important in STEM. Explain.
HAADF imaging. T11.6 What application in ‘light’ optics did the Ronchigram
Q11.6 Can we measure the probe’s intensity distribution inside derive from? Use the internet.
the crystal? Explain your answer. T11.7 Does anyone use a point-projection microscope these
Q11.7 Can we see defects in HAADF? Justify your answer. days? Explore your resources.
Q11.8 Imagine you are Mr. Tompkins in the TEM. What are T11.8 The Fourier transform of the image intensity can give
we talking about? misleading information. Discuss this statement.
Q11.9 The electron is quantum mechanical and finally hits the T11.9 Can you use ABF, ADF, and HAADF at the same time?
detector. Does it know it is there? Discuss.
Q11.10 Discuss the possible confusion over the phrase ‘objec- T11.10 Explain why we have to discuss contrast in a scanned
tive aperture’ in STEM. image as a separate topic.

Chapter 11
Q11.11 What is the relationship between Airy and Raleigh? T11.11 How do thermal vibrations affect ‘longitudinal’ and
Q11.12 How did Scherzer’s classical approach differ from to- ‘transverse’ coherence?
day’s Cs-correction approach? T11.12 Is dynamical scattering relevant in STEM? Explain your
Q11.13 We noted that ‘the channeling does not make the La answer.
atoms appear smaller, only brighter’. Explain this point. T11.13 What does the 2s state have to do with channeling?
Q11.14 What do we mean by the term ‘depth sectioning’? T11.14 Has channeling contrast been used to study defects in
Q11.15 The depth resolution for extended objects is much crystals? Discuss.
poorer than for a point object. Explain why this is so. T11.15 Frozen appears in this chapter more than once. Discuss
Q11.16 Explain why Z-contrast imaging can tolerate a much its appropriateness.
lower beam current than spectroscopic imaging.
Q11.17 STEM imaging is fundamentally limited by the bright-
ness of the source. Why is this so and what can we do
about it?
 338 11 Imaging in STEM

imaging in aberration corrected scanning transmission elec-

References tron microscopy. Phys Rev Lett 91(10):105503 (Examples of
full image simulations)
Part I

Batson PE (1993) Simultaneous STEM imaging and electron

energy-loss spectroscopy with atomic-column sensitivity. Na-
General References ture 366:727–728 (First atomic column EELS data)
van Benthem K, Lupini AR, Kim M, Baik HS, Doh S, Lee
Born M, Wolf E (1999) Principles of Optics. Cambridge Univer- JH, Oxley MP, Findlay SD, Allen LJ, Luck JT, Pennycook
sity Press, Cambridge SJ (2005) Three-dimensional imaging of individual haf-
Kirkland EJ (2010) Advanced Computing in Electron Micros- nium atoms inside a semiconductor device. Appl Phys Lett
copy, 2nd edn. Springer, New York, Dordrecht, Heidelberg, 87:034104(1–3)
London Borisevich AY, Lupini AR, Pennycook SJ (2006a) Depth sec-
Pennycook SJ (2011) Scanning Transmission Electron Micros- tioning with the aberration-corrected scanning transmission
copy. Springer, New York. (Much more detail) electron microscope. Proc Natl Acad Sci USA 103:3044
Borisevich AY, Lupini AR, Travaglini S, Pennycook SJ (2006b)
Depth Sectioning of Aligned Crystals with the Aberra-
tion-Corrected Scanning Transmission Electron Microscope.
Specific References J Electron Microsc 55:7–12
Borisevich A, Ovchinnikov OS, Chang HJ, Oxley MP, Yu P, Se-
Bak P (1986) Icosahedral crystals – where are the atoms. Phys idel J, Eliseev EA, Morozovska AN, Ramesh R, Pennycook
Rev Lett 56:861–864 (A key theme in this chapter) SJ, Kalinin SV (2010a) Mapping octahedral tilts and polar-
Beck V, Crewe AV (1975) High-resolution imaging properties of ization across a domain wall in BiFeO3 from Z-contrast scan-
STEM. Ultramicroscopy 1:137–144 (The Founding Editor, ning transmission electron microscopy image atomic column
Elmer Zeitler was also in Chicago) shape analysis. ACS Nano 4:6071–6079 (Influence of defects
Crewe AV, Wall J, Langmore J (1970) Visibility of single atoms. and surfaces on local polarization)
Science 168:1338–1340 (The annular detector) Borisevich AY, Chang HJ, Huijben M, Oxley MP, Okamoto S,
Feynman RP (1992) There’s plenty of room at the bottom. J Niranjan MK, Burton JD, Tsymbal EY, Chu YH, Yu P, Ra-
Microelectromechanical Systems 1:60–66 (Reprinted article mesh R, Kalinin SV, Pennycook SJ (2010b) Suppression of
of his 1959 talk which was almost ignored at the time. The octahedral tilts and associated changes in electronic proper-
paper/talk actually began to be noticed in the early 1990s) ties at epitaxial oxide heterostructure interfaces. Phys Rev
Howie A (1963) Inelastic scattering of electrons by crystals .1. Lett 105:087204 (Influence of defects and surfaces on local
Theory of small-angle inelastic scattering. Proc R Soc Lond polarization)
A 271:268–287 (Preservation of contrast under inelastic scat- Bosman M, Watanabe M, Alexander DTL, Keast VJ (2006)
tering) Mapping chemical and bonding information using multivar-
Ronchi V (1964) 40 years of history of grating interferometer. iate analysis of electron energy-loss spectrum images. Ul-
Appl Opt 3:437–451 (One of his 900 papers) tramicroscopy 106:1024–1036 (Using multivariate statistical
Rose H (1974) Phase-contrast in scanning-transmission elec- analysis)
tron-microscopy. Optik 39:416–436 (The ABF detector) Bosman M, Keast VJ, Garcia-Munoz JL, D’Alfonso AJ, Findlay
Warren BE (1990) X-Ray Diffraction. Dover, New York. (A SD, Allen LJ (2007) Two-Dimensional Mapping of Chemical
classic) Information at Atomic Resolution. Phys Rev Lett 99:086102
(2D maps. Using multivariate statistical analysis)
Browning ND, Chisholm MF, Pennycook SJ (1993) Atomic-res-
Background References olution chemical analysis using a scanning transmission elec-
tron microscope. Nature 366:143–146 (First atomic resolu-
Chapter 11

Abe E, Pennycook SJ (2005) Ultrahigh-resolution Scanning tion spectroscopy)

Transmission Electron Microscopy with Sub-Ångstrom-Sized Browning ND, Chisholm MF, Pennycook SJ (2006) Atomic-res-
Electron Beams. J Cryst Soc Jpn 47:26–31 olution checmical analysis using a scanning electron micro-
Abe E, Pennycook SJ, Tsai AP (2003) Direct observation of a scope. Nature 444:235 (Atomic resolution spectroscopy)
local thermal vibration anomaly in a quasicrystal. Nature Cowley JM (1975) Coherent and incoherent imaging in scan-
421:347–350 (Enhanced thermal vibration amplitude) ning-transmission electron-microscope. J Phys D 8:L77–L79
Abe E, Yan YF, Pennycook SJ (2004) Quasicrystals as cluster (The hole in the detector problem)
aggregates. Nature Mater 3:759–767 (A review paper) Cowley JM (1976) Scanning-transmission electron-microscopy
Ade G (1977) On the incoherent imaging in the scanning trans- of thin specimens. Ultramicroscopy 2:3–16 (The hole in the
mission electron microscope (STEM). Optik 49:113–116 detector problem)
(The hole in the detector problem) Cowley JM (1979) Adjustment of a STEM instrument by use of
Allen LJ, Findlay SD, Lupini AR, Oxley MP, Pennycook SJ shadow images. Ultramicroscopy 4:413–418 (Adjusting the
(2003) Atomic-resolution electron energy loss spectroscopy astigmatism)
 Appendix References 339

Findlay SD, Shibata N, Sawada H, Okunishi E, Kondo Y, Yama- crystal using STEM and EELS. Nature 450:702–704 (2D
moto T, Ikuhara Y (2009) Dynamics of annular bright field maps)
imaging in scanning transmission electron microscopy. Appl Kirkland EJ (2010) Advanced Computing in Electron Micros-

Part I
Phys Lett 95:191913 (Using different contrast transfer func- copy, 2nd edn. Springer, New York, Dordrecht, Heidelberg,
tions) London
Findlay SD, Shibata N, Sawada H, Okunishi E, Kondo Y, Iku- Krivanek OL, Nellist, Dellby N, Murfitt MF, Szilagyi Z (2003)
hara Y (2010) Dynamics of annular bright field imaging in Towards sub-0.5 Å electron beams. Ultramicroscopy 96:229
scanning transmission electron microscopy. Ultramicroscopy Krivanek OL, Dellby N, Keyse RJ, Murfitt MF, Own CS, Szi-
110:903–923 (Using different contrast transfer functions) lagyi ZS (2008) Advances in Aberration-Corrected STEM
Garcia-Barriocanal J, Rivera-Calzada A, Varela M, Sefrioui Z, and EELS. In: Hawkes PW (ed) Aberration-Corrected Elec-
Iborra E, Leon C, Pennycook SJ, Santamaria J (2008) Colos- tron Microscopy, vol 121. Elsevier, Amsterdam (On the probe
sal ionic conductivity at interfaces of epitaxial ZrO2:Y2O3/ size effect in equation 11.19)
SrTiO3 heterostructures. Science 321:676–680 (Perovskite Krivanek OL, Ursin JP, Bacon NJ, Corbin GJ, Dellby N, Hrncirik
O sublattices melting) P, Murfitt MF, Own CS, Szilagyi ZS (2009) High-energy-res-
Gazquez J, Luo W, Oxley MP, Prange M, Torija MA, Sharma M, olution monochromator for aberration-corrected scanning
Leighton C, Pantelides ST, Pennycook SJ, Varela M (2011) transmission electron microscopy/electron energy-loss spec-
Atomic-resolution imaging of spin-state superlattices in nano- troscopy. Philos Trans A Math Phys Eng Sci 367:3683–3697
pockets within cobaltite thin films. Nano Lett 11:973 (Map (Chromatic aberrations)
spin states) Krivanek OL, Chisholm MF, Nicolosi V, Pennycook TJ, Corbin
Haider M, Uhlemann S, Zach J (2000) Upper limits for the re- GJ, Dellby N, Murfitt MF, Own CS, Szilagyi ZS, Oxley MP,
sidual aberrations of a high-resolution aberration-corrected Pantelides ST, Pennycook SJ (2010a) Atom-by-atom struc-
STEM. Ultramicroscopy 81:163 tural and chemical analysis by annular dark-field electron
Hall CR, Hirsch PB (1965) Effect of Thermal Diffuse Scattering microscopy. Nature 464:571–574
on Propagation of High Energy Electrons Through Crystals. Krivanek OL, Dellby N, Murfitt MF, Chisholm MF, Pennycook
Proc Roy Soc A286:158 (The frozen lattice) TJ, Suenaga K, Nicolosi V (2010b) Gentle STEM: ADF im-
Hytch MJ, Stobbs WM (1994) Quantitative comparison of high aging and EELS at low primary energies. Ultramicroscopy
resolution TEM images with image simulations. Ultramicros- 110:935–945
copy 53:191 (The Stobbs factor) LeBeau JM, Stemmer S (2008) Experimental quantification of
Ishikawa R, Okunishi E, Sawada H, Kondo Y, Hosokawa F, Abe annular dark-field images in scanning transmission electron
E (2011) Direct imaging of hydrogen-atom columns in a crys- microscopy. Ultramicroscopy 108:1653–1658 (Gaussian
tal by annular bright-field electron microscopy. Nature Mater source size contribution)
10:278–281 LeBeau JM, Findlay SD, Allen LJ, Stemmer S (2008) Quan-
James EM, Browning ND (1999) Practical aspects of atomic titative Atomic Resolution Scanning Transmission Electron
resolution imaging and analysis in STEM. Ultramicroscopy Microscopy. Phys Rev Lett 100:206101 (Gaussian source size
78:125–139 contribution)
Jesson DE, Pennycook SJ (1993) Incoherent imaging of thin LeBeau JM, D’Alfonso AJ, Findlay SD, Stemmer S, Allen LJ
specimens using coherently scattered electrons. Proc Roy Soc (2009) Quantitative comparisons of contrast in experimental
Lond Ser A441:261–281 (Width of the detector function in and simulated bright-field scanning transmission electron mi-
real space) croscopy images. Phys Rev B 80:174106 (No Stobbs factor)
Jesson DE, Pennycook SJ (1995) Incoherent Imaging of crystals Lee J, Zhou W, Pennycook SJ, Idrobo JC, Pantelides ST (2013)
using thermally scattered electrons. Proc Roy Soc Lond Ser Direct visualization of reversible dynamics in a Si6 cluster em-
A449:273–295 (Independently vibrating packets) bedded in a graphene pore. Nat Commun 4:1650 (Graphene
Kadavanich AV, Kippeny TC, Erwin MM, Pennycook SJ, Rosen- nanopore)
thal SJ (2001) J Phys Chem B 105:361–369 Loane RF, Kirkland EJ, Silcox J (1988) Visibility of single heavy

Chapter 11
Kim YM, He J, Biegalski MD, Ambaye H, Lauter V, Christen atoms on thin crystalline silicon in simulated annular dark-
HM, Pantelides ST, Pennycook SJ, Kalinin SV, Borisevich field STEM images. Acta Cryst A44:912–927
AY (2012) Probing oxygen vacancy concentration and ho- Loane RF, Xu PR, Silcox J (1991) Thermal vibrations in con-
mogeneity in solid-oxide fuel-cell cathode materials on the vergent-beam electron diffraction. Acta Cryst A47:267–278
subunit-cell level. Nature Matt 11:888–894 (Vacancies in (The frozen phonon)
oxide thin films) Loane RF, Xu P, Silcox J (1992) Incoherent imaging of zone
Kim YM, Kumar A, Hatt A, Morozovska AN, Tselev A, Biegal- axis crystals with ADF STEM. Ultramicroscopy 40:121–138
ski MD, Ivanov I, Eliseev EA, Pennycook SJ, Rondinelli JM, (Longitudinal coherence)
Kalinin SV, Borisevich AY (2013) Interplay of octahedral tilts Lupini AR, Borisevich AY, Idrobo JC, Christen HM, Biegal-
and polar order in BiFeO3 films. Adv Mater 25:2497–2504 ski M, Pennycook SJ (2009) Characterizing the two- and
(Influence of defects, surfaces on local polarization) three-dimensional resolution of an improved aberration-cor-
Kimoto K, Asaka T, Nagai T, Saito M, Matsui Y, Ishizuka K rected STEM. Microsc Microanal 15:441–453 (Locating Bi
(2007) Element-selective imaging of atomic columns in a atoms in Si [100])
 340 11 Imaging in STEM

McBride JR, Kippeny TC, Pennycook SJ, Rosenthal SJ (2004) Nellist, Pennycook SJ (2000) The Principles and Interpretation
Aberration-Corrected Z-Contrast Scanning Transmission of Annular Dark-Field Z-Contrast Imaging. Adv Imag Elect
Electron Microscopy of CdSe Nanocrystals. Nano Lett Phys 113:147–203
Part I

4:1279–1283 Nellist PD, Xin Y, Pennycook SJ (1997) Direct structure determi-

McBride J, Treadway J, Feldman LC, Pennycook SJ, Rosenthal nation by atomic-resolution incoherent STEM imaging. Inst
SJ (2006) Structural basis for near unity quantum yield core/ Phys Conf Ser 153:109–112 (Pt atoms on γ-alumina)
shell nanostructures. Nano Lett 6:1496–1501 Nellist, Behan G, Kirkland AI, Hetherington CJD (2006) Con-
McGibbon AJ, Pennycook SJ, Jesson DE (1999) Crystal struc- focal operation of a transmission electron microscope with
ture retrieval by maximum entropy analysis of atomic reso- two aberration correctors. Appl Phys Lett 89:124105 (Two
lution incoherent images. J Microsc 195:44–57 aberration correctors)
McMullan D, Rodenburg JM, Murooka Y, McGibbon AJ (1990) Nellist, Cosgriff EC, Behan G, Kirkland AI (2008) Imaging
Parallel EELS CCD detector for a VG HB501 STEM. Inst modes for scanning confocal electron microscopy in a dou-
Phys Conf Ser 98:55–58 ble aberration-corrected transmission electron microscope.
Merli PG, Missiroli GF, Pozzi G (1976) On the statistical aspect Microsc Microanal 14:82 (Two aberration correctors)
of electron interference phenomena. Am J Phys 44:306–307 Oxley MP, Pennycook SJ (2008) Image simulation for electron
Molina SI, Varela M, Ben T, Sales DL, Pizarro J, Galindo PL, energy loss spectroscopy. Micron 39:676–684 (Examples of
Fuster D, Gonzalez Y, Gonzalez L, Pennycook SJ (2008) A EELS image simulations)
method to determine the strain and nucleation sites of stacked Oxley MP, Varela M, Pennycook TJ, van Benthem K, Findlay
nano-objects. J Nanosci Nanotechnol 8:3422–3426 (Cali- SD, D’Alfonso AJ, Allen LJ, Pennycook SJ (2007) Inter-
brated by high resolution X-ray diffraction) preting atomic-resolution spectroscopic images. Phys Rev
Molina SI, Sales DL, Galindo PL, Fuster D, Gonzalez Y, Alen B, B 76(064303):1–8 (Examples of EELS image simulations)
Gonzalez L, Varela M, Pennycook SJ (2009) Column-by-col- O’Keefe MA, Allard LF, Blom DA (2005) HRTEM imaging
umn compositional mapping by Z-contrast imaging. Ultrami- of atoms at sub-Angström resolution. J Electron Microsc
croscopy 109:172–176 (Calibrated by high resolution X-ray 54:169–180 (More on Sparrow and Rayleigh)
diffraction) Peng YP, Nellist, Pennycook SJ (2004) HAADF-STEM imaging
Mory C, Colliex C, Cowley JM (1987) Optimum defocus for with sub-angstrom probes: a full Bloch wave analysis. J Elec-
stem imaging and microanalysis. Ultramicroscopy 21:171– tron Microsc 53:257–266 (Probe propagation)
177 (Using the integrated current for resolution measurement) Peng Y, Oxley MP, Lupini AR, Chisholm MF, Pennycook SJ
Muller DA, Tzou Y, Raj R, Silcox J (1993) Mapping sp2 and sp3 (2008) Spatial resolution and information transfer in scan-
states of carbon at sub-nanometre spatial resolution. Nature ning transmission electron microscopy. Microsc Microanal
366:725–727 (First maps of bonding configurations) 14:36–47
Muller DA, Edwards B, Kirkland EJ, Silcox J (2001) Simulation Pennycook SJ (2002) Structure Determination through Z-Con-
of thermal diffuse scattering including a detailed phonon dis- trast Microscopy. Adv Imag Elect Phys 123:173–206
persion curve. Ultramicroscopy 86:371–380 Pennycook SJ, Boatner LA (1988) Chemically sensitive structure
Muller DA, Kourkoutis LF, Murfitt M, Song JH, Hwang HY, Sil- imaging with a scanning transmission electron microscope.
cox J, Dellby N, Krivanek OL (2008) Atomic-scale chemical Nature 336:565–567 (Incoherent characteristics)
imaging of composition and bonding by aberration-corrected Pennycook SJ, Jesson DE (1990) High-resolution incoherent
microscopy. Science 319:1073–1076 (2D maps. Using multi- imaging of crystals. Phys Rev Lett 64:938–941 (Incoherent
variate statistical analysis) characteristics; the probe propagating)
Nellist, Chisholm MF, Dellby N, Krivanek OL, Murfitt MF, Szi- Pennycook SJ, Jesson DE (1991) High-resolution Z-contrast
lagyi ZS, Lupini AR, Borisevich A, Sides WH, Pennycook imaging of crystals. Ultramicroscopy 37:14–38 (Incoherent
SJ (2004) Direct sub-angstrom imaging of a crystal lattice. characteristics; the probe propagating)
Science 305:1741–1741 Pennycook SJ, Jesson DE (1992) Atomic resolution Z-contrast
Nellist, Pennycook SJ (1996) Direct imaging of the atomic con- imaging of interfaces. Acta Metall Mater 40:S149–S159
Chapter 11

figuration of ultradispersed catalysts. Science 274:413–415 Pennycook SJ, Nellist PD (1999) Z-Contrast Scanning Transmis-
Nellist, Pennycook SJ (1998) Accurate structure determination sion Electron Microscopy. In: Rickerby DG, Valdré U, Valdré
from image reconstruction in ADF STEM. J Microsc-Oxford G (eds) Impact of Electron and Scanning Probe Microscopy
190:159–170 on Materials Research. Kluwer Academic Publishers, Dor-
Nellist, Pennycook SJ (1998) Subangstrom Resolution by Under- drecht, the Netherlands, pp 161–207
focussed Incoherent Transmission Electron Microscopy. Phys Pennycook SJ, Jesson DE, Chisholm MF, Ferridge AG, Seddon
Rev Lett 81:4156–4159 MJ (1992) Sub-Ångstrom Microscopy through Incoherent
Nellist, Pennycook SJ (1999) Subangstrom resolution imaging Imaging and Image Restoration. In: Hawkes PW (ed) Sig-
using annular dark-field STEM. Inst Phys Conf Ser 161:315– nal and Image Processing in Microscopy and Microanalysis,
318 Scanning Microscopy Supplement, vol 6. Scanning Micros-
Nellist, Pennycook SJ (1999) Incoherent imaging using dy- copy International, SEM Inc., AMF O’Hare, Chicago
namically scattered coherent electrons. Ultramicroscopy
78:111–124
 Appendix References 341

Pennycook SJ, Jesson DE, Chisholm MF, Browning ND (1993) Scherzer O (1949) The theoretical resolution limit of the electron
Atomic-Resolution Imaging and Analysis with the STEM. microscope. J Appl Phys 20:20–29
Inst Phys Conf Ser 130:217–224 Shibata N, Pennycook SJ, Gosnell TR, Painter GS, Shelton WA,

Part I
Pennycook SJ, Jesson DE, McGibbon AJ, Nellist (1996a) High Becher PF (2004) Observation of rare-earth segregation in
angle dark field STEM for advanced materials. J Electron silicon nitride ceramics at subnanometre dimensions. Nature
Microsc 45:36–43 428:730–733
Pennycook SJ, Browning ND, McGibbon MM, McGibbon AJ, Sohlberg K, Rashkeev S, Borisevich AY, Pennycook SJ, Pan-
Jesson DE, Chisholm MF (1996b) Direct Determination of telides ST (2004) Origin of anomalous Pt-Pt Distances in Pt/
Interface Structure and Bonding with the Scanning Transmis- Alumina catalytic system. Chemphyschem 5:1893–1897
sion Electron Microscope. Philos T Roy Soc A 354:2619– Sparrow CM (1916) On spectroscopic resolving power. Astro-
2634 phys J 44:76–86 (Two-point resolution)
Pennycook SJ, Lupini AR, Kadavanich A, McBride JR, Rosen- Spence JCH, Cowley JM (1978) Lattice imaging in STEM. Op-
thal SJ, Puetter RC, Yahil A, Krivanek OL, Dellby N, Nellist tik 50:129–142 (Diameter of the probe-forming aperture)
PDL, Duscher G, Wang LG, Pantelides ST (2003) Aberra- Steinhardt PJ, Jeong HC, Saitoh K, Tanaka M, Abe E, Tsai AP
tion-corrected scanning transmission electron microscopy: (1998) Experimental verification of the quasi-unit-cell model
The potential for nano- and interface science. Z Metalkd of quasicrystal structure. Nature 396:55–57
94:350–357 Thust A (2009) High-Resolution Transmission Electron Mi-
Pennycook SJ, Varela M, Lupini AR, Oxley MP, Chisholm MF croscopy on an Absolute Contrast Scale. Phys Rev Lett
(2009) Atomic-resolution spectroscopic imaging; past, pres- 102:220801(1–4) (Stobbs factor)
ent and future. J Electron Microsc 58:87–97 (Perovskite O Tonomura A, Endo J, Matsuda T, Kawasaki T, Ezawa H (1989)
sublattices) Demonstration of single-electron buildup of an interference
Pennycook TJ, Beck MJ, Varga K, Varela M, Pennycook SJ, pattern. Am J Phys 57:117–120 (The effect of increased de-
Pantelides ST (2010) Origin of colossal ionic conductivity in tection time at fixed beam current)
oxide multilayers: interface induced sublattice disorder. Phys Urban K (2008) Studying Atomic Structures by Aberration-Cor-
Rev Lett 104:115901(1–4) (O sublattice melting) rected Transmission Electron Microscopy. Science 321:506–
Pennycook TJ, Oxley MP, Garcia-Barriocanal J, Bruno FY, Leon 510
C, Santamaria J, Pantelides ST, Varela M, Pennycook SJ Varela M, Lupini AR, van Benthem K, Borisevich A, Chisholm
(2011) Seeing oxygen disorder in YSZ/SrTiO3 colossal ionic MF, Shibata N, Abe E, Pennycook SJ (2005) Materials char-
conductor heterostructures using EELS. Eur Phys J-Appl acterization in the aberration-corrected scanning transmission
Phys 54:33507–33517 electron microscope. Annu Rev Mater Res 35:539–569
Pennycook TJ, McBride JR, Rosenthal SJ, Pennycook SJ, Pan- Varela M, Oxley MP, Luo W, Tao J, Watanabe M, Lupini AR,
telides ST (2012) Dynamic Fluctuations in Ultrasmall Nano- Pantelides ST, Pennycook SJ (2009) Atomic-resolution
crystals Induce White Light Emission. Nano Lett 12:3038– imaging of oxidation states in manganites. Phys Rev B
3042 (Photon absorption) 79:085117(1–14) (Using multivariate statistical analysis)
Perovic DD, Rossouw CJ, Howie A (1993) Imaging inelastic Wang ZL (1998) An optical potential approach to incoherent
strains in high-angle annular dark field scanning transmission multiple thermal diffuse scattering in quantitative HRTEM.
electron microscopy. Ultramicroscopy 52:353–359 Ultramicroscopy 74:7–26 (Excitation and annihilation of
Prange MP, Oxley MP, Varela M, Pennycook SJ, Pantelides ST phonons)
(2012) Simulation of Spatially Resolved Electron Energy Wang ZL (2003) Thermal diffuse scattering in sub-angstrom
Loss Near-Edge Structure for Scanning Transmission Elec- quantitative electron microscopy-phenomenon, effects and
tron Microscopy. Phys Rev Lett 109:246101(1–5) (Solid state approaches. Micron 34:141–155 (Excitation and annihilation
bonding for EELS) of phonons)
Puetter RC, Gosnell TR, Yahil A (2005) Digital Image Recon- Wang SW, Borisevich AY, Rashkeev SN, Glazoff MV, Sohlberg
struction: Deblurring and Denoising. Annu Rev Astron Astr K, Pennycook SJ, Pantelides ST (2004) Dopants adsorbed

Chapter 11
43:139–194 as single atoms prevent degradation of catalysts. Nat Mater
Rafferty B, Nellist PD, Pennycook SJ (2001) On the Origin of 3:143–146 (On an alumina flake)
Transverse Incoherence in Z-Contrast STEM. J Electron Mi- Warren BE (1990) X-Ray Diffraction. Dover, New York
crosc 50:227–233 Witte C, Findlay SD, Oxley MP, Rehr JJ, Allen LJ (2009) The-
Rayleigh L (1896) On the theory of optical images with special ory of dynamical scattering in near-edge electron energy
reference to the microscope. Philos Mag 5(42):167–195 (His loss spectroscopy. Phys Rev B 80:184108-1-15 (Solid state
original paper) bonding)
Reimer L (1985) Scanning Electron Microscopy. Springer, Ber- Xin Y, Pennycook SJ, Browning ND, Nellist, Sivananthan S,
lin (For example, the edge resolution test) Omnes F, Beaumont B, Faurie JP, Gibart P (1998) Direct ob-
Roberts KG, Varela M, Rashkeev S, Pantelides ST, Pennycook servation of the core structures of threading dislocations in
SJ, Krishnan KM (2008) Defect-mediated ferromagnetism GaN. Appl Phys Lett 72:2680–2682 (Dislocations emerging
in insulating Co-doped anatase TiO2 thin films. Phys Rev B at the surface of GaN)
78:014409(1–6)
 342 11 Imaging in STEM

Xin Y, James EM, Arslan I, Sivananthan S, Browning ND, Pen- Yu ZH, Muller DA, Silcox J (2004) Study of strain fields at
nycook SJ, Omnes F, Beaumont B, Faurie JP, Gibart P (2000) a-Si/c-Si interface. J Appl Phys 95:3362–3371 (Accurate
Direct experimental observation of the local electronic struc- simulations)
Part I

ture at threading dislocations in metalorganic vapor phase Zhou W, Kapetanakis M, Prange M, Pantelides S, Pennycook
epitaxy grown wurtzite GaN thin films. Appl Phys Lett S, Idrobo JC (2012a) Direct Determination of the Chemical
76:466–468 (Dislocation cores studied by EELS) Bonding of Individual Impurities in Graphene. Phys Rev Lett
Yan YF, Pennycook SJ (2000) Alloys: Atomic structure of the 109:206803(1–5)
quasicrystal Al72 Ni20 Co8. Nature 403:266–267 Zhou W, Oxley MP, Lupini AR, Krivanek OL, Pennycook SJ,
Yan YF, Pennycook SJ (2001) Chemical Ordering in Al72 Ni20 Idrobo JC (2012b) Single atom microscopy. Microsc Micro-
Co8 Decagonal Quasicrystals. Phys Rev Lett 86:1542–1545) anal 18:1342–1354
Yan YF, Pennycook SJ, Terauchi M, Tanaka M (1999) Atomic
Structures of Oxygen-associated Defects in Sintered Alumi-
num Nitride Ceramics. Microsc Microanal 5:352–357
Chapter 11
Electron Tomography
12

Part I
Chapter Preview

It is the nature of ‘transmission’ that we can never overlook men from a series of 2D micrographs, usually acquired at a
the finite thickness of our specimens. We will always lose range of tilts. Electron tomography was briefly mentioned
information about the specimen in the beam direction; this in Sect. 1.3.B and 29.1 of W&C and specifically in relation
can be both concealing and misleading. However, in some to energy loss tomography in Sect. 40.9; this section seeks
cases in order to successfully characterize the structure of to expand on this brief introduction. While electron tomog-
nanoscale materials this information is invaluable. So how raphy was developed primarily as a tool for the examination
do we return the missing information, and resolve three-di- of macromolecular assemblies and cellular structure, in this
mensional (3D) information from what is essentially a two chapter we will explore how this technique can be applied
dimensional (2D) technique? The problem of determining in the physical sciences. This chapter will address some of
higher dimensionality information from lower dimension- the most important considerations when applying this tech-
ality data arises in many technical disciplines; from radio nique for materials science. We will cover in some detail
astronomy to medical imaging to geophysics. Tomography, the choice of imaging technique, introduce the experimental
from the Greek ‘tomos’ – to slice, describes a broad class of limitations of the microscopy, cover the processing of the
techniques to solve this problem. Electron tomography, spe- tomography data from tilt series to volume, and discuss the
cifically, can be used to restore the 3D structure of a speci- care that you must take in interpreting the results.

Chapter 12

© Springer International Publishing Switzerland 2016

C. B. Carter, D. B. Williams (Eds.), Transmission Electron Microscopy, DOI 10.1007/978-3-319-26651-0_12 343

Licensed to ([email protected])
 344 12 Electron Tomography

12.1 Theory of Projection As we live in an imperfect, experimental world any sampling of

Radon space is actually discrete and the numbers of projections
acquired always limited by practical considerations. Therefore
Part I

As in many other areas of microscopy research, the theory of to- any reverse transform will be imperfect; as with any finite ap-
mography outpaced its experimental implementation. The theory proximation to an infinite function.
of projection was first formulated by the Austrian mathematician
Johan Radon in 1917 (Radon 1917). However, the first demon- The Challenge
stration of a practical application of this theory didn’t occur un-
til 1957 when Ronald Bracewell used Radon’s theory in order
to reconstruct 2D radio astronomy data from 1D line integrals Obtaining the most accurate reconstruction from discrete
(Bracewell 1956). experimental data is the central challenge of tomographic
reconstruction. 9
Radon’s theory describes for us how an object D in real space
f(x,y) can be described by a line integral through all possible
lines L through that object: So how does this imperfection in sampling manifest in real sys-
Z tems? We can elegantly describe the effects of limited sampling
Rf = f .x; y/ds by using the ‘projection slice theorem’; a projection of an object
(12.1) at an angle is mathematically identical to a section through the
L
Fourier transform of that object at the same angle. This also
where ds is the unit length of L, see Fig. 12.1. The projection puts a different spin on the description of how tomographic re-
operator R, the Radon transform, is best explored by describing construction is possible from a tilt series of projections. Each
its geometry more explicitly. We can simplify this by defining projection at each angle gives us information about the object in
a new set of coordinates that are perpendicular (l) and parallel Fourier space in a particular direction. By taking projections at
(z) to the projection/transform direction defined by the line L. many angles we can describe the whole object in Fourier space,
We define the angle θ as that from the horizontal to the line and also in real space by a simple inverse Fourier transform.
normal of L, usually termed the ‘projection angle’. The ge-
ometry of these terms is illustrated in Fig. 12.2a. We can then This theorem instantly suggests one potential method of tomo-
directly relate the intensity in real and radon space by these graphic reconstruction: by simply summing the tableau of pro-
two equations; jections in Fourier space (at the correct angles) and applying an
inverse Fourier transform. While this ‘direct’ Fourier approach
Z1 p can be used for tomographic reconstruction it has largely been
Rf .l;  / = f . l2 + z2 + tan−1 .zl//dz if l ¤ 0 (12.2) superseded, due to ease of implementation, by real-space tech-
−1 niques based on backprojection, which we will discuss at length
later in the chapter.
Z1 
n
Rf .0;  / = f z;  + if l = 0
2 (12.3) Fourier and Radon space
−1

We use the Radon transform R to convert an object in Cartesian The relationship between Fourier and Radon space is
co-ordinates (x,y) into an object in Radon co-ordinates; where invaluable for both conceptualizing reconstruction and
the object is described in terms of the angle θ of the normal of describing the limitations incurred during experimental
the projection angle to the x-axis and l the line perpendicular to tomography. 9
the projection direction. When we carry out the Radon transform
a point in real space (x,y) becomes a line in Radon space (l,θ)
linked through the equation l = rcos(θ–φ), such as the one marked We have already discussed how resolution in two dimensions can
in Fig. 12.2b. A model example of this transformation is shown be expressed as the highest spatial frequency with statistically
for a 2D test object in Fig. 12.3. In the Radon-transformed ob- significant sampling transfer in Fourier space; the same in true in
ject the bright spots have been converted into sine curves; for three dimensions. This statement also describes one of the most
this reason the Radon space image is often termed a ‘sinogram’. important limits to tomography, the number of sample points in
Sometimes it is also known as a ‘tomogram’. three dimensions is the square of those in two dimensions; log-
ically therefore it is much harder for us to achieve a particular
A complete sampling of Radon space is achieved by recording resolution in 3D than in 2D. While imaging resolution in 2D TEM
projections over the continuum of angles. With a sufficient num- has passed the 1 Å barrier in a single image, the highest homo-
ber of projections spread across the entire range of θ, we can geneous resolution achieved in 3D TEM reconstruction is 3.3 Å.
Chapter 12

achieve a reconstruction of an object can by simply applying the However, achieving this resolution in tomography requires ap-
inverse Radon transform. proximately 20,000 images of a single repeating structure (Zhang
et al. 2010) (for the ‘single particle reconstruction’ method).
 12.1 Theory of Projection 345

Fig. 12.1 The Radon trans-

form defines the projection
of an object D in Cartesian

Part I
space (x,y) through an in-
finite number of lines (L)

Fig. 12.3 Demonstration of the Radon transform, over a full range of tilt ± 180°,
applied to a test object. A point object is Cartesian space (a), circled, becomes a
sinusoidal line in Radon space (b), indicated by the dashed lines. The amplitude
of the line in l is a function of its average distance from the center of the volume

Fig. 12.2 The Radon transform converts objects in Cartesian space (a) to Radon
space, (b) measured in terms of the projection angle (θ) and the distance from
the origin (l). A point in real space (x,y) is converted into a line with the equation
l = rcos(θ-φ) Fig. 12.4 The effect of the tomographic point spread function has both positive
and negative components. Here the PSF for ± 60° has been convolved with a
small Gaussian, shown in the inset, to illustrate its effect. The negative com-
We need a certain sampling, number of projections, in 3D fre-
ponent is much weaker, and the above plot has been boosted by 4× to allow
quency space to reach a particular resolution. By assuming a
comparison with the positive component. a The limited tilt range serves to elon-
perfect imaging system, with a complete absence of noise, it is
gate the point in the z-direction and reduces its intensity in the perpendicular
possible to derive a simple equation relating resolution (d) to direction, with two negative peaks on either side of the object. b, c The PSF also
sampling. shows weaker +ve/-ve “fan” artefacts, at the angle of maximum tilt, that are
D characteristic of tomographic reconstruction
d= (12.4)
N
An example of point spread functions is shown in Fig. 12.4. This
where N is the number of projections (N) and (D) is the diameter figure shows the distinctive ‘fan’ artefacts present in all tomo-
of the volume to be reconstructed. This simple relationship is graphic reconstructions from limited tilts. Probably the most sig-
known as the Crowther criterion (Crowther et al. 1970), and is nificant feature of this PSF is the vertical elongation, a reflection
regularly used to give a rough estimate of the resolution which of the poor sampling in the z direction. An estimation of the
can be attained from a particular tilt series. degree of this elongation factor (e), as a factor of our maximum
tilt angle (α), has been made by Radermacher (1988).
In deriving the Crowther criterion we make the implicit assump- r
tion that the projections are evenly spaced throughout Fourier e=
˛ + sin ˛ cos ˛
space. However, in electron tomography this is almost exclu- ˛ − sin ˛ cos ˛  (12.6)
sively not the case. The need to clamp and support the specimen
almost always introduces a range of angles which cannot be For a maximum tilt angle of 70º, not uncommon in electron to-
sampled, and in practice tilt series rarely extend beyond ± 75°. mography, our reconstructions are subject to an elongation factor
This results in a ‘missing wedge’ of information in Fourier space, of ~ 30 %. It is worth stressing that e is a convolution, rather than
reducing the sampling and hence resolution in the depth (z) di- a stretch, so accurate geometry cannot be returned simply by
rection. A more quantitative description of the effects of both resampling the volume.
limited tilt and sampling can be made by calculating the point-
spread- unction (PSF) in the reconstruction plane (y,z) for a given While the Crowther criterion is based on an equi-axed recon-
maximum tilt angle (θmax); struction volume, TEM samples are more often extended thin
Chapter 12

slabs with far larger dimension in plane (x,y) than in depth (z).
1
h.y; z/ = (12.5) A valid hypothesis in this situation is that while we have less z
 2 .z2 tan max / − y2 =.tan max / direction information in a tilt limited series, for such a reduced
 346 12 Electron Tomography

volume we do not need as much information! The effect of this In practice we carry out the backprojection operator by using a
can be estimated by the use of a modified diameter (D), which real space discrete estimation of the infinite integrais 12.2 and
is a factor of the thickness of the slab (T) and the maximum tilt 12.3 to estimate the inverse Radon transform. There are a num-
Part I

angle (α) ber of strategies that have evolved for this estimation, and they
include brute force rotate and sum, reconstruction by convolu-
D = T cos ˛ (12.7) tions, orthogonal functions, and using Riemann sums. These real
space techniques have largely superseded the Fourier approach
Therefore, fewer projections are required to attain the same over- for a number of reasons, including mathematical stability, ease
all resolution, leading to a less pessimistic resolution prediction of implementation, insensitivity to noise and, perhaps most im-
than the basic Crowther equation. portantly, they are more suited to the regular grid applied by
digital processing.

Whilst backprojection avoids using the projection slice theo-

12.2 Back-Projection rem for reconstruction, its consequences are used to improve
the quality of the reconstruction. Even if each individual image
has uniform sampling of spatial frequencies, each projection is
Whereas the projection slice theorem allows us to explore the a central section through 3D Fourier space, and a tilt series will
possibilities of reconstruction in Fourier space, the theory of sample that space unevenly. We will have greater sampling close
backprojection allows us to do the same for reconstruction in to the center of Fourier space and less sampling at the extremi-
real space. If a projection is the act of summing an object along ties; low frequencies will be over-sampled and high-frequencies
a given line L, a backprojection is what will arise if we sum that under-sampled. If uncorrected this will give us a blurred recon-
projection back through a volume at the projection angle. This struction. In order to restore the correct object the frequencies
leads to a set of rays, of varying intensity, crossing the volume of the reconstruction will need to be weighted to take account
in a single direction. If we then sum two such backprojections of this limitation. As the exact geometry of the sampling is
from different directions, an increase in total intensity will be known, it is simple to generate a frequency space filter which is
observed where the two sets of rays cross, which will coincide weighted to correct the average radial sampling, also known as
at a possible location of the original object. Combine enough of an r-weighting filter. This can be applied in Fourier space to the
these backprojections and the density distribution of the original original projections, or to the final reconstruction with equivalent
object should be restored. This process is illustrated for a test ob- effect. This frequency restoration is illustrated for a test object in
ject in Fig. 12.5. What becomes quickly apparent is that the more Fig. 12.6. Weighted backprojection is the standard reconstruction
complex the object, the more projections we require to achieve technique used for electron tomography; it is fast, robust, easy
an adequate reconstruction through backprojection; a practical to implement and retains the relationship between sampling and
illustration of the Crowther criterion. resolution. However, by setting the zero frequency data to zero
(using a typical ramp filter) we lose information about the overall
‘black level’ of the image.
Chapter 12

Fig. 12.5 Backprojection acts to sum each projection as a single ray through the reconstruction. A single projection becomes a ray, as above, and where rays cross
the increase in intensity suggests an object. But as the object becomes more complex, as here from 1 to 6 Gaussians, more projections are required before any
certainty about the object can be reached. With just 2 projections the single Gaussian is well described however, for the more complex object it takes 16 projections
to reach a similar certainty
 12.3 Constrained Reconstruction 347

jection of a reconstruction typically shows large variations

in intensity when compared to the original projections. This
variation is characteristic of the reconstruction technique and

Part I
the deficiencies of the tilt series as a whole, in effect each pro-
jection is blurred by the imperfect contributions from other
projections. By constraining the reconstruction to match the
original data it should be possible to minimize these incon-
sistencies and improve the final reconstruction. This approach
Fig. 12.6 Weighted backprojection. Here a test object, left, has been sampled requires us to apply several iterations of constraint, due to the
by 141 projections with a 1° increment over ± 70°. This is fairly typical for real corrections themselves subject to the same imperfect process
tilt series. A raw backprojection, center, of this data leads to a hazy and indis- of reconstruction as the original images. The simplest way we
tinct reconstruction. However, when we r-weight the reconstruction, right, to can implement this constraint corrects the reconstruction to
take account of the uneven spatial sampling of the original data the returned
match each tilt in sequence, the algebraic reconstruction tech-
reconstruction is much improved. This example also demonstrates the effects
nique (ART) (Gordon et al. 1970). With model datasets ART
of the tomographic PSF; the reconstruction shows features elongated in Z and
can yield significant improvements in the reconstruction, but
distinctive “fan” artifacts
it proves unstable in the presence of small amounts of noise.
For real, noise containing, data a more robust approach is to
A Word of Warning compare all projections simultaneously; generating an average
difference for each iteration (Gilbert 1972). This simultaneous
iterative reconstruction technique (SIRT) is extremely stable,
Regarding constrained reconstructions – as soon as any and is particularly suited to tilt series with poor signal-to-noise-
of these techniques are applied the resultant reconstruc- ratios. A demonstration of the improvement in reconstruction
tion is almost always a non-unique solution. Many of quality between SIRT and weighted backprojection is shown
these techniques rely on a subjective choice regarding the in Fig. 12.7 for a test dataset.
nature of the constraints or somewhat ill-defined choices
regarding the reconstruction. 9
SIRT and ART

While SIRT and ART do not, theoretically, offer any

12.3 Constrained Reconstruction resolution improvements over weighted backprojection,
we can use them to generate reconstructions that show
significant reductions in reconstruction artifacts. Such a
In a real experimental situation the limit on projection numbers reduction will simplify any subsequent interpretation and
is well below that required to give us a reconstruction with the segmentation of the reconstruction volume; and in that
same resolution as the original images. When this is combined way may result in an improvement in the “useful” reso-
with the limited angular sampling of the TEM, with further com- lution. 9
plications arising due to the type of interpolation used in the
reconstruction, the use of a non-optimal weighting filter and by
deficiencies in the signal-to-noise (SNR) ratio of the original There are a large number of other variants on applying the pro-
data, a typical reconstruction will be far from ideal. In carry- jection constraint, which modify typically how the projection
ing out tomographic reconstruction we already make use of the information is utilized, but they typically take the form of a mod-
assumption that a discrete estimation is a valid approximation ified ART/SIRT approach. Variations include using multiplica-
to the infinite Radon transform. However, we can make use of tive/additive approaches to determining/applying differences, the
number of other valid assumptions, which can be applied as con- number of iteratons used and using a middle ground between
straints to improve the overall reconstruction quality. single/average comparisons. All these techniques are actually
specific examples of the ‘projection onto convex sets’ (POCS)
You should always view the final reconstruction from these tech- approach to solution.
niques with healthy skepticism! Consider any conclusions you
draw from them very carefully. It is worth explaining that while ART and SIRT were developed
in the 1970s there has been a resurgence of interest in these
approaches, especially in materials science; it appears that an
assumption of positivity, implicit in ART/SIRT, can offer us sig-
12.3.1 Constraint by Projection Consistency nificant gains in reconstruction fidelity when combined with high
SNR data available from imaging techniques such as HAADF
Chapter 12

STEM (Midgley and Weyland 2003). Another advantage of

We can assume, not unreasonably, that a tomographic recon- materials science datasets is that often the object to be recon-
struction generated from a series of projections should be structed is surrounded by low (or even zero) intensity regions;
entirely consistent with those projections. However, a re-pro- it is ‘isolated’ much like the test object in Figs. 12.3–12.7. This
 348 12 Electron Tomography

to the reconstruction, which should, in theory, provide a better

reconstruction with the same data. This approach is known as
discrete reconstruction. In practice, any reconstruction from a
Part I

finite tilt angle will always contain gray levels as a consequence

of the reconstruction PSF. The result of which is that forcing a
discrete reconstruction will cause most iterative algorithms to
blow up (mathematically speaking). A successful implemen-
tation of this technique, known as DART (discrete algebraic
reconstruction technique), gets around these problems by re-
stricting not only the gray levels but by only allowing operation
on voxels near boundaries (Batenburg et al. 2009). An example
of the improvement possible is shown in Fig. 12.8, showing the
potential of DART to return a clean, artefact free reconstruc-
tion. This has been successfully applied in a number of real
materials systems, however it remains difficult to implement
in general due to the need for prior selection of gray levels and
limitations due to sampling limitations in tilt-angle limited tilt
series. For example, in the Al/Sn system above there is also the
presence of surface oxides/contaminants that do not resolve
into easy discrete steps. Despite this, discrete reconstruction has
one extremely attractive aspect: it can combine reconstruction
and segmentation in a single step, as demonstrated in Fig. 12.8.
Fig. 12.7 Tomographic reconstructions from a test object, sampled by 71 pro- Segmentation, and the potential for discrete reconstruction, will
jections over 2° increments, from a tilt range of ± 70°. The weighted backprojec- be discussed later.
tion demonstrates classical “fan” artifacts, a symptom of the missing wedge, es-
pecially outside the reconstruction object. Iterative, constrained, reconstruction
techniques (SIRT/ART) can greatly reduce these effects. Unfortunately ART, which
offers arguably the best reconstruction, breaks down in the presence of noise, 12.3.3 Constraint by Symmetry
whereas SIRT proves more robust

isolation aids iterative reconstruction, in an isolated object all the By definition symmetry implies redundancy in reciprocal space,
intensity being redistributed in the iterative loop can be restored. reducing the sampling required to fully describe an object. As
This is because the tilt series contains all the important Fourier such symmetry is a powerful constraint that can be applied in
components for the object. By contrast in a continuous object, both the reconstruction and the experimental acquisition of tilt
such as a biological section, there are always unsolvable Fourier data. The first ET reconstruction used the helical symmetry of the
components. This also might contribute to the limited success of tail of a bacteriophage (De Rosier and Klug 1968) to reconstruct
iterative solution in bio ET. In algorithmic terms this condition from a single image. There are many examples of symmetry in
is known as ‘compact support’, and is well known to be critical the realm of biological macromolecules, such as the capsids of
for iterative solution in a variety of fields. viruses or the arrangement of protein chains in molecular ma-
chines. Such regular symmetry is unusual in materials science
at the nanoscale, but exists in abundance at the atomic scale
(but not conversely in the biological sciences where individual
12.3.2 Constraint by Discrete Methods proteins have very little useful symmetry). The use of such sym-
metry constraints for atomistic reconstruction will be discussed
in Sect. 12.10.
One property of materials specimens that we can readily take
advantage of by constraint is that we often know, a-priori,
which constituent phases are present in the reconstruction vol-
ume. As such the number of different gray levels within the 12.3.4 Metric-Based Constraint
reconstruction volume will also be strictly limited. For exam-
ple, we may be studying tin in aluminum; at room tempera-
ture these two elements have little or no solid solubility – as What is evolving very rapidly in the field of tomographic re-
such the volume will consist of only volumetric elements that construction in electron microscopy is the understanding that
contain Al, Sn, or open space. The resultant reconstructed vol- all constrained reconstruction algorithms are simply part of the
Chapter 12

ume (assuming some kind of projection related to density or much larger framework of the solution of linear equations with
atomic number) should contain only three discrete intensity finite data/samples. These issues are relevant across a broad
levels. This vastly simplifies the number of possible solutions range of digital imaging problems, most notably in image com-
 12.3 Constrained Reconstruction 349

Fig. 12.8 A comparison between

SIRT and the discrete algebraic recon-
struction technique (DART). Sections,

Part I
in x and z, through the HAADF STEM
tomographic reconstruction of a low
atomic number carbon nanotube
containing a higher atomic number
Cu/Cu2O catalyst. a–c are sections
through the reconstruction using SIRT,
whereas (d–f) are the same sections
from the DART reconstruction. Note
both the clarity of the reconstruction
and the uniform intensity distribu-
tions, essentially pre-segmenting the
volume

pression. As such, all constraint based techniques can be pa-

Sparsity
rameterized in terms of sparsity in some metric – whether that
metric be symmetry, gradient (defined by edges), or discrete
gray levels. The term is well understood in the image processing com-
munity but almost unknown in the EM world.  9
Some of the most recent applications of tomographic reconstruc-
tion make explicit use of this more general view, making use of
reconstruction algorithms by the approach of ‘compressed sens- Sparsity is a parameter that describes the amount of informa-
ing’ (Donoho 2006; Candes et al. 2006) (here compressed refers tion required for solution of some function in comparison to the
to ‘compression’ of image information). The research of Saghi amount of information required to solve a maximally complex
and co-workers (Saghi et al. 2011) laid down the framework for function in the same sample. In this way it defines how much
such a general approach, as well as demonstrating the use of better we can expect to reconstruct than we would expect based
sparsity in the gradient domain. Compressed sensing electron on traditional sampling theory (the Crowther criterion in other
tomography (CSET) has started to see wide application, as it is words).
robust and requires very little user input (in contrast to discrete
methods). An example of CSET is shown in Fig. 12.9. This tech-
Chapter 12

nique has also been applied successfully to achieve a general

approach to atomic resolution reconstruction; its application will
be discussed in Sect. 12.10.
 350 12 Electron Tomography

Fig. 12.9 A comparison between

(a–c) SIRT and (d–f) compressed
sensing tomography (CST) recon-
Part I

struction. The same slice is shown

for a range of numbers of pro-
jections; (a), (d) 27 projections,
(b), (e) 13 projections and (c), (f)
only 9 projections. This demon-
strates both the potential improve-
ment of CST over SIRT, but also
its robustness to small number of
projections

12.4 Other Reconstruction Another, more speculative, approach to tomographic recon-

struction is ‘big bang tomography’, which is an attempt to find
Approaches an approach to reconstruction of non-crystalline materials at
atomic resolution (Van Dyck and Chen 2012). This approach is
There are a number of emerging techniques for tomographic re- not based on tilting but on focal series of phase contrast TEM
construction that won’t be covered here in detail that also show images, and use of algorithms originally coded for determining
promise. These include equally sloped tomography (EST), where the relative distance of stars to determine the relative distance of
projection angles are chosen precisely to allow transformation atoms. Clearly, atoms are not independent scatterers in the same
between Radon and real space without the need for interpolation way as stars, but the promise of solution to a problem that can’t
(one of the largest sources for error in tomographic reconstruc- be practically solved in any other way is enticing.
tion) (Miao et al. 2005). This approach has been applied to only a
limited number of studies, but shows strong promise for achiev-
ing high reconstruction resolutions (Scott et al. 2012).
12.5 Meeting the Projection
One technique which initially showed promise for 3D recon-
struction was confocal scanning transmission electron micros- Requirement
copy (CSTEM). While this approach had been used to image
buried features in extremely thick specimens (Frigo et al. 2002) At the heart of tomography, electron or otherwise, is an assump-
the arrival of aberration-corrected instruments with large open- tion so simple that you could be forgiven for not even consider-
ing angles and consequent restrictions in depth of field raised ing it; does the imaging technique we are using generate images
the possibility of depth sectioning in the STEM at close to which are projections of an object’s structure? Essentially, you
atomic resolutions without the need for tilting (van Benthem should ask whether the image generation process is consistent
et al. 2005). Unfortunately, it has proven to be much more com- with the Radon transform; either directly or with some analyt-
plicated to set up this experiment for general use, and current ical relationship. This consideration, known as the ‘projection
approaches produce vertical stacks with resolutions comparable requirement’, suggests that the recorded signal be at least a
to very limited tilt series (on the order of ± 4°!) (Xin and Muller monotonic function of some aspect of the structure. For most
2010). While in the future this will be a powerful technique optical and X-ray tomography there is no doubt this is the case,
for imaging at high 2D resolution inside thick specimens, it is both absorption and attenuation are clearly consistent. The range
Chapter 12

doubtful whether it will be a generally applicable technique for of possible interactions between high-energy incident electrons
3D reconstruction. and the periodic/aperiodic atomic potentials in the TEM seems
 12.6 STEM Tomography 351

another case entirely. However, despite this, tomographic re- number of techniques available to the physical sciences. The
construction has been successfully demonstrated across a broad strongly scattering nature of our specimens, and their robust
range of biological and, in recent years, materials systems. nature when compared to biological thin sections, allows us to

Part I
employ a wide range of alternative imaging techniques, many of
Classically, electron tomography has been applied to resolve which, as we shall see, are arguably more suited to tomography
the structure of cellular material and biological macromole- than BF TEM.
cules. This has been done almost exclusively by bright field
TEM of stained/unstained biological thin sections. In this sit- Materials vs Biological sciences
uation we can rely on several assumptions about the nature of
the specimens and subsequent electron scattering that meet the
projection requirement. Biological specimens are typically non The practice of electron tomography in materials science
crystalline, at least on short (< 1 nm) length scales, and weakly is substantially different to that in biological science.
scattering. In this situation contrast arises from two contribu- Thanks to the stability of materials specimens we can
tions: amplitude contrast due to electrons scattered outside the use the instrument to attack a broader range of problems,
objective aperture and phase change induced by elastic scat- related not only to structure and morphology but also
tering. The former, mass-thickness, contrast is related to the to the chemical, magnetic and electrical properties. We
inverse log of the atomic number, and the thickness of material have a choice of a wide range of techniques to solve these
in projection; clearly meeting the projection requirement. When problems; the first question to answer in electron tomog-
the biological section is stained, usually with high atomic num- raphy of materials is which imaging technique to use?9
ber metals such as uranium and lead, this contrast dominates.
For stained material this limits us to resolution on the order of
5 nm, due to limits in the localization of the stain and the scat- However, BF tomography still has important applications in
tering events. For resolution better than 5 nm unstained material materials science; there are a significant number of materials
has to be used, which gives very little amplitude contrast, and systems which are both amorphous and of low enough density
we rely on a combination of phase change in the specimen and to be comparable in nature to biological specimens. The first
defocus to generate contrast. For a sufficiently thin, weakly materials systems to be studied by electron tomography were
scattering, specimen we can rely on the phase grating and col- complex 3D networks of polymers (Spontak et al. 1988); unsur-
umn approximations, which suggest that the exit wave electron prisingly as they share many physical and structural similarities
function is a linear multiplicative projection of the specimen with biological materials.
potentials, the details of which are covered by Hawkes (1992).
These approximations start to break down, though not cata- An example of the reconstruction from a block copolymer sys-
strophically, in biological specimens greater than 20 nm thick tem is shown in Fig. 12.10. Bright field electron tomography has
(at 200 kV). In addition the exit wave is then convoluted with been applied to crystalline materials, to study both semiconduc-
the contrast transfer function (CTF) of the post specimen lens, tor devices and nanoscale alloys, and it is interesting to note that
which can complicate the contrast considerably (even more so successful results have been achieved for these systems. How
with a tilted specimen). These complications can be corrected is this possible given the limitations discussed above? This is
by a range of new methods that have been developed to cal- still an area of ongoing study and no definitive answer has yet
culate an experimental CTF from tilt series (Fernandez et al. been given. The reconstructions in question seem to show some
2006) and correct for its effects, even on highly tilted specimens accuracy in localizing the position of interfaces and external
(Voortman et al. 2011) structure, but the intensities of the reconstruction are not mean-
ingful and may be misleading. In this case we can make use of
These approximations are all very well for biological material, them to study the general morphology of the system, but due to
but for the bulk of materials science specimens they clearly do the unreliability of the intensities, conclusions must be reached
not apply. Our specimens are typically strongly scattering, usu- with great care.
ally both moderate to high Z and density, crystalline, often with
multiple orientations and packed with defects. This clearly in-
validates the phase grating approximation but it will also lead to
significant Fresnel contrast at specimen edges and interfaces. Our 12.6 STEM Tomography
biggest problem is arguably diffraction contrast. Being related to
the orientation of the crystal with respect to the incident beam
it cannot easily be related to the amount of material in projec- Of the alternative imaging techniques for electron tomography
tion. The situation is complicated further by the contrast induced it is arguably HAADF STEM that generates contrast which is
by strain contrast around crystalline defects; but the situation most suited for reconstruction. Under optimal conditions the
is not as difficult for us as it may initially seem. While phase/ image generated will show contrast proportional to Zn, where n
Chapter 12

mass-thickness contrast BF imaging is the dominant technique approaches 2 for an ideal detector, and proportional to the pro-
for imaging in biological specimens, it is only one of a large jected thickness; known as Z-contrast. It is difficult to imagine a
 352 12 Electron Tomography
Part I

Fig. 12.10 Comparison between (a) an experimental reconstruction and (b) an ideal gyroid structure for an SIS tri-block co-polymer.

more suitable relationship for tomography. In addition, the con- decrease with increasing Z, the average scattering angle will be
trast we observe is almost completely incoherent in nature; the high enough that more electrons will begin to be scattered out-
annular detector serves to exclude all the lower-order diffraction side the detector than we will record. This will give us a contrast
information and averages over higher-order scattered beams. In inversion above a certain thickness, a loss of the monotonic na-
some conditions coherent effects can be significant, especially ture of the signal with thickness and which results in a deviation
when we orient the crystal in order to carry out atomic resolution from the projection requirement. While it is very difficult, if
HAADF STEM. But given the narrow cone of angles at which not impossible without a significant redesign of the microscope
we can carry out high resolution STEM the odds of it having a column, to build an ideal HAADF detector (which we could
significant effect on tomography are extremely low. The only imagine would wrap completely around the specimen!), it is
significant limitations to the technique, other than those imposed fairly easy to build a detector which would be the inverse of this
by automating acquisition, which we will discuss later, are the perfect detector. This detector is bright field, in that it collects the
need for some difference in Z to generate contrast and the abil- unscattered beam, but with a high enough outer angle to average
ity of the specimen to withstand the significant beam current in over Bragg beams and hence generate an incoherent image. This
our converged STEM probe. This technique, first demonstrated incoherent bright field (IBF) approach to STEM allows us to
in 2001 (Midgley et al. 2001, Weyland et al. 2001), has been study thick (over 1 µm) specimens of high Z material (Ercius
successfully applied across to nearly every class of material. et al. 2006). An example of this technique being applied to the
Figure 12.11 demonstrates the application of HAADF STEM study of thick sections of Ta and Cu, in a semiconductor device,
tomography to reveal the internal structure of a heterogeneous is shown in Fig. 12.12.
catalyst. This flexibility suggests that it could become the de-
facto technique for materials tomography, much like BF TEM is IBF
for biological tomography.

There is one further condition in which we find HAADF STEM Incoherent bright field (IBF) STEM is extremely useful
tomography limited, and that is in the study of extremely thick for specimens which are thick enough to be problematic
and/or high Z specimens. As a specimen increases in thickness for HAADF STEM. However, the contrast for HAADF
the average scattering angle also increases simply due to the image is significantly greater than for the comparable IBF
statistics of scattering. In addition due to mechanical limitations image, so it is still the preferable technique for thinner
specimen.
Chapter 12

imposed by the microscope design there is an upper limit to the 9

collection angle. As a result at a certain thickness, which will
 12.6 STEM Tomography 353

Fig. 12.11 STEM HAADF electron to-

mography by of a heterogeneous cat-
alyst; a nanostructured mesoporous

Part I
silica support with Pt/Ru bimetallic
nanoparticles. a, b Voxel (volume
pixel) projections, silica is gray and
nanoparticles are red, from two dif-
ference directions. This shows the
semi-regular hexagonal nature of the
pores, and that the particles are well
dispersed. c A surface render of single
pore containing two nanoparticles
extracted from the volume (Weyland
et al. 2006)

A further limitation occurs when applying STEM tomography

to thick specimens due to the convergent nature of the imaging.
The STEM probe is formed by focusing an image of the con-
denser aperture to a point on the specimen, forming a cone of
illumination with a characteristic convergence semi-angle (α).
Convergence angles are small, fractions of a degree, but they
lead to a finite depth of field (DOF). This loss of DOF is a con-
cern both for thick specimens and for aberration corrected STEM
where the convergence angle is increased to improve the lateral
resolution. The effects of the loss of DOF in STEM imaging Fig. 12.12 Incoherent bright field (IBF) for ET of high mass thickness speci-
have been explored by Hyun et al. (2008), and it is shown that mens. Here a double lined MOSFET transistor, the via width is ~ 150 nm, has
care must be taken to tune the convergence angle to the volume been reconstructed from both (a) HAADF and (b) IBF images. c, d The HAADF
being studied. This has been explicitly applied to thick biologi- reconstruction shows contrast reversal in the Ta layer, whilst the IBF reconstruc-
cal sections where small convergence (2 mrad) angle STEM has tion is more faithful. This deviation from linearity also affects the morphology of
the reconstruction, in particular the electromigration void at the via foot is well
been shown to allow successful reconstruction of specimens up
reconstructed by (f) IBF, but not by (e) HAADF
to 1-micron thick, where TEM fails to provide adequate recon-
struction (Aoyama et al. 2008).
Chapter 12

Licensed to ([email protected])
 354 12 Electron Tomography

12.7 Element-Selected Tomography

Part I

Bright field TEM, for the correct specimen, and HAADF/IBF

STEM make use of elastic/quasi-elastic scattering to provide
contrast related to the projected atomic number of the speci-
men. If we wish to image a specific element, or combination
of elements, we have a number of options which make use of
inelastic scattering either indirectly via energy dispersive X-ray
(EDX) analysis or directly by electron energy loss spectroscopy
(EELS). Previous chapters demonstrated that, given appropriate
consideration of the relative cross-sections of different elements
(and ZAF corrections for X-ray), the signal in an elemental map
can be related directly to the amount of elemental species being
excited by the beam. In theory we can think of an elemental map
as a projection of the amount of a particular species; suggesting
that tomographic reconstruction should be possible from such
a dataset. Fig. 12.13 XEDS tomography of a 28 nm high K transistor. The collection of
entire spectra at each pixel at each tilt allow reconstruction of volumes from any
While the nature of the signal generated by STEM EDX map- peak in the spectra. This allows separation of chemically distinct regions from
ping may be ideal for tomography, historically there are large the complex overlapping structure of this transistor. The reconstruction volume
problems with the practical application of the technique and is 150 nm across
early demonstrations of the technique served to highlight sig-
nificant challenges in application. The root of this problem is
poor collector efficiency, a consequence of the small collector mapping that can be problematic to reconstruction. Adapting
solid angle subtended by the EDX detector. This results in slow EFTEM for tomography is not without its challenges, how-
acquisitions, on the order of tens of minutes to hours for modest ever. For our purposes, inelastic scattering is largely incoher-
pixel size images, and perhaps more importantly a very poor ent and meets the projection requirement, but given the small
signal to dose efficiency. This makes it difficult to acquire a cross-section for core losses, relative to elastic scattering, it
single elemental map of moderate SNR without damaging the is difficult to avoid a combination of elastic/inelastic scatter-
specimen, let alone the many images we need in tomography. ing. As such energy-loss images show a mixture of contrast
A slow acquisition time will also lead to problems with sample related to the chemistry (electronic structure) of the specimen
drift and image distortion; reconstruction algorithms are not for- and the diffraction condition (bending and strain) of the crys-
giving of deviations from geometric accuracy. As a consequence tal, the latter being non-monotonic. This component can be
despite early interest in EDX tomography (Mobus et al. 2003), minimized by using jump-ratio imaging, where we generate a
there was very little progress beyond proof-of-principle. How- map by dividing the post-edge signal by the pre-edge signal.
ever, the recent introduction of large solid angle SDD detectors However, this puts significant restrictions on our specimen
(ca. 1 sr) has vastly improved the efficiency of X-ray collection. thickness, jump-ratio imaging becoming non-monotonic at
In conjunction with the availability of high brightness electron moderate thicknesses, as well as generating only qualitative
sources, this has transformed EDX tomography from a novel results. For thicker specimens, therefore, we have to use a
curiosity it to a practical technique. The first results of this ap- multiple-window method, where pre-edge images are used to
proach were published recently, demonstrating its applicability extrapolate the background under post-edge window/s. Un-
to a broad range of materials systems; shown in Fig. 12.13 is fortunately, this does not compensate for diffraction contrast
one such set of EDX tomography results from a semiconductor although dividing by a zero loss image may help. A further
device (Lepinay et al. 2013). limitation is achieving sufficient SNR in EFTEM elemental
maps, especially for high energy loss edges and/or low con-
At a first glance EELS seems a promising technique for centrations, to achieve a high fidelity tomographic reconstruc-
3D compositional mapping. Not only does it have an intrin- tion. While we can increase both the illumination intensity
sically high collection efficiency (orders of magnitude better and/or count for longer, both of these increase the chance for
than EDX), the acceptance aperture collecting almost all in- beam damage of our specimen. Despite these problems, we
elastically scattered electrons, those of a particular energy loss can use EFTEM tomography to generate accurate 3D maps
range can be reformed into a spatially resolved TEM image of multiple elements by careful selection of acquisition pa-
using an energy filter. The ability to acquire spatial informa- rameters. An example of such a multiple element analysis is
tion in parallel allows us to acquire an elemental map using shown in Fig. 12.14.
Chapter 12

energy filtered transmission electron microscopy (EFTEM)

in seconds or minutes. Using EFTEM also allows us to avoid When carrying out two-dimensional EFTEM our optimal route
the distortion and drift problems inherent to STEM elemental is not to acquire two or three images but a whole series of im-
 12.7 Element-Selected Tomography 355

ages, an ‘image spectrum’, over the edge/edges of interest. This

allows more accurate removal of the pre-edge background and
optimal choice of post edge integration window; it also allows

Part I
us to form an energy-loss spectrum at each pixel. If carried out at
each tomographic tilt increment it should be possible to generate
a four-dimensional volume (x, y, z, ∆E) and extract spectra for
each volume element. This has been demonstrated for low-loss
bulk plasmon imaging (Gass et al. 2006); Fig. 12.15 shows dif-
ferentiation of the different volume spectra from two allotropes
of carbon, in nylon and carbon nanotubes, which show distinc-
tive plasmon energies.

Mapping and Dose

Elemental mapping is typically considered a compara-

tively high-dose technique, especially compared to con-
ventional TEM, even for a single 2D image. With tomog-
raphy we need as many projections as possible, however
with sufficient care 3D elemental mapping should be
possible for most systems; it has even seen successful
application in biological specimens. 9

Spatially resolved energy loss information can also be acquired

as EELS spectrum images (with an EEL spectrum at each pixel
in a 2D map at each tilt), with the potential advantage of offering
post-facto analysis to optimize mapping and reveal unexpected
features. This approach has been compromised in the past by
slow spectrometer acquisition and poor signal-to-noise ratios,
leading to reconstruction-compromising drift. However, the
advent of ultrafast spectrometers and high brightness electron
sources now makes this approach more practical and this ap-
proach has been successfully demonstrated for core-loss elec-
trons by Jarausch et al. (2009). The results of this study were
not only 3D chemical maps from a range of elements, but also
extraction of 3D maps of the bonding state of silicon by match-
ing fine structure of the Si-K edge. These results are shown in
Fig. 12.16. Similarly, 3D maps from low loss electrons have
been reconstructed showing the localization of surface plasmon
resonances on metallic nanostructures (Nicoletti et al. 2013).

Fig. 12.14 a Zero loss bright field image and two-dimensional elemental maps
of nickel, iron, and oxygen recorded from a chain of FeNi nanoparticles chosen
for EFTEM tomography acquisition. The lower elemental maps are generated
by re-projecting the tomographic reconstructions in the zero tilt direction. The
upper elemental maps are the equivalent raw conventional elemental maps.
b RGB voxel (volume pixel) projection of the combined reconstructions of the
FeNi nanoparticles. c Three ‘core samples’ have been extracted from the central
volume of the three largest particles (marked 1, 2, and 3, respectively in (b)).
The ‘core samples’ have been averaged in the two shorter dimensions to give
average line traces of all three elements. Plotted for all three samples are the
Chapter 12

nickel/iron ratios (black) and the equivalent oxygen line traces (green). d A slice
through the oxygen tomographic reconstruction showing how the oxide covers
the ensemble of nanoparticles rather than individual nanoparticles
 356 12 Electron Tomography

Both examples illustrate an intrinsic advantage of EELS in that,

unlike EDX, the electronic and chemical nature of the specimen
can be derived from the data. Combining EELS and EDX in
Part I

a simultaneous tilt series acquisition provides perhaps the best

overall data set (Haberfehlner et al. 2014).

12.8 Dark-Field TEM Tomography

While diffraction contrast can complicate a tilt series, and is

best avoided when trying to reconstruct areas of different atomic
number and/or chemistry, it is possible to use diffraction for to-
mography and reconstruct features differentiated by changes in
crystal structure. This can be carried out in a number of modali-
ties; BF TEM, BF STEM, or low-angle ADF STEM.

When we choose a particular Bragg reflection from which to

form a dark field TEM image, we are essentially setting up im-
aging conditions to provide a projection of only those areas of
a specimen at a particular diffraction condition. With certain
kinds of features it is possible for us to maintain this diffraction
condition through a range of tilt by careful positioning of the
objective aperture, compensation with a secondary (β) tilt and/or
specimen rotation. This offers us the prospect of a very sensitive
method for differentiation between phases/structures in three di-
mensions. Its central limitation is in the difficulty of application;
we need a system where it is experimentally feasible to track
such a diffraction feature through tilt. For arbitrarily oriented
crystals in a similarly arbitrarily oriented, or amorphous, ma-
trix this would be all but impossible. But for structures with a
tightly defined relationship, say to a host crystal, it is possible,
if difficult, for us to maintain the required angular accuracy.
Two examples of these structures are dislocations, as studied
by Barnard et al. (2006) and shown in Fig. 12.17, and coherent
precipitates in Ni based super alloys, reported by Kimura et al.
(2005).

DF Tomography

The experimental difficulty of DF tomography is its cen-

tral limitation. However, it is potentially the only suit-
able technique for 3D imaging of structures for which
we cannot achieve contrast through any other technique;
Fig. 12.15 3D EFTEM tomography of a nylon–nanotube composite, this tech- where the only difference between an object and its sur-
nique allowed differentiation of elementally similar regions by the differenc- roundings are purely structural. 9
es in bonding. a Zero loss image; b 22 eV/28 eV plasmon loss ratio image;
c 28 eV/22 eV plasmon ratio loss image; d 3D surface rendered reconstruction
of the composite, showing differentiation of nylon from nanotube. e Low loss
One of the central difficulties of the DF TEM approach is that
volume spectra extracted from each distinct type of carbon
maintaining a particular Bragg condition through tilt is highly
challenging, and requires precise alignment during tilting and
when loading the specimen into a specimen holder. An alterna-
Chapter 12

tive approach, and more forgiving in terms of the exact Bragg

condition, is to acquire data in a STEM configuration (Tanaka
et al. 2008). Using a low-angle annular dark field (ADF) detector
 12.8 Dark-Field TEM Tomography 357

Part I
Fig. 12.16 EELS SI tomography of a W to Si semiconductor device contact. a Extracted spectra from different forms of Si are used to generate (b) multiple linear
least squares maps through tilt that can then be used to reconstruct (c) 3D reconstructions from the different forms of Si in the device

Fig. 12.17 A dark field tomographic

reconstruction of an array of dislo-
cations in a GaN epilayer. Threading
dislocations align to form low-an-
gle domain boundaries (DB) while
threading dislocations close to a
crack (C) turn over and glide away
leaving an in-plane segment that is
a pure screw. One of these has inter-
acted directly with another threading
dislocation (K) (Image courtesy of
Dr J. Barnard)
Chapter 12
 358 12 Electron Tomography

In the case of electrostatic potentials, in the absence of magnetic

fields, in the absence of diffraction contrast, and in the absence
of stray fields outside the specimen, the phase change is directly
Part I

proportional to a product of projected thickness, t, and the mean

inner potential of the specimen, V0
'.x; y/ = CE V0 .x; y/t.x; y/ (12.9)

It is therefore possible to use electron holography in a tomo-

graphic way determine the 3D distribution of electrostatic poten-
tials. This approach in demonstrated in Fig. 12.19 reconstructing
the potential distribution in a device containing a p-n junction in
3D, and reveals, quite dramatically, the influence on the device’s
electronic properties through surface modification by FIB-based
sample preparation methods.

This technique has been successfully applied to a number of

nanoscale device materials, and has shown to be particularly
Fig. 12.18 The implementation of dark field TEM tomography requires precise useful where the device is dimensionally confined, such as the
control of the diffraction geometry through tilt. Shown above is an alternative results on semiconductor nanowires by Wolf et al. (2013), where
approach using low-angle annular dark field STEM, where Bragg reflections 2D results would be especially compromised.
(here as an example a 220 reflection) are maintained on the ADF detector while
tilting the specimen
Reconstruction

allows dislocation contrast over a broader range of angles (due

to the convergent nature of STEM illumination). The geometry For vector quantities in three dimensions this requires
of this experimental setup is shown in Fig. 12.18. Another inno- more than a single tilt series. Why? One way of seeing
vation of the same group is the design of a specialized specimen this is that having vectors gives a greater degree of free-
holder with an extra degree of tilt freedom, this ‘HATA’ holder, dom for the value of each voxel; more possible informa-
allows both high tilts and precise adjustment of the diffraction tion means that more sampling is required to uniquely de-
condition. termine it. Sampling a scalar quantity from the opposite
direction (180° tilt) will give, all other things being equal,
the same result. The same is not true of vector quantities,
such as the magnetic induction, where sampling from two
12.9 Holographic Tomography different directions will give unique results. 9

Using the above techniques we can determine the structure and/ However, it is also desirable to recover the magnetic induction
or the chemistry of a specimen in three dimensions. However, B in addition to V0. Projected components of the B-field are rou-
we should also be able to take advantage of the phase change tinely determined by electron holography, which is covered in
inherent in elastic scattering to map magnetic and electrostatic Sect. 29.11 of W&C. Unfortunately because the B-field is a vec-
fields inherent to the specimen in 3D using off-axis electron ho- tor quantity, a single tilt series of holograms does not contain
lography. In general, we can describe the reconstructed phase enough information to determine all three vector components.
image as a projection of the 3D electrostatic or (component of
the) magnetic potential. In general, the reconstructed phase im- A number of theoretical approaches to solving this problem have
age ∆φ (x, y) is been published (Stolojan et al. 2001, Lade et al. 2005; Phatak
Z Z Z et al. 2008), but the difficulties of acquiring a suitable dataset
e
'.x; y/ = CE Vcryst .x; y; z/dz − B.x; y/dS (12.8) have prevented experimental implementation. The first success-
h ful experimental reconstruction (Phatak et al. 2010) showed the
magnetic induction and potential outside a magnetic permalloy
where the first integral is taken along a trajectory parallel to the plate. Figure 12.20 shows this reconstruction.
beam direction z, Vcryst is the crystal potential, B is the magnetic
flux (or induction), and S is normal to the area mapped out by
the trajectories of the electrons going from source to detector
Chapter 12

through the vacuum and the sample. CE is a wavelength-depen-

dent constant.
 12.10 Atomistic Tomography 359

Fig. 12.19 a, b 3D reconstruction of the electrostatic

potential near a p-n junction in a silicon device, recon-
structed from a series of off-axis holograms, for two

Part I
different thicknesses. c–d Plots through the 3D potential
shows an increasing depletion width near the surface
(gray and blue points) compared to the interior (black
points). This surface depletion is caused by damage of
the specimen surfaces due to the focused ion beam
preparation

12.10 Atomistic Tomography

Some time has already been spent describing the limitations in

reconstruction resolution attainable due to the paucity of sam-
pling in any practical ET dataset. On this basis alone, atomic
resolution ET would seem unattainable. However, we have also
seen that, given sufficient constraints (on any number of appli-
cable metrics), resolution is not limited by simple sampling con-
siderations.

On the atomistic scale it is possible to make a number of reason-

able assumptions that should simplify reconstruction. Atoms are
by definition discrete in nature, both in terms of their ‘physical’
location and their atomic mass (and associated cross-sections).
We are also often interested in atoms in crystalline (periodic) ar-
rangements. As such, crystals have well-defined symmetries, and
hence structural redundancies that can be unpicked by well-de-
signed sampling (Qin and Fraundorf 2003). (While it would be
attractive to image non-crystalline (amorphous) assemblies of
atoms, such as the atomic arrangement of protein chains in mac-
romolecular machines, this does not currently seem practical).
In addition free energy considerations vastly reduce the number
of stable configurations of surfaces and interfaces at the atomic
scale, as such continuously varying morphologies are rare. As
such these structures are sparse in both gradient and configura-

Fig. 12.20 Vector field electron tomography (VFET) of the (a) 3D magnetic in-
Chapter 12

duction and (b) the 3D vector potential outside of a thin permalloy island. This
reconstruction was carried out by combining four tilt series of in line holograms;
two in orthogonal directions followed by another two after an inversion of the
sample
 360 12 Electron Tomography

tional metrics and 3D atomistic studies can use different classes The majority of demonstrated atomistic reconstructions take
of constrained reconstruction. the form of determining the arrangement in 3D of all the atoms
in an isolated nanostructure. Here symmetry constraints and pe-
Part I

In addition, significant efforts have been made to quantify the riodicity simplify reconstruction, either explicitly or implicitly.
STEM image intensities. It is now possible to be quantitative The outer bound for the crystal (its surface) provides support
about the number of atoms in projected columns (LeBeau et al. for the solution of the positions of the atoms in the nanocrystal.
2008) and for mixed atomic columns the average chemical oc- What is more interesting, and much more relevant in materi-
cupancy of a given atomic column (Dwyer et al. 2011) from als science terms, is whether such techniques can restore the
2D projections (given some knowledge of the structure and ex- 3D arrangement of atoms in defective crystals or with com-
tensive knowledge about the microscope). plex interfaces, or in continuum specimens without 3D support,
where simple symmetry breaks down and it is harder to apply
These assumptions and imaging approaches have been com- quantitative imaging techniques. One of the first examples of
bined to allow atomistic reconstruction in three dimensions this at atomic resolution was by the Antwerp group (Goris et al.
from a number of materials systems. The first practical exam- 2012) in which the 3D strain in a nanowire was determined.
ple was applied to resolve the 3D atomic structure of a silver Recently, equally-sloped tomography was used to determine
nanoparticle in metallic Al (Van Aert et al. 2011), as shown the atomic structure around dislocations in a Pt nanoparticle
in Fig. 12.21. This approach essentially applies quantitative (Chen et al. 2013).
STEM to a series of images taken at high symmetry directions
(rather than a continuous series), providing 2D atom counts
for each image, these can then be numerically combined (by
discrete or CSET reconstruction) to reach an atomistic solution 12.11 Experimental Limitations
in 3D.

In our discussion of tomographic reconstruction we have already

Atomic arrangements in three dimensions are inherently introduced the theoretical consequences of a limited tilt range
sparse in many metrics, however the number of potential and numbers of projections. Therefore, it makes sense for us to
arrangements of atoms in even a small nanostructure is explore the practical reasons behind these limitations and inves-
very high. As such reconstructions at atomic resolution tigate approaches to minimizing their effects.
should be treated as a single plausible, non-unique solu-
tion from many possible solutions. While they promise In discussing the design of the objective pole piece we have
information about a specimen that cannot be determined already introduced the conflict between minimizing spherical
in any other way, the plausibility of that information aberration (small pole piece gap) and achieving specimen tilt
should be examined as rigorously as possible. 9 (large pole piece gap). With conventional pole piece and holder
designs we reach a maximum tilt angle when the holder contacts

Fig. 12.21 a Atomistic 3D reconstruction of a Ag GP

zone in an α-Al matrix (not shown). This reconstruction
was generated from quantitative imaging of two HAADF
Chapter 12

STEM projections in the [101] and [100] zone axes, the

error between the reconstruction and original projec-
tions is shown in (b) and (c)
 12.11 Experimental Limitations 361

Part I
Fig. 12.23 Tilt limits due to the geometry of the sample holder shadowing the
Fig. 12.22 Relative geometries of the objective pole piece and specimen holder
specimen at high tilt. a Conventional specimen holders are designed for imaging
at zero tilt and at the contact limited maximum tilt. This illustrates the impor-
at zero tilt. b When tilted the design of the holder and specimen clamping can
tance of the thickness (t) and width (w) of the specimen holder and the pole
severely limit the maximum unshadowed area
piece gap (g) in determining the maximum tilt angle (θ). The rotation axis is
marked in the center of the specimen (eucentric)

How High to Tilt?

one pole/aperture, as illustrated in Fig. 12.22. This conflict be-
comes even more prominent when we wish to tilt our specimen
over the large angular ranges required for tomography. In the For electron tomography we really require unrestricted
biosciences wide pole piece gaps are the norm as they provide tilting to as a high a tilt as possible. In practice ± 70°
both high contrast from weakly scattering specimens, due to the seems to be the de facto standard; however given the cosθ
large CS, and high tilt for tomography. relationship of the resolution every extra degree above
70° is valuable! 9
However, the opposite is the case for materials science where
the drive for higher and higher resolution TEM and STEM has
led to the prominence of narrow gap pole pieces. Given you In practice shadowing, blocking the incident or transmitted elec-
may wish to form a small STEM probe and have sufficient tilt tron beam, more often defines our maximum tilt than mechani-
for electron tomography at the same time, the only options are cal limitations. For conventional specimens there is always the
to redesign either the specimen holder or the specimen. The need to provide specimen clamping and support, and this will
second limiting factor for tilt range is the basic need to transmit often block itself in the beam at high tilts. Figure 12.23 describes
electrons through our specimen without the beam becoming how these limitations occur at a range of length scales. We can
blocked; by the holder, the specimen support or by the spec- minimize the effect of the sample holder by both reducing the
imen itself. thickness (t) and careful design of the shape of specimen holder
in the direction normal to the rotation axis. When carrying out
In theory, the arrival of aberration corrected electron microscopy specimen preparation we must also make sure that we make
and removed the need for a narrow pole piece gap to achieve the correct choice of sample support and/or orientation. If our
high resolution. However, very few instruments have been de- specimen is supported on a copper grid the finite thickness of
signed with much wider pole piece gaps specifically to take ac- the grid comes into effect, and as such we should choose low
count of this; instead improved tomographic sample holders are thickness, low background, grids with large grid squares. Typi-
now available to maximize tilt range. cally, 400 mesh (number of grid squares per inch) is sufficient,
however there are a number of manufacturers who provide grids
In examining Fig. 12.22 it becomes obvious that for conventional with elongated holes that when oriented perpendicular to the
specimen holders maximizing mechanical tilt is a matter of lim- tilt axis provide very little shadowing. The ultimate shadowing
iting both the width (w) and thickness (t) of the specimen holder limiter is the specimen itself, which we can do very little about
tip. For ‘analytical’ pole pieces, typically with gaps between 4 other than to be careful how an area is chosen for tomography.
to 6 mm in size, it is possible for us to engineer a holder which is
narrow, and thin, enough to rotate the holder completely through It is not possible for us to achieve 360° rotation with conven-
360° inside the gap. And indeed there are several commercial tional electron microscopy specimens; there is always some kind
specimen holders available that can achieve this. With gaps of of support and/or specimen in the plane normal to the tilt axis
less than 3 mm, in ‘high resolution’ pole pieces, we are limited which will block the beam with tilt. However, by preparing a
by the width of the specimen itself, and no holder design will specimen that is a narrow needle of material mounted in the
allow us complete freedom of tilt. It is possible to prepare a direction parallel to the tilt axis it is possible to avoid this lim-
smaller specimen, but this may be limited by sample preparation itation. In this situation, which is much like the way specimens
Chapter 12

considerations and typically this would also require the design are mounted for other tomographic techniques, we can in theory
of a bespoke specimen holder in which to mount your smaller achieve full coverage of frequency space and achieve significant
specimen. improvements in reconstruction fidelity. This approach sounds
 362 12 Electron Tomography

ideal but sample preparation is a challenge: the sample should gets successively smaller as we increase the tilt angle. We can
be undamaged and only a few hundred nanometers in diameter. calculate the next tilt angle θn+1 in the Saxton scheme by
The demonstrated examples of this approach have made use of
Part I

focused ion beam (FIB) milling to prepare the needle, in a sim- n+1 = n + cos.n /n (12.10)
ilar way to how ‘lift out’ specimen preparation is described in
Sect. 10.7 of the W&C. For metal specimens it is also possible where θn is the current tilt angle and θ0 is the initial tilt increment.
to prepare a needle using electropolishing, using an approach This results in a higher density of sampling at higher angles
pioneered for field ion microscopy (FIM) and 3D atom probe where more information is needed as well as to compensate for
(3DAP). These kinds of specimens clearly offer us the ultimate the ever increasing thickness of our specimens. For the same start
in tomographic fidelity, as demonstrated for an electropolished increment this will clearly result in more projections; however
needle of an Al-Li-Mg-Cu alloy in Fig. 12.24, with the cross-sec- for an identical number of projections a Saxton scheme will lead
tions in x–z demonstrating none of the missing wedge artefacts to an improved reconstruction. If you plan to use this scheme
present in all other sample geometries. with weighted backprojection you need to take account of the
irregular sampling when generating the weighting filter.
Another way we can improve our sampling of Fourier space
is to use an alternative sampling geometry; which can be done A significant limitation to single-axis tilting is the anisotropy
without the need for expensive specimen preparation equipment. between resolution in the direction parallel and perpendicular to
Perhaps the most simple is to switch from a regular tilt increment the tilt axis, this can lead to serious problems with specimens
to a Saxton scheme (Saxton et al. 1984), where the tilt increment oriented in certain directions. This is demonstrated quite clearly
Chapter 12

Fig. 12.24 HAADF STEM tomography of an electropolished needle specimen of an Al-Li-Mg-Cu alloy. The needle, top left, can be rotated fully 360° allowing acquisi-
tion without the missing wedge. The (SIRT) reconstruction shows homogenous resolution, lower right, with very few artefacts. This allows unambiguous identification
of features and greatly aids segmentation
 12.11 Experimental Limitations 363

Part I
Fig. 12.25 a is a reconstruction of CdTe tetrapods from a single tilt series and shows that some of the legs of the tetrapods are missing, as indicated by arrows, due
to the effects of the missing wedge; b shows a reconstruction of the perpendicular tilt series showing a different set of missing legs, again indicated by arrows; the
tilt axes in (a) and (b) are parallel to the direction of the arrows; c is a dual-axis reconstruction of the two data sets and illustrates that no legs are missing because
the missing information has been greatly reduced; the inset shows a view of the tetrapod boxed in (e), d is a dual-axis reconstruction of CdTe ‘mutipod’

in Fig. 12.25 for the tomographic reconstruction of CdSe/CdTe possibility of incorrect interpretation and conclusions. If, however,
Chapter 12

tetrapods; due to the anisotropy the legs perpendicular to the tilt we acquire two tilt series from the same area with perpendicular
axis have reconstructed with much lower fidelity than those par- tilt axes a combination of the reconstructions should show homo-
allel. For specimens with unknown morphologies there is a strong geneous signals in the x–y plane. In practice this is achieved by

Licensed to ([email protected])
 364 12 Electron Tomography

rotating the specimen by 90°, either ex-situ or in the instrument ery case beam sensitivity will be both different in mechanism and
using a tilt-rotate holder. The effect on the fidelity of the tetrapods degree. We can broadly divide specimens into the kind of damage
in using this approach is dramatic. Obviously, if you are dose lim- mechanism which dominates; knock-on damage or electron beam
Part I

ited, you need to spread this dose across these two series. There heating. Knock on damage occurs when a single scattering event
are also significant difficulties combining two separate reconstruc- generates enough energy to knock an atom from its position, break-
tion volumes into a single reconstruction (Mastronarde 1997). ing its bonds. This ionization event will only occur above a partic-
ular energy, and by operating the instrument below that threshold
energy, hence below a certain voltage, we can greatly reduce the
amount of knock-on damage. The classical example for this is for
12.12 Beam Damage crystalline silicon, where this threshold voltage is ~ 240 kV.
and Contamination However, lowering the accelerating voltage of your TEM also
has a significant impact on spatial resolution; the effects of spher-
Whilst the range of sampling is limited by both the specimen and ical and chromatic aberration become increasingly important at
mechanical design of the instrument, the density of that sampling low voltages. The arrival of aberration correction has opened up
is entirely a factor of the specimen. The ultimate limitation is the this realm of operation, with practical operation voltages drop-
dose which our sample can withstand before it is modified; by the ping to 80 kV and below (Sawada et al. 2010). This ‘drive to the
effects of beam damage and/or contamination. It should be fairly bottom’ is being driven by the interests in carbonaceous materi-
obvious to the reader that any significant modification during a als, such as nanotubes and graphene (Kaiser et al. 2011). It also
tilt series will invalidate our whole dataset. has the potential to enable electron tomography of soft materials
at higher resolutions. However, resolution at low voltages goes
Unsurprisingly for electron tomography of biological speci- hand in hand with a reduction of depth of field, which may crip-
mens, which are easily damaged under the beam, dose is the ple ET experiments. The potential and pitfalls of low kV ET has
central concern during acquisition. Indeed, it has been the avail- yet to be explored in depth.
ability of computerized, low-dose, acquisition systems that has
allowed any progress to be made in the field. In any specimen, The other significant damage mechanism is caused by electron
be it organic or inorganic, there is a theoretical upper limit to beam heating, occurring due to the collective excitation of a
the amount of dose which can be used, the ‘critical’ dose. It also specimen by many lower energy scattering events. If the speci-
stands to reason, therefore, that there is an upper limit to the at- men is insufficiently conducting, electron beam heating can lead
tainable resolution. Somewhat surprisingly this is not dependent indirectly to bond scission; this is the case for many polymers
on the number of projections that can be acquired but by on the where this damage mechanism dominates. As we are well aware
signal-to-noise ratio of the acquired images. This is known as the total cross-section for inelastic scattering decreases with in-
the dose-fractionation theory. First controversially espoused by creasing voltage; so in this case we should increase the operating
Hergl and Hoppe in 1976 (Hegerl and Hoppe 1976), this states voltage to minimize beam damage. Exactly the opposite case
that critical dose can be divided between as many projections than for knock on damage! The central point to take away is that
as practicable yet still result in the same SNR in the image/re- for every material we need to re-think our strategy for applying
construction. Our maximum number of projections is then de- tomography. A complete review of the different kinds of beam
termined by the need to align a tilt series to a common tilt axis damage mechanisms is a subject for a book in its own right;
(more on this later) or to have significant image contrast given however Sect. 4.6 of W&C and the review by Egerton (Egerton
an imperfect detector. The upshot of which is that given the com- et al. 2004) should be consulted.
mon nature of the bonding in nearly all biological material, and
hence similar limits of dose, the resolution of biological tomog- While the removal of material due to electron beam damage
raphy is limited, for a sufficiently thin specimen, to about 2–3 nm concerns us, so should the opposite; deposition of material un-
(Baumeister et al. 1999). Stained biological material, while more der the electron beam. Such contamination is usually caused by
beam stable, will actually give rise to lower resolution limits due the cracking of mobile hydrocarbons under the electron beam,
to delocalization of both scattering and stain position. the heat generated by beam/specimen interactions providing a
significant energy for diffusion. This will, at best, reduce feature
One of the most interesting aspects of applying electron tomog- contrast progressively through a tilt series or, at worst, introduce
raphy in materials science is the potential for matching, and ex- extra structure and invalidate the entire tilt series. The latter is
ceeding, the resolution demonstrated in the biological sciences. a particular problem with STEM where the strongly localized
Our specimens are, typically, many times more beam stable than probe intensity can cause carbon pillars to grow in a few sec-
those of our colleagues in the biological sciences. However, the onds! An example of this kind of contamination structure is
nature of beam damage in electron microscopy of materials is also given in Fig. 12.26. Cleanliness in specimen preparation can
a large unknown factor in applying tomography. Virtually all bio- reduce the amount of free hydrocarbons, but the most effective
Chapter 12

logical specimens share common bonding, chemistry and atomic approach is to plasma clean, in an Ar/O plasma, the specimen
arrangements. On the other hand, materials can contain many and holder prior to insertion into the instrument (Isabell et al.
combinations of different elements, bonding, and structure. In ev- 1999). Unfortunately the plasma will remove any carbon from
 12.13 Automated Acquisition 365

12.13 Automated Acquisition

Part I
One of the important concepts for the reader to grasp about to-
mographic acquisition is the importance of useful dose, accrued
in acquiring the tilt series data. This is compared with the tilt
‘overhead’, that dose expended when translating and focusing
the specimen before and during the tilt series but which does
not directly contribute to the final dataset. While we can ac-
quire tilt series manually, by tilting the specimen re-centering
and re-focusing our specimen at every tilt increment this is
not really very dose efficient. Achieving even modest results in
biological tomography of asymmetrical specimens has been de-
pendent on the use of computerized acquisition routines which
significantly reduce the overhead for acquiring a tilt series. As
we have just discussed, in the physical science we are lucky to
have comparatively beam stable specimens which allow us to
be somewhat less concerned about dose. But there are several
issues that make automated acquisition the preferable option
even in materials science. Probably the most important is the
that automation is simply more time efficient; a modern auto-
mated acquisition can acquire 140 projections in as little as half
an hour. A manual tilt series of the same size could take you
the best part of a day to acquire, a day consisting of repetitive
tilting tracking and focusing. These kinds of repetitive tasks can
make even the best scientist prone to fatigue and the resultant
experimental error.

Fig. 12.26 Effect of contamination in STEM tomography. a An electron beam The ability to drive the goniometer stage by computer, com-
was left in the middle of the field of view for a few seconds prior to the tilt series bined with digital image acquisition and analysis, was the key
acquisition, resulting in a bright ‘spot’ of carbon growth. b When tilted this is advance allowing the development of automated tomography.
revealed to be an elongated structure. c After reconstruction it becomes clear This allows the computerized system to mimic our manual ac-
that the structure is actually two ‘pinnacles’ one on either side of the specimen quisition techniques, with a fraction of the dose overhead. At
reflecting the original beam path each tilt increment a small number of images (1–3) are acquired
for which are then compared by cross-correlation to a reference
image from the previous increment to correct spatial displace-
our specimens, so it is not really suitable for specimens which ment between increments. For parallel beam techniques we can
are made of carbon or are supported on carbon films! In this case, implement auto-focusing by the use of the ‘shift-with-tilt’ meth-
or supplementary to plasma cleaning, we can carry an electron odology as devised by Koster et al. (1987). Assuming you have
beam ‘flood’. This is done by illuminating a wide area, such as an a well-aligned optical system, at focus a tilt of the beam induces
entire grid square, with a parallel electron beam for an extended no discernable image shift. As a consequence if we tilt the beam
period, typically 20–30 minutes. This causes thin homogeneous and do see a shift then the size of that shift is proportional to
cracking across the grid square sequestering any mobile hydro- the absolute defocus. This allows an accurate determination of
carbons, reducing the potential for further contamination. the focus with as few as two images, though more often three
are used for increased accuracy. For STEM techniques there is
Manual or Automatic? unfortunately no analogously simple technique for auto focusing.
Some form of maximum contrast analysis is typically applied
over a defocus series to determine focus. Under these conditions
For all tomography we need to be very careful about we have to make a careful selection of the range for the defocus
accurate determination of the eucentric height. Small series, large enough to be sure to include the optimal position
deviations from the correct position will result in large yet fine enough to provide an accurate measurement (typically
stage movements in the direction perpendicular to the tilt more than 5 but less than 12). A combination of these automated
axis as well as z, making tracking and focusing difficult. routines can reduce the tilt series acquisition time to a few hours
Also more modern automated acquisition routines make (Dierksen et al. 1992).
predictions of stage movement that are invalid for large
Chapter 12

errors in eucentric position. 9 A further revolution in automated tomography was brought

about by the realization that although our goniometers are far
 366 12 Electron Tomography

from perfect, they are at least reliably imperfect: in essence

Zeise and co-authors (Zeise et al. 2001) demonstrated it was
possible to calibrate how the goniometer moves, using a coarse
Part I

tilt series, and apply those calibrations in subsequent tilt series

to vastly reduce the tilt overhead. For tilt series at lower magni-
fications this calibration can be accurate enough to allow us to
acquire without further tracking and focusing; in this manner a
tilt series, of 141 images from ± 70°, can be acquired in as little
as 30 minutes. We should be mindful however that pre-calibra-
tion is sensitive to the starting position for stage calibration.
The stage calibration will vary dependent on the accuracy of the
eucentric height setting but, perhaps more worryingly, also with
the starting x/y stage position. We will always need therefore
further corrections during the acquisition for specimen areas
well away from the calibration position. A different approach
to automated acquisition is not to pre-calibrate but to use the
initial stage movement in a tilt series, determined by tracking
and focusing, to predict future stage movement. This predictive
methodology has the advantage that it is insensitive to starting Fig. 12.27 Schematic of the alignment of the tilt axis for tomographic tilt series.
position, as it makes no prior assumptions, and in theory in a The actual tilt axis usually away from the volume of interest. During acquisi-
requires us to carry out only a few tilt increments of tracking/ tion, and subsequent alignment, this axis is reset (b) to rest inside the volume by
focusing. However, stage calibration curves are rarely smooth the successive repositioning of the object/projections. The tilt axis, a line in the
and often show rapid changes in slope. An intelligent predic- projection of the volume, is identified and the projections rotated (c) to bring the
tive methodology, as suggested by Mastronarde (2005), moni- axis direction perpendicular to a major axis. d The spatial alignment can then
tors deviations from the predicted behavior and is able to alter effectively reposition the axis origin, marked o, to any point inside the recon-
prediction on-the-fly and even revert to traditional track and struction volume as long as the tilt axis remains perpendicular to this main axis
focus methodology if required. Whichever method is applied,
pre-calibration or predictive, the tilt overhead can be reduced
dramatically. This clearly has important consequences; more Interestingly, there is no one axis unique to our dataset; there
dose being available for acquiring images leads us, following are multiple solutions and any object that is present in all our
the dose fractionation theory, to improved resolution recon- tilt images can become the center of rotation. The only thing set
struction. about the tilt axis is its direction and determining this direction is
a key part of the alignment process. A summary of this process is
shown in Fig. 12.27. We can divide approaches to alignment into
three classes; those based on least squares tracking of fiducial
12.14 Tilt-Series Alignment markers, those which determine image shift between successive
images through cross-correlation and a new third class based on
optimization of reconstructions. These techniques are not mutu-
During acquisition of tilt series we have already made innumer- ally exclusive and using some combination is typical.
able corrections of the specimen position in order to retain the
area of interest in the field of view and in focus. However, before We implement marker tracking by relying on a dispersion of
we can carry out a reconstruction we have to go one stage further fiducial markers which can be identified in every single tilt im-
and ensure all our images are rotating about a common tilt axis. age. These markers are usually small, unreactive, high-density
What does this mean in practice? If observed as a movie the spherical particles, typically gold or silver in a colloidal solu-
corrected tilt series will appear to be a smooth rotation of the tion to prevent clustering. These markers provide us with a fixed
object about a single axis. reference for the 3D geometry of our reconstruction volume;
there being only one correct solution for their 3D positions which
Alignment Is critical will satisfy all of the projection data. We obtain this solution
by solving a series of linear equations by a least squares min-
imization technique (Lawrence 1992). As with any such solu-
We cannot overstate the importance of accurate tilt se- tion of unknowns the achievable accuracy is dependent on the
ries alignment. Even a small deviation from optimal will amount of information we start with; in this case the number of
give you confusing reconstructions filled with spurious fiducial markers and projection images. A distinct advantage of
features. You will find that reaching an optimal, sub-pixel this approach over cross-correlation is that it provides not only
accuracy, alignment can sometimes be a tedious task, but
Chapter 12

the x-y registration of the images but also any distortions in the
one that is entirely unavoidable. 9 individual images such as shears, rotations and/or magnification
changes. And usefully the direction of the tilt axis also falls out
 12.14 Tilt-Series Alignment 367

Part I
Fig. 12.28 Example of cross-correlation to determine the relative shift between two images. a, b Two images of a nanostructured block-copolymer from a HAADF
STEM tilt series. c the cross-correlation image between the two images, showing the high intensity peak at (–2.19) pixels from the image center. This corresponds to
the shift between the two images. (d–f) are the same series of image, but show the improvement in the sharpness of the cross-correlation peak on application of a
high pass filter to the original images

of the least squares solution. However, as we add more factors to mismatch between the two images. We can achieve a quantitative
solve we also require more fiducial markers; in practice however description of this difference, and determine the shift, by cal-
around 15–20 markers are sufficient. The method is not without culating the cross-correlation between images. This is demon-
its weaknesses however. Requiring a certain number of markers strated in Fig. 12.28. The attraction of this approach is that it can
for alignment instantly excludes large areas of the specimen, a be applied to tilt series of any arbitrary object shape, requires no
serious limitation if there are few areas to study on our speci- manual selection of features and avoids any reconstruction arte-
mens. The alignment accuracy is also limited by the accuracy of facts associated with high intensity markers. We can achieve the
the marker position, which is usually determined semi-manu- most accurate track by applying a spatial filter or edge detection
ally. Ideally these would also be point objects, however at higher algorithm, enhancing image features of selected frequencies, to
magnifications the finite size of the fiducial marker can reduce the original images. Figure 12.28 clearly shows the increase of
the accuracy of the least squares fit. In addition, the presence of sharpness, and hence certainty in position, of the cross-correla-
high-density objects in the projection images can causes object tion peak. Perhaps most serious limitation to this approach is that
features in the same reconstruction plane to be hidden/masked it does not, unlike marker tracking, determine the direction of
due to the PSF for reconstruction. the tilt axis and as such we need to apply additional processing
before we can reconstruct. Likewise is does not solve or deter-
If we watch a tilt series as a movie we see huge changes in image mine any second-order effects such as magnification changes,
features from one extreme of tilt to another, however between image distortions or rotations. Another recognized limitation to
Chapter 12

any two consecutive images there is very little change. This cross-correlation is that it can be insensitive to small sub-pixel
small change between subsequent projections means that the dif- shifts. Between any pair of images the effect will be small, but
ference between the two images will be dominated by the spatial compounded quarter pixel error per image over 140 images will
 368 12 Electron Tomography
Part I

Fig. 12.29 Tilt axis direction determination by series summation. a A single STEM HAADF image, at zero tilt, from a tilt series acquired from a catalyst structure of a
carbon matrix embedded with Pd particles. b The direction of the tilt axis is determined using a tilt series summation, or (c) its Fourier transform

be catastrophic for the final reconstruction. The application of features; find the plane where these features are conserved and
a robust sub-pixel approach to alignment, and simply aligning you have found the plane normal to the tilt axis direction. If we
the series from the middle rather than from one extreme of tilt to carry out a reconstruction with our tilt axis undetermined, or at
the other, can minimize this effect. In addition cross-correlation least non-optimal, it will be blighted by characteristic arcing fea-
alignment is often applied in multiple passes as the apodization tures; these arise as a rotation introduces a positional uncertainty
typically applied to the images will centrally weight the images to the image intensity. These arcing features will also appear if the
features and lead to errors with large mismatches. reconstruction is re-projected; and one approach to optimization
can be made by iterative adjustment of rotational alignment and
The third method of alignment, based on refinement of the re- reconstruction (Winkler and Taylor 2006). Another option is to
sultant reconstruction, is a relatively new approach (Houben reconstruct a small number of 2D slices at different tilt axis di-
and Bar 2011) that has been enabled by the availability of fast rections and observing the effect on the resultant reconstructions
computing. Here the alignment and reconstruction are iterated estimate the correct axis direction.
to maximize contrast in the reconstruction, and shows promise
to bridge the precision gap between cross-correlation and mark- Biological electron tomography typically makes use of fiducial
er-based alignment. tracking techniques. This is largely because the uniformly low con-
trast and signal-to-noise ratios achieved from BF TEM images of
Automatic determination the tilt axis direction is one the key thin, unstained, biological sections; it is difficult for us to achieve
advantages of the marker tracking approach to alignment. How- accurate cross-correlation under these conditions. In materials sci-
ever, if we are aligning a tilt series without fiducial markers we ence we often only have a limited number of suitable areas to study
still have to determine this direction in order to be able to recon- in our specimens, this makes the random nature of fiducial marker
struct of series. There are a number of solutions to this problem, distribution a barrier to their use. In addition the higher contrast
none of which has yet to become the de facto standard. Generally available from strongly scattering specimens, irrespective of the
they fall into two categories, those which work on the common imaging technique, leads to significant increases in the accuracy of
line theory and those which analyze the resultant reconstruction. cross-correlation. Whichever technique you choose for alignment
The common line approach is based on the simple reasoning that will therefore depend on the specimen under study, and a degree
given that a spatially aligned, single axis, tilt series all significant of pre-experimental planning is undoubtedly required!
movement of image features through the series should be perpen-
dicular to the tilt axis direction. Probably the simplest way we
can make use of this theory is to sum the image series and look
for the direction of any linear features caused by this movement, 12.15 Visualization of Three-
which should be perpendicular to the tilt axis direction. This is
demonstrated in Fig. 12.29. A more sophisticated approach ana- Dimensional Datasets
lyzes the Fourier transforms of the tilt series images that, given
the foreshortening of image features in closely spaced tilt images, We have already spent a large portion of this book discussing the
Chapter 12

should show a smear of the cross-correlation peak in the direction interpretation of two-dimensional micrographs, and the inherent
of the tilt axis. Another option is for us to analyze the tilt series challenges within. However, the analysis of three-dimensional
in Radon space (Liu et al. 1995), searching for distinctive sinθ volumes presents us with a whole new series of difficulties. The
 12.16 Segmentation 369

most significant of which is probably the sheer volume of data;

from which making valid conclusions takes great care.

Part I
The most rigorous way we have of displaying our 3D data is,
ironically, to reveal the internal reconstruction intensity in 2D by
taking individual ‘slices’ of the volume. Isn’t this just a backward
step? Here we must stress the difference between the intensity in
a ‘slice’ and that in a projection. As we well know the intensity in
a projection is difficult to interpret as we lose information about
its distribution though the specimen. However, the intensity in
a slice shows that distribution without this ambiguity. For this
reason, it is important to show slices when presenting your to-
mographic results, and to evidence a healthy skepticism to any
results presented without them! We can slice along the primary
axes (x,y,z), or produce slices in arbitrary planes using some form
of three-dimensional interpolation. An example of slicing of a
tomographic dataset is shown in Fig. 12.30.

In collecting the original tilt series, tilting and projecting, we

achieved a visualization of a 3D volume. It is perfectly possible
to carry out the same task in the computer, rotating the data- Fig. 12.30 a A 3D volume can be sliced along three major orthogonal axes
set and then projecting (summing) in a single direction; except (b–d). These reveal the internal structure of the object. These are sometimes
without any need for all the tedious tracking and focusing. This termed axial (z,y), sagittal (z,x) and coronal (y,x) directions following medical
is known as voxel projection, a voxel being a three-dimensional imaging terminology. Slices can also be generated in oblique directions
pixel, or sometimes just direct volumetric rendering. In the com-
puter we have complete freedom to rotate about any axis and
produce a projection from any direction, and we can view any of structures clearly and allows a more intuitive ‘physical’ con-
sub volume of interest without its surroundings simply by crop- ceptualization of the reconstruction. Different structures within the
ping. But perhaps most powerfully we are not limited to view- volume can also be differentiated out by variations of color and/
ing a simple sum of the data. We can apply any kind of optical or transparency. A comparison between surface rendering and vol-
condition we wish to the projection using the intensity values of ume visualization is shown in Fig. 12.32. The challenge becomes
the volume. These include, but are not limited to; adding color, how we best choose the surface to display. A simple approach,
modifying the opacity, exclusion to render certain features invis- and that demonstrated in Fig. 12.31b, is to form a surface at a
ible and changing the transfer function. Voxel projection gives single intensity value in the reconstruction. Such an isosurface is
us a highly flexible method of visualization, but also a range of quick to generate and, if chosen carefully, should correspond to
opportunities to generate misleading images! In addition voxel a meaningful structure within our volume. However, we quickly
projection is comparatively resource intensive, though with mod- find that an accurate value is not so easy to select with most tomo-
ern computers, this is becoming less of a limitation. An example graphic reconstructions. The effect of the tomographic PSF acts
of a voxel projection is shown in Fig. 12.31a. to spread out the intensity of even the most abrupt interfaces and,
more seriously, its anisotropy leads to varying interface intensities
Selective Abstractions! in different directions inside our reconstruction. With sufficient tilt
range, and strong contrast between structures, an isosurface can
provide an adequate approximation. However, in many cases, and
The reader should never lose sight of the fact that most certainly for nearly all biological reconstructions, we will not find
3D volume visualizations, while often aesthetically pleas- isosurfaces terribly informative In these cases we need to employ
ing, are selective abstractions of the dataset as a whole. some more advanced method of differentiating meaningful vol-
There is a real danger when carrying out this abstraction umes within our reconstructions; a process known as segmentation.
of generating exactly what you wish to see! 9

Voxel projection, while informative, tends to share the central 12.16 Segmentation
limitation of the original projections; it is often difficult to relate
the visualization with internal structure. It is for this reason that a
more popular method of 3D visualization is surface rendering. By Interestingly segmentation, which we can define as the separa-
Chapter 12

creating an opaque polygonal surface of some structure contained tion of grayscale pixels (or voxels) into regions of structural,
within the reconstruction visualized by adding lighting, shade, functional, or chemical similarity, is one of the central challenges
depth, and color. This technique reveals the surface morphology of machine vision across many fields of research, not just for
 370 12 Electron Tomography
Part I

Fig. 12.31 A comparison of (a) voxel projection and (b) isosurface rendering for the same volume of a HAADF STEM tomography reconstruction of precipitate plates
in an Al alloy. The voxel projection is analogous to the original micrograph, and allows visualization of buried structures, but can be equally hard to interpret. The
isosurface gives a real impression of the 3D shape of the reconstruction, but the lack of a definitive surface intensity leads to holes in structures and ‘peppering’ of
small objects that may or may not be real especially at the volume edges

electron tomography (Russ 2002). As a consequence, there are a nique that can applied to all specimens. Interestingly, most of the
broad range of techniques for automatic segmentation of images; difficulties in applying these techniques are related to the poor
but it is quite telling that none have seen wide application in contrast and SNR of the average BF biological reconstruction.
electron tomography. Indeed, most segmentation is carried out It would seem that with the reconstructions we can achieve from
manually. Manual segmentation is a laborious process; for every high contrast imaging techniques and strongly scattering materi-
slice of the reconstruction we mark boundaries, usually using a als, automatic segmentation should be re-evaluated.
cursor, around pixels which we determine to belong to particular
structures. The advantage of manual segmentation is that surface Objectivity in Segmentation
visualization of these volumes can be informative and aestheti-
cally pleasing, as shown in Fig. 12.32. A distinct disadvantage is
that in drawing boundaries we make many objective decisions, The potential for objectivity is one of the most attrac-
and as such it is dangerous to rely on segmented volumes alone. tive aspects of automated segmentation. An individ-
Visualizations such as these are usually partnered with 2D slices ual segmentation might offer a less than accurate in-
of the reconstruction data, especially when we wish to make dividual measurement of a particular microstructural
conclusions about the precise structure of the specimen. feature, but a ‘hands-off’ automated procedure should
allow differences between volumes to be quantified in
Developing a reliable approach to automatic, or even semi-au- an unbiased way that could never be carried out with
tomatic, segmentation would seem a logical next step. And, in- manual segmentation. 9
deed, a number of different research groups have demonstrated
such approaches. We find examples of segmentation carried out
by statistical evaluation of the voxel intensities (Frangakis and Perhaps the most promising approach to a general approach
Hegerl 2002) and others who have applied segmentation based to segmentation comes from the application of discrete tomo-
on interconnectedness (Volkmann 2002) or by searching for graphic reconstruction, see Sect. 12.3.2. By constraining the
sharp gradients within the image data (Bajaj et al. 2003). In each reconstruction to discrete intensity levels, this approach carries
case, applying these approaches to reconstructions has resulted out reconstruction and segmentation in a single step. In the limit
Chapter 12

in successful segmentations, but also in each case the technique this generates a completely objective, reproducible segmenta-
has to be tuned to the specimen in order to give accurate results. tion result. In practice things are not quite so simple. Subjective
We are therefore a long way from a robust segmentation tech- choice is still used to determine the number, and voxel value, of
 12.17 Quantitative Analysis of Volumetric Data 371

Part I
Fig. 12.32 Segmentation and quantitative analysis of volumes, for a HAADF STEM tomography reconstruction of precipitate plates in an Al alloy. The volume shown
is the same as visualized by voxel and isosurface rendering in Fig. 12.31. In this case segmentation allows clear differentiation in the visualization of different families
of precipitates (determined by their relative orientation), identification of unknown phases as well as direct measurement of volume fractions

the discrete steps. It will also always offer a discrete reconstruc- us reach the solution to many classical problems in 2D micro­
tion with these steps; there is no allowance for deviations, this scopy; which can, in turn, lead to insights into both synthesis and
in itself is a subjective choice. physical properties of materials. A simple summation over the
segmented volumes will give a measure of true volume fraction,
as shown in Fig. 12.32. Given that we know the positions of all
the voxels we can also quickly calculate volumes of individual
12.17 Quantitative Analysis features inside the reconstruction, such as measurements of over-
age particle size and size distributions. We should also be able
of Volumetric Data to measure of other basic physical properties from a segmented
volume such as aspect ratios, surface areas, ordering, roughness,
Demonstrations of the segmentation of tomographic volumes, densities, and homogeneity.
whether automatic or manual, have typically been carried out in
order to allow improved visualization of biological macromole- The segmentation of data also allows more subtle analysis of
cules. Their complex, overlapping and low contrast features are the specimen morphology, and can lead us to insights into the
difficult for us to visualize in any other way. In materials science, fundamentals of a materials performance. For example, Ward
however, the process of segmentation leads us to datasets which et al. (2007), examined the local curvature of mesoporous silica
Chapter 12

are also eminently suited to quantitative analysis. Segmentation heterogeneous catalyst supports. The shape of these supports,
is effectively a means to partition voxels into groups belonging and the relationships between support and active nanoparticles,
to one structure, phase or chemical element. This should allow is critical in modeling the performance of these systems. Af-
 372 12 Electron Tomography

ter segmentation analysis found this particular support showed In materials science we are always seeking the correlation be-
strong fractal character, and that the active nanoparticles showed tween microstructure and materials properties, and in many cases
a preference for surfaces with concave ‘saddle’ shaped portion we need to describe it in quantitative terms to be of real value
Part I

of the support. These findings can drive future models for cata- to predict properties and guide synthesis. And while qualitative
lytic activity, as well as decisions on support synthesis. Having a visualization is undeniably attractive, and can be of extraor-
numeric handle on physical characteristics it also becomes pos- dinary useful, it will be by providing quantitative analysis of
sible to relate them to the macroscale physical properties of the microstructure that electron tomography will be of most value
material. The resistivity of rubber composites had been seen to in materials science. Indeed, for many problems at the nano-
vary with the volume fraction of carbon black, and the research structural level it can provide site-specific information which is
of Ikeda et al. (2004) has shown that the resistivity is directly unattainable by any other technique.
correlated to the nearest neighbor distance of carbon particles.

Chapter Summary

No matter how thin you make your specimen it will always the tools to resolve these complexities for a wide range of
have a finite thickness; this makes it impossible for you to materials. In applying the technique you should pay partic-
completely ignore, no matter how tempting it is, the exis- ular attention to the following four points.
tence of the third dimension. In many cases it may have – Selection of the most suitable imaging technique for the
little impact on your results, especially when you carefully material.
prepare and orient you specimen. However, it would be an – Considering, and hence minimizing, the effects of dose
atypical microscopist who didn’t spend at least some of and contamination.
their professional life scratching their heads while trying to – Optimizing tilt series alignment – no matter how tedious!
reach conclusions from micrographs of complex overlap- – Understanding the power, and limitations, of visualization
ping structures! Electron tomography can provide you with and segmentation.
Chapter 12
 Appendix 373

Appendix

Part I
People Q9.13 Describe two different constraints that can be applied in
tomographic reconstruction. Include a justification for
their use.
As microscopists, we may think of tomography as being a nat- Q9.14 Compare and contrast slicing and surface rendering as
ural progression from stereo pairs, in which case you might be means of visualizing tomographic reconstructions.
interested to read about Sir Charles Wheatstone (born February 6, Q9.15 What is the ‘projection requirement’ and why is it so
1802, died October 19, 1875) and Sir David Brewster (born De- important when carrying out electron tomography on
cember 11, 1781, died February 10, 1868); you’ve heard of each engineering materials?
before in very different contexts! Explore the meaning of the word Q9.16 Why is HAADF STEM an ideal imaging mechanism for
planography. Then look at the invention of Godfrey Newbold ET?
Hounsfield (born August 28, 1991, died August 12, 2004), who Q9.17 List the challenges in carrying out electron tomography
used computers and X-rays and shared a Nobel Prize in 1979. using secondary X-ray signals. Suggest means to over-
come these challenges.

Self-Assessment Questions
References
Q9.1 Describe the importance of the Radon transform to (elec-
tron) tomography.
Q9.2 Draw a diagram to illustrate what is meant by the pro- General Tomography References
jection slice theorem. Explain its implications for tomo-
graphic reconstruction. Deans SR (2007) The Radon transform and some of its appli-
Q9.3 Describe one experimental situation/specimen (physical cations. Dover Publications, Mineola, New York (The most
or biological) where electron tomography WOULD be of comprehensive and readable reference regarding the mathe-
value. matics of the Radon transform. Best of all it contains an En-
Q9.4 Suggest one experimental situation/specimen (physical or glish translation of Radon’s original German paper. Unavail-
biological) where electron tomography WOULD NOT be able for many years, this is a very welcome, and reasonably
of value. priced, reprint)
Q9.5 Explain some of the limitations that might be imposed Egerton RF, Li P, Malac M (2004) Radiation damage in the TEM
by the electron microscope and specimen holder on the and SEM. Micron 35(6):399–409 (Electron beam damage is a
acquisition of tilt series. Suggest some means for over- central concern in ET, as such comprehensive review of beam
coming these limitations. damage is invaluable)
Q9.6 Describe some of the limitations that might be imposed Frank J (2006) Electron Tomography; Methods for Three-Di-
by the specimen on the acquisition of tilt series. Suggest mensional Visualization of Structures in the Cell, 2nd edn.
some means for overcoming these limitations. Springer, New York. (The standard ET ‘textbook’ for the
Q9.7 Before tomographic reconstruction can occur the tilt se- biological sciences, in the absence of an equivalent for the
ries needs to be aligned with respect to a common tilt materials sciences this is the best general reference)
axis. Describe the TWO parts of this alignment. Midgley PA, Ward EPW, Hungría AB, Thomas JM (2007)
Q9.8 Why is there more than one valid tomographic alignment Nanotomography in the chemical, biological and materials
for each tilt series? sciences. Chem Soc Rev 36:1477–1494 (A review of na-
Q9.9 List the advantages and limitations of alignment by track- noscale electron tomography with comparisons with compa-
ing of fiducial markers. rable X-ray experiments)
Q9.10 List the advantages and limitations of alignment by Saghi Z, Midgley PA (2012) Electron Tomography in the (S)
cross-correlation. TEM: From Nanoscale Morphological Analysis to 3D Atomic
Q9.11 Derive the equation to generate an r-weighting filter for Imaging. Annual Rev Mater Res 42:59–79 (A review focus-
tomographic reconstruction. ing primarily on STEM-based tomography. The volume also
Q9.12 Under what circumstances would r-weighting reconstruc- contains reviews on confocal STEM, cryo-tomography, X-ray
Chapter 12

tion generate a ‘perfect’ reconstruction? tomography and atom probe tomography)

Licensed to ([email protected])
 374 12 Electron Tomography

Specific References Fernandez JJ, Li S, Crowther RA (2006) CTF determination

and correction in electron cryotomography. Ultramicroscopy
Van Aert S, Batenburg KJ, Rossell MD, Erni R, Van Tendeloo 106(7):587–596
Part I

G (2011) Three-dimensional atomic imaging of crystalline Frangakis AS, Hegerl R (2002) Segmentation of two- and
nanoparticles. Nature 470:374–377 three-dimensional data from electron microscopy using ei-
Aoyama K, Takagi T, Hirase A, Miyazawa A (2008) STEM to- genvector analysis. J Structural Biol 138(1–2):105–113
mography for thick biological specimens. Ultramicroscopy Frigo SP, Levine ZH, Zaluzec NJ (2002) Submicron imaging of
109:70–80 buried integrated circuit structures using scanning confocal
Bajaj C, Yu ZY, Auer M (2003) Volumetric feature extraction and electron microscopy. Appl Phys Lett 81:2112–2114
visualization of tomographic molecular imaging. J Structural Gass MH, Koziol KK, Windle AH, Midgley PA (2006) 4-dimen-
Biol 144(1–2):132–143 sional spectral-tomography of carbonaceous nano-compos-
Barnard JS, Sharp J, Tong JR, Midgley PA (2006) High-Reso- ites. Nano Lett 6(3):376–379
lution Three-Dimensional Imaging of Dislocations. Science Gilbert P (1972) Iterative Methods for the Three-dimensional
313:319 Reconstruction of an Object from Projections. J Theoretical
Batenburg KJ, Bals S, Sijbers J, Kübel C, Midgley PA, Hernan- Biol 36:105–117
dez JC, Kaiser U, Encina ER, Coronado EA, Van Tendeloo G Gordon R, Bender R, Herman GT (1970) Algebraic Reconstruc-
(2009) 3D imaging of nanomaterials by discrete tomography. tion Techniques (ART) for Three-dimensional Electron Mi-
Ultramicroscopy 109:730–740 croscopy and X-ray Photography. J Theor Biol 29:471–481
Baumeister W, Grimm R, Walz J (1999) Electron tomography of Goris B, Bals S, Van den Broek W, Carbó-Argibay E, Gó-
molecules and cells. Trends Cell Biol 9(2):81–85 mez-Graña S, Liz-Marzán LM, Van Tendeloo G (2012) Atom-
van Benthem K, Lupini AR, Kim M, Baik HS, Doh S, Lee JH, ic-scale determination of surface facets in gold nanorods. Na-
Oxley MP, Findlay SD, Allen LJ, Luck JT, Pennycook SJ ture Mater 11:930–935
(2005) Three-dimensional imaging of individual hafnium at- Haberfehlner G, Orthacker A, Albu M, Li J, Kothleitner G (2014)
oms inside a semiconductor device. Appl Phys Lett 87:034104 Nanoscale voxel spectroscopy by simultaneous EELS and
Bracewell RN (1956) Strip Integration in Radio Astronomy. Aust EDS tomography. Nanoscale 6:14563–14569
J Phys 9:297–314 Hawkes PW (1992) The Electron Microscope as a Structure Pro-
Candes EJ, Romberg J, Tao T (2006) Robust uncertainty prin- jector. In: Frank J (ed) Electron tomography: three-dimen-
ciples: Exact signal reconstruction from highly incomplete sional imaging with the transmission electron microscope.
frequency information. IEEE Trans Information Theory Plenum Press, London, pp 17–39
52:489–509 Hegerl R, Hoppe W (1976) Influence Of Electron Noise On
Chen CC, Zhu C, White ER, Chiu CY, Scott MC, Regan BC, 3-Dimensional Image-Reconstruction. Z Naturforschung
Marks LD, Huang Y, Miao J (2013) Three-dimensional im- Section A 31(12):1717–1721
aging of dislocations in a nanoparticle at atomic resolution. Houben L, Bar SM (2011) Refinement procedure for the image
Nature 496:74–77 alignment in high-resolution electron tomography. Ultrami-
Crowther RA, de Rosier DJ, Klug A (1970) The reconstruction croscopy 111:1512–1520
of a three-dimensional structure from projections and its Hyun JK, Ercius P, Muller DA (2008) Beam spreading and spa-
application to electron microscopy. Proc Roy Soc Lond A tial resolution in thick organic specimens. Ultramicroscopy
317:319–340 109:1–7
Dierksen K, Typke D, Hegerl R, Koster AJ, Baumeister W (1992) Ikeda Y, Katoh A, Shimanuki J, Kohjiya S (2004) Nano-struc-
Towards Automatic Electron Tomography. Ultramicroscopy tural observation of in situ silica in natural rubber matrix by
40(1):71–87 three dimensional transmission electron microscopy. Macro-
Donoho DL (2006) Compressed Sensing. IEEE Trans Informa- molecular Rapid Commun 25(12):1186–1190
tion Theory 52:1289–1306 Isabell TC, Fischione PE, O’Keefe C, Guruz MU, Dravid VP
Dwyer C, Weyland M, Chang LY, Muddle BC (2011) Combined (1999) Plasma cleaning and its applications for electron mi-
electron beam imaging and ab initio modeling of T(1) precip- croscopy. Microsc Microanal 5(2):126–135
itates in Al-Li-Cu alloys. Appl Phys Lett 98:201909 Jarausch K, Thomas P, Leonard DN, Twesten R, Booth CR
Van Dyck D, Chen FR (2012) ‘Big Bang’ tomography as a (2009) Four-dimensional STEM-EELS: Enabling nano-scale
new route to atomic-resolution electron tomography. Nature chemical tomography. Ultramicroscopy 109:326–337
486:243–246 Jinnai H, Nishikawa Y, Spontak RJ, Smith SD, Agard DA,
Egerton RF, Li P, Malac M (2004) Radiation damage in the TEM Hashimoto T (2000) Direct measurement of interfacial cur-
and SEM. Micron 35(6):399–409 vature distributions in a bicontinuous block copolymer mor-
Ercius P, Weyland M, Muller DA, Gignac LM (2006) Three phology. Phys Rev Lett 84(3):518–521
-dimensional imaging of nanovoids in copper interconnects Kaiser U, Biskupek J, Meyer JC, Leschner J, Lechner L, Rose
using incoherent bright field tomography. Appl Phys Lett H, Stoger-Pollach M, Khlobystov AN, Hartel P, Muller H,
Chapter 12

88(24):243116 Haider M, Eyhusen S, Benner G (2011) Transmission electron

microscopy at 20 kV for imaging and spectroscopy. Ultrami-
croscopy 111:1239–1246
 References 375

Kimura K, Hata S, Matsumura S, Horiuchi T (2005) Dark-field De Rosier DJ, Klug A (1968) Reconstruction of Three Di-
transmission electron microscopy for a tilt series of order- mensional Structures from Electron Micrographs. Nature
ing alloys: toward electron tomography. J Electron Microsc 217:130–134

Part I
54(4):373–377 Russ JC (2002) The Image Processing Handbook. CRC Press,
Koster AJ, Van Den BA, Van Der MKD (1987) An autofocus London
method for a TEM. Ultramicroscopy 21:209–222 Saghi Z, Holland DJ, Leary R, Falqui A, Bertoni G, Sederman
Lade SJ, Paganin D, Morgan MJ (2005) Electron tomography of AJ, Gladden LF, Midgley PA (2011) Three-Dimensional Mor-
electromagnetic fields, potentials and sources. Optics Com- phology of Iron Oxide Nanoparticles with Reactive Concave
mun 253:392–400 Surfaces. A Compressed Sensing-Electron Tomography
Lawrence MC (1992) Least-Squares Method of Alignment Using (CS-ET) Approach. Nano Lett 11:4666–4673
Markers. In: Frank J (ed) Electron tomography: three-dimen- Sawada H, Sasaki T, Hosokawa F, Yuasa S, Terao M, Kawa-
sional imaging with the transmission electron microscope. zoe M, Nakamichi T, Kaneyama T, Kondo Y, Kimoto K,
Plenum Press, London., pp 197–204 Suenaga K (2010) Higher-order aberration corrector for an
LeBeau JM, Findlay SD, Allen LJ, Stemmer S (2008) Quantita- image-forming system in a transmission electron microscope.
tive atomic resolution scanning transmission electron micros- Ultramicroscopy 110:958–961
copy. Phys Rev Lett 100:206101 Saxton WO, Baumeister W, Hahn M (1984) 3-Dimensional Re-
Lepinay K, Lorut F, Pantel R, Epicier T (2013) Chemical 3D construction of Imperfect Two-Dimensional Crystals. Ultra-
tomography of 28 nm high K metal gate transistor: STEM microscopy 13(1–2):57–70
XEDS experimental method and results. Micron 47:43–49 Scott MC, Chen CC, Mecklenburg M, Zhu C, Xu R, Ercius P,
Liu Y, Penczek PA, McEwen BF, Frank J (1995) A Marker- Free Dahmen U, Regan BC, Miao J (2012) Electron tomography
Alignment Method For Electron Tomography. Ultramicros- at 2.4-angstrom resolution. Nature 483:444–491
copy 58(3–4):393–402 Spontak RJ, Williams MC, Agard DA (1988) 3- Dimensional
Mastronarde DN (1997) Dual-axis tomography: An approach Study Of Cylindrical Morphology In A Styrene Butadiene
with alignment methods that preserve resolution. J Structural Styrene Block Copolymer. Polymer 29(3):387–395
Biol 120(3):343–352 Stolojan V, Dunin-Borkowski RE, Weyland M, Midgley PA
Mastronarde DN (2005) Automated electron microscope to- (2001) Three-dimensional magnetic fields of nanoscale ele-
mography using robust prediction of specimen movements. ments determined by electron-holographic tomography. Elec-
J Structural Biol 152(1):36–51 tron Microsc Anal 2001:243–246
Miao JW, Forster F, Levi O (2005) Equally sloped tomography Tanaka M, Higashida K, Kaneko K, Hata S, Mitsuhara M (2008)
with oversampling reconstruction. Phys Rev B 72:052103 Crack tip dislocations revealed by electron tomography in
Midgley PA, Weyland M (2003) 3D electron microscopy in the silicon single crystal. Scripta Mater 59:901–902
physical sciences: the development of Z-contrast and EFTEM Twitchett-Harrison AC, Yates TJV, Dunin-Borkowski RE, Midg-
tomography. Ultramicroscopy 96(3–4):413–431 ley PA (2008) Quantitative electron holographic tomography
Midgley PA, Weyland M, Thomas JM, Johnson BFG (2001) for the 3D characterization of semiconductor device struc-
Z-Contrast tomography: a technique in three-dimensional tures. Ultramicroscopy 108:1401–1407
nanostructural analysis based on Rutherford scattering. Chem Volkmann N (2002) A novel three-dimensional variant of the wa-
Commun 2001:907–908 tershed transform for segmentation of electron density maps.
Mobus G, Doole RC, Inkson BJ (2003) Spectroscopic electron J Structural Biol 138(1–2):123–129
tomography. Ultramicroscopy 96:433–451 Voortman LM, Stallinga S, Schoenmakers RHM, van Vliet LJ,
Nicoletti O, de la Pena F, Leary RK, Holland DJ, Ducati C, Rieger B (2011) A fast algorithm for computing and correct-
Midgley PA (2013) Three-dimensional imaging of localized ing the CTF for tilted, thick specimens in TEM. Ultramicros-
surface plasmon resonances of metal nanoparticles. Nature copy 111:1029–1036
502:80–84 Ward EPW, Yates TJV, Fernandez JJ, Vaughan DEW, Midgley
Phatak C, Beleggia M, De Graef M (2008) Vector field electron PA (2007) Three-dimensional nanoparticle distribution and
tomography of magnetic materials: Theoretical development. local curvature of heterogeneous catalysts revealed by elec-
Ultramicroscopy 108:503–513 tron tomography. J Phys Chem C 111:11501–11505
Phatak C, Petford-Long AK, De Graef M (2010) Three-Dimen- Weyland M, Midgley PA, Thomas JM (2001) Electron tomog-
sional Study of the Vector Potential of Magnetic Structures. raphy of nanoparticle catalysts on porous supports: A new
Phys Rev Lett 104:253901 technique based on Rutherford scattering. J Phys Chem B
Qin W, Fraundorf P (2003) Lattice parameters from direct-space 105(33):7882–7886
images at two tilts. Ultramicroscopy 94:245–262 Weyland M, Yates TJV, Dunin-Borkowski RE, Laffont L, Midg-
Radermacher M (1988) 3-Dimensional Reconstruction of Single ley PA (2006) Nanoscale analysis of three-dimensional struc-
Particles From Random and Nonrandom Tilt Series. J Elec- tures by electron tomography. Scripta Mater 55:29–33
tron Microsc Techniq 9(4):359–394 Winkler H, Taylor KA (2006) Accurate marker-free alignment
Chapter 12

Radon J (1917) Uber die Bestimmung von Funktionen durch with simultaneous geometry determination and reconstruc-
ihre Intergralwerte langs gewisser Mannigfaltigkeiten. Ber tion of tilt series in electron tomography. Ultramicroscopy
Verh K Sachs Ges Wiss Leipzig, Math-Phys Kl 69:262–277 106(3):240–254
 376 12 Electron Tomography

Wolf D, Lubk A, Roeder F, Lichte H (2013) Electron holo- method of data collection for automated electron tomogra-
graphic tomography. Current Opinion Solid State Mater Sci phy based upon pre-calibration of image shifts and defocus
17:126–134 changes. Microsc Microanal. Springer, Long Beach, CA
Part I

Xin HL, Muller DA (2010) Three-Dimensional Imaging in Ab- Zhang X, Jin L, Fang Q, Hui WH, Zhou ZH (2010) 3.3 angstrom
erration-Corrected Electron Microscopes. Microsc Microanal Cryo-EM Structure of a Nonenveloped Virus Reveals a Prim-
16:445–455 ing Mechanism for Cell Entry. Cell 141:472–482
Zeise U, Janssen R, Geerts W, van der Krift T, van Balen A, de
Ruijter H, de Jong K, Verkleij A, Koster B (2001) A novel
Chapter 12
 EFTEM
13

Chapter 13
Chapter Preview

When we view a conventional TEM image in the electron cess is termed ‘energy-filtering’, and energy-filtered TEM
microscope, we observe contrast, which is comprised of (EFTEM) provides a simple but powerful means for utiliz-
both elastic and inelastic scattering information. The elas- ing the energy-loss spectrum to reveal additional informa-
tic components may be filtered by their angular range by tion about our specimens, whether by a simple improvement
means of inserting an objective aperture to produce, for in resolution or as an element- (or even chemical-state) spe-
example, a bright- or dark-field image. Likewise, the in- cific map. This chapter discusses the topic of energy-filtered
elastic component of the transmitted image can be filtered transmission electron microscopy (EFTEM). It reviews the
in terms of the energy-loss range by use of an imaging filter technique, combining theory with a number of applications
with an energy-selecting slit. Hence, in analogy with dark from materials science to highlight the progress made in
field imaging, an energy-loss image may be produced com- the subject.
posed of a discrete range of the EELS spectrum. This pro-

© Springer International Publishing Switzerland 2016

C. B. Carter, D. B. Williams (Eds.), Transmission Electron Microscopy, DOI 10.1007/978-3-319-26651-0_13 377
 378 13 EFTEM

13.1 Introduction acquisition of (typically) a number of images separated in energy

loss (Fig. 13.1).
Chapter 13

Since the introduction of the commercial imaging filter for the In principle, the scanning approach is far more efficient since
transmission electron microscope in the early 1990s, EFTEM the majority of information at each point is recorded, unlike
has emerged as a powerful and highly versatile tool for materials EFTEM in which only the region defined by the energy-select-
analysis. By combining the parallel imaging capabilities of the ing slit is sampled at each acquisition. Accordingly, the total
TEM with the abundant information contained in the electron electron dose for the scanning techniques is considerably less
energy-loss spectrum, two- (and even three-) dimensional maps than in EFTEM mapping, reducing possible radiation damage
regarding the material's elemental, chemical and physical state to the sample, which may be of importance in many beam-sen-
can be obtained in a fast, parallel fashion, resolving features up sitive materials. In practice, however, the non-negligible readout
to the nanometer scale with all the simplicity and convenience time per spectrum of EELS detectors results in relatively long
of using fixed beam illumination. acquisition times for maps acquired using the scanning approach
containing a large number of pixels. Total acquisition times for
As described previously in Chap. 37 of W&C, the electron ener- EELS spectrum-images can be tens of minutes for typical image
gy-loss spectrum contains a wide variety of additional informa- acquisitions, although advances in both EELS detector and elec-
tion about the material under analysis. The presence and relative tron-source technology continues to reduce this considerably.
size of core-loss ionization edges can provide both qualitative The recording time in EFTEM image acquisition is almost inde-
and quantitative compositional data. Chemical and solid-state pendent of the number of pixels, allowing elemental distribution
information is encoded in the shape of the ionization edges, and maps to be acquired in a matter of minutes. Hence, a number of
physical properties such as the local thickness of the specimen analyses may be performed within a non-prohibitive time scale.
can be calculated from the scattering statistics of the whole spec- The spatially parallel nature of EFTEM allows sample drift be-
trum. In principle, the same methods developed for EELS for tween individual acquisitions to be corrected with considerably
extracting local information about the specimen can be applied greater ease than with scanning methods. In addition, although
in a spatially parallel manner using EFTEM, albeit typically at the total electron dose of the specimen is larger in EFTEM,
a lower energy resolution. This ability of EFTEM to efficiently the current density of the probe is considerably smaller. Hence,
record spatially-resolved information represents an important dose-rate-dependent damage experienced in scanning probe
advantage; namely that calculated quantities can, in principle, studies, such as hole-drilling and the movement of segregants,
be computed and displayed as maps. can be minimized by using EFTEM.

In recent years, much has been done to develop techniques to im- EFTEM offers similar spatial resolution to EELS spectrum-im-
prove EFTEM analysis. In particular, emphasis has been placed aging (approximately 1 nm for a typical experiment), with de-
on the determination of chemical composition and distribution localization effects providing the fundamental limiting factor.
at the nanometer level, in a quantitative manner with improved It also generally gives superior spatial resolution to XEDS
confidence levels. This chapter describes the key elements of spectroscopy with considerably better light-element sensitiv-
EFTEM, the relevant quantification of the energy-loss spectrum, ity. Since EFTEM is performed in conventional TEM mode
and the instrumentation needed, and details some key examples (i.e., under parallel illumination) it is not dependent on your
that illustrate the ability of EFTEM to provide maps of com- microscope having STEM or small analytical probe capabili-
positional and electronic information at high spatial resolution. ties. Further, whereas XEDS spectroscopy records the second-
ary excitations created by the fast electron interacting with the
atomic electrons, EELS-based techniques utilize the primary
signal. Since excitation events are recorded independently of
13.2 Why Use EFTEM? their de-excitation mechanism, EELS and EFTEM yield infor-
mation on both the momentum and energy transfer between the
incident and atomic electrons, and can provide additional infor-
There are a number of analytical electron microscopy techniques mation regarding the local environment of the atomic electrons
that you can use for accurate chemical mapping at the nanometer (such as chemical state).
level. In scanning probe-based techniques, such as energy-disper-
sive X-ray spectroscopy (XEDS) and EELS spectrum-imaging, Hence EFTEM analysis offers a number of advantages over
a focused electron beam is scanned across your specimen’s area both XEDS mapping and EELS spectrum-imaging. However,
of interest, acquiring spectral (and hence chemical) data at each as a result of the complexity of the electron energy-loss spec-
point in order to build up spatial information in a serial manner trum, examples of quantitative EFTEM mapping have, until
(Hunt and Williams 1991). Conversely, EFTEM analysis records relatively recently, been few and far between. Advances in the
energy-selected images using parallel, fixed-beam illumination methods used to capture EFTEM data, and in particular the
in TEM mode, with spectral information integrated over a par- availability of commercial software for the acquisition and
ticular energy-loss range as defined by an ‘energy-selecting slit’. analysis of EFTEM spectrum-image series, now allow quanti-
So, spatial information is recorded in a parallel manner, whilst tative analyses using EFTEM to be performed, as will be dis-
spectral (and hence elemental) information is probed serially by cussed in this chapter.
 13.3 Instrumentation for EFTEM 379

a E b E of modern TEMs (e.g., 80 kV and above) ensure long inelastic

mean free paths, λ, for scattered electrons; the relative sample
thickness is reduced, and therefore the detrimental effect of plu-

Chapter 13
ral scattering is minimized. Of course, key to providing EFTEM
capabilities is the presence of an imaging filter, the main designs
of which are covered in the following section.
x x

y y 13.3.2 The Imaging Filter

EFTEM analysis – STEM analysis –
Image-Spectroscopy Spectrum-Imaging

Energy filtered imaging can be regarded as a five stage process.

Fig. 13.1 Comparison of energy-loss data acquisition using (a) an energy-fil- The TEM itself fulfils the first stage of this process, namely illu-
tered TEM, and (b) a STEM equipped with a parallel detection spectrometer. mination of your sample to provide an unfiltered image or diffrac-
Energy-filtered TEM acquires spatial information in parallel, while sampling en- tion pattern. The imaging filter takes care of the remaining four:
ergy-loss space in a serial manner. Conversely, STEM EELS spectrum-imaging
captures parallel spectra whilst serially stepping the electron probe zz Energy dispersion and focusing
zz Energy-selection
zz Projection
zz Detection

13.3 Instrumentation for EFTEM Almost all electron energy-loss spectrometers and imaging filters
operate on the principle that, in an electric or magnetic field,
electrons may be dispersed according to their kinetic energy,
13.3.1 General TEM Considerations which is related directly to their energy-loss E. Imaging filters
allow a region of the energy-loss spectrum to be selected, using
an energy-selecting slit of width δ, from which the corresponding
Unlike the majority of chemical analysis techniques in modern energy-filtered image is recovered. The energy-filtered electrons
TEM, EFTEM is performed in conventional imaging mode and, are then reformed as an image and projected onto a 2D detector
therefore, does not require STEM capabilities or the ability to (typically CCD) for readout. In the process, corrections are made
form a small probe. Hence, the majority of modern transmission to minimize the aberrations inherent to the process.
electron microscopes are suitable, in principle, for performing
EFTEM studies. Key characteristics are the ability to image with In the main, there are generally two popular designs of imag-
reasonably high resolution, good sample stability (i.e., low spatial ing filter that are commercially available: the post-column and
drift), and with high overall brightness. The objective lens of a in-column (or Omega) type filters, as shown schematically in
modern microscope has a low spherical and chromatic aberration Fig. 13.2. These were described already in Chap. 37 of W&C.
coefficient, ensuring high spatial resolution is available in most
modes of operation. The requirement for high overall brightness Despite the significant differences in filter location and electron
in imaging mode means that, while the high energy-resolution and optics between the various types of imaging filter, the principles
small probe capabilities of field-emission guns (FEG) make these of operation remain the same. In normal EFTEM operation, the
the electron source of choice for dedicated EELS studies, either microscope is operated in TEM mode with close to parallel il-
a FEG or LaB6 electron source is perfectly capable of performing lumination. The energy loss, E, to be imaged is defined by the
EFTEM with good count rates and at high spatial resolution. energy loss of the electrons passing through the energy-selecting
slit. The energy-selecting slit is a mechanical aperture that can
EFTEM Is Imaging be set from software to a specific energy width, typically from
a few eV up to 50 eV, and wider. The angular range of electrons
that contribute to the image, defined as collection semi-angle β,
EFTEM is performed with the microscope in imaging is determined by the objective aperture.
mode (i.e., parallel illumination). Since a small analytical
probe is not required, EFTEM can be used to perform
elemental mapping at the nanoscale even with a LaB6- In EFTEM, the collection semi-angle β is determined by
equipped microscope. 9 the size of your objective aperture. It is essential to ensure
you have an objective aperture inserted when performing
EFTEM to limit blurring from chromatic aberration. The
Further, a significant consideration for any EELS signal-based objective aperture should have a collection semi-angle of
analysis is minimizing sample thickness relative to the inelastic typically 10 mrad. 9
mean free path, t/λ. The relatively high accelerating potentials
 380 13 EFTEM
Chapter 13

Fig. 13.2 Schematic diagrams showing the main filter designs available commercially: the Gatan imaging filter (GIF) post-column filter (a) and the Omega in-column
filter (b)

In practice, when performing EFTEM the energy – loss is se- 50 eV) where sufficient signal is present to allow focusing to be
lected not by changing the imaging filter optics but by changing performed on a ‘live’ CCD camera view.
the microscope operating voltage to E0 + ∆E in order to keep the
optical path constant, hence keeping the imaged electrons in Focussing
focus. Whilst this greatly reduces chromatic blurring, it can have
the undesired effect of subtly altering the illumination conditions
that must be corrected by adjusting the strength of the last con- Always refocus your EFTEM image at an energy-loss us-
denser lens. This ‘condenser correction’ can be calibrated and is ing the CCD camera view. Typically using an energy loss
then automatically corrected for in the acquisition process. Since of 200 eV with 50 eV slit width will give good results.9
there is no high-voltage offset at the filter, the energy-selecting
slit always images electrons of approximately the same kinetic
energy and focus. However, it is often necessary to readjust the
focus via the objective lens at higher energy losses to obtain the
optimal focal setting. As a result of the narrow angular distribu-
tion of elastic intensity, the apparent depth of focus is very large 13.3.3 Detector Considerations
for zero-loss images in comparison to inelastic images. Finer
focusing (and even tweaking of the objective stigmation to get
the best results) should, therefore, be performed on the inelastic Imaging filters are generally equipped with a single charge-cou-
image, even if the elastic image is centrally positioned and at pled device (CCD) camera for data capture. CCD cameras offer
Gaussian focus prior to applying the energy offset. Since the sig- numerous advantages over photographic plates, giving a prac-
nal intensity reduces exponentially as a function of energy-loss, tically linear response over a wide dynamic range, allowing
this slight refocusing is performed typically at a relatively low data to be read out in a digital manner, and offering higher sen-
energy-loss (e.g., 200 eV) with a reasonably large sit width (e.g., sitivity. A summary of how these detectors work can be found
 13.4 Limitations and Artefacts 381

in Chap. 7 of W&C. A number of considerations must be taken cient signal. This sample drift will manifest in a single EFTEM
into account when choosing the CCD camera acquisition pa- image as a spatial blurring in the direction of the sample drift.
rameters for performing EFTEM imaging, such as exposure Drift that occurs during the acquisition of an individual image

Chapter 13
time and pixel ‘binning’. Both saturation of the CCD camera cannot be corrected, so steps should be taken prior to any data
and image degradation through sample drift can set an upper acquisition to minimize the effects of drift. For example, after
limit for the exposure time, though typically saturation isn’t moving the sample or changing the illumination conditions you
too much of an issue when acquiring images at mid to high should allow a settling period for any resulting drift to subside.
energy losses. Conversely, low count levels should be avoided This mechanical instability places an upper limit on the signal
since ‘shot-noise’, resulting from statistical fluctuations in the integration time of the detector, the extent of which should be
number of electrons incident on the detector, can result in noise established by the user prior to data acquisition. However, drift
dominated images. This inherent Poisson noise follows √I (for between consecutive acquisitions may be identified and removed
raw images), where I is the number of incident electrons per by means of a conventional cross-correlation algorithm as de-
pixel. In addition to this, there will be a time-dependent ‘dark- scribed later in Sect. 13.4.3.
noise’ component, and also a non-time-dependent ‘readout’
noise component, both of which would dominate in the ab-
sence of any signal. Clearly it is important for you to ensure
that your image signal exceeds these detector artefacts in order 13.4 Limitations and Artefacts
to be useful.

Typical signal integration times for EFTEM vary from a frac- 13.4.1 Spatial Resolution in EFTEM Images
tion of a second for the relatively intense signal in ‘unfiltered’
or ‘zero-loss’ imaging, to up to a minute or so for much weaker
core-loss imaging. In principle, the lower limit for acquisition In energy-filtered TEM, spatial resolution is often limited by
is generally limited by the speed of the electromagnetic, elec- aberrations of the image forming lenses. Usually it is assumed
trostatic, or even mechanical beam shutter (usually in the range that the main contributions to spherical and chromatic aberration
of tens of milliseconds). For zero- or low-loss imaging, where arise from the objective lens, since the angular divergence of
signals are usually aplenty, CCD acquisition is performed with electrons θ from the optic axis is greatest at this point (Egerton
typically a fraction of a second exposure time with no ‘detector and Crozier 1997). In the following sections, the coefficients
binning’ applied, producing images at the full resolution of the Cc and Cs refer to the chromatic and spherical aberration of the
CCD sensor (typically 2k × 2k pixels). For much less intense objective lens, respectively.
core-loss imaging, the upper exposure time limit is constrained
purely by the loss of resolution that can be tolerated by the drift A number of articles describing the spatial resolution limit in
rate of the sample. Even at relatively long acquisition times EFTEM may be found in the literature. Generally, three main
(i.e., tens of seconds), it may still be necessary to increase the factors are considered to limit the spatial resolution in core-loss
relative sensitivity of the CCD detector by applying ‘detector images (Krivanek et al. 1995).
binning’. Detector binning is a technique that groups together
adjacent pixels to, in effect, create a larger pixel. For example, zz The delocalization (or impact) parameter, brms. Delocaliza-
applying 2× binning groups together 4 adjacent pixels, in the tion reduces the spatial resolution particularly at low energy
process making the detector 4× more sensitive. This, however, losses, ∆E, and high collection angles, β. This may be de-
is at the expense of a factor of 2 loss in pixel resolution. Modern scribed in a simplified form as
EFTEM systems provide both auto-exposure and auto-binning −1=2
determination, which makes it considerably easier for you to brms = „vˇ=E .ˇ 2 + E2 / ln.1 + .ˇ 2 =E2 // (13.1)


determine the optimal parameters for your EFTEM image ac-

quisition. where E0 is the incident beam energy, and θE is the characteristic
angle for inelastic scattering as described in Chap. 38 of W&C.
Generally, short exposure times and no detector binning zz The chromatic aberration from the objective lens, dc. This
are appropriate for acquiring EFTEM images containing can be measured through the coefficient Cc, arising from the
the zero-loss signal. For core-loss images, it will be nec- energy spread of electrons admitted by the energy-selecting
essary for you to increase the exposure time and also the slit of width δ, and can be approximated to
detector binning to ensure adequate SNR. 9
dc = .Cc ı=E0 /ˇ (13.2)

In any TEM acquisition, sample drift will occur as a result of zz At very small collection semi-angles, the spatial resolution is
instabilities of the microscope or specimen. This is particularly dominated by the aperture diffraction limit
problematic for acquiring EFTEM core-loss images since rela-
tively long acquisition times often need to be used to get suffi- dd = 0:6=ˇ (13.3)
 382 13 EFTEM

Fig. 13.3 Spatial resolution plots for

EFTEM imaging plotted as a function
of objective aperture semi-angle β,
Chapter 13

calculated using the approach of

Krivanek. a Theoretical spatial res-
olution, with chromatic aberration,
delocalization and aperture diffrac-
tion limit contributions as indicated
(E0 = 400 keV, Cc = 2.0 mm, δ = 10 eV,
∆E = 500 eV). The spatial resolution
will be affected by changes in (b)
accelerating voltage E0, (c) energy
loss E (delocalization increases with
decreasing energy-loss), and (d) en-
ergy-slit width δ (chromatic aberra-
tion increases with slit width)

Spherical aberration arising from the objective lens is considered duced count rates. For example, by choosing a smaller energy-se-
to add an essentially constant background in this approach, and lecting slit or objective aperture the signal contributing to the
is assumed to be negligible under optimized imaging parameters image is reduced. Likewise, ionization cross-sections are lower
for minimum chromatic aberration. The individual contributions for both high energy-loss edges and for increased incident beam
may be added in quadrature to give the total theoretical spatial energies, so again the gain in resolution may be outweighed by
resolution, dtotal the resultant decrease in signal. Optimization of these parame-
ters in a quantitative manner is difficult to implement as other
factors, such as sample drift and beam damage, also play an
q
dtotal = .2 brms /2 + dc2 + dd2 (13.4)
influential role. However, calculation of such spatial resolution
plots for the edges of interest gives a good qualitative guide of
Theoretical resolution plots, showing the effects of varying each the optimal acquisition parameters, in particular allowing the
parameter within this approach, are shown in Fig. 13.3. As you most suitable objective aperture size to be determined for the
can see, higher spatial resolution in EFTEM may be achieved acquisition conditions.
for increased incident beam energies. Also, ionization edges at
higher energy loss give a more localized signal than their low-
loss counterparts and, hence, offer improved spatial resolution. Increasing the slit width and/or the collection (objective)
Of specific importance to EFTEM, decreasing the width of the aperture angle increases the SNR, but decreases the spa-
energy-selecting slit and choosing an objective aperture of an tial resolution. Hence, you must balance the need for
appropriate (and generally smaller) semi-angle further improves good signal versus high resolution. 9
spatial resolution. However, these factors will also reduce the
amount of signal contributing to your EFTEM images and so
the improvement in spatial resolution must be carefully balanced One thing to consider when you’re selecting the optimal ob-
against an increase in statistical uncertainty resulting from re- jective aperture size for performing EFTEM acquisition at a
 13.4 Limitations and Artefacts 383

core-loss edge is ensuring the majority of the available edge misleading results. Approaches exist for the removal of this
intensity is collected in order to optimize SNR. The characteris- effect from acquired data, which can be useful for performing
tic scattering angle θE is approximately proportional to the edge high-resolution EFTEM experiments with a very small slit width

Chapter 13
energy loss and represents the mean inelastic angle for a given (Schaffer et al. 2006). However, most modern imaging filters
energy-loss. As described in Chap. 37 of W&C, you should use will automatically measure and correct for these aberrations up
a collection semi-angle β of at least 2-3θE in order to collect the to several orders, and a correctly aligned filter is more than ade-
majority of the available EELS edge signal. Since β is deter- quate for your general EFTEM work.
mined in EFTEM by the objective aperture, it is necessary to
balance the aperture size required for optimal spatial resolution
with that for optimizing the SNR by capturing the majority of the
signal. In practice, the two values are generally commensurate, 13.4.3 Sample Drift
and the greater of the two can be used – an objective aperture
with approximately 10 mrad semi-angle generally satisfies both
criteria and provides good results for the majority of use cases. As a result of instabilities of the microscope or specimen, sample
drift may occur during the acquisition of a series of EFTEM
A final consideration when acquiring EFTEM data at the max- images. Before data acquisition, steps may be taken to minimize
imum attainable spatial resolution is to ensure an appropriate the effects of drift. For example, after moving the sample or
magnification is selected at the microscope. As a rule of thumb, changing the illumination conditions, a settling period should
the spatial sampling per pixel on the CCD camera should be be allowed for any resulting drift to subside.
approximately three times higher than the expected theoretical
resolution. Working at a higher magnification than this will result Drift that occurs during the acquisition of an individual image
in over-sampling (and unnecessarily noisy data), whilst working cannot be corrected, and the result will be the introduction of
at a lower magnification will give an attained spatial resolution a directional blurring in the final image. This mechanical in-
that is limited by the pixel sampling. stability places an upper limit on the signal integration time of
the detector, which in turn can limit the SNR of the captured
image. You should establish the extent of the spatial drift prior
to data acquisition, and select an exposure time such that the
13.4.2 Non-Isochromaticity drift during the exposure does not exceed the spatial sampling
of your detector.

The non-isochromaticity of an imaging filter is a measure of the Cumulative drift between consecutive acquisitions can be
difference in energy at a point in an energy-filtered image with identified and removed after acquisition. This is desirable
respect to the center of the image (i.e., the optic axis). From if the one or more acquired images are to be used in some
a practical perspective, it determines the minimum slit width subsequent calculations that assume the images are spatially
that can be used for energy filtered imaging before chromaticity aligned; for example, in computation of elemental maps. The
artefacts become an issue. It is generally measured using auto- spatial alignment of images is usually automated by means of a
mated tuning software by scanning the zero-loss peak over the cross-correlation algorithm, which is provided as standard with
energy-selecting slit, and recording the intensity changes at each commercial software. A problem remains in that the intensity
energy-loss on the CCD camera. The resulting ‘chromatic figure’ distribution of core-loss images can differ to a large extent,
is then computed from this scan as a two-dimensional image and especially when comparing images located before and after an
is a measure of the relative chromatic shift at each point in the ionization edge which can show a reversal of image contrast.
field of view. Typically, this figure is used to tune the imaging Such differences in contrast can affect severely the cross-cor-
filter at the start of a microscope session in order to minimize the relation procedure.
energy displacement over the CCD image area (see Fig. 13.4).

Sample drift during an EFTEM image acquisition will

To Begin
result in blurring in the direction of the sample’s motion.
Drift accumulating between consecutive EFTEM acqui-
At the start of your EFTEM session, run the imaging-fil- sitions can (and should) be corrected for prior to any sub-
ter system’s automatic tuning software to ensure aberra- sequent processing (e.g., summing or map computations).
tions are minimized for optimal performance. 9  9

However, even a well-tuned filter will have a degree of non-iso- This problem may be partially overcome by pre-processing the
chromaticity that will, in practice, limit the minimum slit width images using an edge-enhancement filter (e.g., a Sobel filter).
that can be used without introducing noticeable artefacts into Generally, when aligning more than two images, a single refer-
the acquired data. Isochromaticity artefacts typically manifest ence image is selected that most strongly represents the average
as changes in image intensity at the corners, which can yield intensity distribution of all the images, to which each image in

Licensed to ([email protected])
 384 13 EFTEM

Fig. 13.4 The measured non-iso-

chromaticity figures for a Gatan
Imaging Filter before (a) and after
Chapter 13

(b) alignment (GIF model Quantum

ER). Corresponding plasmon images
of Al3Li precipitates in an Al matrix
acquired using the uncorrected filter
(c) and the corrected filter (d). The
precipitates have a plasmon energy
of 13.5 eV, which is lower than that
of the aluminum matrix (15 eV), and
hence appear bright at the imaged
energy-loss. The aberration artefacts
highlighted in the corners of the im-
age are typical for a badly aligned fil-
ter and indicate that the filter needs
tuning before data acquisition

the sequence is then aligned. However, automated alignment range ∆E as the energy-filtered image. This correction accounts
routines can fail in some instances, at which point the user must partially for elastic scattering intercepted by the angle-limiting
revert to manually aligning the images. This can be achieved to a aperture. However, as shown by Hofer et al. (1997), this method
surprisingly high accuracy by using a rapidly alternating overlaid can introduce serious artefacts for crystalline specimens. Other
display of the images to be aligned, allowing you to manually techniques exist for the removal of diffraction contrast in the case
shift the images into spatial register with single pixel accuracy of elemental maps, as will be discussed later.
or better. Such routines are built into commercial software for
performing EFTEM; for example, in Gatan’s EFTEM suite for
DigitalMicrograph.
13.4.5 Illumination Convergence

13.4.4 Diffraction Contrast As mentioned above, when performing EFTEM an offset is ap-
plied to the microscope’s high tension, and not on the imaging
filter’s prism itself. This is to ensure that the energy-loss elec-
Images acquired using energy-filtered TEM suffer from the ef- trons that are being imaged are always at the same energy as the
fects of diffraction contrast in the same way as images acquired microscope was aligned at, hence keeping the energy-loss image
using conventional bright-field imaging. Accordingly, conven- at the correct focus. While this ensures that the post-specimen
tional approaches for avoiding or reducing the effects of diffrac- electrons that are subsequently transmitted through the energy-se-
tion contrast are equally well suited to EFTEM data acquisition lected slit being maintained at the at nominal microscope voltage,
(for example, you should tilt the sample away from strongly it also results in the pre-specimen electrons having a higher than
diffracting orientations, and effects can be reduced further by nominal energy E0 + ∆E. This has the undesirable side effect of
using rocking-beam illumination). In instances where diffraction subtly changing both the illumination convergence and, to a lesser
cannot be avoided, the most common way to minimize these un- degree, the illumination position on the sample. This effect can
wanted effects is to perform a normalization by division of a low- be quite significant, particularly over large energy losses, and
loss image, containing the zero-loss and possibly plasmon-loss typically manifests as an encroachment of the edge of the illu-
intensity, acquired using the same energy-slit width δ and angular mination disk into the EFTEM image at higher energy losses.
 13.5 Application of EFTEM 385

This effect can be corrected for using a so-called ‘condenser cor- a b

rection’. This needs to be pre-calibrated by you, the user, using
the EFTEM acquisition software’s in-built correction routine.

Chapter 13
This routine will measure the required offset to be applied to
both the condenser lens and the beam shifts in order to keep the
illuminated area constant as a function of applied energy loss.
Once measured, these calibrated values are stored and applied
whenever an offset is applied to the microscope high tension
during EFTEM. This correction isn’t so important when per-
forming simple energy-loss imaging, or even when acquiring a
single EFTEM elemental map, but it is important to apply when
performing EFTEM experiments spanning typically hundreds
of eV energy loss (for example, EFTEM-SI or elemental maps
of multiple elements).

Caution unfiltered zero loss

c
If you notice the edge of the illumination disk encroach silicon {110} slice
on your image during EFTEM imaging, particularly at
higher energy losses, then the illuminated area is most
likely changing as the HT offset changes. You should Si Si
check that your EFTEM condenser correction is properly
configured. 9
Si Si

13.5 Application of EFTEM

Fig. 13.5 A comparison of convergent beam electron diffraction (CBED) pat-
terns recorded at the <110> zone axis of a diamond sample. a The pattern has
It is important to keep in mind that EFTEM is, quite literally, not been energy filtered and the contrast is very weak. b The effect of filtering
parallel imaging a discrete portion of the EELS spectrum from out all the inelastically-scattered electrons to form a zero loss CBED pattern.
your specimen. As a result, the majority of features or proper- Such patterns can be used to determine structure factors to very high accuracy
ties that can be measured from a regular EELS spectrum can and to allow the bonding charge density to be revealed, as shown in (c) for a
be imaged using EFTEM, albeit with significantly less spectral {110} slice through the silicon structure. The yellow peak shows the covalent
information. The most commonly used EFTEM applications are bond between the silicon atoms
described in this section.

The great advantage in removing the inelastic scattering is to

improve image contrast by reducing the blurring effect of chro-
13.5.1 Zero-Loss Imaging and Diffraction matic aberration. Since the chromatic aberration component
arises from the inelastically scattered electrons, the benefit is
more pronounced for specimens of higher thickness or atomic
One of the simplest, yet perhaps most important uses of EFTEM number. Further, the benefits of the removal of chromatic blur-
is simply to remove the vast majority of inelastic scattering and ring by zero-loss filtering are not restricted to use in imaging
form a 'zero-loss' image. Zero-loss imaging is covered exten- mode. An extreme example of the reduction of this aberration
sively in Chap. 38 of W&C, and so will not be covered here is shown in Fig. 13.5 where an unfiltered convergent beam
in detail. To summarize, zero-loss imaging involves placing an electron diffraction pattern from a diamond specimen about
energy-selecting slit symmetrically about the zero loss peak in 1 μm thick is compared with the same pattern taken with a
the energy loss spectrum to form a zero-loss (or elastic) image. 10 eV slit centered around the zero loss peak. Such filtered pat-
The choice of slit width depends very much on the experiment, terns can be compared with simulated patterns using dynam-
but is typically 5–10 eV. It is possible to use a smaller slit (for ical diffraction theory and are so sensitive that small (< 5 %)
example, 1–2 eV) but a single image using too narrow an energy changes to the structure factors of low-order reflections can be
window will generally suffer from non-isochromaticity artefacts, used to determine the bonding charge density (Midgley and
i.e., a variation in the energy loss across the field of view, as Saunders 1996).
described in Sect. 13.4.2.
 386 13 EFTEM

13.5.2 Measuring Relative Thickness can be obtained by simply a recording a low-loss spectrum and
(t/λ Mapping) separating the zero-loss peak from the inelastic signal, a process
that is fully automated in commercial software. The t/λ value
Chapter 13

provides an ideal measure of the suitability for the local sample

Knowledge of the local sample thickness is required for a num- area for performing EELS studies since its value provides a
ber of reasons – for example, to evaluate if a sample region is direct indication of the degree of signal degradation by plural
a suitable thickness for performing EELS-based studies, or to scattering independent of the sample of the sample composition.
convert areal densities measured by EELS or EFTEM to true
concentrations. This may be found using conventional analytical The same approach can be performed in EFTEM to generate a
TEM techniques, for example by analysis of CBED diffraction two-dimensional t/λ map. In the two-window log-ratio method
patterns or using the thickness fringes from a crystal wedge sam- for computing a thickness map, an unfiltered image and a ze-
ple. However, these techniques are generally time consuming ro-loss image are acquired in succession (and usually in that
and may require a specific type of specimen (e.g., crystalline). order). The t/λ map is then computed by taking the logarithm of
Fortunately, the energy loss spectrum can provide a convenient the ratio of unfiltered to zero-loss images. The zero-loss signal
estimate of thickness from both crystalline and amorphous ma- is recorded as a zero-loss image, using an energy-selecting slit to
terials in terms of the total inelastic mean free path by straight- exclude the inelastically scattered signal, allowing only zero-loss
forward integration of the energy loss signal. electrons to pass through. The unfiltered image is recorded by
acquiring an image under identical conditions, except with the
Computation of the relative sample thickness can be performed energy-selecting slit retracted, allowing both elastic and inelastic
very easily using the EELS low-loss spectrum; please refer to electrons to fall on the detector.
Chap. 38 of W&C for details. EELS thickness measurement uti-
lizes the probability of plural scattering, which follows Poisson An example of t/λ mapping is shown in Fig. 13.6 for a com-
statistics to give the ratio of the specimen thickness t to the total mercial semiconductor device. As can be seen, the t/λ value
inelastic mean free path λ as varies significantly for various components of the device. Since
  the inelastic mean free path λ decreases for materials of higher
t It atomic number, the t/λ map contains a combination of thickness
= ln (13.5)
 I0 and compositional information, and hence you should interpret
it with some caution. However, t/λ mapping is an invaluable
where It is the total integrated spectral intensity and I0 is the tool, particularly at low magnifications, for determining which
zero-loss integral. The inelastic mean free path λ varies with regions of the sample are suitable for subsequent EFTEM or
both sample composition and beam energy, and represents the EELS investigations. When performing EFTEM analysis, it is
average distance the electrons travels through the sample be- a prudent first step to acquire a t/λ map from your region of in-
tween inelastic scattering events. This method is used routinely terest to ensure that the local relative thickness is thin enough to
in EELS studies to obtain a relative measure of the local sample perform further EFTEM analysis. Also, you get a zero-loss and
thickness, and is often the first property measured before com- unfiltered image acquired as part of the thickness measurement,
mencing with core-loss spectroscopy. In EELS, the t/λ value which are often of use anyhow.

Fig. 13.6 An example of t/λ mapping using EFTEM (commercial semiconductor sample). Unfiltered (a) and zero-loss (b) images are acquired, and used to compute
the corresponding t/λ map (c). The t/λ map show the sample region is approximately 1.2 inelastic mean free paths thick with significant variation over the field of
view arising mainly from compositional changes (slit width ∆ = 10 eV, 0.5 s exposure)
 13.6 Core-Loss Elemental Mapping 387

Acquire a t/λ Map ground contribution Ib corresponding to a variety of inelastic

events occurring at lower energy losses. Therefore, to obtain
an elemental distribution map from a core-loss image, the spec-

Chapter 13
Do acquire a t/λ map to confirm that the local relative tral background contribution must be taken into consideration.
sample thickness is suitable for subsequent EFTEM mea- This key consideration when performing EFTEM mapping is, of
surements. Don’t assume an increase in t/λ necessarily course, an essential requirement for elemental analysis in EELS
means an increase in thickness – it could also arise from also. This is covered in detail in Chap. 39 of W&C. In EELS
a local increase in average atomic number. 9 spectroscopy, the acquired spectrum generally contains sufficient
spectral detail to allow accurate fitting of a background model
using a pre-edge region. However, in EFTEM the spectral infor-
The log-ratio approach for generating t/λ maps is both simple mation can only be sampled serially (that is, image by image),
to implement and rapid, requiring only two images and modest making background fitting and its subsequent removal consid-
computational power for online analysis. However, chromatic erably more challenging.
blurring as a result of the large energy-loss range of the elec-
trons contributing to the unfiltered image will degrade spatial Although a number of routines have been proposed to remove
resolution of the computed map, an effect that worsens with the influence of the background contribution in EFTEM map-
increasing specimen thickness. Generally, this is not an issue ping, two simplistic methods in particular have been shown to
since the t/λ map is only used to survey whether the local sam- give routinely satisfactory results, using a minimum of ener-
ple thickness is suitable for subsequent EFTEM acquisitions. A gy-filtered images and without the need for reference spectra.
preferable method for acquiring t/λ maps without losing spatial These methods are the three-window technique for elemental
resolution utilizes the EFTEM spectrum-imaging technique as mapping, which may be regarded as quantitative in nature, and
described later in this chapter. By acquiring a continuous series also the qualitative two-window jump-ratio approach. More re-
of images from zero-loss up to an appropriately high energy cently, the EFTEM spectrum-imaging approach, in which an ex-
(about 200 eV loss is more than sufficient), the t/λ value may tended series of energy-filtered images is acquired, has become
be calculated on a pixel by pixel basis, using the log-ratio pro- an increasingly popular method for capturing the information to
cedure for EELS described in Chap. 38 of W&C. Typically, calculate one or more core-loss maps from a single image series
a slit width, δ, and energy-step, ∆E, of 10 eV are used, with a with significantly improved spectral detail.
total acquisition time of approximately 1 min. This method is
computationally more demanding (and also has a much higher
total electron dose!), but it yields t/λ maps with significantly
lower chromatic blurring. 13.6.1 Elemental Mapping
(Three-Window Method)
To obtain an absolute measurement of the sample thickness, the
inelastic mean free path must be determined independently. An
expression has been derived which allows the parameterized In EFTEM the simplest approach for isolating the core-loss
inelastic mean free path to be calculated using a mean energy signal from the spectral background is to characterize the pre-
loss Em, which in turn can be approximated from the effective edge region by acquiring a minimum number of images, and
atomic number of the sample under investigation. In practice, to capture the core-loss signal in one image positioned at or
however, the calculation of an absolute thickness map is a dif- beyond the core-loss edge threshold. The three-window method
ficult problem, as thickness cannot be estimated using parame- of elemental mapping, as proposed originally by Jeanguillaume
terized inelastic mean free paths without knowing the absolute et al. (1978), does just this by requiring two ‘pre-edge’ images
composition on a pixel-by-pixel basis (which is the quantity such and a post-edge image. The problem here is that there are only
investigations often aim to determine). two energy-loss intensity values per pixel with which to model
the background. While this would be trivial for, e.g., a linearly
extrapolated background model, it is not so straightforward for
the more accurate power-law background model I(E) = AE−r
13.6 Core-Loss Elemental Mapping generally favored in EELS analysis (where r is the background
slope exponent and A is a scaling term). Hence the limited spec-
tral information contained in a simple three-window EFTEM
One of the most popular uses of EFTEM is for the rapid ac- experiment means that an alternative background fitting method
quisition of elemental distribution maps using core-loss edges. is required.
By the use of EFTEM, an electron spectroscopic image may
be acquired at, or beyond, an ionization-edge absorption energy The approach which is now generally used simply divides the
corresponding to a specific elemental species. Such a post-edge background-fitting region into two sections of equal width and
image contains information regarding the spatial distribution allows the pre-edge regions to be placed apart; the only require-
and concentration of the chosen element in the form of the in- ment being the regions must be of equal width (Berger and Kohl
tegrated edge intensity Ik, with an additional underlying back- 1993). By applying least-squares fitting, the value of the back-
 388 13 EFTEM

J(E) the value of the background exponent r decreases monotonically

with increasing energy loss, limiting the range of validity of the
power-law model. Hence a compromise between systematic and
Chapter 13

statistical errors must be made when deciding on the pre-edge

window positions; commercial software for EFTEM acquisition
I1 will provide reasonable default values for each core-loss edge. In
I2 AE-r practice, the choice of pre-edge fitting region is often restricted
by preceding edges, which must be avoided in order to avoid
introducing edge-overlap artefacts into the computed map.
E1 E2 energy-loss
The optimal position and width for the post-edge energy win-
dow will be where the signal-to-noise ratio is maximized. This
Fig. 13.7 Background extrapolation using the two-window method for back- depends on the shape of the edge, and, for example, with delayed
ground modelling. The background-fitting region consists of two regions of edges the optimum position could be 10’s of eV higher than the
equal energy width edge onset. Again, commercial software will generally provide
a reasonable default window position.

ground slope exponent r and A may then be obtained by applying

the approximate formulae Three-window EFTEM maps can be considered ‘quan-
titative’ since the background has been properly mod-
elled and removed. Keep in mind, though, that the actual
   
I1 E1
r = ln
I2
= ln
E2 (13.6) mapped quantity is the integrated edge intensity, Ik, which
can be affected by both sample thickness and diffraction
and contrast. 9
    
1 I1 E1
ln A = ln + ln (13.7)
2 I2 E2 An example of an elemental map acquired using the three-win-
dow approach is shown in Fig. 13.8. Here, an O K map is cal-
where I1 and I2 are the intensities at each pixel in the pre-edge culated to show the oxygen distribution in the same device as
images 1and 2, and E1 and E2 are the energies corresponding to the used in the previous section. The three-window method produces
centre of the energy window of each image, as defined in Fig. 13.7. images that may be considered as true elemental maps, with the
The corresponding background contribution Ib is then found by in- edge intensity Ik integrated over the post-edge window width
tegration of I(E) = AE−r over the signal integration range ∆. Hence measured in each pixel being related to the areal density of the
the power-law background contribution may be calculated using element analyzed, as described by Eq. 39.6 in W&C. However,
the two pre-edge windows but since the background contribution as sample thickness and diffraction contrast will vary across the
will depend on many locally varying factors, such as composi- elemental map, these effects should be taken into account when
tion, orientation and specimen thickness, the background fitting evaluating relative map intensities.
parameters must be calculated for each individual pixel within
the image. As a result an A and r map are generated, allowing the In principle, absolute quantification of a three-window map may
background contribution corresponding to the energy-range of the be performed by further dividing by the partial inelastic scatter-
post-edge image to be calculated. Subtraction of this background ing cross-section and zero (or low)-loss intensity, yielding a map
contribution from the post-edge image leaves the core-loss signal that reflects the elemental concentration. In general, however,
remaining in the form of an elemental map. the three-window mapping technique is used to obtain elemen-
tal distribution maps to be used for a qualitative evaluation of
The position of the first pre-edge window should be sufficiently chemical distribution.
distant from any preceding edges so that they will not affect
the background fit, i.e., the background may be fitted using a
power-law model. The presence of any preceding edge intensity
within the background fitting region will have the detrimental 13.6.2 Jump-Ratio Mapping (Two-Window
effect of introducing a systematic error into the background ex- Method)
trapolation process. The position of the second pre-edge window
should then be placed as close to the edge threshold as possible
whilst avoiding the edge onset, since this region has the greatest An alternative method for obtaining elemental distribution maps
influence on the accuracy of the extrapolation procedure. To max- is to divide the post-edge image by the corresponding pre-edge
imize the accuracy of the background fit, one might think that the image to give a jump-ratio image. This procedure has been
separation of the two windows should be as large as possible so shown by Hofer et al. (1995) to yield maps with enhanced con-
as to minimize statistical error. However, this is not the case since trast that are sensitive to low concentrations with minimum addi-
 13.7 EFTEM Spectrum-Imaging 389

a b However, keep in mind that maps obtained in this way are not
O K Pre-edge 1 (484±15eV) O K Pre-edge 2 (514±15eV) quantitative; instead they are only indicative of elemental distri-
bution. Further, jump-ratio images are particularly susceptible to

Chapter 13
artefacts as a result of their sensitivity to changes in the preceding
background. For example, modulations in the pre-edge intensity
arising from variations in the strength of a preceding ionization
O K Map
edge will manifest themselves as a change in the jump-ratio,
regardless of whether there is any variation in the concentration
of the element of interest. Therefore, it is widely recommended
to compute jump-ratio images in addition to quantitative elemen-
200 nm tal maps, allowing a direct comparison to be made to confirm
if features are present in both maps. Fortunately, a jump-ratio
c d
map can be computed from the post edge and either pre-edge
O K Post-edge (547±15eV) image acquired when performing three-window mapping. Fur-
ther, forming a ratio map of two pre-edge images may give an
additional check for artefacts; any common features between the
two ratio maps may be dismissed as artifacts.

Jump-ratio imaging is a simple way to improve the SNR

of computed elemental maps, particularly in biological
specimens. It is recommended, however, to acquire a
three-window map to allow the background to be properly
accounted for – you can still compute a jump-ratio map in
addition from the three-window map source images. 9
Fig. 13.8 Example of elemental mapping of oxygen in a commercial semi-con-
ductor device using the three-window technique. Two ‘pre-edge’ images (a, b),
acquired before the O K edge at 532 eV, are used to model and remove the back- An example of jump-ratio mapping is shown in Fig. 13.9. The
ground contribution to a ‘post-edge’ image. c, d The resulting map contains only jump-ratio image in this example agrees reasonably well with
the O K-edge signal, which is directly related to the projected oxygen atomic the corresponding three-window map (shown in Fig. 13.8) and
concentration (slit width ∆ = 30 eV, 11 s exposure) is arguably less noisy. However, on closer inspection differences
can be observed (e.g., intensity in the tungsten plugs, whereas
the three-window map shows none), and only qualitative mea-
tional noise. In general, images produced in this way are less af- surements should be taken from this type of map.
fected by diffraction contrast and thickness variations compared
with their three-window counterparts, with diffraction contrast
eliminated almost completely by the additional use of conical or
rocking-beam illumination. Also, jump-ratio images provide a 13.7 EFTEM Spectrum-Imaging
convenient means for obtaining elemental distribution maps in
situations where the pre-edge background is difficult to model,
for example in the case of overlapping edges. The simple two- or three-window methods described above
provide a fast and very efficient tool for rapid analytical char-
acterization using EFTEM. However, such techniques provide

Fig. 13.9 Jump-ratio imaging of a b c

the O K edge at 532 eV loss using O K Pre-edge (514±15eV) O K Post-edge (547±15eV) O K Rao Map
the same source data as Figure 13.8.
Dividing a ‘post-edge’ image ac-
quired after the ionization threshold
(b) with a ‘pre-edge’ image (a) yields
the O jump-ratio map (c), (∆ = 30 eV,
11 s exposure)

200 nm
 390 13 EFTEM

Fig. 13.10 Analysis of a core-loss a b

edge using an extended image-series. e s
ag J(E) N data channels
N images are acquired (a), allowing im
Chapter 13

N
individual ‘image spectra’ to be ex-
tracted, after correction for sample x Is
drift during acquisition (b)

Ib
E

AE --r
Extract image-
spectrum

y Em En
energy loss

very limited spectral information, prohibiting the use of stan- zz Thirdly, since a number of post-edge images are acquired,
dard spectrum processing techniques, such as deconvolution each with typically a narrower energy slit width ∆ than used
of plural scattering and separation of overlapping edges using in the three-window approach, the core-loss signal can be
multiple linear least squares (MLLS) fitting. Further, such tech- integrated over a relatively wide energy loss range without
niques require the user to have a priori knowledge of which the introduction of excessive chromatic blurring since this is
species are present and of the optimal window widths and dependent only on the individual slit width applied.
placement, and provide no further means to examine for un-
suspected elements or confirming the validity of an unexpected EFTEM spectrum-imaging is shown schematically in
feature in a map. Fig. 13.10. The number of spectral channels is determined by
the number of energy-steps performed, while the number of
pixels is dependent on the CCD readout size. Typically, the
Simple two- or three-window EFTEM mapping provides energy-selecting slit width and energy-step size are set to be
no useful spectral information. Hence, its use is limited to equal but this is not essential. The image series is acquired se-
somewhat ‘known’ samples and should be supplemented quentially with a short HT settling delay between energy steps.
with point EELS or XEDS spectroscopy where unknowns Acquisition is typically from low energy loss to high energy
are suspected. 9 loss, although the reverse direction may be preferential to first
capture the weak signal and hence limit the influence of residual
detector signal from scintillator afterglow. The more advanced
A more advanced approach extends the acquisition of just two or ‘Smart EFTEM-SI’ acquisition method as proposed by Wata-
three images to the acquisition of a continuous series of images nabe and Allen (2012) further alleviates the residual effects of
across the energy range of interest. This technique, first developed detector afterglow in acquired EFTEM spectrum images by
by Lavergne et al. (1992) for the analysis of biological specimens, refreshing the dark reference prior to each image captured in
is sometimes referred to in the literature as ‘image-spectroscopy’ the series.
by analogy to ‘spectrum-imaging’ in STEM. The technique is
now more generally known as ‘EFTEM spectrum-imaging’ (EFT-
EM-SI). Some main advantages of this approach in comparison EFTEM-SI provides the best of both worlds, allowing a
to the simple three-window technique are as follows. large number of pixels to be imaged rapidly whilst pro-
viding spectral information for each pixel. The overall
zz Firstly, the total number of CCD counts in an extended sample dose can be high, though, and spectral resolution
EFTEM series can be considerably greater than for three-win- is typically limited to 5–10 eV. 9
dow acquisitions. Hence, the signal-to-statistical-noise con-
tent is improved, which is advantageous for jump-ratio imag-
ing and, in part, for elemental mapping. Once the EFTEM-SI series has been acquired, you must take
zz Secondly, and far more importantly, the extent of spectral care to accurately remove the sample’s spatial drift throughout
information is increased dramatically. This increased spectral the series, which is usually performed by cross-correlating each
information allows each pixel through the image-series to be image plane with a reference plane and shifting each image in
treated as an individual EELS spectrum, since in principle the stack accordingly. Schaffer et al. (2004) have shown that
both EFTEM and EELS probe the same three-dimensional improved accuracy for spatial drift correction can be obtained
data space, I(E, x, y). This allows signal extraction parameters by aligning all images in the stack with each other, and applying
(i.e., window positions) to be optimized post acquisition, and statistical analysis to the subsequent drift matrix to determine the
also enables examination for spectral features not assumed to overall drift, a technique termed statistically determined spatial
be present at the time of acquisition. drift (SDSD) correction. For experiments requiring a high degree
 13.7 EFTEM Spectrum-Imaging 391

850eV 860eV 870eV 880eV 890eV 900eV 910eV

Chapter 13
920eV 930eV 940eV 950eV 960eV 970eV 980eV

990eV 1000eV 1010eV 1020eV 1030eV 1040eV 1050eV

1060eV 1070eV 1080eV 1090eV 1100eV 1110eV 1120eV

Fig. 13.11 A montage showing an EFTEM image-series taken from a grain boundary in an AlZnMgCu high strength alloy, illustrating the change in feature intensity
with increasing energy-loss, corresponding to regions of differing chemical composition. Contrast changes occur at the Cu and Zn L edges, at 932 eV and 1020 eV
loss, respectively

of energy alignment, such as low-loss analyses performed with corresponding core-loss energy; for example, there is copper
a small energy slit, it may also be necessary to correct for the segregation at the grain boundary which becomes apparent
effects of non-isochromaticity (Schaffer et al. 2006). DigitalMi- beyond the Cu L23 threshold energy at 932 eV loss. You can
crograph scripts are readily available for doing both of these confirm the presence of an ionization edge by inspection of the
corrections. extracted spectrum at that point. Once the edge is identified, the
optimal background removal and signal integration parameters
Once the series is acquired and aligned, the spectral informa- can be determined for the extraction of the core loss signal as
tion can be extracted for each pixel in the field of view. Each a 2D map. Since the filter-slit width determines the chromatic
‘image-spectrum’ contains energy-loss information with the blurring contribution to the acquired image, the core-loss signal
same spectral resolution as the acquisition slit width or step integration window can be integrated over many 10’s of eV loss
size (whichever is larger). EFTEM-SI spectra typically have far while still maintaining optimal spatial resolution. Further, the
fewer spectral channels (and accordingly lower spectral resolu- increase in spectral information with the two or three-window
tion) than their EELS counterparts, but nonetheless generally approaches not only results in an improvement in quantitative
enough for you to make out spectral features. The spectrum evaluation, but also increases confidence that any feature ob-
may span over multiple ionization edges, allowing analysis of served is, in fact, really attributable to a core-loss edge and al-
a number of elemental species within a single acquisition. The lows the identification of core-loss edges after data acquisition.
method is also equally well suited for data acquisition over the This is a significant advantage over three-window mapping,
low-loss region, where the relatively high intensity allows rapid where often the user is acquiring ‘blind’ of what spectral fea-
acquisition using a small slit width and energy step size, yielding tures are really present.
image series with surprisingly high energy-resolution (1–2 eV
is typical). The Cu L23 elemental map as computed from the full image-spec-
trum series is shown in Fig. 13.12. You can use the exact same
An EFTEM image-spectrum series for an aluminum alloy is tools and procedure for generating maps from EFTEM-SI series
shown in Fig. 13.11. As you can see, the appearance of par- as you would for regular STEM-EELS spectrum-images – for
ticular features can be observed by visual inspection at the example, by regular background extrapolation and signal inte-
 392 13 EFTEM

Table 13.1 A comparison of the merits of EFTEM-SI and STEM-EELS SI

EFTEM-SI STEM EELS-SI

Chapter 13

Parallel mapping of energy-loss High-speed parallel acquisition of

plane for a narrow energy range energy-loss data for a single pixel
60–4k energy planes/hour 3k–5M pixels/hour
65k–4M pixels/plane 1k–2k energy channels/pixel

Serial acquisition of energy-loss Serial acquisition of pixels

planes Minimizes specimen dose
High overall dose Energy resolution of 0.1–1.5 eV
Resolution of 1–20 eV typical Instrumentation artefacts fully
Instrumentation artefacts partially correctable
correctable Highest quality spectra
Low quality spectra
Fig. 13.12 Copper distribution maps, calculated from an EFTEM image-series
Spatial Resolution Spatial Resolution
using the L23 edge at 931 eV threshold energy. a Copper elemental map, and 200 kV FEG: 0.2–1 nm Cs-corrected FEG: 0.07–0.15 nm
(b) jump-ratio map 200 kV LaB6: 0.5–1 nm 200 kV FEG: 0.2–1 nm
120 kV LaB6: 1–2 nm 200 kV LaB6: 1–5 nm
120 kV LaB6: 3–10 nm
gration, or using more advanced means such as MLLS fitting.
Generally, maps computed from an EFTEM-SI series show
higher SNR and also better spatial resolution than maps acquired
using the three-window approach. In practice, a larger slit width The information in the low-loss spectrum can, of course,
and signal integration time is typically used for three-window be used for EFTEM imaging, albeit with relatively coarse
mapping in order to increase the signal levels, which can result spectral resolution. By using a narrow energy-selecting slit
in a reduction in the overall spatial resolution through chromatic (typically 1–2 eV in width) placed at the plasmon energy
blur and sample drift. for a known phase, a single EFTEM image can often suf-
fice to show the distribution of that material. For example,
Finally, it worth considering the merits of EFTEM-SI in con- both aluminum and boron have well-defined plasmon ener-
trast with its scanning mode counterpart, STEM EELS-SI. A gies (at 15.0 and 22.7 eV loss, respectively). The example
comparison of the two techniques is given in Table 13.1. While in Fig. 13.13 shows how the Al and B phases in aluminum
STEM EELS-SI technique provides by far the best spectral boride can be mapped effectively by their plasmon excitations
performance, EFTEM-SI provides the advantage in that spatial alone, revealing the underlying phase structure in a way that
information is acquired in parallel. Hence, for elemental map- is not discernible from the zero-loss image. Often it can be
ping studies which require high pixel counts and do not require a useful preliminary analysis stage for you to view the plas-
high spectral resolution, EFTEM-SI can be a simple but effective mon image at different energies in real time, using a TV rate
choice, particularly for LaB6 equipped machines or those without camera or fast CCD acquisition mode, to help differentiate
fast STEM EELS capabilities. between phases and their possible chemical composition.
However, this approach yields very limited quantitative in-
formation and it can be difficult to identify phases that have
similar plasmon energies.
13.8 Low-Loss Imaging
A more accurate approach is to use EFTEM spectrum-imaging
to acquire an image series across the low-loss region. Using
Whilst the core-loss signal provides the most direct means a small slit width and energy step, an EFTEM series can be
to quantify the composition of materials using the EELS sig- acquired over the entire low-loss region with good energy res-
nal, the weak intensity of signal at high energy losses has olution within a short time (the total acquisition time is of the
prompted many people to consider what information can order of about 1 min). Once acquired, the position and width
be gleaned from the high-intensity, low-energy losses, via of the plasmon can be modelled accurately and mapped. This
low-energy edges or plasmon excitations. The relatively high is illustrated in Fig. 13.14, where different chemical phases of
intensity of the low-loss region allows the elemental or chem- silicon-based compounds are mapped by means of their specific
ical distribution of a material to be mapped quickly using, for plasmon energies (Sigle, 2003). Once acquired, the plasmon
example, the plasmon energy to differentiate between chem- energy can be measured from the EFTEM stack by, for example,
ical phases. For a thorough description of the low-loss EELS mathematically fitting a suitable model to the low-loss region
spectrum and the information contained within it, please refer to yield a plasmon energy map. The ability to rapidly record
to Chap. 38 of W&C. images with high pixel counts makes low-loss EFTEM-SI an
ideal approach for plasmonic studies. This approach was applied
by Schaffer et al. (2006), who used a monochromated TEM to
 13.9 Alternative Imaging Techniques for Biological Specimens 393

Fig. 13.13 Plasmon imaging of alu-

minum boride. The Al and B phases,
unidentifiable in the zero-loss image

Chapter 13
(a), are clearly visible in energy-fil-
tered images acquired at the Al (b)
and B plasmon peak energies (c),
respectively. The dark regions at the
edge of the image are thick regions,
resulting in a shift in plasmon inten-
sity to second order (or higher) plas-
mon scattering events

Fig. 13.14 a Color-coded plasmon a b

energy map of a sample with a
mixture of silicon nitride and silicon
carbide. Large orange regions cor-
respond to Si3N4 grains whereas the 24 eV
large blue areas are SiC grains. b EEL Si3N4
image spectra extracted from the en- 23 eV
ergy-filtered series. Curves a (Si3N4)
and b (SiC) were obtained from the 22 eV
areas marked by boxes in (a)
21 eV SiC
20 eV

directly image surface plasmons in gold nanoparticles at around Pre-Carbon Imaging. The dominant element in almost all biolog-
1 eV energy loss with surprisingly high energy resolution by use ical specimens is carbon. By acquiring an EFTEM image using
of a narrow slit width and energy. Alternatively, multiple linear an energy window just below the C K edge (at 283 eV) it is pos-
least-squared (MLLS) fitting provides a powerful way to map sible to highlight all the non-carbon species, including staining
the distribution of all the phases present by using their low- elements such as Pt and U if the section has been stained. This
loss spectrum as a fingerprint, though generally this requires the leads to a form of negative (bright) contrast, highlighting, e.g.,
plural scattering to be removed from the EFTEM series first to organelle membranes in cells, that is stronger than the equivalent
ensure that the components are free of deviations arising from positive (dark) contrast seen in an equivalent zero loss image.
thickness variations. An example of pre-carbon imaging from a biological specimen
is shown in Fig. 13.15.

Contrast Tuning. By changing the energy loss passing through

13.9 Alternative Imaging Techniques the energy window, it is often possible to see very clear reversals
of contrast, either in the low loss regions, near the plasmon res-
for Biological Specimens onance, or at higher losses. This arises because of the different
energy loss characteristics of different parts of the specimens
Biological specimens are often prepared in the form of sections and especially, in thick specimens, the variation in the gradient
which can be relatively thick, perhaps 100–300 nm or more in of the background loss signal, i.e., differences in the decrease
thickness. These thicknesses result from the need to ultramicro- in the background intensity with increasing energy loss. This
tome the specimens and the difficulty in preparing thin sections is often best optimized by simply watching the change in the
free from cutting artefacts. The thickness of the sample coupled loss image as the energy loss accepted by the energy window
with the beam sensitivity (and thus the need to minimize ex- is varied.
posure) of many biological specimens has led to a number of
alternative imaging techniques that are discussed briefly here. Most-Probable Loss Imaging. For very thick specimens, dis-
For a more in-depth discussion, see Reimer (1995). crete plasmon losses and clear ionization edges merge into a

Licensed to ([email protected])
 394 13 EFTEM

use of the partial inelastic scattering cross-section σK(β,∆) and

the quantification formulae given in Eq. 39.6 of W&C
Chapter 13

IK .ˇ; /
N=
Ilow .ˇ; /K .ˇ; / (13.8)

Where N is the areal density of element K (atoms per unit area),

and Ilow is the integrated low-loss intensity including the ze-
ro-loss peak. In practice, this kind of analysis is very difficult
as it is often highly problematic to acquire the low-loss intensity
under the same conditions as the core-loss signal of interest.
Particularly in the case of EFTEM, care must also be taken
to ensure illumination conditions remain constant between ac-
quisitions of zero-loss and core-loss information in order to
achieve acceptable accuracy (see the ‘condenser correction’
discussed in Sect. 13.4.5). The illumination will generally be
optimized for imaging the weak core-loss signal, which often
makes it difficult, if not impossible, to capture the zero-loss im-
age under identical conditions without saturating the detector. It
is, however, perfectly acceptable for you to reduce the exposure
time, and even the detector binning, to assist in acquiring the
low-loss image, providing an appropriate scaling factor is used
to accommodate the change in detector sensitivity. However,
capturing all the information required for quantification high
energy-loss edges is a challenge on the majority of systems
Fig. 13.15 Example of a stained biological sample (a cell nucleus from hu- equipped with electromagnetic beam shutters since the mini-
man autopsy tissue), imaged over different energy-loss ranges: unfiltered (a), mum CCD exposure time is relatively high (typically 50 ms).
plasmon loss image (b), pre-carbon image (c), and pre-carbon image inverted This limitation, however, is effectively removed for modern
(d). Comparisons between the inverted pre-carbon image and unfiltered imag- imaging filters equipped with electrostatic shutters which can
es show similarities, with the pre-carbon image showing improved contrast for provide exposure times as low as 1 μs with excellent linearity
features outside of the cell nucleus. The difference is more pronounced for un- (see Gubbens et al. 2010).
stained samples
Collectively these factors serve to restrict the accuracy of abso-
lute quantification, and, as a result, more reliable results can be
simple loss function, described by the Landau distribution with obtained by comparative approaches such as relative quantifica-
the most probable loss (i.e., the maximum in the loss spectrum) tion. Relative quantification involves measuring the atomic ratio
rising monotonically with the product of density and thickness of two or more elements (e.g., A and B), in which case the above
(‘mass-thickness’). The high intensity afforded by the most-prob- equations can be rewritten as
able loss image increases the usefulness of specimens with high
NA IA .ˇ; /B .ˇ; /
mass-thickness. This contrast mechanism is shown to good effect = (13.9)
in Fig. 13.16, where most probable loss imaging is used to image NB IB .ˇ; /A .ˇ; /
features in a very thick (~ 1 μm) biological specimen which could
not be easily resolved in the corresponding unfiltered or zero-loss This approach, first used by Leapman (1986) for quantification
filtered images. of elemental maps of biological specimens, has been shown to
have the advantage of cancelling the effects of several artefacts
including thickness effects and diffraction contrast in EFTEM
maps (Hofer et al. 1997). It is analogous to the ‘k-factor’ de-
13.10 Quantitative Elemental Mapping termination of elemental ratios in XEDS analysis, except that
in EELS the quantities dealt with are atomic rather than weight
ratios. The low-loss region need not be measured unless it is re-
Quantitative elemental mapping from STEM-EELS data is dis- quired for removal of plural scattering by deconvolution, which
cussed in Chap. 39 of W&C. Here, we expand on that within is generally not necessary using this method since plural scat-
the context of EFTEM mapping. The most commonly required tering effects tend to cancel to a good approximation for edges
quantity in quantitative analysis is an absolute concentration in of similar shape providing the same signal integration range ∆
number of atoms per unit area (the areal density) or, more con- is applied to each edge.
veniently, per unit volume. In theory, absolute quantification of
EELS or EFTEM core-loss maps can be accomplished by direct
 13.10 Quantitative Elemental Mapping 395

The obvious limitation to this approach is that the resultant

atomic ratio map contains only relative values. This works well
if the sample has two species that are always found together.

Chapter 13
However, it fails to be useful for regions where one element is
found in the absence of the other. An alternative approach that
has been proposed is to acquire a full EFTEM spectrum-image
series including the zero-loss image and all the major core-loss
edges of interest (Thomas and Midgley 2001), enabling absolute
quantification directly following Eq. 13.8. As discussed, the re-
quirement for continuous spectral information over an extended
energy range (e.g., 0–1000 eV) poses a significant problem for
EFTEM spectrum-image acquisition, placing unrealistic de-
mands on the dynamic range of the CCD camera as a result of
the very high zero- to core-loss intensity ratio. This problem can
be overcome by adaptively adjusting exposure times and CCD
hardware binning between successive acquisitions in order to
optimize the ratio between the energy-loss signal and the sta-
tistical detector noise. In effect this extends the dynamic range
of the camera by several orders of magnitude. For example,
allowing the exposure time to vary from 0.1 to 12 s and changing
the CCD hardware binning from single to eight times can extend
the ‘dynamic range’ of a 14-bit camera from 214 to 227 counts.
However, this gain in detector sensitivity is in exchange for
lower spatial resolution at a fixed magnification, as a result of
coarser spatial sampling through increased effective pixel size.
Fig. 13.16 EFTEM imaging of a thick biological specimen (HL 60 cell, thickness It is, therefore, important to consider the pixel per nanometer
approximately 1 μm). a The unfiltered image displays high chromatic blurring, resolution required at the coarsest binning to be used before se-
b whilst the elastic filtered image shows low contrast variation and ultramicro- ries acquisition, and operate at an appropriate microscope mag-
tome knife marks. c The most probable loss image at 120 eV provides excellent nification. For modern systems with high-speed electrostatic
image contrast by comparison, revealing features that would otherwise be diffi- shutters, the necessity to dynamically change the CCD binning
cult to resolve. d The image acquired beyond the carbon K edge shows reduced is effectively removed since the minimum exposure time is low
image contrast. (imaged at 200 kV) enough to acquire the zero-loss image without saturation even
at high CCD binning. Once acquired, the recorded images are
scaled and aligned appropriately to give one continuous EFTEM
In EFTEM analysis, this approach allows the calculation of spectrum-image data-set which may be visualized and analyzed
‘atomic ratio’ maps, which may be regarded as quantitative in using conventional EELS spectrum-imaging techniques, for ex-
nature. An example of quantitative atomic ratio mapping by ample removal of plural scattering by deconvolution or perform-
Hofer et al. (1997) illustrates the technique. Here, the relative ing full quantitative analysis.
concentrations of the constituents of CVD grown titanium car-
bonitride layers were investigated using the atomic ratio ap- An example data series acquired using this approach is shown
proach. Elemental maps were acquired for Ti, C and N using the in Fig. 13.18. An EFTEM image series was acquired from a
three-window approach, over a region with varying thickness 316 stainless steel, from 0–800 eV loss using a slit size and en-
and significant amounts of diffraction contrast. The atomic ratio ergy step of 10 eV. The energy-range was selected to cover the
map was computed by dividing one elemental map by another most suitable core-loss edges for the species of interest. Since
and normalizing using the computed partial ionization cross-sec- the lower energy-loss range has relatively high intensity, the
tions following Eq. 3.10 above. The computed atomic ratio map acquisition time for such data series is non-prohibitive (typically
showing the C/Ti ratio, along with the 1D histogram, is shown less than 10 min in total). Once acquired, the data-set can be
in Fig. 13.17. As can be seen, the thickness variations and dif- examined by extracting spectra from selected areas, allowing
fraction contrast present in the zero-loss image are not apparent a list of species present to be compiled. Plural scattering was
in the ratio map. Further, the histogram plot shows two very removed from the series by Fourier-log deconvolution, before
well-defined intensity peaks, indicating that there are two main performing quantitative analysis following Eq. 13.8 to yield the
phases present. From the histogram values, it can be seen that areal density maps. The absolute concentrations computed via
these correspond to C:Ti ratios of approximately 0.36 and of this approach show good agreement with the known compo-
unity, corresponding to Ti(C,N) and TiC phases, respectively. sition of the material, with the measured chromium and iron
Analysis of the region using the absolute quantification approach concentrations in the matrix agreeing with the expected values
produced maps that contained high levels of thickness variation within the error margin for the computed cross-sections (10 %).
and diffraction contrast artefacts. This example illustrates that quantitative EFTEM can be per-
 396 13 EFTEM

Fig. 13.17 Quantitative analysis of a

TiC/Ti (C,N) interface using atomic-ra-
tio EFTEM. The bright field image (a)
Chapter 13

contains diffraction contrast. The C:Ti

atomic ratio map (b) contains intensi-
ty values which directly represent the
local atomic ratios and is relatively
free of artefacts. Histogram analysis
(c) reveals that two main phases are
present, with C:Ti ratios of 0.36 and
1.03

Fig. 13.18 Example of extended

EFTEM data-series acquisition. The
series contains a precipitate cluster
in an aged 316 stainless steel (a),
and was acquired from 0–800 eV
loss with a slit-width and step size of
10 eV. The image-spectrum line-trace
(b) is taken along the line indicated
and shows the captured core-loss
edges (plural scattering has been re-
moved and a log-scale used for clari-
ty). The Cr and Fe areal density maps
(c, d) were calculated from the L23
edges using absolute quantification,
yielding elemental distribution maps
in units of atoms per unit area

formed in practice. However, such analyses are non-trivial and series is sufficient to allow the capture of this electron loss near-
are not performed routinely. edge structure for use in chemical mapping.

EFTEM chemical-state mapping is demonstrated in Fig. 13.19,

where the ELNES of the carbon K edge at 284 eV loss is used to
13.11 Chemical State Mapping Using perform chemical bond maps of carbon in various forms (Martin
et al. 1996). The diamond-like carbon signal was obtained by di-
ELNES viding an energy loss image recorded at 287 eV by one recorded
at 275 eV using a 5 eV slit width for each image. The diamond
Chemical state mapping using the core-loss ELNES is described signal was obtained by dividing an energy loss image recorded
in Chap. 40 for EELS spectra. By use of a small slit width δ and at 327 eV by one recorded at 312 eV using a 5 eV slit width and
energy step size ∆E, the spectral resolution attainable in EFTEM a 30 s exposure for each image.
 13.12 Hybrid EFTEM Modes 397

Fig. 13.19 Carbon phase mapping

using the near-edge ELNES structure
of the carbon C edge in a sample

Chapter 13
composed of a number of forms
of carbon. The diamond and dia-
mond-like carbon (DLC) signals were
emphasized by dividing one EFTEM
image by another

Fig. 13.20 a A high-resolution image of a LaCaMnO3 / SrTiO3 sandwich structure. b A spectrum-image showing the Ca K-edge and La and Ti L-edges dispersed in
energy along the x-axis and in real space along the y-axis

Such studies are relatively demanding and are limited to the trum (Walther 2003). A slit is used to delineate a small part of
lower end of the energy-loss spectrum, but they serve to illustrate the overall image and in this case it is essential to ensure that the
the versatility of the EFTEM technique. interface normal is perpendicular to the dispersion, otherwise the
spectral information will be blurred across the interface.

Relevant parts of the resulting EFTEM line-spectrum are shown

13.12 Hybrid EFTEM Modes(ω-q, Line in Fig. 13.20b, indicating how the Ti signal appears to be lo-
calized to the thin STO barrier layer confirming a lack of Ti
Spectrum EFTEM) diffusion into the surrounding LCMO layers. This mode of
line-spectrum acquisition has the distinct advantage that sub-
The ability to create a line focus in an imaging filter means that sequent spectra will not suffer from energy drift since they are
it is possible to create a parallel array of energy loss spectra from acquired in parallel. However, it should be noted that, unlike
an extended region of a specimen. Figure 13.20 illustrates how regular EELS acquisition, only the energy-loss electrons located
this can be achieved by placing a thin slit at the entrance aperture at the mid-point of the energy slit will be spatially in focus, and
of the imaging filter which selects a thin sliver of the image to energy channels away from this point will suffer chromatic blur-
then disperse in energy and create a hybrid image in which the ring in the spatial direction in the same way as for a regular
vertical axis is real space but the horizontal axis is energy loss. EFTEM image acquisition.
In essence this enables a ‘ω-r pattern’ or ‘line spectrum-image’
to be acquired in parallel on the CCD camera. In the example of A similar geometry can be established for diffraction to form an
Fig. 13.20a a small part of a high resolution image of a sandwich ‘ω-q pattern’ or ‘spectrum-diffraction pattern’. Again a narrow
of LCMO and STO is dispersed in energy to form an image-spec- slit is used to delineate a small part of the diffraction pattern
 398 13 EFTEM

(perhaps along a systematic row of reflections) and this is then The other dispersive feature is a hybrid state: above 12 eV the
dispersed in energy. Such patterns are particularly useful to show excitation is surface plasmon-like in character, where ε1 < 0 and
the dispersion of low-loss features, the Bethe ridge over a large below 12 eV the excitation changes to a waveguide mode where
Chapter 13

energy range, or to identify the character of peaks (arrowed) ε1 > 0. These features, when integrated over the scattering angle,
in the low-loss region of the spectrum, such as illustrated in give rise to the peaks in the spectrum in the 5–12 eV range, and
Fig. 13.21. Here, an energy loss spectrum from diamond reveals are therefore not related to the density of states as might have
a series of peaks in the energy loss region up to 30 eV (Midgley first been thought. An interband transition occurs at 23 eV and
1999). Many EELS studies of semiconductors are now trying the volume plasmon is seen at about 33 eV which at this scale is
to determine band gaps directly by studying the low loss, but non-dispersive. By placing an energy-selecting slit in the spectral
one has to be careful not to confuse band edges with other low- plane it is possible to form an energy-filtered ‘diffraction pattern’
loss phenomena, such as Cerenkov radiation which has disper- in the image plane. Figure 13.21d shows a montage at different
sion but over angles of tens of microradians. Thus a method is energy losses using a 2 eV slit to ‘image’ the Cerenkov radiation
needed to allow such high angular resolution energy-loss spec- and waveguide mode (the double annulus of scattering).
troscopy. Figure 13.21b shows how high angular resolution can
be achieved by using a narrow convergence angle and raising
the sample in height (defocus) so that the sample is no longer
conjugate with the image plane. The image plane now reveals a 13.13 EFTEM Tomography
diffraction pattern whose angular resolution is equal to the probe
size divided by the sample defocus, i.e., ∆θ = d/z. This way an
angular resolution of a few microradians can be achieved. It is now possible to extend EFTEM into three dimensions using
electron tomographic techniques. This has been explored in the
Figure 13.21c shows an ω-q pattern recorded from diamond to- Electron Tomography chapter of this book and interested read-
gether with a graphical explanation of the features seen in the ers are asked to consult that chapter for more details. It is clear
pattern. The dispersion of the Cerenkov radiation is clear. The that 3D imaging at high resolution is becoming a key method
Cerenkov radiation disappears where ε1β2 > 1 (ε1 is the real part of characterization and chemical mapping in 3D will play a big
of the dielectric function and β = v/c, the ratio of the electron part in this. EFTEM is well positioned to be adapted to 3D im-
speed with the speed of light). For 200 keV electrons in diamond aging and to yield 3D compositional maps which are reliable
this cut-off is about 12 eV. and quantitative.
 13.13 EFTEM Tomography 399

Chapter 13

Fig. 13.21 a A typical low loss spectrum (shown on a log scale) from a diamond sample, recorded near the <001> zone axis, in image-coupled mode with a
collection angle of 3 mrad. Arrows highlight key features in the spectrum. b Schematic diagram showing the geometry of the acquisition of a w–q pattern. c A
w–q pattern from diamond taken at the same orientation as (a). A schematic diagram explains the features seen above. d A montage of images taken with a 2 eV
energy window showing the features illustrated in the w–q pattern in (c). The double annulus corresponds to the two main dispersive features seen in (c) in the
0–12 eV energy range
 400 13 EFTEM

Chapter Summary
Chapter 13

Energy-filtered transmission electron microscopy (EFTEM) TEM imaging and without the need for a small probe. When
is a versatile and invaluable tool for the analysis and char- performed as EFTEM-SI, spectral information can also be
acterization of materials up to the nanometer scale. It can acquired, sometimes with surprisingly good energy resolu-
be used quite simply as a TEM image contrast enhance- tion. Although EFTEM does not normally have the energy
ment technique using zero-loss filtering or by imaging at resolution of EELS mapping, it enables 2D and (sometimes
an appropriate energy loss. It enhances diffraction images 3D) compositional maps over a large field of view to be
by removing the inelastic ‘fog’. The three-window ele- acquired in a relatively short space of time. Finally, EFTEM
mental mapping approach provides the ability to acquire can be used in hybrid modes to allow either EFTEM line-
element specific maps at high spatial resolution and with scan or angular resolved spectra to be acquired.
a high number of pixels, all with the simplicity of regular
 Appendix 401

Appendix

Chapter 13
Self-Assessment Questions Specific References
Berger A, Kohl H (1993) Optimum imaging parameters for ele-
Q13.1 What advantage(s) does zero-loss imaging have over mental mapping in an energy-filtering TEM. Optik 92:175–
regular bright field imaging in the TEM? 193
Q13.2 What differences are there between three-window ele- Egerton RF, Crozier PA (1997) The effect of lens aberrations on
mental mapping and jump-ratio imaging of a core-loss the spatial resolution of energy-filtered TEM image. Micron
edge? Which should be used for quantitative mapping? 28:117–124
Q13.3 Why is the microscope high tension used to perform Gubbens AJ, Barfels M, Trevor C, Twesten R, Mooney P,
EFTEM at an energy loss? Thomas PJ, Menon NK, Kraus B, Mao C, McGinn B (2010)
Q13.4 In addition to thickness, what other physical properties The GIF Quantum, a next generation post-column imaging
from the specimen affect the intensity in a t/λ map? energy filter. Ultramicroscopy 110:962–970
Q13.5 What effect does increasing the energy-selecting slit Hofer F, Warbichler P, Grogger W (1995) Imaging nanome-
width have on i) the spatial resolution and ii) the re- ter-sized precipitates in solids by electron spectroscopic im-
corded signal have when performing core-loss imag- aging. Ultramicroscopy
ing? 59:15–31
Q13.6 Why is an objective aperture inserted for EFTEM im- Hofer F, Grogger W, Kothleitner G, Warbichler P (1997) Quan-
aging? titative analysis of EFTEM elemental distribution images.
Q13.7 What advantages does EFTEM-SI offer over simple Ultramicroscopy 67:83–103
two- or three-window EFTEM mapping? Hunt JA, Williams DB (1991) Electron energy-loss spectrum-im-
Q13.8 What effect does diffraction-contrast have on EFTEM aging. Ultramicroscopy 38:47–73
core-loss maps? What steps can be taken to reduce these Jeanguillaume C, Trebbia P, Colliex C (1978) About the use
effects? of electron energy-loss spectroscopy for chemical mapping
of thin foils with high spatial resolution. Ultramicroscopy
3:237–242
References Krivanek OL, Gubbens AJ, Dellby N (1991) Development in
EELS instrumentation for spectroscopy and imaging. Microsc
Microanal M2:315–332
Krivanek OL, Kundmann MK, Kimoto K (1995) Spatial resolu-
General References tion in EFTEM maps. J Microsc 180:277–287
Lavergne J, Martin J, Belin B (1992) Interactive electron ener-
Egerton RF (2011) Electron Energy-Loss Spectroscopy in the gy-loss mapping by the “imaging-spectrum” method. Microsc
Electron Microscope, 3rd edn. Springer Science & Business Microanal M3:517–528
Media, New York (A classic text, primarily for EELS but is Leapman RD (1986) Microbeam Analysis 1986. In: Romig AD,
also of great use for EFTEM) Chambers WF (eds) Quantitative electron energy loss spec-
Grogger W, Varela M, Ristau R, Schaffer B, Hofer F, Krishnan troscopy and elemental mapping in biology. San Francisco
KM (2005) J Electron Spectrosc Related Phenom 143:139– Press, San Francisco, pp 187–191
147 (A review article on EFTEM focusing on nanometer Martin JMM, Vacher B, Ponsonnet L, Dupuis V (1996) Chemi-
resolution) cal bond mapping of carbon by image-spectrum EELS in the
Reimer L (1995) Energy-Filtering Transmission Electron Mi- second derivative mode. Ultramicroscopy 65:229–238
croscopy. Springer Series in Optical Sciences, vol. 71. Midgley PA, Saunders M (1996) Quantitative electron-diffrac-
Springer-Verlag, Berlin (This is a fine book with a number of tion – from atoms to bonds. Contemp Phys 37:441
chapters written by experts in the field of EELS and EFTEM) Schaffer B, Grogger W, Kothleitner G (2004) Automated spatial
Sigle W (2005) Annu Rev Mater Res 35:239–314 (This is a long drift correction for EFTEM image series. Ultramicroscopy
review article that deals with many of the topics covered in 102:27–36
this chapter and some at a higher level.) Schaffer B, Kothleitner G, Grogger W (2006) EFTEM spec-
trum imaging at high-energy resolution. Ultramicroscopy
106:1129–1138
Sigle W, Krämer S, Varshney V, Zern A, Eigenthaler U, Rühle
M (2003) Plasmon energy mapping in energy-filtering trans-
mission electron microscopy. Ultramicroscopy 96:565–571
 402 13 EFTEM

Egerton RF (1982) Principles and practice of quantitative elec-

Thomas PJ, Midgley PA (2001) Image-Spectroscopy 2: the Re- tron energy-loss spectroscopy. In: Heinrich KFJ (ed) Mi-
moval of Plural Scattering from Extended Energy-Filtered crobeam Analysis. San Francisco Press, San Francisco
Chapter 13

Series by Fourier Deconvolution Techniques. Ultramicros- Egerton RF (1991) Factors affecting the accuracy of elemen-
copy 88:187–194 tal analysis by transmission EELS. Microsc Microanal M
Walther T (2003) Electron energy-loss spectroscopic profiling 2:203–213
of thin film structures: 0.39 nm line resolution and 0.04 eV Egerton RF (1997) The effect of objective-lens aberrations
precision measurement of near-edge structure shifts at inter- on EFTEM, STEM and SEM images. Inst Phys Conf Ser
faces. Ultramicroscopy 96:401–411 153:149–154 (EMAG 97)
Watanabe M, Allen FI (2012) The SmartEFTEM-SI method: De- Egerton RF, Crozier PA (1997) The effect of lens aberrations on
velopment of a new spectrum imaging acquisition scheme for the spatial resolution of energy-filtered TEM image. Micron
quantitative mapping by energy-filtering transmission elec- 28:117–124
tron microscopy. Ultramicroscopy 113:106–119 Gubbens AJ, Barfels M, Trevor C, Twesten R, Mooney P,
Thomas PJ, Menon NK, Kraus B, Mao C, McGinn B (2010)
The GIF Quantum, a next generation post-column imaging
References for Further Reading energy filter. Ultramicroscopy 110:962–970
Haking A, Troester H, Richter K, Crucifix C, Spring H, Tren-
Ahn CC, Krivanek OL (1983). EELS Atlas, Gatan Inc., 780 Com- delburg MF (1999) An approach to an objective background
monwealth Drive, Warrendale, PA 15086. subtraction for elemental mapping with core-edges down to
Bazettjones DP, Ottensmeyer FP (1981) Phosphorus distribution 50 eV: description, evaluation and application. Ultramicros-
in the nucleosome. Science 211:169–170 copy 80:163–182
Berger A, Kohl H (1992) Elemental mapping using an imaging Hofer F (1987) EELS quantification of M edges by using oxidic
filter: image formulation and resolution limits. Microsc Mi- standards. Ultramicroscopy 21:63–68
croanal M(3):159–174 Hofer F, Kothleitner G (1993) Quantitative microanalysis using
Berger A, Kohl H (1993) Optimum imaging parameters for ele- electron energy loss spectrometry. I. Li and Be in oxides.
mental mapping in an energy-filtering TEM. Optik 92:175– Microsc Microanal M 4:539–560
193 Hofer F, Warbichler P (1996) Improved imaging of secondary
Blake DF, Freund F, Krishnan KFM, Echer CJ, Shipp R, Buch phases in solids by energy-filtering TEM. Ultramicroscopy
TE, Tielens AG, Lipari RJ, Hetherington CJD, Chang S 63:21–25
(1988) The nature of interstellar diamond. Nature 332:611– Hofer F, Wilheim P (1993) EELS microanalysis of the elements
613 Ca to Cu using M23edges. Ultramicroscopy 49:189–197
Brink HA, Trevor C, Hunt J, Mooney PE (2000) A new high Hofer F, Golob P, Brunegger A (1988) EELS quantification of the
performance detector for electron energy loss spectroscopy. elements Sr to W by means of M45edges. Ultramicroscopy
Microsc Microanal 6:212–213 25:81–84
Brydson R, Sauer H, Engel W, Zeitler E (1991) EELS as a finger- Hofer F, Grogger W, Kothleitner G, Warbichler P (1997) Quan-
print of the chemical co-ordination of light elements. Microsc titative analysis of EFTEM elemental distribution images.
Microanal M(2):159–169 Ultramicroscopy 67:83–103
Cheng SC, Egerton RF (1985) Signal/background ratio of ioni- Hofer F, Warbichler P, Grogger W (1995a) Imaging nanome-
sation edges in EELS. Ultramicroscopy 16:279–282 tre-sized precipitates in solids by electron spectroscopic im-
Colliex C (1985) An illustrated review of various factors gov- aging. Ultramicroscopy 59:15–31
erning the high spatial resolution capabilities in EELS micro- Hofer F, Warbichler P, Grogger W, Lang O (1995b) On the ap-
analysis. Ultramicroscopy 18:131–150 plication of energy-filtering TEM in materials science: I. Pre-
Colliex C, Jeanguillaume C (1989) Spectrum-image: the next cipitates in a Ni/Cr-alloy. Micron 26:377–390
step in EELS digital acquisition and processing. Ultramicros- Hunt JA, Williams DB (1991) Electron energy-loss spectrum-im-
copy 28:252–257 aging. Ultramicroscopy 38:47–73
Crozier PA (1995) Practical limits on the spatial resolution in en- Isaacson M (1980) A poor man’s approach to semi-quantitative
ergy-filtered mapping. In: Bailey GW, Ellisan MH, Hennigar analysis with electron energy loss spectroscopy Proc 38th
RA, Zaluzec NJ (eds) Proc Microsc Microanal, pp 304–305 EMSA meeting., pp 110–111
Crozier PA (1995) Quantitative elemental mapping of materials Jeanguillaume C, Trebbia P, Colliex C (1978) About the use
by energy-filtered imaging. Ultramicroscopy 58:157–174 of electron energy-loss spectroscopy for chemical mapping
Daniels J, Festenberg CV, Raether H, Zeppenfeld K (1970) Op- of thin foils with high spatial resolution. Ultramicroscopy
tical constants of solids by electron spectroscopy. Springer 3:237–242
Tracts Modern Phys 54:78–135 Johnson DE (1979) Energy-loss spectroscopy for biological
Egerton RF (1978) Formulae for light element analysis by elec- research. Introduction to Analytical Electron Microscopy.
tron energy loss spectrometry. Ultramicroscopy 3:243–251 Plenum Press, New York, pp 245–258
Egerton RF (1979) K-shell ionization cross-sections for use in Joy DC, Maher DM (1981) The quantitation of electron energy
microanalysis. Ultramicroscopy 4:169–179 loss spectra. J Microsc 124:37–48
  References 403

Joy DC, Egerton RF, Maher DM (1979) Progress in the quan- Liu D-R, Brown LM (1987) Influence of some practical factors
tification of electron energy-loss spectra. In: O’Hare AMF on background extrapolation in EELS quantification. J Mi-
(ed) Scanning Electron Microscopy. SEM Inc.,, Illinois, pp crosc 147:37–49

Chapter 13
817–826 Malis T, Tichmarsh JM (1985) A k-factor approach to EELS
Joy DC, Romig AD, Goldstein JI (1986) Principles of analytical analysis. Inst Phys Conf Ser 78:181–182
electron microscopy. Plenum Press, New York Malis T, Cheng SC, Egerton RF (1988) EELS log-ratio technique
Kohl H, Berger A (1995) The resolution limit for elemental map- for specimen measurement in the TEM. J Electron Microsc
ping in energy-filtering transmission electron microscopy. Ul- Techniq 8:193–199
tramicroscopy 59:191–194 Marien J, Plitzko JM, Spolenak R, Keller RM, Mayer J (1998)
Körtje KH (1994) Image-EELS: simultaneous recording of mul- Quantitative electron spectroscopic imaging studies of mi-
tiple electron energy-loss spectra from series of spectroscopic croelectronic metallization of layers. J Microsc 194:71–78
images. J Microsc 174:149–159 Martin JM, Lavergne JL, Vacher B, Inoue K (1995) Interactive
Kothleitner G, Hofer F (1998) Optimisation of the signal to noise image-spectrum EELS: Application to elemental mapping of
ratio in EFTEM elemental maps with regard to different ioni- lubricant colloids. Microsc Microanal M 6:53–63
sation edge types. Micron 29:349–357 Martin JMM, Vacher B, Ponsonnet L, Dupuis V (1996) Chemi-
Krivanek OL, Mooney PE (1993) Applications of slow-scan cal bond mapping of carbon by image-spectrum EELS in the
CCD cameras in TEM. Ultramicroscopy second derivative mode. Ultramicroscopy 65:229–238
49:95–108 Mayer J, Plitzko JM (1995) Mapping of ELNES on a nanometre
Krivanek OL, Gubbens AJ, Dellby N (1991) Development in scale by electron spectroscopic imaging. J Micros 183:2–8
EELS instrumentation for spectroscopy and imaging. Microsc Mayer J, Eigenthaler U, Plitzko JM, Dettenwanger F (1997)
Microanal M 2:315–332 Quantitative analysis of electron spectroscopic imaging se-
Krivanek OL, Gubbens AJ, Dellby N, Meyer C (1992) Design ries. Micron 28:361–370
and first applications of a post-column imaging filter. Microsc Meyer RR, Kirkland AI (1998) The effect of electron and photon
Microanal M 3:187–199 scattering on signal and noise transfer properties of scintil-
Krivanek OL, Gubbens AJ, Kundmann MK, Carpenter GC lators in CCD cameras used for electron detection. Ultrami-
(1993) Elemental mapping with an energy-selecting imaging croscopy 75:23–33
filter. In: 51st Proc EMSA, pp 586–587 Midgley PA (1999) A simple new method to obtain high angular
Krivanek OL, Kundmann MK, Kimoto K (1995) Spatial resolu- resolution ω–q patterns. Ultramicroscopy 76:91–96
tion in EFTEM maps. J Microsc 180:277–287 Midgley PA, Saunders M (1996) Quantitative electron-diffrac-
Lanio S, Rose H, Krahl D (1986) Test and improved design of tion – from atoms to bonds. Contemp Phys 37:441
a corrected imaging magnetic energy filter. Optik 73:56–68 Otenmeyer FP, Andrew JW (1980) High-resolution Microanal-
Lavergne J, Martin J, Belin B (1992) Interactive electron ener- ysis of Biological Specimens by electron energy loss spec-
gy-loss mapping by the “imaging-spectrum” method. Microsc troscopy and by electron spectroscopic imaging. J Ultrastruct
Microanal M 3:517–528 Res 72:336–348
Lavergne JL, Foa C, Bongrand P, Seux D, Martin JM (1994) Ap- Otten MT (1993) Schottky versus cold field emission. Philips
plication of recording and processing of energy-filtered image Electron Optics Special Bulletin. Philips, Eindhoven, pp 4–7
sequences for the elemental mapping of biological specimens: Özkaya D, Yuan J, Brown LM, Flewitt PEJ (1995) Segrega-
Imaging-spectrum. J Microsc 174:195–206 tion-induced hole drilling at grain boundaries. J Microsc
Leapman RD (1986) Quantitative electron energy loss spec- 180:300–306
troscopy and elemental mapping in biology. In: Romig AD, Pearson DH, Ahn CC, Fultz B (1993) White lines and d-electron
Chambers WF (eds) Microbeam Analysis 1986. San Fran- occupancies for the 3d and 4d transition metals. Phys Rev
cisco Press, San Francisco, pp 187–191 47:8471–8477
Leapman RD, Hunt JA (1991) Comparison of detection limits for Pennycook SJ (1982) High resolution electron microscopy and
EELS and EDXS. Micros Microanal M 2:231–244 microanalysis. Contemp Phys 59:371–400
Leapman RF, Hunt JA (1995) Compositional imaging with elec- Plitzko JM, Mayer J (1996) Mapping of ELNES on a nanometre
tron energy-loss spectroscopy. J Microsc Soc America 3:93 scale by electron spectroscopic-imaging. J Microsc 183:2–8
Leapman RD, Swyt CR (1983) Electron energy-loss imaging Plitzko JM, Mayer J (1999) Quantitative thin film analysis by
in the STEM: systematic and statistical errors. In: Gooley energy-filtering transmission electron microscopy. Ultrami-
R (ed) Microbeam Analysis 1983. San Francisco Press, San croscopy 78:207–219
Francisco Pun T, Ellis JR, Eden M (1984) Optimized acquisition param-
Leapman RD, Swyt CR (1988) Separation of overlapping core eters and statistical detection limit in quantitative EELS. J
edges in electron energy loss spectra by multiple – least- Microsc 135:295–316
squares fitting. Ultramicroscopy 26:393–404 Van Puymbroeck J, Jacob W, Van Espen P (1992) Methodology
Liang WY, Beal AR (1976) A study of the optical joint densi- for spectrum evaluation in quantitative electron energy-loss
ty-of-states function. J Phys C 9:2823–2832 spectrometry using the Zeiss CEM902. J Microsc 166:273–
286

Licensed to ([email protected])
 404 13 EFTEM

Reimer L (1995) Energy-Filtering Transmission Electron Mi- Sigle W, Krämer S, Varshney V, Zern A, Eigenthaler U, Rühle
croscopy. Springer Series in Optical Sciences, vol. 71. M (2003) Plasmon energy mapping in energy-filtering trans-
Springer, Berlin mission electron microscopy. Ultramicroscopy 96:565–571
Chapter 13

Reimer L (1998) Energy-filtering imaging and diffraction. Mater Thomas PJ, Midgley PA (2001) Image-Spectroscopy I: The Ad-
Trans 39:873–882 vantages of Increased Spectral Information for Compositional
Rez P (1982) Cross-sections for energy loss spectroscopy. Ultra- EFTEM Analysis. Ultramicroscopy 88:179–186
microscopy 9:283–288 Thomas PJ, Midgley PA (2001) Image-Spectroscopy 2: the Re-
Rez P (1989) Inner-shell spectroscopy: an atomic view. Ultra- moval of Plural Scattering from Extended Energy-Filtered
microscopy 28:16–23 Series by Fourier Deconvolution Techniques. Ultramicros-
Rez P, Weng X, Ma H (1991) The interpretation of near-edge copy 88:187–194
structure. Microsc Microanal M2:143–151 Thomas PJ, Midgley PA, Spellward P (1999). Compositional
Rez P, Bruley J, Brohan P, Payne M, Garvie LAJ (1995) Review mapping in the EFTEM using Image-Spectroscopy. Inst Phys
of Methods for calculating near-edge structure. Ultramicros- Conf Ser (EMAG 99).
copy 59:159–167 Trebbia P, Bonnet N (1990) EELS elemental mapping with un-
De Ruijter WJ (1995) Imaging properties and applications of conventional methods I. Theoretical basis: image analysis
slow-scan charge-coupled device cameras suitable for elec- with multivariate statistics and entropy concepts. Ultrami-
tron microscopy. Micron 26:247–275 croscopy 34:165–178
Schaffer B, Grogger W, Kothleitner G (2004) Automated spatial Walther T (2003) Electron energy-loss spectroscopic profiling
drift correction for EFTEM image series. Ultramicroscopy of thin film structures: 0.39 nm line resolution and 0.04 eV
102:27–36 precision measurement of near-edge structure shifts at inter-
Schaffer B, Kothleitner G, Grogger W (2006) EFTEM spec- faces. Ultramicroscopy 96:401–411
trum imaging at high-energy resolution. Ultramicroscopy Warbichler P, Hofer F, Hofer P, Lotofsky E (1998) On the ap-
106:1129–1138 plication of energy-filtering TEM in materials science: III.
Schaffer B, Hohenester U, Trügler A, Hofer F (2009) High-res- Precipitates in steel. Micron 29:63–72
olution surface plasmon imaging of gold nanoparticles by Weikenmeier AL, Nüchter W, Mayer J (1995) Quantitative char-
energy-filtered transmission electron microscopy. Phys Rev acterisation of point spread function and detection quantum
B 79:041401 efficiency for a YAG scintillator slow scan CCD camera. Op-
Schweitz KO, Ratzke K, Foord D, Thomas PJ, Greer AL, Geisler tik 99:147–154
H, Chevallier J, Bottiger J (2000) The microstructural devel- Williams DB, Edington JW (1976) High resolution microanaly-
opment of Ag/Ni multilayers during annealing. Philos Mag sis in materials science using electron energy loss measure-
A 80:1867–1877 ments. J Microsc 108:113–145
Scott J, Thomas PJ, MacKenzie M, McFadzean S, Wilbrink J, Wooten F (1972) Optical properties of solids. Academic Press,
Craven AJ, Nicholson WAP (2008) Near-simultaneous dual New York
energy range EELS spectrum imaging. Ultramicroscopy Zanchizx G, Kihn Y, Sevely J (1982) On aberration effects in
108:1586–1594 the chromatic plane of the Omega-filter. Optik 60:427–436
Calculating EELS
14

Part I
Chapter Preview

Chapter 14
Electron energy-loss spectroscopy (EELS) is a unique tool electronic structure and, from there, a simulation of the
to measure electronic structure at the sub-nanometer or expected EELS data. (Following W&C, we’ll always say
even atomic scale. From such measurements we can de- EELS even when EEL might be grammatically correct – it
velop an understanding of materials’ properties and cor- is just too awkward) This chapter aims to introduce the
relate them with the structure and composition. To achieve physical concepts behind how these calculations are done.
this we have to interpret the fine details in the spectrum We’ll discuss the fine structure in the ionization edges (the
in terms of the electronic structure and bonding. Our abil- energy-loss near-edge structure (ELNES)) and, to a lesser
ity to perform this interpretation can be significantly en- extent, valence EELS.
hanced by using a calculation, from first principles, of the

© Springer International Publishing Switzerland 2016

C. B. Carter, D. B. Williams (Eds.), Transmission Electron Microscopy, DOI 10.1007/978-3-319-26651-0_14 405
 406 14 Calculating EELS

14.1 Introduction
Part I

As discussed in Chap. 40 of W&C, the EELS spectrum contains

a wealth of information in addition to the composition. There are
fine details in the spectrum, in the energy-loss near-edge structure
(ELNES) and in the low-loss region. We can connect this informa-
tion to the electronic structure of the material and thus go beyond
measuring structure–composition correlations to understanding
the way in which the atoms in the material are bonded together.
Mechanical, electrical, and optical properties are all intrinsically
related to the electronic structure. EELS is increasingly used to
gain an insight into these material properties and will continue to
Chapter 14

be as more materials are fabricated on very small length scales.

Our challenge is to understand what the observed EELS fine

structure means in terms of bonding and properties. In a few spe-
cial cases the interpretation of the spectra is relatively straight-
forward. For example, in carbon-based materials there are dis-
tinct features in both the ELNES (as shown in Fig. 40.5 in W&C)
and the low-loss region of the spectrum that can be connected
to sp2 and sp3 bonding types. Because the properties of amor-
phous carbon materials, such as density and hardness, can be
connected to the sp2/sp3 ratio, we can use EELS to characterize Fig. 14.1 The evolution over time of the simulation of the calculated spectrum
these materials (see Fig. 40.11 in W&C). In some other materials for the C K-edge in diamond as compared to an experimental spectrum
a fingerprinting strategy can be useful; we can draw conclusions
about the bonding in the material of interest by comparing our
experimental spectra to spectra from standard materials where more sophisticated calculation approaches has driven this trend.
the nature of the bonding is well understood. Published experimental measurements of EELS fine structure
are becoming more routinely accompanied by a corresponding
In general you’ll find it desirable to go beyond such approaches theoretical simulation.
and obtain a full and predictive understanding of the origin of all
of the features in the EELS spectrum. The key to developing this Today there is a wide variety of approaches and programs avail-
deeper understanding is to calculate the electronic structure using able to calculate electronic structure. Unfortunately, many of
the underlying physics and, from this, to predict what your EELS these programs were written for other purposes, so they do not
spectrum should look like. In this way, you can connect features in take the next step and calculate the EELS spectrum from the
the experimental spectrum with aspects of the electronic structure. electronic structure. Notable exceptions include the WIEN2k
program, CASTEP, and FEFF; the last of these was explicitly
The importance of calculations was recognized as early as 1974 written for ELNES calculations (or, rather, for the X-ray equiv-
when Egerton and Whelan compared their experimental spec- alent). Practical guides to using WIEN2k and CASTEP have
trum from diamond to calculations of the unoccupied electron been published, and you should read them if you are planning
states. In the following 40 years, calculations were often used to to use these programs. Many research groups have developed
qualitatively explain the features in the spectrum. For many years their own computational approaches; we will discuss some of
the success of the calculations in reproducing experimental spec- these in later sections.
tra was somewhat limited. Occasionally the calculations would
fail to predict peaks and features in the spectrum, or they would An in-depth interpretation of the spectra requires an understand-
appear at the wrong energy. Usually the relative intensities of the ing of the underlying physics. That is, we need to go beyond
various features were poorly predicted. So be careful when you using the computer program as another ‘black-box’ to obtain a
use the literature but don’t ignore earlier studies. simulated spectrum. In this chapter we’ll provide the background
In recent years we have seen remarkable advances and it is not and introduce the principles of theoretical simulation of EELS
uncommon to see calculated spectra that show excellent agree- fine structure. To do so, within the constraints of a single chap-
ment with experimental spectra. Figure 14.1 shows a number of ter, requires the assumption that you already have some work-
calculations of the C-K edge of a diamond from the literature, ing knowledge of quantum mechanics, solid-state physics, and
illustrating a general trend of improved agreement of the rela- EELS. Chaps. 2–4 in W&C contain some of the basics, and the
tive peak positions and intensity, when compared with experi- general reference list at the end of this chapter has some sugges-
ment. The combination of increased computational speed and tions for background reading.
 14.2 Density Functional Theory (DFT) 407

A Literature Comment the spectrum. Simulations of the valence EELS are also being
successfully applied for some materials.

Part I
When searching the literature for reference spectra for The common starting point for our calculations of both ELNES
your ELNES measurements, be sure to also look for XA- and the valence EELS is a ground-state calculation of the elec-
NES/NEXAFS spectra. 9 tronic structure; i.e., we determine the energy of the occupied
and unoccupied states for the electrons of the material in the
absence of any electron-induced excitation. We then calculate
Chaps. 38 and 39 in W&C divided the EELS spectrum into two the spectra by including the interaction of this ground-state
regions: the high-energy, or core-loss EELS (> 50 eV), contain- electronic structure with the fast incident electron. A review of
ing the ionization edges, and the low-energy or valence EELS the theory behind the calculation of the ground-state electronic
(< 50 eV) which contains interband transitions and plasmons. Re- structure in solids, generally known as density functional theory
call from W&C, Chapters 4, 39 and 40, that the ionization edges (DFT), is a natural place for us to start.

Chapter 14
are a consequence of the transition of an atomic electron from a
tightly bound core-state to an unoccupied energy state above the
Fermi level. The energy of the initial core-state is well defined
and so the fine structure in the edge, ELNES, is related to the 14.2 Density Functional Theory (DFT)
details of the unoccupied electron states (see Fig. 40.2 in W&C).
There are also oscillations at higher energy above the edge onset
relating to the atomic co-ordination, called the extended ener- 14.2.1 Introduction to DFT
gy-loss fine structure (EXELFS), but we will not discuss this
topic. ELNES is essentially equivalent to X-ray absorption near
edge structure (XANES), which you’ll also see called near-edge The problem of calculating the ground-state electronic structure
X-ray absorption fine structure (NEXAFS). Some of the calcula- is one that is encountered in many branches of chemistry and
tion methods we discuss in this chapter were initially developed solid-state physics, and is a highly developed field. There are
for comparison with XANES measurements. several approaches with varying levels of sophistication, speed,
and accuracy. If we are to calculate the ground-state electronic
On the Border structure, we must determine the behavior of all of the electrons
in the material. That is, we must find a solution to Schrödinger’s
equation for every electron in the system, taking into account
The border between the core-loss and valence regions of the interaction with every other electron and with the underlying
the spectrum is somewhat nominal. A low-energy ioniz- atomic nuclei. This is not trivial!
ation edge could arguably be described as a high-energy
interband transition. 9 The first approximation we must make is to assume that the nu-
clei are ‘frozen’ and that they are not influenced by the electrons.
They only contribute a fixed, background charge distribution
The low-loss region measures excitations of the valence electrons. and associated potential. We call this the Born–Oppenheimer
These can be single-electron transitions from a valence state to an approximation.
unoccupied state, i.e., interband transitions, or a collective oscilla-
tion of the valence electrons called a plasmon. Unlike core states, Then we can write the time-independent Schrödinger equation
the valence states are not in a narrow, well-defined energy range. (seen in W&C, Eq. 14.1) as:
Rather there is a broad band of possible initial states. Interband
transitions are related to the details of both the valence band and
2 3
X „2 1 e2
the unoccupied states (the conduction band), and so this region of
X X
4− r2 + ˇ+ Vext .ri /5  .fri g/
the spectrum can be quite complex. In addition, interband transi-
ˇ
i
2m ri 8"0 ˇri − rj ˇ
i
ij.i¤j/
tions and the plasmons can have a similar energy of excitation and
they influence one another in the way they appear in the spectrum. = E .fri g/ (14.1)

You will see that the connection between the valence EELS and where the term in square brackets is the Hamiltonian, H. The first
the electronic structure is not straightforward and so interpreta- term in the Hamiltonian is a kinetic energy term for the electrons
tion can be difficult. The difficulty of interpretation is one of the and Vext is the potential energy term for each electron due to the
reasons why valence EELS has not been as popular as ELNES nuclei. The second term describes the inter-electron repulsion
for measuring electronic structure. However, with the introduc- between the electrons.
tion of electron monochromators, which can provide high-en-
ergy resolution spectral data in the low-loss region, there is a The big problem with Eq. 14.1 is that it represents a complex
resurgence of interest in measuring and interpreting this part of many-body problem, with the wave-function ψ({ri}) describing
 408 14 Calculating EELS

the system as a whole. It is a many-body wave function that as if they were the true electron wave functions, sometimes with
depends on the coordinates, ri, of every single electron. That is, surprising success.
there is one equation that is dependent on N variables, where N
Part I

is the number of electrons in the system. We can only solve this Fictitious Wave Functions
equation exactly for the hydrogen atom.
The Kohn–Sham wave functions generated by DFT are
Computational Methods
not the true wave functions that would describe the be-
havior of each individual electron in the solid. The elec-
DFT belongs to a class of computational methods referred tron density is, however, the true electron density. 9
to as first principles methods or ab initio methods. Since
the Latin term ab initio means ‘from the beginning’, the
idea is that the calculations build only on the laws of The last term in the Kohn–Sham equation, Vxc, is an effective po-
Chapter 14

physics without reference to empirical parameters. 9 tential which depends on n(r) and is called the exchange-correla-
tion potential. It aims to take into account that the electrons do
not move independently and includes all the difficult many-body
We can simplify Eq. 14.1 in a few different ways to deal with parts of the problem. The Vxc is that part of the potential that has
systems with more than one electron. The underlying principle been neglected to obtain the non-interacting parts in the Ham-
is to reduce the many-particle Schrödinger equation to a set of iltonian of Eq. 14.2. Exchange is associated with the fact that
single-particle equations, where the wave functions involved in electrons of the same spin must be separated (the Pauli exclusion
each of these equations describes only one electron. The most principle, W&C, Section 40.2.C). The remainder of Vxc takes
widely used approach for solids, where there are very large num- account of the fact that the electrons do not move independently;
bers of electrons, is DFT. their motions are ‘correlated’ with one another.

Using DFT we assume that all aspects of the electronic structure As you can imagine, the exact choice for the exchange-correla-
of a system of interacting electrons are completely determined tion potential does not have a straightforward solution. It could
by the ground-state electron density n(r). We then rewrite the be said that we have traded the impossible problem of solving the
problem as N separate equations, each for a single wave func- N-electron Schrödinger equation for the impossible problem of
tion, ψn(r): knowing what Vxc should be. Fortunately, some relatively simple
descriptions for Vxc have been found to give reasonable results
n.r0 / 0
 2
e2
Z 
h
− r2 + dr + Vext .r/ + Vxc .r/ n .r/ when compared to various experimental variables.
2m 4"0 jr − r0 j
= En n .r/ Reasons for Failure
(14.2)

The first term is again the kinetic energy term and Vext the poten- When DFT calculations fail to reproduce an experimental
tial energy due to the nuclei. The second term is now the classical quantity, it can be hard to know if this represents a failing
Coulomb contribution from the other electrons, assuming the due to the Kohn–Sham wave-functions not being the true
electrons are acting independently. wave-functions, a failure in the Vxc, or that the experiment
is not well described by a ground-state calculation. 9
The electron density can be written in terms of the wave func-
tions as
N
You can see from Eq. 14.2 and 14.3 that the solutions to the
X Kohn–Sham equations, ψn(r), also appear in the equation itself,
n.r/ = (14.3)

n .r/ n .r/
via n(r). Therefore, we must obtain the solution by an iterative
n=1
process. We make an initial guess for n(r), then solve the Kohn–
Sham equations to find ψn(r), and a new n(r) is calculated using
We call Eq. 14.2 the Kohn–Sham equation, and the solutions, Eq. 14.3. We solve the Kohn–Sham equations again with this
ψn(r), are called the Kohn–Sham wave functions. We should re- new n(r), and so on, repeating the process until the difference
member that the Kohn–Sham wave functions are not the true between the solutions for successive iterations is negligible. This
wave functions for the electrons in the material; they are just is an example of a self-consistent field (SCF) approach; the final
wave functions that reproduce the correct density. That is, we density is consistent with the field generated by that density.
have described the system by a set of ‘fictitious’ wave functions
for non-interacting electrons but these wave functions have a The usual techniques for solving Schrödinger’s equation are also
density that is identical to the density of the ‘real’ electrons. De- used to solve Eq. 14.2 and find the ψn(r). Often, the solutions are
spite this fiction, we often use the Kohn–Sham wave functions described in terms of a basis of other simpler functions, bj(r),
 14.2 Density Functional Theory (DFT) 409
X
n .r/ = cnj bj .r/
j
(14.4)

Part I
and so solving for ψn(r) simplifies to finding the coefficients cnj
which we can do with matrix techniques. For a periodic material
you should work in reciprocal space and the Kohn–Sham wave
functions, ψn(r), will also be Bloch states (see Chap. 14 in W&C).

Starting the SCF cycle

A typical initial guess for the starting charge density is

one that is generated by adding together the isolated atom

Chapter 14
charge densities. 9

You can use a number of different approaches to find the ground-

state electronic structure using DFT. The differences between
them are in three areas: Fig. 14.2 Calculated Cu L3 ionization edges for pure face-centered-cubic copper.
The LDA and GGA approximations for the exchange-correlation potential, Vxc are
zz The choice of the expression for the exchange-correlation compared. The differences between the spectra are negligible. The spectra have
potential Vxc. been vertically offset from one another
zz Any approximations made to the electronic and/or external
potential.
zz The choice of the basis functions.
metal compounds and in materials with 4f and 5f electrons. In
those instances, we expect strong correlation between electron
behavior, so approximations based on a homogenous electron
14.2.2 The Exchange Correlation Potential gas may not be accurate. A related problem is that the classical
Coulomb part of the Hamiltonian includes a non-physical self-in-
teraction contribution – the electron experiences a Coulomb re-
The choice of what to use for exchange correlation potential, Vxc, pulsion from itself! This contribution needs to be corrected for in
in order to best describe the many-body effects has been called the Vxc term and it is not achieved with LDA or GGA. There are
the ‘holy grail’ of electronic structure calculations. It is an area more sophisticated approaches to describe exchange and correla-
of ongoing research that awaits a satisfactory answer. tion (see the review by Kümmel and Kronik (2008)), but many
of these are computationally expensive. There has been recent
The simplest way to include exchange and correlation is called promise in a new Vxc called the modified Becke–Johnson poten-
the local density approximation (LDA). In LDA we assume that tial (TB-mBJ), which does not add a significant computational
n(r) varies slowly with position and that the contribution to the burden. There has been little other exploration of the effect on
Vxc from a small region of space is the same as it would be in a the ELNES of going beyond LDA or GGA for the Vxc.
uniform electron gas with the same electron density. A version of
LDA, called the local spin-density approximation (LSDA) uses
different densities for electrons with different spins. The slightly
more complex generalized gradient approximation (GGA) is 14.2.3 Approximations to the Potential
more accurate in many situations and so is widely used. GGA
includes not just the local density, but also the local gradient in
the charge density. When we include all of the electrons in the system, as outlined in
Eq. 14.2, it is called an all-electron or full-potential calculation.
Both LDA and GGA have been quite successful in predicting Not surprisingly, such all-electron calculations are quite compu-
many material properties and, so far, nearly all ELNES simula- tationally intensive. Despite today’s high-speed computers, the
tions using DFT have relied on one of these approximations. For calculations are prohibitively slow for systems with large num-
ELNES calculations, as shown in Fig. 14.2, you can see that the bers of atoms. One way we can improve the speed of the calcu-
differences between LDA and GGA are minimal. lation is to simplify the expressions for the nuclear and electronic
potential (the second and third terms in Eq. 14.2). Originally,
However, we find that LDA and GGA do not accurately model when there was not the current level of computational power,
highly localized electron states, such as those in some transition full-potential calculations were not used at all.
 410 14 Calculating EELS

Writing Now, Calculations Are Slow, But …

Part I

Although at the time of writing full-potential calculations

for crystal structures with more than 100 atoms in the unit
cell are prohibitively slow, the rapid transformation in
computing power in the past decade has revolutionized
the field. It could be anticipated that 2025 will see us
performing spectroscopy calculations of a sophistication
that is unimaginable now. 9

One of the earliest approximations was the ‘muffin-tin’ poten-

Chapter 14

tial (see Sect. 40.5.A of W&C). We divide the space around the
atoms into spherical regions where the potential is chosen to be
the atomic potential, which we calculate by solving Schröding-
er’s equation for an isolated atom. Between these regions, the
‘interstitial’ region, we assume the potential is zero or constant. Fig. 14.3 Schematic diagram illustrating the full potential, V(r) versus the pseu-
This type of muffin-tin potential is not a fully SCF approach dopotential Vp(r) as a function of distance, r, from the center of the atom. It can
since the potential cannot be consistent with the potential that be seen that using the pseudopotential removes the highly oscillatory part of the
we would calculate from the wave functions. We can bring a wave functions φ(r) near the core, replacing them with much smoother wave
degree of self-consistency into the muffin-tin potential by using functions, φps(r), which will be more easily described by a plane wave basis set.
a spherically averaged potential in the muffin-tins, derived from The potentials and the wave functions are identical beyond a core radius, rc (from
the wave-functions. You have probably deduced that this muf- Singh and Nordstrom 2005)
fin-tin potential will not accurately describe the bonding between
the atoms when it is highly directional in nature! For this reason,
muffin-tin potentials are not considered a good representation Pseudopotentials also offer us an additional benefit. The highly
of the real potential and are no longer widely used in DFT cal- oscillatory nature of the wave functions in the core region is
culations. replaced by smooth functions and this allows for plane-wave
basis sets to be used (see Sect. 2.4 below). Pseudopotential meth-
When we bring isolated atoms together to form a solid, the core ods are significantly faster than full-potential methods and for
electrons remain relatively unaffected. However, in the full po- this reason are very common. They are particularly useful for
tential description of Eq. 14.2, these core electrons are included. systems with large numbers of atoms. We use pseudopotential
One way we can considerably decrease the computational effort calculations widely for structure relaxations and in molecular
is to exclude these core electrons from the calculation by re- dynamics simulations. An obvious issue arises when simulations
placing the nuclear Vext near the atomic core with a pseudopo- of the ELNES are required; by using a pseudopotential we have
tential that describes the nuclear and electronic potential of the not explicitly calculated the core states.
core states. Since the word ‘pseudo’ means false, fraudulent,
or pretending to be something it is not (!) what we are doing Of course, there are numerous approaches to constructing the
is falsely to move some of the terms in the inter-electron part pseudopotential. Originally, we used empirical approaches, but
of the Hamiltonian into Vext and fewer electron states need to most modern methods try to construct the pseudopotential from
be calculated. The pseudopotential in the core regions should first principles. Determining the best way to construct the pseu-
evolve continuously into the real potential away from the core. dopotential is a non-trivial business, and you must make some
Figure 14.3 shows an illustration of how the potential and wave assumptions. For example, you must decide which electrons to
functions are changed when a pseudopotential is used. include in the pseudopotential and what the ionization state of
the atom is.
The Full Potential

One of the advantages of the full-potential approaches 14.2.4 Basis Sets

for a novice user is that it is not necessary for you to
come to grips with the subtleties of choosing and using
pseudopotentials. For this reason, many of us are willing Which basis set, bj(r), we choose is one of the defining charac-
to accept the associated compromise on computational teristics of a calculation method and it is often the origin of the
speed of full-potential calculations. 9 acronym used for that approach. The quality of the basis set can
be important in determining the accuracy of the calculations.
 14.2 Density Functional Theory (DFT) 411

Four important characteristics of a basis set are a straightforward conceptual link between the bonds formed
between atoms and features in the spectrum. MO calculations
zz Completeness – a basis set is complete if the real wave-func- are widespread in chemistry, but have only found limited ap-

Part I
tion can be expressed exactly by a linear combination of the plication in ELNES simulations.
basis functions.
zz Efficiency – the number of basis functions that are needed to The orbitals that we find as solutions to Schrödinger’s equation
achieve a given level of accuracy. in a muffin-tin potential can themselves be used as a basis, in a
zz Simplicity – the difficulty in using the basis set to solve so-called linear muffin-tin orbital (LMTO) approach, and can
Eq. 14.2. give relatively accurate calculations. Originally we used the muf-
zz Bias – whether or not the basis set favors certain regions of fin-tin potential and muffin-tin basis together, but in recent years
space over others. it has been more common to use the muffin-tin orbitals as a basis,
but use a full-potential description. Nevertheless, this approach
Obviously, completeness and efficiency are interlinked; a com- is not widely used.

Chapter 14
plete basis is likely to include a very large number of basis func-
tions and, therefore, will be inefficient. Plane wave basis sets are attractive because we can express any
function as a combination of plane waves; i.e., an infinite set
Clear candidates for basis functions are atom-like orbitals (AOs) of plane waves is a complete description. However, for the full
because the wave functions in the solid may not be all that dif- crystal potential this approach is not efficient. We need large
ferent than those found in an atom. The final wavefunctions will, numbers of plane waves to model rapidly varying wavefunctions
therefore, be a linear combination (LC) of atomic orbitals, hence in the region near the nucleus. Hence, we usually combine plane
the acronym (if you can say it) as the LCAO approach. AOs are wave basis sets with pseudopotentials as we described above.
a complete basis if all possible AOs are included. Usually, we The number of plane waves used in the basis set is usually de-
use only a subset of AOs, limiting the accuracy. fined by the ‘cut-off energy’. The cut-off energy is the energy
that a free electron would have, if it had the same wavevector as
The tight-binding approximation, which is discussed in a num- the largest wavevector used in the basis set. You’ll find there are
ber of books and articles, is a variation on the LCAO method. a wide variety of readily available plane wave-pseudopotential
It uses a small set of atom-like basis functions and we make programs available for the calculation of electronic structure and
additional approximations in solving the Schrödinger equation. materials properties. Of these, CASTEP is currently the only one
Essentially we assume that only interactions between neighbor- that we are aware of that has the ELNES calculations already
ing atomic sites significantly influence the bonding properties. built in and well supported.
This is not considered a fully ab initio method and is not gener-
ally used to model ELNES. However, it is computationally fast Plane wave basis sets are also not particularly good at repre-
and has been used to successfully predict a number of material senting tightly bound valence states such as d or f electrons.
properties. There are a group of calculation approaches that overcome this
limitation by adding to (or augmenting) the plane wave basis
We can use a variant on the LCAO method for the simulation set to better model the tightly bound states. In these so-called
of ELNES edges. In this approach we represent the AOs them- augmented plane wave (APW) methods we divide the crystal
selves by a set of Gaussian-type orbitals. The core states are into two regions: i) spheres centered on the atom sites, and
eliminated from the problem by an orthogonalization process, ii) the remaining interstitial region. Within the spheres and in
which speeds the calculation, and, not surprisingly, this method the interstitial region we have to use different basis sets. You
is called the orthogonalized LCAO (OLCAO) method. However, should note that no similar division is necessary for the po-
if we removed all of the core states we would prevent the full tential – the full potential can be used. It is only the nature
ELNES being calculated. You can use a supercell and retain the of the basis set that changes between the inside and outside
core states in the excited atom to overcome this limitation. You of the spheres. Inside the spheres, we use radial solutions to
can find further details on the OLCAO method in the review by Schrödinger’s equation, multiplied by spherical harmonics as
Ching and Rulis (2009). the basis (an atom-like expansion). In the interstitial region we
use plane waves. In the original scheme the basis functions de-
When atoms are brought together to form a molecule, the pended on the energy, leading to a non-linear problem that was
atomic orbitals (AOs) of the isolated atoms interact with each computationally expensive. So, when a linearization scheme
other to form new molecular orbitals (MOs). Examples include is introduced, it allows for energy-independent basis functions
the π and σ states in C-based materials and the eg and t2g states to be set up (hence the initials LAPW). When we combine an
in transition metal complexes. For a molecule or solid, if we LAPW basis set is combined with a full-potential calculation
assume that the nearest-neighbor interactions are the most im- we have FLAPW. The WIEN2k code is a FLAPW method and
portant, these MOs make a natural choice for basis functions. is widely used for ELNES calculations. In recent versions the
We express the MOs as the sum of a relatively small set of option of including additional local orbitals in the basis can
AOs. You’ll find that MO approaches offer the advantage of improve the performance of the basis set.
 412 14 Calculating EELS

14.2.5 The Korringa–Kohn–Rostoker (KKR) The differential cross-section (introduced back in W&C,
Method Sect. 2.4) for this interaction can be determined using Fermi’s
golden rule.
Part I

ˇ ˇ2
The KKR method relies on a different approach to solving d2 
 2  ˇZ ˇ
4 k ˇˇ X ˇ
the Kohn–Sham equation than that outlined above. Instead = 2
 
f .fr i g/ exp.iq  rj / i .fr i g/dr 1 : : : dr z
ˇ
d˝dE a0 q4 k0 ˇ ˇ ˇ
of expressing the wave-functions as a linear combination of j ˇ
basis functions and then solving a matrix equation, we use 2
ˇ ˇ ˇ2 ˇ
4 4 2 k ˇˇk+ E
  2  ı.E ˇZ ˇ i − Ef / X ˇ ˇ2
d 
scattering theory to find the solutions. You can imagine a va- d2 =
Z ˇ ˇ
= 2 4
ˇ  .fr g/ X
i
f  .fri g/ exp.iq  r
exp.iqj / ri .fr
/ i g/dr
.fr 1 : : : dr
: : z
:
ˇ ˇ
ˇdrz ˇ
lence electron as moving throughout the crystal and beingd˝dE a q k 1
2 4 0
ˇ ˇ
f  j i i g/dr
d˝dE 0 a0 q k0 ˇ ˇ ˇ j
j
ˇ ˇ
scattered by the crystal potentials of the atomic nuclei. By
analyzing the scattering process you can derive the energy of ı.Eı.E+ E+i −EE−f /E / (14.5)
i f
the electron states.
Chapter 14

The KKR method belongs to a class of methods for solving dif- where a0 is the Bohr radius and γ is the relativistic factor. You
ferential equations that is known as Green’s function methods. should note that this equation describes the excitation of the tar-
By use of a special function called a Green’s function, we trans- get and so the wavefunctions, ψi and ψf, are the many particle
form the Kohn–Sham equations from the differential form of states that describe the target as a whole. They are a function of
Eq. 14.2 into an integral form. We can understand this new inte- the co-ordinates {ri} of all of the electrons in the atom or solid.
gral expression in terms of scattering concepts and the Green’s Equation 14.1 is for a particular transition, so if there are a num-
function does have a physical meaning as a ‘propagator’. The ber of transitions with the same energy then you need to make a
KKR method, with the reliance on scattering theory, has links summation over all the transitions with energy Ef − Ei.
to the multiple scattering method (Sect. 14.3.4), as this is also
a Greens function method. Indeed, the KKR method and the In order to proceed further we make the one-electron approxima-
multiple scattering methods are formally equivalent. tion. The excitation is approximated as a transition of a single elec-
tron from an initial to a final excited state. Equation 14.5 becomes:
Solution with KKR is relatively simple if the potential has a ˇ2
d2 
 2  ˇZ
muffing-tin shape. Application of KKR methods to calculating =
4 k ˇˇ
 .r/
exp.iq r/ .r/dr
ˇ
f
ELNES/XANES have, so far, only used muffin-tin potentials.
 i
ˇ
(14.6)
d˝dE 2 4 k0
a0 q ˇ ˇ
Full-potential KKR methods have been developed but we ı.E + Ei − Ef /
await further publications on ELNES simulations. However,
KKR methods are finding increased application in other elec-
tronic structure calculations, particularly for defect states in The wavefunctions and energies are now one-electron wavefunc-
materials. tions and energies, with ψi and Ei describing the initial state and
ψf and Ef describing the final state for the electron undergoing
excitation. The initial state is a core-electron state, with a readily
determined atom-like wave function and energy. We then make
14.3 Calculations of the ELNES further approximations in order to determine the final state wave
function and energy. In the independent electron approximation
we assume that the ionization of the core electron has no effect
14.3.1 ELNES Theory on the other electron states in the solid. That is, the ground-state
unoccupied electron states are used for ψf. If you had used a
DFT calculation to obtain the ground state electronic structure,
Once you have a suitably accurate ground-state electronic the Kohn–Sham wavefunctions have to be used, even though,
structure, you then have to predict how the fast incident elec- strictly speaking, these are not the real electron states.
tron will interact with this ground state and undergo the energy
losses that produce the ELNES. A good first approximation is Perhaps not surprisingly the independent one-electron approx-
to assume that the incident electron is sufficiently energetic imation does not actually hold very well to describe the EELS
compared to the interaction energy that we can approximate spectrum of many materials; i.e., the ionization of the core state
it as a plane wave. (We call this the first Born approxima- has significant influence on the other electron states in the mate-
tion). The fast electron has an incident wave-vector k0 and rial. These multi-electron effects are called core-hole effects and
is scattered into a solid angle, dΩ, with a wave vector k. The we will discuss these in more detail in Sect. 14.3.2.
momentum transfer is q = k − k0. The target undergoes a tran-
sition from an initial state ψi, with energy Ei to a final state ψf, Equation 14.6 is the differential cross-section for a particular tran-
with energy Ef. The corresponding energy lost by the incident sition, whereas an EELS spectrum will be described by a summa-
electron is E = Ef − Ei. tion over all possible transitions and in some of the literature this
 14.3 Calculations of the ELNES 413

summation is explicitly included in Eq. 14.6. In a solid there is a there are suitable electron wavefunctions and energies, you should
continuum of unoccupied states and possible transitions. Equa- convert these to EELS spectra by the calculation of Eq. 14.6. Only
tion 14.6 is often rewritten as product of the ‘matrix term’ M(E) a few of the available DFT programs include this option; for the

Part I
and a density of states (DOS) term, ρ(E) (see W&C, Sect. 40.2). others you will need to so some further programming.
d2  Many users are understandably reluctant to take this next step,
/ jM.E/j2 .E/ (14.7)
d˝dE which may account for why experimental spectra have some-
times been directly compared to the unoccupied DOS. The ma-
Strictly, this approximation to Eq. 14.6 can only be done if we trix term in Eq. 14.7 and 14.8, which is an overlap integral, will
make an average over the direction of q, which fortunately oc- be slowly varying in energy. This means that the intensity in
curs in most experimental acquisition conditions. the ionization edge is approximately proportional to the unoc-
cupied DOS, and so such an approach has some justification.
Another approximation that we sometimes make is the dipole Figure 14.4 shows how the influence of the matrix elements on

Chapter 14
approximation where it is assumed that exp(iq · r) ≈ 1 + iq · r; the prediction of the ionization edge intensity can be quite small.
i.e., the higher terms in the Taylor expansion are neglected. The So some users simply broaden the calculated DOS to account
orthogonality of the states gives zero for the first term in the for experimental broadening and for broadening due to the finite
expansion and the matrix term becomes; lifetime of the excited states.
Z
M.E/ = f .r/.q − r/i .r/dr (14.8) Just Calculate It

A consequence of the dipole approximation is that dipole selection Although it may not be necessary to calculate the final
rules will apply as discussed in Chap. 40 of W&C. This means that spectrum in order to interpret your ELNES (DOS alone
the change in angular momentum number, ∆l, must be equal to may be sufficient for this purpose), given the ready avail-
± 1. For example, for K edges, where the initial state is an s-state ability of well-documented and supported computer
(l = 0), only transitions to unoccupied p-states (l = 1) are allowed. codes and the speed of modern computers, there is not
The dipole approximation will only hold for small scattering an- really any excuse for neglecting to do so. 9
gles (i.e., small q). It can be possible for non-dipole transitions to
appear in the spectrum under certain experimental conditions. In
addition, the initial state, ψi is a core state and is localized on the ELNES involves a transition from a core state to an unoccupied
atom undergoing ionization and so the matrix term (in the form state, so you have an immediate problem with using a pseudo-
of an overlap integral between the initial and final states) will potential method for ELNES calculations. The core states have
only have a significant value when final state, ψf, also has some not actually been calculated; restricting us to comparison to the
magnitude at the excited atom. Therefore, when you measure the unoccupied DOS only and neglecting the matrix term. This lim-
ionization edge intensity, you do not measure the entire (unoccu- itation has been overcome for you in the plane wave pseudopo-
pied) DOS, but a site- and symmetry-projected DOS. tential program CASTEP. This is achieved by using core states
from an all-electron atomic calculation and making corrections
If we assume that the atom is isolated and use the atomic bound to the pseudopotential wavefunctions using a method called the
and continuum states as ψi and ψf in Eq. 14.6, then we simply cal- projector-augmented wave (PAW) approach.
culate the cross-section for the ionization. This is the cross-sec-
tion that we use for quantitative compositional analysis in EELS, For anisotropic materials, the measured ELNES may depend on
as described in detail in W&C, Chap. 39. This cross-section ne- the orientation of the electron beam relative to the crystal axes.
glects all the bonding effects between atoms and does not repro- You include this effect when you calculate Eq. 14.6 explicitly.
duce the details of ELNES. Integration over the different momentum transfers, q, associated
with the finite convergence and collection angles should also be
To reproduce ELNES we must include the bonding effects be- included. If you are comparing spectra to the unoccupied DOS
tween the atoms. As the initial state for the ionization edges is a only, then you must use the appropriate symmetry-selected DOS.
well-defined core-electron state, Eq. 14.6 reveals that theoretical Equation 14.6 does not fully account for relativistic effects in
simulation of ELNES boils down to working out all the possible anisotropic materials, and so we have to modify it. As a result,
electron wavefunctions and energies for the unoccupied states in for example, we obtain a better prediction or the collection angle
your specimen. DFT allows you to calculate the electron wave- at which orientation effects can be ignored (the so-called ‘magic
functions in a material, and so there is a plethora of possible cal- angle’). The most recent versions of ELNES calculations imple-
culation methods and computer codes. However, you’ll find that mented in WIEN2k can perform the full relativistic calculations
some of these are not suitable for EELS analysis, such as when for anisotropic materials, as a function of scattering angle or
the results cannot be decomposed by site or symmetry, or where collection angle. This software also allows calculations beyond
the core or the unoccupied states are not calculated. Even when the dipole approximation to be performed.

Licensed to ([email protected])
 414 14 Calculating EELS

14.3.2 The Core Hole

Part I

All of our discussion so far has been on the basis that Eq. 14.6
is an accurate representation of the excitation process. That is,
we assume that the excited electron makes a transition from a
core state to some unoccupied state and all of the other electrons
in the material do not notice. Not surprisingly, it turns out that
this is often not a good approximation. As described in W&C,
Chap. 40, the empty state left behind (which we call the ‘core
hole’) acts as an effective positive charge that will influence the
other electron states, including the final state of the electron that
has been ejected. This influence occurs because the relaxation
Chapter 14

time for the core hole (i.e., the time it takes for this hole to be
filled by another electron) is much longer than the time required
for the original ionization excitation.

The electron that is ejected into the unoccupied states ‘sees’ the
core hole, as do all the other electrons in the material. In severe
cases the ejected electron and core hole can form a bound-pair Fig. 14.4 The Cu L23 edge for pure face-centered cubic copper. The influence
state called an exciton. In the best case, such as a free-electron of the matrix term, M(E), on the shape of the spectrum is illustrated and in
like metal, the valence electrons are relatively free and move this instance is shown to be minimal. The shape of the ionization edge is well
around to compensate for the core hole. We call this process described by the unoccupied d-DOS,ρ(E). Although transitions to s-states are
also possible, the matrix term for these transitions is very small (not shown),
‘screening’. So the effect of the core hole depends on your spec-
and only the d-DOS needs to be considered. Upon addition of a broadening
imen! If it occurs, it generally shifts intensity in the ionization
term, the agreement with experiment is shown to be quite reasonable. In this
edge from higher energies towards the edge onset, due to the
instance the energy scale has been given with respect to the Fermi energy, not
attractive nature of the core hole. the ionization edge onset

If we were to attempt a full and correct description of the excitation

process we would actually find it very difficult because it would in- agreement is obtained with a full core hole for AlN but half a
volve having to solve a time-dependent, many-particle Schröding- core hole for GaN. For copper, as we saw back in Fig. 14.4, rea-
er-like equation. For ELNES we usually take the approach that sonable agreement was observed without a core hole at all. Most
retains the simplicity of Eq. 14.6 but manages to capture much of commonly, a full core hole is used in the literature for ELNES cal-
the physics of the excitation process. We approximate the excited culations of non-metals. Therefore, your choice of how much of
state by calculating the final state wave function in the same way a core hole to include depends on the material. This arbitrariness
as is done for a ground state calculation, but with the core hole in the choice for how to include core hole is rather unsatisfactory
already in place. This approach is called the sudden approximation and detracts from the predictive capability of our calculations. You
or the final-state rule and you’ll find that it often works quite well. should note that the supercell used in the calculations of Fig. 14.5
was relatively small, and there is evidence that with a larger su-
For muffin-tin or pseudopotential calculations this approxima- percell half a core-hole may also provide good agreement with
tion can be achieved by replacing the excited atom by one with experiment for AlN. This example nicely illustrates some of the
an extra proton, which we call the (Z + 1) approximation. In a difficulties and subtleties that you must consider when performing
full-potential calculation an electron can be removed from the a core-hole calculation. Although AlN and GaN are very similar
appropriate core state of the excited atom and placed into a va- materials, the Ga 3d electrons probably provide additional screen-
lence state or considered as a background charge. When you use ing of the core hole, allowing smaller supercells to be used.
calculations relying on the periodicity of the lattice, you should
perform the core-hole calculation on a supercell that is much The Core Hole
larger than the primitive unit cell, so that the core holes are far
enough apart to not interact. Sometimes we use a Slater transition
where only half an electron is removed from the core state. We Including a core hole into a calculation is a significant
could justify this in a hand-waving manner by thinking of the simplification of the full many-body excitation process
electron as being in transition, i.e., ‘half in and half-out’ (see of EELS. In due course we can expect that the rigorous
Paxton et al. 2003 for a more rigorous explanation.) calculation of these effects will become commonplace.
In the meantime you will find that core-hole calculations
Figure 14.5 shows some calculations of AlN and GaN, with and often perform well in reproducing the main features of
without a core hole, and both compared to experimental spectra. experimental spectra. 9
As you can see, under the same calculation parameters, the best
 14.3 Calculations of the ELNES 415

The question of whether to use a full or partial core hole ap-

pears to be one with little consensus in the literature. Remember
that both approaches, in describing a static situation, are only

Part I
an approximation to the full, dynamic, many-body excitation
process. If the sudden approximation were indeed correct then,
in principle, a DFT calculation with an exact Vxc and a full core
hole in place should accurately predict the ELNES. It may be
that a partial core hole acts to compensate for the dynamic nature
of the excitation where the full screening processes of the other
electrons have not had time to sufficiently develop. In order to
know how to predict this effect, we would need to determine the
dynamic screening properties of the system, not just the static
screening. The other question of whether a more accurate Vxc

Chapter 14
could improve core-hole calculations is a topic that may warrant
further study.

Fundamentally, when we use the single-particle approximation

to go from Eq. 14.5 to Eq. 14.6 we do not correctly describe the
excitation process because many-body effects can be important.
There are cases where, when we introduce a core hole, we still
fail to predict the spectral shape adequately. Such cases include
the L23 edges of transition materials where there is strong cou-
pling between the core-hole and the tightly-bound d-electrons. In
some light elements there are strong excitonic effects observed
in the K-edges. Specifically, in these cases, we cannot express
the final excited state wave function as a Slater determinant of
single-particle wavefunctions.

There are a number of approaches that go beyond the indepen-

dent one-electron approximation in a more rigorous fashion. At
the current time (2015) these are being developed and applied
primarily for the excitation of valence electrons (i.e., for opti-
cal absorption and low-loss EELS), and we will discuss these
very briefly in Sect. 14.4. They are considerably more compu-
tationally complex and more time consuming, but calculations
of ELNES based on these methods are beginning to appear. We
anticipate, given the ever-increasing computational power, we
should see more of these rigorous inclusions of multi-electron
effects in the near future.

14.3.3 Multiplet Theory Fig. 14.5 Calculated and experimental spectra for (a) AlN and (b) GaN. Calcu-
lations with no core hole, half a core hole (Slater’s transition-state approxima-
tion), and a full core hole (final state approximation) are all shown. For AlN, best
The application of multiplet theory to describe the ELNES in agreement with the experimental spectrum is obtained using a full core hole,
solids is a partly empirical approach that can include many-body whereas with GaN best agreement is obtained using half a core hole
effects and has been successfully applied for a number of years.
It has been particularly successful in describing the L23 edges
for the transition metals where the d-bands retain much of their
atomic character, and there is a strong interaction between the whole rather than as single electron transitions. For example, in
core wavefunctions and these narrow, partially occupied bands. the single-electron picture we describe the L23 edge for a first
In a similar way, there has been success in describing the M45 transition series element with N d-electrons as an electron mov-
edges (3d to 4f transitions) of the rare earth elements. ing from a core 2p state to one of the unoccupied d states. In
an atomic picture, the initial state for the entire atom has the
In multiplet theory we return to Eq. 14.5 to describe the exci- electron configuration of 2p63dN and it is excited into a final state
tation process, and we consider the excitation of the atom as a configuration of 2p53dN+1.
 416 14 Calculating EELS

cessfully included into atomic multiplet theory. If crystal-field

effects are only a perturbation of the atom-like description then
we can obtain good agreement with experiment. The disadvan-
Part I

tage is that you need to know in advance how strong the effect
of the crystal field should be, and make a decision via compar-
ison with experiment. This is the reason why we describe the
method as semi-empirical. Similarly, the addition of delocalized
states can be included but we must make some prior assump-
tion about the nature of the bands. Comprehensive reviews of
the multiplet approach are available (see the two by de Groot
2001, 2005), and Fig. 14.6 compares a multiplet calculation
with a DFT calculation. You can find a freely available program
called CTM4XAS for the multiplet calculations of transition
Chapter 14

metal L edges.

14.3.4 Multiple Scattering (MS) Methods

MS methods are conceptually quite different from the band-struc-

Fig. 14.6 The Ca L23 ionization edge for CaF2 comparing the DFT (ground state ture approaches described above. In the MS method, we consider
(GS) and core hole (ch)) calculations with multiplet calculations and with the the electron ejected from the core state as an outgoing spheri-
experimental spectra. For this material, the multiplet calculations give excellent cal wave, where the wavelength is determined by the energy.
agreement with the experimental data, whereas the DFT calculations, even with This outgoing wave is then elastically scattered by the potential
a core hole, are unsatisfactory of the surrounding atoms, with a phase shift. The interference
between the outgoing and scattered waves, back at the site of
the excited atom, produces the oscillation in the ionization edge
An appropriate starting point for us is the atomic Schrödinger (recall Fig. 40.1 in W&C).
equation, and we need to explicitly consider spin–orbit cou-
pling. Because there are a number of p and d states in the atom, If the energy of the excited electron is high, it has a short mean-
you’ll see there are a number of possible alternate configu- free path and only one scattering event is likely to occur before
rations, or multiplets, for the electrons in the atomic states, interference is observed at the site of the excited atom. This sin-
designated by the values for the total, orbital and spin moments gle scattering gives rise to the EXELFS. At lower energies, i.e.,
(J, L, and S). Therefore, when the atom makes a transition, closer to the edge onset, the mean-free path of the outgoing elec-
such as from a 2p63dN to a 2p53dN+1 state, there are a number of tron is much longer and therefore many multiple-scattering pro-
possible transitions that it can make, where some of these will cesses can occur (giving the ELNES). To calculate ELNES fully
be degenerate in energy. Whether a transition actually occurs we must consider the full multiple-scattering processes. One way
between the different possible states will also depend on se- for you to think of the problem is to look back to Eq. 14.6 – the
lection rules. Peaks appear in the spectrum associated with the final state of the excited electron is determined by the way it is
energies of the possible atomic transitions and their intensity scattered by the surrounding potential of the crystal.
will be related to the degree of degeneracy of the initial and
final atomic configurations and on the matrix element term for The Green’s function, G(r, r′, E), introduced when we discussed
the transition. the KKR method, has an intuitively clear role in MS theory as
a ‘propagator’. This function describes the propagation of the
In solids, the neighboring atoms can have a significant effect electron of energy, E, from position r to a different position r′.
on the energy of the atomic states. The L and M edges for the We can the rewrite Eq. 14.6 in a form that includes the G(r, r′,
rare-earth materials are well described by the atomic approach, E) and the initial states, ψi.
but for the transition metals we must introduce the effect of the
d2 
Z
neighboring atoms, especially for the 3d series. Firstly, the pres- / Im i .r0 .q  r0 /G.r; r0 ; E/.q  r0 /i .r/dr0 (14.9)
ence of the neighboring atoms changes the local symmetry of the d˝dE
excited atom (the crystal-field or ligand-field effects). Secondly,
the delocalization of the electrons must be considered because Predicting ELNES can be performed by finding Green’s func-
they are no longer confined to a single atomic site. tion, instead of explicitly calculating the final state wave-func-
tions, ψf. We derive the Green’s function from the Hamiltonian
By introducing the crystal field we have removed the degen- for the final state. This final state should include the screened
eracy of some of the states and these effects have been suc- potential due to the core-hole and an energy-dependent contri-
 14.4 Calculating Low-Loss EELS 417

bution called the self-energy, which accounts for the effects of

exchange and correlation between the final excited state and the
other electrons in the system.

Part I
We can express G(r, r′, E) as a summation over different scatter-
ing paths with free-electron propagators, G0, describing propaga-
tion between scattering centers (atoms) and scattering matrices,
t, calculated from the potential, at the centers. Inelastic losses
of the excited electron and the core-hole lifetime mean that the
range of the excited electron is limited, and so calculations can
be performed within a real-space cluster of finite size. For full
MS, which is required for accurate description of the ELNES,
we must include all of the possible scattering paths within the

Chapter 14
cluster.

In nearly all of the MS calculations applied to date, a muffin-tin

potential is used to describe the crystal potential and, from there,
the scattering matrices. Originally the muffin-tin potential was
derived from a superposition of atomic potentials, but more re-
cently the potential is determined self-consistently. This is possi-
ble because the Green’s function can also be used to calculate the
charge density n(r). Even though the potential is self-consistent,
you still make a muffin-tin approximation to the shape and this is
one of the limitations. However, comparisons between MS cal-
culations with muffin-tin potentials and full potential band-struc-
ture calculations have shown that the effects are usually quite
subtle and you would only expect them to occur for materials
with strongly directional bonding.

MS methods have been widely used for ELNES interpretation,

although they were originally developed for (and have been more Fig. 14.7 A multiple scattering calculation of the oxygen K-edge in rutile (TiO2)
widely applied to) XANES. With the equivalence of XANES and for increasing cluster sizes. The first four-atom cluster shows that the first peak
ELNES measurements, MS calculations are equally applicable in the spectrum is identified as scattering from the first coordination shell of
here, and you can find comprehensive reviews of the methods titanium atoms. The second peak~ 15 eV above the edge onset appear when
in the papers by Rehr. MS calculations can be more computa- the next shell of oxygen atoms is included and results from scattering from this
tionally efficient than band-structure methods, particularly when shell. However, the splitting and finer details in the spectrum are not revealed
core-hole effects are to be included and for complex or non-pe- until larger cluster sizes, when the crystal field splitting due to the longer range
riodic structures. The latest version of the FEFF code for MS symmetry becomes effective
calculations explicitly calculates ELNES (as well as XANES),
and allows you to include momentum-dependent and relativistic
effects. The cluster geometry of MS methods offers you a helpful proach cannot predict the plasmonic features of the spectrum,
way of interpreting the data. By calculating clusters of increas- but may be adequate if the plasmon energy is quite far from the
ing size, you can relate features in the ELNES to the shells of region of interest.
neighboring atoms around the excited atom. Figure 14.7 shows
an example of such an analysis. However, in general, we start with classical electrodynamics.
We describe our solid as a medium with a complex dielectric
function ε(q, E) = ε1 + iε2. The dielectric function describes how
that medium will respond to an external electric field (such as the
14.4 Calculating Low-Loss EELS one provided by the incident electron or from incident electro-
magnetic radiation). There will be an induced field in response
to the external field, and the dielectric function predicts what that
Calculating low-loss EELS could proceed from the independent induced field will be. The energy lost by the incident electron
one-electron approximation, as is done for ELNES, by consid- is due to this induced field and the differential cross section for
ering transitions from all of the possible valence states to the energy loss becomes:
unoccupied states. In ELNES the features in the spectrum are
d2 
   
connected to the unoccupied DOS, whereas for low-loss EELS 1 1 −1
= Im (14.10)
we have to consider the joint DOS (JDOS) instead. Such an ap- d˝dE  2 a0 m0  2 na  2 + E2 ".q; E/
 418 14 Calculating EELS

where v is the speed of the incident electron, na is the number that the optical community has developed can be applied for
of atoms per unit volume and θE is the characteristic scattering low-loss EELS, at least for small scattering angles where we can
angle (θE = E/γm0v2). The term Im(-1/ε) is called the loss func- safely approximate q = 0. In addition, some of the available codes
Part I

tion. You can also use optical response measurements to obtain allow for the calculation of non-zero q.
ε(0, E), where q = 0 because photons are unable to transfer mo-
mentum. The simplest approximation that we can make for the dielectric
function is given by:
Whilst we might be relieved to find there is a relatively simple
16 X
equation to describe low-loss EELS, the situation is not neces- "2 .!/ = jh jj c ij2 ı.Ec − E − „!
! 2 ;c (14.11)
sarily improved at all. For a bulk material, where the dielectric
function is known, it is a trivial matter to predict the EELS spec-
trum. However, if the dielectric function is already known, there where we are summing over valence and conduction band states
may not be a lot of point to doing the EELS measurement in the (given by the subscripts v and c respectively). The valence and
Chapter 14

first place because it doesn’t provide any new information! conduction wave functions and their energies can be obtained
from a ground-state DFT calculation. From ε2 it is a then sim-
The benefit of EELS over optical measurements is that it can be ple matter to obtain ε1 using Kramers–Kronig analysis and,
performed with nanometer resolution and so we can obtain the from there, the EELS spectrum. As was the case for Eq. 14.6,
response from particles, defects, interfaces, or other small struc- Eq. 14.11 also has the form of Fermi’s golden rule. At large
tures. Once structures are sufficiently small, the EELS response energy loss, ε2 is small and ε1 approaches 1, so Im(−1/ε) ≈ ε2,
will no longer be a bulk-like response. To proceed further in the returning to our earlier description of the spectrum being sim-
theoretical derivation, we have to divide space into regions of ply related to a (matrix modified) JDOS. You cannot ignore the
different dielectric functions with the assumption of perfectly influence of ε1 on the final EELS spectrum in most instances, as
sharp interfaces between them. We then solve Maxwell’s equa- is illustrated in Fig. 14.8.
tions, using the boundary conditions, to work out the induced
field (and thus the energy loss) produced by such a multi-com- There are three significant approximations embedded in
ponent system. You need prior knowledge about the dielectric Eq. 14.11:
function for the materials in the different regions. We can also
predict surface plasmons using such an approach. zz The random phase approximation (RPA).
zz Neglecting local field effects (LFE).
Analytical solutions are only available for relatively simple ge- zz That q → 0.
ometries, such as spheres or planar interfaces; Wang’s (1996)
text provides an overview of these. For anything more complex, You can find details about how to perform calculations with
you have to use numerical methods. There is considerable in- these approximations and the WIEN2k code in Keast’s (2013)
terest in this area at present due to the emerging technological paper. Incidentally, the name RPA is not particularly helpful for
significance of surface plasmons in applications such as sin- our understanding the origin of this approximation. However, if
gle-molecule detection and sub-wavelength optics. the ground-state wavefunctions and energies are used, this ap-
proximation can be more helpfully described as the independent
In the remainder of this section, we will address the question of one-electron approximation.
how to perform a quantum-mechanical calculation of the dielec-
tric function, ε. There are two situations where you might need Don’t Neglect LFEs in Nanostructures
to do this. Firstly, EELS measurements may have identified a
region of the specimen consisting of a metastable or previously
undiscovered phase, meaning the dielectric function will not be Do not use a computer code that neglects local field ef-
available from optical measurements. Then you would have to fects (LFEs) to calculate the optical response or valence
make a calculation in order to interpret your EELS spectrum. EELS of a nanostructure. 9
Secondly, a more general scenario would be when, for your ma-
terial of interest, you do not yet fully understood the dielectric
function in terms of the electronic structure and your calculations Neglecting LFE is equivalent to averaging the response function
are intended to develop this connection. over the unit cell. That is, we do not include the effects of local
variations in any induced field. It follows that this approximation
The optical community has been tackling the problem of cal- will be inappropriate when the responses of interest are those
culating the dielectric function from first-principles for many that are inhomogeneous within the unit cell. This is an important
years now, and considerable progress has been made. The key point, as any attempt to calculate the EELS or optical properties
reference in this area is the review paper of Onida et al. (2002), of a nanostructure using a periodic supercell will give incorrect
although it is not easy reading for the beginner. Once you have results if you do not include LFE. For bulk materials, the most
calculated the dielectric function, it is a simple matter to predict significant effects of the LFE are usually seen in spectral features
the energy-loss spectrum. The techniques and computer codes that are associated with the more tightly bound electron states.
 14.4 Calculating Low-Loss EELS 419

Part I
Chapter 14
Fig. 14.9 The calculated low-loss EELS versus experiment for rutile (Ti02). The
main spectrum shows the results including local field effects (LFE) whereas the
inset shows the result when the LFE are neglected. Overall the agreement be-
tween the experiment and calculation are good, both with and without LFE,
except for the M23 edge at ~ 46 eV, where the intensity is greatly overestimated
when LFE are neglected

ture, the sudden approximation, that we used for ELNES, is not

practicable.

There are two broad classes of approaches to going beyond the

ground-state RPA. One approach is based in many-body pertur-
bation theory and, in the first instance, can correct for the inter-
Fig. 14.8 Calculations using the RPA and neglecting LFE for silver. The (ma- action of the valence hole and the excited electron with the other
trix selected) JDOS/E2, the dielectric function (ε1 and ε2) and the calculated and states in the solid. These are called quasiparticle corrections.
experimental low-loss EELS are all shown. It can be seen that the peaks in the The quasiparticle wave functions and energies can be used with
JDOS correspond to peaks in ε2, as expected. However, although there are similar the RPA of Eq. 14.11. The outcome of quasiparticle corrections
peaks in the loss function, the relative intensities and positions have been shift-
is usually an approximately linear stretch and shift of the valence
ed [because Im(−1/ε) = ε2/(ε12 + ε22)]
and conduction band energies. These results lend credence to
using the so-called scissors corrections to ground-state RPA,
where we apply a simple empirical shift and stretch of the bands
That is, mostly at high energies in the spectra, as illustrated in in order to obtain the best agreement with experiment. In order
Fig. 14.9. to include excitonic effects we must solve a two-body interacting
problem, via the Bethe–Salpeter equation. The other class of
As an independent one-electron approximation, it is not unex- approaches to addressing the many-body problem is a time-de-
pected that Eq. 14.11 sometimes fails to provide an acceptable pendent reformulation of DFT (i.e., TDDFT).
description of the dielectric function and low-loss EELS. The
excited electron and the valence hole interact with the other In all cases, the calculations involve a substantial increase in
electron states in the solid. They also interact with each other in theoretical complexity and in computational effort. Although
the form of an excitonic interaction. That is, the real excitation considerable progress has been made for optical calculations and
process is a many-body interacting problem. As a consequence, computer codes are available to perform such calculations, at-
both the peak positions and intensities in the spectrum may be tempting to go beyond the RPA can only be recommended when
incorrectly predicted. In general we find that for metallic ma- you have a lot of experience in the field. The other difficulty is
terials the overall shape of the spectrum appears to be mostly that the quantum-mechanical computer codes for optical calcu-
correct, but a well-known failing of DFT for non-metallic ma- lations cannot, as yet, address the question of spatially resolved
terials is that they underestimate the band gap. Because of the EELS. So there is considerable room for new developments in
large number of possible initial states and their delocalized na- the calculation methods!
 420 14 Calculating EELS

Chapter Summary
Part I

In this chapter we have given you an introduction to the the- loss EELS relies on first calculating the dielectric function
ory behind the calculation of the fine structure in the EELS and, at the current time, most calculations also rely on an
spectrum. Calculation of the ground-state electronic struc- independent-particle approximation. However, the develop-
ture via DFT is the foundation on which most EELS simula- ment of methods to go beyond the independent particle ap-
tions are built. Choosing a computational approach relies on proximations is an active area of research and there are good
selecting which description of the crystal potential you want prospects for further improvements in EELS simulations.
to use and its corresponding basis set. Most ELNES calcula- Finally, you may have encountered entirely new vocabulary
tions use the independent one-electron approximation, and in this chapter. You need to know what the new terms mean
make use of the sudden approximation for the inclusion of even if you only want to understand the literature.
Chapter 14

the core-hole. The quantum-mechanical calculation of low-

 Appendix 421

Appendix

Part I
People Q14.8 What is meant by site and symmetry projected DOS and
why is it important for ELNES?
Q14.9 How does multiplet theory differ from the independent
Hans Albrecht Bethe was born in Strasburg, Germany on July 2, one-electron approximation?
1906 and died in Ithaca, NY, on March 6, 2005. He was one Q14.10 What advantages does the multiple scattering approach
of the pioneers in understanding the atomic nucleus and is fre- offer over band structure approaches for the calculation
quently cited for one Bethe potential or another. His Ph.D. advi- of ELNES?

Chapter 14
sor was Arnold Sommerfeld (who suggested he work on electron Q14.11 Why are the approaches used to calculate ELNES not
diffraction), he did an assistantship with Erwin Madelung, and suitable for low-loss EELS?
his father-in-law was P.P. Ewald! Read more about his fascinat- Q14.12 How is the valence EELS connected to optical proper-
ing life even if you can’t understand all of his papers! ties?
Q14.13 Why is the RPA often inadequate to describe valence
Felix Bloch was born in Zurich on October 23, 1905 and died EELS?
there on September 10, 1983. He was awarded the Nobel Prize
for physics in 1952. His Ph.D. advisor was Werner Heisenberg
(December 5, 1901 – February 1, 1976), who was awarded the
same prize in 1932. Text-Specific Questions
DFT was developed in the 1960s by Pierre Hohenberg, Walter
Kohn, and Lu Sham. The DFT reformulation of the many-body T14.1 What feature of the Kohn–Sham (Eq. 14.2) allows it
Schrödinger equation into a soluble set of independent particle to be used to solve for the electronic structure of solids
equations can be ranked as one of the great successes of modern where the Schrödinger equation (Eq. 14.1) cannot?
physics. Walter Kohn (born in Vienna on March 9, 1923) shared T14.2 Using appropriate equations, explain why it is necessary
the Nobel Prize for Chemistry in 1998 for the development of for a DFT calculation to go through a SCF cycle.
DFT. T14.3 In what kinds of materials might the LDA or GGA be
expected to be inadequate?
Erwin Schrödinger was also born in Vienna, a little earlier, on T14.4 List and describe three different basis sets used in DFT
August 12, 1887 and died there on January 4, 1961. He was calculations.
awarded the Nobel Prize for Physics in 1933. T14.5 Distinguish between the one-electron approximation
and the independent electron approximation.
The WIEN2k code also originated in Vienna! It was originally de- T14.6 Figure 14.4 shows that d-DOS provides a good repre-
veloped by Peter Blaha and Karlheinz Schwarz from the Institute sentation for the features in the Cu L23 edge. Use the
of Materials Chemistry of the Vienna University of Technology. dipole approximation to explain why.
T14.7 Why is Eq. 14.6 not an entirely accurate representation
for the ELNES? How is this limitation usually over-
come?
Self-Assessment Questions T14.8 With reference to Fig. 14.5, describe and explain how
the effects of the core hole usually appear in the spec-
trum.
Q14.1 What distinguishes the electronic excitations responsi- T14.9 Figure 14.7 shows multiple scattering calculations for
ble for spectral details in the core-loss region and the the O K edge in rutile. Use the evolution of the spectral
valence region of the EELS spectrum? details to connect spectral features with particular in-
Q14.2 What is meant by ground-state electronic structure and teratomic bonds.
how is it relevant to EELS fine structure? T14.10 Figure 14.9 shows that only the peak at 40 eV is signifi-
Q14.3 What is the Born–Oppenheimer approximation? cantly changed when LFE are included. Explain why
Q14.4 What is the exchange correlation potential? this is so.
Q14.5 What are pseudopotentials and why are they used? T14.11 As a relaxing challenge for readers interested in history:
Q14.6 What must be done to go from the ground state elec- how are George Green (1793–1841), William Rowan
tronic structure to a prediction of the ELNES? Hamilton (August 3, 1805–September 2, 1865), and
Q14.7 What is the core hole and how is it usually accounted John Clarke Slater (December 22, 1900–July 25, 1976)
for in ELNES calculations? related?
 422 14 Calculating EELS

Garcia de Abajo FJ (2010) Optical excitations in electron mi-

References croscopy. Rev Modern Phys 82:209–275 (Some forefront
research on surface plasmons (Section 14.4))
Part I

de Groot F (2001) High-resolution X-ray emission and X-ray

absorption spectroscopy. Chem Rev 101:1779–1808
General References de Groot F (2005) Multiplet effects in X-ray spectroscopy. Coord
Chem Rev 249:31–63
Egerton RF (1996) Electron-Energy Loss Spectroscopy in the Hetaba W, Blaha P, Tran F, Schattschneider P (2012) Calculat-
Electron Microscope, 2nd edn. Plenum Press, New York ing energy loss spectra of NiO: Advantages of the modified
Fox M (2001) Optical Properties of Solids. Oxford University Becke-Johnson potential. Phys Rev B85:205108 (For more
Press, Oxford on the modified Becke-Johnson potential)
Kaxiras E (2003) Atomic and Electronic Structure of Solids. Hébert C (2007) Practical aspects of running the WIEN2k code
Cambridge University Press, Cambridge (This author coined for electron spectroscopy. Micron 38:17–28 (As the title says
Chapter 14

the phrase „the holy grail“ for the exchange correlation po- for WIEN2k)
tential) Jorissen K, Rehr JJ, Verbeeck J (2010) Multiple scattering
Kittel C (2005) Introduction to Solid State Physics, 8th edn. John calculations of relativistic energy loss spectra. Phys Rev
Wiley and Sons, Hoboken B81:155108 (Discussion of the state-of-the-art in FEFF cal-
Liboff R (2003) IntroductoryQuantum Mechanics, 4th edn. Addi- culations (Section 14.3.4).)
son-Wesley, Boston (Or any other introductory quantum text) Keast VJ (2013) An introduction to the calculation of valence
Martin RM (2004) Electronic Structure: Basic Theory and Meth- EELS: Quantum mechanical methods for bulk solids. Micron
ods. Cambridge University Press, Cambridge 44:93–100
Singh DJ, Nordstrom L (2005) Planewaves, Psuedopotentials Kümmel S, Kronik L (2008) Orbital-dependent density func-
and the LAPW method, 2nd edn. Springer, New York (Dis- tionals: Theory and applications. Reviews of Modern Physics
cussion of the available pseudopotentials is beyond the scope 80:3–60
of this chapter, but this is very good reference to consult. Mizoguchi T, Tanaka I (2000) Core-hole effects on theoretical
This book also provides a good introduction to these LAPW electron-energy-loss near-edge structure and near-edge x-ray
methods) absorption fine structure of MgO. Phys Rev B61(3):2180–
Wooten F (1972) Optical Properties of Solids. Academic Press, 2187 (One example of the application of MO calculations in
London and New York ELNES simulations)
Moreno MS, Jorissen K, Rehr JJ (2007) Practical aspects of
electron energy-loss spectroscopy (EELS) calculations with
FEFF8. Micron 38:1–11 (As the title says for FEFF)
Specific References Nelhiebel M, Louf PH, Schattschneider P, Blaha P, Schwarz
K, Jouffrey B (1999) Theory of orientation sensitive near-
Ankudinov AL, Ravel B, Rehr JJ, Conradson SD (1998) Re- edge fine-structure core-level spectroscopy. Phys Rev
al-space multiple scattering calculation and interpreta- B59(20):12807–12814 (Rigorous treatment of orientation
tion of x-ray absorption near-edge structure. Phys Rev and relativistic effects in ELNES)
B58(12):7565–7576 (For more detail on FEFF) Onida G, Reining L, Rubio A (2002) Electronic excitations:
Ching WY, Rulis P (2009) X-ray absorption near edge structure/ density-functional versus many-body Green’s-function ap-
electron energy loss near edge structure calculation using the proaches. Rev Modern Phys 74:601–659
supercell orthoganized linear combination of atomic orbitals Paxton AT (2005) Theory of the near-edge structure in electron
method. J Phys: Condens Matter 21:104202 (A useful review energy loss spectroscopy. J Electron Spectros 143:53–66
of OLCAO) (Muffin-tin orbitals but full-potential description)
Clark SJ, Segall MD, Pickard CJ, Hasnip PJ, Probert MJ, Refson Paxton AT, Craven AJ, Gregg JM, McComb DW (2003) J Mi-
K, Payne MC (2005) First principles methods using CASTEP. crosc 210:35–44 (A more rigorous explanation than the ‘half
Z Kristallogr 220:567–570 (A planewave pseudopotential in and half-out’ explanation of the Slater transition in Section
code where ELNES calculations are already built in) 14.3.2)
Egerton RF, Whelan MJ (1974) Electron energy loss spectra of Rehr JJ, Albers RC (2000) Theoretical approaches to x-ray ab-
diamond, graphite and amorphous carbon. J Electron Spec- sorption fine structure. Rev Modern Phys 72(3):621–654
trosc 3:232–236 Rehr JJ, Kas JJ (2009) Prange, MP, Sorini AP. Takimoto Y, Vila
Gao SP, Pickard CJ, Perlov A, Milman V (2009) Core-level F An initio theory and calculations of X-ray spectra. CR Phy-
spectroscopy calculation and the plane wave psuedopoten- sique 10:548–549.
tial method. J Phys 21:104203 (For more detail on CASTEP Rez P, MacLaren JM, Saldin DK (1998) Application of the layer
and the projector augmented wave (PAW) approach (Sec- Koringa-Kohn-Rostoker method to the calculation of near-
tion 14.3.1)) edge structure in x-ray absorptiona and electron-energy-loss
spectroscopy. Phys Rev B57(4):2621–2627 (As the title says:
 Appendix References 423

applying KKR methods to calculating ELNES/XANES (us- de Groot F (2005) Multiplet effects in X-ray spectroscopy. Co-
ing muffin-tin potentials)) ordination Chem Rev 249:31–63
Schattschneider P, Hébert C, Franco H, Jouffrey B (2005) Aniso- Hébert C (2007) Practical aspects of running the WIEN2k code

Part I
tropic relativistic cross sections for inelastic electron scatter- for electron spectroscopy. Micron 38:17–28
ing, and the magic angle. Phys Rev B72(4):045142 (As the Jorissen K, Rehr JJ, Verbeeck J (2010) Multiple scattering
title says: the relativistic correction) calculations of relativistic energy loss spectra. Phys Rev B
Schwarz K, Blaha P, Madsen GKH (2002) Electronic structure 81:155108
calculations of solids using the WIEN2k package for mate- Kümmel S, Kronik L (2008) Orbital-dependent density func-
rial sciences. Computer Phys Commun 147:71–76 (For more tionals: Theory and applications. Rev Modern Phys 80:3–60
details on WIEN2k) Mizoguchi T, Tanaka I (2000) Core-hole effects on theoretical
Vast N, Reining L, Olevano V, Schattschneider P, Jouffrey B electron-energy-loss near-edge structure and near-edge x-ray
(2002) Local field effects in the electron energy loss spectra of absorption fine structure of MgO. Phys Rev B 61(3):2180–
rutile TiO2. Phys Rev Lett 88(3):037601 (Significant effects 2187

Chapter 14
of the LFE (Section 14.4)) Moreno MS, Jorissen K, Rehr JJ (2007) Practical aspects of
Wang ZL (1996) Valence electron excitations and plasmon oscil- electron energy-loss spectroscopy (EELS) calculations with
lations in thin films, surfaces, interfaces and small particles. FEFF8. Micron 38:1–11
Micron 27:265–299 Nelhiebel M, Schattschneider P, Blaha P, Schwarz K, Jouffrey B
(1999) Theory of orientation sensitive near-edge fine-structure
core-level spectroscopy. Phys Rev B 59(20):12807–12814
Background References Paxton AT (2005) Theory of the near-edge structure in electron
energy loss spectroscopy. J. Electron Spectrosc Related Phe-
Ankudinov AL, Ravel B, Rehr JJ, Conradson SD (1998) Re- nom 143:53–66
al-space multiple scattering calculation and interpreta- Rehr JJ, Albers RC (2000) Theoretical approaches to x-ray ab-
tion of x-ray absorption near-edge structure. Phys Rev B sorption fine structure. Rev Modern Phys 72(3):621–654
58(12):7565–7576 Rehr JJ, Kas JJ, Prange MP, Sorini AP, Takimoto Y, Vila F (2009)
Ching WY, Rulis P (2009) X-ray absorption near edge structure/ Ab initio theory and calculations of X-ray spectra. CR Phy-
electron energy loss near edge structure calculation using the sique 10:548–549
supercell orthoganized linear combination of atomic orbitals Rez P, MacLaren JM, Saldin DK (1998) Application of the layer
method. J Phys: Condens Matter 21:104202 Koringa-Kohn-Rostoker method to the calculation of near-
Clark SJ, Segall MD, Pickard CJ, Hasnip PJ, Probert MJ, Refson edge structure in x-ray absorptiona and electron-energy-loss
K, Payne MC (2005) First principles methods using CASTEP. spectroscopy. Phys Rev B57(4):2621–2627
Z Kristallogr 220:567–570 Schattschneider P, Hebert C, Franco H, Jouffrey B (2005) Aniso-
Egerton RF, Whelan MJ (1974) Electron energy loss spectra of tropic relativistic cross sections for inelastic electron scatter-
diamond, graphite and amorphous carbon. J Electron Spec- ing, and the magic angle. Phys Rev B72:045142
trosc 3:232–236 Schwarz K, Blaha P, Madsen GKH (2002) Electronic structure
Gao SP, Pickard CJ, Perlov A, Milman V (2009) Core-level calculations of solids using the WIEN2k package for material
spectroscopy calculation and the plane wave psuedopotential sciences. Computer Phys Comm 147:71–76
method. J Phys: Condens Matter 21:104203 Singh DJ, Nordstrom L (2005) Planewaves, Pseudopotentials,
de Groot F (2001) High-resolution X-ray emission and X-ray and the LAPW Method, 2nd edn. Springer, New York
absorption spectroscopy. Chem Rev 101:1779–1808 Williams DB, Carter CB (2009) Transmission Electron Micros-
copy: A Textbook for Materials Science, 2nd edn. Springer,
New York

Licensed to ([email protected])
Diffraction & X-ray Excitation
15

Part I
Chapter Preview

When you tilt your crystal specimen close to a zone-axis entations. This approach allows us to gain insight into the
orientation, you can set up a situation where the incident ways that specific atoms substitute onto specific crystal sites
beam channels preferentially down the rows of atoms. (sublattices) in ordered compounds such as intermetallics,
When this happens and the Bloch waves are maximized on many ceramics, and minerals. The technique is known by
the atom column, anomalous X-ray excitation occurs. The the most famous acronym in the TEM field, ALCHEMI. We
characteristic X-rays are generated with intensities that you can also use ALCHEMI to form channeling patterns which
can quantify. These atomic concentrations reflect the com- can be correlated with the more traditional backscattered
positions of the specific rows of atoms, which often differ electron channeling patterns. We understand this whole
significantly from the average composition of the crystal, process sufficiently well that we can accurately calculate
which you would measure by XEDS in non-zone-axis ori- the experimental spectra and channeling patterns directly.

Chapter 15

© Springer International Publishing Switzerland 2016

C. B. Carter, D. B. Williams (Eds.), Transmission Electron Microscopy, DOI 10.1007/978-3-319-26651-0_15 425
 426 15 Diffraction & X-ray Excitation

15.1 Introduction
Part I

In W&C Chaps. 11–15 you saw how electron diffraction not only
results in diffracted beams of electrons, but also under strong ‘dy-
namical’ conditions changes the spatial distribution of electrons in
your crystal specimen. In Chaps. 4 and 34 you learned how inelas-
tic interactions between the fast, beam electrons and the atoms in
the specimen produce characteristic X-rays, which are invaluable
for elemental chemical analysis. In this chapter, we bring together
these two aspects: when we combine the redistribution of electrons
with X-ray excitation we get a powerful analytical technique.

In particular, in this chapter we explain how electron diffraction

and subsequent X-ray excitation in an ordered crystal can be ex-
ploited through the technique we usually call ALCHEMI. Inelas-
tic scattering followed by diffraction (i.e., the other way round)
results in Kikuchi lines and is described in W&C, Chap. 19. You
can find a brief description of ALCHEMI in W&C, Sect. 35.9.

15.2 ALCHEMI
Chapter 15

When you orient your crystalline specimen in a TEM such that

the electron diffraction is strong (‘dynamical’), the spatial distri-
bution of the beam electrons takes on the periodicity of the crys-
tal and varies across the projected unit cell (see W&C, Sect. 14.6
and Fig. 14.2). For example, when one set of planes only is dif-
fracting (‘systematic conditions’) we showed in W&C, Sect. 15.6 Fig. 15.1 The influence of orientation on electron distribution in a crystal. a In
that when the deviation parameter sg (W&C, Sect. 12.6) is nega- a disordered crystal the electrons are channeled along the atoms when sg is –ve.
tive, the beam electrons are channeled along the planes of atoms. b In an ordered crystal – for example NiAl at <001> – for sg – ve the electrons
When sg is positive, the electrons are channeled between the atom are channeled along the heavier atoms, here unshaded. In a real ALCHEMI ex-
planes. Because inelastic scattering is associated preferentially periment, systematic conditions would be used for the {100} planes, not a low
with the centers of the atoms, this is how the familiar dark (low index zone axis as shown here
intensity) bright-field bend contours arise (as shown in W&C,
Fig. 24.8). These contours are mainly due to diffuse incoherent ment which sublattice it has chosen to inhabit. How we do this
elastic scattering by the atoms. Equally, we find that character- will be explained below.
istic X-ray emission, which is also localized to near the atom
nuclei, is enhanced for sg – ve. In very general terms, if an ordered crystal consists of n sub-
lattices, we can measure the composition of each using these
In a chemically ordered crystal – e.g., a ceramic or an interme- dynamical diffraction effects.
tallic compound – all or some of the reflections are affected by
the state of order of the compound. For example, in an ordered Why ‘ALCHEMI’? ALCHEMI stands for Atom Location by
intermetallic alloy, a superlattice reflection is one which would Channeling Enhanced Microanalysis and is thus both a pun (on
not exist if the crystal were to disorder. The crystal planes giv- ‘alchemy’) and an acronym at once. It is due to Spence and
ing rise to superlattice rows (systematic rows containing super- Taftø (1983).
lattice reflections) will have different chemical compositions,
containing different proportions of the sublattices making up the
crystal. Figure 15.1 shows the {100} planes in a NiAl crystal.
The crystal planes are alternately pure Ni and pure Al. If you 15.3 Gedanken ALCHEMI
tilt a NiAl specimen about the {100} Bragg position, the beam
electrons are channeled along the Ni atoms (sg – ve) or the Al
atoms (sg + ve). (The origin is conventionally taken on the heavier Imagine that an element C is dissolved into a B2 (CsCl) inter-
atom, to make the extinction distances positive.) Thus for sg – ve metallic AB. (For example, chromium dissolved into NiAl.) The
Ni X-rays preponderate and for sg + ve, Al. If we dissolve a third specimen is oriented – ve of the {100} Bragg position and the A,
element into the NiAl, we can deduce from this type of experi- B, and C X-rays are counted for a given time period. You then
 15.3 Gedanken ALCHEMI 427

Table 15.1 X-ray counts in an ALCHEMI experiment Table 15.2 Table 15.1 normalized to 1,000 A X-ray counts. You can think of this
as the results from two experiments where the beam is turned off when the ‘A’
sg (100) A B C counts reach 1,000. The bold numbers are used below

Part I
Negative (specimen close to 2000 16 000 60 sg A B C
100 Bragg position)
Negative 1000 8000 30
∞ (kinematical non-diffracting 1000 2000 10
conditions) ∞ 1000 2000 10

Ratio -/∞ 1 4 3
tilt the specimen so it is then oriented well away from the {100}
Bragg position and you count the X-rays again. The results are
given in Table 15.1. Table 15.3 Table 15.1 slightly elaborated

Although the absolute numbers have some significance, let us sg (100) A B C

for the moment consider the A : B : C ratios only of the X-ray
Negative 2000 16 000 60
counts (Table 15.2).
∞ 1000 2000 10
What can we say about the crystal before we start the analysis? Ratio RA RB RC

zz B is clearly the heavier element.

zz The fact that the ratio A:B changes is consistent with the fact
that {100} is a superlattice row and is why the reflection ex- Given the way Table 15.2 has been normalized, we have only
ists in the first place. used the red numbers

Chapter 15
30 10  4
These two points are just setting the scene: they are not what we =p+ .1 − p/ (15.2)
really want to know. 10 10

zz C must be present on both sublattices; on both the A sublat- Generalizing, then (but staying with a B2 crystal: AB + C),
tice and the B sublattice. If C were all on A, the ratio A:C
A B C
could never change. By the same token, if C were all on B,
the C counts would have risen by four times, like the B counts. Ratio negative=1 1 RB RC
In fact, the C X-rays have gone up by an intermediate ratio: RC = p + RB .1 − p/ (15.3)
3. What proportion of C is on the A sublattice and what pro- p= Rc −RB
portion on B? 1−RB

Imagine a fraction p of C is on A and (1–p) on B. p is a partition [If the A dose ratio is left as RA, instead of being set to 1, i.e., if
coefficient. we use Table 15.1 above directly (Table 15.3).

We have normalized the numbers in the table to the same elec- Then
tron dose (not current) for the A atoms. Under these circum-
stances, for sg negative the B atoms see four times the dose which RC
− RB
they see in the kinematical experiment (Table 15.2). Of the 10 C
RA RA
P= RC
X-ray counts under non-diffracting conditions, 10p come from 1− RA
the C atoms on the A sublattice and 10 × (1-p) come from the C Rc − RB
atoms on the B sublattice. Under dynamical conditions the total =
RA − RB (15.4)
dose along the A planes is the same. Thus, of the 30 counts, 10p
still come from the A sublattice. The electron dose given to the
B planes has gone up by four times and so 10 × (1-p) × 4 counts This is like the lever rule in phase diagrams and is the formula
come from the C atoms in the B planes. Thus we can write: given by Spence and Taftø (1983).]

30 = 10p + 40.1 − p/ You must ensure that one of the measurements is under kine-
matical conditions (far from the Bragg position), otherwise you
1 will never know how much of the C X-ray signal corresponds
and p =  (15.1) to each sublattice.
3
This means that one third of the C atoms sit in the A sublattice So far we have taken here a very simple, well-defined situation.
and two thirds sit in the B sublattice. (Notice we have said noth- We knew that all the A was on one sublattice and that all the B
ing about the total concentration of element A.)
 428 15 Diffraction & X-ray Excitation

in a two-phase region of a ternary phase diagram and Hou and

Fraser (1997) called such a line an ordering tie line. Although
we can measure the slope of the tie line, we cannot measure its
Part I

extent, without another independent datum. This extra datum

point can be anything depending on the state of order (like the
superlattice extinction distance) or can simply be a calculation
of the extent of channeling.

It is probably better if we analyze all but the simplest situations

using the general analysis.

Notice that the results of the two different types of analysis

are different. One gives us a partition coefficient and the other,
which requires an extra measurement, gives two absolute sub-
lattice occupancies. You’ll find that many analyses exist which
are intermediate between the two outlined above, including that
named ‘statistical ALCHEMI’ by its inceptors (Rossouw et al.
1989). The intermediacy is achieved usually by fixing the ends
of the tie-line to the periphery of the triangle (or whatever) – for
example by insisting that C partitions between A and B, but that
Fig. 15.2 In a CsCl AB-C crystal, for example NiAl with C dissolved in it, the ends
A and B do not mix: there are, however, many variations on this
of the Ordering Tie Line (OTL) correspond to the chemical compositions of the A
and B sublattices. A kinematical measurement gives the average crystal chemi-
theme. The statistical aspect is equivalent to fitting a straight
cal composition, midway between them. Electron channeling via a superlattice
line through the experimental data plotted in such a way that
Chapter 15

systematic row causes the apparent chemical concentration to move towards data deriving from stronger channeling is given more weight.
one or other end of the OTL We explain this in more detail in Sect. 15.6. You’ll see that many
of these approaches are rather heavy on algebra and none of
them strictly adds anything to the simple Spence–Taftø or OTL
approaches outlined above. Unless you intend to devote a large
was on the other. We can relax these stringent conditions, but part of your life to this type of measurement, it might be wiser
another measurement is usually required, the algebra becomes to use one of the two simple methods (as appropriate) and to
complex and the beautiful simplicity of ALCHEMI is lost. We add any necessary refinements as your common sense dictates,
can retain the simplicity in these more complicated situations and accordingly.
by taking a different approach to the problem, first presented
by Matsumura et al. (1991). In this approach, we make no as-
sumptions whatsoever, but unambiguously require an additional,
independent measurement. 15.4 Two Examples
Let’s go back to Table 15.1 and imagine we know nothing what-
soever about where the A, B and C atoms are. The kinematical A worked example is worth a lot of abstract theorizing. Let us
measurement gives us our overall (standardless) composition therefore examine two analyses; one illustrating the dilute solu-
(using some proprietary XEDS software, as usual) or, alterna- tion/partition coefficient approach of Spence and Taftø and the
tively, if we know the overall composition (which we usually other illustrating the concentrated solution general approach of
do), the kinematical measurement is our standard. Either way, Matsumura.
we can plot it on an A, B, C ternary diagram as in Fig. 15.2.

The other, dynamical, measurement clearly favors the B atoms.

Let us call the sublattice the B atoms preponderantly occupy the 15.4.1 Dilute Solution/Partition Coefficient
‘B’ sublattice. Thus, if the compositions of the two sublattices are Analysis
included in the plot (these are what we are actually trying to mea-
sure), the apparent composition under dynamical sg conditions
moves from the overall composition towards the B sublattice end Most of the ALCHEMI work reported in the literature uses this
of the line (Fig. 15.2). partition-coefficient approach. If you read Jones (2002) you’ll
find lists of work published before that year, including much
If we move to sg + ve, the apparent composition will move in the involving intermetallic compounds. (We’ll discuss the reason
opposite direction along the line towards the A sublattice compo- for this in Sect. 15.9.) For simplicity, we will stay with the same
sition. The line connecting the chemical compositions of the two system used to illustrate Sect. 15.3: NiAl-X.
sublattices is like a tie line connecting two phase compositions
 15.4 Two Examples 429

From Table 2A of Jones (2002) there were eight reports of AL- Again, the absence of a kinematical measurement precludes
CHEMI applied to NiAl-X. In order of date of publication they are: quantification of p, but the authors infer from the data above
that Cr is on the Al sublattice (i.e., that p = 1 or 0, depending on

Part I
1. Darolia et al. (1989) definition). This time, <111> dislocations were introduced by
2. Munroe and Baker (1990) the alloying, but still with no noticeable effect on the ductility.
3. Munroe and Baker (1992)
4. Field et al. (1991) Anderson et al. (1995) examined four NiAl-Fe alloys:
5. Anderson et al. (1995) Ni50Al40Fe10, Ni40Al50Fe10, Ni47.3Al50.7Fe2 and
6. Bastow and Rossouw (1998) Ni49.75Al49.75Fe0.5. Taking the first alloy, for example, clearly,
7. Wilson and Howe (1999) if all the iron is not on the Al sublattice, either Al vacancies must
8. Anderson et al. (1999) be introduced, or some of the Ni must substitute for Al. In fact,
for NiAl, on the Ni-rich side of stoichiometry, Ni substitutes
Darolia et al. were interested in improving the ductility of NiAl for Al, but on the Al-rich side, Ni vacancies appear (Bradley
by alloying it with third elements. Calculations had suggested and Taylor 1937). Anderson et al. use ‘statistical ALCHEMI’
that adding V to NiAl decreased the anti-phase boundary (APB) (see Sects. 15.3 and 15.6). A straight line is fitted to the strength
energy to a rather greater extent if the V inhabited the Ni sublat- of the channeling as reflected in the ratios of the intensities of
tice rather than the Al sublattice. The phase diagram suggested the X-rays emitted to the kinematical values. The gradient of
to the authors that V inhabited the Al sublattice (see Sect. 15.8). the line gives the partition coefficient of the Fe and the analysis
They therefore performed ALCHEMI experiments on a NiAl-V corrects in some sense for delocalization (but see Sect. 15.5).
alloy of unspecified composition using an unspecified reflection. The results are:
In fact the alloy was NiAl + 3%V (see below). The reflection they
used must have been a superlattice reflection and probably was Ni Al Fe p (fraction of Fe on Ni sublattice)
{100} since the smallest g vector generally gives the strongest 50 40 10 .253 ± .015 5.1 ± 0.3 % Al sites

Chapter 15
channeling (as we show in Sect. 15.6). They report that ‘the contain Ni
intensities of the Al and V peaks were enhanced for positive sg, 40 50 10 .880 ± .018 4.7 ± 0.7 % Ni
while those of the Ni peaks were higher for negative sg’. They sites = vacancies
probably meant, in fact, that the ratios Al/Ni and V/Ni rose for 47.3 50.7 2 .941 ± .032 3.2 ± 0.2 % Ni
sites = vacancies
positive sg as compared with negative sg while Al/V was con-
stant. Common sense tells us that V must therefore inhabit the 49.75 49.75 0.5 .545 ± .062
Al sublattice. If you refer back to Sect. 15.3 you’ll see that we
ought to have the kinematical measurements. In fact, if p = 0 Bastow and Rossouw (1998) also applied ‘statistical ALCHEMI’
or 1 as it does here, it does not matter and the situation is quite to two NiAl-Cu alloys: Ni50Al48Cu2 and Ni48Al50Cu2, this
unequivocal. On this basis, then, the authors went on to make up time using X-ray emission or ‘incoherent’ channeling patterns
a series of Ni(Alx,V1–x) alloys, but, in fact, their hope/expectation (thus ICPs) (see Sect. 15.7). The Cu partition coefficient is de-
of encouraging 12 h111i : : : 12 h111i superlattice pairs and thus in- termined by matching the Cu ICP to a linear sum of the Ni and
creased ductility was not fulfilled. Al ICPs. The result is that for Ni50Al48Cu2, all the Cu (± 3 %)
is on the Al sublattice, but for Ni48Al50Cu2, 75 % (± 3 %) only
Munroe and Baker re-examined NiAl-V (in fact literally the is on the Ni sublattice. These measurements agreed well with
same alloy, since it was provided by Darolia). They specified independent NMR measurements.
its composition as Ni47Al3V and, using symmetry and kine-
matical orientation measurements and Eq.(15.2), reported the Wilson and Howe (1999) examined five NiAl alloys using ‘sta-
partition coefficient ‘p’ for V between the Al and Ni sublattices as tistical’ ALCHEMI.
0.96 ± 0.05 (Munroe and Baker 1990) and 0.945 ± 0.035 (Munroe
and Baker 1992). This can reasonably be taken as 1. Munroe and Ni50Al46Cu1Ti3
Baker report decreased accuracy with increasing beam voltage Ni50Al44Cu3Ti3
and with using an axial geometry (Sect. 15.5), both of which Ni50Al41Cu6Ti3
they ascribe to delocalization (Sect. 15.5). You should note that Ni49Al47Cu1Ti3
the composition (Ni47Al3V) of itself is quite persuasive in plac- Ni47Al47Cu3Ti3
ing V on the Al sublattice (Sect. 15.9).
Broadly speaking, they found that Ti occupied the Al sublattice
Field et al. (1991) carried out a study similar to that of Darolia and Cu occupied either sublattice, as necessary to prevent Ni/Al
et al., described above, but this time on Ni49Al1Cr. Their results point defects (substitutional or vacancy as appropriate).
are as follows:
Finally, in this summary of ALCHEMI work applied to NiAl
sg INi IAl ICr ICr/INi ICr/IAl alloys, Anderson et al. (1999) examined Ni50-xAl50Fex,
negative 160 639 51 692 2020 .0126 .0391 Ni50Al50-xFex and Ni50-x/2Al50-x/2Fex with x = .25, 2, 5 and
negative 102 603 72 466 2468 .0241 .0341 10 using statistical ALCHEMI and a delocalization correction.
 430 15 Diffraction & X-ray Excitation
Part I

Fig. 15.4 The three different types of systematic row in Ti2AlNb

dilute solution type (mainly the earlier papers) where it is not

necessary to know the overall composition, and analyses inter-
mediate between this and the full-blown concentrated solution
OTL approach. Before turning to the topics listed above (delo-
calization, etc.), we’ll present a second example, involving a
different alloy system, NbAlTi, where we make no assumptions
about the compositions of the sublattices. We chose this example
Chapter 15

because it involves three sublattices rather than two and therefore

helps signpost the extension of ALCHEMI to more complicated
systems of scientific and technological relevance.

15.4.2 Concentrated Solution/OTL Analysis

The unit cell of Ti2AlNb is orthorhombic. As is illustrated in

Fig. 15.3 An 001 projection of the Ti2AlNb unit cell. The green atoms are Ti (8g), Fig. 15.3, there are three different atom locations in the unit
the red are Nb (4c1), and the blue are Al (4c2)
cell, with Wyckoff designations g, c1 and c2. There are eight
g positions in the unit cell and four each for c1 and c2. In the
ideal ordering scheme determined by Banerjee et al. (1988),
Fe has a slight preference for the Ni sublattice but, in contrast titanium atoms occupy the 8 g positions, Al the 4c1 and Nb the
to the Cu and Ti discussed in the previous example, tends to 4c2. Sarosi et al. (2003) used the OTL approach to determine the
displace Ni and Al, forming point defects. true state of order in Ti-23Al-21Nb-2Mo-0.35Si as a function
of temperature.

Summary So Far In Ti2AlNb there are broadly three types of systematic row. The
This selection of ALCHEMI papers has introduced some new type I systematic row (e.g., 020, 002, 022, 221, 310, 131, 223)
concepts: might be described as a fundamental row: the structure factors
of all the reflections are unaffected by the individual chemical
zz delocalization compositions of the g, c1 and c2 sublattices. In type IIa (200,
zz axial channeling 202, 021, 023) the c1 and c2 sublattices project on top of each
zz effect of composition (e.g., A50B47C3 vs A47B50C3) and other, but separately from the g sublattice (see Fig. 15.4). In
nature of solute (bonding considerations) type IIb (110, 111, 112, 130, 311, 113, 312, 132, 023, 332) (see
zz ICPs Fig. 15.4 again) the 16 atoms in the unit cell project as (4g + 4c1)
zz statistical ALCHEMI and (4g + 4c2). Thus the nature of the type IIa systematic rows
zz other, independent methods depends on the chemical compositions of g and the average of c1
zz vacancies and c2, respectively, and that of the type IIb systematic rows de-
pends on c1 and c2, respectively, since the g sublattice projects in
We will elaborate on these (and others) in Sects. 15.5–15.9. The an even-handed, neutral, way. It is the systematic rows of types
analyses described are a mixture of the partition coefficient, IIa and IIb which are used for the ALCHEMI experiment. For
 15.5 Delocalization and Axial Channeling 431

Part I
Fig. 15.5 Apparent chemical compositions measured in Ti-23Al-21Nb-2Mo- Fig. 15.6 Ordering tie triangles in Ti-23Al-21Nb-2Mo-0.35Si measured by AL-
0.35Si using the systematic rows indicated CHEMI using the experimentally determined 130g and 200g OTLs The ALCHEMI
experiment, of itself, can only give the shape of the triangle. An iterated n-beam
Bloch wave type calculation establishes the true, absolute, triangle to be the

Chapter 15
outer, red, one
IIa, 200 was chosen and for IIb, 110 and 130. The experimen-
tal (diffraction, etc.) conditions were optimized as described in
Sect. 15.6. The apparent chemical compositions were normalized
to the known true composition via a kinematical measurement. beam, electron passes through the atom in question, its prob-
ability of ionizing a particular orbital maximizes at a distance
Figure 15.5 shows the ALCHEMI results using the 200, 110, which, in classical terms, is called the impact parameter ‘b’. We
and 130 systematic rows. As you might expect, the (type IIa) call this phenomenon ‘delocalization’. Simple expressions for b
200 results lie on one line and the (IIb) 110 and 130 results have been given by Bourdillon et al. (1981):
lie on a different, but the same, straight line. The 200 OTL
1:24hv
defines the g − c1 + c2 line and the 110 and 130 OTLs, the b=
(g + c1) / 2 − (g + c2) / 2 line (see above). Thus, we can deduce 2E (where v = the speed of a fast electron and ∆E =
the shape of the ordering tie triangle (Fig. 15.6). In this case ionization energy)
the independent ‘measurement’ consisted of an n-beam Bloch
wave calculation (see W&C, Chaps. 14 and 15). The specific and by Pennycook (1988), who distinguished between X-ray
measurements shown in Fig. 15.5 were used, according to the excitation:
geometry of Fig. 15.6, along with foil thickness measurements 1
and the Bloch wave calculation iterated until it reproduced the
    
h E 16 E 2
= ln ln
experimental results. Iteration is necessary because the diffrac- 2Et Et Et (where E is the electron
tion also depends on the chemical compositions of the sublat- energy and Et the threshold for ionization) and EELS:
tices. The absolute ordering tie triangle thus deduced was the
red outer one shown in Fig. 15.6. For the results at the other − 12
temperatures, and their implications, you should read the orig-
 
hv 4E
v= ln
inal paper. 2 E

These expressions were used by, for example, Anderson et al.

15.5 Delocalization and Axial (1999) (see Sect. 15.4.1) but were criticized by Oxley and Al-
len (2000). You’ll find it is generally agreed that delocalization
Channeling increases as the X-ray energy decreases (larger orbital) and as
the beam energy increases and is greater for axial (zonal, 2D)
It has been implicit thus far that the original event, which leads channeling as compared with systematic (1D) channeling. So
to the production of the characteristic X-rays, occurs at a point, it’s a good idea for you to minimize these two aspects in your
effectively the center of the atom. This is not true. As the fast, experiments and, as far as possible, use X-ray peaks of similar
 432 15 Diffraction & X-ray Excitation

energy. Otherwise, you need to carry out a full calculation in the Then it is straightforward to show that
style of Oxley and Allen (2000), including thermal vibrations, ˛A
which also spread the ionization event.
Part I

p=
˛A + kA (15.6)
kC

where kA,C are X-ray generation/proportionality factors (see

15.6 Optimizing ALCHEMI: W&C Sect. 35.2).
‘Statistical’ ALCHEMI You should note that Anderson’s (1999) expressions also include
a delocalization correction (Anderson et al. 1999).
You’ve seen that ALCHEMI clearly works better when the chan-
neling is stronger. Ideally we’d like to have perfect channeling, In the OTL context this is equivalent to fitting a straight line
where the beam travels entirely along the A or B atoms, depend- through your raw data. Those points more towards the ends of
ing on the crystal orientation. This, however, is never true and the OTL (stronger channeling) exert more leverage.
we simply have to make the best of a real situation.

As far as your experiment is concerned, Jiang et al. (1999) have

shown that for systematic diffraction your specimen should be 15.7 Incoherent Channeling Patterns
tilted to the symmetry orientation, its thickness chosen to be
between ξg / 4 and ξg / 3, the beam convergence kept below the
Bragg angle and your beam voltage altered until the dynamical Figure 15.7 shows incoherent channeling patterns (ICPs) from
interaction parameter: TiAl – 1%Ta. An ICP is the X-ray signal around a zone axis. In
this case, the data were analyzed according to Eq. 15.3 above
Chapter 15

to yield the Ta sublattice occupancies. The plus point of this

p g
ws = 2+
22g approach is that the process is computer controlled over a long
2
period and, therefore, a significant statistical weight is built up.
(Remember, ws = 2U g
g
and measures the dynamical interaction be- The minus point is that when you use a zone axis orientation you
tween the beams along a systematic row in a given situation; Ug exacerbate the delocalization problem.
is the g Fourier component of lattice potential – see W&C, 14.2),
ICPs can also be used in their entirety – for example, Bastow
In a real-life situation you should manipulate (i.e., translate, tilt) and Rossouw (1998) as described in Sect. 15.4.1. You can also
your specimen so as to maximize the change in the ratio of the calculate them (Fig. 15.7b) and thus apply them to the problem
A and B X-ray peaks (but not using axial channeling). You’ll of interstitial location (Sect. 15.8 below).
find that a typical optimum channeling efficiency (e.g., the max-
imum length of OTL which can be reproduced) is significantly
less than 0.5. The longer you count for, the better, of course (as
emphasized in W&C, Sect. 34.2). Depending on the vintage of 15.8 Vacancies and Interstitials
your TEM, this approach will probably be limited by contami-
nation (W&C, Sect. 8.6). Cooling your specimen both reduces
the contamination rate and diffuse scattering. If you have a large enough concentration of constitutional va-
cancies in your specimen (usually revealed by SRO bands),
So much for the conduct of your experiment. As for the analysis they might be sufficient to affect your ALCHEMI results. We
of your results, the situations where the channeling is greatest can regard such vacancies as a separate chemical species, but
are clearly the most significant. This is reflected in ‘statistical’ without an X-ray signature, and therefore you have to infer their
ALCHEMI (Rossouw et al. 1996; Anderson and Bentley 1994) presence from a combination of the overall composition and
where the results are plotted against the strength of channeling the sublattice occupancies. This is a clear outcome in the OTL
and a straight line (plane, etc.) is fitted to the data points via lin- approach.
ear regression. The results corresponding to stronger channeling
therefore carry more weight (as they should). In the case of interstitials, there is no reference host sublattice,
and so some theoretical simulation is necessary. For example,
Thus, returning to the partition of element C between two sublat- Rossouw and Miller (1999a,b) identified the site of interstitial
tices in AB (Sect. 15.3 above), if NA, NB and NC are the measured Cr atoms in a mullite by comparing experimental ICPs with
numbers of A, B, C X-rays, then you can determine the coeffi- a series of calculations corresponding to different interstitial
cients αA and αB by a least squares fit of locations.
NC = ˛A NA + ˛B NB (15.5)
 15.6 Optimizing ALCHEMI: ‘Statistical’ ALCHEMI 433

Fig. 15.7 a Experimental channeling

patterns from three different X-ray
lines (Al K, Ta L and Ti K) and from

Part I
backscattered electrons (BSE) as well
as a large-angle convergent beam
electron diffraction pattern (LACBED)
for the L10 <111> zone of TiAl con-
taining 1 % Ta. The Ta ICP is similar to
the Ti ICP and dissimilar to the Al ICP.
Therefore, Ta is on the Ti sublattice.
b Calculated version of the various
patterns in (a)

Chapter 15
 434 15 Diffraction & X-ray Excitation

15.9 Chemistry Intermetallic compounds, and metallic bonding, are more flex-
ible chemically. In AB–C the behavior of C will depend on the
AB, AC, and BC bond energies. If, for example, the AB bonds
Part I

The results of ALCHEMI experiments – the chemical compo- are strong compared with AC and BC, the C atoms will just fill
sitions adopted by the different sublattices – depend on the var- up the deficit sublattice, whichever it is. You should note that in
ious bond energies, most easily interpreted on a pairwise basis. the NiAl-X results presented above, and particularly in the later
In ceramics we find that the situation is often fairly black and work, the series Ni50–xAl50Xx, Ni50–x/2Al50–x/2Xx, and Ni50Al50–xXx
white. In Fe2O3, for example, you might expect Cr3+ to dissolve is frequently examined, the middle one of the series being the
into the Fe3+ sublattice, but not at all in the O2− sublattice. How most informative. When comparing two different compositions a
strictly Cr3+ would simply replace Fe3+, maintaining the exact sensible route is to derive the underlying bond energies. Simple
(Fe,Cr)2O3 stoichiometry, would depend on the tolerance of each Bragg–Williams calculations are usually quite accurate; in fact,
sublattice for vacancies and of the whole lattice for interstitials. we can’t find any obvious example in the literature of an experi-
Table 1 of Jones (2002) suggests that, largely, solute elements mental measurement warranting a calculation more sophisticated
choose to inhabit sublattices consisting of chemically similar than Bragg–Williams (e.g., CVM, cluster variation method).
elements. The bonding in the ceramics reported is mainly ionic, You’ll find that phase diagrams, because they also depend on the
with some covalency. three bond energies, will probably suggest the correct sublattice
occupancy, as will the periodic table.

Chapter Summary
Chapter 15

Simple ALCHEMI results are easy and convenient for you and direct-beam, or real space, techniques, where we direct
to obtain. However, if you want to increase their precision, a narrow beam of electrons along a particular row of at-
this involves a substantial increase in effort. There is also no oms in the crystal. Aberration correction (see Chap. 3) has
obviously received truth about the best way to improve ac- strengthened the direct-beam techniques.
curacy. Nor do we find any indication that such an improved
accuracy can be exploited, either via thermodynamics or ab EELS, as an alternative to XEDS in ALCHEMI, has not
initio calculations. More complex systems, however, such as been exploited much, although clearly, useful results can
might be useful for structural (probably high-temperature) be obtained (Taftø and Krivanek 1982). One reason is that
situations, or for magnetic or electronic applications, would the electrons carrying the energy loss signal can be scattered
be of current interest. away, and as a result do not contribute to the ALCHEMI
signal. This is also true, of course, for the real-space EELS
There is a competition between ALCHEMI, where we use experiment, where some spectacular successes have been
electron diffraction to alter the electron distribution across recorded (e.g., Klie et al. 2005).
the whole volume of crystal irradiated by the electron beam,

Licensed to ([email protected])
 Appendix 435

Appendix

Part I
People Ni73Al25Re2 s<0 s>0 kin R−/∞ R+/∞
Al K 50 000 50 000 50 000 1.000 1.000
Ni K 353 685 279 999 267 209 1.324 1.048
Since this is such a young field we decided not to highlight any Re M 4084 3977 4090 0.999 0.972
particular historical characters.
Ni/Al 7.07 5.60 5.34
Re/Al 0.082 0.079 0.082
Re/Ti 0.012 0.014 0.015
Questions on Analyzing Data
Q15.3 The following intensity data from an ALCHEMI ex-
periment on Ti47.5Al47.5Mn5 were reported by Holmestad
The questions for this chapter are all in the form of exercises. et al. (1995). A 001 systematic row was used. Where are
The references for each question are listed on the book’s web- the Mn atoms?
site.
Ti47.5Al47.5Mn5 s<0 s>0 kin R−/∞ R+/∞
Q15.1 Shindo et al. (1986) reported the following X-ray counts Al K 475.9 461.7 667.3 0.713 0.692
from an ALCHEMI experiment in Ti43Al55 Nb2 using Ti K 1261.4 960.4 1834.1 0.688 0.524

Chapter 15
the {110} systematic row. What can you deduce about Mn K 132.6 116.0 184.3 0.719 0.629
the Nb occupancy of the two sublattices (Al : A; T i: B)?
Ti/Al 2.65 2.08 2.75
Ti43Al55Nb2 s<0 s>0 kin R−/∞ R+/∞ Mn/Al 0.279 0.251 0.276
Al K 54 154 51 713 47 139 1.149 1.097 Mn/Ti 0.105 0.121 0.100
Ti K 112 471 55 584 61 924 1.816 0.898
Nb L 6702 3275 3800 1.764 0.862 Q15.4 Nakata et al. (1991) reported the following X-ray data
collected during an ALCHEMI experiment using a
Ti/Al 2.080 1.070 1.310 {100} systematic row in B2 Ti48.5Ni48.5Co3. What can
Nb/Al 0.124 0.063 0.081 you deduce?
Nb/Ti 0.060 0.059 0.061
Ti48.5Ni48.5 s<0 s>0 kin R−/∞ R+/∞
Co3
Q15.2 Miyazaki et al. (1994) reported the following AL-
Ti K 53 993 50 803 53 086 1.017 0.957
CHEMI data for 3 Ni3Al-Re alloys (110 systematic
Ni K 44 698 56 126 54 676 0.818 1.027
row). What can you conclude?
Co K 3303 4192 4036 0.818 1.039
Ni75Al23Re2 s<0 s>0 kin R−/∞ R+/∞
Al K 50 000 50 000 50 000 1.000 1.000 Ni/Ti 0.828 1.105 1.030
Ni K 346 577 302 348 338 567 1.024 0.893 Co/Ti 0.061 0.083 0.076
Re M 4874 4471 4668 1.044 0.958 Co/Ni 0.074 0.075 0.074

Ni/Al 6.93 6.05 6.77 Q15.5 Gu et al. (1997) report the following for a {110} row in
Re/Al 0.097 0.089 0.093 Ni73.9Al25.4Zr0.7. Comment on these results.
Re/Ti 0.014 0.015 0.014
s<0 s>0
Al/Ni 0.273 0.179
Ni74Al24Re2 s<0 s>0 kin R−/∞ R+/∞ Zr/Al 0.883 0.877
Al K 50 000 50 000 50 000 1.000 1.000
Zr/Ni 0.241 0.157
Ni K 386 035 317 837 347 658 1.110 0.914
Re M 1955 2071 1994 0.980 1.039

Ni/Al 7.72 6.36 6.95

Re/Al 0.039 0.041 0.040
Re/Ti 0.005 0.007 0.006
 436 15 Diffraction & X-ray Excitation

Q15.6 The following table shows some of the results reported

by Horita et al. (1995). Where does the Ta sit? References
Part I

Ni75Al21Ta4 s<0 s>0 kin R−/∞ R+/∞

Al K 15 590 14 060 15 201 1.026 0.925
Ni K 161 533 130 984 149 790 1.233 0.874
General References
Ta M 10 840 9742 10 706 1.013 0.910
Jones IP (2002) Determining the Locations of Chemical Species
Ni/Al 10.4 9.3 9.9
in Ordered Compounds: ALCHEMI. Advances in Imaging
Ta/Al 0.70 0.69 0.70
and Electron Physics 125:63–117
Ta/Ni 0.067 0.074 0.071
Spence JCH, Taftø J (1983) ALCHEMI: a New Technique for
Locating Atoms in Small Crystals. J Microsc 130:147–154
Q15.7 The same group as in Q15.6 (but a different paper by Taftø J, Krivanek OL (1982) Site-Specific Valence Determina-
Horita et al. (1997)) reported some axial channeling tion by Electron Energy-Loss Spectroscopy. Phys Rev Lett
ALCHEMI results on the same alloy. What is the an- 48:560–563
swer now?

Ni75Al21Ta4 Axial kin Rax/∞

Al K 2020 1746 1.157 Specific References
Ni K 15 282 8843 1.728
Ta L 814 631 1.290 Anderson IM, Bentley J (1994) ALCHEMI formulated for delo-
calization and anti-site defects Proc. ICEM13, Paris, France.
Ni/Al 7.57 5.06
vol. 1., p 609
Anderson IM, Duncan AJ, Bentley J (1995) Determination of
Chapter 15

Ta/Al 0.403 0.361

Ta/Ni
Site Occupancies in Aluminide intermetallics by ALCHEMI.
0.053 0.071
High Temperature Ordered Intermetallic Alloys, vol. 64.
MRS, Pittsburgh PA, pp 443–448
Q15.8 Chen et al. (1990) reported the following NiK/AlK Anderson IM, Duncan AJ, Bentley J (1999) Site-distributions
X-ray intensity ratios for Ni3Al. What do you deduce? of Fe Alloying Additions to B2-ordered NiAl. Intermetallics
s<0 kin 7:1017–1024
110 7.4 5.2 Banerjee D, Gogia AK, Nandi TK, Joshi VA (1988) A New Or-
111 6.1 dered Orthorhombic Phase in a Ti3AlNb Alloy. Acta Metall
36:871–882
Bastow TJ, Rossouw CJ (1998) Lattice Site Determination of
Q15.9 Leonard et al. (2000) reported the following X-ray in- Dilute Cu in β-phase NiAl by 63 Cu Nuclear Magnetic Res-
tensity ratios for the Nb50Ti25Al25 {100} B2 systematic onance and X-ray Emission Channelling Patterns. Philos Mag
row. What are the chemical compositions of the two Letters 78:461–467
sublattices? Bourdillon AJ, Self PG, Stobbs WM (1981) Crystallographic
Nb50Ti25Al25 s<0 s>0 kin Orientation Effects in Energy Dispersive X-ray Analysis. Phi-
Al K/Nb L 0.288 0.427 0.384 los Mag A 44:1335–1350
Ti K/Nb L 0.625 0.515 0.585 Bradley AJ, Taylor A (1937) An X-ray analysis of the nickel-al-
Al K/Ti K 0.460 0.829 0.656 uminium system. Proc Roy Soc A159:56–62
Darolia R, Lahrman DF, Field RD, Freeman AJ (1989) Alloy
Modeling and Experimental Correlation for Ductility En-
Q15.10 Analyzing the B2 alloy Ti47V30Cr14Al10, Li et al. (1998) hancement in NiAl High Temperature Ordered Intermetallic
made the following ALCHEMI measurements. What Alloys III. MRS, Pittsburgh PA, pp 113–118
can you say about the compositions of the two sublat- Field RD, Lahrman DF, Darolia R (1991) The Effect of Alloy-
tices? ing on Slip Systems in Oriented NiAl Single Crystals. Acta
Ti47V30Cr14Al10 s<0 s>0 kin Metall Mater 39:2961–2969
Cr/Ti 0.282 0.283 0.268 Hou D-H, Fraser HL (1997) The Ordering Scheme in Nb Alu-
Al/Ti 0.079 0.101 0.095 minides with the B2 Crystal Structure. Scripta Mater 36:617–
V/Ti 0.568 0.682 0.645 623
Al/V 0.140 0.148 0.147
Jiang N, Hou D-H, Jones IP, Fraser HL (1999) Optimizing the
Cr/V 0.496 0.414 0.416
ALCHEMI Technique. Philos Mag A 79:2525–2538
Jones IP (2002) Determining the Locations of Chemical Species
in Ordered Compounds: ALCHEMI. Advances in Imaging
and Electron Physics 125:63–117
 Appendix References 437

Klie RF, Arslan I, Browning ND (2005) Atomic Resolution Elec- Pennycook SJ (1988b) Impurity Lattice and Sublattice Location
tron Energy-Loss Spectroscopy. J. Elect Spect Rel Phenom by Electron Channeling. Scanning Microscopy 2:21–30
143:107–117 Rossouw CJ, Miller PR (1999a) Analysis of Incoherent Chan-

Part I
Matsumura S, Morimura T, Oki K (1991) An analytical elec- nelling Patterns Formed by X-ray Emission from Host Lat-
tron-diffraction technique for the determination of long-range tice Species and Interstitial Cr in Mullite. J Electron Microsc
order parameters in multicomponent ordered alloys. Mat 48:849–864
Trans JIM 32:905 Rossouw CJ, Miller PR (1999b) Location of Interstitial Cr in
Munroe PR, Baker I (1990) An ALCHEMI Investigation of Ter- Mullite by Incoherent Channeling Patterns from Character-
nary Site Occupancy in NiAl-based Alloys. Proc XIIth ICEM, istic X-ray Emission. Am Mineral 84(5–6):965–969
Seattle, U.S.A., pp 448–449 Rossouw CJ, Forwood CT, Gibson MA, Miller PR (1996) Sta-
Munroe PR, Baker I (1992) Effect of Accelerating Voltage on tistical ALCHEMI – General Formulation and Method With
Planar and Axial Channeling in Ordered Intermetallic Com- Application to Ti-Al Ternary Alloys. Philos Mag A 74:57–76
pounds. J Mater Res 7:2119–2125 Sarosi PM, Hriljac JA, Jones IP (2003) Atom Location by
Oxley MP, Allen LJ (2000) Atomic Scattering Factors for K- Channelling-enhanced Microanalysis and the Ordering of
shell and L-shell Ionization by Fast Electrons. Acta Cryst Ti2AlNb. Philos Mag 83:4031–4044
A56:470–490 Wilson AW, Howe JM (1999) Statistical ALCHEMI Study of
Pennycook SJ (1988a) Delocalization Corrections for Electron the Site Occupancies of Ti and Cu in NiAl. Scripta Mater
Channeling Analysis. Ultramicroscopy 26:239–248 41:327–331

Chapter 15
X-ray and EELS Imaging
16

Part I
Chapter Preview

Collecting X-ray or EELS data has traditionally involved images can then be retrospectively analyzed and re-analyzed
many decisions regarding how many point analyses should as many times as needed to fully understand the chemis-
we acquire and from where should we acquire them. Alter- try of the specimen. Spectral imaging in the TEM/STEM
natively, simple maps were acquired telling us how indi- comprises a powerful set of microanalytical tools for the
vidual spectral features (e.g., elements) are distributed in comprehensive chemical characterization of all classes of
a region of our specimen. Point analyses give us typically materials including biological specimens.
very good spectra to analyze but the question that remains
is, did we appropriately sample our specimen? Did we miss In this chapter we will define a spectral image, describe the
something? Alternatively we might ask, is what we thought common methods of acquisition as well as typical acquisi-
we were mapping really present in the specimen or is it an tion conditions for X-ray and electron energy-loss spectral
artifact? images. We illustrate the various modes of acquisition with
several examples. Additionally, we will discuss methods
Today we can take the best parts of point analyses and map- of analysis for what can be extremely large resulting data
ping and combine them into what is called ‘spectral imag- sets. For some spectral imaging background look at W&C,
ing’ where a complete spectrum is acquired from each of a Chaps. 33 and 35, pages 605, 619–20, and 650 for XEDS
series of points in a two-dimensional array. These spectral and Chap. 37, page 680 for EELS.

Chapter 16

© Springer International Publishing Switzerland 2016

C. B. Carter, D. B. Williams (Eds.), Transmission Electron Microscopy, DOI 10.1007/978-3-319-26651-0_16 439
 440 16 X-ray and EELS Imaging

16.1 What Are Spectral Images and

Why Should We Collect Them?
Part I

The analytical electron microscope has evolved into a powerful

tool for microanalysis. The primary advantage to analyzing thin
foils of materials is that the interaction volume for the generation
of X-rays is decreased substantially over that for bulk analyses
and, therefore, spatial resolution can be improved. The AEM
allows us to effectively span a range of practical microanalytical
resolutions from 100 nm down to less than 1 nm. It is important
to note that while nm-scale resolution may be important for some
analysis problems, lower resolution can be extremely useful for
a large range of practical AEM-type problems. In fact, focusing
only on the highest resolutions can mean important features in
our specimens may be overlooked. Modern AEMs allow us to
tune the probe and pixel sizes in STEM as well as the effective
magnification and therefore pixel size in EFTEM to easily span
two or more orders of magnitude in effective spatial resolution.
For general analyses this entire range may be required to answer
critical problems.

Spectral vs Spectrum Fig. 16.1 Schematic of the X-ray and EEL spectral image data cube acquired
in STEM mode. The spectral image consists of a full spectrum of 1,000 or more
You’ll see and hear the term ‘spectrum imaging’ as often energy/energy-loss channels in each pixel in a 2D array
as ‘spectral imaging’. The difference is purely grammat-
ical! 9
from each point in a 2D array and, therefore, post-acquisition or
retrospective processing can be performed to extract maps or
The traditional methods employed to perform microanalysis perform more sophisticated analyses from the spectral image
on thin foils include point analyses as well as mapping. With data set. Figure 16.1 is a schematic of the X-ray and EEL spectral
point analyses, the operator images the specimen, visually iden- image data cube as acquired in STEM mode. Simple X-ray map-
tifies locations for point-spectral-acquisition and then acquires ping is prone to artefacts, for example, pathological overlaps can
Chapter 16

a relatively high signal spectrum from each of a limited number lead to erroneous intensity in a given X-ray-derived elemental
of points. Features not showing strong contrast in the imaging map. Additionally, higher-Z elements produce a more intense
modes available can potentially be overlooked with this analy- Bremsstrahlung that can be mistaken for the presence of a low-Z
sis approach. Additionally, it is rarely clear how many analysis element in its respective map. Later in this chapter, effective
points should be chosen to adequately survey a microstructure. strategies for these types of analyses will be discussed.

As an alternative to point analyses, mapping has traditionally The problems with microanalysis in the STEM can best be il-
been used in the AEM. A map consists of an image for a range lustrated by an example from a microelectronics failure analysis
of energy channels with the intensity in the map corresponding (Kotula and Keenan 2006). Figure 16.2 is an annular dark-field
to the number of counts in that limited spectral range. Additional STEM image of two metal vias separated by dielectric material
processing may be applied to maps to subtract non-element-spe- plus selected point analyses and elemental maps. The transistor
cific spectral background. Mapping has the advantage that, in is more than a micron below the bottom of the image. In this
principle, elemental images of the specimen can be acquired, image we have contrast from different metals plus the dielectric
showing the spatial distribution of the chosen elements. Early material. The question that arises is how many points do we need
implementations of mapping on STEM systems required that to collect X-ray spectra from to fully characterize this material?
the elements to be mapped be known in advance so that if an How can we be sure? Alternatively we may ask which elements
element present in the specimen were not mapped another map we should acquire maps for.
would have to be acquired. This limitation was really a limitation
of the older computer systems collecting the data. Figure 16.2 illustrates the inherent difficulties in this process,
showing selected X-ray spectra and X-ray maps. This particular
In present-day systems, this has been overcome with the advent analysis will be addressed in detail later in the chapter.
of spectral imaging methods, where the entire spectrum is stored
 16.2 Some History 441

Part I
Fig. 16.2 STEM-ADF image of a cross-section of metal vias and interlayer di-
electrics (ILD) in a microelectronic device, with several analysis points chosen
and spectra from some of those points, as well as Ti and O X-ray maps

Fig. 16.3 Schematic of an EFTEM spectral image acquired with broad illumina-
A Note on Nomenclature
tion, energy plane by energy plane

Spectral images as defined previously in this chapter are

sometimes called hyper-spectral images to specifically area of the specimen to build up the final spectral image (XEDS),
denote a spectral image with more than 20 contiguous or energy plane by energy plane as in EFTEM spectral imaging
energy channels. The spectral images described in this with a region of the specimen broadly illuminated with electrons

Chapter 16
chapter typically have in excess of 1,000 or as few as (Fig. 16.3).
~ 100 energy channels. This is contrasted with the term
multi-spectral images or spectroscopic images which Regardless of how we acquire the data, each approach will result
both denote data sets with fewer than 20 energy-chan- in a data set of dimensions x by y pixels by n energy-channels.
nels or possibly a discontinuous energy-axis consisting In this chapter, we will discuss both acquisition and analysis
of only a series of pre-chosen elemental maps. 9 methods for spectral images acquired in the TEM/STEM and
retrospective analysis methods to get at the information in the
spectral images.
Traditionally however, analysts would use a combination of
point analyses and mapping to solve this problem based upon
significant prior knowledge, always with the possibility of over-
looking something. One solution to this problem is to combine 16.2 Some History
the power of point analyses with imaging into spectral imag-
ing. Conceptually, spectral imaging can really be thought of as
a merging of the best parts of point analysis with conventional The historical developments behind spectral imaging track with
mapping approaches without their respective limitations. A spec- the development of early computers. With greater computing
tral image consists of a series of spectra in a 2D array as shown power came the ability to consider the acquisition of full spectra
schematically in Fig. 16.1. The spectra are acquired to cover at an array of pixels. The historic development of X-ray mapping
an area of the specimen. The spectra may be acquired in serial and spectral imaging has been summarized by Lyman and Friel
fashion as a focused electron probe (e.g., in STEM) is scanned (2006). In 1956 Cosslett and Duncumb (1956) published the first
once across the specimen (XEDS and EELS), in quasi-parallel example of X-ray mapping in an electron-beam instrument. In
fashion with the beam being scanned multiple times across an this case, an electron beam was scanned as specific X-rays were
 442 16 X-ray and EELS Imaging

acquired with a low-resolution XEDS based on a gas-propor- 16.3 Acquisition and Analysis
tional counter.
of Spectral Images
Part I

The intervening time has seen the development of the XEDS

spectrometer (Si(Li)) and application to SEMs and STEMs. In The STEM and XEDS/EELS system parameters we choose for
a ground-breaking piece of work in 1979 Legge and Hammond acquisition of X-ray spectral images will largely depend upon
demonstrated the concept of X-ray spectral imaging in a proton the ultimate information we desire from the material being an-
microprobe with the full X-ray spectral data recorded as events alyzed, the probe size and current, the specimen thickness, and
– tagged with x and y positions of the scan as well as the photon pixel size. The primary advantage of performing microanalysis
energy (Legge and Hammond 1979). These data were streamed in the STEM versus the SEM is the improvement in spatial res-
to magnetic tape for storage. In order to reconstruct complete olution due to the thin specimens used. The tradeoff for X-ray
spectra from points/regions or maps, the entire tape had to be re- spectral images for example is fewer counts for comparable SEM
played looking specifically for events corresponding to a specific probe-currents and detector solid angles. The number of counts
x, y pair/range of pairs or energy range respectively. The entire desired in the spectral image will largely be determined by the
data processing step was limited by slow data storage methods type(s) of analysis to be performed after acquisition.
and limited computing power. Legge and Hammond did not use
the term spectral imaging and their work went largely unnoticed In this following section we will describe the interplay of mi-
by the electron-beam microanalysis community until 1995. The croscope and XEDS/EELS system parameters to perform both
proton microprobe community however, seems to have been high- and low-total-signal spectral images, as well as the choices
early adopters of spectral imaging. of pixel and probe sizes for various scales of information desired.

Early integration of computers with STEMs can be followed in a

special issue of Ultramicroscopy (1989), although this work was
in many ways less sophisticated than that of Legge and Ham- 16.3.1 Sampling and the Effect of Probe Versus
mond (and not cited therein as well). Gorlen et al. (1984) demon- Pixel Size (STEM-XEDS/EELS)
strated combined STEM-XEDS and EELS data acquisition with
on-the-fly data processing and only processed maps were stored. or Magnification (EFTEM)
In 1989 Jeanguillaume and Colliex (Jeanguillaume and Colliex
1989) described the concept of EELS spectral images and coined One of the advantages of spectral imaging over point analyses is
the term ‘spectrum imaging’ with Hunt and Williams (1991) our ability to select a regular array of analysis points to sample
showing the first EELS spectral images in 1991. In 1992, Weiss an area of our thin specimen and then have a computer drive the
et al. (1992) demonstrated the combined acquisition of multiple beam to each location automatically acquiring spectra at each
signals (XEDS and EELS) and full spectral series and spectral pixel. We must not forget, however, that our specimen has a
image acquisition and storage. The data were collected via a finite thickness and as we remember from Chap. 36 of W&C,
Chapter 16

serial, single-pass scan of the beam. the beam will broaden as it passes through the specimen. This
has the effect of lowering our spatial resolution.
In 1995, however, with the wide availability of relatively inex-
pensive computers, Mott and co-workers at Princeton Gamma In the following discussion though, we will consider effective
Tech, Inc. described the first commercial (and practical) X-ray resolution to be equivalent to the probe size. If however, you
spectral imaging system (Mott et al. 1995). This system collected know your specimen thickness, you can use a better value for
data via the method described by Legge and Hammond (1979) effective spatial resolution based upon the methods described
by acquiring only X-ray events as the beam was rastered multiple in Chap. 36 (W&C). If we know our pixel size we can then use
times across the specimen, so-called position-tagged spectrom- probe and pixel sizes to determine how we are sampling our
etry or event streaming (a quasi-parallel acquisition mode). It is microstructure. If our probe is smaller than our pixel size then
interesting to note that in commercial XEDS systems on most we are under-sampling our specimen. If our probe is larger than
SEMs and STEMs this is the predominant X-ray spectral image our pixel size then we are over-sampling our specimen. This is
data acquisition method. shown schematically in Fig. 16.4.

Commercial variations were also developed in the mid-to-late To determine the pixel size in STEM, simply define a scan region
1990s by EMiSPEC (ESvision) and Gatan (Digiscan) based of a fixed size (e.g., 200 nm × 200 nm) and then set the number of
upon serial (single-pass scan) spectral image acquisition of pixels in the scan (e.g., 100 × 100 pixels). The pixel size is then
both X-ray as well as EELS spectral image data. This is the only calculated with the following simple formula:
method implemented so far for STEM-EELS spectral image
acquisition and will have some ramifications in the presence of pixel size = spectral image x − dimension
specimen contamination, beam damage or uncorrected specimen (16.1)
=number of pixels in the x − direction
drift.
 16.3 Acquisition and Analysis of Spectral Images 443

Table 16.1 Ratio of probe to pixel size with associated scan areas assuming a
100 by 100 pixel scan

Part I
Scanned Pixel size 50 nm 20 nm 5 nm 1 nm
region [nm] [nm] probe probe probe probe

5000 50 1 0.4 0.1 0.02

2000 20 2.5 1 0.25 0.05

1000 10 5 2 0.5 0.1

Fig. 16.4 Schematic representation of probe versus pixel size, and the concept 500 5 10 4 1 0.2
of over- and under-sampling 100 1 50 20 5 1

50 0.5 100 40 10 2
For the example above we take 200 nm and divide by 100 pixels
to get 2 nm/pixel. Of course, we assume here that our pixels are
square, which should be periodically checked with a suitable
standard. Table 16.1 shows the ratio of the probe to pixel size for of spectral imaging over prolonged point analysis, namely that
a number of different probe and pixels-size combinations. Num- if we dwelled for long times on this specimen, the plastic would
bers less than 1 indicate undersampling, while numbers greater tear. By limiting our dose at any one point we can still effectively
than 1 indicate oversampling. The gray boxes in Table 16.1 in- sample the chemistry of the specimen without destroying it.
dicate the largest recommended oversampling. Undersampling,
however, can be a useful way to sample the chemistry of large The following section will examine the issues with signal as re-
areas of material without having to re-tune the probe-forming lated to the overall size and acquisition times of spectral images.
system of the microscope. It also can be useful for materials that
rapidly beam damage by physically separating analysis points so
that the damage is localized and does not continually accumulate.
16.3.2 Signal: Count Rate, Dwell Time, Spectral
Be aware of the fact that when driving the microscope’s scan Image Size, and Acquisition Time
coils with an external scan generator, for example with an exter-
nal XEDS system controlling the microscope, that the external
system must also be checked for the squareness of the pixels. The X-ray signal we measure at our detector for a given solid an-
Keep in mind that the scan region itself however need not be gle (typically ~ 0.1 sr in most older STEMs and surpassing 0.7 sr
square. We might, for example, want to scan a rectangular region in newer systems) depends upon the probe current, specimen
or even user-defined polygons or other shaped areas. makeup (atomic number and density), and specimen thickness.

Chapter 16
In a commercially-available field-emission gun STEM with an
There are several strategies with respect to sampling and spectral 0.13 sr Si(Li) detector and a 100 nm-thick specimen we might
imaging. In the first example, we can collect a spectral image typically get 1 kcps to 5 kcps even for moderately small (~ 2 nm)
which under-samples our specimen. In this case, we can image probes. It is useful to set the XEDS pulse-processor’s time con-
our specimen with a relatively small probe to form a good STEM stant to a small value to keep the detector dead-time low (below
image and then collect a spectral image covering a large area 10 %). For a typical Si (Li) XEDS system this is 15 μs or lower
resulting in a large pixel size. This is an effective strategy for at the expected specimen thicknesses (~ 100 nm)/count rates
covering large areas to survey our specimen without having to (< 10 kcps). For silicon-drift detectors (SDD) the shaping time
realign the probe. In the second case, we can just decrease the is 1,000 nsec or less and count-rates up to 100 kcps to 250 kcps
scan area we collect our spectral image from, leaving the probe are possible but probably indicate a specimen that is too thick
size the same, to over-sample a smaller region of our specimen. for an 0.1 sr system but still probably OK for a 0.7 sr or higher
This way we can be sure to not miss any fine microchemical system employing probes with currents of 2 nA to 5 nA. If thicker
features in that region of our specimen. specimens are used, consider using an even smaller pulse-pro-
cessor time constant for a Si(Li) detector of less than 5 μsec as
Figure 16.5 shows two images of biological thin sections, after the count rates could easily approach 20 kcps.
spectral image acquisition, in which we have intentionally under
sampled the specimen. The spectral image dwell time in both Figure 16.6 shows two spectra extracted from the spectral image
cases was 100 msec per point, which resulted in a small con- of the microelectronics circuit shown in Fig. 16.2. This spectral
tamination spot at each pixel measured. It should be noted that image was acquired from a rectangular region 800 nm by 200 nm
another way to change the probe size in a STEM, other than by and 256 pixels by 64 pixels for a pixel size of ~ 3 nm with a Si(Li)
changing the spot size is to defocus the probe by a small amount. XEDS with a nominal solid angle of 0.13 sr (Kotula and Keenan
The example shown in Fig. 16.5 also illustrates another attribute 2006). The dwell time was 100 ms for a total acquisition time of
 444 16 X-ray and EELS Imaging
Part I
Chapter 16

Fig. 16.6 Raw spectra extracted from the spectral image in Fig. 16.2. a From the
Ti metallization layer, containing 592 total X-ray counts. b From the thinner Si-O
ILD containing 191 total X-ray counts

the Si-O ILD (see the tip below). Table 16.2 illustrates the spec-
tral image acquisition times for different size spectral images for
a total per-pixel dwell time of 100 msec. If our output count rate
is 1–6 kcps then our spectra will have 100 to 600 total counts,
which are sufficient for subsequent conventional or advanced
Fig. 16.5 Annular dark-field STEM images showing contamination spots on a processing as will be described below.
microtomed biological thin section after spectral image acquisition. Two differ-
ent spot sizes and spacings, both undersampling the microstructure were uti- It is clear from Table 16.2 that the 128 by 128 pixel spectral
lized, (a) 20 nm probe (defocused) with a 35-nm spacing. b A 10 nm probe with image size is a reasonable trade-off between number of pixels
a 40 nm spacing and time to collect the data at about 30 min. Smaller data sets can
readily be acquired but the resultant maps or processed images
are clearly pixilated. Collecting images larger than 256 × 256 pix-
~ 30 min. The two spectra chosen for Fig. 16.6 illustrate the wide els quickly results in long acquisition times with spectral images
range of count rates possible from the same nominally 100 nm- taking about 7.5 h, which is prohibitive unless acquiring data
thick FIB specimen, almost 6 kcps from the Ti and 2 kcps from overnight. It should be noted that after years of commercial
 16.3 Acquisition and Analysis of Spectral Images 445

Table 16.2 Acquisition times for spectral images of different sizes assuming to worry about. EELS spectrometers are sophisticated electron
100 msec or 10 msec total dwell time per pixel. Times do not include system optical devices in their own right, arguably as complicated as
overhead or drift correction, which typically adds several to tens of minutes,
most electron microscopes! Firstly, as stated above, we do not

Part I
depending upon the acquisition time and the specific system
want to saturate the signal on the CCD. Secondly, we need to be
Pixel dimen- Total Acquisition Acquisition aware of the probe convergence angle and detector collection
sions in SI number of time [min] time [min] angle. We can look to W&C, Chaps. 37–40 to re-familiarize
spectra 100 msec/pixel 10 msec/pixel ourselves with EELS and with the optimum collection angles
total dwell total dwell for different parts of the EEL spectrum. We can control the
collection angle in STEM mode by adjusting the size of the
64 × 64 4096 7 0.7
entrance aperture to the EEL spectrometer as well as the dif-
128 × 128 16,384 27 2.7 fraction camera length. Remember, in STEM mode we have
256 × 256 65,536 109 (1.8 h) 11 a convergent-beam diffraction pattern on the viewing screen,
which directly shows us the convergence angle. The conver-
512 × 512 262,144 437 (7.3 h) 44 gence angle is set by a beam-defining aperture (and possibly
1024 × 1024 1,048,576 1748 (29 h) 175 (2.9 h) other condenser optics) and as such changing the spot size will
not change the convergence angle. Consult with your micro-
scope’s manufacturer for the numbers needed to calculate the
detector collection angle for various entrance apertures and
TEM/STEM manufacturers stuck at X-ray detectors with nomi- camera lengths. Importantly for our discussion of EELS-SI
nal solid angles of 0.1 sr, 0.7 sr or larger SDDs are now available set-up is that adjusting spot size is the best way to adjust the
from several manufacturers. Increased solid angle allows us to signal intensity. We want the most signal possible in our data
collect more of the limited number of X-rays escaping from the without saturation. The low-loss is much more intense than core
sample and with the advent of artefact-free silicon drift detectors loss regions and as such the settings can be quite different for
(Strüder et al. 1998 and Niculae et al. 2006) image sizes above the two respectively unless we have a fast shuttering system
512 by 512 pixels have now become practical in an AEM just as which allows very short dwell times. If we adjust the camera
they are already practical in SEM (Kotula and Michael 2006). length or entrance aperture to adjust the signal intensity we will
affect the acceptance angle.
In systems such as newer SDD-based ones, the data are for the
most part collected in the so-called event streaming mode where Tip
the beam is continuously rastered over the sample and only the
X-ray events detected are recorded tagged with the position the
beam was at the time (Mott et al. 1995). In this mode, instanta- To determine the number of X-ray counts per spectrum,
neous dwell times are typically 10 μsec to 50 μsec and as such put the microscope in STEM mode with the desired probe
the counts per pixel (spectrum) for such short dwell times is less size, use the live-view mode of the XEDS system at the

Chapter 16
than one on average. Even if the sample is producing 100,000 desired total XEDS dwell time and position the beam
counts per second, a 10 μsec dwell time would result in a spec- on the different regions of the specimen on each of the
trum with perhaps one count! For this reason spectral images higher and lower atomic number materials (or regions of
in event-streaming mode are acquired by summing subsequent different thickness). The output count-rate can be read
frames, where we keep scanning the same area recording only off of the XEDS-system interface. Just multiply the out-
the detected X-ray photons. In such a way longer total dwell put count rate (in counts per second) by the total desired
times can be achieved. The total per-pixel dwell time is then dwell time (in seconds) to get the counts per spectrum in
just the product of the number of frames and the instantaneous the spectral image. Alternatively, to get 100 counts per
per-pixel dwell time. spectrum in the spectral image, simply divide 100 by the
average count rate from the sample to arrive at the de-
You should note that for EELS spectral images in systems sired total per-pixel dwell time. 9
which rely on a CCD detector only single-pass spectral imag-
ing is done since the CCD has such a slow read out and event
streaming would be complicated and inefficient in this case. One area we tend to be more concerned with in EELS versus
The most important thing to remember is choosing appropriate XEDS is specimen thickness. Familiarize yourself with using
dwell times for EELS spectral images is not to saturate the de- EELS to measure specimen thickness (W&C, Sect. 39.5) so you
tector. Additionally, set-up can be helped by moving the beam can make sure you’re analyzing the thinnest part of your sample.
around on the sample at a modest dwell time while looking at This can be done concurrently as you check to make sure you’re
the live EELS spectrum. If collecting the low-loss data, dwell not saturating the detector. As a rule of thumb we want to have
times will tend to be much shorter than, for example, collecting our sample no thicker than 0.5 λ (inelastic mean-free path) for
a spectral region around the O-K peak, for example. It is almost the materials we’re analyzing. The main problem with very thick
always true to say the EELS set up is more time consuming samples is that we lose signal to inelastic scattering and also
than that of XEDS. This is due to a larger number of variables make the spectral background higher.

Licensed to ([email protected])
 446 16 X-ray and EELS Imaging

The following section deals with drift correction, which is essen-

tial for long acquisitions and high-resolution analyses.
Part I

16.3.3 Drift Correction and Beam Damage

X-ray spectral imaging systems are available from virtually all of

the XEDS system manufacturers although they differ in the way
the data are acquired. The two methods implemented are serial or
single-pass scanning and pseudo-parallel or multiple pass scan-
ning (event streaming). In the first method, the beam is scanned
once across the specimen with a fixed dwell time at each pixel
but can be periodically stopped to perform a drift correction. Fig. 16.7 Annular dark-field STEM image of a FIB cross-section of a (Al,Ga)As
One of the disadvantages of this approach over multiple-pass test structure. The two red squares show the initial and final positions of the
spectral image scan. The specimen has drifted over a micron during the scan
scanning is that each location on the specimen receives the entire
although the path may not have been a straight line
electron dose all at once. This may result in beam damage up to
and including hole drilling for some specimens. In contrast, the
multiple-pass scanning method allows the beam to be scanned
across the specimen, typically dwelling for as little at 10 μsec We can summarize this section quite simply: the essential steps
at each pixel per frame for SDD systems. The spectral image is of spectral imaging comprise knowing our approximate probe
then built up frame by frame until the desired total dwell time is size and the number of counts this probe generates from speci-
reached. This has the advantage over single-pass scanning that men of interest.
the total electron dose on the specimen is distributed over the
total acquisition time for the data. Additionally, drift correction If our detector is saturating (dead times above 50 %) then we
can be performed either by simultaneously monitoring the STEM should either reduce our probe current (smaller spot size, con-
image or by stopping the beam and scanning another region. denser aperture, or gun settings) or our detector pulse-processor
It should be noted that for EELS spectral images in systems, shaping time. This then will determine the relevant total per pixel
which rely on a CCD detector, only single-pass spectral imaging dwell time such that we collect on average 100–300 counts per
is done since the CCD has relatively slow read out so that event spectrum as a rule of thumb. We can decide upon the number of
streaming would be inefficient. spectra or pixels in our spectral image such that the total acqui-
sition time is tractable. One single spectral image is rarely suffi-
Figures 16.7 and 16.8 show an extreme example of the effect of cient to cover all of the important length scales so collecting one
Chapter 16

specimen drift on spectral image acquisition. To produce this SI at low-resolution undersampling the specimen but covering
level of drift (over 1 μm in 30 min), the specimen holder was a large area followed by successive SIs at higher resolution but
removed from the microscope and left sitting in the room for smaller areas is a strategy to characterize a specimen, which will
30 min. It was then loaded in the microscope and spectral im- be illustrated below for analysis of a precipitate in a meteorite
age acquisition was started immediately. This was done solely specimen.
for the purposes of illustrating the effect of drift on spectral
image acquisition and does not represent good general practice.
In most cases, however, utilizing drift correction can allow data
acquisition, for moderate resolutions, to begin immediately after 16.3.4 Conventional Data Analysis Methods
specimen insertion.

Figure 16.7 is an image of a layered test specimen showing the Now that we have a spectral image acquired from our speci-
cumulative drift of the specimen over a 30-min period during men, the next question we have to ask is what do we do with
the spectral image acquisition. Figure 16.8 shows X-ray maps it? How do we analyze the spectral image to determine what
extracted from the spectral images acquired without and with is going on in our specimen? The beauty of spectral imaging is
active drift correction. It is clear that drift has occurred as the in- that we can analyze the data retrospectively. This section will
terfaces, in the case of no drift correction, are no longer straight cover some strategies for spectral image analysis based upon
but rather curved. The STEM control and spectral image acqui- conventional and advanced statistical methods. It is helpful
sition software used in this case made a single serial scan over before proceeding to review the Chapters in W&C: 34/35 for
the 1500 nm by 1500 nm region as the specimen was drifting. X rays and 38/39 for energy-loss electrons. Much of what is
If we were using the more common quasi-parallel (short dwell- covered in those chapters will be of use as we analyze the
time multiple pass) XEDS system, the maps would be blurred data and then interpret the results of our analyses all while
rather than distorted. avoiding pitfalls.
 16.3 Acquisition and Analysis of Spectral Images 447

Fig. 16.8 X-ray maps extracted from

spectral images of the (Al,Ga)As test
structure shown in Fig. 16.7 both

Part I
without and with drift correction.

Al Map Uncorrected Al Map Drift Corrected

0 20 40 60 0 20 40 60

Chapter 16
Ga Map Uncorrected Ga Map Drift Corrected

50 100 150 200 50 100 150 200

As Map Uncorrected As Map Drift Corrected

0 50 100 150 200 0 50 100 150 200

 448 16 X-ray and EELS Imaging

additional analysis requiring several passes through the data to

verify the accuracy of the analysis.
Part I

Spectral image data is at its heart no different from collecting

data from a series of points in our specimen and collecting maps
directly from our specimen, except that we can do both retro-
spectively with one data set. In a conventional sense, we have
many tools available to us to interrogate the spectral image data.
Legge and Hammond (1979) described the basics of conven-
tional spectral image analysis in their work.

Figure 16.9b illustrates schematically how we can extract a

sum spectrum from a region of interest on our specimen and
Fig. 16.9b shows how we can sum a series of energy channels
to form an elemental map. Keep in mind however that all of the
pitfalls of point analyses and mapping still exist with spectral
image analysis. For example, we might think we have Ti in our
specimen, and that the Ti-map describes its distribution in the
specimen only to find out that our specimen has barium in it as
well. This well-known pathological overlap of X-ray lines is only
one of many such possible problems with mapping. We might
also think a map shows a real distribution of an element only to
find out that it is Bremsstrahlung from a higher Z element in the
specimen.
Fig. 16.9 Schematic of extracting (a) a spectrum summed from a group of
We can follow a simple strategy to convince ourselves that we
pixels comprising a region of interest in the spectral image and (b) X-ray map
are correct in our interpretation of an analysis of a spectral image
where a contiguous series of energy-slices or channel images are summed to-
by using the data itself as a check. The basic procedure described
gether
here can also be found in the literature (e.g., Kotula and Keenan
2006).

In beginning an analysis of a material we always ask ourselves Some of the tools described here are available within the XEDS
what we know about the material to start and what we need manufacturers’ built-in data analysis software (e.g., mapping,
microanalysis to tell us about our materials. If our question is extracting spectra from regions, and even multi­variate statis-
Chapter 16

for example, where the Li is distributed in my specimen then tical analysis), while others are available in off-line public do-
we can immediately say that XEDS is not useful for answering main (Bright 2006) and commercial (partial listing) (Matlab
this question. We should always be open to learning new and 2015; ENVI 2013; PLS Toolbox 2016; Mathematica 2015)
perhaps unexpected information about our samples and that is data processing software packages. The author of this chap-
precisely the strength of the spectral imaging approach. In the ter and several of his colleagues have developed multivariate
previous sections of this chapter we have learned how to set up statistical analysis software called Automated eXpert Spectral
the microscope to collect a spectral image with sufficient counts Image Analysis (AXSIA) (Keenan and Kotula 2003, 2004b).
to perform both conventional and advanced statistical analysis, For the purposes of this discussion, we have used Matlab with
both qualitatively as well as quantitatively. The following will the Image Processing Toolbox, Statistics Toolbox, and AXSIA,
deal with some specific spectral image data analysis approaches. although the other packages could work as well for certain
There are many methods that will not be covered in this section calculations.
that can, under the right circumstances, yield valid and useful
representations of the spectral image data while the reader is en- The first step in spectral imaging and analysis is collecting a
couraged to search the literature cited here and references therein spectral image or series of spectral images. Once a spectral
more widely. image is acquired, there are several conventional tools that can
be used to interrogate the data. Figure 16.9 illustrates two such
The spectral image itself consists of a collection of a large num- tools: extraction of a single spectrum or sum-spectrum from
ber of individual spectra and our goal is to somehow represent a group of pixels and extraction of maps. Most commercial
this in a more compact form. The raw spectral image is cumber- XEDS systems will at a minimum allow these processes to
some to display in its entirety and typically contains much noisy be performed. The public domain and commercial stand-alone
redundant data. Since no analysis is independent of interpre- data analysis packages require that you export your data from
tation, the process can be iterative with new information from the XEDS manufacturer’s data format into a more generic
 16.3 Acquisition and Analysis of Spectral Images 449

format supported by the stand-alone data analysis packages.

Remember that the spectral image data you acquire on your
XEDS or EELS system is your data and the equipment man-

Part I
ufacturer should always give you access to the data. Several
of the manufacturers either support the export of files into a
generic binary format or give you enough information about
their file format to allow you to be able convert it yourself into
a generic file format.

Programs such as Matlab (as well at the others listed above)

have a broad array of input/output commands to facilitate this
process. These can typically be implemented as scripts without
the need to take programming classes. When in doubt about a
spectral image file format, ask the technical support people at the
XEDS company to help you. Most XEDS spectral image data
can readily be stored as one-byte or two-byte unsigned integers
as long as no single data element (x, y, energy) exceeds 256
or 65,536, respectively. EELS spectral-image data (especially
low loss) routinely exceed even the two-byte limit and requires
higher-precision formats such as four-byte integers or real num-
bers. If storing background-subtracted or deconvolved data, real
numbers might be required.

In addition to looking at spectra from pixels or regions show-

ing distinct contrast in annular dark-field or other images and
maps of expected elements, it is also helpful to look at the
mean spectrum and the log-mean spectrum. As an example, in
Matlab, if the spectral image data is stored as a matrix called D
with dimensions of n-channels by m-pixel, the mean spectrum
is simply calculated (using Matlab syntax here and elsewhere)
Fig. 16.10 a Mean spectrum of the spectral image region shown in Fig. 16.2.
as b Log-mean spectrum to show weaker spectral features. c MaxPixel spectrum
(Bright and Newbury 2004)
mean spectrum = mean .D; 2/ (16.2)

Chapter 16
Here we operate on the second dimension of the spectral im- Figure 16.11 shows a selection of background-subtracted maps
age, the pixels and thus calculate the mean spectrum. This can from the microelectronics example of Fig. 16.2. The sum image
then be plotted with a linear y-axis or log y-axis quite readily as is calculated with Matlab as
shown in Fig. 16.10a and b, respectively, for the microelectronics
example of Fig. 16.2 and as published elsewhere (Kotula and sum Image = sum .D/ (16.4)
Keenan 2006). From these spectra, we can clearly see that we
have Si, O, Ti, Ar, Ga, Pt, Cu, and possibly Al. Alternatively, we The result here must be reshaped into an image of the appropriate
can calculate the so called max-pixel spectrum as described by number of pixels and can then be displayed as in Fig. 16.11a. The
Bright and Newbury (2004) and as shown in Fig. 16.10c, which sum or equivalently mean images can be useful for specimens
interestingly shows Al but not several of the other signals from with porosity, and also finding high-Z particles in a low-Z ma-
the mean spectrum. terial. Background subtracted maps can be simply calculated by
using methods described in Chap. 35, taking care to avoid possible
The max-pixel spectrum can quite readily be calculated in Mat- peaks in the background regions of interest. Figure 16.11b–g show
lab as a selection of maps from the microelectronics example. One po-
tential complication of this method for quantitative analysis is that
max − pixel spectrum = max .D; 2/ (16.3) there are quite often very few counts in the background regions of
interest, as the individual spectra in the spectral image are noisy.
In addition to extracting real spectra or derived spectra (e.g., This can be overcome by the use of multivariate statistical analysis
mean, sum, etc.) from the spectral image we can similarly work methods, as will be described later. Using a Matlab, an X-ray map
in the image domain by extracting maps, with or without back- for element A can be extracted from the spectral image as follows
ground subtraction and perform similar operations to produce
derived images such as mean and sum images. X−ray MapA = sum .D.StartEnergy W EndEnergy; W// (16.5)
 450 16 X-ray and EELS Imaging
Part I
Chapter 16

Fig. 16.11 a Sum image. b Oxygen map, 0.48–0.58 keV. c Aluminum map, 1.44–1.54 keV. d Silicon map, 1.69–1.79 keV. e Platinum map, 2.00–2.20 keV. f Titanium
map, 4.40–4.65 keV. g Gallium map, 9.19–9.29 keV

The second colon indicates that the sum is taken over all the X−ray MapA−Bckd = sum .D.StartEnergy W EndEnergy; W// − sum.D
pixels in the image and the StartEnergy and EndEnergy indicate
X−ray Map =.BackgroundStartEnergy W// − sum.D W//(16.6)
W BackroundEndEnergy;
sum .D.StartEnergy W EndEnergy;
the start and end channels of the map, respectively. Again, ifA−Bckd X−ray MapA−Bckd = sum .D.StartEnergy W EndEnergy; W// − sum.D
we wish to display this as an image we must re-shape it from
.BackgroundStartEnergy W BackroundEndEnergy;
.BackgroundStartEnergy W BackroundEndEnergy;
W// W//
m-pixels to rows by columns. Similarly, if we wish to subtract
the background, which can be useful for removing the effect of Alternatively, we could estimate the background using the other
Bremsstrahlung from light-element maps for example, we mod- methods described in W&C, Chap. 35 and apply them here as
ify the above equation as: well.
 16.4 Multivariate Statistical Analysis Methods 451

The above methods really describe only the first pass at a con-
ventional analysis of a spectral image. We need to examine both
extracted spectra as well as maps to convince ourselves that we

Part I
understand the specimen. Figure 16.12 shows the results of a few
image-processing-based additional calculations to the spectral
image from Fig. 16.2.

Firstly, the outstanding question from the analysis of this spectral

image was “What was the contaminant at the interface between
the two dielectric layers?” (Kotula and Keenan 2006). To help
answer this, we first extracted the spectrum from the most intense
‘Al’ pixel. Figure 16.12a shows the most intense ‘Al’ raw spec-
trum as determined from the Al map. To find this pixel, we start
with the background-subtracted Al map. We then find the most
intense pixel in this map by using the ‘max’ command in Matlab
with the index of the pixel’s position as an additional output.
We can then go back to the raw spectral image and extract the
spectrum of that index. If instead of just a single pixel we wish
to sum all of the pixels that are above a threshold intensity in a
map (in other words, the spectra in the spectral image), we again
start with the background-subtracted map.

Our goal is to transform the map into a binary image, one that
consists of ones where the intensity map is above the threshold
and zeros elsewhere. Choosing this threshold can take several tri-
als, keeping in mind that the resultant binary image can be filtered
to remove unwanted pixels. One easy way to help in choosing
the threshold is to look at the map’s histogram. Given that the
map is typically noisy given the suggested acquisition conditions,
this binary image may need to be noise-filtered, for example by
smoothing. Other noise filters would also work to reduce the effect
of noise, which would include spectra away from the features of
interest. In this case, if we make the Al map into a binary image Fig. 16.12 a Spectrum extracted from the most intense Al pixel in the data set as
and then smooth it, we substantially reduce the off-interface pix- determined from the Al map. b Binary image, produced by thresholding/smooth-

Chapter 16
els. Figure 16.12b shows the result of these operations on the Al ing the Al-map, of the interface between the two ILD materials. c The composite
map. This is a binary mask, which can be multiplied with the spec- spectrum extracted from the spectral image of the interfacial region shown in b
tral image to give us just the spectra from the interface. We can
then sum the interfacial spectra to produce the composite spectrum (1998), Hannequin (1989), Titchmarsh et al. (1995), Titchmarsh
from the interface, as shown in Fig. 16.12c. In this case, a distinct and Dumbill (1996), Titchmarsh (1997) and (1999), Anderson
Al peak appears indicating that our original assessment of Al at the (1998), Kotula and Keenan (2000), (2003a) and (2003b), Kotula
boundary between the dielectric layers was correct. The following et al. (2003), Mayer et al. (2003), Oliver et al. (2003), Boyce
section details the application of advanced multivariate statistical et al. (2004), Goods et al. (2004), Loehmann and Kotula (2004),
analysis methods to analyze the same spectral image. and Wilson et al. (2004).

MSA

16.4 Multivariate Statistical Analysis

MSA is a method of analyzing spectra – many spectra.
Methods We separate it here because it is a little complex, we give
several examples of the technique; it’s a hot topic and
As should be clear from the previous discussion of conventional it’s extremely powerful if you understand what you are
methods for spectral image analysis, you have to take great care doing. 9
to confirm that maps come from the ‘real’ elements! With the
relatively recent advent of commercial spectral imaging hard-
ware, researchers began to look for better methods to analyze Perhaps the most promising method for analysis of spectral-im-
the increasing amounts of data (see the papers by, e.g., Bonnet age data is multivariate statistical analysis (MSA). (Perhaps we
 452 16 X-ray and EELS Imaging

discuss the basic steps in MSA, which consist of scaling the data
for Poisson statistics, determining the number of components to
keep in the data model, several methods for factoring the scaled
Part I

data, and inverse scaling of the data.

As the first step in MSA of X-ray and EEL spectral images, it

is critical that we account for the nature of the noise in the im-
ages. The strength of MSA methods is that they typically have
no bias, that is, we need not assume the presence or absence of
spectral features. The problem is that if our data contain noise,
as data does in the type of counting experiments we are doing,
then MSA methods will do an excellent job of fitting noise at
the expense of weaker (but more important) chemical signals.
Say, for example, we have a spectrum in our spectral image
with several peaks, one peak with on-average 100 counts and
one with an average of 4 counts. The magnitude of the noise
in the 100-count peak will be ± 10 counts (i.e., √100). This
noise is of greater magnitude than the signal in our weak spec-
tral feature and MSA methods will find it more profitable, in
a least squares sense, to preferentially fit noise in this case. To
overcome this problem due to the Poisson nature of the noise
in the data we can use the computationally simple rank-one
approximation described by Keenan and Kotula (2004a). This
can be simply written as

Dweighted = .Mean Image/−1=2 D.Mean Spectrum/−1=2 (16.7)

where Dweighted is the Poisson-weighted data, D is the raw spectral

image, Mean Image is a matrix with the mean image of the spec-
tral image along its diagonal and Mean Spectrum is a matrix
Fig. 16.13 Two schematic representations of the dimensional reduction real- with the mean spectrum along its diagonal. The correct weight-
ized by MSA techniques. We assume that the factors we calculate linearly super- ing of the spectral image will ensure that our components rep-
impose to give us a model for our raw spectral image resent variations in the data due to chemistry rather than noise,
and that we can more easily determine the number of chemically
Chapter 16

relevant components p to retain in our model of the data. The

are also biased but we’re right.) Multivariate refers to the fact remaining n-p components will describe noise in the data and
that we have many variables, i.e., energy channels that make can be excluded from further calculations (i.e., the dimensional
up the data. Typical X-ray and EEL spectral images have 1,000 reduction discussed above).
energy channels but typically tens of thousands or more spec-
tra. The primary challenge that MSA techniques overcome is Counting
a dimensional reduction. If our spectral image has 1000 chan-
nels, then the data consist of 1000 dimensions or variables. It
is very unlikely, however, that these 1000 variables will vary The basic steps of MSA of spectral images consisting of
independently of one another! count data are

We can, therefore, reduce the dimensionality of the solution by 1. Scale the data for Poisson statistics.
correlating channels in each pixel throughout the image. We can 2. Determine the number of components, p, to retain in
think of this for the case of X-ray spectral image analysis in the model.
terms of multiple channels representing the same peaks in the 3. Factor the scaled data matrix (D).
spectra as well as multiple spatial regions having the same peaks. 4. Inversely scale the data for Poisson statistics. 9
Figure 16.13 shows the result of MSA schematically, where we
start with our raw data, m-pixels by n-channels unfolded, and
factor it into p chemically relevant factors or components. The original rank of a spectral image will be the smaller of
the number of pixels (m) or channels (n), which in practice is
It should be clear from Fig. 16.13 that instead of having to ex- typically (although not always) the number of channels. Thus,
amine all of the channel images or all of the spectra in the image, the number of noise components is typically n-p as described
we can examine just p component image/spectrum pairs. We will above.
 16.4 Multivariate Statistical Analysis Methods 453

Fig. 16.14 Comparison of Eigena-

nalysis with (blue) and without
(green) Poisson weighting. The first

Part I
non-component for the weighted
data is indicated by the red circle

Figure 16.14 shows the results of weighting on the Eigenanalysis Eigenvalues representing Poisson noise will lie on a straight
of the data. A good way to determine p is by calculating the Ei- line on this plot while those that deviate positively represent
genvalues of the data weighted for Poisson statistics as described non-noise information which is typically chemical in nature. So
above. We then sort the Eigenvalues in descending order and far, we have weighted the data and determined the number of

Chapter 16
display them on a semi-log plot, commonly referred to as a scree components to retain in our model. We now need to perform the
plot. A straightforward way to calculate the eigenvalues is to use factor analysis.
Matlab (with the Statistics Toolbox) and the function ‘princomp’
will output the Eigenvalues as well as the principal components. There are many different factor analysis methods and a good
The Eigenvalue plot describes the linearly independent sources background to the different possibilities would be the book by
of information in the data in an easy to interpret way where the Malinowski (2002) and the book chapter by Keenan (2007). As
amount of variance described by each component is given by the previous references give extensive details of the math under-
the Eigenvalue. lying several powerful ways to analyze spectral images, we will
not go into the math in very great detail here. Rather, we will
Figure 16.14 shows that if we do not weight for Poisson statistics illustrate the application of several methods on the same data
(green curve) then we see an Eigenvalue plot with one prominent set from the microelectronics example seen previously in this
and several minor break points. In this case, there is no clear chapter. If we look at the equation we are trying to solve
break point between chemical information and noise informa-
tion. This is due to the fact that noise and chemical information Dweighted = CST (16.8)
Eigenvalues are mixed, and in many cases noise components
are more important from a variance standpoint. In contrast, if where Dweighted is the raw data scaled for Poisson statistics of size
we weight the data for noise (blue curve), we clearly see five m pixels by n channels, C is a matrix of concentrations or com-
significant components describe this data. That is, in a linear ponent images spatially unfolded as m pixels by p factors, S is a
model of the data, there are five independent components (com- matrix of spectral shapes p factors by n channels, and T denotes
ponent image and spectrum pairs) that describe the non-noise the matrix transpose. This is shown schematically in Fig. 16.13.
and chemically relevant information in the data. Simple examination of this equation shows that there are two
unknowns and, therefore, an infinite number of solutions. This is
known as the rotational ambiguity in the MSA literature.
 454 16 X-ray and EELS Imaging

The key then in solving this factor analysis problem is to firstly Given the constraints of PCA described above the possible
reduce the dimensionality of the problem via the Eigenanalysis sources of variance in the data, the solution is rarely directly in-
described above and secondly to factor or transform the data into terpretable. It is better utilized as a model of the raw data, which
Part I

a useful and/or meaningful form. has been de-noised. This can provide a significant computational
benefit, supposing that PCA has been calculated correctly, by
The dimensionality (and equivalently the rank or number of making a compact representation of the data. Anywhere we need
variables) of a spectral image is typically the number of spectral to know the raw data we can use our reduced rank (p-factor)
channels, n, which is 1,000 or more. Via MSA, we reduce the model of the data instead. This includes additional MSA calcu-
dimensionality of the data in order to describe the chemistry lations as well as quantification, which will be discussed below.
with a smaller number of components or factors, which consist
of linear combinations of the original variables (channels). We To illustrate the effect of weighting on PCA we again look at the
will assume that the components linearly recombine to model our microelectronics example for which our weighted Eigenanalysis
original data set. We then factor the data subject to these reduced showed five non-noise components, as seen in Fig. 16.14. If we
dimensions given various possible constraints. do not scale the data for noise prior to PCA and we look at the
first five principal components in this case; the result is shown
in Fig. 16.15. The solution is displayed as principal component
image and spectral-shape pairs. We see that the first component,
16.4.1 Principal Components Analysis (PCA) a, is like the mean (i.e., the mean is the best one-component
model of a data set) and the second component, b, is Ti positive
in image and spectral intensity, and Si is negative in image and
PCA (Jollife 2002) is probably the most utilized of the MSA spectral intensity.
methods but the reader is cautioned that often no consideration is
taken of the noise structure of the underlying data or an inappro- Again, it is important that you realize that the principal compo-
priate data pre-treatment is performed. An example is described nent images can be both positive and negative, as can the corre-
below in which PCA is performed both without and with taking sponding spectral shapes, which makes them difficult to directly
into account the noise structure of the data to illustrate the for- interpret.
mer. PCA, as it has traditionally been applied, has been used in
attempts to gain insight into both X-ray and EEL-spectral images Further complicating such an attempted interpretation is the fact
(see, for example, Anderson 1998). The important properties of that every peak can be in every component. The last three prin-
PCA are that it serially maximizes the variance described by each cipal components, c–e, all appear to be noise related and are,
principal component and the components are orthogonal to each in fact, illustrative of PCA doing what it was designed to do,
other. This is significant in that a one-component PCA model of which is fit variance in the data with as few factors as possible.
the data will be the best one-component model of the raw data in If we were to look at more factors we would certainly see weaker
a least-squares sense. The same holds true for a two-component but chemically relevant components mixed with noise factors.
Chapter 16

PCA model and so on. However, if we calculate the first 100 principal components in
this case, we cannot see the Al signal from manual inspection of
The way PCA is typically performed on spectroscopic imaging the spectral shapes and corresponding images.
data considers the spectral channels the variables and the pixels
the observations or objects although thinking in opposite terms Figure 16.16 shows the first five principal components from
can provide important insight into some data analysis problems the weighted data. While this solution has captured the rele-
as described by Keenan (2009). In the former case, PCA will vant chemistry of this sample in as few components as possible,
produce spectral shapes (or loadings as they’re typically referred it is still not straightforward to interpret. Importantly though,
to in the literature) that are mutually orthonormal (orthogonal Fig. 16.16e shows a clear Al signal (negative in both the image
vectors of unit length) and component images (so-called scores and spectral shape).
in the literature) that are mutually orthogonal.
Comment
One outcome of the above constraints that the components se-
rially maximize variance and have the properties stated above
(spectral shapes are orthonormal and images are orthogonal) is Often in the texts and the literature of PCA, the first step
that the principal component images and spectra will contain in the analysis is to mean-center the data. This is equiv-
both negative and positive values and will, in general, be ab- alent to an offset or adding a constant to the data, which
stract. Given this, it is important to then make note of some then reduces the rank of the data by one. For spectral im-
of the possible sources of variance in the data: changes in ele- age data this is not generally useful as the mean must then
mental makeup; Poisson noise, breakdown of the linearity as- later be recombined with the solution for reconstructing
sumption (for example, due to preferential absorption of low de-noised spectra for example and also for quantification.
energy X-rays in some part of the specimen), and artefacts (e.g.,  9
spectrometer instability).
 16.4 Multivariate Statistical Analysis Methods 455

Part I
Chapter 16
Fig. 16.15 First five principal components from an analysis of data not weight-
ed for Poisson statistics. Note the energy scales in a–b are different from c–e

Fig. 16.16 First five principal components from an analysis of data weighted
16.4.2 Factor Rotations for Poisson statistics. Although the factors are abstract, there is still chemical
information that can readily be visualized from this analysis

Again the reader interested in more mathematical detail is di-

rected to the book chapter and paper by Keenan (2007, 2009) nent images or alternatively the spectral shapes to make them
and references therein. simple or high contrast. We then perform the inverse of this ro-
tation in the complementary domain in order to maintain the
One strategy to improve the interpretability of PCA is via factor product of the two matrices, which is the model of the raw data
rotations where we perform an orthogonal rotation (using the itself. Finally the inverse of the Poisson weighting is applied to
Varimax procedure, for example) of either the principal compo- the factors to bring them back to counts.

Licensed to ([email protected])
 456 16 X-ray and EELS Imaging

cedure applied to make the spectral domain simple in structure,

we could have started with PCA calculated with orthonormal
images and orthogonal spectral shapes and then rotated the im-
Part I

ages to a simple structure (Keenan 2009).

By ‘simple’, in the case of the calculation which produced

Fig. 16.17, we mean high contrast or alternatively consisting of
mostly zeros in the spectral channels in most of the components
with ‘contrast’ in only a small group of channels in a single com-
ponent. While the orthogonal rotation applied to the orthonormal
principal component spectral shapes retains that property in the
rotated spectral shapes, the inverse of that rotation applied to the
principal component images has the effect of relaxing orthogo-
nality in the rotated images.

More importantly, this solution is directly interpretable with each

factor essentially capturing the spatial distribution of a single
element. It does, however, still contain some negative image and
spectral intensities in the components. The product of the rotated
spectral shapes and images in this case is as good a fit to the raw
data as the unrotated principal components; thus we have not
lost the ability to accurately retain counts for later quantification
should we desire to do so.

16.4.3 Multivariate Curve Resolution (MCR)

Thus far, our primary assumption in the MSA methods described

has been that our data have a noise structure (Poisson) which
we must account for. We can also apply additional constraints
to MSA such as non-negativity of the component images and
spectral shapes via an approach known as multivariate curve
Chapter 16

resolution (MCR, Tauler and de Juan (2013)). One way to im-

plement MCR (see, for example, Keenan (2007) and the Matlab
pseudo-code (Keenan 2009)) is via an alternating least squares
(ALS) approach. Starting with Eq. 16.8, Dweighted = CST, we first
take an initial guess of S and then calculate C subject to no neg-
ative image intensities!

Comment

If your goal is to perform quantitative analysis of a

spectrum image, you must first perform a comprehen-
Fig. 16.17 First five factors from principal components rotated to a simple spec- sive qualitative analysis either by conventional or MSA
tral structure from an analysis of data weighted for Poisson statistics. The factors methods. 9
are no longer abstract and for the most part consist of single elements

We then use that C to calculate S subject to no negative spectral

intensities. The process proceeds iteratively until a convergence
We can illustrate the benefit of this approach on the microelec- criterion is reached. A good initial guess for S is the result shown
tronics example by first calculating the principal components as in Fig. 16.17, which has been truncated so as to be non-negative.
in Fig. 16.16 (Poisson-scaled data, orthonormal spectral shapes
and orthogonal images … conventional PCA). It is important to Figure 16.18 shows the results of the MCR-ALS on the same
note, however, that while Fig. 16.17 shows the result of this pro- data set and it’s apparent that neither the component images nor
 16.4 Multivariate Statistical Analysis Methods 457

Fig. 16.18 Results from the MCR

calculation on the spectral image.
Spectral intensities are normalized

Part I
to sum to one count so that the com-
ponent images' intensities describe
counts in a respective component.
a Si–O component consisting mostly
of Si. b Si–O component enhanced
in O between the metallization lines.
c Ti component. d Pt-Ga component
resulting from re-sputter in the FIB.
e Al component due to interfacial
contaminant

Chapter 16
spectral shapes have negative intensities. The first two factors (a PCA or rotated PCA solution. One should, therefore, be careful
and b) both contain Si and O in different ratios and reflect the using the output of an MCR calculation to extract counts for
breakdown of the linearity assumption (implicit in the MSA) due subsequent quantification.
to preferential absorption of the soft O–K X-rays in the thicker
part of the specimen beneath the Ti. Both of these factors des­
cribe the dielectric, SiO2 but two factors are needed to describe
this due to the change in O-signal in the different thicknesses 16.4.4 Quantification
of the sample. The last three factors, c–e, describe the Ti, Pt
(re-sputter from the FIB) and Al at the boundary between two
dielectric layers, respectively. So far we have covered conventional and MSA methods for qual-
itative analysis of spectral images. This is a critical first step
The MCR solution in general will be more straightforward to and one best done with an open mind. One the advantages of
interpret at the expense of not fitting the raw data as well at the the MSA methods described above is that they use a minimum
 458 16 X-ray and EELS Imaging

number of assumptions (Poisson noise and perhaps non-nega- 16.5 Example of X-ray and Electron
tivity) and, furthermore, allow the data to self-model itself from
several possible viewpoints (e.g., spectrally or spatially simple Energy-Loss Spectral Image
Part I

factors). Once we are convinced we understand our data quali- Acquisition and Analysis
tatively we can then add assumptions to go from counts to com-
position. The counts from the qualitatively determined elements
plus measured Cliff-Lorimer (Cliff and Lorimer 1975) k-factors This section is also part of our analysis but we are treating it
(Goldstein et al. 2007; Parish et al. 2008; Kotula et al. 2012) or separately because ‘spectral imaging’ is the way of the future.
alternatively ζ-factors described by Watanabe and co-workers You will use this approach to analysis in your research and it is
could be used (Watanabe and Williams 2003; Watanabe et al. best understood by studying examples.
2006) to go from counts to composition.

We refer to Chap. 17 for methods to measure k-factors or ζ-fac-

tors. In order to extract counts for given elements identified in 16.5.1 Fe-Ni Spectral Image Acquisition
our qualitative analysis we can in the simplest case, where there and Quantification
are no close peak overlaps, simply use Eq. 16.6 or something
similar to extract background-subtracted counts for a given el-
emental line. Goldstein and co-workers (Goldstein et al. 2007) provide an
example of MSA analysis of a spectral image of followed by
An MSA alternative to this for such simple cases would be to quantification of a Fe-Ni particle from a chondritic meteorite. A
use MCR on a sub-set of channels from a spectral image which TEM specimen was extracted from metal particle A2 in the CH
contain only the peaks of interest (Goldstein et al. 2007), which chondrite Northwest Africa (NWA) 739. The FIB sample was
will be shown below. A more general approach (Watanabe et al. analyzed in a FEI Company Tecnai F30-ST operated at 300 kV
2006; Parish et al. 2008) is to use Poisson weighting (Keenan and equipped with a field emission source and an EDAX r-TEM
and Kotula 2004a, 2004c) followed by PCA (Eigenanalysis to Si(Li) energy-dispersive X-ray detector with a nominal solid an-
reduce the rank) to de-noise the data followed by inverse Poisson gle of 0.13 sr. The probe was measured to be 2 nm FWTM and
weighting. This then provides a model for the data from which had a current of 1.5 nA.
counts can be extracted. In the case of Parish and co-workers
(Parish et al. 2008) counts were extracted from a top-hat filtered Typical regions of the specimen generated count rates of about
PCA de-noised model of a data set by a multiple linear least 10,000 counts per second. Several spectral images were acquired
squares (MLLS) fit of top-hat-filtered measured pure elemental to firstly give a large field-of-view overview and secondly a high-
reference peak shapes. Such a method is suited to general anal- er-resolution look at an interesting precipitate identified from
ysis cases where peaks may be partially or heavily overlapped the overview. Three overview spectral images were acquired at
with those from other elements in the sample (pathological over- 20 nm/pixel, each covering 4 μm by 4 μm regions with 200 by
Chapter 16

laps, for example, such as analysis of Ba and Ti in BaTiO3, as 200 pixels. The dwell time per pixel was 300 msec. The spectral
shown in Shelton et al. (2012)). images were firstly analyzed in their entirety with MCR and aside
from the major elements of Fe and Ni, small Cr-rich Fe-S particles
One note of caution in the use of PCA de-noising is illustrated were also identified. As these sulfide particles were not of interest,
in the supplemental materials from Kotula et al. (2012). In this only Fe and Ni were considered in the quantification. MCR was
analysis we show that use of PCA on spectral image correlated then performed on a sub-set of the original spectral image cover-
the signals from different families of lines from the same element ing 6 to 9 kV, which includes the Fe–K lines and Ni–K lines. Two
Y–L and Y–K lines. In this case, the quantitative elemental im- factors resulted from the MCR analysis, Fe and Ni.
ages derived from fitting to the Y–L and Y–K separately from a
PCA model derived from the entire spectral image looked vir- The spectral shapes were scaled to sum to one count with the
tually identical. Only if we use PCA on sub-sets of the spectral corresponding component images scaled to keep the product of
data containing only the elemental lines from which the k-factors C and S constant.
were measured (e.g., Y–K and Ti–K in that case) can we derive
the true count statistics and, therefore, the true measurement As a result, the component images now represent counts for the
uncertainty. An example of this will be shown below for a Mn- respective elements. A correction was applied to remove the ef-
doped grain boundary in SrTiO3 (Yang et al. 2013) where PCA fect of the Cu grid fluorescence peak. For quantification, kNiFe
rotated to a simple spectral structure of a sub-set of the data was was measured from a standard sample consisting of 25 wt. %
used to accurately extract counts for Ti, Mn, and Sr–K lines for Ni in Fe and found to be 1.10 ± 0.05 (weight-percent basis). The
subsequent quantification. Cliff–Lorimer procedure was then performed on a pixel-by-pixel
basis to convert from counts to composition.

Figure 16.19 shows the high-angle annular dark-field (HAADF)

STEM image of the region from which the spectral images were
 16.5 Example of X-ray and Electron Energy-Loss Spectral Image Acquisition and Analysis 459

acquired as well as the quantification results in montage form.

Little of the Ni variation is obvious from the HAADF image.
This overview, where we have significantly under sampled the

Part I
microstructure (20 nm pixel spacing with a 2 nm probe), can then
direct us to look at smaller regions at higher resolution and this
approach is a reasonable strategy which balances analysis time
while statistically sampling a microstructure.

The analysis flow described above is shown in more detail in

Fig. 16.20, which is a higher-resolution analysis of the region
circled in Figs. 16.19a and b. Figure 16.20a is a HAADF-STEM
image of the region analyzed where the red box shows the re-
gion of the spectral image. The spectral image was acquired
from a region 500 nm by 500 nm with a pixel spacing of 4 nm
(i.e., 128 pixels by 128 pixels). The dwell time per pixel was
300 msec, which resulted in approximately 3,300 counts per
spectrum in the 6 kV through 9 kV spectral region. While this
may seem high compared to the rule of thumb of 100 counts
per spectrum stated previously in this chapter, our goal here is
pixel-by-pixel quantification and, therefore, more counts in each
spectrum is desirable.

As before, we analyzed just the spectral region containing the

peaks of interest with MCR, which resulted in two factors, Fe–K
and Ni–K. Figure 16.20b and c show the component images
scaled to counts for Ni and Fe, respectively. Figure 16.20d shows
the corresponding pure-elemental spectral shapes. After applying
the Cliff–Lorimer procedure to the count images (Figs. 16.20b Fig. 16.19 a HAADF-STEM montage image from the FIB sample of NWA 739.
The inset red boxes show the regions from which the spectral images were ac-
and 16.20c) we arrive at Fig. 16.20e, which is the Ni composi-
quired. b Resulting Ni compositional image montage after MCR and application
tional image. The box indicates the region from which a profile
of a measured k-factor for Ni-Fe. Circles in both a and b highlight an interesting
shown in Fig. 16.20f (integrated over 20 pixels laterally) was
Ni-rich precipitate from which a higher-resolution spectral image was acquired
extracted. (Fig. 16.20)

The profile shows a characteristic M-shape and from this we

Chapter 16
can estimate that the metal particle from which the FIB sample
was extracted experienced reheating sometime after formation to region at the boundary with 300 by 300 pixels which implies a
about 900 K for less than a week followed by cooling to below pixel size of 0.012 nm. The probe size was smaller than 0.136 nm
800 K during a similar period of time, and thus from such data (as verified on Si [011]), and had a convergence angle of 18 mrad
we can understand how such structures formed and were later and a current of ~ 120 pA.
altered.
This spectral image significantly oversampled our resolution
(probe size), which is necessary in this case for the X-ray detec-
tor system’s (Bruker EspritTM) drift-correction routine to func-
16.5.2 Mn-Doped SrTiO3 Grain Boundary tion optimally. The data were acquired in a series of several
Spectral Image Acquisition hundred frames where we continuously raster the beam over
the sample, storing only the X-ray events and dwelling for
and Quantification 40 μsec per pixel per frame. Between frames a drift correction
is performed. Given that our specimen is approximately 30 nm
Yang and co-workers (Yang et al. 2013) analyzed a Mn-doped thick, these analytical conditions do not result in many counts
Σ13 (510)/[001] grain boundary in SrTiO3 with a FEI Titan per pixel. In fact, the average number of counts per spectrum
80-200 with ChemiSTEM technology which was operated at was 8.4 in this case.
200 kV and is equipped with a high-brightness Schottky emitter
(X-FEGTM), a spherical aberration corrector on the probe-form- Since we oversampled the spatial information in our image, we
ing optics (D-COR by CEOS, GmbH) and 4 silicon-drift ener- can afford to spatially bin the data post facto by 3 by 3. This re-
gy-dispersive X-ray spectrometers with a combined solid angle sults in a 9 times increase in signal and 76 counts per spectrum
of 0.7 sr (SuperXTM). An X-ray spectral image was acquired with on average. The resulting binned pixels are now 0.036 nm. This
a microscope magnification of 7.2 MX, from a 3.6 by 3.6 nm follows the approach described by Kotula and co-workers (Kot-
 460 16 X-ray and EELS Imaging

Fig. 16.20  a HAADF-STEM image

the Ni-rich precipitate highlighted in
Fig. 16.19. b, c Component images
Part I

from the MCR calculation scaled to

represent counts. d Corresponding
spectral shapes scaled to sum to
one count for ‘b’ (Fe-blue) and ‘c’
(Ni-green). e Ni compositional im-
age with inset region from which the
integrated profile shown in (f) was
extracted
Chapter 16

ula et al. 2012) as does much of the subsequent analysis. An ini- Figure 16.21a shows the HAADF image and MSA analysis with
tial MCR analysis of the entire spectral image was performed to the spectral shapes color-coded to the corresponding component
verify that there were no unexpected elements present. We then images inset as a color overlay. Figures 16.21b–d and e–g are
re-analyzed (PCA rotated to a simple spectral structure) only the the component images scaled to counts and the compositional
data in the spectral regions of interest which included Sr-L, Ti- images in atomic fraction, respectively. From the latter composi-
K, Mn-K and Sr-K. In order to quantify the data, kSrTi = 1.34 was tional images we can extract a line profile by averaging the com-
measured from an undoped SrTiO3 specimen of similar thickness positions across the images. This is shown in Fig. 16.22a, where
at the [001] zone axis. Since a suitable standard for Mn to Ti we can see that away from the grain boundary, Sr is roughly
was not readily available, a value was calculated from k-factors 50 at.% while the sum of Ti and Mn is also roughly 50 at.%. This
measured at 200 kV (on a different microscope) referenced to Fe would suggest that Mn substitutes for Ti and should, therefore,
(kTiMn = 0.9, W&C, Chap. 35). have a + 4 charge state.
 16.5 Example of X-ray and Electron Energy-Loss Spectral Image Acquisition and Analysis 461

Part I
Chapter 16

Fig. 16.21 a MSA analysis of a spectral image acquired from the inset HAADF image showing the spectral shapes color-coded to their respective component images
in the inset color overlay. b–d Component images from the MSA analysis scaled to represent counts for the elements: Sr (b), Ti (c) and Mn (d). e–g Compositional
images in atomic fraction: Sr (e); Ti (f); and Mn (g). The field of view for all of the images is 3.6 nm
 462 16 X-ray and EELS Imaging

Figure 16.21b shows EELS spectra taken at the Mn-L32 edge from
at the boundary, 5 nm away and far from the boundary in bulk
material. Using the procedure describe in the work by Yang and
Part I

co-workers (Yang et al. 2013 and references therein) to extract the

Mn-L3 and Mn-L2 intensities (shown schematically in Fig. 16.21c),
the charge state of the Mn can be determined. This measurement
shows that, consistent with the compositional analysis, the Mn
is +4 (substituting Ti) away from the boundary, while at the bound-
ary it is +2, likely substituting Sr and also sitting at interstitial sites.

16.5.3 Plasmon Mapping of AG Nanorods: EELS

Spectral Image Analysis

The last example in this chapter examines a much different data

analysis challenge, namely of effectively mapping surface plas-
mons of nanorods with low-loss EELS spectral images. Taken
from the work of Guiton and co-workers (Guiton et al. 2011)
we start with a low loss EELS spectral image of an Ag nanorod
acquired on a VG Microscopes HB501UX STEM operated at
100 kV with Nion aberration corrector and a Gatan Enfina EEL
spectrometer. A low-loss EEL spectral image was acquired with
a dispersion of 0.05 eV/channel and a dwell time of 50 msec. The
energy resolution of this system was approximately 450 meV to
500 meV as measured from the FWHM of the zero-loss peak.

Prior to MSA the zero-loss peak was fit to a Gaussian plus Loren-
tian using Gatan Digital Micrograph, the spectra were all aligned
to have the zero-loss peak centered at 0 eV followed by subtraction
of the zero-loss peak. MSA consisted of noise scaling (Keenan
and Kotula 2004d) and then PCA where the spectral shapes were
rotated to a simple structure and then inversely scaled for noise.
Chapter 16

This method allows the data to self-model itself with the con-
straint that the spectral shapes be simple. With no similar con-
straint on the component images, they are free to overlap sig-
nificantly with each other. The MSA results were compared to
simulations of the optical response and local field enhancements
described in the paper by Guiton et al. (2011). Figure 16.23a
shows the HAADF image of a silver nanorod with the com-
ponent images and corresponding spectral shapes in B and C,
respectively. The results of the calculations shown in Fig. 16.23d
compare well with the MSA-derived components.

Fig. 16.22 a Integrated line profile calculated from Figs. 16.21e–g showing
that away from the interface Sr is about 50 % of the cations with the other
50 % the sum of Ti and Mn suggesting that Mn substitutes Ti. At the boundary,
Mn and Ti increase while Sr decreases. b EELS spectra taken at the Mn-L3,2 edge
indicating that at the boundary Mn has a valence state of +2 while away from
the boundary it is +4 which is consistent with compositional analysis. c Method
for measuring the intensities of the L3 and L2 peaks
 16.5 Example of X-ray and Electron Energy-Loss Spectral Image Acquisition and Analysis 463

Part I
Chapter 16
Fig. 16.23 EELS data and electrodynamics calculations for rod 3. a ADF image of the rod: length = 202 nm, diameter = 40 nm, aspect ratio = 5.1. b MSA score
images, and (c) loading spectra, for each component, interpreted as plasmon maps and energies, respectively. From a total of 12 components found by MSA, two
components were considered overlapping and combined to form the E = 1.2 eV mode, two were combined to form the E = 3.2 eV mode, and two are displayed as
found. d DDA calculated electric field plots displaying the field generated by a plane wave optical excitation at the energies and polarizations given on each panel.
e Summed EEL spectrum

Chapter Summary

Spectral imaging both with X-ray and EELS is a powerful ventionally or with the MSA methods described here (or
method for sampling the chemistry and chemical informa- other approaches), should be performed with an open mind
tion from materials. The challenge comes in both acquir- or, in other words, with as few assumptions as possible.
ing the appropriate signal levels and then analysis of the Then quantification involves the application of additional
data sets. This chapter has provided some general guidance knowledge and assumptions in order to go from counts to
for setting up spectral image acquisition. Whether or not composition for the desired elements.
quantification is the goal, a qualitative analysis, either con-
 464 Appendix

Appendix
Part I

People Legge GJF, Hammond I (1979) Total quantitative recording of

elemental maps and spectra with a scanning microprobe. J
Microsc 117:201–210
Peter Duncumb joined the Tube Investments Lab in 1959 and is Lyman CE, Friel JJ (2006) X-ray mapping in electron-beam in-
credited with first converting the TEM into an AEM by adding a struments. Microsc Microanal 12:2–25
spectrometer – the new instrument was called EMMA. Malinowski ER (2002) Factor Analysis in Chemistry, 3rd edn.
Wiley, New York.
John J. Friel died aged 70 on February 17, 2015, after slipping See papers in Ultramicroscopy Vol. 28, Issues 1–4, Pages
on ice. He was a pioneer in bringing modern spectral imaging to 1–368 (1989) from the NSF/CNRS Workshop entitled, Elec-
the microanalysis community. tron-beam induced spectroscopies with high spatial resolu-
tion, Aussois, France 1988.

Self-Assessment Questions Specific References

Anderson IM (1998) Multivariate statistical analysis of low-volt-
In setting up for an XEDS spectral image we have an instanta- age EDS spectrum images. Microsc Microanal 4(2):272–273
neous count rate from our sample of 3000 counts per second. Bonnet N (1998) Multivariate statistical methods for the analysis
How long do we have to acquire data from each pixel (total per of microscope image series: applications in materials science.
pixel dwell time) to get 100 counts per spectrum? J Microsc 190(1–2):2–18
Bonnet N, Hannequin P (1989) Chemical mapping and multi-
variate statistical analysis prospect. Ultramicroscopy 28(1–
4):248–251
Text-Specific Questions Bonnet N, Brun N, Colliex C (1999) Extracting information from
sequences of spatially resolved EELS spectra using multi-
variate statistical analysis. Ultramicroscopy 77(3–4):97–112
Why can we event-stream or make multiple passes to collect Boyce BL, Michael JR, Kotula PG (2004) Fatigue of metallic
XEDS-SI data but not for an EELS-SI? microdevices and the role of fatigue-induced surface oxides.
Chapter 16

Acta Mater 52(6):1609–1619

We’re setting up for an EELS-SI and notice that we are saturating Bright DS (2006) Lispix public domain image analysis program
the detector. Aside from decreasing the dwell time, what else can for Windows. http://www.nist.gov/lispix
we do to decrease our signal on the detector without changing Bright DS, Newbury DE (2004) Maximum pixel spectrum: a new
the detector collection angle? tool for detecting and recovering rare, unanticipated features
from spectrum image data cubes. J Microsc 216:186–193
Cliff G, Lorimer GW (1975) The quantitative analyses of thin
specimens. J Microsc 103:203–207
References Cosslett VE, Duncumb P (1956) Microanalysis by a flying-spot
X-ray method. Nature 177:1172–1173
ENVI (2013) Commercial generic and multivariate data visual-
ization software. http://www.ittvis.com/envi/
General References Goldstein JI, Jones RH, Kotula PG, Michael JR (2007) Micro-
structure and thermal history of metal particles in CH chon-
Jollife IT (2002) Principal Component Analysis, 2nd edn. drites. Meteoritics Planetary Sci 42:913–933
Springer, New York Goods SH, Buchheit TE, Janek RP, Michael JR, Kotula PG
Keenan MR (2007) Multivariate analysis of spectral images (2004) Oxide dispersion strengthening of nickel electro-
composed of count data. In: Grahn H, Geladi P (eds) Tech- deposits for microsystem applications. Metall Mater Trans
niques and applications of hyperspectral image analysis. John 35A(8):2351–2360
Wiley & Sons, Chichester Gorlen KE, Barden LK, Delpriore JS, Fiori CE, Gibson CC,
Keenan MR (2009) Exploiting spatial-domain simplicity in spec- Leapman RD (1984) Computerized analytical electron-mi-
tral image analysis. Surf Int Anal 41:79–87 croscope for elemental imaging. Rev Sci Instr 55(6):912–921
 Appendix References 465

Guiton BS, Iberi V, Li SZ, Leonard DN, Parish CM, Kotula aging. In: Bailey GW, Ellisman MH, Hennigar RA, Zaluzec
PG, Varela M, Schatz GC, Pennycook SJ, Camden JP (2011) NJ (eds) Proc. Microscopy Microanalysis. Jones and Begell
Correlated Optical Measurements and Plasmon Mapping of Publishing., New York, pp 592–593

Part I
Silver Nanorods. Nano Lett 11(8):3482–3488 Niculae A, Lechner P, Soltau H, Lutz G, Strüder L, Fiorini C,
Hunt JA, Williams DB (1991) Electron Energy-Loss Spec- Longoni A (2006) Optimized readout methods of silicon drift
trum-Imaging. Ultramicroscopy 38(1):47–73 detectors for high-resolution X-ray spectroscopy. Nuclear In-
Jeanguillaume C, Colliex C (1989) Spectrum-image: The next struments and Methods. Physics Research A 568:336–342
step in EELS digital acquisition and processing. Ultramicros- Oliver JM, Pfeiffer JR, Surviladze Z, Steinberg SL, Leiderman
copy 28:252–257 K, Sanders ML, Wofsy C, Zhang J, Fan H-Y, Andrews N,
Keenan MR, Kotula PG (2003) Apparatus and System for Mul- Bunge S, Boyle TJ, Kotula PG, Wilson BS (2003) Membrane
tivariate Spectral Analysis. US Patent # 6(584):413 Receptor Mapping: The Membrane Topography of FcεRI Sig-
Keenan MR, Kotula PG (2004a) Accounting noise in the mul- naling. In: Quinn PJ (ed) Membrane Dynamics and Domains.
tivariate statistical analysis of TOF-SIMS data. Surf Interf Subcellular Biochemistry, vol 37. Kluwer Academic Press/
Anal 36:203–212 Plenum Publishers., New York, pp 3–34
Keenan MR, Kotula PG (2004b) Method of Multivariate Spec- Parish CM, Brennecka GL, Tuttle BA, Brewer LN (2008) Quan-
tral Analysis. US Patent # 6(675):106 titative X-Ray Spectrum Imaging of Lead Lanthanum Zir-
Keenan MR, Kotula PG (2004c) Multivariate Statistical Anal- conate Titanate PLZT Thin-Films. J Am Ceram Soc 91:3690–
ysis of EEL-Spectral Images. Microsc Microanal 10(Suppl. 3697
2):874CD–875CD PLS_Toolbox (2016) Commercial multivariate data analysis
Keenan MR, Kotula PG (2004d) Optimal Scaling of ToF-SIMS software. http://eigenvector.com
spectrum-images prior to multivariate statistical analysis. Shelton CT, Kotula PG, Brennecka GL, Lam PG, Meyer KE,
Appl Surf Sci 231–232:240–244 Maria J-P, Gibbons BJ, Ihlefeld JF (2012) Chemically Homo-
Kotula PG, Keenan MR (2000) Automated unbiased information geneous Complex Oxide Thin Films Via Improved Substrate
extraction of STEM-EDS spectrum images. IoP Conference Metallization. Adv Functional Mater 22(11):2295–2302
Series 165:147–148 Strüder L, Meidinger N, Stotter D, Kemmer J, Lechner P, Leu-
Kotula PG, Keenan MR (2003a) Automated analysis of large tenegger P, Soltau H, Eggert F, Rohde M, Schulein T (1998)
(>4Gb) spectral images with efficient out-of-core RAM al- High-resolution X-ray spectroscopy at close to room tempera-
gorithms. Microsc Microanal 9(Suppl. 2):152–153 ture. Microsc Microanal 4:622–631
Kotula PG, Keenan MR (2003b) X-ray Spectral Imaging in the Tauler R, de Juan A (2013) Multivariate curve resolution home
STEM for Microelectronics Failure Analysis. Microsc Micro- page. http://www.cid.csic.es/homes/rtaqam/tmp/WEB_MCR/
anal 9(Suppl. 2):998–999 welcome.htm
Kotula PG, Keenan MR (2006) Application of Multivariate Sta- Titchmarsh JM (1997) Extraction of Information from STEM-
tistical Analysis to STEM X-ray Spectral Images: Interfacial EDX Segregation Profiles Using Multivariate Statistical
Analysis. Microsc Microanal S2:937–938

Chapter 16
Analysis in Microelectronics. Microsc Microanal 12(6):538–
544 Titchmarsh JM (1999) EDX spectrum modeling and multivariate
Kotula PG, Keenan MR, Michael JR (2003a) Automated analysis analysis of sub-nanometer segregation. Micron 30:159–171
of EDS spectral images in the SEM: a powerful new micro- Titchmarsh JM, Dumbill S (1996) Multivariate statistical anal-
analysis technique. Microsc Microanal 9(1):1–17 ysis of FEG-STEM EDX spectra. J Microsc 184(3):195–207
Kotula PG, Klenov DO, von Harrach HS (2012) Challenges to Titchmarsh JM, Dumbill S, Vatter IA (1995) Investigation of
Quantitative Multivariate Statistical Analysis of Atomic-Res- Interfacial Segregation in Steels Using Multivariate Statis-
olution X-Ray Spectral. Microsc Microanal 18:691–698 tical Analysis of EDX Spectra. Microbeam Analysis. VCH
Kotula PG, Michael JR (2006) Spectral Imaging and Multivari- Publishers, New York, NY, pp 259–260
ate Statistical Analysis from Thin Specimens in the SEM with Watanabe M, Williams DB (2003) Quantification of elemental
a Four-Channel Silicon Drift Detector. Microsc Microanal segregation to lath and grain boundaries in low-alloy steel by
12(Suppl. 2):1390–1391 STEM X-ray mapping combined with the ζ-factor method. Z
Loehman RE, Kotula PG (2004) Spectral imaging analysis of in- Metallk 94:307–316
terfacial reactions and microstructures in brazing of alumina Watanabe M, Ackland DW, Burrows A, Kiely CJ, Williams DB,
by a Hf-Ag-Cu alloy. J Am Ceram Soc 87:55–59 Krivanek OL, Dellby N, Murfitt MF, Szilagyi Z (2006) Im-
Mathematica™ (2015) Commercial generic and multivariate provements in the X-ray analytical capabilities of a scanning
data analysis software. http://wolfram.com transmission electron microscope by spherical-aberration
Matlab™ (2015) Commercial generic and multivariate data anal- correction. Microsc Microanal 12:515–526
ysis software. http://mathworks.com Weiss JK, Rez P, Higgs AA (1992) A computer-system for im-
Mayer TM, Elam JW, George SM, Kotula PG, Goeke RS (2003) aging and spectroscopy in analytical electron-microscopy.
Atomic-layer deposition of wear-resistant coatings for micro- Ultramicroscopy 41:291–301
electromechanical devices. Appl Phys Lett 82(17):2883–2885 Wilson BS, Pfeiffer JR, Surviladze Z, Steinberg SL, Zhang J,
Mott RB, Waldman CG, Batcheler R, Friel JJ (1995) Position Lieberman K, Kotula PG, Oliver JM (2004) High resolution
tagged spectrometry: A new approach for EDS spectrum im- imaging shows distinct and independent behavior for GPI-an-

Licensed to ([email protected])
 466 16 X-ray and EELS Imaging

chored proteins, gangliosides and receptors in native mem-

branes. Mol Biol Cell 15(6):2580–2592
Yang H, Kotula PG, Sato Y, Ikuhara Y, Browning ND (2013)
Part I

Segregation of Mn2+ Dopants as Interstitials in SrTiO3

Grain Boundaries. Mater Res Lett. http://dx.doi.org/10.1080
/21663831.2013.856815
Chapter 16
 
Practical Aspects
17

Part I
and Advanced Applications
of XEDS

Chapter Preview

The details and applications of X-ray analysis in TEM – Simulation of X-ray spectra for thin-film analysis.
were described comprehensively in Chaps. 32–36 of W&C. – Details of quantitative X-ray analysis of thin-films by the
These chapters cover all essential topics you should know ζ-factor method.
prior to your own applications of X-ray analysis; from the – Contemporary topics including recent advances in X-ray
fundamentals of spectrometers, qualitative elemental anal- analysis.
ysis, and quantitative composition determination through
advanced spatial resolution and analytical sensitivity of Each section is independent of the others, but closely related
XEDS. The main objective of this chapter is to complement to corresponding chapters in W&C. In addition to Chaps.
W&C by introducing more advanced XEDS approaches in 32-36 of W&C, you can also read Lyman et al (1990),
the TEM while not increasing the level of difficulty. This Williams and Goldstein (1991) and Zemyan and Williams
chapter consists of four sections (1995) on fundamentals of X-ray analysis in TEM.
– Practical characterization of XEDS detectors in TEM.

Chapter 17

© Springer International Publishing Switzerland 2016

C. B. Carter, D. B. Williams (Eds.), Transmission Electron Microscopy, DOI 10.1007/978-3-319-26651-0_17 467
 468 17 Practical Aspects and Advanced Applications of XEDS

17.1 Performance Parameters of before applying any measurement of the following fundamental
system parameters in AEM. By using the NiOx test specimen,
XEDS Detectors you can easily calibrate the energy dispersive range with the
Part I

Ni Kα peak at 7.48 keV, in combination with either the O Kα

As we suggested in Chaps. 32 and 33 of W&C, you need to know peak (0.52 keV) or the Ni Lα peak (0.85 keV). If you need to
about your XEDS detector in terms of its fundamental performance calibrate the energy range > 10 keV then you can use the Mo
parameters before utilizing it for any practical analysis. You should Kα peak at 17.42 keV. Although any characteristic X-ray signal
measure some detector parameters at least once when your system from the grid bar is always undesirable (since it proves there are
is installed or significantly upgraded. Other parameters also need to spurious X-ray signals in your spectrum), the high-energy X-ray
be measured time to time to monitor your system performance in a peak is actually useful for this calibration.
consistent manner. Fortunately, you can measure most of the XEDS
characteristics using Egerton’s NiOx test specimen (Egerton, Cheng, Detector Energy, Resolution: We define the energy resolution of
1994); this is why W&C strongly recommends that you purchase the an XEDS detector as the full-width at half-maximum (FWHM)
NiOx film or similar NiO films – the x is nicely ambiguous! value at the Mn Kα peak position (5.9 keV). You can derive this
value from the FWHM value of the Ni Kα peak in an XEDS
The NiOx Test Specimen spectrum from the NiOx film as described by Bennett and
Egerton (1995)

It consists of ~ 50-nm-thick NiO film on ~ 20-nm-thick

amorphous carbon support on a 200 mesh Mo grid. Be
RFWHM .MnK ˛/ = 0:926 RFWHM .NiK ˛/ (17.1)
careful if you make your own! 9
The FWHM value of the Ni Kα peak can be best determined by
fitting to a Gaussian, as shown in Fig. 17.1a. You can see that
You can use this test specimen for calibrations of the camera the Gaussian fit curve deviates slightly from the Ni Kα peak on
length, aperture angles, EELS analysis, etc. Besides those cali- the low-energy side. This deviation from the Gaussian is mainly
brations, the following parameters can be determined for XEDS- caused by incomplete charge collection, which can be evalu-
based applications from a single XEDS spectrum of the NiOx ated if you measure the ratio of the full-width at tenth-maximum
thin film (FWTM) to the FWHM values of the peak (see Sect. 32.9A in
W&C).
zz Detector energy resolution: described in Sect. 32.8 of W&C.
zz Peak-to-background ratio in general and the Fiori definitions: Take Time
mentioned in Sect. 33.4A of W&C.
zz Inverse hole-count (a.k.a film count): explained in Sect. 33.3A
of W&C. The longer the process-time setting, the better the detec-
zz Mo K/L ratio: described in Sect. 33.3A of W&C. tor energy resolution. A higher number on the x-axis of
zz Detector collection angle: described in Sect. 32.10A of W&C. the plot gives a lower energy range. 9
zz Detector efficiency: defined in Sect. 33.4B of W&C.

You can find detailed descriptions about these various determi- The detector energy resolution depends on how you set the pro-
nations in Egerton’s original papers. In addition, you can also cess time (or time constant). If you choose a shorter process
monitor some of the detector parameters (such as thicknesses time, the energy resolution is degraded despite having higher
of ice and carbon in front of the active area) if you gather sev- count rates. Figure 17.1b shows the detector energy resolution
eral spectra from the NiOx film over different time periods. The measured from three different AEMs with a Si(Li) detector op-
detection of X-rays from light elements (which produce low-en- erated at 100, 200 and 300 kV, plotted against the process-time
Chapter 17

ergy soft X-rays) is significantly influenced by the thicknesses of setting.

these various layers. You can measure quantitatively the growth
of ice and carbon layers, which directly degrade detection of Peak-to-Background Ratio: We can use the peak-to-background
these soft X-rays. In this section, we will show you how to mea- (P/B) ratio as a measure to describe how well the XEDS-AEM
sure these parameters in practice and will introduce software that system is configured. Typically a higher P/B value indicates a
allows you to carry out these measurements easily. better-configured system. As described in Sect. 33.4A of W&C,
there are a number of definitions for P/B. The most common
definition is to use the full peak intensity and the background
intensity with the same energy window as the peak.
17.1.1 Detector, Fundamental Parameters
The above definition of P/B may depend on the energy res-
olution of the spectrometer system since the energy window
Energy Calibration: The X-ray energy range and dispersion size for the background varies with the energy resolution. To
must be calibrated before performing any X-ray analysis or even compare the P/B values taken from different instruments (or
 17.1 Performance Parameters of XEDS Detectors 469

Part I
Fig. 17.1 a Example of energy resolution determination using the Ni Kα peak from the NiOx test specimen. For the determination, the NiOx plug-in in Gatan Dig-
italMicrograph was used. b The determined energy resolution of XEDS systems of three different instruments (100, 200, and 300 kV), plotted against the process
time setting

even at different process time setting) fairly, you should use the Table 17.1 Typical energy windows of characteristic X-ray lines for the NiOx
Fiori definition for P/B (Fiori et al. 1982), in which the average (Lyman et al. 1994)
background intensity at a single channel (e.g., 10 eV) is used
X-ray line Energy window range (keV)
rather than the full background intensity. The typical energy
windows of characteristic X-ray lines for the NiOx are sum- Peak (P) Back­ground 1 Back­ground 2
marized in Table 17.1. The background intensities are ideally (B1) (B2)
determined by averaging two intensities at lower and higher
energy regions of the peak. According to Bennett and Egerton, O Kα 0.38–0.64 1.10–1.36
this value for the Ni Kα from the NiOx test specimen can be Ni Lα 0.68–1.00 1.10–1.42
as high as 3,000 in modern commercial 200-kV-AEMs. If a
Ni Kα 7.16–7.76 6.50–7.10 8.60–9.20
pure-element film such as a Cr film is used, your measured P/B
value should be higher than that from the NiOx film because Mo Lα 2.16–2.50 1.10–1.44 3.00–3.34
the Cr Kα intensity corresponds to 100 wt% in the pure Cr film, Mo Kα 17.04–17.80 16.10–16.86 18.14–18.90
whereas the Ni Kα intensity represents 78.6 wt% in the NiOx
film (50 : 50 in atomic ratio).

Inverse Hole Count (Film Count): The inverse hole count (IHC) described in Sect. 33.3.A of W&C. Stray X-rays would produce
is another useful measure we can use to evaluate the presence of a high Mo K/L ratio since high-energy bremsstrahlung X-rays
stray electrons and X-rays in an AEM column. We define IHC as preferentially fluoresce the high-energy Mo Kα line. In contrast,
the intensity ratio of a major X-ray line in the specimen (which stray electrons would excite the Mo Lα rather than the Mo Kα.
should be detected) to an X-ray line from a grid material (which Generally, the Mo K/L ratio decreases with an increase of the
should not be detected in an ideal column). In the NiOx test probe-forming aperture size and increases with an increase in
specimen, IHC is given as the ratio of the intensities of the Ni Kα the incident electron energy. Thus, using the Mo K/L value, you
to the Mo Kα peaks above background. In modern commercial can diagnose stray irradiations of the illumination system in your
AEMs, this value is usually in range from 3–7. instrument.

Chapter 17
Using the P/B and IHC values determined from a single spec- Detector Collection Angle: As mentioned in Sect. 32.10A of
trum in the NiOx test specimen, you can evaluate the perfor- W&C, the collection angle for X-ray detection (Ω) is the most
mance of your specific XEDS-AEM. Figure 17.2 shows a plot important parameter in all of X-ray analysis. We generally want
of P/B (the Fiori definition) against the IHC, measured in three to detect more X-rays, so a larger Ω value is always preferable.
different instruments operated at 100, 200 and 300 kV (these In fact, you may use the latest silicon drift detectors (SDDs)
are actually the same instruments in Fig. 17.1b). The error bars with significantly improved solid angle as described later in
indicate a 99 % confidence limit (3σ). As we mentioned above, Sect. 17.4. You can measure the Ω value nominally from the
higher values both of P/B and IHC are preferable. Therefore, geometric configuration of your AEM-XEDS interface, using
instruments at the top right-hand corner should give supe- Eq. 32.4 in W&C. However, such a practical measurement of
rior X-ray analyses. Ω is not very straightforward. The Ω measurement requires
that you take a reference spectrum from a known standard in a
Mo K/L Ratio: You can use the ratio of the Mo Kα to the Mo Lα specific instrument with calibrated Ω in order to compare X-ray
intensities to evaluate stray irradiation in your instrument as was intensities.
 470 17 Practical Aspects and Advanced Applications of XEDS
Part I

Fig. 17.2 Determined P/B (Fiori definition) plotted against the IHC, measured Fig. 17.3 Comparison of X-ray spectra obtained from the same NiOx test spec-
from three different instruments operated at 100, 200, and 300 kV imen in the same instrument at different periods of time

Table 17.2 A prefactor C1 for Ω determination at a given an incident beam

17.1.2 Monitoring Detector Contamination
energy E0 (Bennett and Egerton, 1995).

E0 (kV) 80 100 120 200 300 400

The detection performance of your XEDS detector may
C1 2.6 2.9 3.2 4.0 4.8 5.3
change considerably with time, especially in its ability to de-
tect low-energy, soft X-rays. For example, Fig. 17.3 compares
two X-ray spectra acquired from the NiOx film in the same
Fortunately, we can use an empirical formula for the Ω determi- instrument with a windowless XEDS detector at different time
nation based on a single NiOx spectrum periods (one was recorded just after detector conditioning and
other was a few months later). As you can see, although there
˝ = C1 cos  INiK =.t  Ip / (17.2) is no noticeable difference in the Ni Kα line, the intensity
level is significantly reduced in the lower-energy O Kα and Ni
where C1 is a factor depending on the incident beam energy Lα peaks. This difference in the X-ray detection efficiency at
(summarized in Table 17.2 or you can calibrate it beforehand), lower energy regions may lead to serious problems, especially
INiK is the Ni Kα intensity above the background, t is the film if you are carrying out quantitative analysis using lower-energy
thickness, τ is the acquisition time and Ip is the probe current, soft X-rays.
respectively. In addition, θ is the total tilt angle including tilt
angles in both x and y directions (θx and θy, respectively) and Residual Gases
simply given as:

In a typical microscope column, the major residual gases

q 
 = tan−1 tan2 x + tan2 y (17.3) are water vapor and hydrocarbons. 9

Relative Detector Efficiency: Once you have measured a Such degradation of soft X-ray detection can be caused by
more accurate Ω value through Eq. 17.2, the relative detector contamination (mainly in front of the detector window) and
efficiency can be determined. We define this relative detector the degradation rate is strongly related to the vacuum condi-
efficiency in terms of how many X-ray counts per second are tions of your instrument. The active area of your detector (an
Chapter 17

collected from a standard specimen (such as the pure Cr film and XEDS detector should be cooled by liquid N2 or by a Peltier
the NiOx specimen) in the measured collection angle (sr) with device) easily condenses water vapor into an ice layer, and
the incident probe current (nA). The relative detector efficiency hydrocarbons may also be condensed as a carbonaceous layer.
can then be given as:
Figure 17.4 shows the absorption ratio of the O Kα, Ni Lα and
.relative detector efficiency/ = INiK =. Ip ˝/.cps=nA=sr/ (17.4) Ni Kα lines, plotted against thickness of the ice (a) and carbon
(b) layers. The ice layer build-up influences the Ni Lα line
most severely. When the ice layer reaches ~ 1 μm, 50 % of the
You should measure all the parameters described above at least Ni Lα intensity is absorbed in the ice layer. For the O Kα, the
once, ideally when your system is installed. If you don’t know ice build-up is less sensitive. Over 5 μm of ice layer absorbs
these parameters, it is not too late to determine them now. The ~ 50 % of O Kα intensity. In contrast, the carbon layer signifi-
determination is straightforward since only a single spectrum cantly influences both the O Kα and Ni Lα lines, as shown in
needs to be acquired from your NiOx test specimen. Fig. 17.4b.
 17.1 Performance Parameters of XEDS Detectors 471

with
OK OK NiK
O = f.=/O − .=/O gice

Part I
OK OK NiK
C = f.=/C − .=/C gC
(17.7)
NiL NiL NiK
O = f.=/O − .=/O gice

NiL NiL NiK

C = f.=/C − .=/C gC

If you take ratios of the above intensity ratios (IOK/INiK and INiL/
INiK) between two independent measurements (subscripts 1 and
2) you can derive the following equations:
.IOK =INiK /1
= exp −OK OK
 
ROK = O tice − C C
.IOK =INiK /2
.INiL =INiK /1 (17.8)
= exp −NiL NiL
 
RNiL = O tice − C C
.INiL =INiK /2

where ∆tice and ∆tC are thickness differences in ice and carbon
layers between measurement 1 and 2, respectively, i.e., the rela-
tive layer growth of ice and carbon between your two measure-
ments. Then, ∆tice and ∆tC can be derived as:

NiL OK
C ln.ROK / − C ln.RNiL /
Fig. 17.4 Calculated absorption ratios of the O Kα, Ni Lα and Ni Kα lines, plot- tice =
ted against the ice thickness (a) and the carbon thickness (b) in front of the C O − NiL
OK NiL
C O
OK

detector active area (17.9)

OK NiL
C ln.RNiL / − C ln.ROK /
tC =
C O − NiL
OK NiL
C O
OK

With a 1 μm carbon layer, 90 % of the O Kα and 50 % of the Ni In the above equation, the only parameters you need are the
Lα X-rays are absorbed. Conversely, absorption of the Ni Kα densities of ice and carbon, and the MACs, which you can find
X-rays is negligible both in the ice and the carbon (up to 10 μm). summarized in Table 17.3. Using these parameters, Eq. 17.9 can
In addition to the reduction of soft X-ray detection, accumulation be simplified thus:
of the ice and carbon layers may also cause (1) a decrease in the    
background intensity even at higher energies (which results in tice = 481:4 ln
.IOK =INiK /1
− 1689:8 ln
.INiL =INiK /1
an increase in your P/B ratio) and (2) a decrease in dead time .IOK =INiK /2 .INiL =INiK /2
during X-ray acquisition. Therefore, it is essential that you mon-
   
.INiL =INiK /1 .IOK =INiK /1
itor the accumulation of ice and carbon layers on your detector. tC = 69:0 ln − 397:2 ln
.INiLw =INiK /2 .IOK =INiK /2 (nm)

X-ray absorption due to the ice and carbon layers can be modeled  (17.10)
as (Hovington et al. 1993)
In the derivation of Eq. 17.10, we used the MAC values reported
A = exp −.=/X X
 
O ice tice − .=/C C tC (17.5) by Heinrich 1987. If you want to determine the contamination

Chapter 17
layer thicknesses, you have to take two spectra in almost iden-
Where .=/X O and .=/C are the mass absorption coefficients
X
tical conditions, i.e., with the same specimen tilt, time constant,
(MACs) of X-ray X in oxygen and carbon, ρice and ρC are the etc. However, by taking the double ratio, you can cancel out any
densities of ice (0.917 Mg/m3) and carbon (2.267 Mg/m3), and tice differences in the acquisition time and the probe current.
and tC are the thicknesses of ice and carbon layers. You should
note that the MAC value for oxygen is used instead of that in ice
in the above equation because the MAC value for hydrogen is
very low for all X-rays. In particular, for the NiOx test specimen, 17.1.3 Software to Determine Detector
the following equations are relevant: Parameters
.IOK =INiK / / exp OK OK
 
O tice − C tC
(17.6) You can measure many detector parameters including the de-
.INiL =INiK / / exp NiL NiL
 
O tice − C tC
cay of detector performance by acquiring a single spectrum
 472 17 Practical Aspects and Advanced Applications of XEDS

Table 17.3   Summary of parameters for determination of contamination layer

growth at an XEDS window. The MAC values are calculated from Heinrich’s for-
17.2 X-ray Spectrum Simulation –
mula (Heinrich 1987) a Tutorial and Applications
Part I

For ice For carbon of DTSA

ρice = 0.917 Mg/m3 ρC = 2.267 Mg/m3
Planning your experiments is one of the most important tasks
O = 119.3 m /kg
.=/OK C = 1,168.6 m /kg
.=/OK
2 2
for all microscopists because preparation for TEM experiments,
O = 680.3 m /kg
.=/NiL 2
C = 333.3 m /kg
.=/NiL 2 including specimen preparation and actual microscopy sessions,
requires significant amounts of time and money – the cost of run-
O = 1.395 m /kg
.=/NiK C = 0.545 m /kg
.=/NiK
2 2
ning and maintaining a TEM – it’s not free! In order to plan (or
ideally to optimize) your experiments, it would be very efficient
to predict i) the feasibility of each experiment, ii) difficulties of
measurements, and iii) the fundamental limits of planned analy-
or a set of spectra from the NiOx test specimen. To perform ses. Theoretical simulation is one of the best approaches to con-
these measurements routinely and consistently, use the plug-in firm your experimental limits before you conduct actual exper-
for Gatan DigitalMicrograph. To run any functionalities in this iments if, of course, appropriate simulation tools are available.
plug-in package, you have to have Gatan DigitalMicrograph (see
Chap. 6). Next, we’ll introduce two additional plug-ins, in which Fortunately, for XEDS, several tools are available to simulate an
all the evaluation procedures of detector performance described X-ray spectrum: some are based on a first principles approach
above are implemented. such as DTSA (Fiori et al. 1992) while others are combined
with Monte Carlo techniques, e.g., the electron flight simulator
NiOx: this plug-in determines all the fundamental detector pa- (Brundle et al. 1996). (See References for Software for both.)
rameters described in Sect. 17.1.1 above. Figure 17.5a shows In the current version of DTSA-II (Ritchie 2008), however, the
the dialog of the NiOx plug-in. All you need to do is to i) open function for spectrum simulation is only available for bulk sam-
a spectrum from the NiOx test specimen, ii) select the detector ples. While writing, we are still waiting for a thin-film version
parameters you want to determine by clicking the check boxes, of the spectrum generation. Therefore, in this section, we will
and (iii) click the ‘Calculate’ button. If you select the ‘Solid explain X-ray spectrum simulation using DTSA instead. Perhaps
angle and Relative detector efficiency’, an additional sub-dialog the thin-film version of X-ray spectrum simulation will be imple-
appears (Fig. 17.5b) so you can input several instrumental and mented in either DTSA-II or other modern software packages by
experimental parameters. An example of the output is shown in the time you are reading this chapter.
Fig. 17.5c.
DTSA
NiOxIceC: this plug-in determines the accumulated ice and car-
bon thicknesses from two spectra measured in different time pe-
riod from your NiOx test specimen. As we already mentioned The most recent version of DTSA is DTSA-II; it includes
above, you must acquire these spectra in the same instrument un- both a first principles approach and Monte Carlo tech-
der almost identical experimental conditions. Figure 17.6 shows niques. Use it to generate X-ray spectra before you go to
the dialog of the plug-in (A) and an example of the output (B) TEM. 9
from the two spectra shown in Fig. 17.3. In this particular case,
the accumulated ice and carbon layers are 1.67 ± 0.04 μm and
53.1 ± 8.5 nm with 99 % confidence limits (3σ), respectively. In order to simulate an X-ray spectrum properly, you need to un-
derstand several fundamental parameters related to X-ray genera-
In addition to these plug-ins, this package also contains a func- tion and detection. After we introduce DTSA in this section, we’ll
Chapter 17

tion to import a spectrum in the EMSA/MAS file format (EMSA illustrate the fundamental parameters for X-ray analysis by a brief
file import). The EMSA/MAS file format is the one most com- tutorial for the X-ray spectrum simulation. Then, we’ll explain
monly used in the electron microscopy community and most further details of X-ray generation and detection by simulating
XEDS manufactures support this format in their systems. You several different stages of X-ray spectra, (which are never actually
should check the system manual or consult the manufacturer to seen in practice) for example what an X-ray spectrum looks like
find out how to save a spectrum in the EMSA format. As long before it is emitted from your specimen and before it is processed
as your spectrum is saved in the EMSA format, you can convert by the XEDS detector electronics, etc. We can only demonstrate
it to DigitalMicrograph and your detector parameters can be de- this feature by a first principles simulation, which is very useful
termined via the above plug-ins. You can find a standardized to help you understand the physics behind X-ray generation and
description of the EMSA format through the International Orga- detection.
nization for Standardization.
Once you understand the appropriate procedures for X-ray spec-
trum simulation using DTSA, you will be ready to apply it to
 17.2 X-ray Spectrum Simulation – a Tutorial and Applications of DTSA 473

Part I
Fig. 17.5 a Main dialog of the NiOx plug-in package. b Sub-dialog for determination of the collection angle and the relative detector efficiency, and c An example
of the output for the determination of the collection angle and the relative detector efficiency in a 200 kV instrument

Fig. 17.6 a Main dialog of the NiOxIceC plug-in package. b An example of the output for determination of the accumulated layer thicknesses of ice and carbon,
measured in a 300 kV UHV instrument with a windowless XEDS detector

Chapter 17
predict your experimental spectra. Finally, we’ll show you four 17.2.1 What Is DTSA?
practical applications of the X-ray spectrum simulation at the end
of this section. These four are
The function of an electron-excited X-ray spectrum generator
zz Confirmation of peak overlap (related to Sect. 34.3 of W&C), is one of the main features of DTSA, which was implemented
zz Evaluation of X-ray absorption in a thin specimen (related to as a ‘dry lab’ by Fiori to predict experimental limits. This
Sect. 35.6 of W&C), generator is available both for bulk samples and for thin films.
zz Demonstration of AEM-XEDS interfaces (related to Besides this powerful feature, DTSA provides several use-
Sect. 33.4A of W&C), ful functions for X-ray analysis such as peak deconvolution,
zz Estimation of the detectability limits (see Sect. 34.4 and 36.4 background subtraction, quantitative matrix corrections for
of W&C). bulk samples and Cliff–Lorimer quantification for thin speci-
 474 17 Practical Aspects and Advanced Applications of XEDS
Part I

Fig. 17.7 Screen shot of DTSA running via SheepShaver under a Windows platform

mens, along with a comprehensive database related to X-ray las Ritchie at NIST, and hence can run in Windows, Mac OS X
Chapter 17

analysis. and Linux. Therefore, in the near future, all the descriptions in
this section may be replaced with those adjusted for DTSA-II,
Fiori developed the original version of DTSA using the Pas- once DTSA-II provides the spectrum generation function for
cal program language for an Apple Macintosh II and the last thin films!
versions (2.5 or 3.0.1) run in a Power Macintosh under classic
OS 8 and 9. DTSA does run under the so-called classic mode in We should mention that you can run the PowerMac version (2.5)
OS X if the CPU is either G4 or G5. However, the DTSA runs of DTSA either in Windows (as shown in Fig. 17.7) or in Mac
neither in a newer Mac with an Intel CPU nor in any Windows/ OS X, or even in Linux, via a PowerMac emulator called ‘Sheep­
Linux PC. So, we keep an iMac (generation 2000) just to run Shaver’ (see References to find this software package). So, while
DTSA! While many of you will not have access to such museum it is a temporary solution for now, it should also work for Mac
pieces, fortunately, DTSA-II (the successor to DTSA) has been OS X and Linux systems.
developed as a JAVA-based, platform-free, software by Nicho-
 17.2 X-ray Spectrum Simulation – a Tutorial and Applications of DTSA 475

17.2.2 A Brief Tutorial of X-ray Spectrum

Simulation for a Thin Specimen Using

Part I
DTSA

Once you successfully launch DTSA, select ‘Generate > Thin

Target Spectrum’ from the pull-down menu to simulate an
X-ray spectrum for a thin specimen. Then, the main dialog
for thin-specimen spectrum generation appears as shown in
Fig. 17.8. In this main dialog and subsequent dialogs, you can
fill in or select the parameters required for simulation. Note that
you can also simulate an X-ray spectrum for a bulk sample when
‘Bulk Target Spectrum’ is selected. The difference between the
bulk and thin targets is minimal with respect to parameter set-up
for spectrum simulation. Fig. 17.8 Screen shot of the main dialog for thin-specimen spectrum generation
Detector Parameters: If you click the ‘Detector Parameters’ in DTSA
button in the main dialog, you’ll see that a dialog for detector
parameters appears (Fig. 17.9a). In this dialog, you can (A)
select/input several detector configuration parameters and (B) If you prefer to change the energy dispersion (eV/channel), click
define the detector-window parameters. Select your detector the ‘Channel Spec’ button to select the eV per channel and the
type (either Si or Ge) and input the energy resolution that you number of channels (Fig. 17.9c). For a typical X-ray spectrum
should have determined using the NiOx specimen and any from a thin specimen, 10 eV per channel with 2,048 channels
other parameters associated with your detector and window are good values. Then, click ‘OK’ to return to the main dialog.
configurations. If you are not sure about your specific detector
configuration, consult the lab manager or the manufacturer, or Specimen Information: Specific information about your
take values from Table 32.1 in W&C. You can also load typical specimen should be input in the main dialog. All you need to
values by clicking the ‘Load Typical Values’ button (remember input is the constituent elements and their compositions, the
however, these typical values could be very different from those specimen density and its thickness. For element information,
for your particular instrument). It should be mentioned that many either an atomic number such as ‘8’ or an atomic symbol such as
XEDS detectors attached to modern AEMs are Si(Li) detectors ‘Mg’ can be used. You can also select several different units for
or SDDs with an atmospheric thin-window (ATW), which has compositions. Note that the spectrum simulation does not require
a Moxtek polymer layer of ~ 0.3 μm and an Al contact layer of the density value for the bulk target (one of the major differences
~ 0.04 μm, as shown in Fig. 17.9a. between the bulk and thin targets simulations). As an example,
in Fig. 17.8, information about the NIST standard reference
material (SRM) glass, thin film, 2063a is shown. This specimen
Typical Detector Configurations has had its composition, density and thickness already certified
by the NIST SRM program – hence by the US government! You
In modern instruments the take-off angle is 20–25o and can also save the input information to a Database in DTSA.
the specimen-detector distance is ~ 15 mm or less. 9
Experimental Information: Experimental conditions can also
be set up in the main dialog (Fig. 17.8). Only three parameters
After setting up the above parameters, click the ‘Detector-Spec- are required for the spectrum simulation: the incident electron
imen Geometry Dialog’ button (C). Another sub-dialog appears energy, probe current, and acquisition time.
in which you can input the geometry configuration between the

Chapter 17
XEDS detector and instrument, as shown in Fig. 17.9b. There Physics Choice: The last parameters you need to set up are
are two most important parameters you have to input in this dia- related to the physics of X-ray generation. These parameters can
log, namely the ‘Detector Elevation Angle’ corresponding to the be selected by clicking the ‘Physics Choice’ button available in
X-ray take-off angle and the ‘Specimen-Detector Distance’ that the main dialog. An independent sub-dialog appears, as shown
controls the X-ray collection angle (see Sect. 32.10A of W&C). in Fig. 17.10. In this sub-dialog, models of the ionization cross-
Again, you should consult the manufacturer if you are not aware section for K-, L- and M-shells, and a cross-section model for
of these parameters. Alternatively, you can reverse-calculate the continuum X-rays can be selected. In addition, you can also
specimen-detector distance from the detector collection angle us- select scaling-factor values for each of the cross-section models.
ing Eq. 32.4 in W&C, if your collection angle has been determined As we will describe in the following subsection, electron-
using standard specimens such as NiOx or NIST standard spec- induced X-ray signals are directly proportional to magnitudes
imens, which we describe below. If every parameter looks OK, of the cross-sections. Therefore, the physics settings chosen in
you can return to the main dialog by clicking the ‘OK’ buttons. this sub-dialog are very important to match with experimental

Licensed to ([email protected])
 476 17 Practical Aspects and Advanced Applications of XEDS

Fig. 17.9 Screen shots of the dia-

logs for detector parameters (a), for
detector geometry configuration (b),
Part I

and for channel settings (c)

Chapter 17
 17.2 X-ray Spectrum Simulation – a Tutorial and Applications of DTSA 477

Fig. 17.10 Screen shot of the dialog

for physics choice

Part I
spectra. However, unless you have any specific idea in terms 17.2.3 Details of X-ray Simulation in DTSA
of the selection of models and their scaling factor, or are very
familiar with these models and concepts, you can leave them
with default values or just click the ‘Load Defaults’ button. You A thorough understanding of X-ray generation and detection is
can obtain preferential choices of the cross-section models with useful if you need to interpret X-ray spectra acquired from a
the corresponding scaling factors in determination of the ζ-factor thin specimen and also for further quantitative analysis. New-
method, as we explain in the next section. After setting up the bury et al. (1995; see their 2014 article for DTSA-II) reviewed a
physics choices, click the ‘OK’ button to return the main dialog. series of simulation procedures for an X-ray spectrum in DTSA
in which the spectrum simulation for bulk targets is mainly fea-
Finally, an X-ray spectrum can be generated by clicking the tured. Unfortunately, we don’t know of any detailed description
‘Generate’ button in the main dialog. Then, the generated X-ray available for the spectrum simulation for thin targets. So, here,
spectrum is displayed as shown in Fig. 17.11a. This spectrum is we explain step-by-step the details of the X-ray spectrum simu-
ideal and clean, i.e., it has none of the spurious effects described lation for a thin target.
in Chaps. 32 and 33 of W&C. At least, noise can be added to Characteristic X-ray Generation: In a bulk sample, the
this spectrum, which is essential when you need to estimate an- generation of characteristic X-rays is a very complicated series
alytical sensitivity. To add noise, you should select ‘Math > Add of phenomena and many aspects need to be taken into account
Poisson Noise to Work’ from the pull-down menu, and then an- for spectrum simulation. In contrast, the X-ray generation
other spectrum will appear with noise as a ‘Results’ spectrum, process in a thin specimen is much more straightforward. In a
as shown in Fig. 17.11b. (Note: DTSA can display ten spectra thin specimen, we can make a few key assumptions:
simultaneously.) Yes, we add noise to the simulation rather than
remove noise from the experimental data! zz The energy loss of the incident electrons is negligible, be-
cause the energy loss can be as low as only 3 % (in up to

Chapter 17
As we just described, one of the great advantages of DTSA is 500-nm-thick specimens) even at 100 kV.
that you can go into the software dialog and try setting a whole zz Electron backscattering loss is also ignored since most of the
different set of parameters. This approach allows you to see what scattering events in TEM are forwardscattering.
effect the different values have on the generated spectrum of your zz The depth distribution of X-ray production is unity.
particular specimen on which you are depending for your re-
search. You’ll immediately find which parameters are important With these approximations the characteristic X-ray generation
and which do not affect your spectrum significantly. This aspect in a thin specimen can be simplified to be proportional to the
of spectrum simulation will of course help you in your general mass thickness (ρt) and the characteristic X-ray intensity Ic at
operation of the AEM. You’ll soon learn which parameters you a given beam current Ip and an acquisition time τ can be ex-
should select carefully and which you can generally leave to pressed as
take care of themselves. Let’s now explore further details of the
si Qi !aCN De
spectrum simulation. IC = t (17.11)
A
 478 17 Practical Aspects and Advanced Applications of XEDS
Part I
Chapter 17

Fig. 17.11 Example of simulated X-ray spectra from the NIST SRM2063a thin specimen. a Noise-free ideal spectrum. b Spectrum with random noise
 17.2 X-ray Spectrum Simulation – a Tutorial and Applications of DTSA 479

Part I
Fig. 17.12 Simulated result for a generated X-ray spectrum from the SRM2063a
thin specimen

where Qi is the ionization cross-section, si is the scaling factor

Fig. 17.13 Comparison between generated (black) and emitted (red) X-ray
of Qi, ω is the fluorescence yield, a is the line weight (or the
spectra from the SRM2063a thin specimen (a) and the X-ray signal loss in % due
relative intensity ratio), C is the weight fraction, A is the atomic
to X-ray absorption within the specimen, plotted against the X-ray energy (b)
weight, Nv is the Avogadro number, and De is the total electron
dose defined as
De = Ne Ip  (17.12)
Fig. 17.11. Note that this generated spectrum is never seen in
where Ne is the number of electrons (electron counts) in a unit practice since the X-rays are not yet emitted from the specimen.
electric charge. In addition, the characteristic X-ray peaks are much sharper than
a single channel width (10 eV/channel in this simulation). The
Continuum X-ray Generation: The continuum background actual generated peak width is narrower than 1 eV. You should
of an X-ray spectrum can ideally be generated only by note that intensities of all simulated spectra are normalized with
bremsstrahlung events. The intensity of the continuum X-ray the detector collection angle to make it easy to compare the spec-
Ib can be expressed similarly to Eq. 17.11 as a function of the tra.
X-ray energy E
X-ray Emission after Being Absorbed in a Thin Specimen: Not
X sC QC;j .E/Cj N De all the generated X-rays are emitted from your specimen
Ib .E/ = t (17.13) since some fraction of the generated X-ray is absorbed, even
Aj
j
in a thin specimen. We can express the relationship between
the generated and emitted X-ray intensities in a differential
where Qc,j(E) is the continuum X-ray cross-section at an X-ray form:
energy E, sc is the scaling factor of Qc,j, and j is an index for each
element included in the thin specimen. So, the continuum X-ray Iemit = Igen exp Œ−.=/z cosec ˛ (17.14)
intensity is the sum of contributions from all the elements. In
general, Qc,j is based on the classic description by Kramers as
described in Eq.(35.7) in W&C. Especially for a thin specimen where ∆Iemit and ∆Igen are the emitted and generated X-ray inten-

Chapter 17
in an AEM, the angular distribution of continuum X-rays is not sities from a small mass-thickness segment ∆(ρt) at α depth of ρz
homogeneous, as was shown in Fig. 33.6 in W&C. We can han- from the incident surface, respectively, and α is the X-ray take-
dle this angular dependence of the continuum X-ray generation off angle. In DTSA, the total emitted X-ray intensity is calculated
by including the detector elevation angle, and the specimen tilt by integrating all the X-ray intensities, including absorption, in
angles in the proposed cross-section models of continuum X-ray the individual segments. We compare an emitted spectrum from
generation in thin specimens. SRM 2063a (red solid line) with a generated one in Fig. 17.13a.
Fig. 17.13b, the fraction of X-ray absorption is plotted in percent
The generated characteristic X-ray peaks are merged with the as a function of the X-ray energy. Despite the very thin SRM
generated continuum X-ray spectrum in the first step of the spec- 2063a specimen (only 76 nm), you can see that a considerable
trum simulation. Figure 17.12 shows an example of a generated fraction of the X-rays generated at energies lower than 1 keV
spectrum from SRM2063a (thickness of 76 nm), simulated in (soft X-rays) is absorbed into the specimen and not emitted. For
the same microscope and for the experimental conditions in instance, more than 30 % of the generated O Kα is not emitted
 480 17 Practical Aspects and Advanced Applications of XEDS
Part I

Fig. 17.14 Comparison between emitted (red) and detected (blue) X-ray spec- Fig. 17.15 Comparison of X-ray spectra from the SRM2063a thin specimen
tra from the SRM2063a thin specimen (a) and the X-ray signal loss in percent before (blue) and after (orange) broadening due to the detector electronics
due to absorption in the detector, plotted against the X-ray energy (b) (a) and the detector energy-resolution plotted against the X-ray energy (b)

from the film due to absorption. In contrast, X-ray absorption is In Fig. 17.14a, we compare the detected spectrum (blue solid
negligible for X-rays > 2 keV. line) with the emitted spectrum. As we mentioned above, the
intensities of both spectra were normalized with the collection
Influence of the Detector Collection Angle: X-rays are emitted angle Ω/(4π). In Fig. 17.14b, detection losses due to window
radially into a 4π steradian, spherical space. However, only a absorption and crystal transmission are plotted (in percent)
small portion of the X-rays can be collected, depending on the as a function of the X-ray energy. Again, you can see that the
XEDS detector geometry, defined by the collection solid angle Ω. soft X-rays below 2 keV are reduced significantly for a Si(Li) de-
As we described earlier, only 0.1 ~ 3 % of all the emitted X-rays tector (even with an atmospheric ultra-thin window (UTW)). The
are collected in commercial AEMs unless you use the latest O Kα intensity drops to only ~ 25 % of the generated intensity
SDDs with large solid angles. The X-ray intensity collected by due to absorption in the specimen and losses during detection.
the detector can be expressed as: If your detector window is contaminated with carbon and ice,
  further intensity loss will occur.
˝
Icol = Iemit (17.15)
4 Influence of Detector Electronics: The peaks of detected
characteristic X-rays are still very sharp in the detected spectrum
As we also mentioned above, the background intensity is depen- (Fig. 17.14). However, these sharp peaks are blurred by the X-ray
dent on the take-off angle of the detector, which is included in detection process and the detector electronics. This blurring is
the cross-section models of the continuum X-rays. proportional to the X-ray energy and expressed as

X-ray Absorption and Transmission in the XEDS Detector: The

p
RFWHM .E/ = 2:5.E − EMnK ˛ / + RFWHM .MnK ˛/2 (17.17)
collected X-rays reach the active area of detector crystal
Chapter 17

through the detector window. In this process, the low-energy

soft X-rays may be absorbed by your detector window and by any where EMnKα and RFWHM(MnKα) are the energy and detector reso-
contamination layers in front of the window. In addition, higher- lution at the Mn Kα peak. The spectrum blurred by the detector
energy, hard X-rays may transmit through the detector crystal, electronics (orange solid line) is superimposed on the detected
especially if it is a Si(Li) crystal or an SDD crystal. We describe spectrum at the top in Fig. 17.15. Fig. 17.15b, the energy res-
the loss of X-ray intensity due to the window absorption and olution RFWHM(E), calculated using Eq. 17.17, is plotted against
crystal transmission in terms of the detector efficiency ε, which the X-ray energy. Although the characteristic peak intensities
is given as Eq. 35.15 in W&C. We can express the detected X-ray remain the same, you can see that the height of each peak (hence
intensity as: the peak-to-background ratio) is significantly reduced due to the
blurring. With such blurring, DTSA finally produces a simu-
Idet = Icol " (17.16) lated X-ray spectrum which has the familiar shape we usually
see on our computer display.
 17.2 X-ray Spectrum Simulation – a Tutorial and Applications of DTSA 481

Fig. 17.16 Screen shot of the dialog

for output choice

Part I
Using DTSA, you can simulate an X-ray spectrum at each detec- major question is “Is it possible to identify the existence of a S Kα
tion step. In the main dialog for thin-specimen spectrum gener- peak in the presence of Mo?”. The answer is “yes” – if a sufficient
ation (Fig. 17.8), click the ‘Output Choice’ button and then you amount of S is contained in the analyzed region. You can confirm
can select appropriate indices for desired outputs in the output the S presence by comparing (1) the Mo K/L intensity ratio from
choice dialog (Fig. 17.16). a pure Mo standard thin specimen and (2) the energy resolution
of the peak at 2.3 keV. In contrast, if you try to identify S with a
low amount in the presence of Mo, the situation becomes much
harder. (Look at the chapter on spectral imaging.)
17.2.4 Application 1: Confirmation of Peak
Overlap Overlapping Peaks

Once you understand the X-ray simulation process in DTSA, If there are several characteristic peaks at very similar
you can apply the spectrum simulation feature for your specific energy ranges, the peaks might overlap one another and
research needs. You should perform the spectrum simulation be- it is important to know this before you carry out your
fore doing the experiment but after reading this chapter and those experiments: DTSA can help. 9
in W&C! (Then read the chapters again and repeat the process!)
In the rest of this section, we will introduce several useful appli-
cations of spectrum simulation. If your specimens are not self-supported (such as nanoparticles
or FIB lift-out coupons) you must use supporting grids. Selec-
The first application you should always carry out is the confirma- tion of the supporting grids is another important task prior to

Chapter 17
tion of the energy positions of all the characteristic X-ray peaks any practical experiments. Again, you can evaluate suitable
from your specimen. As described in Sect. 34.3 of W&C, you can supporting grids with the aid of a spectrum simulation. For ex-
confirm any peak overlaps simply by the spectrum simulation in ample, a simulated spectrum from HfO2 is shown in Fig. 17.18.
DTSA. Figure 17.17 shows a simulated X-ray spectrum with a HfO2 is used in optical coatings and in DRAM capacitors as a
selected energy range of 1–4 keV from MoS2, which is one of high-k dielectric. It has recently attracted attention as a good
the famous ‘pathological overlaps’. To confirm the peak energy candidate for the gate oxide in field-effect transistors to replace
positions, KLM markers of the Mo L and S K line families are current silicon oxides. Therefore, HfO2 is often analyzed in
also shown in Fig. 17.17. You can see that the Mo Lα peak is a TEM, either on its own and as one of the components in a
superimposed on the S Kα peak. In addition, the S Kβ peak lies semiconductor device. In Fig. 17.18, the KLM markers of Cu
between the minor Mo Lβ4 and β3 peaks as well. You can easily and Mo (both are common support-grid materials) are also
confirm whether the peak at 2.3 keV is the Mo L or S Kα by shown. Both the Cu Kα and Kβ lines overlap with the Hf L
checking the presence of the Mo Kα line at 18 keV. However, the peaks. Obviously, you can see that the (more popular) Cu grids
 482 17 Practical Aspects and Advanced Applications of XEDS
Part I

Fig. 17.18 Simulated X-ray spectrum from HfO2 with KLM markers for Cu and
Fig. 17.17 Simulated X-ray spectrum with a selected energy range of 1–4 keV
Mo. Since the Hf L-line family is superimposed on the Cu L-line family, a Cu grid
from MoS2, showing pathological overlaps between the Si Kα (red) and Mo Lα
is not advisable for X-ray analysis of materials containing Hf
(blue) peaks

should not be your primary choice for XEDS analysis of Hf. In Table 17.4, the MACs of the Ni Kα, Ni Lα and Al Kα lines
Instead, you can use a Mo grid to mount a FIB-cut specimen in Ni, Al and NiAl are summarized. The red dashed line in
from a semiconductor device which contains HfO2. In addition Fig. 17.19b indicates a 10 % loss of the total intensity due to
to the grid materials, you may check other X-ray peak overlaps absorption in the specimen. When more than 10 % of absorp-
with Ga and also with protective deposition materials such as tion occurs, the absorption correction may be required for good
W or Pt. quantitative analysis. We define the specimen thickness which
causes 10 % absorption loss as the critical thickness. As shown
in Fig. 17.19b, the critical thicknesses for the Al Kα and Ni Lα
are 54 and 97 nm, respectively, again confirming the dangers of
17.2.5 Application 2: Evaluation of X-ray absorption from lighter elementss, even in thin films.
Absorption into a Thin Specimen
The absorption path length is dependent on the X-ray take-off
angle α and given as tcosecα for a specimen thickness t. How-
We have stressed that X-ray absorption, even in the thinnest ever, the path length varies with the specimen tilt-angle. You can
specimens, may not be negligible for lower energy X-rays. As reduce the absorption path if you tilt your the specimen toward
we explained in Eq. 17.14, how X-ray absorption affects your the XEDS detector, or increase it by tilting away. If we define
spectrum depends on the thickness, density, and mass absorption the tilt angle toward the detector as θ, we can express the total
coefficient of the specimen and the X-ray take-off angle. You can take-off angle as α + θ. Fig. 17.20 shows X-ray spectra from a
evaluate the expected X-ray absorption of your specimen using 100 nm thick NiAl film simulated under conditions of no-tilt,
the spectrum simulation in DTSA before you start your actual +15º tilt (toward the detector, the total take-off angle: 40º) and
experiments. −15º tilt (away from the detector, the total take-off angle: 10º).
Although the peak intensity of Ni Kα is almost unchanged, the
Figure 17.19a shows three spectra simulated at different thick- peaks of both the Ni Lα and Al Kα vary significantly with the
nesses of 50, 100, and 300 nm of NiAl intermetallic compound tilt conditions. Therefore, you can control the absorption path
in a 200-kV-instrument with an X-ray take-off angle of 25º. To length, i.e., the absorption effect, to some degree, by tilting your
ensure a fair comparison, we have normalized these spectra to specimen. In addition, you should know the proper configura-
the maximum peak intensity of the Ni Kα line. As the specimen tions of the specimen-detector geometry in your instrument, if
Chapter 17

thickness increases, you can see that the peak heights of the Ni you still insist on tilting your specimen for analysis.
Lα and Al Kα lines are relatively reduced. The absorption loss
of the Ni Kα, Ni Lα, and Al Kα lines are shown as a function Be Aware
of the specimen thickness in Fig. 17.19b. For calculation of the
absorption loss, we used an X-ray take-off angle of 25º. We can
calculate the MAC of the individual X-ray lines in the specimen You decrease the P/B ratio as you tilt the specimen. 9
by the weighted mean of the individual MAC values of the con-
stituent elements:
X
.=/Xsp = Cj .=/jX
j
(17.18)
 17.2 X-ray Spectrum Simulation – a Tutorial and Applications of DTSA 483

Table 17.4 Summary of MAC values of the Ni Kα, Ni Lα and Al Kα lines in Ni, Al
and NiAl intermetallic compound. The MAC values in the pure elements are cal-
culated from Heinrich’s formula (Heinrich 1987) and those in NiAl are obtained

Part I
from Eq. 17.18 using the values in the pure elements

Absorber MAC (m2/kg)

Ni Kα Ni Lα Al Kα

Ni 6.0 181.1 454.3

Al 6.1 179.8 39.7

NiAl 6.0 180.8 323.8

Fig. 17.19 a Comparison of three X-ray spectra simulated at different thick-

nesses of 50, 100, and 300 nm of Ni3Al intermetallic compound in a 200 kV in-
strument with an X-ray take-off angle of 25º. These spectra were normalized by Fig. 17.20 Comparison of three X-ray spectra of NiAl simulated at different tilt
the maximum peak intensity of the Ni Kα line. b The signal losses (%) of the angles of −15º, 0º, and +15º
Ni Kα, Ni Lα and Al Kα lines due to the absorption into the specimen, plotted
against the specimen thickness
of P/B of Ni Kα peak using the Fiori definition. The P/B value is
much lower in the exit-surface side, and hence the higher take-
17.2.6 Application 3: Evaluation of the AEM- off angle is desirable since we want to obtain higher P/B values.
XEDS Interface It’s impractical to change the detector geometry in your micro-
scope instrument (unless you have more money than sense). In
contrast, you can easily vary the take-off angle in a DTSA simu-
Using DTSA, you can also simulate an X-ray spectrum in un- lation. Thus spectrum simulation is a very handy tool to support
usual conditions – sometimes these conditions may be unreal- all your XEDS experiments.
istic. In Sect. 33.3 of W&C, we discuss why an XEDS detector
is mounted on the incident-beam side of the specimen, despite
the fact that characteristic X-rays are emitted uniformly over
a 4π steradian sphere. The major reason for this choice is the 17.2.7 Application 4: Estimation
angular dependence of the continuum X-ray emission: the back- of the Detectability Limits
ground-intensity distribution is maximized in the forward scat-

Chapter 17
tering direction and is reduced as the scattering angle increases
toward backscattering directions. Using the DTSA spectrum sim- One of the most useful applications of the spectrum simulation
ulation, you can confirm this angular dependence of the back- is to evaluate detectability limits. This is why this particular ap-
ground intensity distribution. Figure 17.21a shows a comparison plication was the major motivation for the original development
of two spectra from our standard NiOx test specimen, simulated of DTSA (Fiori, Swyt 1989; Newbury et al. 1995). A commer-
at 200 kV with take-off angles of ± 70º. Although the peak in- cialized software package (EELS Advisor for Gatan DigitalMi-
tensity of the Ni Kα line remains almost the same, you can see crograph) expands the concept of detectability evaluations to
significant difference in the background intensity. As expected, find experimental configurations for appropriate acquisition of
the background intensity is much higher on the exit-surface side EELS spectra and EFTEM elemental images. You can find the
of the specimen, especially at lower energies. detailed procedures for detectability evaluation in the literature
(e.g., Watanabe and Williams 1999). Here, we will show how
Remember, the absorption-path length is the same in both geom- you can evaluate three different parameters related to the ana-
etries. In Fig. 17.21b, you can see the take-off angle dependence lytical sensitivity, using the spectrum simulation.
 484 17 Practical Aspects and Advanced Applications of XEDS
Part I

Fig. 17.21 a A comparison of two spectra from a NiOx test specimen, simulated
at 200 kV for take-off angles of ± 70º. b The take-off angle dependence of P/B of
the Ni Kα peak (using the Fiori definition)

Look Before you Look!

It’s obviously crucial to find out if you can actually detect

what you are looking for, before you embark on specimen Fig. 17.22 a Comparison of two X-ray spectra from a 100-nm-thick Fe-0.15
wt%P simulated with and without noise in a 200 kV instrument for 100 s. Two
preparation and all the other time-consuming steps. 9
highlighted regions indicate the net peak (I) and background (b) intensities.
The peak visibility (b) and detectability (c) ratios of the P Kα line in Fe, plotted
against the P composition CP for 100-nm-thick films. The error bars represent the
Peak Detectability: After acquisition of an X-ray spectrum from 3σ (99 % confidence limit) range determined from 50 simulated spectra with
your specimen, you may see that an element of interest only different random noise conditions
exhibits a small peak. In this situation, you must judge whether
the small peak is actually visible or even detectable (is it just
Chapter 17

a fluctuation in the background noise?). Detailed descriptions intensities are measured in the same channel width, as shown in
of when a peak is visible or not can be found in Sect. 34.5 of the shadowed areas in Fig. 17.22a. Then, the net peak intensity
W&C. Here we’ll explore the peak visibility and detectability in (I) is determined by subtracting B from P. As we described p in
practice using the spectrum simulation. Sect. 34.5 of W&C, if the net peak intensity exceeds 3 B, the
peak is visible
p with a 99 % confidence limit (3σ). In addition,
As an example, let’s consider phosphorus detection in Fe. P is when I >3 2B (slightly stricter than the visibility limit), the peak
one of the impurities that can cause brittle failure in steels by is detectable
p as mentioned
p in Sect. 36.4B of W&C. Let’s denote
segregating to grain boundaries. Therefore, the detection of small I=3 B and I=.3 B/ aspthe visibility p and detectability ratios,
amounts of P (and other segregants) to GBs has long been one the respectively. So, if I=.3 B/ or I=.3 2B/ > 1 in terms of the P
major problems addressed by XEDS in the AEM. Figure 17.22a Kα peak, then the peak is visible and the phosphorus is detect-
shows a typical spectrum simulated for a 100-nm-thick foil of able in Fe, respectively. Figures 17.22b, c show the visibility
Fe-0.15wt%P in a 200-kV instrument for an acquisition time and detectability ratios of the P Kα line in Fe, plotted against the
of 100 s. From the spectrum, the peak (P) and background (B) P composition (CP) for 100-nm-thick films. As you can see, the
 17.2 X-ray Spectrum Simulation – a Tutorial and Applications of DTSA 485

P Kα peak is visible at CP = 0.11wt% and becomes detectable at

CP = 0.16wt%.

Part I
These simulated results are calculated assuming ideal condi-
tions and could be very different from realistic situations since
we have not included any noise associated with signal gener-
ation and detection. In DTSA, we can add such random noise
to simulated spectra, based on Poisson counting statistics. In
Fig. 17.22a, we show another simulated spectrum with random
noise as a red line. Obviously, it becomes harder to recognize
such a small peak under the noise, which is more similar to prac-
tical cases, and hence the peak visibility and detectability are
degraded when we add noise to the spectrum simulation. We Fig. 17.23 MMF of P in Fe, determined from simulated X-ray spectra of a thin
determined the 3σ range of the peak visibility and detectability foil of Fe-5 wt% P using the GRM equation, plotted against the specimen thick-
from 50 simulated spectra with different random noise condi- ness
tions for each P composition and these are shown as error bars
in Fig. 17.22b, c. The results from the noise-free spectra appear
close to the average of those with random noise. described in Sect. 36.4D of W&C, the MDA in AEMs is now
at or approaching the single atom level as we show below. This
You should note that the above definitions for the peak visibility value is better than any other analysis technique. (Atom-probe
and detectability are approximately equivalent to the decision microscopy is the only competition.)
(Lc) and detection (Ld) limits in Currie’s definitions for the an-
alytical sensitivity limits (described with Eqs. (36.16)–(36.20) To calculate the MDA values by translating from the MMF,
in Sect. 36.4C of W&C). Furthermore, the above visibility limit you need first to calculate the analyzed volume. To do this, you
also corresponds to the visibility of a 1D image (i.e., the spec- need to know your probe size and your specimen thickness. Fig-
trum!) in the Rose criterion. You should be able to work out how ure 17.24a shows the analyzed volume of Fe determined based
these definitions are related to one another in some degree. on the Gaussian broadening model (see next section) for LaB6-
AEM (probe size, d = 10 nm), FEG-AEM (d = 2 nm) and aberra-
Minimum Mass Fraction (MMF): The detectability limit we tion-corrected (AC) FEG-AEM (d = 0.2 nm), plotted against the
described above is essentially the same as the MMF, which specimen thickness. If you were to analyze a thinner region, the
is the minimum detectable composition. You can calculate incident probe size becomes the dominant factor rather than the
the MMF value from the peak and background intensities of beam broadening. This smaller analyzed volume is more signif-
a standard specimen with a known composition using the icant if the AC instrument is used. However, the benefit of using
Goldstein–Romig–Michael (GRM) equation (Eq. 35.15) first an AC instrument is obviously degraded when you are analyzing
discussed in Sect. 36.4B of W&C. The MMF values for P in specimens thicker than ~ 50 nm. So if you’ve paid for an expen-
Fe, determined from simulated X-ray spectra of Fe-5wt%P by sive AC AEM, don’t waste time putting thick specimens in it!
the GRM equation, are plotted against the specimen thickness
in Fig. 17.23. The simulation parameters are same as those in You can convert the analyzed volume to the total number of
Fig. 17.22 (accelerating voltage: 200 kV and acquisition: 100 s) atoms using basic crystallographic information. For example,
and the error bars are determined from 50 spectra simulated, with Fe has a bcc structure with 2 atoms in the unit cell (i.e., the
noise, at each thickness condition. As you can see, in comparison unit volume per atom can be given by (a3)/2, where a is the lat-
with the detectability evaluation shown in Fig. 17.22c, the error tice parameter). Then, by multiplying the atomic fraction of the
bars are much smaller. This is because the CP(MMF) values are MMF value (you have to convert the MMF value (in wt%) to
calculated using simulated spectra with the high P composition. the atomic fraction), the MDA can be determined. The MDA of

Chapter 17
The calculated CP(MMF) value decreases as the specimen P in Fe translated from the MMF in three different instruments
thickness increases simply because the X-ray signal increases. In is shown in Fig. 17.24b as a function of the specimen thickness.
the 100-nm-thick film, CP(MMF) approaches ~ 0.16 wt% under
these conditions. Therefore, the CP(MMF) value calculated using Single-atom AEM!
Eq.(36.15) agrees well with the detectability limit determined
from simulated spectra with trace amounts of elements.
We can identify single atoms! (See the paper by Lovejoy
et al. 2012) 9
Minimum Detectable Mass (MDM) or Minimum Detectable
Atoms (MDA): Another expression of the analytical sensitivity
is the MDM, which is more conveniently defined as the Despite having a poorer MMF (higher mass fraction) in thinner
minimum number of atoms detectable in the analysis volume regions, the MDA value is improved as the specimen thickness
in your thin specimen in the AEM, i.e., MDA. As we already decreases. According to these results, only ~ 20 P atoms are de-

Licensed to ([email protected])
 486 17 Practical Aspects and Advanced Applications of XEDS

worsened. Furthermore, the X-ray absorption correction for thin

specimens in AEM is harder than that for bulk samples in SEM/
EPMA because you need to know some essential prior infor-
Part I

mation about the specimen thickness and the density at every

individual analysis point.

Several methods are proposed for thickness determination in

Chap. 36 of W&C. However, these thickness measurements need
to be performed independently of the X-ray analyses and doing
this may introduce additional errors, which could also be an ad-
ditional limiting factor for quantitative X-ray analysis. So, the
conclusion we draw is that although applying the Cliff–Lorimer
ratio method is straightforward, it cannot achieve the necessary
levels of accuracy and precision in quantitative X-ray analysis
in many thin specimens. The ζ-factor method is a new and im-
proved quantitative procedure for thin specimens that overcomes
some of the limitations in thin-film X-ray analysis listed above
by making local thickness determination an integral part of the
process. In this section, we go into much greater depth than you
saw in Sect. 35.5 of W&C and explore further details of the
ζ-factor method.

17.3.1 Why Bother with Quantification?

Fig. 17.24 a Analyzed volume of Fe determined based on the Gaussian broad-
ening model for LaB6-AEM (d = 10 nm), FEG-AEM (d = 2 nm) and AC FEG-AEM Before explaining what quantitative X-ray analysis via the ζ-fac-
(d = 0.2 nm), plotted against the specimen thickness. b The MDA of P in Fe calcu- tor method is, it’s worth asking whether such an involved quan-
lated from the MMF in Fig. 17.23 with the analyzed volume size tification process is really required for materials characterization
or not. In some cases, you’ll find that elemental distributions
can be determined, or a phase can be identified, by qualitatively
tectable by using the FEG-AEM under these specific conditions. analyzing a single spectrum. A typical example is shown in
This value may be improved to a single-atom level in the AC Fig. 33.11 in W&C. With prior knowledge of preferential ele-
instrument. If you use longer acquisition times or the X-ray col- mental distributions, you can easily distinguish types of carbides
lection efficiency is improved with better detector geometries in steel. If you can identify the different carbides, you often don’t
(e.g., larger collection angles), it’s possible that the MDA value need further quantification.
might be further improved.
As another example, a set of X-ray intensity maps of Pb L
and O K lines taken from a triple point of a Pb-based oxide
Pb(Mg1/3Nb2/3)O3-35 mol%PbTiO3 is shown in Fig. 17.25a along
17.3 The ζ-factor Method: a New with an ADF-STEM image (Gorzkowski et al. 2004). From these
elemental maps, you might infer that both Pb and O are depleted
Approach for Quantitative X-ray along the boundaries constituting the triple point. Using these
Chapter 17

Analysis of Thin Specimens elemental maps, quantification was performed by the ζ-factor
method and a different set of composition maps of Pb and O
obtained, as we show in Fig. 17.25b. In addition, a thickness map
In Chap. 35 of W&C, different approaches to quantitative X-ray was obtained through the ζ-factor quantification process. As we
analysis are described for thin specimens in the AEM. Since the inferred from the elemental map, depletion of O is confirmed
development of the ratio method by Cliff and Lorimer (1995), at the boundaries and the triple point. However, quantification
this particular method has become the standard approach and shows that Pb is actually segregated to the triple point, which is
is widely used due to its simplicity. Unfortunately, even in contradictory to the original conclusion from the elemental map.
thin specimens, X-ray absorption may need to be corrected for
quantification and the simple Cliff–Lorimer approach is found According to the thickness map determined by the ζ-factor
wanting. As we’ve noted many times, X-ray absorption is more method, the thickness decreases along the boundaries and at the
serious in some materials than others, especially if lighter el- triple point, which reduces the X-ray intensity of the Pb L line.
ements are present, and then the accuracy of quantification is Therefore, the discrepancy between the qualitative elemental and
 17.3 The ζ-factor Method: a New Approach for Quantitative X-ray Analysis of Thin Specimens 487

Part I
Fig. 17.25 a Set of X-ray intensity maps of Pb L and O K lines taken from a triple point in a Pb-based oxide Pb(Mg1/3Nb2/3)O3-35 mol%PbTiO3 and an ADF-STEM
image. b A set of composition maps of Pb and O with a specimen thickness map quantified by the ζ-factor method (Gorzkowski et al. 2004)

quantitative compositional maps of Pb has arisen. The thickness 17.3.2 What Is the ζ-factor?

Chapter 17
reduction along the boundaries constituting the triple point was
caused by preferential thinning during specimen preparation by
FIB. From this example, you should conclude that qualitative In a thin-film specimen, the measured characteristic X-ray in-
analysis, or simply looking at X-ray intensity changes may result tensity I is proportional to the mass thickness ρt and the com-
in wrong conclusions. Quantification is an essential process for position (weight fraction) C unless the generated X-ray signals
materials characterization! You will be ready to use quantifica- are affected by absorption and fluorescence. We can establish
tion procedures via the ζ-factor once you understand and agree the following relationship between the mass thickness and the
with this example! measured X-ray intensity by normalizing with the total electron
dose De (i.e., the number of electrons hitting the analysis region
during X-ray acquisition)
IA
t = A
C A De (17.19)
 488 17 Practical Aspects and Advanced Applications of XEDS

where ζA is a proportionality factor to connect IA to ρt and CA. 17.3.3 Quantification Procedure in the ζ-factor
If X-ray absorption and fluorescence are negligible because the Method
specimen is thinner than the critical thickness, the generated
Part I

characteristic X-ray intensity can theoretically be described as

Fortunately, as we’ll show, the quantification procedure in the
QA !A aA
g
IA = N CA tDe (17.20) ζ-factor method is quite simple. Since a similar relationship to
AA Eq. 17.19 holds for element B independently, we can express CA,
CB, and ρt as follows, assuming CA + CB = 1 in a binary system
where Nv is Avogadro’s number, QA is the ionization cross-­
A IA B IB A IA + B IB
section, ωA is the fluorescence yield, aA is the relative transition CA = CB = t =
probability (i.e., the relative line weight) and AA is the atomic A IA + B IB A IA + B IB De
weight. You should note that Eq. 17.20 is essentially the same (17.25)
as Eq. 17.11, which we introduced in the previous section –
except for the scaling factor to adjust the absolute magnitude Therefore, you can determine CA, CB, and ρt simultaneously by
for ionization cross-section. By taking into account the X-ray measuring their X-ray intensities. This approach can easily be
collection efficiency terms such as the collection angle Ω rel- expanded to any multi-component system as long as the assump-
ative to the whole sphere (4π) and the detector efficiency εA, tion of ΣCi = 1 is reasonable. Once you know the ζ factors, it is
we can express the measured X-ray intensity IA as (see, e.g., straightforward to determine compositions.
Armigliato 1992)
 
˝ QA !A aA ˝ A Warning
g
IA = IA
4
"A = N
AA
CA tDe
4
"A (17.21)
You need to know the electron dose term De for accurate
By comparing Eq. 17.19 with Eq. 17.21, the ζ factor can be thickness determination: so beam current measurement
defined as is essential for quantification via ζ factors. 9

AA
A = (17.22)
N QA !A aA Œ˝=.4/ "A The need for this measurement is the major drawback to the
ζ-factor method because measurement of the beam current is
The ζ factor contains terms for both X-ray generation and detec- not always easy in TEM (unlike the SEM/EPMA). In some TEM
tion. In other words, the ζ factor is equivalent to the atomic-num- instruments, it is straightforward to measure the in-situ beam
ber correction term (Z) in the ZAF matrix correction procedure, current because there is an in-built Faraday cap (similar to SEM/
which is widely employed for quantitative X-ray analyses of EPMA). If you have an EELS detector on your AEM, you can
bulk samples in SEM/EPMA. also monitor the beam current by prior calibration of the detector
readout system. Fortunately, if you have a modern instrument
Since the same relationship described in Eq. 17.21 is true for any with a Schottky FEG source, the beam-current fluctuation over
other element (e.g., B), we can derive the following equation by typical time periods for analysis should be less. In this case, the
taking a ratio between elements A and B and rearranging: beam current does not need to be monitored frequently and can
be calibrated easily through the readout from the viewing screen
CA AA QB !B aB "B IA A IA IA
= = = kAB (17.23) etc. We should also note that in the ζ-factor method, the mass
CB AB QA !A aA "A IB B IB IB thickness can be determined as shown in Eq. 17.25. To determine
the absolute thickness, you need to know the specimen density at
Obviously, this relationship is exactly same as the definition of individual measured points. You can find out more about density
the Cliff–Lorimer ratio equation. Therefore, we find the relation- determination in the Appendix.
Chapter 17

ship between the k factor and ζ factors is:

Information about the mass thickness of your specimen at in-
dividual measured points is very useful when you need to cor-
A
kAB = (17.24)
B rect for X-ray absorption, even in thin specimens. The absorp-
tion-correction term for a single X-ray line from a thin specimen
Once you know one of the ζ factors (e.g., from a reference ele- is
ment such as Si and Fe), you can determine other ζ factors from
existing k-factor databases (e.g., Tables 35.1 and 35.2 in W&C). .=/A
sp t cosec ˛
You should be clear that the ζ-factor method does not eliminate
AA =
1 − expŒ−.=/A (17.26)
sp t cosec ˛
the widely used Cliff–Lorimer ratio method but expands the way
we do quantitative X-ray analysis as described in the following
sections. where .=/A sp is the mass absorption coefficient of the charac-
teristic X-ray line in the specimen and α is the X-ray take-off
 17.3 The ζ-factor Method: a New Approach for Quantitative X-ray Analysis of Thin Specimens 489

Part I
Fig. 17.27 ζ-factors of K lines determined in a 200-kV JEOL JEM-ARM200CF,
plotted against the X-ray energy. The open circles are the measured values from
the SRM2063a glass thin-film; the closed circles indicate the estimated values
by parameter optimization based on the measured values (Watanabe and Wade
2013)

Being able to use thin films of pure single-element materials

is a major advantage and overcomes any limitations associated
with thin-film standards since you can relatively easily fabricate
pure-element thin films by evaporation, electron deposition or
sputtering Such pure-element films are routinely available (in
contrast to the tremendous difficulties to prepare multi-element,
thin-film, standards with known compositions for k-factor deter-
mination). The only disadvantage you’ll find is that the thickness
Fig. 17.26 Flow chart of the quantification procedure for the ζ-factor method
with the X-ray absorption correction (Watanabe and Williams 2006)
of your standard film needs to be determined, prior to ζ-factor
determination. However, if you do a bit of research you’ll find
that independent thickness determination is relatively easy in
single-element, thin films.
angle. We incorporate this absorption-correction term into the ζ
factor by multiplying it by the corresponding X-ray intensities in The use of pure element thin standards expands the degrees of
Eq. 17.25, via an iteration process. A complete flow chart for the freedom for the ζ-factor determination. Unfortunately, you’ll
quantification procedure via the ζ-factor method is summarized still find it very tedious to determine the many ζ factors from
in Fig. 17.26. The calculation usually converges within five iter- standard thin specimens when you need to analyze multi-ele-
ation steps. We should also mention that the local mass-thickness ment materials. This situation can be improved when you use
value is useful for the fluorescence correction on the very rare the NIST-issued SRM 2063 or the re-issued 2063a glass thin-
occasions when it is necessary. In the Appendix, you’ll find the film standard. (Does your lab have such a sample?) Not only
various steps in the error calculation for the iteration process the compositions but also the thickness including the density
summarized. of the SRM2063/2063a are well characterized. By measuring a
single spectrum from the SRM2063/2063a, we can determine the
ζ factors of the five major components. Then, based on the five
ζ-factors determined from the SRM2063/2063a specimen, other

Chapter 17
17.3.4 Determination of ζ factors ζ factors can be estimated by fitting uncertainty parameters such
as absolute values of the ionization cross-section and window
thicknesses of the detector.
Similar to the Cliff–Lorimer k factors, the accuracy of your quan-
titative analysis is directly affected by the reliability of your ζ Figure 17.27 shows the ζ-factors measured from a single spec-
factors. You have to determine reliable ζ factors if you want reli- trum of the SRM2063a specimen (open circles) in a JEM-
able quantification. For the experimental determination of k fac- ARM200CF aberration-corrected STEM at 200 kV with a large
tors, you should know that standard thin specimens with known collection-angle XEDS detector, plotted against X-ray energy.
composition (such as alloys and compounds) are required. Such The closed circles in Fig. 17.27 indicate a series of ζ factors
standard thin specimens are also required for ζ-factor determi- estimated from the measured values by the parameter optimiza-
nation. However, unlike the k-factor determination, you can use tion. Although the NIST SRM2063a specimen is not available
pure element thin-films as standards for ζ-factor determination currently, similar types of materials can be utilized as standards
as might be expected from Eq. 17.19. for ζ-factor determination.
 490 17 Practical Aspects and Advanced Applications of XEDS

Table 17.5   Comparison of measured compositions of Pb-based oxide with

and without X-ray absorption correction to nominal values (Gorzkowski et al.
2004)
Part I

Measured Measured (un- Nominal

(corrected) corrected)

Mg 1.6 1.2 1.7

Nb 12.7 12.8 12.7

Pb 65.4 75.6 65.3

O 15.1 5.5 15.1

Ti 5.2 4.9 5.2

Fig. 17.28 Comparison of the absorption loss of the oxygen Kα intensity in Total 100.0 100.0 100
various oxides, plotted against the specimen thickness. Dashed and dot-dashed
lines represent 5 % and 10 % losses, respectively

Table 17.5 compares experimentally determined compositions

17.3.5 Applications of ζ-factor Method with the theoretical values of the Pb-based oxide (this is the
same specimen as shown in Fig. 17.26). As shown in Fig. 17.28,
the quantified oxygen composition is ~ 1/3 unless we apply the
As we noted, there are several advantages to the ζ-factor over the absorption correction to the oxygen Kα. In the ζ-factor method,
Cliff–Lorimer ratio method. In this section, we’ll highlight some the X-ray absorption correction is incorporated as we described
of these advantages, which are mainly due to the availability of above and the ζ-factor method has already been shown to work
simultaneous thickness determination. well for several light-element (oxide) quantifications, even for
Light Element Analysis: Quantitative analysis of the many these difficult Pb-based oxides.
important materials systems that include light elements
(e.g., the commercially important oxide, carbide and nitride Extraction of Thickness-Related Information: Knowing the
systems) is always challenging. You know that X-ray emission local composition and thickness via the ζ-factor method allows
from light elements below Si (Z = 14) in the periodic table is you to extract further information related to your specimen
generally very low due to their low fluorescence yields. In thickness, such as the beam broadening and the spatial resolution,
addition, as we’ve now described several times, these light as described in Sect. 36.1 in W&C. For example, the spatial
elements only produce soft X-rays (with energies below 2 keV) resolution R can be estimated using the equation proposed by
which are severely absorbed not only in the detector window Van Cappellen and Schmitz (1992), derived from the Gaussian
but also in the specimen itself. The first effect reduces the beam-broadening model
detector efficiency and the second reduces the accuracy of our
quantifications. Obviously this latter effect is more serious and 1=2
R = q  2 + ˇ.t/3 =2 (17.28)

has to be corrected! Figure 17.28 compares the absorption loss
of the oxygen Kα intensity in various oxides, plotted against
the specimen thickness. We calculate this absorption loss for an where the scaling parameters q and κ are dependent on the cho-
XEDS detector with the take-off angle of 20o using the following sen fraction of the incident intensity, e.g., q = 4.29 and κ = 0.68
equation for the spatial resolution which contains 90 % of the incident
intensity, which corresponds to the beam diameter at a thickness
of 0.68 t from the top surface. The terms σ and β are associated
h i
Chapter 17

1 − exp .=/X
sp t cosec ˛
LX (17.27) with the incident-beam size and the beam broadening, respec-
A =1−
.=/X
sp t cosec ˛ tively, and are given by
 2  
4ZN 
 = dTM =4:29; ˇ = 500 (17.29)
The dashed and dot-dashed lines in Fig. 17.28 indicate the 5 and E0 MN
10 % absorption limits respectively. If the X-ray absorption ex-
ceeds 10 %, the accuracy of quantification degrades correspond-
ingly. If absorption reduces the accuracy that we reasonably where dTM is the incident beam diameter at full-width-tenth-max-
expect from the quantification procedures, we need to correct imum (FWTM), E0 is the incident beam energy (in eV), ZN and M N
for it. The oxygen Kα line is more severely absorbed in oxides are the averaged atomic number and atomic weight, respectively.
with heavier elements. For example, 10 % of the X-ray intensity Since we can determine all the terms related to the specimen by
is absorbed at ~ 20 nm in Pb2O3. It is rare that we can expect our the ζ-factor method, we can easily extract R at individual mea-
specimens to be so thin. sured points if dTM and E0 are known. You should note that the
 17.3 The ζ-factor Method: a New Approach for Quantitative X-ray Analysis of Thin Specimens 491

Gaussian beam-broadening model is the best description of the

interaction volume, as was determined by careful evaluation of
experimental and simulated concentration profiles. Figure 17.29

Part I
shows a set of maps of impurity segregation in a low-alloy steel.
Map (a) is an ADF-STEM image in a vicinity of a grain bound-
ary. Composition maps of (b) Ni and (c) Mo were determined
by the ζ-factor method (this particular set of maps is also shown
in Fig. 35.11 in W&C), together with (d) thickness and (g) R
maps. The R map was determined using Eq. 17.28 and shows
that the R value around the GB is ~ 4 nm, which is consistent
with the experimentally determined spatial resolution from the
concentration profiles across the GB.

The knowledge of local compositions and thickness determined

by the ζ-factor method allows us to deconvolute any thickness
contribution (i.e., the beam broadening) from the compositions
we measure. You’ll find that such deconvolution of the thickness
contribution can be essential for quantification of fine features
embedded in a matrix or on support materials. For example, in
quantifying impurity segregation levels at an interface or a grain
boundary, the impurity composition you determine may vary
depending on the incident probe size, the specimen thickness, the
accelerating voltage, etc. The boundary enrichment of impurity
element A, which is defined as a number of segregant atoms per
unit area Γex, can be determined as (see Alber et al. 1997).

Stating the Obvious – Don’t Ignore It

If your results vary with the experimental conditions,

then the composition measurements can’t be quantitative.
 9

CA AB V
Aex + Abk = Nbk (17.30)
CB AA A

where Γbk is the number of solute atoms per unit area in the bulk,
Nbk is the number of atoms per unit volume in the surrounding
bulk region, and AA and AB are the atomic weights for A and B. In
addition, V and A are the interaction volume and the area of the
boundary inside the interaction volume, respectively. Obviously,
the major requirement to determine Γex is knowledge of V and
A, which are dependent on the specimen thickness. Since the

Chapter 17
diameter of the interaction volume in the Gaussian beam-broad-
ening model is given as q(σ2 + βz3 / 2)1/2 at the depth z from the
beam-entrance surface (see Van Cappellen and Schmitz 1992),
both V and A can be given as
Zt
2 q2
 q. 2 + ˇz3 =2/1=2 =2 dz = .8 2 t + ˇt4 /

V=
32
0 Fig. 17.29 Summary of an application of the ζ-factor method to quantification
(17.31) of boundary segregation in a low-alloy steel. a ADF-STEM image in the vicinity of
Zt a grain boundary, b Ni composition, c Mo composition, d thickness, e Ni bound-
 2 ary excess, f Mo boundary excess, and g R maps
q. + ˇz3 =2/1=2 dz

A= (17.32)
0
 492 17 Practical Aspects and Advanced Applications of XEDS

In Fig. 17.29, maps of the boundary enrichment Γex for Ni (e) and
Mo (f) determined using Eq. 17.30 are also shown. The average
Γex values for Ni and Mo determined from the maps are 3.5 ± 0.4
Part I

and 4.8 ± 0.1 atoms/nm2, respectively. Since a single monolayer is

~ 19 atoms/nm2 in the closed packed (110) plane for bcc Fe, the
enrichments of Ni and Mo can be estimated as less than ~ 25 %.
Since the thickness contribution is deconvoluted in the form of the
boundary enrichment, the composition can now be transformed to a
truly quantitative measure. This thickness deconvolution approach
is also applicable when you need to extract the true composition of
a precipitate/particle embedded in matrix/support materials.

Evaluation of Analytical Sensitivity: In Sect. 36.2 in W&C,

determination of the MMF for the ratio method is summarized.
By spectrum simulation in DTSA, we can predict the analytical Fig. 17.30 Detector efficiency curve for a JEOL JEM2-ARM200CF with a JEOL
sensitivity as we just described. The MMF value can also be Centurio X-ray detector (Watanabe and Wade 2013)
determined in the ζ-factor method. The criterion for the minimum
detectable peak-intensity in an X-ray spectrum is defined as mation. Using the thickness values of the detector window, we can
(Romig, Goldstein 1979) estimate the detector efficiency. Figure 17.30 shows the detector
p efficiency of the large-angle SDD calculated from the determined
I > 3 2B (17.33) thickness values of detector window materials, plotted as a func­
tion of the X-ray energy. Note that the decrease of the detector
where I is the X-ray peak intensity above the background and efficiency after ~ 10 keV is due to the thinner crystal in the SDD.
B is the background intensity. We can incorporate this criterion
into the ζ-factor method to describe the MMF So in summary, there are several clear advantages to the ζ-factor
p method over the conventional Cliff–Lorimer ratio method; not
A 3 2BA only for quantification but also for more detailed analysis of
MMFA = (17.34)
De t specimens. In addition, you can more easily estimate the XEDS
detector parameters and compare the performances of individual
In addition to the MMF, the other descriptions of analytical sen- instruments as well.
sitivity, i.e., MDM or MDA can also be derived by the ζ-factor
method. As we described in the previous section, information
on the analyzed volume is required if you wish to determine the
MDA. Since the analyzed volume is given in Eq. 17.31, you can 17.4 Contemporary Aplications
determine the total number of atoms in the analyzed volume as
N V=AN (where AN is the averaged atomic weight). Therefore, of X-ray Analysis
you can estimate the MDA for individual elements by multiply-
ing the total number of atoms in the analyzed volume with MMF In recent years the development of aberration-correction technology
values in units of atomic fraction. in both TEM and STEM has brought a paradigm shift in materials
characterization. Just in terms of image resolution, sub-Å resolu-
Determination of Detector Parameters: As with the Cliff– tion is now routinely available (to anyone with enough money) and
Lorimer k factor, the ζ factor also depends on the X-ray detection currently the latest aberration-corrected microscopes offer 0.5 Å
efficiency. Since we define the ζ-factor for an individual element, resolution both in STEM and TEM modes (Erni et al. 2009; Sawada
the X-ray collection efficiency (controlled by the detector et al. 2009). Especially in STEM, we now use an incident probe
Chapter 17

collection angle in different instruments) can be compared with refined by aberration correction to improve the spatial resolution
the ζ-factor values. As shown in Fig. 17.27, the ζ-factor values of analysis by both EELS and XEDS.
in the JEOL ARM are ~ 1/3 of those in the VG HB603. We can
interpret this to show that the analytical sensitivity in the ARM More importantly, we find that the analytical sensitivity can also
is ~ three times higher than that in the HB603. From the reported be improved because more current can be included in the refined
value of the X-ray collection angle in the HB603, we can estimate probe while maintaining fine probe dimensions. Since these ab-
the solid angle of the large-angle SDD in the ARM as ~ 0.53 sr erration-correction STEM instruments appeared, X-ray analysis
by taking into account the difference in the ionization cross- is now seeing a resurgence of interest bolstered by the simple
section at different kV. (note: by further optimization of the SDD nature and robustness of the quantification procedures which we
position, the solid-angle increases as high as 0.64 sr in the ARM) have just described in great detail.

Furthermore, as mentioned above, we can determine the thickness Both the hardware and software related to X-ray analysis, espe-
values of the detector window materials during the ζ-factor esti- cially in aberration-corrected STEM, continue to improve. In this
 17.4 Contemporary Aplications of X-ray Analysis 493

Part I
Fig. 17.31 Comparison of input/output count rate of a large solid-angle SDD at
three different time constants on an aberration-corrected STEM JEM-ARM200CF

section, we’ll introduce you to the latest instrumentation related

to X-ray analysis in STEM. Then, we’ll give you two examples
of more recent applications of X-ray analysis.

17.4.1 Renaissance of X-ray Analysis

Aberration-corrected STEM instruments have triggered further

interest in X-ray analysis as we just above. However, in com-
parison to EELS, the collectable fraction of the generated X-ray
signals remains extremely limited (at most ~ a few percent). This
miserable collection efficiency is still the major limitation of
X-ray analysis, even in the aberration-corrected instruments. To
some extent we can mitigate this drawback by using a silicon drift
detector (SDD). The latest SDD technologies are relatively flexi-
ble and so we can fabricate more complex detector arrangements.
Several new detector geometries have been proposed. For exam-
ple, an annular detector geometry (similar to a BSE detector in
Fig. 17.32 a HAADF-STEM image of TiO2-supported Au nanoparticles and X-ray
SEM) has been proposed by Kotula et al. (2009). Zaluzec (2009)
maps of Au (left) and RGB color overlay with O, Au and Ti (right), acquired using
has proposed a post-specimen geometry for X-ray collection in an JEOL JEM-ARM200F with the large collection angle SDD system: total acqui-
the forward-scattering direction of electrons, which allows the sition times are (b) 1 min and (c) 6 min
collection angle of π sr out of the 4π sr sphere (more than 10 times
higher than the conventional detector geometry in AEM!).

Chapter 17
As we show in Fig. 17.31, the high throughput capability of technologies. If we acquire these maps for a bit longer (still only
the SDD can handle high count rates (up to ~ 100 kcps) taking 6 min!) we can easily see the distribution of Au nano particles in
advantage of the new geometries. Furthermore, if we integrate Fig. 17.32c. In the following sections, we summarize two novel
the new detector configurations into aberration-corrected in- applications of X-ray analysis, which again can only benefit from
struments, e.g., the FEI Chemi/STEM technology and the JEOL aberration correction and SDD technology.
large-angle SDD system, we get detector solid-angles up to ~ 1 sr
(3–10 times higher than conventional systems). Figure 17.32
shows two sets of X-ray maps acquired from TiO2-supported Au
nano particles in a JEOL JEM-ARM200F aberration-corrected
STEM with the large collection-angle SDD system. The Au map
and color overlay map (B) were acquired for 1 min in total. While
these maps are still noisy, obtaining such X-ray maps with such
very short acquisition times is an amazing advance over prior
 494 17 Practical Aspects and Advanced Applications of XEDS
Part I

Fig. 17.33 Tilt series of Ti Kα maps around a W contact-plug in a pillar-shaped specimen at different tilt angles between 0° and 170° with 10° steps. Signals in these
elemental maps were enhanced by applying the MSA reconstruction to the original X-ray SIs

butions in the z direction through your specimen. Fortunately

Beginning – A Big Change
both EELS and XEDS can give you 3D elemental distributions.
However, XEDS-based tomography has seen little usage be-
We can maximize the X-ray collection efficiency by using cause of the poor efficiencies of X-ray generation and detection
arrays of SDDs (FEI, Harrach et al. 2009) by using an (again!). In addition, if you use a conventional TEM holder for
SDD with a very large active area (JEOL, Ohnishi et al. tomographic XEDS analysis, the path length of the X-rays and
2011) 9 hence the X-ray absorption change as you change the tilting an-
gle (Möbus et al. 2003). Once again, at larger tilt angles, X-ray
absorption in your specimen becomes the limiting factor.

17.4.2 XEDS Tomography for 3D Elemental We can solve the problem of differences in X-ray absorption at
Distribution different tilt angles by using a pillar-shape specimen with a spe-
cially designed specimen holder. Figure 17.33 shows a tilt series
of Ti Kα maps acquired from a pillar-shaped Si device around
In TEM, we usually view the projected image of a thin speci- a W contact plug, using a Hitachi HD-2300 dedicated STEM.
Chapter 17

men, which means that information in the optic-axis direction The tilt-series of X-ray SIs were acquired over a tilt range from
(the z direction) is not obtainable unless we tilt the specimen. 0 to 175o with five steps. To compensate for the weak X-ray sig­
If we want to observe the microstructure through the depth of nals, multivariate statistical analysis (MSA) was applied. (See
the specimen, we have to use either stereo microscopy (see Chap. 16 for much more on MSA.) The Ti distributions show the
Sect. 29.1 in W&C) or electron tomography (introduced for metallization layer around the W plug. You can see this layer more
EELS in Sect. 40.9 in W&C). This latter approach requires that clearly in the reconstructed Ti distribution shown in Fig. 17.34.
multiple images be recorded at various tilt angles to give you the The complex Ti distribution, especially at the bottom part of the
most accurate 3D reconstruction. Much more information can be pillar, is barely observable in a single X-ray map. So in a con-
found in Chap. 12 on electron tomography. ventional XEDS configuration, weak signal enhancement, e.g.,
by MSA, is essential if you want to see an XEDS tomographic
Although you can image the 3D microstructure by electron to- image. Not surprisingly, XEDS tomography is much more suited
mography, it would be still be difficult to infer elemental distri- to an aberration-corrected STEM with an SDD X-ray system.
 17.4 Contemporary Aplications of X-ray Analysis 495

atomic resolution is still a challenge for X-ray analysis because

of its inherently poorer signal collection efficiency (~ 100 times
worse than EELS). Therefore, if we want to approach atom-

Part I
ic-level resolution X-ray analysis, either we need much greater
beam currents (which may induce radiation damage and de-
grade spatial resolution) or we need to collect our signals for
much longer acquisition time (which means more spatial drift).
Fortunately, in the best instruments today, the mechanical and
electrical stabilities have been significantly improved and
large-angle SDDs are routine. In fact, as shown in Fig. 35.11
in W&C, the spatial resolution of X-ray analysis (in 2009)
was ~ 0.4 nm. Additionally, as you can now see in Fig. 17.24b,
DTSA simulation suggests that the analytical sensitivity in
terms of MDA may already have reached the single-atom level
in an aberration-corrected STEM.

Figure 17.36 shows a set of X-ray maps with (a) a HAADF-

STEM image from interfaces in a [100]-projected LaMnO3/
SrTiO3 multilayer thin-film, acquired by using the JEM-
ARM200CF with a large collection-angle XEDS detector. The
X-ray maps were acquired in the SI mode with 256 × 256 pixels
for a frame time of 50 ms. The total acquisition time was ~ 40 min
and spatial-drift correction was applied during acquisition. The
bright and slightly fainter spots appearing in the HAADF-STEM
image correspond to heavy atomic columns of La or Sr and to
Fig. 17.34 a 3D-reconstructed Ti distribution around the W contact-plug creat-
Ti-O or Mn-O columns in the perovskite structure, respectively.
ed using 37 Ti K maps at tilt angles between 0°–180° with a 5° step, partially
shown in Fig. 17.33: a at 45° from all primary x, y and z axes and 2D projected
From the extracted elemental maps, two RGB color-overlay im-
Ti distributions along the z axis (b), x axis (c), and y axis (d) ages were constructed, as you can see in Fig. 17.36 (g, Red: La
L, Green: Mn K and Blue: O K) and (h, Red: Sr L, Green: Ti
K and Blue: O K), which represent LaMnO3 and SrTiO3 lay-
Figure 17.35 shows another example of XEDS tomography from ers, respectively. So we can obtain elemental distributions from
an In-doped GaN nano pyramid structure for light emitting diode much larger fields of view, if we take advantage of the improved
applications. In Fig. 17.35a, HAADF-STEM image and a set of stability in the latest instruments, such as the JEM-ARM200CF.
elemental maps of Ga, N and In are shown, which were obtained
using an instrument with an SDD X-ray collection system. You An atomic-resolution X-ray SI data set was acquired from a
can see that the In dopant is mainly located at the top of the pyr- [100]-projected GaAs specimen using the aberration-corrected
amid structure. In this field of view, the XEDS tomographic se- STEM JEM-ARM200CF. Once we obtain such atomic resolution
ries was acquired by recording two orthogonal-axis tilt-series of elemental maps in the latest instruments, can we quantify them?
X-ray spectrum images with ± 60o in 3o steps (in total, 80 XEDS To see how well we can do this, the maps were analyzed by the
spectrum images). This took a total acquisition time of ~ 8 h. ζ-factor method. Figure 17.37 shows a set of ζ-factor-processed
If you use an instrument with improved X-ray collection effi- elemental and compositional X-ray maps along with an HAADF-
ciency, you can now acquire dual-axis XEDS tomography series STEM image. In the [100]-projection of GaAs, Ga and As layers
in such reasonable times. Selected slices of the In distribution are alternatively separated. As shown in the elemental maps of
reconstructed from the dual-axes XEDS tomography datasets Ga and As with their color overlay, atomic layers of Ga and As

Chapter 17
are shown in Fig. 17.35b. You can see that In is located not only are separated as expected. Conversely in the quantified composi-
at the top of the nano pyramid structure but also along the sides tional maps, the compositions do not reach the expected 0 at% or
of the pyramid. Again, this information is not obtainable from a 100 at% in the corresponding atomic layers. The average compo-
single 2D-projected image or X-ray map. sitions of whole Ga and As maps were calculated to be 50.8 and
49.2 at%, respectively, which are very close to the nominal value
(50:50). Thus, the quantification itself was performed correctly.

17.4.3 Atomic Resolution X-ray Mapping So why do we not see the correct values for the individual atomic
columns? The deviations in compositions from the expected
values (0 or 100 at%) are partially due to the beam broadening.
If you have access to an aberration-corrected STEM, atomic According to the thickness map (Fig. 17.37), there appear to be
resolution EELS mapping can be routine. Conversely, such relatively large thickness variations between on-column and

Licensed to ([email protected])
 496 17 Practical Aspects and Advanced Applications of XEDS
Part I
Chapter 17

Fig. 17.35 a HAADF-STEM image and a set of elemental maps of Ga, N, and In obtained from an In doped GaN nano pyramid structure using an instrument with an
improved X-ray collection system. b Selected slices of the In distribution around one of the pyramid structures reconstructed from the dual-axes XEDS tomography
datasets (two orthogonal-axis tilt-series of X-ray spectrum images with ± 60o in 3o steps)
 17.4 Contemporary Aplications of X-ray Analysis 497

Part I
Chapter 17
Fig. 17.36 a An HAADF-STEM image from a LaMnO3/SrTiO3 interface, elemental maps of (b) La, (c) Sr, (d) O, (e) Mn, (f) Ti, (g) RGB color-overlay image of LaMnO3
and RGB color-overlay image of SrTiO3
 498 17 Practical Aspects and Advanced Applications of XEDS
Part I
Chapter 17

Fig. 17.37 Set of quantitative X-ray maps from a [100]-projected GaAs specimen: a HAADF-STEM image, b Ga K intensity, c As K intensity, d color overlay of Ga K
(red) and As K (green), e Ga composition, f As composition, and g thickness
 17.4 Contemporary Aplications of X-ray Analysis 499

off-column regions: ~ 60 nm and 30 nm at on- and off-column When de-channeling of the incident electrons occurs, they
regions, respectively. It is most unlikely that local specimen generate X-ray signals from neighboring columns. So, we can
thickness changes of this magnitude would occur from column conclude that, in addition to regular beam broadening de-chan-

Part I
to column. Because the specimen thickness was determined di- neling is another reason for the apparent deviations in compo-
rectly from the X-ray intensities of Ga K and As K lines, such sitions from the expected single column values. When we are
thickness enhancement at the on-column regions indicates that quantifying such atomic resolution X-ray maps, we might not
abnormal X-ray emission occurs due to channeling. Because the need to include the correction for X-ray absorption because the
map was obtained in a highly symmetric zone axis orientation, we specimens are thin enough for atomic resolution imaging and
might expect the incident beam propagation to be influenced by analysis. Instead, we now have to consider the channeling cor-
the atomic arrangement, i.e., the incident electrons are channeled rection in quantitative X-ray analyses at the atomic level, which
and de-channeled. is why you’ve already read about channeling in several previous
chapters.
In AEM

Always keep in mind beam broadening and channeling.

 9

Chapter Summary

In this chapter, we reviewed four independent subjects re- you use DTSA to first simulate the X-ray spectra from your
lated to the X-ray analysis of thin specimens in TEM to experimental specimen prior to actual analyzing them, you
augment the contents of Chaps. 32–36 in W&C. would save time (and perhaps money).

In the first section, we summarized the systematic pro- In W&C, several procedures for quantitative X-ray anal-
cedures to characterize XEDS spectrometers attached to ysis are summarized, including a brief introduction to
TEMs. You can determine several fundamental parameters the ζ-factor method. So in the third section, we went into
such as the energy resolution (as a function of the process much greater detail concerning this quantitative analysis
time), IHC, P/B, etc., from a single spectrum of a standard method. In addition, we compared the ζ-factor method
NiOx thin specimen. We recommend that these parameters with the conventional Cliff–Lorimer k-factor approach and
be characterized at least once (preferably prior to first oper- showed that the ζ-factor was much better if the specimen
ating the microscope). In addition, you should also monitor was thick, if absorption was a problem (which it is for light
the detector performance by periodically measuring spectra element X-rays) and if k-factor standards were difficult to
from the NiOx. obtain (which they usually are!).

The second section comprises a comprehensive tutorial of In the last section of this chapter, we introduced several con-
X-ray simulation through the NIST DTSA software pack- temporary applications of X-ray analysis. Because of the re-
age. Then, we described the fundamental physics related cent availabilities of aberration-correction columns in com-
to X-ray generation and detection through simulated X-ray bination with the latest developments of SDD technology,
spectra, which are never seen in practice. From these simu- the popularity of X-ray analysis has been revived. With such

Chapter 17
lated spectra, you can learn detailed aspects of X-ray spec- instrumentation X-ray analysis can now be used to obtain
trometry. Additionally, we went through four applications 3D elemental distributions and atomic-resolution elemental
of X-ray spectrum simulation including the estimation of distributions. When W&C was first penned, such advanced
X-ray absorption and calculation of analytical sensitivity. If applications were merely a dream. Now they are a reality!
 500 Appendix

Appendix
Part I

People probe size, detector collection angle etc.). State any fur-
ther assumptions.
Q17.11 Simulate X-ray spectra to confirm the conditions (the
Chuck Fiori (1938–September 15, 1992). The DeskTop Spec- beam current, specimen thickness, acquisition time and
trum Analyzer (DTSA) software package was originally devel- accelerating voltage) for 1.0 wt% and 0.3 wt% detection
oped by the late Chuck Fiori with Bob Myklebust and Carol levels of a high Z element in a relatively low Z material,
Swyt. They worked at the National Institutes of Standards and e.g., Cu in Al.
Technology (NIST) and the National Institutes of Health (NIH) Q17.12 Similar to the above question; simulate X-ray spectra
in late 1980s. to confirm the conditions for 1.0 wt% and 0.3 wt% de-
tection levels of a low Z element in a relatively high Z
Joseph (Joe) I. Goldstein was born in Syracuse, NY, on Janu- material, e.g., P in Ga.
ary 6, 1939 and died on June 27, 2015; he founded the Lehigh Q17.13 Questions 17.11 and 17.12 are for estimation of mini-
short course, inspired a generation using AEM, and recruited a mum mass fraction (MMF). Based on your estimated
young DBW to join him. conditions in the above question, how many solute at-
oms are included. The number of solute atoms is equiv-
alent to the minimum detectable mass (MDM).
Q17.14 Plot the absorption loss curves of major X-ray lines in
Questions your materials systems using Eq. 17.27 and estimate the
critical specimen thickness below 10 % absorption.
Q17.15 Using Eq. 17.28, estimate spatial resolution values
Q17.1 Compare and contrast the k-factor and the ζ-factor ap- for your specimen in conventional and aberration-cor-
proaches to quantification. rected AEMs, and plotted as a function of the specimen
Q17.2 What is the single most important variable that affects thickness. This plot is essentially same as Fig. 36.5c in
your decision to use either of these two approaches? W&C. Using this plot, determine the required specimen
Explain why you chose that variable. thickness especially for the aberration-corrected AEM.
Q17.3 Why should you simulate the spectra that you hope will
be generated from your specimen before proceeding to
gather them experimentally?
Q17.4 What can you do to minimize ice and carbon contami- Appendix 1. Error Analysis in the ζ-factor
nation on your XEDS detector Method
Q17.5 Why has it proven so difficult to detect single atoms in
thin foils using XEDS while EELS has been able to do
this for many years? As described in Sect. 17.3.3, we need an iterative calculation for
Q17.6 Distinguish the several different definitions we use for determination of compositions and specimen thickness including
analytical sensitivity. the absorption correction in the ζ-factor method. It is not very
Q17.7 Why does a single column of atoms in a thin foil not straightforward to estimate errors in such an iterative process
give rise to an XEDS spectrum containing the signal but there is an alternative approach for the error calculation. In
from these atoms alone? an n component system, we determine compositions and thick-
Chapter 17

Q17.8 What does your answer to question 17.5 lead you to ness from n characteristic X-ray intensities via n ζ factors in the
conclude about the real spatial resolution of analysis ζ-factor method. Obviously, both the n X-ray intensities and n
by XEDS? ζ factors are independent variables, and their errors need to be
Q17.9 Why are there so many characteristics of your XEDS taken into account if you want to determine the error estimation
detector that you have to determine prior to XEDS independently. Let’s denote the errors in X-ray intensity and the
analysis when, by comparison, an EEL spectrometer is ζ factor for the jth component as ∆Ij and ∆ζj, respectively.
relatively free of such requirements?
Q17.10 If you have a single atom of element B in an analyzed First, we determine the error-free composition(s) Ci and thick-
volume containing 100 atoms of element A, can you ness t from n intensities and n ζ factors without their errors.
estimate, to a first approximation, how long you need to Then, we calculate the compositions and thickness with an error
gather a spectrum in order to say with 99 % confidence contribution of jth X-ray intensity by substituting Ij + ∆Ij for Ij.
that that atom of A is present. Choose a reasonable set The composition and thickness with the error of jth intensity are
of experimental variables (including kV, beam current expressed as Ci(∆Ij) and t(∆Ij), respectively. You have to repeat
 Appendix 501

this process for all X-ray intensities independently. Similarly, the

errors in each individual ζ-factor are incorporated by substituting
ζj + ∆ζj for ζj, and composition and thickness with errors of the

Part I
z-factor are expressed as Ci(∆ζj) and t(∆ζj), respectively. Finally,
the errors in the compositions and thickness are given as:
v
u n  n
uX
2 X  2
Ci = t Ci .j / − Ci + Ci .Ij / − Ci
j=1 j=1
v (17.35)
u n  n
uX
2 X  2
t = t t.j / − t + t.Ij / − t
j=1 j=1

This approach requires 2n times extra calculations of composi- Fig. 17.38 Comparison of calculated density values by weighted (dashed line)
tions and thickness after determination of the error-free values and harmonic (solid line) means with those determined from reported lattice
(yes, it is a bit complicated and tedious!). However, we can easily parameters summarized by Okamoto et al (1987)
adapt this approach to computational codes and it is applicable to
any iterative calculation (e.g., the matrix correction procedures
for bulk-sample analysis in an EPMA such as ZAF and ϕ(ρz)).
The full error analysis procedures for the ζ-factor determination References
and estimation can be found in the paper by Watanabe and Wil-
liams (2006).

General References
Appendix 2. Calculation of the Specimen Lyman CE, Newbury DE, Goldstein JI, Williams DB, Romig
Density AD Jr., Armstrong JT, Echlin PE, Fiori CE, Joy D, Lifshin
E, Peters KR (1990) Scanning Electron Microscopy, X-Ray
Microanalysis and Analytical Electron Microscopy; A Labo-
In the ζ-factor method, we first determine the specimen thick- ratory Workbook. Plenum Press, New York
ness as the mass thickness ρt as we described above. To convert Williams DB, Goldstein JI (1991) Quantitative X-ray Microanal-
the mass thickness to the absolute specimen thickness, we need ysis in the Analytical Electron Microscope. In: Heinrich KFJ,
values of the specimen density at individual analysis points. The Newbury DE (eds) Electron Probe Quantification. Plenum
specimen density can be estimated from Eq. 35.27 in W&C, Press, New York., pp 371–398
i.e., the mass divided by the unit-cell volume. So we need some Zemyan SM, Williams DB (1995) Characterizing an Energy-Dis-
crystallographic information to determine the unit-cell volume. persive Spectrometer on an Analytical Electron Microscope.
Otherwise, the density can be calculated as a first approxima- In: Williams DB, Goldstein JI, Newbury DE (eds) X-Ray
tion by taking a weighted mean (ρ = ΣCj ρj) or a harmonic mean Spectrometry in Electron Beam Instruments. Plenum Press,
(1/ρ = ΣCi/ρj) from the density values of the individual compo- New York., pp 203–219
nent elements.

For example, Fig. 17.38 shows the composition dependence of Specific References
the specimen density in the Au-Cu system. All the symbols in

Chapter 17
this figure represent the densities calculated from reported lattice Alber U, Müllejans H, Rühle M (1997) Improved Quantification
parameters using Eq. 35.27 in W&C. The dashed and solid lines of Grain Boundary Segregation by EDS. Ultramicroscopy
indicate the estimated values from the simple weighted mean 69:105–116
and the harmonic mean, respectively. The ρ values estimated by Bennett JC, Egerton RF (1995) NiO test specimen for analytical
the harmonic mean describe the density very well. In fact, the electron microscopy: round-robin results. J Microsc Soc Am
harmonic-mean approach may work especially well for close- 1:143–149 (Following up on using NiOx in the journal now
packed condensed systems, such as metallic alloys and interme- known as Microscopy & Microanalysis)
tallic compounds. For other materials systems such as ceramics Van Cappellan E, Schmitz A (1992) A Simple Spot-size Versus
and glasses (even not crystalline), the density value needs to be Pixel-size Criterion for X-ray Microanalysis of Thin Foils.
estimated differently. Ultramicroscopy 41:193–199
Cliff G, Lorimer GW (1975) The Quantitative Analysis of Thin
Specimens. J Microsc 103:203–207 (Historical classic)
 502 17 Practical Aspects and Advanced Applications of XEDS

Egerton RF, Cheng SC (1994) The use of NiO test specimens in tion procedure and usage details in the help file, which comes
analytical electron microscopy. Ultramicroscopy 55:43–54 with the plugin package.
(All about using NiOx) Watanabe has summarized how to install DTSA in Windows on
Part I

Lovejoy TC, Ramasse QM, Falke M, Kaeppel A, Terborg R, Zan his web site (http://www.lehigh.edu/~maw3/msh/dtsaonwin-
R, Dellby N, Krivanek O (2012) Single atom identification top.html ).
by energy dispersive X -ray spectroscopy. Appl Phys Lett
100:154101-1-4
Newbury DE, Ritchie NWM (2015) Performing elemental mi- 17.1 – XEDS Detector Characterization
croanalysis with high accuracy and high precision by scan-
ning electron microscopy/silicon drift detector energy-dis- Fiori CE, Swyt CR, Ellis JR (1982) The Theoretical Character-
persive X-ray spectrometry (SEM/SDD-EDS. J Mater Sci istic to Continuum Ratio in Energy Dispersive Analysis in
50(2):493–518 (A very useful review) the Analytical Electron Microscope. In: Microbeam Analy-
Rose A (1970) Quantum limitations to vision at low light levels. sis-1982. Ed. Heinrich KFJ, San Francisco Press, San Fran-
Image Technol 12:13–15 (See also pp. 30–31) cisco, CA., pp 57–71
Watanabe M, Williams DB (1999) Atomic-Level Detection by Heinrich KFJ (1987) Mass absorption coefficients for electron
X-ray Microanalysis in the Analytical Electron Microscope. probe microanalysis. In: Brown JD, Packwod RH (eds) Proc
Ultramicroscopy 78:89–101 11th Int Cong on X-Ray Optics Microanalysis. University of
Watanabe M, Williams DB (2003) Quantification of Elemen- Western Ontario, London., pp 67–377
tal Segregation to Lath and Grain Boundaries in Low-alloy Hovington P, L’Espérance G, Baril E, Rigaud M (1993) A stan-
Steel by STEM X -ray Mapping Combined with the ζ-factor dard procedure for the modeling of the decrease in detection
Method. Z Metallk 94:307–316 (We think that this and the efficiency with time for low-energy EDS spectra. Microsc
2006 paper are worth reading!) Microanal 2:277–288
Watanabe M, Williams DB (2006) The Quantitative Analysis of
Thin Specimens: a Review of Progress from the Cliff-Lorimer
to the New ζ-Factor Methods. J Microsc 221:89–109 (As in 17.2 – X-ray Spectrum Simulation (see also
the 2003 paper, you’ll find more details and references to the
original work.) References for Software above)
We include more references than usual because this is such a new
field. We’ll add other references to the website. Fiori CE, Swyt CR (1989) The use of theoretically generated
spectra to estimate detectability limits and concentration vari-
ance in energy-dispersive X-ray microanalysis. In: Russell
References for Software PE (ed) Microbeam Analysis-1989. San Francisco Press, San
Francisco, CA.
Newbury DE, Myklebust RL, Swyt CR (1995) The use of sim-
(http://www.cstl.nist.gov/div837/837.02/epq/dtsa2/index.html).
ulated standards in quantitative electron probe microanaly-
Brundle D, Uritsky Y, Chernoff D (1996) Real-time simulation
sis with energy-dispersive X-ray spectrometry. Microbeam
for X-ray microanalysis. Solid State Technology 39(3):105–
Analysis 4:221–238
111 (Electron Flight Simulator is commercialized at http://
www.small-world.net/efs.htm)
Find a standardized description of the EMSA format through
the International Organization for Standardization (ISO 17.3 – ζ-factor Method
22029:2003). ISO 22029-2003 2003 is the Standard file
format for spectral data exchange. Also available at ANSI Armigliato A (1992) X-ray Microanalysis in the Analytical Elec-
(American National Standard Institute) web site (www.ansi. tron Microscope. In: Merli PG, Antisari MV (eds) Electron
org). Microscopy in Materials Science. World Scientific, Singa-
Chapter 17

Find SheepShaver at http://sheepshaver.cebix.net/ pore, pp 431–456

Fiori CE, Swyt CR, Myklebust RL (1992) NIST/NIH Desk Top Gorzkowski EP, Watanabe M, Scotch AM, Chan HM, Harmer
Spectrum Analyzer. Public domain software available from MP (2004) Direct Measurement of Oxygen in Lead-Based
the National Institute of Standards and Technology, Gaithers- Ceramics Using the ζ-factor Method in an Analytical Electron
burg, MD. http://www.cstl.nist.gov/div837/Division/outputs/ Microscope. J Mater Sci 39:6735–6741
DTSA/DTSA.htm Lyman CE, Goldstein JI, Williams DB, Ackland DW, von
Ritchie NWM (2008) DTSA-II. Public domain software avail- Harrach HS, Nicholls AW, Statham PJ (1994) High Perfor-
able from the National Institute of Standards and Technology, mance X-ray Detection in a New Analytical Electron Micros-
Gaithersburg, MD copy. J Microsc 176:85–98
Watanabe has developed a simple software package as a set of Romig AD Jr., Goldstein JI (1979) Detectability Limit and Spa-
plug-ins for Gatan DigitalMicrograph. This plug-in package tial Resolution in STEM X -ray Analysis: Application to
is freely available through his home page (http://www.lehigh. Fe-Ni. In: Newbury DE (ed) Microbeam Analysis – 1979.
edu/~maw3/msh/xutilmain.html ). You can find the installa- San Francisco Press, San Francisco, CA., pp 124–128 (The
 Appendix References 503

criterion for the minimum detectable peak-intensity in an Zaluzec NJ (2009) Innovative Instrumentation for Analysis of
X-ray spectrum) Nanoparticles: The π Steradian Detector. Microscopy Today
Watanabe M, Wade CA (2013) Practical Measurement of X-ray 17(4):56–59

Part I
Detection Performance of a Large Solid-Angle Silicon Drift
Detector in an Aberration-Corrected STEM. Microsc Micro-
anal 19(Suppl. 2):1264–1265 17.4 – XEDS Tomography
Möbus G, Doole RC, Inkson BJ (2003) Spectroscopic electron
17.4 – New Detector Configurations tomography. Ultramicroscopy 96:433–451
Yaguchi T, Konno M, Kamino T, Watanabe M (2008) Observa-
Erni R, Rossel MD, Kisielowski C, Dahmen U (2009) Atom- tion of Three-dimensional Elemental Distributions of a Si-de-
ic-Resolution Imaging with a Sub-50-pm Electron Probe. vice Using a 360-degree-tilt FIB and the Cold Field-emission
Phys Rev Lett 102:096101 ((4 pages).) STEM System. Ultramicroscopy 108:1603–1615 (The Hita-
von Harrach HS, Dona P, Freitag B, Soltau H, Niculae A, Rohde chi approach)
M (2009) An Integrated Silicon Drift Detector System for FEI Zaluzec NJ (2012) The Confluence of Aberration Correction,
Schottky Field Emission Transmission Electron Microscopes. Spectroscopy and Multi-Dimensional Data Acquisition. Proc.
Microsc Microanal 15(Suppl. 2):208–209 (The FEI approach) European Microscopy Congress
Kotula PG, Michael JR, Rohde M (2009) Results from Two
Four-Channel Si-drift Detectors on an SEM: Conventional
and Annular Geometries. Microsc Microanal 15(Suppl. 17.4 – Atomic Resolution X-ray Analysis
2):116–117
Ohnishi I, Okunishi E, Yamazaki K, Aota N, Miyatake K, Nakan- Watanabe M (2013) Microscopy Hacks: Development of Var-
ishi M, Ohkura Y, Kondo Y, Yasunaga K, Toh S, Matsumura ious Techniques to Assist Quantitative Nanoanalysis and
S (2011) Development of a Large Solid Angle SDD for TEM Advanced Electron Microscopy. Microscopy 62(2):217–241
and its Applications. Microsc Microanal 17(Suppl. 2):Late
Breaking 22 (The JEOL approach)
Sawada H, Tanishiro Y, Ohashi N, Tomita T, Hosokawa F, 17. Appendix 2 – Calculation of Specimen Density
Kaneyama T, Kondo Y, Takayanagi K (2009) STEM Imaging
of 47-pm-Separated Atomic Columns by a Spherical Aber- Okamoto H, Chakrabarti DJ, Laughlin DE, Massalski T (1987)
ration-Corrected Electron Microscope with a 300-kV Cold The Au-Cu (Gold-Copper) System. Bull Alloy Phase Dia-
Field Emission Gun. J Electron Microsc 58:357–361 grams 8:454–473
Watanabe M (2011) Chapter 7: X-ray Energy Dispersive Spec-
trometry in Scanning Transmission Electron Microscopes.
In: Pennycook SJ, Nellist PD (eds) Scanning Transmission
Electron Microscopy: Imaging and Analysis. Springer, New
York., pp 291–351

Chapter 17
 Figure and Table Credits

Chapter 1 Figure 2.8 Reprinted from Shorokhov D, Zewail A (2008) 4D

Electron imaging: principles and perspectives. Phys Chem
Chem Phys 10(20), 2869–3016. Figure 2. Reproduced by per-
Figure 1.6 Courtesy Bronsgeest MS and Delft University of mission of Royal Society of Chemistry.
Technology.
Figure 2.9 Reprinted from Kulovits A, Wiezorek JMK, LaGrange
Figure 1.7 Courtesy Bronsgeest MS and Delft University of T, Reed BW, Campbell GH (2011) Revealing the transient states
Technology. of rapid solidification in aluminum thin films using ultrafast in
situ transmission electron microscopy. Philos Mag Lett 91, 287–
296. Figure 2. Reproduced by permission of Taylor and Francis.

Chapter 2 Figure 2.10 Reprinted from LaGrange T, Reed BW, McKe-

own JT, Santala MJ, Dehope WJ, Huete G, Shuttlesworth RM,
Campbell GH, Microsc. Microanal. 19, 1154 (2013) Reprinted as
Figure 2.1 Reprinted from Svensson K, Olin H, Olsson E (2004) Figure 2 in: Movie-mode Dynamic Electron Microscopy, MRS
Nanopipettes for Metal Transport Phys Rev Lett 93(14) 145901- Bulletin 40(1), 22–28. Reproduced by permission of Cambridge
1-4, Fig. 4. With permission of the American Physical Society. University Press.

Figure 2.3 Robertson IM, Ferreira PJ, Dehm G, Hull R, Stach Figure 2.11 Lobastov VA, Weissenrieder J, Tang J, Zewail AH
EA (2008) Visualizing the behavior of dislocations – seeing is (2007) Ultrafast Electron Microscopy (UEM: Four-Dimensional
believing. MRS Bulletin 33, 122–131. Figure 5. Reproduced by Imaging and Diffraction of Nanostructures during Phase Trans-
permission of Cambridge University Press. formations. Nano Lett 7(9), 2552–2558. Fig. 1. Reproduced by
permission of American Chemical Society.
Figure 2.4 Rudolph AR, Jungjohann KL, Wheeler DR, Brozik
SM (2014) Drying Effect Creates False Assemblies in DNA- Figure 2.12 Adapted after Boyes ED, Gai PL (2014) Aberration
coated Gold Nanoparticles as Determined Through In Situ Liq- corrected environmental STEM (AC ESTEM) for dynamic in situ
uid Cell STEM. Microsc Microanal 20, 437–444. Figures 4 & 7. gas reaction studies of nanoparticle catalysts. J Physics: Conf
Reproduced by permission of Cambridge University Press. Ser 522, 012004. Reproduced by permission of IOP Publishing.

Figure 2.5 Panciera F, Chou Y-C, Reuter MC, Zakharov D, Stach Figure 2.13 Reprinted from Boyes ED, Gai PL (2014) Visu-
EA, Hofmann S, Ross FM (2015) Synthesis of nanostructures in alising reacting single atoms under controlled conditions:
nanowires using sequential catalyst reactions. Nature Materials 14, Advances in atomic resolution in situ Environmental (Scan-
820–826. Reproduced by permission of Nature Publishing Group. ning) Transmission Electron Microscopy (E(S)TEM. CR
Phys 15, 200–213. Fig. 1 Reproduced by permission of Else-
Figure 2.6 Reprinted from Creemer JF, Helveg S, Hoveling vier France.
GH, Ullmann S, Molenbroek AM, Sarro PM, Zandbergen HW
(2008) Atomic-scale electron microscopy at ambient pressure, Figure 2.14 Courtesy of (a) Protochips, (b) Poseidon electro-
Ultramicroscopy 108, 993–998. Figure 3b. With permission from chemical cell, (c) Hummingbird Scientific, (d) Hitachi. Dens-
Elsevier. Solutions: The Ocean series

Figure 2.7 Reprinted from LaGrange T, Campbell GH, Reed BW, Figure 2.15 Reprinted from Creemer JF, Helveg S, Hoveling
Taheri M, Peravento JB, Kim JS, Browning ND (2008) Nano- GH, Ullmann S, Molenbroek AM, Sarro PM, Zandbergen HW
second time-resolved investigations using the in situ of dynamic (2008) Atomic-scale electron microscopy at ambient pressure.
transmission electron microscope (DTEM. Ultramicroscopy 108, Ultramicroscopy 108, 993–998. Figure 3b. Reproduced by per-
1441–1449. Fig. 1. With permission from Elsevier. mission of Elsevier.

© Springer International Publishing Switzerland 2016

505
C. B. Carter, D. B. Williams (Eds.), Transmission Electron Microscopy, DOI 10.1007/978-3-319-26651-0
506 Figure and Table Credits

Figure 2.16 Reprinted from Yoshida H, Omote H, Takeda S Figure 2.27a–e Reprinted from Saka H, Kamino T, Ara S, Sasaki
(2014) Oxidation and reduction processes of platinum nanopar- K (2008) In Situ Heating Transmission Electron Microscopy.
ticles observed at the atomic scale by environmental transmis- MRS Bull, 33, 93–100, Fig. 1. Reproduced by permission of
sion electron microscopy. Nanoscale 6, 13113–13118. Repro- Cambridge University Press.
duced by permission of Royal Society of Chemistry.
Figure 2.27f Reprinted from Gandman M, Kauffmann Y, Koch
Figure 2.17 Reprint from Alan T, Yokosawa T, Gaspar J, Pan- CT, Kaplan WD (2013) Direct Quantification of Ordering at a
draud G, Paul O, Creemer F, Sarro PM, Zandbergen HW (2012) Solid-Liquid Interface Using Aberration Corrected Transmis-
Micro-fabricated channel with ultra-thin yet ultra-strong sion Electron Microscopy. Phys Rev Lett 110, 086106. Fig. 1.
windows enables electron microscopy under 4-bar pressure. Reproduced by permission of American Physical Society.
APL, 100, 081903-1-4, Fig. 5. With permission from AIP Pub-
lishing LLC. Figure 2.28 Reprinted from Kasama T, Dunin-Borkowski
RE, Matsuya L, Broom RF, Twitchett AC, Midgley PA, New­
Figure 2.18 After Jungjohann KL (2012) Ph.D. Thesis, UC Da- comb SB, Robins AC, Smith DW, Gronsky JJ, Thomas CA,
vis. Nanoscale imaging and analysis of fully hydrated materials. Fischione PE (2005) A versatile three-contact electrical bi-
Figure 3.10. asing transmission electron microscope specimen holder for
electron holography and electron tomography of working de-
Figure 2.19 Photo by Carter CB courtesy Furuya K. vices. Mater. Res. Soc. Symp. Proc., 907E, MM13.02. “In situ
Electron Microscopy of Materials”, Eds Ferreira PJ, Robertson
Figure 2.20 Reprint from Mehraeen S, McKeown JT, Desh­ IM, Dehm G, Saka H. Fig. 8. Reproduced by permission of
mukh PV, Evans JE, Abellan P, Xu P, Reed BW, Taheri ML, Cambridge University Press.
Fischione PE, Browning ND (2013) A (S)TEM Gas Cell Holder
with Localized Laser Heating for In Situ Experiments. Microsc Figure 2.29 Reprinted from Asoro MA, Kovar D, Ferreira PJ
Microanal 19(2), 470–478, Fig. 1. Reproduced by permission of (2013) In Situ Transmission Electron Microscopy Observa-
Cambridge University Press. tions of Sublimation in Silver Nanoparticles. ACS Nano, 7(9),
7844–7852, Figs 1&2. Reproduced by permission of American
Figure 2.21 Reprinted from Niu K-Y, Park J, Zheng H, Alivisatos Chemical Society.
AP (2013) Revealing Bismuth Oxide Hollow Nanoparticle For-
mation by the Kirkendall Effect. Nano Letters 13, 5715–5719. Figure 2.30 Reprinted from Delalande M, Guinel MJ-F, Allard
Reproduced by permission of American Chemical Society. LF, Delattre A, Le Bris R†, Yves Samson Y, Bayle-Guillemaud P,
Peter Reiss P (2012) L10 Ordering of ultrasmall FePt Nanopar-
Figure 2.22 Reprinted from Wang C, Qiao Q, Shokuhfar T, Klie ticles Revealed by TEM in situ annealing. J Phys Chem C 116,
RF (2014) High-Resolution Electron Microscopy and Spectros- 6866–6872. Fig. 7. Reproduced by permission of American
copy of Ferritin in Biocompatible Graphene Liquid Cells and Chemical Society.
Graphene Sandwiches. Advanced Materials 26, 3410–3414.
Fig. 1. Reproduced by permission of John Wiley and Sons. Figure 2.31 Reprinted from Vendelbo1 SB, Elkjær CF, Falsig H,
Puspitasari I, Dona P, Mele L, Morana B, Nelissen BJ, van Rijn
Figure 2.23 Reprinted from White ER, Singer SB, Augustyn V, R, Creemer JF, Kooyman PJ, Helveg S (2014) Visualization of
Hubbard WA, Mecklenburg M, Dunn B, Regan BC (2012) In Situ oscillatory behaviour of Pt nanoparticles catalysing CO oxida-
Transmission Electron Microscopy of Lead Dendrites and Lead tion. Nature Mater 13(9), 884–890. Reproduced by permission
Ions in Aqueous Solution. ACS Nano, 6(7), 6308–17, Fig. 1. Re- of Nature Publishing Group.
produced by permission of American Chemical Society.
Figure 2.32 Reprinted from Evans JE, Jungjohann KL, Wong
Figure 2.24 Reprinted from Mortensen PM, Hansen TW, Wagner PCK, Chiu P-L, Dutrow GH, Arslan I, Browning ND (2012) Vi-
JB, Jensen AD (2015) Modeling of temperature profiles in an sualizing macromolecular complexes with in situ liquid scanning
environmental transmission electron microscope using compu- transmission electron microscopy. Micron 43 1085–1090. Fig. 3.
tational fluid dynamics. Ultramicroscopy 152, 1–9, Fig 7 and 10. Reproduced by permission of Elsevier.
Reproduced by permission of Elsevier.
Figure 2.33 Wirix MJM, Bomans PHH, Friedrich H, Sommerdijk
Figure 2.25 Reprinted by permission from Macmillan Publish­ers NAJM, de With G (2014) Three-Dimensional Structure of P3HT
Ltd: Gao Y, Bando Y (2002) Carbon Nanothermometer Contain- Assemblies in Organic Solvents Revealed by Cryo-TEM. Nano
ing Gallium. Nature 415, 599. Copyright 2002. Fig. 1. Repro- Lett 14, 2033–2038. Fig. 2. Reproduced by permission of Amer-
duced by permission of Nature Publishing Group. ican Chemical Society.

Figure 2.26 Courtesy Basu J, Ravishankar N, Carter CB. Un- Figure 2.34 Tai K, Liu Y, Dillon SJ (2014) In Situ Cryogenic
published research. Transmission Electron Microscopy for Characterizing the Evo-
lution of Solidifying Water Ice in Colloidal Systems. Microsc

Licensed to ([email protected])
 Figure and Table Credits 507

Microanal 20, 330–337. Fig. 1a. Reproduced by permission of electrical properties measured inside a transmission electron
Cambridge University Press. microscope. J Mater Sci 43, 1460–1470. Fig 6. Reproduced by
permission of Springer Publishing Company.
Figure 2.35 Haque MA, Saif MTA (2002) In situ tensile test-
ing of nano-scale specimens in SEM and TEM. Experimental Figure 2.45 Liu XH, Huang JY (2011) In situ TEM electro-
Mech 42(1), 123–128, Fig. 1. Reproduced by permission of chemistry of anode materials in lithium ion batteries. Energy
Springer Publishing Company. & Environmental Sci, 4(10), 3844–3860, Fig. 1. Reproduced by
permission of Royal Society of Chemistry.
Figure 2.36 Reprinted from Ye J, Mishra RK, Pelton AR,
Minor AM (2010) Direct observation of the NiTi martensi- Figure 2.46 Wang CM, Li X, Wang Z, Wu W, Liu J, Gao F,
tic phase transformation in nanoscale volumes. Acta Mater Kovarik L, Zhang J-G, Howe J, Burton DJ, Liu Z, Xiao X,
58, 490–498. Fig 1b & 2b. Reproduced by permission of El- Thevuthasan S, Baer DR (2012) In Situ TEM Investigation of
sevier. Congruent Phase Transition and Structural Evolution of Nano-
structured Silicon/Carbon Anode for Lithium Ion Batteries. Nano
Figure 2.37 Reprinted from Espinosa HD, Bernal RA, Filleter Lett 12, 1624–1632. Fig. 4. Reproduced by permission of Amer-
T (2012) In Situ TEM Electromechanical Testing of Nanowires ican Chemical Society.
and Nanotubes. Small 8(21), 3233–3252, Fig. 4. Reproduced by
permission of John Wiley and Sons. Figure 2.47 Liu Y, Liu XH, Nguyen BM, Yoo J, Sullivan JP,
Picraux ST, Huang JH, Dayeh SA (2013) Tailoring Lithiation Be-
Figure 2.38 Reprinted from Nowak JD, Mook WM, Minor AM, havior by Interface and Bandgap Engineering at the Nanoscale.
Gerberich WW, Carter CB (2007) Fracturing a Nanoparticle. Nano Letters, 13(10), 4876–4833, Fig. 1, 3. Reproduced by per-
Philos Mag 87(1), 29–37, Fig. 1. Reproduced by permission of mission of American Chemical Society.
Taylor and Francis.
Figure 2.48 Leenheer AJ, Jungjohann KL, Zavadil KR, Sullivan
Figure 2.39 Reprinted from Bernal RA, Filleter T, Connell JP, Harris CT, Lithium electrodeposition dynamics in aprotic
JG, Sohn K, Huang J, Lauhon LJ, Espinosa HD (2014) In situ electrolyte observed in Situ via Transmission Electron Micros-
electron microscopy four-point electromechanical characteri­ copy, Fig. 2. Reproduced by permission of American Chemical
zation of freestanding metallic and semiconducting nanowires. Society.
Small 10 (4), 725–733. Reproduced by permission of John Wiley
and Sons. Figure 2.49a Zhang C, Tian W, Xu Z, Wang X, Liu J, Li S-N,
Tang D-M, Liu D, Liao M, Bando Y, Goldberg D (2014) Pho-
Figure 2.40 Reprinted from Lau JW, Schofield MA, Zhu Y, Neu- tosensing performance of branched CdS/ZnO heterostructures
marl GF (2003) In situ Magnetodynamic Experiments Achieved as revealed by in situ TEM and photodetector tests. Nanoscale
with the design of an In-plane Magnetic Field Specimen Holder. 6, 8084–8090, Fig. 4, 5. Reproduced by permission of Royal
Microsc. Microanal. 9, (Suppl 2), 130–131. Reproduced by per- Society of Chemistry.
mission of Cambridge University Press.
Figure 2.50 Zhang L, Miller BK, Crozier PA (2013) Atomic Level
Figure 2.41 Reprinted from Budruk A, Phatak C, Petford-Long In Situ Observation of Surface Amorphization in Anatase Nano-
AK, De Graef M (2011) In situ Lorentz TEM magnetization crystals During Light Irradiation in Water Vapor. Nano Lett 13,
studies on a Fe–Pd–Co martensitic alloy. Acta Mater 59, 6646– 679–684, Fig. 1, 2, 4. Reproduced by permission of American
6657, Fig. 1. Copyright 2011, with permission from Elsevier. Chemical Society.

Figure 2.42 Reprinted from Wang J, Zeng Z, Weinberger CR, Figure 2.51 Reprinted from Picher M, Mazzucco S, Blankenship
Zhang Z, Zhu T, Scott X. Mao (2015) In situ atomic-scale S, Sharma R (2015) Vibrational and optical spectroscopies in-
observation of twinning dominated deformation in nanoscale tegrated with environmental transmission electron microscopy.
body-centred cubic tungsten Nature Mater 14, 594–560. Fig- Ultramicroscopy 150, 10–15, Fig. 1, 2, 4c. Copyright 2015 with
ure 13 of Supplementary data. Reproduced by permission of permission from Elsevier.
Nature Publishing Group.
Figure 2.52 Reprinted from Hillerich K, Dick KA, Wen C-Y,
Figure 2.43 Reprinted from Spence JCH (1988) A Scanning Reuter MC, Kodambak S, Ross FM (2013) Strategies To Con-
Tunneling Microscope in a Side-Entry Holder for Reflection trol Morphology in Hybrid Group III–V/Group IV Hetero-
Electron-Microscopy in the Philips EM400, Ultramicroscopy structure Nanowires. Nano Lett 13, 903–908. Reproduced by
25, 165–170, Fig. 4. Copyright 1988 with permission from permission of American Chemical Society.
Elsevier.
Figure 2.53 Reprinted from Simonsen SB, Chorkendorff I,
Figure 2.44 Reprinted from Costa PMFJ, Golberg D, Shen G, Dahl S, Skoglundh M, Sehested J, Helveg S (2010) Direct Ob-
Mitome M, Bando Y (2008) ZnO low-dimensional structures: servation of Oxygen-induced Platinum Nanoparticle Ripening
508 Figure and Table Credits

Studied by In Situ TEM. JACS 132, 7968–7975, Fig. 3. Repro- Figure 4.8 as Figure 4.3
duced by permission of American Chemical Society.
Figure 4.13 as Figure 4.3
Figure 2.54 Reprinted Woehl TJ, Evans JE, Arslan I, Risten-
part WD, Browning ND (2012) Direct in Situ Determination Figure 4.11 From: Tanaka M, Saito R, Sekii H (1983a). Point-
of the Mechanisms Controlling Nanoparticle Nucleation and Group Determination by Convergent-Beam Electron Diffraction.
Growth. ACS Nano 6(10), 8599–8610, Fig. 1. Reproduced by Acta Cryst., A39, 357–368. Reproduced by permission of John
permission of American Chemical Society. Wiley and Sons.

Figure 2.55 Reprinted from Schneider NM, Norton MM, Men- Figure 4.16 From: Morniroli JP (2002) Large-Angle Conver-
del BJ, Grogan JM, Ross FM, Bau HH (2014) Electron-water gent-Beam Electron Diffraction. Paris, Société Française des
interactions and implications for liquid cell electron microscopy. Microscopies, Figure VI.21. Reproduced by permission of John
J Phys Chem C 118, 22373–22382, Fig. 1, 2, 5 Reproduced by Wiley and Sons.
permission of American Chemical Society.
Figure 4.17 From: Eades JA, Kiely CJ (1987) Convergent-beam
Figure 2.56 Reprinted from Hattar K, Bufford DC, Buller DL diffraction. EMAG 87, Inst. Phys Conf Series, 90, 109–114. Re-
(2014) Concurrent in situ ion irradiation transmission electron produced by permission of Taylor and Francis.
microscope. Nucl Instrum Meth B 338, 56–65, Fig. 2. Repro-
duced by permission of Elsevier. Figure 4.19 From Cherns D, Preston AR (1989) Convergent
Beam Diffraction Studies of Interfaces, Defects and Multilay-
Figure 2.57 Reprinted from Hattar K, Bufford DC, Buller DL ers. J. Electron Microsc. Tech., 13, 111–122. Reproduced by
(2014) Concurrent in situ ion irradiation transmission electron permission of John Wiley and Sons
microscope. Nucl Instrum Meth B 338, 56–65. Fig. 8. Repro-
duced by permission of Elsevier. Figure 4.20 From Tanaka M, Terauchi M, Kaneyama T (1988)
Convergent-Beam Electron Diffraction II. JEOL, Tokyo,
page 165. Reproduced by permission of the authors and JEOL.

Chapter 3 Figure 4.21 From Morniroli JP (2002) Large-Angle Conver-

gent-Beam Electron Diffraction. Paris, Société Française des
Microscopies, Figure XI.31. Reproduced by permission of John
All Chapter 3 figures are unpublished works, with the exception of Wiley and Sons.

Figure 3.11 Sundararaman M, Mukhopadhyay P, Banerjee S Figure 4.23 From Tanaka M, Terauchi M (1985) Conver-
(1988) Precipitation of the δ-Ni3Nb phase in Two Nickel Base gent-Beam Electron Diffraction. JEOL, Tokyo, Page 83. Repro-
Superalloys. Met. Trans., 19A, 453. Reproduced by permission duced by permission of the authors and JEOL.
of Springer Publishing Company.
Figure 4.24 From Humphreys CJ, Eagelsham DJ, Maher DM,
Fraser HL (1988) CBED and CBIM from semiconductors and
superconductors. Ultramicroscopy 26(1–2), 13–23. Reproduced
Chapter 4 by permission of Elsevier.

Figure 4.25 Courtesy Christenson KK, Eades JA.

Figure 4.2 Unpublished work courtesy Hseih KC.
Figure 4.26 With permission from Christoph T. Koch
Figure 4.3 Kaufman MJ, Eades JA, Loretto MH, Fraser HL
(1983) A Study of a Cellular Phase Transformation in the Ter- Table 4.3 After: Eades JA (1988) Symmetry Determination by
nary Ni-Al-Mo Alloy System. Met. Trans., 14A, 1561–1571. Re- Convergent-beam Diffraction. EUREM 88; IOP Conf. Series
produced by permission of Springer. 93, 1, 3–12.

Figure 4.4 as Figure 4.3 Table 4.5 After: Buxton BF, Eades JA, Steeds JW, Rackham GM
(1976) The Symmetry of Electron Diffraction Zone Axis Patterns
Figure 4.5 as Figure 4.3 Phil. Trans. A281, 171–194.

Figure 4.6 as Figure 4.3 Table 4.6 Adapted from: Eades JA, Shannon MD, Buxton BF.
SEM (1983) Crystal Symmetry from Electron Diffraction III,
Figure 4.7 as Figure 4.3 1051–1060.
 Figure and Table Credits 509

Chapter 5 Chapter 11

Figure 5.6: Zuo J, Kim M, O’Keeffe M, Spence JCH (1999) Di- Figure 11.1 Adapted from Pennycook SJ, Browning ND, Mc-
rect observation of d-orbital holes and Cu-Cu bonding in Cu2O. Gibbon MM, McGibbon AJ, Jesson DE, Chisholm MF (1996a)
Nature, 401, 49–52. Reproduced by permission of Nature Pub- Philos. Trans. R. Soc. A. 354, 2619.
lishing Group.
Figure 11.2 From Pennycook SJ, Jesson DE (1992) Acta Metall.
Figure 5.7: Courtesy Mayer J, unpublished. For more details see Mater. 40, S149. Reproduced under US Govt agreement.
Krämer S, Mayer J (1999) Using the Hough transform for HOLZ
line identification in Convergent Beam Electron Diffraction. J Mi- Figure 11.3 Adapted from Pennycook SJ, Jesson DE, Chisholm
crosc 194, 2–11 (1999) MF, Browning ND (1993) Atomic-Resolution Imaging and Anal-
ysis with the STEM, Vol. 130, X-Ray Optics and Microanalysis
Figure 5.9a: Tanaka M et al (2003) CBED. JEOL, Vol. 3. Repro- 1992, edited by Kenway PB, Duke PJ, Lorimer GW, Mulvey T,
duced by permission of the authors and JEOL. Drummond IW, Love G, Michette AG and Stedman M, pp. 217.
Bristol: Institute of Physics.

Figure 11.4 From Pennycook SJ, Jesson DE (1991) Incoherent

Chapter 6 characteristics; the probe propagating Ultramic. 37, 14. Repro-
duced under US Govt agreement.

All Figures are the product of the author. Figure 11.5 Reproduced from Peng Y, Oxley MP, Lupini AR,
Chisholm MF, Pennycook SJ (2008) Microsc. Microanal. 14,
36.

Chapter 7 Figure 11.6 Adapted from Pennycook SJ, Jesson DE, Chisholm
MF, Ferridge AG, Seddon MJ (1992) Sub-Ångstrom Microscopy
through Incoherent Imaging and Image Restoration, Vol. Scan-
All Figures are the product of the author. ning Microscopy Supplement 6, Signal and Image Processing in
Microscopy and Microanalysis, edited by PW. Hawkes, Cam-
bridge, UK: Scanning Microscopy International.

Chapter 8 Figure 11.7 (a) Adapted from Abe E, Pennycook SJ (2005) J.

Crystallog. Soc. Japan 47, 26. (b) is from Steinhardt PJ, Jeong
HC, Saitoh K, Tanaka M, Abe E, Tsai AP (1998) Nature 396, 55.
All Figures are the product of the author. For Figure 7a,b Reproduced under US Govt agreement.

Figure 11.8 (a) From Merli PG, Missiroli GF, Pozzi G (1976)
Am. J. Phys. 44, 306, Reproduced by permission of … (b) from
Chapter 9 Tonomura A, Endo J, Matsuda T, Kawasaki T, Ezawa H (1989)
Am. J. Phys. 57, 117. Reproduced with permission.

All Figures are the product of the author. Figure 11.14 From Nellist PD, Pennycook SJ (1999a) Inst Phys
Conf Ser 315. Reproduced under US Govt agreement.

Figure 11.16 Images a,b and c courtesy of McGibbon MM, Varela

Chapter 10 M, Lupini AR, respectively.

Figure 11.17 Images from (A) Nellist PD, Pennycook SJ (1996)

Most Figures are the product of the author. Science 274, 413 Reproduced under US Govt agreement (B)
Sohlberg K, Rashkeev S, Borisevich AY, Pennycook SJ, Pan-
Figure 10.1: Courtesy EMAT center for Electron Microscopy, Uni- telides ST (2004) Pt atoms on γ-alumina Chemphyschem 5, 1893
versity of Antwerp. Reproduced under US Govt agreement (C) Wang SW, Borise-
vich AY, Rashkeev SN, Glazoff MV, Sohlberg K, Pennycook
Figure 10.8: Courtesy Jia CL, Thust A (1999) Phys Rev Lett 82, SJ, Pantelides ST (2004) On a γ-alumina flake Nature Mat. 3,
5052. Reproduced by permission. 143 Reproduced under US Govt agreement.
510 Figure and Table Credits

Figure 11.18 Adapted from Zhou W, Kapetanakis M, Prange Figure 11.39 Simulated Ronchigrams courtesy Lupini AR.
M, Pantelides S, Pennycook S, Idrobo JC (2012a) Phys. Rev.
Lett. 109, 206803. Figure 11.40 Adapted from Nellist and Pennycook (1998b).

Figure 11.19 Adapted from Zhou W, Kapetanakis M, Prange Figure 11.41 Adapted from Jesson and Pennycook (1995).
M, Pantelides S, Pennycook S, Idrobo JC (2012a) Phys. Rev.
Lett. 109, 206803. Figure 11.42 Adapted from Jesson and Pennycook (1995).

Figure 11.20 Adapted from Lee J, Zhou W, Pennycook SJ, Idrobo Figure 11.43 Adapted from Pennycook et al. (1996b).
JC, Pantelides ST (2013) Graphene nanopore Nature Commu-
nications 4, 1650. Figure 11.44 Adapted from Nellist and Pennycook (2000).

Figure 11.21 Adapted from Borisevich AY, Lupini AR, Penny- Figure 11.45 Adapted from Borisevich et al. (2006a)
cook SJ (2006a) Proc. Nat. Acad. Sci. USA 103, 3044.
Figure 11.46 Adapted from Varela et al. (2005).
Figure 11.22 From Borisevich AY, Lupini AR, Pennycook SJ
(2006a) Proc. Nat. Acad. Sci. USA 103, 3044. Reproduced under Figure 11.48 Courtesy Miyoung Kim.
US Govt agreement.
Figure 11.49 From Perovic DD, Rossouw CJ, Howie A (1993)
Figure 11.23 Adapted from van Benthem K, Lupini AR, Kim M, Ultramicroscopy 52, 353. Reproduced with permission.
Baik HS, Doh S, Lee JH, Oxley MP, Findlay SD, Allen LJ, Luck
JT, Pennycook SJ (2005) Appl. Phys. Lett. 87, 034104. A movie Figure 11.50 From Xin Y, Pennycook SJ, Browning ND, Nel-
version of Figure 11.23. list PD, Sivananthan S, Omnes F, Beaumont B, Faurie JP, Gibart P
(1998) Appl. Phys. Lett. 72, 2680. Dislocations emerging at the
Figure 11.24 From van Benthem K, Lupini AR, Kim M, Baik surface of GaN. Reproduced under US Govt agreement.
HS, Doh S, Lee JH, Oxley MP, Findlay SD, Allen LJ, Luck JT,
Pennycook SJ (2005) Appl. Phys. Lett. 87, 034104. A movie Figure 11.51 Adapted from Kadavanich et al. (2001) McBride
version of Figure 11.23. Reproduced under US Govt agree- et al. (2006) Pennycook et al. (2003).
ment.
Figure 11.52 Adapted from Roberts et al. (2008).
Figure 11.29 From James EM, Browning ND (1999) Practical as-
pects of atomic resolution imaging and analysis in STEM. Ultra- Figure 11.53 From Shibata N, Pennycook SJ, Gosnell TR,
microscopy 78, 125–139 Reproduced by permission of Elevier. Painter GS, Shelton WA, Becher PF (2004) Nature 428, 730.
Reproduced under US Govt agreement.
Figure 11.30 Adapted from Pennycook et al. (2003).
Figure 11.54 Adapted from Pennycook (2002); Yan et al. (1999).
Figure 11.31 From Nellist PD, Chisholm MF, Dellby N, Krivanek
OL, Murfitt MF, Szilagyi ZS, Lupini AR, Borisevich A, Sides Figure 11.55 From LeBeau JM, D’Alfonso AJ, Findlay SD,
WH, Pennycook SJ (2004) Science 305, 1741. Supplementary Stemmer S, Allen LJ (2009) Phys Rev B 80, 214110. No Stobbs
Online Material. Reproduced under US Govt agreement. factor. Reproduced with permission.

Figure 11.31 Courtesy Duscher G. Figure 11.56 From Molina SI, Varela M, Ben T, Sales DL, Pizarro
J, Galindo PL, Fuster D, Gonzalez Y, Gonzalez L, Pennycook SJ
Figure 11.32 Ronchigrams courtesy of Gerd Duscher. (2008) J. Nanosci. Nanotech. 8, 3422. Calibrated by high reso-
lution X-ray diffraction. Reproduced under US Govt agreement.
Figure 11.33b (Pennycook et al. (2009) The original data points
from Browning et al. (1993a). Adapted from Browning et al.
1993a; Browning et al. (1993b)
Chapter 12
Figure 11.34 Adapted from Peng et al. (2008).

Figure 11.35 Adapted from Pennycook and Nellist (1999). Figure 12.8 Reprinted from Ultramicroscopy, Vol 109, Baten-
burg KJ, Bals S, Sijbers J, Kübel C, Midgley PA, Hernandez JC,
Figure 11.36 Adapted from Jesson and Pennycook (1993). Kaiser U, Encina ER, Coronado EA, Van Tendeloo G (2009)
3D imaging of nanomaterials by discrete tomography, 730–740.
Figure 11.38 Simulated Ronchigrams courtesy Lupini AR. Copyright 2009, with permission from Elsevier.
 Figure and Table Credits 511

Figure 12.9 Saghi Z, Holland DJ, Leary R, Falqui A, Bertoni G, structures, 1401–1407. Copyright 2008, with permission from
Sederman AJ, Gladden LF, Midgley PA (2011) Three-Dimen- Elsevier.
sional Morphology of Iron Oxide Nanoparticles with Reactive
Concave Surfaces. Nano Letters, 11, 4666–4673. Reproduced by Figure 12.20 Phatak C, Petford-Long AK, De Graef M (2010)
permission of American Chemical Society. Three-Dimensional Study of the Vector Potential of Magnetic
Structures. Physical Review Letters, 104, 253901. Reproduced
Figure 12.10 Jinnai H, Nishikawa Y, Spontak RJ, Smith SD, by permission of the American Physical Society.
Agard DA, Hashimoto T (2000) Direct measurement of interfa-
cial curvature distributions in a bicontinuous block copolymer Figure 12.21 Van Aert S, Batenburg KJ, Rossell MD, Erni R,
morphology. Physical Review Letters, 84(3), 518–521. Repro- Van Tendeloo G (2011) Three-dimensional atomic imaging of
duced by permission of American Physical Society. crystalline nanoparticles. Nature, 470, 374–377. Reproduced
by permission of Nature.
Figure 12.11 Midgley PA, Thomas JM, Laffont L, Weyland M,
Raja R, Johnson BFG, Khimyak T (2004) Reprinted with per- Figure 12.24 Sample courtesy of Xiong X.
mission of American Chemical Society.

Figure 12.12 Ercius P, Weyland M, Muller DA, Gignac LM

(2006) Three-dimensional imaging of nanovoids in copper in- Chapter 13
terconnects using incoherent bright field tomography. Applied
Physics Letters, 88(24). Reproduced by permission of AIP Pub-
lishing LLC. Figure 13.5 Midgley PA, Saunders M (1996) Quantitative elec-
tron-diffraction – from atoms to bonds. Contemp. Phys., 37, 441
Figure 12.13. Reprinted from Micron, Vol. 47, Lepinay K, Lo- . Reproduced by permission of Taylor & Francis.
rut F, Pantel R, Epicier T (2013) Chemical 3D tomography of
28 nm high K metal gate transistor: STEM XEDS experimental Figure 13.14 Sigle W, Krämer S, Varshney V, Zern A, Eigen-
method and results, 43–49. Copyright 2013, with permission thaler U, Rühle M (2003) Plasmon energy mapping in energy-fil-
from Elsevier. tering transmission electron microscopy. Ultramicroscopy, 96,
565–571. Reproduced with permission of Elsevier.
Figure 12.14 Reprinted from Scripta Materialia, Vol. 55, Wey-
land M, Yates TJV, Dunin-Borkowski RE, Laffont L, Midgley Figure 13.15 Sample provided courtesy of Lin W of the Mayo
PA (2006) Nanoscale analysis of three-dimensional structures by Clinic, Jacksonville.
electron tomography, 29–33. Copyright 2006, with permission
from Elsevier. Figure 13.17 After: Hofer F, Grogger W, Kothleitner G, Warbi-
chler P (1997) Quantitative analysis of EFTEM elemental dis-
Figure 12.15 Gass MH, Koziol KKK, Windle AH, Midgley PA tribution images. Ultramicroscopy, 67, 83–103.
(2006) 4-dimensional spectral-tomography of carbonaceous
nano-composites. Nano Letters, 6(3), 376–379. Reproduced by Figure 13.18 After: Thomas PJ, Midgley PA (2001) Image-Spec-
permission of American Chemical Society. troscopy 2: the Removal of Plural Scattering from Extended
Energy-Filtered Series by Fourier Deconvolution Techniques.
Figure 12.16 Reprinted from Ultramicroscopy, Vol. 109, Ja- Ultramicroscopy, 88, 187–194.
rausch K, Thomas P, Leonard DN, Twesten R, Booth CR (2009)
Four-dimensional STEM-EELS: Enabling nano-scale chemical Figure 13.20 After: Walther T (2003) Electron energy-loss spec-
tomography, 326–337. Copyright 2009, with permission from troscopic profiling of thin film structures: 0.39 nm line resolution
Elsevier. and 0.04 eV precision measurement of near-edge structure shifts
at interfaces. Ultramicroscopy, 96, 401–411.
Figure 12.17 Image courtesy of Barnard J.
Figure 13.21 After Midgley PA (2009).
Figure 12.18 Reprinted from Scripta Materialia, Vol. 59, Tanaka
M, Higashida K, Kaneko K, Hata S, Mitsuhara M (2008) Crack
tip dislocations revealed by electron tomography in silicon single
crystal, 901–902. Copyright 2008, with permission from Elsevier. Chapter 14
Figure 12.19 Reprinted from Ultramicroscopy, Vol. 108,
Twitchett-Harrison AC, Yates TJV, Dunin-Borkowski RE, Figure 14.1 Spectra replotted from the following sources. i)
Midgley PA (2008) Quantitative electron holographic tomog- 1974: Egerton RF, Whelan MJ, J Electron Spectroscopy and
raphy for the 3D characterization of semiconductor device Related Phenomena, 3 (1974) 232; ii) 1989: Weng X, Rez P,
Ma H, Phys. Rev. B 40 (1989) 4175; iii) 2001: Soininen JA,
512 Figure and Table Credits

Shirley EL, Phys. Rev. B 64 (2001) 165112. iv) 1998 and Exp: Figure 16.19 reproduced by permission of The Meteoritical So-
Merchant AR, McCulloch DG, Brydson R, Diamond and Related ciety from Goldstein et al. 2007.
Materials 7 (1998) 1303.
Figure 16.22 B&C reproduced by permission of Taylor &Francis
Figure 14.3 Singh DJ, Planewaves, Psuedopotentials and the from Yang et al 2013.
LAPW Method, Kluwer, Boston (1994) Reproduced by permis-
sion of Springer (formerly Kluwer). Figure 16.23 reproduced by permission of The American Chem-
ical Society from Guiton et al. 2011.
Figure 14.4 After: Keast VJ, Bosman M. Microsc Res
Techn 70:211–219 (2007)

Figure 14.5 Experimental data from Phys. Rev. B 66 (2002) Chapter 17

125319 and Mizoguchi T et al, 34 Micron (2003) 249.

Figure 14.6 Experimental data and multiplet calculations from Figure 17.25 Gorzkowski EP, Watanabe M, Scotch AM, Chan
FM. de Groot et al Phys. Rev. B 41 (1990) 928. HM, Harmer MP (2004) Direct Measurement of Oxygen in Lead-
Based Ceramics Using the ζ-factor Method in an Analytical
Figure 14.7 From Wu ZY, Ouvard G, Gressier P, Natoli CR. Electron Microscope J. Mater. Sci. 39 6735–6741.Reproduced
Phys. Rev. B 55 (1997) 10382. Reproduced by permission of by permission of Springer.
American Physical Society.
Figure 17.26 Watanabe M, Williams DB (2006) The Quantitative
Figure 14.8 Related calculations given in J Electron Spectros- Analysis of Thin Specimens: a Review of Progress from the Cliff-
copy and Related Phenomena 143 (2005) 97–104 Lorimer to the New ζ-Factor Methods J. Microsc. 221 89–109.
As in the 2003 paper, you’ll find more details and references
Figure 14.9 From Vast N et al. Phys. Rev. Lett 88 (2002) 037601. to the original work. Reproduced by permission of John Wiley
Reproduced by permission of American Physical Society. and Sons.

Figure 17.27 From Watanabe M (2013) Microscopy Hacks: De-

velopment of Various Techniques to Assist Quantitative Nano-
Chapter 15 analysis and Advanced Electron Microscopy, Microscopy, 62,
217–241. Reproduced by permission of Oxford University Press.

All Figures are the product of the author. Figure 17.30 From Watanabe M (2013) Microscopy Hacks: De-
velopment of Various Techniques to Assist Quantitative Nano-
analysis and Advanced Electron Microscopy, Microscopy, 62,
217–241. Reproduced by permission of Oxford University Press.
Chapter 16
Figure 17.32 Courtesy of Okunishi E, JEOL Reproduced by per-
mission of JEOL.
Figure 16.2 reproduced by permission of Cambridge University
Press from Kotula & Keenan 2006. Figure 17.33 Yaguchi T, Konno M, Kamino T, Watanabe M
(2008) Observation of Three-dimensional Elemental Distribu-
Figure 16.10a,b reproduced by permission of Cambridge Uni- tions of a Si-device Using a 360-degree-tilt FIB and the Cold
versity Press from Kotula & Keenan 2006. Field-emission STEM System Ultramicrosc., 108, 1603–1615.
Reproduced by permission of Elsevier.
Figure 16.11 reproduced by permission of Cambridge University
Press from Kotula & Keenan 2006. Figure 17.34 Yaguchi T, Konno M, Kamino T, Watanabe M
(2008) Observation of Three-dimensional Elemental Distribu-
Figure 16.12 reproduced by permission of Cambridge University tions of a Si-device Using a 360-degree-tilt FIB and the Cold
Press from Kotula & Keenan 2006. Field-emission STEM System Ultramicrosc., 108, 1603–1615.
Reproduced by permission of Elsevier .
Figure 16.13 (parts) reproduced by permission of Cambridge
University Press from Kotula, Keenan & Michael 2003a. Figure 17.35 Zaluzec N (2012) The Confluence of Aberration
Correction, Spectroscopy and Multi-Dimensional Data Acqui-
Figure 16.18 reproduced by permission of Cambridge University sition. Proc. European Microscopy Congress 2012. Reproduced
Press from Kotula & Keenan 2006. by permission of the author.
 Figure and Table Credits 513

Figure 17.36 Specimen courtesy of Varela M, Lee HN at Oak

Ridge National Lab.

Figure 17.37 From Watanabe M (2013) Microscopy Hacks:

Development of Various Techniques to Assist Quantitative Na-
noanalysis and Advanced Electron Microscopy, Microscopy,
62, 217–241. Reproduced by permission of Oxford University
Press.
 Index

A Basis 410, 411, 420, 421 Chemical doping 29

A2 phase 85 Basis set 152 Child-Langmuir effect 26
Aberration Bend 287, 426 Chromatic aberration 3, 11, 12, 13, 20, 36,
coefficients 257 Bend contour 54, 123 211, 216, 219
correction 222, 256, 295, 300, 434 Berkovich-type indenter 53 Cliff 458
Absorption 11, 67, 150, 155, 157, 161, 163, Bessel 317 Close-packed directions 97
201, 216 Bethe 421 Confocal 350
Accelerating voltage 11, 19, 147, 149, 150, Bethe potentials 154 Continuum 479
151, 154, 155, 162, 210, 211 B-field 217 Converging 292, 293
Achromatic circles 160 Bloch 323, 324, 325, 326, 327, 329, 332, Convolution 185, 186, 202, 204, 213, 216,
Acicular morphology 87, 101 333, 334, 409, 425, 431 219, 244, 246, 268, 284, 305, 307,
Actuator 50, 52, 55, 56, 58, 76 Bloch wave 152 312, 313, 315, 329, 332, 333, 334,
AEM 469, 473 Boersch effect 3, 6, 7 345, 346
AFM-TEM holders 51, 58 Bohr 412 Corrosion resistance 87, 93
Airy 3, 292, 293, 294, 295, 298, 302, 316, Boltzmann constant 4 Cosinusoidal interference pattern 200, 205,
317, 325, 337 Bootstrap method 145, 153 206, 230
disc 292, 293, 294, 295, 298, 302, 316, Bosons 3, 207 Coulomb 2, 6, 7, 9, 13, 46, 225, 237, 314,
317, 325 Bragg 408, 409
function 3, 316, 317 angle 121, 146, 148, 149, 150, 154, 156, interaction 2, 6, 7, 9, 13, 46, 237, 314,
ALCHEMI 425, 426, 427, 428, 429, 430, 157, 158, 159, 160, 317, 322, 352, 409
431, 432, 434, 435, 436 356, 432 repulsion 409
systematic 426 beam intensities 146 Crack dependence 21
Alloy 54, 55, 81, 83, 85, 87, 93, 94, 96, 97, law 109, 154, 159 Critical 482, 488
101, 146, 159, 351, 356, 429, 435, reflections 109, 111, 112, 114, 117, 119, Crystallographic R-factor 158
489, 501 121, 122, 125, 128, 129, 132, 146, CuL 391
Aluminum 142, 211, 348, 384, 391, 392, 149, 150, 156, 158, 160, 162, 317, Cuprite 147, 152, 153
393, 450 356, 426
Amplitude contrast transfer function scattering 109, 122, 132, 146, 147, 148,
(ACTF) 204, 216 150, 154, 156, 157, 158, 159, 160, D
Amplitude normalization 221, 223 162, 322, 352 D0a-ordered orthorhombic structure 87
Angular current density 2, 5, 8, 10, 14 Bravais lattice 148 D03-ordered phase 85
Applied field 20, 56 Bright Dark 239, 268, 284, 314, 318, 320, 328,
Arbitrary waveform generator (AWG) 26 tomography 351, 354 356, 357, 358, 377, 440, 444, 446,
Argand 273, 274, 275 Bright field (BF) 70, 83, 84, 93, 105, 107, 449, 458
ASAC software package 154 112, 114, 115, 117, 118, 119, 123, Dark field (DF)
Atmospheric 475 126, 128, 129, 130, 132, 133, 134, holography 229
Automated 448 136, 137, 138, 142, 160, 174, 189 de Broglie wavelength 205
Automated acquisition 157 Brillouin 322 Deconvolution 162
Axial 429, 430, 431, 432, 436 Brownian motion 34, 35 filter 162
Azimuthally 298 Defect-free region 53, 132, 147
Delocalization 32, 36, 206
C Demagnification 10
B Carbon 4, 10, 17, 18, 21, 27, 33, 35, 36, 38, DeskTop 500
B2-ordered structure 85 42, 43, 45, 46, 54, 58, 60, 61, 63, Detective quantum efficiency (DQE) 162
B32-ordered phase 83, 85 70, 96, 97, 98, 162 Detector 381, 394, 468, 471, 472, 475, 476,
Back-etching 35 nanotubes (CNTs) 18 480, 492
Background 302, 310, 393, 468, 469, 471, thin films 43 Dielectric constant 7
480, 483, 485, 492 Cerenkov 398 Diffraction
Background intensity 132 Chemical 396 mode 136
Bagaryatski 97, 101 pattern 27, 33, 103, 129, 188, 218

515
516 Index

Diffractogram 188, 189, 218, 230, 311 golden 412, 418 162, 211, 213, 216, 224, 225, 229,
Diffuse 284, 334, 337, 432 level 9, 227, 407 230
Diffuse scattering 145, 146, 158, 159, 162 Fermions 3, 207 Image
Digital 384, 391, 462, 469, 472, 483 Ferritic Steels 96, 97 calibration 178, 179, 180
Digital Micrograph 22, 167, 168, 169, 170, Field-enhancement factor 5, 9 display 171, 173, 174, 175, 177, 178,
171, 172, 173, 174, 175, 176, 177, Field ion microscopy (FIM) 83 179, 180
178, 179, 180, 181, 183, 185, 187, Fienup (1982) algorithm 145, 161 filter 162, 185, 187, 189, 191
188, 189, 191, 192, 193, 194, 195 First-order lattice spacing 160 noise 72, 162, 189, 191
Dimensionless X-ray structure factors 151 First-order reflection 159 plane 177, 189, 219
Direct 267, 268, 271, 274, 276, 279, 280, Fraunhofer diffraction 161, 203 In 495
329 Fresnel 56, 202, 203, 205, 218, 236, 351 InAs 334
Discrete 348, 349 contrast 351 Incoherent 3, 82, 93, 94, 97, 101, 148, 163,
Disk of least confusion 12 fringes 202, 205 207, 208, 211, 212, 213, 216, 217,
Disordered A2 (bcc) structure 83, 85 integral 56, 202 219
Dopant distribution 227 propagator 202 Inelastic 36, 38, 64, 67, 132, 138, 149, 150,
Drift Full width at half maximum (FWHM) 3, 159, 163, 211, 212, 217, 223, 224,
correction 24, 446 11, 211 225, 226
Dwell times 69 interaction 211, 212
Infinite 312, 316
G Information loss 178
E Galvanostatic hold 41, 59, 63 Infrared imaging 35, 38
Eigenvalues 331, 453 Gaussian InGaN 226
Eikonal equation 217 broadening 485, 486, 490, 491 Interference
Einstein 322, 323, 326, 333 fit 468 fringe 160
scattering 322, 323, 326, 333 focus 11, 302, 309, 380 pattern 160, 207
Einstein model 159 image 248 Intermetallic 426, 428, 434, 482, 483, 501
scattering 159 image plane 11, 12 International Technology Roadmap for Semi-
Electrochemical impedance spectrosco- noise 153 conductors (ITRS) 153
py 41 orbitals 411 Inverse 468
Electrode materials 40, 59, 60 source 3, 11, 12, 14, 210, 305, 306, 307, Ion-beam deposition (IBD) 37
Electron 261, 378, 387, 427, 446, 472, 479, 334 Iron jump-ratio image 174
487, 488 Global minima 148 Isoplanatic 210
Electron dose 22, 26, 30, 46, 47, 60, 69, 70, Gold-plated titanium mirror 38
188, 224 Goodness-of-fit index 148
Electronic phase transitions 29 Gottfried Möllenstedt 213, 230 J
Electron irradiation 33, 59, 69, 70, 71 G-phase 98, 99 Jump-ratio image 174, 195
Electropolishing 100 Grain boundary, orientations 21
EMiSPEC 442 Grain growth 27, 93
Emittance 2 Green 412, 416, 417 K
Energy G-spacings 33 Kinematic interaction 217
calibration 468 Kirchhoff diffraction integral 197, 199, 201,
distribution 6, 152 202, 203
Ensemble coherence 212 H KLM 481, 482
Environmental gas cell holder 37 Hairpin metals 26 Klug’s approach 46, 75
Euler’s formula 187 Hamiltonian 273, 280, 407, 408, 409, 410,
Ewald 321, 322 416
Ewald sphere 88, 122, 146, 149, 150, 156 Heating holders 41, 42, 43, 44, 45, 65 L
Exchange 408, 409, 421 Heisenberg Uncertainty Principle 198, 206 LaAlO 327, 328
Excitation error 149, 150, 156, 157, 158 Helmholtz equation 198, 199, 205 LaB 379
Ex-situ analysis 20 Hf 302, 304 LaB6 electron source 27
Extinction 111, 112, 113, 119, 123, 127, Higher-order Laue zone (HOLZ) 105, 121, Landau 394
129, 140, 141, 142, 146, 147, 148, 124 Laser 4, 10, 19, 24, 25, 26, 27, 38, 51, 64,
151, 156, 158, 226 High-speed electrostatic deflector 26 65, 67, 207
Extractor potential 5 High temporal-resolution imaging 30 Laws of wave optics 198
Holtsmark regime 7 liquid-N2, 152, 162
Huygens’ principle 202, 213 Log of Modulus 177, 188
F Hysitron picoindenter 51 Longitudinal coherence 4
Fcc Cr-rich M23C6 carbide 93 Lorentz 7, 55, 56, 226, 227, 314, 462
Fcc NbC carbides 94 angular 314
FEFF 417 I force 55
Fermi 3, 9, 207, 227, 337, 407, 412, 414, Illumination 3, 19, 26, 27, 65, 127, 130, lens 55, 226, 227
418 131, 135, 147, 148, 149, 159, 160, mode 55, 227
low 449
 Index 517

Low-loss 35 Pauli 408 Q

Pb 486, 487, 490 Quantitative convergent-beam method
Peak 468, 480 (QCBED) 146
M Peak (X-ray characteristic) 67, 149, 158, Quantitative density map 151
Magnetic flux 55, 218, 230 171 Quasicrystals 288, 289, 327, 337
Mass-thickness contrast 63 Peltier 470
Mathematica™ 3, 24, 170, 175, 176, 183, Pencil-beam regime 7
185, 187, 188, 198, 200, 202, 210, Phase Contrast Transfer Function R
212, 213, 230 (PCTF) 204, 216, 218 Radial function 152
Matlab™ 24 Phase display 178 Radon 344, 345, 346, 347, 350, 373
Maxwell 418 Phonon 149, 156, 158, 162, 163, 211 Rate of evaporation 8
Mean-free path 36 Photoelectric effect 198 Rayleigh 293, 300
Microdiffraction 81, 82, 85, 100, 146, 148, Photo-emission effect 10 Reference wave 219, 220, 221, 222, 226,
159, 160, 163 Piezo 24, 49, 50, 51, 53, 56, 58, 59, 64, 65, 229
Microelectromechanical systems 104, 137, 226 Region-of-interest (ROI) 180, 181
(MEMS) 20, 29 Piezoelectric polarization 226 Relative 394, 470, 472, 473
Micromanipulators 50, 65 Pixon 331 Relativistic electron velocity 217
Minimum 485 Plane wave 148, 152, 162, 198, 199, 200, Relativity-corrected acceleration voltage 3
Möllenstedt biprism 205, 220, 222, 230 201, 202, 203, 205, 213, 217 Relrod 88, 123
Monochromators 3 Plasma cleaning 35 Residual coherent aberrations 216
Most 393 Plasmon 41, 150, 158, 211 Residual gas analyzer (RGA) 32, 39
Mott formula 145, 152 Point P 109, 147, 202, 208 Reversible processes 25, 27
Multipole expansion 148, 152 Point Q 203 RGB values 175
Multiva 456 Point spread function (PSF) 7, 11, 204, 213, Richardson equation 4
216 Riemann 346
Point-symmetric masks 185 Rietveld method 148, 163
N Poisson 145, 148, 151, 157, 255, 381, 386, Ring collapse 8
Nanoelectromechanical systems 56 452, 453, 454, 455, 456, 458, 477, Rocking curve 146, 147, 148, 156, 157, 163
Nanofactory STM holder 51 485 Ronchigrams 145, 154, 159, 160, 161, 163
Nanoindentation 20, 49, 50, 51, 52, 53, 54, equation 148, 157, 255, 453 Rose 485
55 noise 381, 452, 453, 454, 456, 458, 477, Rutherford 284, 314, 315, 332, 337
Nanorods 67, 127, 329, 462 485
AG 462 scaled 452, 456
growth of 67 statistics 145, 151, 381, 386, 452, 453, S
Nanoscale loops 158 455, 456, 485 Sampling Theorem 170
Neutron diffraction 148, 158, 163 weighted 452, 453 Saxton 362
Ni3Nb-type 87 Pole-piece 18, 19, 29, 30, 31, 32, 33, 36, 37, Scaling factor 174
NiAl 429 38, 43, 52, 53, 56, 65, 66, 67, 71, Scanning 350
NiOx 468, 469, 470, 471, 472, 483, 484, 73, 74, 227 Scattered intensity 151
494 Post mortem analysis 20, 49 Scherzer 294, 297, 319
NiOxIceC 472, 473 Potential distribution 219, 227, 228, 230 Schottky effect 4, 5, 13
NIST Standard Reference Data Base 156 Potentiostatic hold 41 emitter 13
Non 383, 384, 385, 391 Powder electron diffraction 156 Schrödinger 407, 408, 410, 411
Nylon gas tubes 39 Poynting vector 198 Scintillator 22, 32, 162, 222
Practical coherence width 3 Secondary Ion Mass Spectroscopy
Pre 393 (SIMS) 227
O Precession method 123, 145, 157, 158 Segmentation 347, 348, 362, 369, 370, 371,
Object transmission function 201, 203 Princip 331, 454 372
Ohmori 96, 98 Principal Components Analysis (PCA) 194 Selected area diffraction (SAD) 81, 111,
Omega 379, 380 Probe 123, 128, 141, 147, 223
Omega (Ω) filter 147, 149, 152, 154, 158, current 12 Shadow image 130, 131, 136, 154, 157,
162 size 11, 148 159, 160, 161
Optical imaging 160, 210 Probe-forming lens 12, 147, 148 Si 286, 288, 303, 308, 310, 312, 321, 326,
Optical pyrometry 41 Profile extraction tool 181, 182, 183, 193 443, 468, 475, 480
Orthogonal 55, 119, 207, 217 Projection 344, 346, 347, 373 Signal-to-noise (SNR) ratio 224, 225
Ostwald ripening 30, 39, 69 Protein monolayers 146 Silicon (111) 7x7 surface reconstruction 22,
Oxygen plasma cleaning 35 Protochips version 45 31, 33, 35, 36, 45, 60, 61, 75, 109,
‘Oxygen-sponge’ crystals 157 Pseudo-Contours 171 134, 136, 137, 141, 146, 149, 155,
Pseudopotential 410, 413 157
Ptychography 145, 159, 160, 162, 163 Simplex algorithm 145, 151
Pythagoras 202 Simultaneous 347
P
Pairwise 238, 239 SiN membranes 33, 34, 35, 38, 39, 40, 70
Paraxial rays 161

d7ec3178-fcbc-4a39-8eac-53c02d449f17
518 Index

Slater 415 Verwey transition 163

Sobel 383 Virtual source 2, 6, 7, 8, 9, 10, 11, 12, 13,
Solid angle 2, 13, 210 14, 55, 148, 159, 160
Space-charge effects 26, 27 Voxel 171
Sparrow 293
Spatial 238, 241, 242, 243, 244, 248, 249,
250, 252, 253, 254, 268, 270, 295, W
305, 318, 319, 320, 325, 344, 346 Warren 322, 337
Spatial coherence 4, 160, 205, 208, 210, Wave
211, 212, 213, 224 amplitude 200, 201, 229
Spatial frequency 162, 187, 188, 189, 195, modulation 201, 216
200, 201, 203, 205, 213, 216, 218, properties 223
219, 220 spherical 199
Spherical aberration 11, 12, 13, 132, 137, Wave transfer function 204, 216
159, 160, 161, 216, 218, 219, 227 Weight coefficient 150
Spherical charge density 152 Wet-etch methods 35
Spherical harmonic function 152 WIEN 411, 418, 421
Spinodal alloy 83, 101
Spinodal decomposition 83, 85, 98, 154
Static cells 40 X
Stereographic projection 100, 101, 113, 120 X-ray Bragg intensities 146, 157
Stobbs 334 X-ray crystallography 41, 113, 145, 146,
Strain 147
mapping 145, 153 X-ray energy dispersive spectroscopy
Superlattice 83, 84, 85, 87, 88, 101, 134 (XEDS) 35
reflection 88 spectroscopy 35
Surface oxide films 100

Y
T YAG scintillator 162
Ta 432 Young 292
Ta disk cathode 26 Young’s fringes 160
Temporal coherence 4, 211, 212, 219
Thermal-diffuse scattering 158, 159
Thermal energy 2 Z
Thermal expansion 38, 42, 43, 54 ZAF 354
Thermionic emitter 6, 7, 13, 25, 26 Z-contrast 160
Thin-film elastic relaxation 154 high-angle scattering 160
Three 387, 388, 389 Ze 321
Total 378, 387, 446, 479, 487 Zeolite structures 157
Total electron dose 47 Zernike phase 203, 218
Transmission 350, 354, 377, 378, 379, 400 Zero-field image-force potential 5
Transmission electron diffraction Zero-frequency 188
(TED) 146 Zero-loss 149, 223
Transmission electron microscopy Zero-order Laue zone 105, 107, 112, 113,
(TEM) 120, 122 114, 121, 123, 124, 126, 141, 150
Transverse 316, 317, 320 Zero scattering angle 151
Tungsten filament 43 Zincblende structure 222, 223
Type 329 duplex stainless steel 98 ZOLZ intensity distribution 153

U
Ultrafast electron microscopes (UEM) 10
Ultra high vacuum (UHV) 18, 36
Uncertainty principle 4, 153, 198, 206

V
Vacanc 429, 430, 432, 434
Vacancy 54, 69, 72, 158, 335
Vanadium-bearing steel 100
Varimax 455