Graduate Texts in Physics

Richard Osgood Jr.

Xiang Meng

Principles
of Photonic
Integrated
Circuits
Materials, Device Physics,
Guided Wave Design
Graduate Texts in Physics

Series Editors
Kurt H. Becker, NYU Polytechnic School of Engineering, Brooklyn, NY, USA
Jean-Marc Di Meglio, Matière et Systèmes Complexes, Bâtiment Condorcet,
Université Paris Diderot, Paris, France
Sadri Hassani, Department of Physics, Illinois State University, Normal, IL, USA
Morten Hjorth-Jensen, Department of Physics, Blindern, University of Oslo, Oslo,
Norway
Bill Munro, NTT Basic Research Laboratories, Atsugi, Japan
Richard Needs, Cavendish Laboratory, University of Cambridge, Cambridge, UK
William T. Rhodes, Department of Computer and Electrical Engineering and
Computer Science, Florida Atlantic University, Boca Raton, FL, USA
Susan Scott, Australian National University, Acton, Australia
H. Eugene Stanley, Center for Polymer Studies, Physics Department, Boston
University, Boston, MA, USA
Martin Stutzmann, Walter Schottky Institute, Technical University of Munich,
Garching, Germany
Andreas Wipf, Institute of Theoretical Physics, Friedrich-Schiller-University Jena,
Jena, Germany
Graduate Texts in Physics publishes core learning/teaching material for graduate- and
advanced-level undergraduate courses on topics of current and emerging fields within
physics, both pure and applied. These textbooks serve students at the MS- or
PhD-level and their instructors as comprehensive sources of principles, definitions,
derivations, experiments and applications (as relevant) for their mastery and teaching,
respectively. International in scope and relevance, the textbooks correspond to course
syllabi sufficiently to serve as required reading. Their didactic style, comprehensive-
ness and coverage of fundamental material also make them suitable as introductions
or references for scientists entering, or requiring timely knowledge of, a research field.

More information about this series at http://www.springer.com/series/8431

Richard Osgood Jr. Xiang Meng
•

Principles of Photonic
Integrated Circuits
Materials, Device Physics, Guided Wave
Design

123
Richard Osgood Jr. Xiang Meng
Columbia University Columbia University
New York, NY, USA New York, NY, USA
Boston University
Boston, USA

ISSN 1868-4513 ISSN 1868-4521 (electronic)

Graduate Texts in Physics
ISBN 978-3-030-65192-3 ISBN 978-3-030-65193-0 (eBook)
https://doi.org/10.1007/978-3-030-65193-0
© Springer Nature Switzerland AG 2021
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, computer 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, expressed or implied, with respect to the material contained
herein or for any errors or omissions that may have been made. The publisher remains neutral with regard
to jurisdictional claims in published maps and institutional affiliations.

This Springer imprint is published by the registered company Springer Nature Switzerland AG
The registered company address is: Gewerbestrasse 11, 6330 Cham, Switzerland
Preface

The field of integrated photonics has advanced rapidly in the last two decades due
to major advances in microfabrication, its design methods, its computer automation,
and the expansion of our understanding of the physics of small-scale optical
structures. Further, the growth in new commercial possibilities and actual devices
has driven these advances through the realities of commercial needs. The excite-
ment of this new technological world has led to the need to capture the field in a
broadly scoped and full textbook for graduate students and those seeking to learn
about this technology.
The materials from this book arose from a graduate course, which was taught for
many years at Columbia University by R. Osgood. The intent of this course was to
provide a full view of the overall field of integrated electro-optic devices. The
authors worked to have a course not welded to the latest materials technology but
rather to give a more pedagogical vehicle that showed how these devices may be
simulated and best operated. Because of their small size and complex design, the
course brings out unusual devices and new perspectives. Not unexpectedly the
course involves an extensive range of ideas and physics, particularly in
guided-wave optics.
The course builds not only on the most recent discoveries, and clearly has its
foundations and insights from earlier classic monographs in this field. These prior
classics include Tamir’s highly respected monograph Guided Waves
Opto-Electronics, the encompassing work on integrated opto-electronics, Optical
Integrated Circuits by Nishihara, Haruna, Suhara, etc. Other texts with much
broader pedagogical scope such as those by Bahaa and Teich, Yariv, or Haus are
drawn upon for insights into new or unusual devices or their physics.
The text is organized and arranged so as to first present or review the basic
electromagnetic fundamentals. It then shows how these electromagnetic “driver”
equations are modified by interfaces and boundaries. In addition, the treatment of
modal behavior and interaction via the coupled-mode equations and their modified
version in the presence of adiabatic evolution during propagation in material media
are then presented. These modified equations allow the initial formulation of
elementary device structures, particularly those of interest for guided-wave

v
vi Preface

phenomena to be developed and presented and their optical response examined.

This introductory section is then followed by understanding of certain device types
such as passive guided wave, i.e., waveguide crossings, Y-branches, bends, etc.,
grating-based devices, imaging devices, or switches and modulators. A major
unique contribution of this book has been the inclusion of a chapter and illustrations
of numerical methods for guided-wave problems. These methods are now the
standards in areas where analytical methods had previously dominated. The pre-
cision and power of numerical calculations now even allow determining precise
device behavior prior to fabrication.

Grateful Acknowledgments
Finally, this book has directly benefited from the input of Dr. Robert Scarmozzino
and Prof. Miguel Levy in sections of earlier drafts of the book. The authors of the
present text have used their substantial contributions and their encouragement in the
years, in which later drafts were completed. In a like manner, Profs. Panoiu, Steel,
and Dadap contributed a strong and exciting research environment, which gave new
ideas and thinking in the content of the text. The text has also been the product of
eager and hardworking graduate students, Zicong Huang, Rui Chen, Shijia Yan,
and Songli Wang, as they moved from MS to Ph.D. in their studies at Columbia and
helped in the preparation of the figures, proofreading, and scientific suggestions.
Note that the book also has been generously helped by comments and suggestions
by Prof. Paiella at Boston University, where R. Osgood spent the last several years
of its preparation, and Prof. Bergman, Lipson, Teich, Herman, Kymissis, and
Englund at Columbia and Prof. Willner of USC.
And finally it is a true joy to salute the grace and affection of Mrs. Alice Osgood
and the Meng Family for their support of the authors during their day-to-day hard
work. Without their work we would not have our tome.

New York, USA Richard Osgood Jr.

Xiang Meng
Contents

1 Integrated Optics: An Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . 1

1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 History of Integrated Optics . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.3 Integrated Optical Circuits: Classification and Advantages . . . . 3
1.4 The Rise of Si and Si-Based Systems . . . . . . . . . . . . . . . . . . . 5
1.5 Contents of This Book . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
1.6 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
2 Materials for PICs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.2 The SiO2-Based Materials . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.3 Polymers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
2.4 Single-Crystal LiNbO3 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
2.4.1 Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
2.4.2 Waveguide Formation . . . . . . . . . . . . . . . . . . . . . . . 19
2.5 GaAs/AlGaAs or InP/InGaAsP . . . . . . . . . . . . . . . . . . . . . . . 22
2.5.1 Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
2.5.2 Index Changes . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
2.6 Single-Crystal Si and Amorphous Si . . . . . . . . . . . . . . . . . . . 24
2.6.1 Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
2.6.2 Materials Properties of SOI . . . . . . . . . . . . . . . . . . . 25
2.6.3 Loss and Patterning . . . . . . . . . . . . . . . . . . . . . . . . 26
2.6.4 Doping-Induced Index Changes and Loss . . . . . . . . . 26
2.6.5 Temperature Variation of the Si Index
of Refraction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
2.6.6 Amorphous Silicon . . . . . . . . . . . . . . . . . . . . . . . . . 27
2.7 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29

vii
viii Contents

3 Dielectric Slab Waveguide . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31

3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
3.2 “Thin-Film” Waveguides . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
3.2.1 Slab Waveguide (2D) . . . . . . . . . . . . . . . . . . . . . . . 31
3.3 Slab Waveguide: Ray Optics Picture . . . . . . . . . . . . . . . . . . . 33
3.4 The Wave Equation for a Slab Waveguide . . . . . . . . . . . . . . . 34
3.4.1 The Wave Equation and Its Boundary
Conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
3.4.2 Effective Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
3.4.3 Polarization: Why TE and TM . . . . . . . . . . . . . . . . 36
3.4.4 TE Guided Modes . . . . . . . . . . . . . . . . . . . . . . . . . 37
3.4.5 TE Radiation Modes . . . . . . . . . . . . . . . . . . . . . . . . 47
3.5 Graded-Index Waveguides . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
3.5.1 Ray Optics Approach for Diffused Waveguides . . . . 49
3.5.2 Numerical Solutions . . . . . . . . . . . . . . . . . . . . . . . . 52
3.6 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
4 Three-Dimensional Waveguide . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
4.2 Types of Channel Guides . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
4.3 Modal Analysis of Three-Dimensional Waveguides . . . . . . . . . 59
4.4 Approximate Methods—Generally for Slab-Like
(Rectangular) Waveguides . . . . . . . . . . . . . . . . . . . ........ 61
4.4.1 Method of Field Shadows . . . . . . . . . . . . ........ 61
4.4.2 Effective-Index Method . . . . . . . . . . . . . . ........ 65
4.4.3 Accuracy of Approximate Techniques
for Channel Waveguides . . . . . . . . . . . . . . . . . . . . . 67
4.5 Wavegiude Dispersion for Channel Guides . . . . . . . . . . . . . . . 70
4.6 The Design of Single-Mode Channel Waveguides . . . . . . . . . . 71
4.7 Graded-Index Channel Waveguides . . . . . . . . . . . . . . . . . . . . 72
4.8 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
5 General Introduction to Coupled-Mode Theory . . . . . . . . . . . . . . . 79
5.1 Introduction: A Simple Example of Mode Coupling . . . . . . . . 79
5.2 General Coupled-Mode Formulation . . . . . . . . . . . . . . . . . . . . 80
5.3 Practical View of Coupling of Modes . . . . . . . . . . . . . . . . . . 82
5.3.1 Propagation Constants for the Coupled Modes . . . . . 83
5.3.2 Power in the Coupled Systems . . . . . . . . . . . . . . . . 84
5.4 Two Specific Examples of Coupling Coefficients . . . . . . . . . . 89
5.5 An Alternate Approach to Obtaining the Coupling
Coefficient . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ..... 91
Contents ix

5.6 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94
6 Optical Couplers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
6.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
6.2 The Standard Two-Waveguide Directional Coupler . . . . . . . . . 96
6.3 Transfer Matrix for a Coupler . . . . . . . . . . . . . . . . . . . . . . . . 98
6.4 More Exact Treatment of Waveguide Couplers . . . . . . . . . . . . 99
6.5 Three Examples of Actual Working Couplers . . . . . . . . . . . . . 101
6.5.1 Lateral LiNbO3 Couplers—A Study
of the Sensitivity to Fabrication . . . . . . . . . . . . . . . . 101
6.5.2 Lateral Si Wire Couplers . . . . . . . . . . . . . . . . . . . . . 104
6.5.3 Vertical Couplers . . . . . . . . . . . . . . . . . . . . . . . . . . 106
6.6 Adiabatic Couplers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106
6.6.1 The Differential Coupled-Mode Equations . . . . . . . . 108
6.6.2 Adiabatic Couplers—An Example . . . . . . . . . . . . . . 110
6.7 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115
7 Passive Waveguide Components . . . . . . . . . . . . . . . . . . . . . . . . . . . 117
7.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117
7.2 Simple Transitions in Waveguide Geometry . . . . . . . . . . . . . . 119
7.2.1 Reflection and Transmission at An Transition
Between Two Waveguides . . . . . . . . . . . . . . . . . . . 119
7.2.2 Modal Coupling in Waveguide Transitions . . . . . . . . 121
7.3 Y-Branches . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 124
7.3.1 Mode Overlap to Analyze Y-Branches . . . . . . . . . . . 125
7.3.2 Guided-Mode Coupling in Y-Branches: Mode
Splitters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 128
7.3.3 Advanced Y-Branch Design . . . . . . . . . . . . . . . . . . 131
7.3.4 Power Loss Due to Coherent Beam Combination
and Splitting in Single-Mode Waveguides . . . . . . . . 135
7.3.5 Example: Asymmetric Y-Branch for Mode
Sorting . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 135
7.4 Abrupt Angle Bends . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 137
7.5 Circular Bends . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 138
7.5.1 Conformal Transformation of Bends into Straight
Waveguides . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 138
7.5.2 Improved Design of Curved-to-Straight
Waveguide Junctions . . . . . . . . . . . . . . . . . . . . . . . 143
7.5.3 An Explicit Equation for Bend Loss . . . . . . . . . . . . 144
7.6 90 Waveguide Crossovers . . . . . . . . . . . . . . . . . . . . . . . . . . 144
7.7 Turning Mirrors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 145
x Contents

7.8 Transfer Matrices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 147

7.9 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 148
Appendix A . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 149
Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 149
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 150
8 Components for Polarization Control . . . . . . . . . . . . . . . . . . . . . . . 153
8.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 153
8.2 Polarizing Elements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 154
8.2.1 Thin Polarization Plates . . . . . . . . . . . . . . . . . . . . . . 154
8.2.2 Thin Absorbing Film on a Buried Waveguide . . . . . 156
8.3 Polarization Splitters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 160
8.4 Polarization Converters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 166
8.4.1 Passive, Plate-Type Converters Placed Normal
to the Waveguide Axis . . . . . . . . . . . . . . . . . . . . . . 167
8.4.2 Integrated Passive Waveguide Converters . . . . . . . . . 168
8.4.3 Active Polarization Converters . . . . . . . . . . . . . . . . . 170
8.5 Polarization Devices for Si Photonics . . . . . . . . . . . . . . . . . . . 172
8.6 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 173
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 174
9 Imaging Devices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 177
9.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 177
9.2 Multimode Interference Devices (MMIs) . . . . . . . . . . . . . . . . 177
9.2.1 Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 177
9.2.2 Self-imaging Principle and Mode Propagation
Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 178
9.2.3 Imaging Modalities: General, Restricted,
and Overlap Imaging . . . . . . . . . . . . . . . . . . . . . . . 180
9.2.4 Properties of MMIs . . . . . . . . . . . . . . . . . . . . . . . . . 186
9.2.5 Materials . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 189
9.2.6 Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 190
9.3 Star Couplers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 195
9.4 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 200
Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 200
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 201
10 Diffraction Gratings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 205
10.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 205
10.2 Collinear Coupling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 205
10.2.1 Contradirectional Waves of a Surface Grating . . . . . 207
10.2.2 Codirectional Coupling . . . . . . . . . . . . . . . . . . . . . . 209
10.3 Grating Dispersion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 210
10.4 Coplanar Coupling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 212
10.4.1 Coupling Coefficients for Gratings . . . . . . . . . . . . . . 214
Contents xi

10.5 High-Q Bragg Grating Structures . . . . . . . . . . . . . . . . . . . . . . 215

10.6 Applications of Bragg Gratings . . . . . . . . . . . . . . . . . . . . . . . 220
10.7 Bragg Diffraction and Raman–Nath Diffraction . . . . . . . . . . . . 221
10.8 Wavelength Selecting via Reection from a Grating
and Grating Fabrication . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 223
10.9 Gratings for Output or Input Coupling with
Guided-Wave Structures . . . . . . . . . . . . . . . . . . . . . . . . . . . . 227
10.10 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 231
Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 232
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 232
11 Wavelength Filtering and Manipulation . . . . . . . . . . . . . . . . . . . . . 233
11.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 233
11.2 General Classes of Wavelength Filtering
and Manipulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 234
11.3 Asymmetric Coupler Filters . . . . . . . . . . . . . . . . . . . . . . . . . . 235
11.4 Devices Based on Diffraction Gratings . . . . . . . . . . . . . . . . . . 241
11.4.1 In-Line Filters Based on Bragg Gratings . . . . . . . . . 241
11.4.2 A Bragg Filter for Channel Dropping . . . . . . . . . . . . 242
11.5 Waveguide-Grating Router . . . . . . . . . . . . . . . . . . . . . . . . . . 244
11.5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 244
11.5.2 Device Structure . . . . . . . . . . . . . . . . . . . . . . . . . . . 246
11.5.3 Basic Functions of the Waveguide Array . . . . . . . . . 246
11.5.4 Free Spectral Range . . . . . . . . . . . . . . . . . . . . . . . . 250
11.5.5 The Frequency-Dependent Transmission
Function of the Router, i.e., The PHASAR
Frequency Response . . . . . . . . . . . . . . . . . . . . . . . . 250
11.5.6 PHASAR Uniformity and Insertion Loss . . . . . . . . . 251
11.5.7 Channel Cross Talk . . . . . . . . . . . . . . . . . . . . . . . . 253
11.5.8 PHASAR Polarization Properties . . . . . . . . . . . . . . . 254
11.5.9 Design of a Multiplexer/Demultiplexer . . . . . . . . . . . 255
11.5.10 More Advanced Design Features . . . . . . . . . . . . . . . 256
11.5.11 MMI Phased-Array Wavelength
(De)multiplexer . . . . . . . . . . . . . . . . . . . . . . . . . . . 256
11.5.12 Applications of PHASARs . . . . . . . . . . . . . . . . . . . 257
11.6 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 258
Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 259
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 260
12 Electro-Optical Modulators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 263
12.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 263
12.2 An Overview of Electro-Optic Modulation . . . . . . . . . . . . . . . 264
xii Contents

12.2.1 Basics of Propagation of Lightwaves in Anisotropic

Crystals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 265
12.2.2 Index Ellipsoid . . . . . . . . . . . . . . . . . . . . . . . . . . . . 266
12.2.3 Dependence of the Electro-Optic Effect
on Crystal Symmetry . . . . . . . . . . . . . . . . . . . . . . . 268
12.3 Electro-Optic Modulation Illustrated for Two
Crystal Types . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 271
12.3.1 III–V (Zincblende) Crystals . . . . . . . . . . . . . . . . . . . 271
12.3.2 LiNbO3 Crystals . . . . . . . . . . . . . . . . . . . . . . . . . . . 273
12.4 Electro-Optic Modulators . . . . . . . . . . . . . . . . . . . . . . . . . . . . 275
12.4.1 Layout of a Basic Modulator . . . . . . . . . . . . . . . . . . 275
12.4.2 Modulator Types . . . . . . . . . . . . . . . . . . . . . . . . . . 277
12.4.3 Free-Carrier Modulations: Ultrasmall Si
Semiconductor Modulators . . . . . . . . . . . . . . . . . . . 288
12.5 Semiconductor Electro-Absorption Modulators . . . . . . . . . . . . 289
12.6 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 291
Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 292
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 292
13 Integrated Optical Switches . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 295
13.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 295
13.1.1 Electro-Optical Switches . . . . . . . . . . . . . . . . . . . . . 296
13.1.2 The Db Directional Coupler Switch . . . . . . . . . . . . . 296
13.1.3 Reversed Db Switch . . . . . . . . . . . . . . . . . . . . . . . . 301
13.2 Interference-Based Switches . . . . . . . . . . . . . . . . . . . . . . . . . . 303
13.3 Modal Interference Switches . . . . . . . . . . . . . . . . . . . . . . . . . 305
13.3.1 Intersecting Waveguide Structures . . . . . . . . . . . . . . 305
13.3.2 MMI Coupler Switches . . . . . . . . . . . . . . . . . . . . . . 306
13.3.3 Digital Optical Switch by Mode Sorting . . . . . . . . . . 307
13.4 Thermo-Optical Switches . . . . . . . . . . . . . . . . . . . . . . . . . . . . 309
13.5 Optical Switch Arrays . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 311
13.5.1 Electro-Optical Switching Arrays . . . . . . . . . . . . . . . 311
13.5.2 LiNbO3 Arrays . . . . . . . . . . . . . . . . . . . . . . . . . . . . 312
13.5.3 Si Arrays: Principally Mach–Zehnder-Based . . . . . . . 312
13.5.4 Thermo-Optical Switching Arrays . . . . . . . . . . . . . . 313
13.6 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 314
Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 314
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 315
14 Numerical Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 317
14.1 Numerical Methods in Nanophotonics . . . . . . . . . . . . . . . . . . 317
14.1.1 Finite-Difference Versus Finite-Element . . . . . . . . . . 317
14.1.2 Time-Domain Versus Frequency-Domain . . . . . . . . . 318
14.2 The Finite-Difference Time-Domain Method . . . . . . . . . . . . . . 319
Contents xiii

14.2.1 Yee’s Algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . 319

14.2.2 FDTD: 1D Example . . . . . . . . . . . . . . . . . . . . . . . . 320
14.2.3 Boundary Conditions . . . . . . . . . . . . . . . . . . . . . . . 323
14.3 Finite-Difference Frequency-Domain (FDFD) . . . . . . . . . . . . . 324
14.3.1 FDFD Using the Wave Equations . . . . . . . . . . . . . . 325
14.3.2 FDFD from Maxwell’s Equations . . . . . . . . . . . . . . 327
14.4 Beam Propagation Method . . . . . . . . . . . . . . . . . . . . . . . . . . . 328
14.4.1 Paraxial Formulation . . . . . . . . . . . . . . . . . . . . . . . . 328
14.4.2 Finite-Difference BPM . . . . . . . . . . . . . . . . . . . . . . 329
14.4.3 BPM Expansions of the Method . . . . . . . . . . . . . . . 330
14.5 Finite-Element Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 331
14.5.1 Boundary-Value Problems . . . . . . . . . . . . . . . . . . . . 331
14.5.2 Implementation of FEM . . . . . . . . . . . . . . . . . . . . . 332
14.6 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 333
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 334

Appendix: Problem Set Solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 337

Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 365
Chapter 1
Integrated Optics: An Overview

Abstract This chapter traces the history of integrated optical devices and gives an
overview of its supporting technology. It also explains why integrated optics is so
important in many areas of high-speed and high-density applications areas, which
grow in significance with modern information systems. The recent growth in silicon
photonics and its advances in fabrication technology is also emphasized.

1.1 Introduction

In the last two decades, the size and scale of photonics technology have been revo-
lutionized by the same miniaturization process that has transformed electronic sys-
tems from collections of bulky solid state devices to nearly atomic-size arrays of
transistors on single-crystal semiconductors. This development in photonics makes
it possible to fabricate or integrate relatively complex optical subsystems on a single
solid microchip. This book will focus on one form of integrated photonic systems,
namely, those which are called photonic integrated circuits.
Table 1.1 makes the advantages of full microchip integration clear by comparing
the optical components typically seen in three “formats” encountered in small-scale
commercial photonic systems. Note that the advantages of integrated optics include
greatly reduced size as scaled by the footprint of an optical waveguide, more robust
optical alignment, i.e., that carried out by the fabrication process itself, and the lower
cost made possible by the massive parallelism of planar processing. Of course, this
lower cost is only realized when a large number of fabrication runs of devices or
optical circuit are warranted.
The design of optical integrated circuits requires a fusion of design techniques
that draw on contributions from three fields: materials, optical, and electronic engi-
neering. Unlike much of the work on free-standing optical devices over the last few
decades, integrated guided-wave designs require careful consideration of physics
and performance of propagating optical waves in assessing device performance and
properties. In particular, the designer must plan a design scheme, which focuses on
the routing of optical waveguides and devices. Factors such as radiative loss, evanes-
cent coupling, optical absorption, etc., must all be anticipated in the design of this
© Springer Nature Switzerland AG 2021 1
R. Osgood jr. and X. Meng, Principles of Photonic Integrated Circuits,
Graduate Texts in Physics,
https://doi.org/10.1007/978-3-030-65193-0_1
2 1 Integrated Optics: An Overview

Table 1.1 Optical circuit technology

Technology Macro-optics Micro-optics Integrated optics
Components Lenses, mirrors LEDs, LD fibers PICs, single-mode LD
Alignment needed needed fixed in fabrication
Connection size cm mm µm
Electrode size cm mm µm (fast, compact)

analog circuit. As a result, the design procedure for these circuits is drastically differ-
ent from that of the usual digital integrated electronics circuits. Further, while design
methods are progressing at a very rapid pace in commercial venues, developing a
standard procedure and even a uniform choice of components for these designs is
still in a state of flux.

1.2 History of Integrated Optics

Integrated optics was first proposed in 1969 by Stuart Miller (then at Bell Labora-
tories), in the context of the rapidly growing field of integrated electronics and the
emerging interest in lightwave communications. This field grew sufficiently rapidly
such that by 1972 the first conference in integrated photonics was held: The Con-
ference on Integrated and Guided-Wave Optics. Since then, integrated optics has
expanded into a field with not only a vigorous research program, but also with com-
mercial markets in communications, cable television, fiber-optic sensors, and radar
control. Initially, these applications used optics with single or small arrays of devices,
but more recent interests in large integrated subsystems have arisen. This chapter is
intended to provide an understanding of the physics and methodology of design for
one class of optical systems that involving dominantly optical functions on a single
chip.
As in integrated electronic design, integrated optical systems have evolved in the
complexity of their design. At the same time, improvements in fabrication technology
have led to a steady increase in simplicity or uniformity in the choice of the basic
materials assembly for the optical chip. This progression caused “integrated” systems
to move from an assembly of diverse materials bonded together, generally by manual
alignment methods, to the present state, in which chips are fabricated by a succession
of large-scale planar processes on a single substrate chip. This evolution is illustrated
in Fig. 1.1, which shows the photonic chips by IBM as they were fabricated after many
years of development of an experimental single platform for optical systems.
1.3 Integrated Optical Circuits: Classification and Advantages 3

Fig. 1.1 IBM demonstrated its first fully integrated wavelength multiplexed silicon pho-
tonics chip. Source https://arstechnica.com/informationtechnology/2015/05/ibmdemos-first-fully-
integratedmonolithic-siliconphotonics-chip/

1.3 Integrated Optical Circuits: Classification and

Advantages

Integrated optical “Chips” have been used for various types of optical systems and
have appeared in several different basic forms, each reflecting their intended appli-
cation. These varying applications have been given several different names or classi-
fications. The first form is basically a standard digital silicon or III-V IC with some
provision for the detection or emission of light. Such a circuit might have, as its opti-
cal function, simply an onboard detector for receiving an optical data stream from
a fiber-optic input. The second type, the opto-electronic integrated circuit (OEIC),
is more clearly a photonic or optical circuit. This type usually involves significant
electronic and optical components and circuits, and typically has important hybrid
mounting methods. A good example is a fiber-optical transceiver, which contains a
laser transmitter, its drive circuitry, a detector, and an amplifier. The integration level
in OEICs has increased drastically over the last few decades; see Fig. 1.2.

Fig. 1.2 The dramatic

change of integration level in
OEICs in the last few years.
The well-known Moore’s
Law is readily seen in these
plots in this figure
4 1 Integrated Optics: An Overview

Fig. 1.3 The cross section of a waveguide showing an example of a typical size of the guided-wave
elements in PICs

Finally, the concept of a photonic integrated circuit (PIC) has been developed over
the last few decades. This circuit is in essence a miniaturized fully photonic system
that has been confined to a planar geometry, thus making copious use of waveguides
and two-dimensional optical elements. For PICs, the design is dominantly concerned
with photonics functionality, since the number of electronic elements is small. Typi-
cally, most of these “circuits” use guided-wave elements, with waveguides acting as
interconnections or “optical wiring,” as well as being integral parts of both simple
and complex devices, such as optical couplers and phase-delay routers, respectively.
The transverse scale size of this “wiring” is small, e.g., 0.5 µm, but not as small as
the wiring on electronic IC (see Fig. 1.3).
In this book, we will be concerned mainly with the design and understanding of
PICs and to a lesser extent the optical portion of OEICs. Thus far, work in PICs has
emphasized passive devices and circuits; this will be our emphasis as well, although
we also present extensive discussions of modulators and switches. An example of
such a PIC is shown in Fig. 1.4.
In terms of engineering integrated optical systems, PICs offer distinct advantages
over hybrid systems, many of which are found only in part in other forms of optical
systems. The first is an extremely broad spectral bandwidth, which can encompass a
large number of optical channels. In practice, this means that many parallel channels
are available for use, each coded at a separate wavelength. This wavelength paral-
lelism can be used for a variety of network or sensor applications. A second advantage
is found in the interfacing of many optical systems in an inexpensive robust manner.
In this case, retaining the optical signal on a signal chip greatly reduces the cost
of alignment and the optical loss for a hybrid system on a Si optical bench. It also
provides greater thermal and mechanical stability than hybrid systems. These advan-
tages are only examples; other advantages are available as well, including easier
access to high modulation rates, more facile interconnect geometries, etc.
1.4 The Rise of Si and Si-Based Systems 5

Laser
Input coupler
Optical ring
resonator
Optical modulator

Photo diode
Optical waveguide

Couplers

Photonic crystal
Optical fiber

Fig. 1.4 A sketch showing the schematic of a photonic integrated circuit (PIC) (not to scale),
showcasing different optical components

1.4 The Rise of Si and Si-Based Systems

Much of the challenge of realizing integrated optical systems is found in the diffi-
culty of selecting a single universally effective material. For example, LiNbO3 or
other metal oxide optical materials make excellent electro-optical switching materi-
als but suffer from a low refractive index and, hence, large characteristic bend radius.
Other materials such as certain III-V crystals are superb as emitters and some passive
devices but are relatively expensive. However, in trying to fix on the most desirable
material, the Si-based materials system allows new possibilities for extensive integra-
tion of various materials (Soref 1993). After all, Si is the basis for the most successful
and mature integrated systems, integrated electronics! In fact, it is reasonable to con-
template manufacture of photonic devices from the standard fabrication processes
at a silicon-chip foundry. The recent major improvement in Si-based opto-electronic
devices makes it worthwhile to focus this chapter on this specific system.
With these factors in mind, as well as the growing interest in using Si, like a
miniature optical bench for glass waveguides (Henry et al. 1989) and other discrete
components, it was suggested, in the 1980s by Soref (1993) and by others, such
as Reed (2008), Jalali and Fathpour (2011) to employ guided-wave optical systems,
which use patterned Si wafers as the light guiding material. Figure 1.5 shows a sketch
of Soref’s original concept for Si integrated photonic circuits. Much of the initial
device work in this area focused heavily on the use of CMOS wafers (Jalali and
Fathpour 2011), which were designed to have waveguides that were weakly guid-
ing. Interestingly this early work did explore many integration issues that are still
of interest and considered a key device concept such as free-carrier modulators, etc.
Subsequent to this important pioneering work, approaches to build high confinement
single-mode waveguides in Si were conceived and developed (Vlasov and McNab
6 1 Integrated Optics: An Overview

Optical amplifier
Optical fiber
Directional coupler

Laser diode
BiCMOS IC
HBT, HEMT or
MODFET

Optical waveguide

2 2 switch

Electrooptic modulator
Integral V-groove
Photodiode

Fig. 1.5 A sketch of Soref’s original concept for Si integrate photonic circuits

Fig. 1.6 An optical microscope image of PIC by PLCC showing the remarkably high-quality
patterning on a silicon on insulator substrates (SOI) chip. Source https://www.extremetech.com/
extreme/201163-ibm-to-demonstrate-first-on-package-silicon-photonics

2004). The most effective material for these devices was SOI (silicon-on-insulator)
substrates, which were available in large-scale planar wafers and for which the meth-
ods for remarkably high-quality patterning had been developed. (see Fig. 1.6)
The most striking advance in Si photonics, however, occurred from the realization
that guiding light in a pure silicon device layer made a very distinctive and potentially
powerful approach to optical integration. This Si-base layer, which has a very high
refractive index contrast, which enables compact bend waveguide systems, would
make extremely small down-scaled photonic systems possible. In addition, while the
array of allowed Si functionality has still certainly not yet been fully realized, it is
1.4 The Rise of Si and Si-Based Systems 7

possible to fabricate a wide array of active devices, including low-power modulators,

high-sensitivity photodetectors, and high-efficiency switches (thermo and electro-
optic). SOI is also an unsurpassed semiconductor material for patterning since it
builds on the massive infrastructure that is in place for nanometer patterning of Si
electronics. The degree of this infrastructure includes the full capability needed for
the manufacturing of microsystems (Chrostowski and Hochberg 2015). Finally, the
spatial resolution of Si patterning is such that spurious roughness is now no longer
an issue.
The excitement of realizing photonics integration based on Si photonics has had a
profound effect on the direction of integrated optics. The first of these directions is the
development of radically new devices, which are enabled by the unique properties of
Si (Xu and Lipson 2006) and its related Group IV material Ge (Liu et al. 2010). These
advances include ultralow energy-per-bit modulators, typically based on micro-ring
resonators, on-chip doped-Si or pure Ge detectors, new approaches to integration,
and precisely defined, efficient grating couplers (Reed 2008); these devices will be
discussed later in the book. The second area of recent advances is in achieving greater
spectral reach for integrated devices and systems. Research is now emerging, which
allows integration at Mid-IR wavelengths for sensing applications and in the near IR
for quantum-computational applications. Finally, major growth is occurring in the
area of large-scale integrated applications such as data systems or multi-wavelength
transceivers (Shen et al. 2019). These systems are characterized by high speed and
orthogonalities in wavelength, mode, and polarization.
The broad goal of one key application of Si photonics is to use the high bandwidth
of optical technology for transmitting large amounts of data between computers,
servers, and data centers. By integrating Si photonic devices into these systems, the
costs of fabrication and electronic power consumption can then be driven down, and
thus inexpensive, highly functional systems can be realized.
Silicon photonics can also improve the response time and allow ultrahigh-speed
data processing. For example, one commercial vendor recently described a photonics
chip, which employs four wavelengths in a single waveguide for data transfer. This
system will enable “digitally sharing of six million images, or downloading an entire
high-definition digital movie in just two seconds.” (Green 2015) In this chip, for
example, each wavelength provides an independent 25 Gb/s optical channel. Thus
for an entire transceiver chip, the four wavelength channels would be multiplexed
so as to obtain a total 100 Gb/s bandwidth in one Si wire waveguide. If the “wire” is
designed to be single-mode, the bandwidth-distance product is no longer limited as
it is in multimode links.
It is important to realize that the use of standard-thickness SOI makes it difficult
to pattern and fabricate nanoelectronic and photonic systems on a single chip; see
discussion in Sun et al. (2015). One solution to this problem is to place photonics
and electronics on separate chips. This solution, while powerful, presents a fabrica-
tion issue in connecting between the two chips. Thus recently methods have been
demonstrated to form a single-chip solution, which enabled the coplanar method for
both electronic and photonic functionality, and methods have now been devised (Sun
et al. 2015).
8 1 Integrated Optics: An Overview

Finally, it is important to conclude this section with a cautionary note to point out
that, as time has progressed, the emphasis on specific choices of photonic materials
has shifted steadily. At present, Si appears to be an ideal solution for the most pressing
of the current photonics system needs. However, a careful look at the commercial
marketplace shows that glass, LiNbO3 , and III-Vs continue to be crucial for use in
integrated optics. Thus to take the broadest of overviews, this book will consider
photonic devices and systems based on each of these materials.

1.5 Contents of This Book

The goal of this book is to provide a summary of the essential devices and device
concepts for photonic integrated circuits. Thus the first chapter lays out the properties
and fabrication of major materials systems and the waveguide structures needed
for these PICs. This book also targets achieving an understanding of the physics
and analytic solutions to waveguiding structures. It is felt that presenting analytic
solutions gives a deeper understanding of the operation principles of these structures.
These analytic methods include a detailed discussion of coupled-mode theory since
this perturbation-based method enables analytical analysis of many guided-wave
structures. However, the advances in computational or numerical design methods
and the computational power itself have been so substantial that they can add new
design tools themselves and add capabilities such as highly accurate imaging. In all
cases, we display many illustrated examples obtained from the numerical methods
for an intuitive understanding of how the devices work. In addition, a review of
the numerical methods used for photonic device and PIC design or simulation is
presented in the text as well.
The book also includes examples of many basic devices. These devices include
grating- and coupler-mediated devices, and adiabatic designs of the devices. In addi-
tion, devices for wavelength filtering and mux/demux are discussed and shown. The
book is presented in a format, which may lend itself to a graduate class in integrated
photonics or to a review for a working-level engineer or applied scientist.

1.6 Summary

The basic introduction and motivation for integrated optical circuits have been pre-
sented in this chapter. Clearly the ultimate driver of any commercial technology is
the nature and extent of its marketplace, and this varies with time. But the basic
argument for PICs is that integration reduces cost by allowing parallel processing
of all devices on the wafer surface and by allowing alignment of the optical system
with the lithography pattern tool.
Finally, note that parallel growth has occurred in the interest in potential devices for
nonlinear-optical integrated photonic devices, particularly in the area of Si photonics
References 9

(Osgood et al. 2009). However, the focus of this book is on linear photonics, in part
due to its more immediate commercial interest. The reader is referred to several
recent reviews on the subject of integrated nonlinear optics (Atabaki et al. 2018).

References

Atabaki, A. H., Moazeni, S., Pavanello, F., Gevorgyan, H., Notaros, J., Alloatti, L., et al. (2018).
Integrating photonics with silicon nanoelectronics for the next generation of systems on a chip.
Nature, 556(7701), 349.
Chrostowski, L., & Hochberg, M. (2015). Silicon photonics design: From devices to systems. Cam-
bridge: Cambridge University Press.
Green, W. (2015). Silicon photonics: The future of high-speed data. IBM Research Blog.
Henry, C. H., Blonder, G., & Kazarinov, R. (1989). Glass waveguides on silicon for hybrid optical
packaging. Journal of Lightwave Technology, 7(10), 1530–1539.
Jalali, B., & Fathpour, S. (2011). Silicon photonics for telecommunications and biomedical appli-
cations. Boca Raton: CRC.
Liu, J., Sun, X., Camacho-Aguilera, R., Kimerling, L. C., & Michel, J. (2010). Ge-on-Si laser
operating at room temperature. Optics Letters, 35(5), 679–681.
Osgood, R., Panoiu, N., Dadap, J., Liu, X., Chen, X., Hsieh, I.-W., et al. (2009). Engineering
nonlinearities in nanoscale optical systems: Physics and applications in dispersion-engineered
silicon nanophotonic wires. Advances in Optics and Photonics, 1(1), 162–235.
Reed, G. T. (2008). Silicon photonics: The state of the art. Hoboken: Wiley.
Shen, Y., Meng, X., Cheng, Q., Rumley, S., Abrams, N., Gazman, A., et al. (2019). Silicon photonics
for extreme scale systems. Journal of Lightwave Technology, 37(2), 245–259.
Soref, R. A. (1993). Silicon-based optoelectronics. Proceedings of the IEEE, 81(12), 1687–1706.
Sun, C., Wade, M. T., Lee, Y., Orcutt, J. S., Alloatti, L., Georgas, M. S., et al. (2015). Single-chip
microprocessor that communicates directly using light. Nature, 528(7583), 534.
Vlasov, Y. A., & McNab, S. J. (2004). Losses in single-mode silicon-on-insulator strip waveguides
and bends. Optics Express, 12(8), 1622–1631.
Xu, Q., & Lipson, M. (2006). Carrier-induced optical bistability in silicon ring resonators. Optics
Letters, 31(3), 341–343.
Chapter 2
Materials for PICs

Abstract The materials technology for integrated optics is complex and thus this
chapter examines the optical properties, the electronic properties, the fabrication, and
pattering of a set of the most common materials. These include polymers, crystalline
silicon, epitaxial three-five thin films, and doped titanium dioxide. In the discussion,
the relative importance of each material is described.

2.1 Introduction

Because the development of PICs is still evolving, the materials technology for
photonic circuits also remains in a state of flux. Thus, a wide variety of new materials
and materials combinations are still being used commercially and there is still major
research into a wide variety of materials technologies. Many of these materials are
ideally suited for a specific photonics application or for realizing one functionality but
then fall short in other areas. Thus the development of a generally accepted and used
materials system, such as is found in the case of SiO2 /Si for CMOS electronics has not
occurred; research into Si photonics has become an increasingly focused choice for
important areas of data communications. In fact, in some sense, the state of materials
choice is even more uncertain than for the comparatively well-developed technology
for microwave circuits, which rely on four different materials technologies: Au/GaAs,
GaAs/AlGaAs, Si:Ge, and SiO2 /Si.
Nevertheless, within the last few years, most commercial PICs have generally
used five basic materials systems: SiO2 ,Si, III-V’s (InP- or GaAs-based), polymers,
and LiNbO3 . As a result this chapter will contain a discussion of each of these five
materials areas.
An overall summary of these five materials systems is contained in Table 2.1. The
table presents data on four characteristics: typical values of n and n, absorption
coefficient α, and the typical planar dimension of the substrate of the five materials
considered here. Note that in two cases (III-Vs and SiO2 /Si), two particular classes of
the materials systems are discussed. Finally, the optical response, which is achievable
with each material system and wavelength is also listed.

© Springer Nature Switzerland AG 2021 11

R. Osgood jr. and X. Meng, Principles of Photonic Integrated Circuits,
Graduate Texts in Physics,
https://doi.org/10.1007/978-3-030-65193-0_2
12 2 Materials for PICs

Table 2.1 Commonly used PIC materials systems and their approximate optical properties and
available planar dimensions
Material Refractive Index contrast Absorption Typical Active optical
index n n α/cm−1 available response
dimension
l/cm
SiO2 /Si 1.46 (P-glass 7×10−3 0.05 ∼10 Thermo-
1.3 µm) electric
modulation
SiO2 /Si3 N4 /Si 1.97 (1.3 µm) 2–4×10−2 0.3 ∼10 Optical
amplification
(with Er
doping)
LiNbO3 2.2(n e ) 10−2 5 × 10−3 ∼2 × 6 Electro-optic
2.24(n o ) modulation
Polymer 1.49 2 × 10−3 – 0.02/0.1 ∼30 Modulation,
10−4 (1.3/1.5 µm) some
emission
GaAs/AlGaAs ∼3.6 5 × 10−2 2 × 10−2 (at 5 Modulation,
n ∼ 1016 cm−3 ) Optical
sources,
Nonlinear
response
InP/GaInAsP ∼3.2 5 × 10−2 2 × 10−2 (at 5
n ∼ 1016 cm−3 )

The reason for the particular choice of properties, presented in Table 2.1, is that
these parameters control the physical scaling of the PIC. The index difference
between the cladding and waveguide core, n, controls modal confinement in a
waveguide. This confinement, in turn, establishes the widths of the waveguide mode
as well as the minimum radius of bends in the waveguide that are achievable without
significant radiative loss. The linear absorption coefficient, α, determines the inser-
tion loss in the waveguide for a given length. Hence a large α will rule out fabrication
of PICs with large linear dimensions. Finally the availability of large-area substrates
of polymer and SiO2 enable large-area circuits to be made of these materials in
comparison to the much smaller size possible with Si or III-V materials.
Some of the optical properties in Table 2.1 are also important for the interfacing
of PICs with an optical fiber link. Specifically, the absolute value of n controls the
magnitude of Fresnel loss, i.e., that due to reflection, at the chip facets. The refractive
index is also important, along with the change in core/cladding refractive index, n,
in determining the mode matching between the fiber and the integrated- waveguide
mode. The radiative loss, which is encountered in going on and off chip due to
imperfect mode matching, can be substantial and is thus an extremely important
quality in designing a practical PIC system. This radiative, mode-mismatch loss is
obviously the same, regardless of chip size. These interfaces will be described more
qualitatively in Chap. 7.
2.2 The SiO2 -Based Materials 13

2.2 The SiO2 -Based Materials

An important passive PIC technology has been developed using the SiO2 -based
materials system on a Si platform; this technology is sometimes called the Si optical
bench technology (Henry et al. 1989). These “bench” materials are grown either by
metal organic chemical vapor deposition (MOCVD), flame hydrolysis (also used for
optical fibers), or by high pressure oxidation of a silicon substrate. Si3 N4 layers can
also be grown by low pressure chemical deposition (LPCVD). This technology is
attractive because it is, in principle, compatible with many phases of Si IC man-
ufacture. This compatibility is due to the fact that it makes use of silicon wafers
and many of the chemical processing steps used in integrated circuit manufacturing.
Despite the importance of this Si-based technology for passive devices, the absence
of an electro-optical response and light-emission capability means that active Si-
based optical circuits, except those based on thermo-optic control, are not possible.
Of course in certain cases thermo-optical modulation, while relatively slow, does
permit switching/modulation operations.
SiO2 -based PICs have a number of important advantages. First, this PIC is
mounted on a Si platform and, thus, pig tailing of the fiber to the dielectric layer
may be realized via Si V-groove technology, which is based on precise crystallo-
graphic etching (see Fig. 2.1). This capability makes optical packaging, which is
generally a costly procedure, somewhat more straightforward. Second, because of
the wide availability of Si wafers, the technology is in principle relatively inexpen-
sive. In addition, the high thermal conductivity of the silicon substrate means that it is
a practical platform for mounting active devices, such as laser diodes, which impose
significant cooling requirements. Finally, the refractive index of this system is close
to that of the fiber, making efficient coupling between fiber and integrated waveguide
modes possible. The core of this waveguide is ∼4 µm × 7 µm (height and width)
for a 1.55 µm wavelength system. This good mode matching has thus been used
for a number of high-quality commercial components such as star couplers, filters,
routers, and splitters.
In the remainder of this section, we will focus on SiO2 on Si (silicon optical bench
used initially by Bell Labs). In this system, there are three SiO2 -based materials sys-
tems of interest: undoped SiO2 , SiO2 :P (phosphorous doped silica), and Si3 N4 . These
systems can be used to make the two types of waveguides shown in Fig. 2.2. In both
types of guides: SiO2 or low P-doped, the SiO2 film functions as the cladding layer.
In the case of SiO2 :P, the variable content of phosphorus dopant atoms can raise
the index above that of the undoped SiO2 . This system (see Fig. 2.2) has indices
which are most compatible with fiber I/O’s. The Si3 N4 film has a much higher index
(∼1.97), and typically can be used in circumstances where greater mode confine-
ment is required. An excellent example of its use is in mode matching a waveguide
interconnect to a semiconductor diode.
The properties of these films determine the comparative advantages of SiO2 for
waveguide materials. A typical refractive index step between the waveguide core
and cladding is n = 0.004 (see Table 2.1). The absolute indices are those of glass
14 2 Materials for PICs

SiO2
54.74°

<100>
Si
<111>

Fig. 2.1 Schematic representation of a v-groove

Fig. 2.2 Two typical waveguides using SiO2 -based materials. Left: SiO2 :P; right: Si3 N4

except that prior to annealing, films can have a lower density than for typical glass.
With such relatively small values of n, it is necessary to decouple the deposited thin
film from the substrate using a ∼15 µm layer. Once decoupled, waveguide losses
are small, i.e.,, <0.05 dB/cm. The residual stress anisotropy in optical waveguides
causes a small but important degree of birefringence in the deposited waveguide of
∼5 × 10−4 for the TE versus the TM polarization, and where the TM polarization
has a higher P-doped index. The nitride-core waveguides have higher index (∼1.97)
and loss (∼0.3 cm−1 ).
These glass layers are grown for waveguides in several ways. First, high pressure
thermal oxidation in a steam ambient is used to form thick optical SiO2 buffer layers
on the Si wafer. This process is sometimes called HiPOx. If P-doped layers are
desired, they can be grown from silicon and oxygen molecular precursors, with a
phosphine dopant source, using LPCVD. A second approach, which yields a more
conformal film, is based on LPCVD using tetroethylorthosilicone and ammonia.
This film is called TEOS or, when doped with phosphorus from phosphine, P-TEOS.
Flame hydrolysis has also been used to grow P-doped SiO2 films. Annealing at higher
temperatures can be used to relieve stress in the film. Finally, use of various nitrogen
precursors in a LPCVD system will deposit a silicon nitride film. These films are
limited by stress to less than 1500 Å in thickness.
2.2 The SiO2 -Based Materials 15

Because of the commercial importance of SiO2 structures, there have been exten-
sive measurements of the optical properties of this materials system. A representative
sample is provided in Fig. 2.2, which shows the measured index versus wavelength
for different P concentration and for undoped SiO2 . Clearly, P-doping does increase
the background index; however, note that the index contrast, n, between the var-
ious glasses remains nearly constant over the wavelength range. Also note that for
undoped SiO2 deposited using TEOS, annealing is needed to cause the index rise to
reach the background index. Annealing changes the index due to glass densification.

2.3 Polymers

Polymers are a particularly interesting materials system for PICs because of the fact
that large-area sheets of polymers can be prepared at very low cost. These thin sheets
can then be readily patterned, by a variety of techniques, to form integrated optical
circuits. At present, the use of polymers is well established for many different passive
devices, such as filters, splitters, and star couplers. (Eldada and Shacklette 2000). In
addition, polymers can, in principle, be a host for electro-optic functionality, as well
as light emitters; this capability enables a nearly fully functional polymeric PIC to be
made. In practice, however, the use of polymers for active components in PICs has
been limited by thermal degradation, or the loss of the poling. This last issue means
that it is very difficult to make a long-term stable electro-optic modulator since loss
of poling eliminates the electro-optical coefficient.
In addition to research work, polymers are also finding wider use in many commer-
cial large low-cost planar optical elements. For example, robust low-cost polymers
have been used in the devices in aircraft fiber-sensor systems. In fact, many of these
materials are being inserted commercially in various approaches to low-cost highly
multimode optical interconnects.
A variety of polymeric materials have been utilized for these optical waveguide
devices. The polymer classes include polyimides, olefins, and acrylates, including
halogenated versions of these building-block molecules. These materials are typi-
cally applied in liquid form as monomers, polymers, or oligomers in solution. The
materials can then be formed into thin sheets by cross-linking, either by exposure
to light, thermal treatment, or desolvation in an oven, respectively. A list of some
of the typical commercial polymeric waveguide materials is given in Table 2.2. The
refractive indices of the materials listed in the table can range from 1.30 to 1.85.
16 2 Materials for PICs

Table 2.2 Examples of polymers used in passive integrated optical devices

Manufacturer Polymer type or trade Patterning technique Optical loss, dB/cm (at
name wavelength, nm)
NIT Halogenated Acrylate RIE 0.02 [830]
0.07 [1310]
1.7 [1550]
NIT Deuterated RIE 0.17 [1310]
Polysiloxane 0.43 [1550]
Amoco Fluorinated Polyimide Photoexposure/wet 0.4 [1300]
[UltradelTM ] etch 1.0 [1550]
DuPont Acrylate Photolocking 0.18 [800]
[PolyguideTM ] 0.2 [1300]
0.6 [1550]
Dow Chemical Benzocyclobutene RIE 0.8 [1300]
[CycloteneTM ] 1.5 [1550]
Dow Chemical Perfluorocyclobutene Photoexposure/wet 0.25 [1300]
[XU 35121] etch 0.25 [1550]
JDS Uniphase [BeamBoxTM ] RIE 0.6 [1550]
Photonics
Allied Signal Acrylate Photoexposure/wet 0.02 [840]
etch, RIE, laser 0.2 [1300]
ablation 0.5 [1550]
Allied Signal Halogenated Acrylate Photoexposure/wet <0.01 [840]
etch, RIE, laser 0.03 [1300]
ablation 0.07 [1550]

By mixing these materials on a single-material substrate such as a glass sheet, it is

possible to “integrate” a wide range of different index polymers. However, chemical
and physical incompatibilities of the different polymer units can set practical limits
on the number of choices for mixing.
Examination of the table shows that a major issue in designing a high-quality
optical polymer is the presence of undesired optical absorption. (Eldada and Shack-
lette 2000) This absorption is generally more significant at infrared than at visible
wavelengths. Figure 2.3 shows that absorption loss in polyimide, a common proto-
typical polymeric material. The calculated points in the figure, which are connected
to give an overall sense of the absorption envelope, show that the absorption maxi-
mum can be accounted for simply by counting the number of vibrational units times
their individual absorption strength. The figure also shows that replacing the polymer
hydrogen with increasingly heavier species, i.e., deuterium, fluorine, etc., shifts the
absorption edge further into the infrared and, hence, lowers absorption in the visible
and near-infrared regions. This strategy has led to low loss in commercial polymers
for integrated optics.
A variety of other materials properties are important for optical applications,
including detection of thermal aging and humidity sensitivity. First, partial thermal
2.3 Polymers 17

Fig. 2.3 Optical loss spectra of single-mode waveguides prepared by perfluorinated polyimide and
partially fluorinated polyimide [S. Ando, 2004]

decomposition of polymers is a particularly severe problem for use at visible or

∼800 nm wavelengths. It can be reduced to negligible amounts at these wavelengths
by using heavily cross-linked polymers. In such a material, however, aging is not an
issue for 1.3 and 1.5 µm wavelengths. Second, halogenation and cross-linking also
have improved the humidity resistance of these polymers to the point where it is less
of an issue.
Similarly, the optical properties of these polymers are now well categorized and
are sufficiently favorable to be useful for a variety of optical applications. For
example, refractive index dispersion is important for many multiple-wavelength-
use devices (Eldada et al. 2000). The material dispersion in a well-designed poly-
mer is roughly comparable to (albeit somewhat higher than) SiO2 . In the case of
halogenated polymers, material birefringence is <1 × 10−6 . This small value is a
result of the non-oriented nature of the polymeric chain. Finally, polymers do have
a larger temperature-dependent refractive index than SiO2 . For example, a value
of −3 × 10−4 /◦ C has been reported for halogenated polymers. While this property
can be detrimental in some applications, it has been used to make a series of very
sensitive thermo-optic switches (Eldada et al. 2000). Note that if problematic, the
detrimental aspects of the index variation with temperature can be ameliorated by
use of a substrate with compensated thermal properties.
A variety of patterning techniques have been used to form devices or compo-
nents in optical polymers. Three of the most promising commercial processes are
photo-patterning, photodelineation, molding, and reactive-ion-etching (RIE). Photo-
patterning typically uses a photolithographic technique for either polymerizing a
monomer or for changing the index in a pre-existing polymer, a process sometimes
called “photolocking.” In the former technique, the monomer is “spun-on,” as in a
photoresist, and then after exposure developed with a solvent. These techniques yield
structures with an index contrast between substrate/cladding and core of n ∼ 10−1 .
This index contrast in photodelineated waveguides is typically ∼1 × 10−2 . Schematic
of polymer photo-patterning is shown in Fig. 2.4.
18 2 Materials for PICs

Fig. 2.4 Schematic of the photodelineation process for SU-8 photoresist

Fig. 2.5 SEM photo of an InP waveguide fabricated using RIE etching. Source https://www.
plymouthgrating.com/about-pgl/pgls-technology/reactive-ion-etching/

The second approach to fabrication uses a molding or embossing tool which is

transparent to UV radiation. In one approach, liquid monomers are pressed with
the embossing tool, and then exposed through the tool with UV light. While this
approach produces uniform surfaces, additional loss due to surface roughness still
limits its applicability. The third patterning approach uses RIE etching of a pre-
existing polymer substrate. Very high-quality devices have been made using this
approach and it is thus useful for fabricating the most advanced devices (see Fig. 2.5).
Finally, as mentioned above, polymers can also be modified to permit their use
as electro-optical and nonlinear-optical media. In their normal state, polymers are
centrosymmetric amorphous materials and, thus, cannot exhibit an electro-optical
response. However, “poling” of polymer materials with a high voltage can provide
2.3 Polymers 19

the molecular orientation for allowing induced dipolar properties. This response
can be further enhanced by adding specific chromophores to the polymer. These
chromophores can be tailored in functionality and varied in concentration. Polymer
modulators based on poled polymers have recently achieved very high modulation
speeds (Chen et al. 1997). The most important issue concerning the practical use of
these polymers has been poling lifetimes. As mentioned earlier in this chapter, thus
far, the results have been promising but not completely satisfactory.

2.4 Single-Crystal LiNbO3

2.4.1 Overview

LiNbO3 was one of the earliest choices for making relatively large integrated-
photonics devices and circuits. It was selected because its low index allows low loss
mode matching to a fiber I/O and because its combination of excellent and repro-
ducible dielectric and electro-optic properties make it ideal for use as a modulator. In
addition, LiNbO3 is transparent over the wavelength range of most commercial fiber
systems. Because it has been an important electro-optical integrated optical material
for many years, the processing and optical properties of the material are extremely
well characterized. As a result, LiNbO3 -based technology is in an advanced state,
with many available commercial products. LiNbO3 is a birefringent crystal, and thus
two indices of refraction must be specified, namely,

n o (λ) = 2.195 + 0.037/λ2 (2.1)

n e (λ) = 2.122 + 0.031/λ2 (2.2)

where λ is in pm. Typically, birefringence is of minimal importance for LiNbO3 ,

even when using z-cut crystals. Note however, the crystal dependence of the index
of refraction is very important for understanding or calculating the effect of electric
field. One final important advantages of LiNbO3 is that it can be grown in large
crystals, with the size of the crystal being as large as 20 cm in diameter and 25 cm in
length. This large size has allowed relatively complex multielement LiNbO3 PICs to
be made.

2.4.2 Waveguide Formation

LiNbO3 waveguides may be formed by several methods; two of which will be dis-
cussed here: impurity doping and proton exchange.
Ti-Diffused Waveguides
Ti “in-diffusion” utilizes standard lithographic-based patterning of a deposited Ti
thin film. These films are then in-diffused via a heat treatment similar to that used
for semiconductor processing. The diffusion of Ti into the LiNbO3 crystal causes a
20 2 Materials for PICs

Fig. 2.6 Ti-diffused LiNbO3

concentration profile Thin film of Ti

LiNbO3

change in the index of the crystal. This method or procedure, which forms a graded-
index structures with relatively low values of n, e.g., 1 − 4 × 10−3 . Because of the
propensity of LiNbO3 to lose volatile LiO2 during heating, a number of important
material chemistry innovations have been made to form reproducible structures.
These are summarized in Korotky et al. (1987). Commercial dopant diffusion is
done using uniform sample heating. As a result, the relative Ti-doping concentration
(i.e., that percentage of the crystal composition) is given by an equation assuming
local thermally equilibrated diffusion:
   
2 τ y 2
C(y) = √ exp − (2.3)
πd d

where d is the diffusion distance into the crystal, y is the vertical coordinate in
the crystal with y = 0 being at the crystal surface, and τ is the thickness of the
deposited Ti film deposited on the surface of a wafer. The above equation assumes
complete incorporation of the deposited film; i.e., no metal film remains on the
surface after the heating step and that the deposited stripe width is much larger
than the metal thickness, so that the “edge” effects from the stripe are negligible.
The diffusion
√ distance, d, in a time, t, is given by the usual diffusion equation,
d = 2 Dt, where D is the diffusion coefficient. In general, D is a temperature-
dependent diffusion coefficient, e.g., D ∼ 0.5 × 10−12 cm2 /s at 1000 ◦ C or ∼ 2 ×
10−12 cm2 /s at 1100 ◦ C. Generally, D can be calculated with the simple equation

D = D0 exp(−T0 /T ) (2.4)

where T0 = 2.5 × 104 K and D0 = 2.5 × 10−4 cm2 /s. An example of the actual
concentration gradient in a Ti-diffused waveguide is shown in Fig. 2.6. After complete
in-diffusion (no free metal remaining on the surface) of the surface, the index profile
is approximately Gaussian.
In fact, in many cases the actual waveguide diffusion profile is more exactly
treated by a two-dimensional diffusion profile. For example, for typical single-mode
waveguides which are defined by diffusion from a stripe of width, w, and thickness
τ , where w/d ∼ 2 − 4, and thus edge effects are important. In this case,

C(x, y) = C0 X (x)Y (y) (2.5)

2.4 Single-Crystal LiNbO3 21

where w
2 τ
C0 = √ erf (2.6)
πd 2d
      
1 1 w 1 w
X (x) = erf x+ − erf x− (2.7)
2erf(w/2d) d 2 d 2

and  2
y
Y (y) = exp − 2 (2.8)
d

where C0 is the peak concentration and X (x) and Y (y) are the lateral- and depth-
dependent concentration profile. In these equations, erf is the error function.
Doping changes the index of refraction as shown in Fig. 2.6. This change has
been attributed to Ti-atom-induced strain in the crystal lattice. The data show that,
conveniently, the change of index depends approximately linearly on the concentra-
tion, n 0 ≈ bC0 (y), where n 0 is the maximum in index change, i.e., from that
at the surface, and b is a constant of proportionality. For λ = 0.6 µm, for example,
b ≈ 0.7 for the extraordinary axis. In the case of the change in the ordinary index,
a linear relation between n and concentration only holds at lower concentrations,
i.e., <0.4%. For values above this concentration, the induced index change begins
to saturate. This proportionality constant, b, also has a wavelength dependence, due
to dispersion in the doped region:

b(λ) = 0.552 + 0.065/λ2 (2.9)

where λ is in µm. When the three-dimensional character of the waveguide is

accounted for, then
n(x, y) = n 0 X (x)Y (y) (2.10)

The relatively low index of the diffused doped region in Ti-diffused LiNbO3 means
that the guided mode has weak confinement in the waveguide, and hence it possesses
a relatively large mode. Note also that the mode width is controlled to a large extent
by the width of the diffused region. The large mode in Ti:LiNbO3 is however not
necessarily a disadvantage; in facet this large size means that there is typically good
overlap between both these waveguides and that of the input fiber mode. This overlap
is crucial for obtaining low loss pigtailed optical systems. Thus Burns and Hocker
(1977) obtained a useful analytic expression for the overlap. Specifically he showed
that the power coupling efficiency η between a fiber with a mode with a 1/e diameter
a and a waveguide of width d x and depth dy can be written conveniently as
 
4(d/a)2
η = 0.93 (2.11)
[(d/a)2 + r ][(d/a)2 + 1/r ]

where d = dx d y and r = dx /d y .
22 2 Materials for PICs

In addition, often, a crystal is covered with a layer of SiO2 on a LiNbO3 surface,

which optically decouples the surface from overlayer effects particularly metal elec-
trodes. With good cladding or with no metal overlayer, a typical LiNbO3 waveguide
crystal loss is ∼0.05 dB/cm for high-quality crystals.
Proton-Exchange Waveguides
Proton exchange provides a technique for making a relatively high-index change in
LiNbO3 . In this method, a patterned substrate containing a surface mask is suspended
in hot benzoic acid, e.g., ∼240 ◦ C. At this temperature, Li+ ions on an unmasked
region of the substrate are replaced with H+ ions from the acid. This compositional
change leads to an index change of n e ∼ 0.12 and n o ∼ −0.04 for the extraor-
dinary and ordinary index of refraction, respectively. As a result, these waveguides
only guide TM waves. Typically, the index profile is more step-like and has a larger
change in index than for Ti-diffused guides (see Fig. 2.6). It can be shown using
expressions for loss in a bending waveguide that the high-index contrast also allows
use of proton exchange possible for the high values required for tight bends in some
PICs. Typical losses for proton-exchange guides are ∼0.5 dB/cm.
The properties of these proton-exchange guides can be altered by adjusting the
processing procedure. For example, post-exchange annealing leads to a reduced
index gradient. In addition, use of a mixed benzoic and lithium-benzoate solution
increases the stability of the index profile in the presence of an electric field more
than for neat benzoic acid. This lack of stability for the neat acid still limits the
use of proton-exchange waveguides for modulators. The use of annealing (annealed
proton exchange) reduces loss and improves the otherwise-degraded electro-optic
coefficient in proton-exchange LiNbO3 .

2.5 GaAs/AlGaAs or InP/InGaAsP

2.5.1 Overview

III-V materials are well developed for electro-optic and integrated optic applications.
This materials system has several important advantages, including the fact that a III-
V PIC can incorporate a high degree of functionality and easily tailorable index
profiles. Note that increased functionality includes the fact that the standard laser
and LED optical sources are made in III-V materials. However, integration of fully
functional PICs does require complex growth steps, typically involving regrowth or
selective-area epitaxy. In addition, the size of most III-V wafers is relatively small.
2.5 GaAs/AlGaAs or InP/InGaAsP 23

2.5.2 Index Changes

Two techniques can be used for changing the refractive index difference in epitaxial
layers of these III-V semiconductors: doping and hetero-layers (Leonberger and
Donnelly 1988). However, while doping may occasionally be useful, it suffers from
an attendant loss due to free-carrier absorption. In fact, optical loss due to background
doping is an important consideration in designing III-V PICs. Note that free-carrier
effects are also present in voltage-induced changes in carriers as well; these will be
discussed in regard to certain optical modulator types.
Doping-induced changes in semiconductors are due to free-carrier-induced change
in the dielectric constant. This negative change can be obtained using a simple free-
carrier plasma calculation of the dielectric constant. For doped semiconductors, the
index is given by n = n 0 − n where n 0 is the index of the undoped material and

n c e2
n = (2.12)
8π 2 0 n 0 υ 2 m ∗

where υ is the optical frequency, m ∗ is the effective mass of carriers in the semicon-
ductor, e is the charge of an electron, n c is the number density of carriers, and 0 is
the dielectric constant of free space. For example, for a semiconductor with m ∗ = 1
and n 0 = 3.5, n ∼ 3 × 10−20 n c λ20 , where λ0 is the free-space wavelength. Thus
for GaAs the index goes down with doping. Unfortunately, this index change comes
with a price! The physical effect causing n also causes a concomitant change in α,
the optical absorption, in fact this change is invariably an increase. In particular, for
doped GaAs,
ge3
α= (2.13)
4π m ∗2 μn 0 c0
2

where g ∼ 1–3 and is related to carrier scattering times, and μ is the carrier mobility.
This loss causes the guide loss in a simple single-mode homojunction guide to be, at a
minimum, 2.5 dB/cm, if the substrate is doped to n c ∼ 1018 cm−3 and λ0 = 1.3 µm.
Note that in this case, the losses result only from the evanescent tail of the optical
mode in the substrate. A fully confined mode in such a substrate would have a loss
of 4.3 dB/cm, also at λ0 = 1.3 µm.
Other doping-induced absorption effects are present near the band edge. These
are due to deep impurity levels, excitons, band filling, and interband absorption,
such as the near edge absorption features in InP, which is doped to different levels
of impurity concentration and led to a significant free-carrier absorption increase at
longer wavelengths as well as a decreasing region just below the band edge due to
interband absorption.
A very important III-V technology that is evolving, which competes with Si
photonics (see below) is that based on InP photonics. This technology has been
the subject of a recent very thorough review article by a European group, Smit et al.
(2014).
24 2 Materials for PICs

2.6 Single-Crystal Si and Amorphous Si

2.6.1 Overview

As discussed in the introduction to this book, Si has, over the last 1–2 decades, become
a major materials platform for integrated optics and PICs in particular. The high-
quality fabrication that is possible using silicon CMOS electronics planar processing
technology is simply unparalleled for realizing the very high spatial image resolution
and large-scale size needed for commercial PICs. Two solid phases of Si material have
been suggested for Si photonics. The first is that of crystalline silicon and is typically
based on silicon-on-insulator (SOI) wafers. The second uses amorphous Si, which
has attractive important advantages for nonlinear Si optics but has consistently shown
degradation in its optical properties over time in working devices. (Foster 2015)
Several important and pressing applications are drivers for the development of
silicon photonics. Thus Si based on SOI is developing into the key materials tech-
nology for truly large-scale and high-performance integrated optics for communica-
tions and computer interconnects (Shen et al. 2019). This development has grown
rapidly, due to recent demonstration of fabricating essential and basic integrated-
photonics Si building blocks, including modulators, lasers, detectors, and passive
optics. In addition, other building-block active devices have been developed includ-
ing current driven switches, very low-energy modulators, simple ultrafast thermo-
optical switches. The scaling down of these devices to the ultrasmall (<0.1 µm2 -area)
dimensions for single-mode waveguides, called Si nanowires (see Chaps. 3 and 4)
has given rise to a new waveguide technology. This technology has been made pos-
sible by using the Si fabrication tools and infrastructure used in modern CMOS Si
fabrication foundries for integrated electronics.
The unique properties of silicon-wire photonic devices are determined by the
high refractive contrast available in the silicon-on-insulator (SOI)-materials platform.
This high contrast permits extremely tight confinement in two dimensions, hence the
name Si wires. (Driscoll et al. 2015) This high contrast allows deep scaling of single-
mode devices so as to have ultrasmall Si waveguide cross sections (<0.1 µm2 ) and
consequently allows for design of compact device areas. Note that the utility of high-
index contrast has been noted in earlier generations of devices using III-V materials
for integrated optical technologies; see, for example, Levy et al. (1999), Huan et al.
(2000). The device size reduction yields a set of more subtle but very important
advantages, such as the possibility for engineering of waveguide dispersion, high
optical-field densities, reduced group velocities and thus group-velocity-enhanced
effects, and unique optical properties of the supported guided modes such as a strong
longitudinal electric field component.
2.6 Single-Crystal Si and Amorphous Si 25

Fig. 2.7 Refractive index

versus wavelength for silicon
at room temperature

2.6.2 Materials Properties of SOI

State-of-the-art silicon devices have generally been made of silicon based on the
silicon-on-insulator (SOI) materials platform. This platform enables large-scale
wafers to be fabricated using ion-induced film separation followed by wafer bond-
ing. Silicon has a very large refractive index (n = 3.5), which, in conjunction with
a low-index waveguide cladding (n = 1 for air or n = 1.45 for silica), permits tight
confinement of light to sub-wavelength dimensions. Because of the ultrasmall dimen-
sions of the oxide, geometrical dispersion can have a major effect on its effective (see
Chap. 3) refractive index. Both of these materials also exhibit significant chromatic
dispersion. The dispersion in the refractive index for Si (Green 2008) is shown in
Fig. 2.7.
As introduced above Si photonic wires are waveguides typically patterned on a SOI
wafer or chip with a 1–3 µm-thick buried oxide layer (BOX), 100–300 nm Si device
layers. Typically many of the photonics devices based on SOI fabrication employ a
220 nm layer because of its modal properties. The thickness of the underlying oxide
layer is important to reduce optical coupling to the substrate. In fact, in some cases
sufficiently low optical loss can only be obtained via under etching of the buried
oxide layer. Finally note that the electrical properties of the relatively thick SOI
can severely limit its use if significant electrical performance is required. In fact, its
thickness does not allow co-integration of CMOS and SI photonics in most cases.
Thus there is currently a major research thrust in using bulk Si for projects needing
full integration.
The crystal orientation of the chip is important for certain fabrication steps for
the chip. The direction of the propagating optical wave is typically oriented so that
the waveguide is aligned along the [1 1 0] crystallographic axis of the Si device
layer (Driscoll et al. 2015). This choice of [1 1 0] orientation is selected to place the
waveguide facet along a cleavage plane of Si for the ready formation of high-quality
26 2 Materials for PICs

Fig. 2.8 SEM micrograph of a silicon photonics waveguide, with an overlay of the cal-
culated TE mode-profile Source https://www.osapublishing.org/oe/fulltext.cfm?uri=oe-23-25-
32452&id=333314

end facets. A typical patterned structure is shown in the SEM micrograph shown in
Fig. 2.8.

2.6.3 Loss and Patterning

Because of the high quality of Si lithography patterning, the optical loss at C-band
wavelengths in Si-wires is largely determined by optical scattering loss induced by
etch-induced sidewall roughness of the etched waveguide. This loss, including a
useful expression for the upper bound of the loss, is discussed in Yamada (2011)
and is generally a result of etch-induced defects. Thus scattering loss can be reduced
by using advanced dry-etching methods such as hydrogen bromide reactive-ion-
etching. In its current state, state-of-the-art waveguide losses for channel waveguides
using such etching and e-beam patterning are typically measured to be α ∼ 1.0 −
3.0 dB/cm for the 1550 nm range. If the operating wavelength is in the mid-IR region
scattering losses can be smaller but substrate absorption will increase (Liu et al.
2010). The high confinement of wavelength-scale features due to Si’s high refractive
index difference reduces optical loss at bends. This effect enables tight folding of Si
PICs and leads to much higher packing density than previously possible.
The origin of loss due to wall roughness on Si wires has been examined using
analytic coupled-mode theory as well as numerically using finite-difference time-
domain (FDTD) computations. In addition it is important that, as mentioned above,
the buried oxide layer sufficiently thick to prevent evanescent coupling to the bulk
Si substrate (Poulton et al. 2006).

2.6.4 Doping-Induced Index Changes and Loss

Si does not possess an intrinsic χ2 and as a result it is not possible to use a simple
intrinsic layer of Si for electro-optical modulation or switching. As a result, alter-
2.6 Single-Crystal Si and Amorphous Si 27

native approaches to modulation have been used. The most common electro-optical
approach involves doping of the Si to form a local concentration of free carriers,
which exhibit the well-known carrier physics seen in basic electronic device books
and discussed earlier in this chapter for the case of compound semiconductors, such
as GaAs or InP.

2.6.5 Temperature Variation of the Si Index of Refraction

In many photonics applications, temperature stability, or conversely, the ability to

change phase with temperature is an important quantity in device design. Crystalline
Si has a significant temperature induced-refractive index shift that is due to changes
in its electronic structure and shifts in carrier physics. This shift, dn 0 /dT , can be
normalized to the value of n 0 , which for Si at λ = 1.55 µm is 5.2 × 10−5 K−1 . This
quantity will be seen later in this text to be important for thermo-optical switches
and phase stability in resonant and non-resonant devices.

2.6.6 Amorphous Silicon

Many applications in Si photonics require interconnects, which are deposited on

nonplanar and fragile substrates; a particular example is the formation of 3D optical
interconnects. This deposition capability is not possible with crystalline Si waveg-
uides, but it is important for low cost, flexible, or even very dense-device fabrication.
In principle it should be possible to use standard polycrystalline Si application.
However pure α-Si cannot be used because of its poor optical properties, particularly
its high optical absorption due to its high concentration of free dangling bonds. On
the other hand, if these free bonds are reacted with hydrogen, optical absorption
drops dramatically. This hydrogenated amorphous silicon (α-Si:H) is deposited using
low temperature (∼100–400 ◦ C, compared to poly) plasma-enhanced chemical vapor
deposition (PECVD). In addition to its flexibility in deposition its use can also reduce
propagation loss (Burns and Hocker 1977) for at least one form of noncrystalline
Si (Foster 2015). As an example of a loss of 3.46 dB/cm has been achieved for
amorphous Si in a 480 nm × 220 nm Si wire at 1550 nm (TE mode) (Soref 1993).
This result indicates that the propagation loss in α-Si:H waveguides are approaching
that of the best Si wire waveguides, i.e., 1 − 3 dB/cm (Kimerling et al. 2004; Reed
2004). Do note however that the crystalline silicon of silicon-on-insulator substrates
is the only possibility for high-index contrast waveguiding.
In addition, α-Si:H waveguides have a high thermal stability, which is useful in
stacked waveguides. It has been reported that these α-Si:H waveguides can be stable
even at room temperature (Soref 1993) and their propagation loss does not increase
after rapid annealing at 550 ◦ C (Burns and Hocker 1977). In contrast the thermal
budget of in CMOS is limited to that of the back end process of ∼450 ◦ C.
28 2 Materials for PICs

2.7 Summary

This section of our text has focused on a short overview of the materials properties
for five choices of integrated optical platforms. Each of these has been used in
commercial devices and thus we have chosen to review each. In our review we have
also attempted to give brief mention of the devices that are important for each material
type. We have also given the typical properties for each material class. Note that the
choice of materials is determined by the PIC application.

Problems

1. Coupling to a fiber-optic input/output for a photonic chip is important for a low

loss optical system.
Find the FWHM of the mode from a single-mode commercial fiber; what is
the approximate shape of the mode? You will have to do some searching (web,
references, books).
2. Determine the Al Composition, i.e.,, x, needed to obtain a n = 0.04 for
GaAs/Alx Ga1−x As structure at λ = 1.3 µm and λ = 0.9 µm, respectively.
3. A Ti diffused waveguide is made in LiNbO3 by annealing the following structure
at 1050 ◦C. If the Ti stripe thickness is 900 Å, how long will it take to change the
index by 1 × 10−3 ? The stripe width is 8 µm. Sketch the index profile at x = 0,
along Y, λ = 1.3 µm (Fig. 2.9).
4. Derive the wave equation for E x , propagating along +z, in the homogeneous
medium having ε, μo .
5. It’s important to get a sense of how the waveguides are actually manufactured
using different materials and techniques. Look into this chapter and search online
to answer the following questions about fabrication techniques:
(a) Sketch and explain how a GaAs/AlGaAs slab waveguide is made.
(b) Sketch and explain how a LiNbO3 channel waveguide is made.

Fig. 2.9 Cross-sectional

view of a Ti-diffused
Waveguide
References 29

References

Burns, W. K., & Hocker, G. (1977). End fire coupling between optical fibers and diffused channel
waveguides. Applied Optics, 16(8), 2048–2050.
Chen, D., Fetterman, H. R., Chen, A., Steier, W. H., Dalton, L. R., Wang, W., et al. (1997). Demon-
stration of 110 GHz electro-optic polymer modulators. Applied Physics Letters, 70(25), 3335–
3337.
Driscoll, J. B., Osgood, R. M., Grote, R. R., Dadap, J. I., & Panoiu, N. C. (2015). Squeezing
light in wires: Fundamental optical properties of Si nanowire waveguides. Journal of Lightwave
Technology, 33(14), 3116–3131.
Eldada, L., & Shacklette, L. W. (2000). Advances in polymer integrated optics. IEEE Journal of
Selected Topics in Quantum Electronics, 6(1), 54–68.
Eldada, L. A., Beeson, K. W., Pant, D., Blomquist, R., Shacklette, L. W., & McFarland, M. J.
(2000a). Polymeric components for all-optical networks. In Optoelectronic integrated circuits IV
(Vol. 3950, pp. 78–90). International Society for Optics and Photonics.
Foster, A. C. (2015). Advances in amorphous silicon waveguides for nonlinear optical signal pro-
cessing. In 2015 IEEE Summer Topicals Meeting Series (SUM) (pp. 90–91). IEEE.
Green, M. A. (2008). Self-consistent optical parameters of intrinsic siliconat 300 K including
temperature coefficients [J]. Solar Energy Materials and SolarCells, 92(11), 1305–1310.
Henry, C. H., Blonder, G., & Kazarinov, R. (1989). Glass waveguides on silicon for hybrid optical
packaging. Journal of Lightwave Technology, 7(10), 1530–1539.
Huan, Z., Scarmozzino, R., Nagy, G., Steel, J., & Osgood, R. (2000). Realization of a compact and
single-mode optical passive polarization converter. IEEE Photonics Technology Letters, 12(3),
317–319.
Kimerling, L., Dal Negro, L., Saini, S., Yi, Y., Ahn, D., Akiyama, S., Cannon, D., Liu, J., Sandland,
J., Sparacin, D., et al. (2004). Monolithic silicon microphotonics. In Silicon photonics (pp. 89–
120). Berlin: Springer.
Korotky, S., Alferness, R., & Hutcheson, L. (1987). Integrated optical circuits and components.
Leonberger, F., & Donnelly, J. (1988). Semiconductor integrated optic devices. In Guided-wave
optoelectronics (pp. 317–395). Berlin: Springer.
Levy, D. S., Park, K. H., Scarmozzino, R., Osgood, R. M., Dries, C., Studenkov, P., et al. (1999).
Fabrication of ultracompact 3-dB 2 x 2 MMI power splitters. IEEE Photonics Technology Letters,
11(8), 1009–1011.
Liu, X., Osgood, R. M, Jr., Vlasov, Y. A., & Green, W. M. (2010). Mid-infrared optical parametric
amplifier using silicon nanophotonic waveguides. Nature Photonics, 4(8), 557.
Poulton, C. G., Koos, C., Fujii, M., Pfrang, A., Schimmel, T., Leuthold, J., et al. (2006). Radiation
modes and roughness loss in high index-contrast waveguides. IEEE Journal of Selected Topics
in Quantum Electronics, 12(6), 1306–1321.
Reed, G. T. (2004). Device physics: The optical age of silicon. Nature, 427(6975), 595.
Shen, Y., Meng, X., Cheng, Q., Rumley, S., Abrams, N., Gazman, A., et al. (2019). Silicon photonics
for extreme scale systems. Journal of Lightwave Technology, 37(2), 245–259.
Smit, M., Leijtens, X., Ambrosius, H., Bente, E., Van der Tol, J., Smalbrugge, B., et al. (2014). An
introduction to InP-based generic integration technology. Semiconductor Science and Technology,
29(8), 083001.
Soref, R. A. (1993). Silicon-based optoelectronics. Proceedings of the IEEE, 81(12), 1687–1706.
Yamada, K. (2011). Silicon photonic wire waveguides: Fundamentals and applications. In Silicon
photonics II (pp. 1–29). Berlin: Springer.
Chapter 3
Dielectric Slab Waveguide

Abstract The origin of guided-wave behavior is presented for the case of slabs
and for thin films of dielectric materials. The discussion examines the origin of the
guided wave using a classical optics approach and that of Maxwell’s equations, as
well as the matching of fields at the interfaces of dielectric layers. The chapter also
includes various approximations and graphical methods for waveguides with abrupt
and diffused geometry.

3.1 Introduction

Waveguides are the wires of photonic integrated circuits. They transport light just as
metallic wires transport electrons. Unlike wires, however, waveguides or an assem-
bly of waveguides can be used as functional passive elements in a circuit. There are
many different forms of waveguides, including those with a variety of geometries
and materials. Generally, however, a waveguide typically has a light-confining struc-
ture obtained by surrounding a high-index material with a low index cladding. Two
examples are shown in Fig. 3.1. Such waveguide functions by virtue of confining the
light in the core through the phenomenon of total internal reflection (TIR).

3.2 “Thin-Film” Waveguides

3.2.1 Slab Waveguide (2D)

Most integrated waveguides are formed from thin-film structures. The structure fab-
rication starts with a substrate, to which is added a higher index film and lower index
cover (which could be air); see Fig. 3.2. In the simplest of these structures, which
is termed a “slab” waveguide, total internal reflection confines light vertically, but
not horizontally. Since a slab waveguide confines light only vertically, it is in itself
not typically very useful for routing light within a PIC. A slab waveguide, however,

© Springer Nature Switzerland AG 2021 31

R. Osgood jr. and X. Meng, Principles of Photonic Integrated Circuits,
Graduate Texts in Physics,
https://doi.org/10.1007/978-3-030-65193-0_3
32 3 Dielectric Slab Waveguide

Fig. 3.1 The cross section of two important waveguide shapes: circular and rectangular waveguides.
Despite its difficulty for analysis, the rectangular waveguide is the most commonly used type for
analysis

Fig. 3.2 The cross section

of a typical thin-film-based
structure. In general the
index of the cover and
substrate region are not
identical

has the advantage of being able to be exactly analyzed: it thus can serve as a model
structure for understanding the basic physics and technology of waveguiding.
In practical PIC and OEIC chips, however, light must also be confined laterally
in two dimensions in order to achieve the two-dimensional routing, that is needed
for the working surface of the light circuit. Lateral confinement is achieved by intro-
ducing changes in the lateral geometry or in the material index of refraction. As a
result, a physics analysis of PICs requires understanding of light propagation in these
3D waveguides; see Fig. 3.3. Three-dimensional guides are surprisingly difficult to
analyze, although they may be fabricated relatively easily. Introduction and analysis
of different types of 3D waveguides will be considered in the next chapter.
A full description of waveguiding requires consideration of several areas of
physics. These include general waveguide phenomena, such as lightwave polar-
ization, guided modes, and radiation modes. In addition, the relation of waveguide
geometry to the properties of the guided wave, such as modal “shape” and its relation
to guided-wave loss, and coupling, need to be considered.
In this chapter, the analysis of basic slab waveguides will be demonstrated. Note,
however, that much of the behavior of 3D guides can be anticipated via a thorough
understanding of 2D waveguides.
3.3 Slab Waveguide: Ray Optics Picture 33

Fig. 3.3 Waveguides that lack (left) and that incorporate (right) lateral wave confinement. The
2D or “slab” waveguide spreads by diffraction as the waveguide beam propagates along in the
x-direction. For the 3D waveguide, index guiding typically leads to a confined structure

3.3 Slab Waveguide: Ray Optics Picture

One of the most readily understandable approaches to slab waveguides is that

obtained from ray optics, combined with the known interaction of a plane wave
with a dielectric interface. This approach is useful for understanding both the phase
conditions for transverse modes to exist, as well as the variation in phase velocity
with mode number. To apply ray optics, we start with a plane wave incident at some
angle and reflecting within the film. The confinement by the waveguide is determined
by the behavior of the plane wave as it strikes the interface.
There are three possible trajectories for a ray, which travels within the slab struc-
ture. In the first, there is total internal reflection (TIR) at the cover and substrate
interface, and a trapped or guided ray is then possible. In the second instance, there
is TIR at the cover interface only, while the ray scatters at the substrate interface
into the “substrate.” Finally, in the third case, there is no TIR at either interface and
radiative loss occurs into both the substrate and the cladding.
Since total internal reflection at either a top or bottom interface is important, we
define two critical angles:
nc ns
θc ≡ sin−1 , θs ≡ sin−1 (3.1)
nf nf

where n c ≤ n s < n f , θc ≤ θs , and where n c,s, f are the index of refraction in the
cladding, substrate, and core region. Now, for a ray to radiate through the top or
bottom surface of the waveguide, θt must be a real angle in the range 0 ≤ θt ≤
90◦ ; this implies sin θt ≤ 1. Thus, for the substrate the existence of refracted beam
requires, by Snell’s law, that θi ≤ sin−1 (n s /n f ) ≡ θcrit . When θi > θcrit , we have
the condition of total internal reflection, in which condition, light is trapped in the
waveguide.
It is possible to continue to use the ray optics approach in conjunction with the
known interfacial phase shifts for a plane wave so as to analyze the dielectric slab
waveguide, and to obtain expressions for modal dispersion, etc. (Kogelnik 1988). This
34 3 Dielectric Slab Waveguide

analysis shows that a mode exists when the total phase shift for a ray, in traversing
the waveguide from the substrate to cladding and back to the substrate, is an integral
number of 2π, say 2mπ, where m is an integer. However, a more useful and complete
approach is obtained by directly using Maxwell’s equations.

3.4 The Wave Equation for a Slab Waveguide

3.4.1 The Wave Equation and Its Boundary Conditions

The modal fields of a slab waveguide, as shown in of Fig. 3.4, are obtained directly by
using the wave equation to solve for the relative fields in the dielectric slab and then
applying the boundary condition in the transverse directions at the slab boundaries.
In this derivation, a simple harmonic, eiωt , time dependence, is assumed, in part,
since monochromatic light signals are typically used in PICs. Next, the waveguide is
assumed to have a medium, which is uniform within, but of different indices, inside
and outside of the waveguide, and of nonmagnetic character. As seen in the sketch
in Fig. 3.4, the slab consists of three layers, each with its owe indices, n c ,n s , and n f .
Application of Maxwells equations, thus,

∇ × E = −iωμ0 H (3.2)

∇ × H = +iω0 n 2 E (3.3)

where

Fig. 3.4 Light confined and propagating in a slab waveguide

3.4 The Wave Equation for a Slab Waveguide 35


n2 ≡ (3.4)
0

Then gives the wave equation

∇ 2 E + k 2 n 2 E = 0 (3.5)

where k, the wavenumber in vacuum, is

√ 2π
k = ω 0 μ0 = (3.6)
λ
This well-known form of the wave equation is called the Helmholtz equation; it
is used repeatedly in waveguide problems.
The geometry of the slab waveguide is unconstrained in the longitudinal or z-
direction, and hence waves in the z-direction are propagating waves. As a result,
the z-dependence is of the form e−iβz for forward-traveling waves, where β is the
propagation constant in the z-direction.
The magnitude of the propagation wavenumber in a free dielectric medium, which
forms the core of the waveguide, is kn. In a waveguide, transverse, i.e., top and
bottom, confinement causes the allowed propagating wave modes to have wavevector
components in the z- and x-direction, such that

k x2 + β 2 = n 2 k 2 (3.7)

One goal of solving the waveguide problem is thus to determine the mode-
dependent values of k x and β.
The Helmholtz equation as shown above assumes a uniform medium; thus it
only holds within each layer but not at the interfaces. At the interfaces, boundary
conditions, which are derived from Maxwell’s equations, must be applied. These
boundary conditions are the continuity of the transverse components of E and H . In
addition, for a slab waveguide, it is assumed that the slab waveguide solutions are
∂ E
uniform in y, the second lateral direction as shown in Fig. 3.4, and thus, = 0.
∂y

3.4.2 Effective Index

Since the z-dependence of the propagating wave in the guide is determined by β, it

is important to examine the implications of having β < nk. For example, since the
effective wave velocity along z is defined as follows:

ω ck c
v= = = (3.8)
β β Neff
36 3 Dielectric Slab Waveguide

the wave velocity in the z-direction can be viewed as resulting from an “effective”
index,
β
Neff ≡ (3.9)
k
which describes the modification of the wave velocity by the waveguide. Note that
this definition encompasses both material- and geometry-dependent effects, through
n and d, respectively, where d is the waveguide thickness.
The allowed values of β are determined by the indices of refraction of the three
layers. Specifically, recall that guided modes exist in the range θs < θ < 90◦ ; thus

β
ns < (or Neff ) < n f (3.10)
k
and similarly, we have
β
< ns (3.11)
k
for any radiation modes.

3.4.3 Polarization: Why TE and TM

Slab waveguide modes are normally classified as either transverse electric (TE)
or transverse magnetic (TM). This choice of terminology is also used for the two
possible polarizations of the fields. TE-polarized light has its electric field only in
a plane parallel to that of the waveguide slab, while for the TM polarization, the
magnetic field lies in this plane. Note that substitution in Maxwell’s equations shows
that these two polarizations are independent of each other. In addition, because of the
waveguide confinement in the x-direction, it can also be shown that the magnetic field
for the TE polarization and the electric field in the TM polarization each possesses
a non-zero component along the z or propagation axis. Thus the waves deviate from
the fully transverse behavior of simple plane waves; this “deviation” can ultimately
be seen to originate from the transverse confinement of the lightwave.
Specifically, if we solve the wave equation for the slab waveguide geometry,
along with the appropriate boundary conditions, we find that, by convention, there
is one solution in which the longitudinal electric field is zero, which implies that E
is only transverse, the TE mode. Similarly, the other solution has a zero longitudinal
magnetic field, the TM mode. These two solutions result from the application of the
two different boundary conditions. If E z and Hz are given, the other components can
be determined, and, in fact, the longitudinal components of E and H determine the
optical field entirely. The wave equations for the two separate polarization cases are
summarized in Table 3.1.
In the following section, we derive the modal properties for each of these two
cases. This derivation is accomplished by solving the appropriate wave equation
3.4 The Wave Equation for a Slab Waveguide 37

Table 3.1 Summary of the wave equations for TE and TM modes

Wave equation Boundary condition
∂2 E y
TE mode (E z = 0) + (k 2 n 2 − β 2 )E y = 0 E y continuous
∂x 2 ∂ Ey
Hz continuous =⇒ cont.
∂x
∂ 2 Hy
TM mode (Hz = 0) + (k 2 n 2 − β 2 )Hy = 0 Hy continuous
∂x 2 1 ∂ Ey
E z continuous =⇒ 2
n ∂x
cont.

with boundary conditions given in Table 3.1. This approach yields both the field
profiles and the dispersive properties, i.e., β versus ω for the two cases.

3.4.4 TE Guided Modes

3.4.4.1 Dispersion Relation

The simplest case to examine is that for the TE guided mode, i.e., for E y . Then,
a solution is sought for E y , which is confined within the slab in the x-direction.
Confinement in the x-direction will yield a mode which is sinusoidal inside the slab
with wavenumber in lateral direction of κ ≡ k x , and has exponential decay outside
as governed by δ and γ. Thus we write
⎧ −δx
⎪
⎨ Ce x >0
E y = A cos κx + B sin κx −d ≤ x ≤ 0 (3.12)
⎪
⎩ +γx
De x < −d

where δ, κ, γ, A, B, C, and D are parameters to be determined from the wave

equation, with application of the appropriate boundary conditions.
Substitution of this form of E y into the Helmholtz equation in each region yields
the following relationships:
−δ 2 = k 2 n 2c − β 2
−γ 2 = k 2 n 2s − β 2 (3.13)
κ =
2
k 2 n 2f −β 2

note that the last equation of (3.13) is identical to (3.7). Further, the boundary
condition on the continuity of E y across the interface gives C = A and De−γd =
A cos κd − B sin κd.
Thus,
38 3 Dielectric Slab Waveguide
⎧ −δx
⎪
⎨ Ae x ≥0
E y = A cos κx + B sin κx −d < x < 0 (3.14)
⎪
⎩
(A cos κd − B sin κd)eγ(x+d) x ≤ −d

∂ Ey
Now, applying the continuity of Hz or across the interface then yields the
∂x
following:
− δ A = κB (3.15)

and
γ(A cos κd − B sin κd) = κ(A sin κd + B cos κd) (3.16)

Equation (3.15) gives B = −(δ/κ)A; inserting this relation into (3.16) then yields
the following relation between δ, γ and κ:

κ(γ + δ)
tan κd = (3.17)
κ2 − γδ

Expressing the relations of δ, γ, and κ in terms of β and k, i.e., (3.13), and inserting
them into (3.17), we obtain an eigenvalue equation for the allowed β’s. In fact, a
careful analysis of (3.17) shows that this equation leads to an equivalent way of
expressing the fact that the phase accumulation of a ray in the waveguide during one
complete transverse path must be as an integer number times 2π.
Equation 3.17 can be solved numerically using commercial mathematical solvers,
or alternatively the equation can be solved graphically.
The particular value of κd given in (3.18) as follows:

κd = kd n 2f − n 2s (3.18)

is such an important physical quantity in designing waveguides that it is often called

the “normalized frequency” of the guide, V see, for example, Nishihara et al. (1989).
Thus, 
V = kd n 2f − n 2s (3.19)

Note that it is a dimensionless quantity. As a practical matter, V is also directly

proportional to the numerical aperture, NA, of the waveguide, since NA = V /kd.
The allowed modes, the cross points of lift and right-hand side of (3.17), are
shown in Fig. 3.5. Note that as the waveguide V increases, the number of modes
also increases; that is, large d (compared to λ) and higher index contrast, i.e., n =
n f − n s lead to a larger number of modes. As V decreases (thus implying a narrower
core, a smaller n, a longer wavelength, or a low optical frequency), the number of
modes decreases and, in fact, a width will be reached where no modes can exist in the
case of an asymmetric slab waveguide. This condition is termed waveguide “cutoff”.
3.4 The Wave Equation for a Slab Waveguide 39

Fig. 3.5 A plot of the

functions in the left and
right-hand side of (3.17).
The cross points are the
discrete solutions in κd for
the allowed guided mode

A more explicit equation displaying the mode number m can be obtained as

an alternately written version of (3.17). This equation expresses the fact that the
boundary-condition equations can have a non-trivial solution if
  κ

κ
κd = (m + 1)π − tan−1 − tan−1 (3.20)
γ δ

where m = 0, 1, 2 . . . This equation is termed the dispersion relation for the waveg-
uides, since it specifies the allowed values of β and ω (or k) for the waveguide
geometry and its optical properties. The equation also expresses the fact that the
total lateral phase shift for each mode is an integral value of 2π, since the phase
shifts at the cladding-core and substrate-core interface are
 
−1 κ
φs = tan (3.21)
γ

and κ

φc = tan−1 (3.22)
δ
respectively, and the phase shift in passing through the waveguide is κd. This dis-
persion relation (3.20), which can also be obtained using the ray optics approach, is
plotted in Fig. 3.6 for the first three modes of a symmetric waveguide, i.e., n c = n s .
The dispersion relation (3.20) implies that a slab waveguide will possess a mode-
dependent group velocity, vg = ∂ω/∂β. This velocity can be obtained by taking the
total derivative directly from the dispersion relation after expressing κ, δ and γ in
terms of k and β. The velocity can be seen graphically as the slope in the dispersion
at a specific ω for each mode as shown in Fig. 3.6.
In practical waveguide (and fiber optic) design, recourse is typically made to
a useful graph which contains quantities that are normalized and thus generally
applicable to any slab waveguide. First, recall that we have previously introduced
40 3 Dielectric Slab Waveguide

Fig. 3.6 The normalized

dispersion curve between
wave vector and frequency.
This plot enables rapid
design of single-mode
behavior

Fig. 3.7 The dispersion

relation between normalized
frequency and normalized
guide index for a slab
waveguide operating on
three different spatial
different modes

one such quantity, i.e., V, as defined above. In addition, two other quantities, “a” and
“b,” can be introduced. The quantity “a” is termed the waveguide asymmetry, since
it quantifies how much the cover and substrate index, n c and n s , differ from each
other, where
n 2 − n 2c
a ≡ 2s (3.23)
n f − n 2s

If n c = n s , then a = 0. The quantity b is typically termed the normalized waveg-

uide index and is defined by
β
2
− n 2s 2
Neff − n 2s
b≡ k = (3.24)
n 2f − n 2s n 2f − n 2s
3.4 The Wave Equation for a Slab Waveguide 41

Table 3.2 Summary of parameters used to describe the behavior of slab waveguides for TE and
TM light

TE TM
 
Normalized thickness V ≡ kd n 2f − n 2s V ≡ kd n 2f − n 2s
n 2s − n 2c n 4f
Asymmetry factor a= am ≡ a
n 2f − n 2s n 4c
 2
(Neff )2 − n 2s nf
Normalized guide index b≡ bm = b,
n 2f − n 2s n s qs

where qs =
   
Neff 2 Neff 2
+ −1
nf ns

Thus when β/k or Neff → n s , b → 0; and when β/k or then Neff → n f , then
b → 1. These three normalized quantities are summarized in Table 3.2.
The definitions in Table 3.2 allow the dispersion relation (3.20) to be written in a
normalized form:
 
√ b b+a
V 1 − b = mπ + tan−1 + tan−1 (3.25)
1−b 1−b

This expression is often alternatively written as

 
√ 1−b 1−b
V 1 − b = (m + 1)π − tan−1 − tan−1 (3.26)
b b+a

One can solve this equation numerically and obtain a general graph of its solu-
tions; this graph is shown in Fig. 3.7. This curve is called the normalized dispersion
curve, since it plots the variation of the normalized guide index, b, versus normalized
frequency, V . The diagram also provides a convenient means for determining modal
cutoff.
The graph in Fig. 3.6 shows that cutoff occurs sequentially for higher order modes
as the waveguide is made thinner or as waveguide confinement is reduced. This cutoff
condition is reached when V is less than its cutoff value Vcm for a specific mode m.
Using (3.25), the normalized frequency at cutoff can be found to occur at
√
Vcm = mπ + tan−1 a (3.27)

Notice that, for a symmetric waveguide, a = 0 and the lowest order mode, m = 0,
mode does not have a “cutoff” value. In addition, (3.27) shows that for a waveguide
with normalized frequency V , the number of allowed modes, m, is given by m ≈
V /n.
42 3 Dielectric Slab Waveguide

3.4.4.2 Approximate Solutions

Very useful approximate, but analytic forms for the above equations can be obtained
from the normalized dispersion curve near certain limiting regions. For example,
near cutoff, b → 0, is found that

√ b √
V 1 − b = mπ + + tan−1 a (3.28)
1−b
or √ √
V 1 − b = Vc 1 − b + b (3.29)

Finally note that other expressions are found in other limiting regions.

3.4.4.3 Normalization of Waveguide Modes

In order to understand how the shape of a waveguide mode varies with its material
properties, the modes local field structure needs to be normalized. Thus recasting
(3.14), as ⎧ −δx
⎪
⎪ Ae x >0
⎪
⎪  
⎪
⎨ δ
A cos κx − sin κx −d ≤ x ≤ 0
Ey = κ (3.30)
⎪
⎪  
⎪
⎪ δ
⎪
⎩ A cos κd + sin κd eγ(x+d) x < −d
κ

δ
where B = − A has been inserted in (3.14).
κ
The arbitrary constant, A, can now be specified by fixing the power carried by the
mode as a constant value, P. In order to do this, integrate the z-component of the
Poynting vector,
1
Sz = Re{ E × H ∗ }·ẑ (3.31)
2
over the transverse cross section of the guide. Thus, after using Maxwell’s equations
to obtain H from the form of E y determined above, an expression for the power can
be found:    ∞
β 1 ∞ ∗ β
P=− E y Hx dx = |E y |2 dx (3.32)
|β| 2 −∞ 2ωμ0 −∞

where the presence of factor β/|β| ensures that the power (P) is always positive.
Substitution for β/|β| then yields the amplitude of the lowest order mode,
3.4 The Wave Equation for a Slab Waveguide 43

Fig. 3.8 A notional plot of

the effective width, Teff , of a
waveguide versus its
physical width, d

4κ2 ωμ0 P
A2 = 1

(3.33)
1
|β|(κ2 + δ 2 ) d + +
γ δ

This relation shows that the peak intensity of the waveguide mode depends on the
confinement of the mode in the waveguide by the quantities 1/γ, 1/δ and d. This
leads to a definition for an effective width of the waveguide, Teff :

1 1
Teff ≡ + +d (3.34)
γ δ

It is instructive to determine how the effective width of a waveguide varies as its

physical width is reduced. This variation is shown in a plot of normalized variables
in Fig. 3.8. Notice that reducing the waveguide width,d, reduces the width of the
mode only up to a certain point. For guides smaller than that, the effective width of
the mode begins to expand. While this behavior implies a lack of confinement, it has
important uses such as to improve mode matching in fiber-waveguide interfaces.

3.4.4.4 Modal Confinement

The adoption of normalized units is also useful for specifying practical engineering
properties for waveguides. For example, it is often important, in diode laser design, to
specify the “confinement factor” of a waveguide. This factor is the ratio of the power
in the waveguide core to the total power in the waveguide. For a slab waveguide, this
quantity (Hutcheson 1987) is
√ √ 
d b b+a
= 1+ + (3.35)
teff V V (1 + a)
44 3 Dielectric Slab Waveguide

where the second and third terms are due to the asymmetric shape of the mode. For
a symmetric waveguide,  reduces to
√
V b + 2b
= √ (3.36)
V b+2

3.4.4.5 Symmetric Slab

Symmetric slab waveguide are useful for easily capturing the essence of the properties
of waveguide. Their dispersion relations are particularly simple, thus allowing a
simple physical picture for modes with odd and even mode numbers. In this case,
when n c = n s , δ = γ and the dispersion relation becomes tan κd = (2κγ)(κ2 − γ 2 ).
The use of trigonometric identities leads to two classes for the modes:

κd γ
tan = for even m (3.37)
2 κ
and
κd κ
tan = for odd m (3.38)
2 γ

In particular, rewriting the expression for the field in the slab by shifting the
coordinate origin to the center of the slab (see Fig. 3.9), we see that even, symmetric
(A cos κx), and odd, antisymmetric (B sin κx), modes exist. Recall also that for a
symmetric slab (n c = n s ), a = 0, and there is no cutoff for m = 0, or the lowest order
mode in a symmetric guide, since we cannot have V < 0, which is not physical!
The field distributions in a symmetric waveguide structure are a simplification of
the earlier expressions for the general structure:
⎧ −δx
⎪
⎪ Ae x >0
⎪ 
⎪ 
⎪
⎨ δ
A cos κx − sin κx −d ≤ x ≤ 0
Ey = κ (3.39)
⎪
⎪  
⎪
⎪ δ
⎪
⎩ A cos κd + sin κd eγ(x+d) x < −d
κ

Finally, note that if the waveguide is not symmetric, the electric fields at the
substrate-core (E s ) and cladding-core (E c ) interfaces are no longer the same. It can
then be shown via the boundary condition that these ratios are

n 2f − Neff
2
E c2 = E 2f (3.40)
n 2f − n 2c
3.4 The Wave Equation for a Slab Waveguide 45

Fig. 3.9 The electrical field distribution calculated for four different TE modes, using a typical
waveguide similar to that shown in Fig. 3.2

n 2f − Neff
2
E s2 = E 2f (3.41)
n 2f − n 2s

where E f is the maximum field amplitude in the core. Notice that the smaller the
value of n c , n s , the smaller the electric field at the cladding, substrate interface.

3.4.4.6 TM Modes

In the derivation above, we have ignored discussion of TM modes. In fact, the deriva-
tions and results follow very closely to those for the TE modes. However, the nor-
malized quantities “a” and “b” are different, although practically speaking “b” is
often identical to that defined for the TE mode if n s and n f are close, that is, if the
waveguide has only a small index contrast. The expression for “a” for TM modes
is typically significantly different from for TE modes and thus is designated by the
symbol am . The expression for am is given in Table 3.2. In addition the normalized
dispersion curves for TM modes will then have a much more complicated form
than that of the TE mode; however, in practice, particularly for low index contrast
materials, the curves generally have a very similar shape.

3.4.4.7 Dispersive Properties of Waveguides

Waveguides are dispersive elements because of the sensitivity of propagation con-

stants to their material properties, waveguide geometry, and mode number. These
dispersive properties can be of direct importance in several different PICs, including
optical delay lines and wavelength routers.
46 3 Dielectric Slab Waveguide

The starting point for discussing dispersion is the group velocity, defined as

dω
νg = (3.42)
dk
This is the velocity at which a pulse of light travels. Using the expression for k in
a medium of index n gives a more useful expression,
c
νg =   (3.43)
dn
n+ω
dω

where it is convenient to define the group index, Ng , as

dn dn
Ng = n + ω =n−λ (3.44)
dω dλ
When several operating wavelengths are present in a device, it is important to
account for the dispersion of the group velocity with wavelength. This dispersion
leads to a difference in arrival time, t, for pulses centered between two wavelengths
separated by λ,
L dNg
t = λ (3.45)
c dλ
dNg
where L is the difference in path lengths and is the group index dispersion.
dλ
The group index dispersion can be rewritten in terms of the second (wavelength)
derivative of the refractive index,

dNg d2 n
= −λ 2 (3.46)
dλ dλ
thus in transparent materials the dispersion can be defined as

−λ d2 n
D= (3.47)
c dλ2
Examples of these quantities for the case of SiO2 are given in Fig. 3.10.
As suggested by an earlier discussion in this chapter, dispersion also can occur
because of the vertical confinement in waveguides. This dispersion arises due to the
sensitivity of β to the wavelength of light in a waveguide, a result apparent from the
waveguide eigenvalue equation. Specifically, the group velocity in the waveguide is
thus given by
dω
νg = (3.48)
dβ
3.4 The Wave Equation for a Slab Waveguide 47

Fig. 3.10 The group index and the dispersion relation for the case of bulk SiO2

where β has replaced the quantity k used in the previous equation. The group index,
Ng , is then
dβ
Ng = (3.49)
dk
As the wavelength of the light changes, this variation causes a change in the
confinement of the mode. For example, considering the right-hand region of Fig. 3.5,
as k is varied, the mode samples more of the cladding at low k and more of core at
high k. As a result, the effective index is seen by the light changes with λ.

3.4.5 TE Radiation Modes

The number of guided modes is finite; thus there must be other solutions of Maxwell’s
equations in order to provide the complete set of modes, which must be used to
describe an arbitrary initial-field configuration. In fact, such an arbitrary field distri-
bution has to include both guided and radiative modes.
Radiative modes were encountered earlier in this chapter when discussing the
slab waveguide using the ray optics picture. For example, recall that for θc < θ < θs ,
or n c < β/k < n s , there are substrate radiation modes, or for 0 < θ < θc , or 0 <
β/k < n c , there are substrate and cladding radiation modes present. For example, if
48 3 Dielectric Slab Waveguide

Fig. 3.11 Ray picture of

radiation modes

the original field inside the waveguide contains k components with an incident angle,
which is too shallow with respect to the interface, it must contain, in part, radiation
modes.
Another viewpoint, which is helpful for understanding the form the solutions will
take, is as follows. Rather than considering the wave as originating from inside the
slab, consider a plane wave incident on the slab from one side, say the substrate (see
Fig. 3.11). For an incident angle shallow enough relative to the interface, light couples
into the film and is reflected at the second air interface. At this second interface, there
is a decaying or evanescent field in the air. But also note that there is then a reflected
wave going into the substrate. The combination of incident and reflected waves in
the substrate leads to a (partial) standing wave, and this will be reflected so as to the
form of a solution for this case.

3.4.5.1 Substrate Radiation Modes

To solve for the radiation modes of the waveguide, the equations and boundary
conditions are the same as used earlier. However, the range of β is such that the
parameters γ and δ may no longer be real and the field then oscillates along the
y-direction; that is, radiation may propagate or radiate away from the waveguide.
This approach leads to developing a set of continuous radiation modes, labeled
by parameter ρ, with ρ in the range:

0 ≤ ρ ≤ k n 2s − n 2c (3.50)

k n 2s − n 2c ≤ ρ ≤ n s (3.51)

However, ρ is also allowed in the range n s ≤ ρ ≤ ∞, (β is imaginary), and the

“substrate-cladding” form of the equations holds true. Note that these modes are
evanescent in z and are particularly needed for describing the field around imperfec-
tions.
3.4 The Wave Equation for a Slab Waveguide 49

The guided and radiation modes that are then found form a complete set. As a
result, any arbitrary field distribution in the waveguide can be written as a superpo-
sition of these modes.

3.5 Graded-Index Waveguides

Thus far in our presentation of slab waveguides, we have only considered abrupt-
index guides. However, many practical guides are made using fabrication techniques
which lead to smoothly merging index profiles. These methods include diffusion, ion
implantation, and optical exposure. Each of these techniques leads to confinement
by a graded-index gradient.
Because of their relative complexity, several approaches have evolved for dealing
with slab waveguides which are confined by a graded-index profile. These include
analogies to well-known problems of ray optics, quantum mechanics, and the WKB
method (see Kogelnik 1988). We will consider only the first here.

3.5.1 Ray Optics Approach for Diffused Waveguides

A ray optics approach can be applied to finding the normalized dispersion curve for
a graded-index waveguide, using the same basic methods as for an abrupt-index slab
waveguide (Nishihara et al. 1989). In this method, a ray trajectory is first “followed,”
and its accumulated phase shifts along the path and summed at the interfaces. The
principal value of this method, which is very nicely discussed by Nishihara et al.
(1989) is that it provides physical insight into the trapping of light in the waveguide
(see Fig. 3.12). If the waveguide is formed such that the index, n(x), decreases with
x, the distance into the cladding-core interface, a ray propagating both in the x- and
z-direction undergoes a continual change in trajectory as measured by the angle θc
such that cos θi = Neff /n(x), where i denotes the ith increment of the ray’s path.
The angle, θi , is the angle between the ray and the z-direction and Neff is the effect
index of the waveguide. At the end of the downward trajectory, x = xt and θi = 0
when n(xt ) = Neff . At this point, the wave turns upward toward the guide, where xt
is the turning point of the ray. Thus xt can be defined as the effective thickness of the
diffused waveguide.
In this waveguide class, the waveguide modes are obtained by quantizing the
cumulative phase change in units of 2π as the ray undergoes one complete path
length of the graded-index film, i.e.,

φi + 2φ0 + 2φxt = 2πm (3.52)
50 3 Dielectric Slab Waveguide

Fig. 3.12 Ray picture of propagation in a graded-index waveguide

where φ0 ≡ φx=0 and φT ≡ φx=xt . These values are obtained from the Fresnel equa-
tions for reflection at a dielectric interface. In this case, using standard formulae for
such a phase-shift and the assumption that n  n s gives the phase at the top and
the bottom of the trajectory,
π
φx=0 = (3.53)
2
π
φx=xt = (3.54)
4
In addition, phase accumulates during each segment of the path,

φi = k xi xi (3.55)

or, substituting for k xi

φi = kn(xi ) sin θi xi
 (3.56)
= k n 2 (xi ) − Neff
2
xi

where we have used the vector relation between k z and k xi as shown in the insert of
Fig. 3.12. The total phase change for a ray going through one entire trajectory is then
given by    xt
φi = 2k n 2 (xi ) − Neff
2
dx (3.57)
0

It is now possible to derive a normalized curve for a graded-index guide just as

seen for an abrupt waveguide. The equation is obtained by using (3.52), along with
the values for each of the phase terms. To obtain a useful expression, it is necessary
to give the depth-dependent index in terms of a characteristic distribution function
x
x
f = f (ζ), where we have defined ζ ≡ for simplicity. Then
d d
3.5 Graded-Index Waveguides 51

Fig. 3.13 The normalized

dispersion curve for
graded-index waveguide

n(x) = n s + n · f = n s + n · f (ζ) (3.58)

d
where d is the depth for which the distribution decreases to 1/e of its peak value;
this is sometimes called the diffusion depth.
For low index contrast, this can be written as

n 2 (x) ≈ n 2s + (n 2f − n 2s ) f (ζ) (3.59)

where n f is defined to be the maximum index in the waveguide. Thus n f ≡ n s + n.

Then, defining the normalized frequency for a graded-index (typically a diffused)
waveguide as 
V g = kd n 2f − n 2s (3.60)

We can write the phase equation (3.52), as

 xt /d 
g 3
2V f (ζ) − b dζ − π = 2mπ (3.61)
0 2

In addition we can define b, i.e., the normalized index, its form is the same as in
(3.24). This equation can be solved analytically for several cases of specific interest
for a diffused waveguide-dopant distribution. For instance, in the case of Ti-diffused
LiNbO3 , the index has a half-Gaussian distribution. In this case n ∼ e−ζ , where
2

again ζ ≡ x/d. For this case, Fig. 3.13 shows the normalized waveguide quantities
obtained by numerical solution of the normalized equation for the existence of modes
(3.61). This diagram is valid for both TE and TM waves, when n  n s , because
of the small birefringence of LiNbO3 crystals.
At cutoff, b = 0 and if we assume a large ratio xt /d  1, then
52 3 Dielectric Slab Waveguide
 +∞

−ζ 2 π
e 2 dζ = (3.62)
0 2
g,m
Thus the value of Vc for mode m at cutoff for a diffused guide is
 
√ 3
Vcg,m = 2π m + (3.63)
4

In comparison, recall that the same quantity for the wave in a slab guide is
√
Vcm = mπ + tan−1 a (3.64)

Finally, Nishihara et al. (1989) has also shown that an analogous ray-tracing
argument can be applied to a graded-index guide with a Gaussian distribution, a
geometry comparable to that seen for the transverse direction in a Ti-diffused channel
guide. This analysis yields the normalized frequency for the full Gaussian, V 2g ,
  
ζ  1
V 2g f (ζ) − b dζ = m + π (3.65)
0 2

for the normalized dispersion curve across a region of halfwidth d. Cutoff for the
g,m
mth mode of this Gaussian guide, Vc , occurs when b = 0, or
  
π 1
Vc2g,m = m+ (3.66)
2 2

3.5.2 Numerical Solutions

The advent of compact powerful digital computers has made numerical solutions
to the mode equation a practical and widely used approach. We will discuss the
numerical approaches later in the text in Chap. ?? and in addition, we give many
examples of computer solutions to integrated optical problems throughout this text.

3.6 Conclusion

We have accomplished several goals in this chapter. First, we have introduced the
physics and terminology of several forms of optical waveguides, including abrupt and
diffused guides. To keep the explanations clear, we have concentrated exclusively on
two dimensional or slab waveguides, since they have simple, closed-form solutions.
Secondly, we have, in the process, obtained several useful formulae and normalized
graphs The normalized graphs (and approximate formulae) are particularly useful
3.6 Conclusion 53

for simple approximate analytical solutions. However, perhaps the most important
consequence of this chapter is that it provides the theoretical underpinning and insight
for the far more complex but more realistic problem of three-dimensional waveguides,
which will be presented in the next chapter.

Problems

1. Consider a symmetric waveguide with y = 0.1 and λ = 0.84 µm (Fig. 3.14)

Fig. 3.14 Cross-sectional view of the symmetric waveguide

Find n f , n c , V , b, γ, for the TE0 mode

2. You fabricated a very wide Ti:diffused slab waveguide by starting with t = 80 nm
thick Ti and annealing for 6 h at 1000 ◦C

(a) Calculate the index profile at 0.6 µm.

(b) How many modes can this slab guide support at λ = 1.3 µm.

3. You make the following slab waveguide (Fig. 3.15):

Fig. 3.15 Cross-sectional view of a slab waveguide

(a) What are n s , n c and n f ?

(b) Sketch the shape of the lowest order modes
(c) How many modes are supported?
(d) How much of the lowest order mode is “in” the GaAs wafer?
54 3 Dielectric Slab Waveguide

4. Derive (3.23)
√  
V 1 − b = mπ + tan−1 b/(1 − b) + tan−1 (b + a)/(1 − b)

starting with (3.15).

5. Using simple physics intuitions, estimate the number of modes of the slab waveg-
uide below (Fig. 3.16).

Fig. 3.16 A slab waveguide

6. Using the modal propagation quantization equation, write an approximate expres-

sion for βm of the highest order mode, m. Assume the waveguide has n f  n c ,
λ = λ0 and width w.
Hint: first determine the transverse propagation constant k x as a function of the
mode number m and width w.
7. A symmetric slab waveguide is given with the parameters as shown in the fol-
lowing figure (Fig. 3.17).

Fig. 3.17 A symmetric slab

waveguide

(a) Design the waveguide with a thickness that is 50% of the cutoff for m = 1
mode. Find the width d.
(b) What are effective-index n eff and propagation constant β for the m = 0 mode
at the value of d you found in (a)?
(c) What is the evanescent decay length in the cover region?
References 55

8. Design a single-mode waveguide, i.e., 10% below m = 1 cutoff, using a buried-

channel waveguide architecture as below. Basically, you need to find out the value
of core thickness d and width w (Fig. 3.18).

Fig. 3.18 A buried-channel

waveguide

References

Haus, H. A. (1984). Waves and fields in optoelectronics. Upper Saddle River: Prentice-Hall.
Hocker, G., & Burns, W. (1975). Modes in diffused optical waveguides of arbitrary index profile.
IEEE Journal of Quantum Electronics, 11(6), 270–276.
Hutcheson, L. D. (1987). Integrated optical circuits and components: Design and applications. In
Optical engineering, No individual items are abstracted in this volume (Vol. 13, 417 p). New
York: Marcel Dekker, Inc.
Kogelnik, H. (1988). Theory of optical waveguides. In Guided-wave optoelectronics (pp. 7–88).
Berlin: Springer.
Nishihara, H., Haruna, M., & Suhara, T. (1989). Optical integrated circuits (Vol. 1). New York:
McGraw Hill Professional.
Pollock, C. R., & Lipson, M. (2003). Integrated photonics, (Vol. 20). Berlin: Springer.
Saleh, B. E., Teich, M. C., & Saleh, B. E. (1991). Fundamentals of photonics (Vol. 22). New York:
Wiley.
Snyder, A. W., & Love, J. (2012). Optical waveguide theory. Berlin: Springer Science & Business
Media.
Chapter 4
Three-Dimensional Waveguide

Abstract In general most PICs use three-dimensionally confined waveguides for

optical transport on an optical chip. Thus this chapter focuses on understating and
controlling the properties of these waveguides and on using these waveguides in
device-like applications. Although the basic principles of 3D waveguides are straight
forward to understand, precise analytic calculations of their properties are not easy
to obtain. However, a set of useful approximations and approaches have been devel-
oped, including designing for single-mode operation in both transverse directions,
the influence of waveguide cross section and materials choice on waveguide disper-
sion, and low loss designs. These design approximations can lead to many important
device ideas including mode-transformation devices and precise delay lines.

4.1 Introduction

While slab waveguides such as described in Chap. 3 are important for their ana-
lytic insight and are even used in some practical diffractive components, most PICs
use three-dimensional or channel waveguides for optical transport on the chip. In
addition, channel guides also form the heart of many complex PIC devices, e.g.,
precise delay lines or mode-transformation devices. Although the basic principles
of channel guides are readily grasped, exact analysis is much more complex than
for two-dimensional waveguides. While this chapter is meant to be very general in
discussions, of applications, references to specific materials types will be made in
several of the sections below.
There are several topics that are recurrent in the analysis of 3D waveguides includ-
ing realization of single-mode operation in both transverse directions, the influence of
waveguide cross section and materials choice on waveguide dispersion, and designs
that result in reduction in waveguide loss. Each of these topics will be covered in this
chapter.

© Springer Nature Switzerland AG 2021 57

R. Osgood jr. and X. Meng, Principles of Photonic Integrated Circuits,
Graduate Texts in Physics,
https://doi.org/10.1007/978-3-030-65193-0_4
58 4 Three-Dimensional Waveguide

4.2 Types of Channel Guides

There are several different types of channel waveguide geometries, all of which
clearly have three-dimensional characteristics. Some of the most common step-index
varieties are shown in Fig. 4.1; the comparative advantages of each step-index type
are given in Table 4.1. As the waveguide types move from strip to rib to strip-loaded,
the conditions for optical lateral confinement of the mode become progressively
more subtle. Also notice that in general n c ≤ n s < n f for all waveguides, while in a
strip-loaded guide, n c < n l < n f .
In addition to these 3D guides with abrupt rectangular geometries, three-
dimensional waveguides have also been made using graded-index structures. These
guides include Ti:diffused and proton-exchanged LiNbO3 and some SiO2 —and
polymer-based structures. The smooth features in graded-index guides reduce the
wall-scattering loss that occurs in many lithographically formed features.
In addition, in the last decade, extremely small (<1 µm wide) crystalline Si and
III–V waveguides have been fabricated using modern pattern-transfer methods; these
are termed “wires” because of their small lateral dimensions. While these guides can
have relatively high lineal loss, their tight confinement and high-index contrast can
allow them to be used in the fabrication of small-area PICs. In addition, improvements

Strip Waveguide Channel Waveguide

nc
nc
nf
nf
ns
ns

Ridge Waveguide Loaded Waveguide

nc
nc
nl
nf nf

ns ns

Fig. 4.1 Cross section of four of the more common step-index waveguides. Subscripts on the
indices are “s” substrate, “c” cladding, “f” film, and “l” loading strip
4.2 Types of Channel Guides 59

Table 4.1 Comparative advantages of certain common channel waveguides

Buried Ridge Strip loaded
• Smooth surfaces possible • Strong confinement • Weak confinement:
• Formed via ions, light or • Wall roughness 10−5 , 10−4
diffusion • Low bend loss • Substantial bend loss

in fabrication methods and new waveguide structures can be expected to make lineal
loss issues less important.

4.3 Modal Analysis of Three-Dimensional Waveguides

The analysis of three-dimensional or channel waveguides is very complex because

their modes are not pure TE or TM: they are hybrid, i.e., TEM. As a result, the
nomenclature for modes in a channel guide is more complex than in slab guides. A
mode having its electric field dominantly oriented along x, i.e., TM-like, is labeled
y
by E xpq , while a mode lying dominantly along y, i.e., TE-like, is labeled by E pq .
In this notation, p and q denote the number of modes in the x- and y-direction,
respectively. An example of the hybrid nature for a common rib waveguide is shown
in Fig. 4.2, where it is seen that electric field components lie along both x and y,
although E x is clearly dominant. Despite the hybrid nature, however, the simple
approximate terms TM and TE are usually retained, because there is typically one
dominant polarization, which allows a single polarization approximation.
The two possible polarization components in channel waveguide modes result
from the fact that the channel waveguide geometry couples the two transverse direc-
tions in the wave equation. As is shown below, approximate solutions using the
scalar-wave equation can be employed successfully if the mode is far from cutoff or

Ex Mode Profile (m=0, neff=2.43994) E y Mode Profile (m=0, neff=2.43994)

Fig. 4.2 Calculated plots of left: E x Mode Profile; right: E y Mode Profile
60 4 Three-Dimensional Waveguide

Fig. 4.3 The normalized

dispersion curve between
guide index (b) and
normalized frequency for
different w/ h. This plot
shows that when the ratio
between width and
thickness, w/ h, becomes
larger, the normalized
dispersion curve approaches
that for a slab waveguide

if the waveguide width, w, is greater than the thickness, t, or w/ h > 1. The crite-
rion of w/ h > 1 is a reasonable approximation since clearly, the greater this ratio,
the more the channel waveguide resembles a slab waveguide. This behavior is illus-
trated by the results of a numerical calculation on a buried-channel waveguide of
height t and width w, shown in Fig. 4.3. Notice that when w/ h > 2, the normalized
dispersion curve approaches that for a slab waveguide, i.e., that with w/ h → ∞.
The root of the difficulty in analyzing three-dimensional waveguides is the fact
that ∇ has both y and x gradients. These gradients introduce additional complexities
into the vector-wave equation by coupling both transverse components of the E and
H fields. In the presence of such a spatially varying dielectric, the wave equations
then become
∇ 2 E + ∇( E·
 ∇ ln ) + ω 2 μ0 E = 0 (4.1)

∇ 2 H + ∇(∇ ln ) × (∇ × H ) + ω 2 μ0 H = 0 (4.2)

These wave equations cannot be reduced to scaler forms and must be used in their
vectorial form. In essence, the (transverse) gradients in the waveguide prevent purely
TE- or TM-polarized guided waves.
Dividing E and H into their transverse (t) and longitudinal (z) components and
substitution in the wave equations yields

 ∇ ln ) + (ω 2 μ0 − β 2 ) E = 0
∇ 2 E + ∇( E· (4.3)

∇ 2 H + ∇(∇ ln ) × (∇ × H ) + (ω 2 μ0 − β 2 ) H = 0 (4.4)

Substitution yields

∇ 2 Et + ∇( Et · ∇ ln ) + (ω 2 μ0 − β 2 ) Et = 0 (4.5)

4.3 Modal Analysis of Three-Dimensional Waveguides 61

∇ 2 Ht + (∇ ln ) × (∇ × Ht ) + (ω 2 μ0 − β 2 ) Ht = 0 (4.6)

where the complete del, ∇, can be used, since all operations involving ∇z are zero.
Careful examination of these results shows that the gradient in  couples only the two
transverse components of E and H ; no longitudinal component appear. Further for
many waveguide types, the terms containing ∇ ln  are negligible. Once the transverse
components are obtained, the longitudinal components are readily found to yield
Maxwell’s equations

jβ E z = ∇· Et + Et · ∇ ln  (4.7)

and
jβ Hz = ∇· Ht (4.8)

Note that once Et (or Ht ) is determined, E z (Hz ) is found, thus, the complete E
and H field is available.

4.4 Approximate Methods—Generally for Slab-Like

(Rectangular) Waveguides

Several approaches are used to obtain approximate solutions for the dispersion and
fields of channel waveguides, including vector perturbation, separation of variables,
method of field shadows, and the effective-index method. We will concentrate on
the latter two methods in this section because of their immediate and widespread
applicability. More recently, numerical techniques have been developed, which are
extremely accurate and are generally widely applicable; they will be discussed in
Chap. 14 and we will see examples of their use throughout the book. In such compu-
tations, the effective-index method is also frequently used to convert a 3D numerical
calculations into a much simpler and faster 2D case. Numerical methods also have
the advantage of providing exact effectively solutions.

4.4.1 Method of Field Shadows

In principle, for certain geometries, it should be possible to apply separation of

variables to solve the 3D wave equation. This approach is useful in the limited cases
in which the permittivities, i.e., the square of the refractive index, can be shown to
be separable into two different orthogonal variables, e.g., n 2 (x, y) = n 20 + n 2x (x) +
n 2y (y). For example, it can be used to obtain an exact solution for the fields and
modes in the presence of a parabolic index distribution for which n 2 (x, y) = n 2f (1 −
(x 2 /x02 − (y 2 /y02 ))). However, this method is not widely applicable to most practical
62 4 Three-Dimensional Waveguide

Fig. 4.4 A plot depicting the basic geometry of the Method of Field Shadows from Marcatili
(1969). By ignoring the shadow region, the separation of variables can be realized

waveguide problems, for which this condition on the permittivity does not hold. Thus,
a more useful approach would be to develop a method which, while approximate,
allows separation of variables to be used more extensively. The “Method of Field
Shadows” is such an approach.
The basic geometry used in the method of field shadows (or as it is sometimes
called, “Marcatili’s Method”) is depicted in Fig. 4.4. The basic idea is to find the
geometric regions around waveguides that can be ignored, so as to allow the desired
separation of variables. In the rectangular geometry of Fig. 4.4, this separation can
be realized by ignoring the “shadow region.” This is a good assumption if light is
confined well in a waveguide, which means the waveguide is far from cutoff. This can
be achieved either by a high enough index contrast between the core and cladding,
or by a large enough transverse geometry. Note that this approach provides only an
fully scalar solution and, thus, is prima facie not exact.
This technique is illustrated by considering a rectangular buried channel (see
Fig. 4.4). In this case, the waveguide can be decomposed into two slabs in order to
seek two independent solutions, each slab corresponding to one of the two orthogonal
variables, x and y,
E(x, y) = X (x)Y (y) (4.9)

where X (x) is the field in the x-slab guide and Y (y) is the field in the y-slab guide.
This separation can be done in two steps: first, ignore the fields in the shadows,
and then divide geometrically the guide into two orthogonally oriented waveguides,
namely, waveguides lying along the x- and y-directions. The electromagnetic fields
in each of these two separate slab waveguides may be found exactly as indicated
earlier in Chap. 3. Specifically, the fields for each of the waveguides are cosinusoidal
in the central region, labeled 1 in Fig. 4.4,

X (x) = A cos(κx x + φx ) (4.10)

Y (y) = B cos(κ y y + φ y ) (4.11)

and exponentially decaying in regions 2, 3, 4, and 5 with an appropriate decay

constant γi for region i. For the solutions of region 1 to be valid, it is required that
4.4 Approximate Methods—Generally for Slab-Like (Rectangular) Waveguides 63

β 2 = k 2 n 21 − κ2x − κ2y (4.12)

again, notice that the orthogonal slab waveguide approach is only an approximate
solution, since the fields in the corner regions are not then properly accounted for.
In the second step, divide the refractive index of the channel guide to allow sep-
arating out of the two slab waveguides. At this point, in order to provide a more
specific example, consider the case of an embedded symmetric channel waveguide.
Thus, regions 2, 3, 4, and 5 have the same index, n s , and the central region has index
n 1 = n f . In this case, two independent slab waveguides can then be constructed if
each of the waveguides has an embedded slab with n 2f /2 and cladding/substrates of
n 2s − n 2f /2. When summed, these two decomposed guides yield that of the original
buried rectangular guide, except for the corner or shadow regions. However, the fact
that these regions contain relatively small fields in many cases allows this to be a
useful method.
Using these two artificial index slabs, eigenvalues of the wave equation can be
solved, just as for the standard slab waveguides. Because the solution is now a product
of X (x) and Y (y), β 2 = k 2 n 2f − κ2x − κ2y , and thus β 2 = βx2 + β y2 and N 2 = N x2 +
N y2 , where βx , β y , N x , N y are the propagation constant and the effective index for
the x- and y-slabs, respectively. This suggests that it is possible to use a variant of the
normalized dispersion variables, including its graphical presentation, that used in 2D
waveguides. The normalized guide frequency of the two component slab waveguides
is then 
Vx = hk n 2f − n 2s ≡ V (4.13)

and  ω
Vy = kω n 2f − n 2s = V (4.14)
h
Notice that by convention, the thinnest waveguide dimension, denoted by h here,
is typically used to specify the overall guide V number. This choice is reasonable,
since the thinnest dimension sets the lowest guide modal cutoff.
The normalized guide index can then be written in both the x- and y-direction as

N x2 − n 2s + n 2f /2
bx = (4.15)
n 2f − n 2s

and
N y2 − n 2s + n 2f /2
by = (4.16)
n 2f − n 2s

where the convention has been adopted that in this chapter, that the symbol N implies
effective index. Since the above relations for V and b are for two slab waveguides,
these waveguides can be designed using the normalized dispersion curve provided
in Chap. 3.
64 4 Three-Dimensional Waveguide

Fig. 4.5 The normalized

dispersion curve between
guide index (b) and
normalized frequency for
different values of m. Note
that the waveguide shows
cutoff before V = 0

Then, since
N 2 − n 2s
b= (4.17)
n 2f − n 2s

and
N x2 + N y2 = N 2 (4.18)

we find
b = bx + b y − 1 (4.19)

Thus the normalized dispersion curve for the two slab waveguides can be used
to find the quantity, b, for the channel waveguide. Once the vertical slab V (=Vx ) is
found, it is also possible to construct a normalized waveguide curve for the chan-
nel guide; however, note that unlike the curve for a simple vertical slab, this curve
is specific to the channel guide geometry. A plot of b versus V for this symmet-
ric channel waveguide, given in Fig. 4.5, shows that this waveguide exhibits cutoff
before V = 0. Since the waveguide is symmetric, we anticipate, on the basis of our
slab waveguide results, that such a symmetric waveguide with proper design can
always allow the fundamental mode to operate. This result indicates that for low V,
Marcatili’s method is no longer accurate. Further comments on its accuracy will be
provided in Sect. 4.4.3.
As mentioned earlier, this method allows the field distribution to be obtained as
well! These fields are obtained by solving for the slab modes for each of the two
waveguides separately and then taking the product to find the final mode shape. For
example, for the above problem, the field distributions of Hy (x, y), namely, the E 00
x
,
x
and E 01 mode (TM-like), appear as shown in Fig. 4.6.
4.4 Approximate Methods—Generally for Slab-Like (Rectangular) Waveguides 65

Fig. 4.6 The calculation of field distribution for Hy (x, y), E 00

x , and E x . These distributions are
01
obtained by solving the slab modes for each of the two waveguides separately

4.4.2 Effective-Index Method

The effective-index method is easily implementable even for comparatively unusual

guides. It is, thus, in many ways the “workhorse” technique for designing channel
waveguides. Again, the approach is to decompose a three-dimensional guide into two
intersecting two-dimensional guides. The essential idea is to replace the indices of
the sometimes complex layered waveguide structure with a single, effective index.
The specific implementation of this idea varies significantly from one waveguide
geometry to another, but the basic approach is the same. To be specific, let us start
with an example, using an abrupt waveguide geometry.
The steps in the approach are shown in Fig. 4.7 for a buried strip guide. As we will
(n f − n s )
find, the method works best for  1 (i.e., weak confinement) and away
nf
from cutoff. The assumption of weak confinement also allow us setting bT M = bT E
for TM modes.
In step I, the same procedure as for any slab waveguide is used;  for example,
for the TM-like mode (i.e., dominant E x ), VI is found from VI = hk n 2f − n 2s . The
normalized dispersion curve, along with the appropriate asymmetry factor, am =
(n f /n c )4 (n 2s − n 2c )/(n 2f − n 2s ) may then be used to obtain b I . The value of b I could
also be determined by using the explicit expression for the normalized TM dispersion
curve for a slab waveguide. The effective index can be determined directly from b I

NI = n s + b I (n 2f − n 2s ) (4.20)

where the definition of normalized index, given earlier, has been used.
66 4 Three-Dimensional Waveguide

Fig. 4.7 An example of the basic steps of implementing the effective-index method on a buried
strip guide. The material quantities are those defined earlier in the book

In step II, a “new” slab waveguide core of index N I and with cladding of n s is
constructed. The TM-like mode for guide I is now TE-like for this second waveguide.
Again, the normalized dispersion curve is used to derive VI I , then b I I . For this
symmetric slab guide with TE modes, a = 0 and the effective index in the slab is
given by 
VI I = kω N I2 − n 2s (4.21)

At this point in the analysis, b I I , the normalized thickness of the slab, can be
found either by substituting in the normalized eigenvalue equation for a symmetric
waveguide, 
 
1 −1 1 − bI I
VI I = √ (m + 1)π − 2 tan (4.22)
1 − bI I bI I

or by using the graphical form b I I . N I I (or β = k N I I ), is determined by


NI I = n 2s + b I I (N I2 − n 2s ) (4.23)

Since the effective-index guide is symmetric, when cutoff occurs at b I I = 0, VI I =

qπ.
A very complete tabulation of the effective-index method for a wide variety of
channel waveguide structures is presented in Kogelnik (1988) and a selection of
4.4 Approximate Methods—Generally for Slab-Like (Rectangular) Waveguides 67

these waveguides are given here in Table 4.2. For example, the application to ridge
waveguide structures reveals the calculation of the effective index for the vertical
slab in both the ridge and adjacent waveguide regions. The equivalent b and V of
the horizontal slab waveguide is then calculated using the effective indices from this
calculation.
The effective-index method is an excellent technique for obtaining the dispersion
relation for a channel waveguide. In addition, the method can also be used to find
the field distribution using the two orthogonal waveguides. However, since a simple
“slab” approximation is being made to otherwise complex geometries in some cases,
e.g., the ridge waveguide, the detailed shape of the distribution will sometimes be at
variance with the actual modal shape. Still, the “effective-index” fields will give at
least some indication of the average spatial extent of the fields.

4.4.3 Accuracy of Approximate Techniques for Channel

Waveguides

It is useful to compare the accuracy of these methods with the generally exact calcula-
tions done with modern numerical techniques. A very interesting study in that regard
has been described by Hocker and Burns (1977) for a buried rectangular channel
waveguide and is shown in the two panels of Fig. 4.8. These graphs in these panels
plot a comparison of calculations based on “exact” numerical techniques with results
obtained using the field shadow and the effective-index methods for two different
ω/t ratios. The buried-channel waveguide was chosen to allow direct comparison
with the results given earlier in Fig. 4.3 for both slab and channel waveguides.

Fig. 4.8 The comparison of

the accurate calculations and
the two approximate
techniques. The solid blue
line represents the exact
calculation while the red and
green dashed lines follow the
field-shadows method and
the effective-index method,
respectively
68 4 Three-Dimensional Waveguide

Fig. 4.9 Left: Mode profile calculated using the numerical Beam Propagation Method (BPM).
Right: Mode profile calculated with effective-index method

In Fig. 4.8, the effective-index method uses V = VI ; it also employs the results
given in Table 4.2 for the expression for the overall b of a buried-channel waveguide
given in Table 4.2, namely, b = b I b I I . In addition, the figure also shows the results
of the field-shadows method that uses the normalized index of the two orthogonal
slabs to obtain b, i.e., b = bx + b y − 1.
The results show that the effective-index method is the more accurate of the two
approximate techniques; however, the field-shadows method becomes increasingly
more accurate as ω/t increases. Note also that both methods are most accurate
far from cutoff and for low mode numbers. The effective-index method tends to
overestimate the normalized index, while the field-shadows method underestimates
the same geometry. This underestimation occurs because the field distribution spreads
over a significant portion of the corner or “shadow” regions. Although the effective-
index method does lead to an overestimation of b near cutoff, it at least gives the
correct reason that one mode is always present in a symmetric channel waveguide.
This fact, plus its easy implementability, causes it to be a common tool in integrated
optics design.
Finally, all of these approximate techniques do have one important inadequacy:
they are incapable of determining any waveguide modulation of the transverse mode
field distribution. In some waveguides, such as the buried-channel waveguide, this
field “fine” structure might not be significant. However, in a surprisingly large number
of practical 3D waveguides, such as rib guides, the modulation in the vertical or
horizontal field can be substantial. The understanding of this structure is crucial for
predicting even the “zeroth-order” performance of devices such as couplers or filters.
Figure 4.9 shows the simulation results of this effect in the field of a rib waveguide
with its effective-index equivalent.
The actual plots of field intensity cannot, in fact, be obtained by approximate
methods. Instead, careful numerical calculation must be used. These methods will
be presented in Chap. 14.
Table 4.2 Effective index parameters for channel guides
Structure Guide height Effective index N I2 − N I2I Channel guide index b

General VI = kh n 2f − n 2s N I2 = n 2s + b f (n 2f − n 2s ) b f (n 2f − n 2s ) − bl (nl2 − n 2s ) b f b I I + bl (1 − b I I )ach

VI I = kh nl2 − n 2f N I2I = n 2s + b f (n 2f − n 2s )

Buried VI = kh n 2f − n 2s N I2 = n 2s + b f (n 2f − n 2s ) b f (n 2f − n 2s ) b f bI I
N I I = ns

Raised VI = kh n 2f − n 2s N I2 = n 2s + b f (n 2f − n 2s ) (n 2s − n 2c ) + b f (n 2f − n 2s ) b f b I I − (1 − b I I )a
N I I = Nc

Rib VI = kh n 2f − n 2s N I2 = n 2s + b f (n 2f − n 2s ) (b f − bl )(n 2f − n 2s ) b f b I I + bl (1 − b I I )

VI I = kl n 2f − n 2s N I2I = n 2s + bl (n 2f − n 2s )

Embedded VI = kh n 2f − n 2s N I2 = n 2s + b f (n 2f − n 2s ) b f (n 2f − n 2s ) b f bI I
N I I = ns

Ridge VI = kh n 2f − n 2s N I2 = n 2s1 + b f (n 2f − n 2s1 ) (1 − bl )(n 2s1 − n 2s2 ) + b f (n 2f − n 2s1 ) b I I (1 + b f aridge ) + bl (1 − b I I )

VI I = kl n 2f − n 2s N I2I = n 2s2 + bl (n 2s2 − n 2s2 )
4.4 Approximate Methods—Generally for Slab-Like (Rectangular) Waveguides
69
70 4 Three-Dimensional Waveguide

4.5 Wavegiude Dispersion for Channel Guides

Thus far, we have been using the scalar-wave equation, plus the matching of boundary
conditions, to obtain the field amplitudes in the waveguide. Note that within this
scalar approximation, there should be no difference in modal propagation constants
because this equation, i.e.,

∇ 2 ψ(x, y) + (n 2 k 2 − β02 )ψ(x, y) = 0 (4.24)

where ψ is the scalar field, does not distinguish the polarization state of the field
 ∇ln) term in
amplitude. Instead, dispersion arises from the inclusion of the ∇( E·
the wave equation:

 ∇ ln ) + (n 2 k 2 − β 2 ) E = 0.
∇ 2 E + ∇( E· (4.25)

While including this transversely varying dielectric constant in the wave equation
makes its solution much more difficult, it is possible to use a perturbation method to
gain insight into its behavior. For example, perturbation theory can be used to obtain
the shift in propagation constants, β, due to waveguide dispersion. The perturbation
treatment finds the dispersive shift in β by comparing the initial scalar-wave equation,
i.e., that without the vector perturbation, with the perturbed full vector-wave equation.
In the presence of this perturbation, E = E0 + E1 , where E0 is the solution to the
scalar-wave equation.

βT2 E = β02 + βT2 E (4.26)

βT2 M = β02 + βT2 M (4.27)

where β0 is obtained from the scalar-wave equation, i.e., with no ∇ ln  term. Using
the difference between the scalar and perturbed equation and (4.26) and (4.27), and
neglecting the second-order perturbation effects, the following shifts are obtained
for the TE- and TM-like modes:

 ∂2
ψ2 (ln ) dxdy
1 ∂ y2
βT2 E =  (4.28)
2 ψ 2 dxdy

 ∂2
1 ψ2 (ln ) dxdy
∂x
2
βT2 M = (4.29)
2 ψ 2 dxdy

where ψ(x, y) is the modal field from the scalar equation. As an example, the appli-
cation of this approach to a parabolic dielectric profile, characterized by half widths
of dx and d y yields
4.5 Wavegiude Dispersion for Channel Guides 71

 2  
λ 1 1
βT2 E − βT2 M − 4 . (4.30)
πn f 4
dx dy

Finally recent developments in Si and Si fabrication (see Chap. 2) allow Si

nanowire waveguides to be made. These waveguides have extremely high-index con-
trast at edges and this high contrast causes additional factors in waveguide design to
be considered; this phenomenon has particularly important effects for dispersion in
the waveguide (Chen et al. 2006). Thus, the strong quartic dependence of the square
of the propagation constant seen in (4.30)suggests that structural dependence is a
major issue for tightly confined waveguides. Thus, structural dispersion is particu-
larly important in the case of Si-wire waveguides, due to the tight confinement of the
mode in these guides. An example of this effect is seen in Fig. 4.2, which compares
the material and structural contributions to the dispersion of a Si-wire waveguide
for a wire with a cross section of 0.2 × 0.5 µm. As is readily seen, the dispersion
is dominated by the structural component. The dominance of structural dispersion
is typical for Si wires. In addition, its importance causes small variations in the
width or height of the wires during fabrication to affect the chromatic response of
many Si waveguide devices, particularly devices with delay lines. These dispersion
effects also are of major concern when designing waveguides for nonlinear optical
experiments, for which phase matching is crucial.

4.6 The Design of Single-Mode Channel Waveguides

The performance of an integrated optical circuit often depends critically on main-

taining single-mode operation in the waveguiding system. For example, higher order
modes cause enhanced loss in waveguide bends and alter the performance of a waveg-
uide coupler. The procedure for designing single-mode devices usually proceeds by
decomposing the problem into two dimensions through the use of the effective-index
method. This approach has been nicely discussed and presented by Nishihara et al.
(1989).
In order to illustrate the design process, consider a simple step-index rectangular
waveguide such as discussed in earlier sections. This waveguide geometry is, for
example, very important in the case of very high contrast Si waveguides such as the
Si wires, mentioned above, and for the related high-index clad waveguides in some
III–V materials.
To design such a waveguide structure, both TM- and TE-like modes must be
considered. As mentioned in previous chapters, this process is made more difficult
since, in principle, the normalized quantities b and a are much more complex for the
case of TM-like modes. If, however, the waveguide has low index contrast, then it
suffices simply to substitute the quantity am for a in the equation for the dispersion
relation, (3.23), since bT E = bT M .
72 4 Three-Dimensional Waveguide

Fig. 4.10 A plot of the

solution to the normalized
dispersion curve for
single-mode operation of
w/ h versus VI normalized
frequency. This graph shows
that a single-mode
waveguide can be made
inside the shadow area

To proceed with an example, consider the buried step-index guide shown earlier
x
in Fig. 4.7 for the case of a TM-like mode, i.e., E 00 mode. The first step consists of
adjusting the height of the vertical slab, waveguide I, so that it is above cutoff for
m = 0 and below that for m = 1, that is, VI is such that
√ √
π + tan−1 am > VI > tan−1 am (4.31)

where VI = hk n 2f − n 2s . In cases where the lateral adjacent regions also have
waveguiding properties, these regions may or may not be in themselves single mode;
it is only important that the central “core” region be single mode for this design
procedure.
In the second step, the “effective-index” symmetric slab is considered, and thus
a = 0. In this case,
0 < VI I < π (4.32)

Substitution then gives

ω π
0< <√ (4.33)
h b I VI

Notice that this equation shows that guide I, and hence, b I and VI , determine
the width-to-thickness ratio for single-mode behavior. In applying this equation, the
dispersion give for single lowest order mode related the value of b I for each value
of VI . The solution to single-mode operation for w/ h versus VI has been given in
graphical form by Nishihara et al. (1989) see Fig. 4.10.

4.7 Graded-Index Channel Waveguides

Thus far we have considered only step-index waveguides, however, channel guides
can also be made in a graded-index medium, with a frequently used example being Ti-
diffused waveguides. In this case, the effective-index method can again be used. In our
4.7 Graded-Index Channel Waveguides 73

Fig. 4.11 A plot of index vs. spatial location for a 3D graded-index waveguide, the quantity dx,y
indicates the diffusion depth in x- and y-direction. As in the case for 3D abrupt-index waveguides,
the effective-index method can still be used

discussion here, we will again follow the excellent discussion given by Nishihara et al.
(1989). The method decomposes the problem into vertical and lateral waveguides,
just as in the case of 3D abrupt-index waveguides; see Fig. 4.11. First, a laterally
uniform but vertically graded slab waveguide with a Gaussian index distribution is
g
assumed in order to find the normalized frequency VI of this vertical guide and, then,
the conditions for single-mode confinement.
To determine the modal properties of the guide, the eigenmode equation is used
for a graded-index guide given in Chap. 3. Specifically, for single-mode behavior,
g
the value of VI must be greater than the cutoff value for m = 0 and less than the
same for m = 1, or, using (3.62),

3√ g 7√
2π < VI ≤ 2π (4.34)
4 4
g
Once VI is fixed, the guide diffusion depth, dx , is found from the usual definition
g
of VI ,
g
V
dx =  I (4.35)
k n 2f x − n 2s
74 4 Three-Dimensional Waveguide

g
In addition, this value of VI can be used in (3.22) to find the effective-index N I of
the vertical guide, if this quantity is needed for further waveguide design calculations,
e.g., computation of dispersion, etc.
Next, the dimension for the lateral guide, shown in Fig. 4.11, has to be found.
To do this, the guide is approximated with a symmetric Gaussian distribution. The
lateral waveguide diffusion depth can be written in terms of the normalized frequency
2g
VI I and the effective index of the vertical guide:

2g
V
dy =  I I (4.36)
k N I2 − n 2s

This equation shows clearly that the design of the vertical waveguides couples to
that of waveguide II, the lateral guide. The condition for a single propagating mode
is then given by the graded-index eigenmode equation, but this time for a symmetric
Gaussian distribution, with m = 0.

π 2g π
< VI I ≤ 3 (4.37)
8 8
g
This equation can be rewritten by dividing through by VI and inserting b I to obtain an
equation which gives the range of the ratio of the lateral, d y , to vertical, dx , distance
g
needed for a specific vertical waveguide design specified by either b I or VI :

π 1 dy π 1
< ≤3 (4.38)
8b I VIg dx 8b I VIg
g
Alternately, this inequality can also be used to determine VI if the ratio d y /dx is
given. As a specific example, consider the important case of d y /dx = 1, i.e., diffusion
from a single “line source.” Substituting this value in (4.38) and manipulating the
equation gives

π g π
< b I VI < 3 (4.39)
8 8

This equation, or its equivalent for other ratios of d y /dx , thus provides a second
g g
set of contrasts on VI above that given in (4.37). To find the values of VI given in
this new inequality, we use the normalized

dispersion curves shown in Fig. 4.12 to
g
determine which value of the product b I VI is valid. This gives the solution

g
√
VI ≈ k0 d x 2n (4.40)

where n ≡ n f − n s , then we can rewrite this inequality as the following criteria

for a single-mode 3D waveguide:
4.7 Graded-Index Channel Waveguides 75

Fig. 4.12 Normalized dispersion curve for graded-index waveguide

 

dx
0.26 < n s n ≤ 0.39 (4.41)
λ

Notice that n and dx play an essential role in setting the single-mode condition.

4.8 Summary

The complexity of 3D waveguide analysis, even for relatively simple geometries,

provides a clear motivation for eventual use of numerical analysis. These techniques
will be described in Chap. 14. However, the methods used in this chapter, especially
the effective-index method, are surprisingly useful even in the case of numerical
analysis since their employment can accelerate the design of complex integrated
circuits.
At this point, the text will begin to analyze the optical properties of multiple
waveguides both in simple coupled systems and in more complicated devices.

Problems

1. Use the effective-index method to determine the dimensions needed for single-
mode operation in the following hetero-structure waveguide:
76 4 Three-Dimensional Waveguide

The lowest order mode should be TM in the vertical direction and TE in the hor-
izontal direction, as we have shown in Chap. 4. Design each dimension in the
figure (i.e., w, d, and h) so that it’s 20% smaller than those needed for cutoff
of the m = 1 mode in the y- and x-directions (i.e., ddesign = 0.8dcutoff ). Assume
h = 2 µm.
2. The same as for Problem 1, except using a strip Si3 N4 /SiO2 structure as below.

3. Consider a Ti-diffused waveguide with two Gaussian distributions in the x- and

y-directions (a half and a full), as shown below.
References 77

Design for single-mode operation (find dx and d y ) using the same 20% rule and
polarization as in Problem 1. As a reminder, the index of Ti-diffused waveguide
satisfies following relation:
⎧
⎪ x2
⎨ n x = n s + n · e− dx2
⎪ x2
⎩ n = n + n · e− d y2
y s

4. Show that the following expression is valid under conditions of low index contrast:

Neff ≈ n s + b(n f − n s ).

References

Chen, X., Panoiu, N. C., & Osgood, R. M. (2006). Theory of raman-mediated pulsed amplification
in silicon-wire waveguides. IEEE Journal of Quantum Electronics, 42(2), 160–170.
Hocker, G., & Burns, W. K. (1977). Mode dispersion in diffused channel waveguides by the effective
index method. Applied Optics, 16(1), 113–118.
Kogelnik, H. (1988). Theory of optical waveguides. Guided-wave optoelectronics (pp. 7–88). Berlin:
Springer.
Marcatili, E. A. (1969). Dielectric rectangular waveguide and directional coupler for integrated
optics. Bell System Technical Journal, 48(7), 2071–2102.
Nishihara, H., Haruna, M., & Suhara, T. (1989). Optical integrated circuits (Vol. 1). New York:
McGraw Hill Professional.
Chapter 5
General Introduction to Coupled-Mode
Theory

Abstract This chapter presents a formalism for showing that small changes in the
structure of an optical waveguide structure can couple optical fields, as they propagate
through the structure. This approach is an effect akin to the perturbation theory seen,
say, in quantum physics; formally this approach is termed as coupled-mode theory.
Coupled-mode theory can be used with a wide variety of geometries or hetero-
materials and has thus been used to analyze periodic structures or a small isolated
surface dent or defect. When waveguide coupling is present, a new set of normal
modes of the coupled system are formed, which are termed as system modal fields.
This coupling can be expressed in terms of the uncoupled fields and the parameters
of waveguide geometry. In addition, this chapter will also consider a variation of the
coupled-mode theory, which is written for a longitudinally varying system.

5.1 Introduction: A Simple Example of Mode Coupling

Coupling of modes is illustrated clearly by the example of the optical fields in two
adjacent waveguides. Coupling occurs because the modal field in one waveguide
overlaps with the dielectric medium of the other. Specifically, as shown in Fig. 5.1, the
field, E a , of waveguide a is perturbed by the dielectric step, which forms waveguide
b. Coupling between modes allows power to flow from one waveguide to another.
The power transfer is maximized when the two modes are nearly synchronous with
each other, i.e., when propagation velocities along the waveguides are comparable.
When waveguide coupling is present, a new set of normal modes of the coupled
system are formed; Fig. 5.1 shows these “system” modal fields, E o,e . These can
be expressed in terms of the uncoupled fields E a and E b and the parameters of
waveguide geometry. For symmetric waveguides, the power in each waveguide varies
sinusoidally with distance (see Fig. 5.2). This behavior can also be interpreted as a
periodic interference between the fields of a set of even and odd system modes, E e
and E o , each propagating with slightly different propagation constants, namely, βe
and βo .
Finally note that our discussion thus far appears superficially to focus on a simple
waveguide structure. However, in fact, the coupled-mode equation applies to a much
© Springer Nature Switzerland AG 2021 79
R. Osgood jr. and X. Meng, Principles of Photonic Integrated Circuits,
Graduate Texts in Physics,
https://doi.org/10.1007/978-3-030-65193-0_5
80 5 General Introduction to Coupled-Mode Theory

Fig. 5.1 When the modal field in one waveguide overlaps with the dielectric medium of the other,
coupling occurs. This figure shows the case that one waveguide couples with another identical
waveguide and forms two new modal fields E o and E e . Here, “o” and “e” stands for “odd” and
“even,” respectively

Fig. 5.2 Plot showing symmetric waveguides, the power in each waveguide varies sinusoidally
with distance

wider variety of guided-wave structures such as diffraction gratings or waveguide

arrays. We will discuss some of these applications later in this chapter and elsewhere
in this book.

5.2 General Coupled-Mode Formulation

The coupled-mode equations relate the first derivative of one mode amplitude to
the amplitude of a second mode via a coupling coefficient, typically denoted by
Greek letter κ. These equations have been derived via several routes. One particular
common approach is to use the Helmholtz equations in conjunction with the slowly
varying wave approximation (Saleh et al. 1991; Hutcheson 1987).
However, a more rigorous and general approach to the coupled-mode equations,
using modal expansions of the fields and vector manipulation of Maxwell’s equa-
5.2 General Coupled-Mode Formulation 81

tions, has been described in the three references (Marcuse 1974; Kogelnik 1988;
Nishihara et al. 1989). Elements of this straightforward derivation will be touched
on and outlined here; more extensive references are found in the references. The
starting point in the derivation is to examine a guided-wave structure and choose the
canonical structures, the modes of which are only “perturbed” in the presence of the
complete structure with perturbations, even major changes. For example, in Fig. 5.1,
the canonical structures are the waveguides and the complete structure is the coupler.
The transverse field of the canonical structure, labeled 1, may then be expanded in
modes of the guide, 
E1t = (a + − 
p + a p ) Et p (5.1)

H1t = (a + − 
p + a p ) Ht p (5.2)

where a + −
p and a p are functions of z (along the guide) and denote the amplitudes of
the forward- and backward-running field, respectively. The functions Et p give the
transverse field, i.e., x distribution. It is typical in many integrated-optic problems
to normalize the waveguide modes such that the power is set equal to unity, which
in our case will be 1W/cm2 . This is a particularly important point when calculating
practical quantities such as the coupling coefficient.
Mode coupling occurs through the presence of a change or perturbation in the
dielectric constant. The change may be scalar,

P =  E (5.3)

or tensoral
Pi = i j E j (5.4)

The former allows coupling among like (e.g., TE) modes. The latter allows mixing
between TE and TM modes as well. Mode conversion via a tensor dielectric change is
important in magneto- and electro-optical devices, and for nonlinear optical effects.
The coupled-mode equations may then be written in a form such as

daq+
+ jβq aq+ = κq p a +
p (5.5)
dz

where for simplicity, the case of coupling only between two forward-moving modes,
q and p, has been considered. A more complete expression including both forward
and backward waves can be obtained (Kogelnik 1988). In many practical cases, only
coupling between two modes is sufficient to provide a rigorous treatment of the
device operation. However, clearly the inclusion of only a limited number of modes
could, in some cases, limit the accuracy of the solution.
Below, we will show that on the basis of simple considerations, relations between
certain coupling coefficients can be derived.
82 5 General Introduction to Coupled-Mode Theory

5.3 Practical View of Coupling of Modes

The coupled-mode equations given above, i.e., in the form such as (5.5), allow the
use of a very simple and more practical approach for solving coupling of modes in
waveguide structures. Before examining this approach, it is important to lay out an
explicit expression for the mode amplitude with a complete expression for its field.
The complete expression for the harmonic field at frequency ω, say in one mode, is


E(x, y, z, t) = at (z)e− jβ1 z f1 (x, y)e jωt (5.6)

where at (z) is the (slowly varying) wave amplitude of the wave labeled by 1, β1
is its propagation constant, and f1 (x, y) is the transverse distribution of the mode.
Notice, in this case, that a purely transverse field has been assumed. This assumption
will be followed throughout this chapter and, except where it is noted or obvious,
throughout the book. In some cases, it is more instructive to lump the amplitude
and the r-dependent-phase term into one term, which is designated by a1 (z); note
that the field is designated without a prime superscript in this case: notice that this
notation was used in the previous section. In this formulation, the slowly varying
spatial dependence of the modal power can be obtained through |a1 |2 , etc.
If modes are labeled by 1 and 2, and they are weakly coupled via coupling coef-
ficients κ12 and κ21 , two coupled-mode equations will describe their intertwined
spatial variation,
da1
= − jβ1 a1 + κ12 a2 (5.7)
dz

da2
= − jβ2 a2 + κ21 a1 (5.8)
dz

The normalization conditions are such that for weak coupling, and |a1 |2 and |a2 |2
represent the powers carried by these particular waves or modes.
Haus (1984) has introduced the notation which gives the total power flow, P, from
these two modes as
P = p1 |a1 |2 + p2 |a2 |2 (5.9)

where p1,2 = ±1 depending on whether the power flow is forward or backward. This
notation thus allows the expectation that the coupled modes can be copropagating or
counterpropagating. Each of these cases will be considered below. In the absence of
power loss or gain, the total power flow of the modes should be such that

dP
=0 (5.10)
dz

and thus
d|a1 |2 d|a2 |2
p1 + p2 =0 (5.11)
dz dz
5.3 Practical View of Coupling of Modes 83

The fact that

d|a1 |2 da ∗ da1
= a1 1 + a1∗ (5.12)
dz dz dz

da1∗ da1
plus the coupled-mode equations for and can then be used to show that
dz dz

p1 κ12 + p2 κ∗21 = 0 (5.13)

Thus power conservation yields a fixed relation between κ12 and κ∗21 , the two
coupling coefficients appearing in (5.7) and (5.8), and depends on the direction of
propagation through the factor p.

5.3.1 Propagation Constants for the Coupled Modes

The coupled-mode equations, (5.7) and (5.8), can now be solved to find the new

propagation constants, β1,2 , of the coupled system. For example, in the coupled
waveguide example given in Sect. 5.1, these “system” modes would contain even and
odd modes shown in Fig. 5.1, having propagation constants β  ≡ βe or βo . Setting
the determinant of the coupled-mode equations equal to zero, we find

(β − β1 )(β − β2 ) + κ12 κ21 = 0 (5.14)

This equation yields the solutions, or eigenvalues, for the propagation constants
of the system modes,

β1 + β2 (β1 − β2 ) 2
β  1,2 = ± − κ12 κ21 (5.15)
2 2

Consider first the case of copropagating waves. In this case, p1 p2 = +1 and then
(5.13), plus some algebraic manipulations, show that κ12 κ21 = −|κ12 |2 . Thus, β1 and
β2 are always real, and two propagating solutions are obtained. Coupling between
the waveguides is only substantial if β1 ≈ β2 , or, specifically, coupling is important
when |β1 − β2 | ≈ |κ12 | .This is the condition for the two waves to be close to phase
matching or synchronous. It is of interest to examine how the system or coupled
modes vary as the difference in the propagation constants of the isolated waveguide
modes, β1 and β2 , evolve. For example, if the isolated waveguide β1 and β2 vary
slowly and linearly with ω, near the point where β1 = β2 , then the system modes

β1,2 will behave as shown in Fig. 5.3. See also the discussion concerning (5.26) and
(5.27), below. Note that there is an avoided “crossing” in the dispersion plot of the two
system modes as a result of the modal coupling term. Also the propagation constants
approach the uncoupled or isolated waveguide values for |β1 − β2 | > |κ12 |. This
84 5 General Introduction to Coupled-Mode Theory

Fig. 5.3 The dispersion

relation when β1 and β2 vary
slowly and linearly with ω,
near the point where
β1 = β2 . The blue lines
represent β1 and β2

behavior is due to the fact that in the limit where the waveguides are sufficiently
asynchronous that their fields do not interact. The use of coupling, over several
propagation lengths, between two waveguides is important in a variety of integrated
optical devices such as switches and routers.
For the case of counterpropagation modes, the same eigenvalue equation (5.14)
can be applied. In addition, p1 p2 = −1 and then κ12 κ21 = −|κ12 |2 . By inspection
of (5.15), this constraint implies that the two waves will have a complex propagation
constant for certain values of |β1 − β2 |, i.e., when

β1 − β2
< |κ12 | (5.16)
2
In this range, the two modes will exhibit attenuation as they propagate into the
coupling region; that is, propagating waves do not exist. The dispersion curve for
counterpropagating modes, having different uncoupled group velocities is shown in
Fig. 5.4. In the center region of this figure, propagation does not occur and thus an
“optical” bandgap in the grating structure exists. This behavior is also typical of
periodic structures such as those used in Bragg reflectors. Photonic crystals are the
same type of periodic structure, but have a larger coupling constant, κ.

5.3.2 Power in the Coupled Systems

The power in coupled systems can be found by solving the appropriate coupled-mode
equations, while taking into account the appropriate initial conditions. For example,
for the case of the copropagating waves, we use the coupled-mode equations given
in (5.7) and (5.8) as well as the modal propagation constants given in (5.15). With
5.3 Practical View of Coupling of Modes 85

Fig. 5.4 The dispersion

curve for counterpropagating
modes, which have different
uncoupled group velocities

this starting point, we can show that for the case of excitation of wave 1 only, i.e.,
a1 (0) = 1, a2 (0) = 0, then
 2
|a1 (z)|2 κ
=1− sin2 βc z (5.17)
|a1 (0)|2 βc

and  2
|a1 (z)|2 κ
= sin2 βc z (5.18)
|a1 (0)|2 βc

where   2
β1 − β2
βc = κ2 + (5.19)
2

and thus  2
κ 1
= (5.20)
βc 1 + (/κ)2

If  ≡ (β2 − β1 )/2. In these equations, we have also assumed that κ12 = κ21 and
that κ is real.
Plots of the power coupling curve in the two guided waves is shown in Fig. 5.5.
The use of |a1,2 (z)|2 removes the rapidly varying, β-dependent phase from the length
variation. Notice that the power coupling is repetitive over a characteristic length
π
given by L c = . For example, for synchronous waveguides, β1 = β2 and L c =
2βc
π
. Also, the maximum possible power transfer, from |a1 |2 to |a2 |2 , is given by
2κ
the ratio (κ/βc )2 ; this term is unity at the phase-matching point, β1 = β2 , and 0 for
|β1 − β2 |  κ.
86 5 General Introduction to Coupled-Mode Theory

Fig. 5.5 Plots of the power coupling curve in two guided waves. Notice that the coupling power
is repetitive over a characteristic length given by L c = π(2βc ). The left is when β1 = β2 and thus
|a1 |2
|a1 |2 = |a2 |2 . The right is when |β1 − β2 | = 6.36κ and thus = (κ/βc )2 = 0.09
|a2 |2

In the case of counterpropagating waves, a different set of equations is obtained.

Further, because the waves are oppositely directed, it is not possible to have phase
matching in the presence of a simple non-periodic coupling term, as was the case
for copropagating modes. If, however, the coupling does have a periodic spatial
dependence, phase matching can be obtained for counterpropagating waves if this
spatial periodicity has a Fourier component equal to |β1 − β2 |. Specifically, if the
waves are coupled through a spatial periodic structure, introduced by, say, a surface
diffraction grating on a single waveguide, then the coupled-mode equation is written
as
da1
= − jβa1 + κ12 a2 e− j K z (5.21)
dz

da2
= jβa2 + κ21 a1 e+ j K z (5.22)
dz

and where K ≡ 2π/,  is the spatial period of the structure, and thus K is the
wavenumber of the periodic coupling element.
In these equations, it is assumed for the sake of simplicity that β1 = β2 = β; that
is, we examine coupling between the forward and backward waves in a single-mode
synchronous structure fabricated on the waveguide. This simple scheme is common
in many practical counterpropagating devices, for example, in a structure where
a grating is etched into the surface of a waveguide. The analogous coupled-mode
equations for the case in which β1 and β2 are different but can also be solved; it is,
however, a straightforward but tedious exercise.
These equations can be reduced to a form analogous to those for the copropagation
coupling by the substitution

a1 (z) = A1 (z)e− j K z/2
(5.23)
a2 (z) = A2 (z)e− j K z/2
5.3 Practical View of Coupling of Modes 87

where A1 (z) is the field amplitude that varies even more slowly than a1 (z) by remov-
ing any field variations at the grating period. This substitution gives
 
d A1 K
= −j β − A1 + κ12 A2 (5.24)
dz 2
 
d A2 K
= j β− A2 + κ21 A1 (5.25)
dz 2

Notice that these equations are identical to those encountered earlier for codirec-
tional coupling (see (5.7) and (5.8), with the substitution of β1,2 → ±(β − K /2).
Also note that for counterpropagating waves, hence by (5.13), κ12 = κ∗21 . Thus by
analogy, phase matching occurs when 2(β − K /2) = 0, or when β = K /2. Note
also that the coupling coefficient κ12 does not include the grating spatial variation.
The fact that we have removed the grating periodicity by adopting (5.23) also means
that in calculating the coupling coefficient, κ12 , we must also remove the grating
spatial variation in the modal fields.
Depending on the device, either β or K can be tuned to achieve a specific func-
tionality, i.e., switching, filtering, etc. A simpler notation can be used which will
be important for examining the behavior of a periodic structure in the presence of a
wavelength-tuned optical source. Specifically, let β be expanded in a Taylor’s series
about the resonant or synchronous frequency for the grating,

dβ
β β(ω0 ) + (ω − ω0 ) (5.26)
dω
ω − ω0
= β(ω0 ) + (5.27)
υg

where we have used the fact that dω/dβ = υg is the group velocity. We then introduce
the detuning of β from the resonant spatial period of the periodic structure,

K
δ≡β− (5.28)
2
ω − ω0
≡ (5.29)
υg

This substitution allows the coupled-mode equations to be rewritten as

d A1
= − jδ A1 + κA2 (5.30)
dz

d A2
= jδ A2 + κA1 (5.31)
dz
88 5 General Introduction to Coupled-Mode Theory

Fig. 5.6 The efficiency of

the transfer of power from
forward a1 to backward a2
propagation in the coupler
sketched in the inset

In these equations, A1 is the amplitude of the forward-propagating mode and A2

the amplitude of the backward-propagating mode. Again, solving the determinant of
these equations gives the propagation constants of the coupled modes:


β1,2 = ± δ 2 − |κ|2 (5.32)

The coupled-mode equations can then also be used to obtain the expressions for the
modal power as a function of distance along a periodic coupler of length L.
To obtain these expressions, we assume A1 (z = 0) = 1 and A2 (z = L) = 0, and
find for the power of the forward and backward waves,

|A1 (z)|2 1 + (κ/βd )2 sinh2 [βd (z − L)]

= (5.33)
|A1 (0)| 2 1 + (κ/βd )2 sinh2 βd L

and
|A2 (z)|2 (κ/βd )2 sinh2 [βd (z − L)]
= (5.34)
|A1 (0)|2 1 + (κ/βd )2 sinh2 βd L
√
where βd = κ2 − δ 2 .
Figure 5.6 shows a plot of these functions for several values of the normalized
length βd L at δ = 0. It shows that as the length increases, the efficiency of the
transfer of power in the backward direction increases. Also, notice that along the
propagation path, i.e., from left to right in the figure, both A1 and A2 decrease along
the periodic structure, since A1 is gradually attenuated by reflection and A2 , the
reflected wave, is the sum of all reflections along z. The reflected power or power
transfer, |A2 (0)|2 /|A1 (0)|2 , from A1 to A2 is 99% when L = π/κ; in general, the
reflected power for no propagation mismatch δ = 0 is

|A2 (0)|2
= tanh(|κ|L) (5.35)
|A1 (0)|2
5.3 Practical View of Coupling of Modes 89

Fig. 5.7 Copropagating mode coupling between the two symmetric slab waveguides

Thus a characteristic coupling length can be defined in the contradirectional case

as well: L c = π/2κ.
Finally, note that while grating-assisted coupling is usually considered in a con-
tradirectional context, a grating can also be used to compensate for differing prop-
agation constants in codirectional coupling. In that case, codirectional coupling can
be accomplished with a grating of spatial frequency K , if β1 − β2 ≈ K . Thus codi-
rectional coupling requires a larger spatial period than contradirectional coupling.
Interestingly, as mentioned earlier in the chapter, the κ for codirectional coupling
is typically stronger than for counterdirectional coupling because of the sign of the
small but finite κz term, which acts against κt for contradirectional coupling. As will
be discussed in a subsequent chapter, the power transfer characteristics for codirec-
tional coupling in a grating are similar to those for a directional coupler.

5.4 Two Specific Examples of Coupling Coefficients

We can illustrate the calculation of the coupling coefficient for two of the more
important cases of the application of coupled mode theory: coupled waveguides and
the coupling of forward and backward waves in a grating.
Consider first copropagating mode coupling between the two symmetric slab
waveguides shown in Fig. 5.7. Using the fields for slab waveguides obtained from
Chap. 3, it can be shown that for these two slab waveguides,

j2γk x2
κ12 = exp(−γ(h − d)) (5.36)
β(k x2 + γ 2 )(d + (2/γ))

In these expressions, we have used the expression k x rather than κ as used in

Chap. 3 to denote the transverse wavevector in the waveguide and to avoid confusion
with the coupling coefficient. Also, γ is the evanescent transverse wavevector in the
slab cladding. Notice the exponential dependence of the coupling on the separation of
waveguides, h − d; this strong, nonlinear dependence makes precise-value couplers
hard to make.
90 5 General Introduction to Coupled-Mode Theory

Fig. 5.8 A grating

waveguide with a simple
spatial variation
d(z) = d0 + d cos K z,
where K = 2π/

For a grating, the index perturbation is spatially periodic.Thus, let us assume here
a simple spatial variation d(z) = d0 + d cos K z as shown in Fig. 5.8. As mentioned
earlier, this spatial periodicity allows the phase-matching condition (i.e., the waves
are synchronous) 2β ≈ K to be achieved for forward and backward moving waves
in the grating coupler.
To obtain the coupling coefficient, we use the known field distributions for a
slab waveguide, given in Chap. 3, and assume TE radiation. The presence of this
surface corrugation alters locally the dielectric constant around x = d such that if
d(z) > d0 , the effect is to add a small volume of dielectric with  = 0 (n 2f − n 2c ),
while if d(z) < d0 , the dielectric change is negative and of opposite magnitude  =
0 (n 2c − n 2f ).
These fluctuations are substituted in (5.22) and (5.23) along with the appropriate
modal fields. Since the field is TE, the longitudinal coupling coefficient of (5.23) is
zero, i.e., κz = 0. However, the transverse coupling coefficient is given by

κt = E y2 dx dy (5.37)

If the grating height corrugation is small compared to γc−1 in the cover or k x−1 in the
waveguide dielectric, then E y (d(z)) ≈ E y (d0 ), a constant independent of integration.
The coupling coefficient, including the sinusoidal z dependence, κ , is then

κ = κt = jω0 E y2 (d0 )(n 2f − n 2c )d cos K z (5.38)

However, as mentioned earlier in this chapter, the usual formulation of the coupled-
mode equations removes the grating-induced rapid spatial variation from the modal
fields (see Sect. 5.3.2), to obtain the usual coupled-mode κ for gratings. In addition,
the value of E y in (5.38) can be related to the maximum field of the waveguide mode
via the relations derived earlier in Chap. 3, e.g. (3.36),

E 2f (n 2f − Ne2f f ) = E c2 (n 2f − n 2c ) (5.39)

where E f is the maximum field in the guided mode and E c is the field at the cover
interface and equals E y (d0 ). Recall from the initial discussion of the modal expression
in Sect. 5.2 that the electric fields used are normalized by setting the waveguide power
equal to 1. This power is obtained from the Poynting vector:
5.4 Two Specific Examples of Coupling Coefficients 91

0 2
P = Ne f f E Te f f (5.40)
μ0 f

1 1
where the average effective thickness of the slab waveguide is Te f f = d0 + + .
δ γ
Then, setting
κ ≡ κe− j K z (5.41)

an expression for the coupling coefficient is realized,

π d n f − Ne f f
2 2
κ= j (5.42)
λ Te f f Ne f f

5.5 An Alternate Approach to Obtaining the Coupling

Coefficient

The general coupled-mode theory given earlier in Sect. 5.2 made it possible to derive
a rigorous and very general expression for the coupling coefficient. In order to gain
more insight in the physical nature of the coupling, we consider a more direct but
simplified approach to the equation for the coupling coefficient. This approach yields
the coupling coefficient, κi j via a derivation based on general power flow arguments;
the discussion of the approach is made more accessible by considering the spe-
cific geometry of two symmetric coupled waveguides (Haus 1984). A sketch of this
waveguide geometry is shown in Fig. 5.9. Physically, the coupling from waveguide
1 into waveguide 2 is due to the polarization current of the dielectric in waveguide
2 which is caused by the presence of a field in waveguide 1. If f1 (x, y) is the nor-
malized transverse field distribution of waveguide 1 and a1 its amplitude (see (5.6)),
d P21
then the cross waveguide polarization current is, by definition,
dt

d P21 d( P21 (x, y) · e− jωt )

= = − jω P21 = jω( f − )a1 f1 (x, y)e− jωt (5.43)
dt dt
where  is the cladding dielectric outside of the waveguides,  f is the dielectric in
guide 2 and where we have subtracted off the current, which would exist in the
absence of waveguide 2.
Using the Poynting vector, the power transferred can then be determined to be
 
1 jω
E2∗ · (− jω) P21 dx dy = a1 a2∗ ( f − ) f1 · f2∗ dx dy + c.c. (5.44)
4 A 4 A

where A is the area of the waveguide. But we know from coupled-mode theory (see
above) that
92 5 General Introduction to Coupled-Mode Theory

Fig. 5.9 A sketch of two symmetric coupled waveguides: the mode propagates in left waveguide
overlap with the right waveguide

d|a2 |2
= κ21 a1 a2∗ + κ∗21 a1∗ a2 (5.45)
dz

Equating the left-hand sides of (5.44) and (5.45) thus shows that

jω
κ21 = ( f − ) f1 · f2∗ dx dy (5.46)
4 A

Notice that this equation is a simplified version of the general equations given
above in Sect. 5.2. In addition, it is easy to show that

jω
κ12 = ( f − ) f2 · f1∗ dx dy (5.47)
4 A

A somewhat more general form of this equation is then


jω
κi j = (x, y, z) fj (x, y) fi∗ (x, y) dx dy (5.48)
4 A

where i, j are mode labels and  ≡  f − . In many cases, this equation can be
directly applied to determine κi j .

5.6 Summary

This chapter has introduced the formulation of coupled-mode theory and has shown
its utility in describing the operating physics of several common device structures for
integrated optics, i.e., a waveguide coupler and a diffraction grating (with in plane
propagation). The discussion here has focused on simple problems, which involve
the coupling between the two lowest order modes. In many cases, however, coupling
schemes can be more elaborate, including coupling of higher order modes in devices
such as mode sorters and switches. These may be approached theoretically using the
5.6 Summary 93

same principles as discussed above. In the next chapter, a more thorough discussion
of devices for simple optical couplers will be presented.

Problems

1. Using the expression for κ12 between two slab waveguides (5.36), calculate h
needed to have L c = 300 µm. Assume n slab = 3.34, n cover = 3.30,λ = 1.3 µm.
First, Fix d, the slab thickness so that the V parameter is 20% below cutoff for the
m = 1 mode. Note that this type of coupler is used for vertical coupling in III-V
devices.
2. Use the effective index method to estimate h for a polymer synchronous coupler
of length 3 mm.

(a) Design the height of the waveguide to be single-mode (TM-like) so that V is

20% below Vcm=1
(b) Design the width to be single mode using same Vc criterion (the mode is now
TE-like, of course)
(c) Find the spacing h
3. What is the maximum power transfer for a codirectional coupler if (β2 − β1 ) ≈
3κ?
4. Derive (5.36) using (5.48) and the modal fields of the slab waveguide.
5. Use (5.42) and the waveguide in Problem 1 to design the modulation depth for a
grating coupler with κ = 2 mm−1 .
6. A coupler made with two symmetric slab waveguides to have coupling coefficient
κ. If an optical wave of power of magnitude P enters waveguide 1 at z = 0, try
to answer the following questions.
(a) Plot the power in waveguide 1 and 2 versus distance z in the following two
cases:
(i) β1 = β2
(ii) β1  β2
where β1 and β2 are the mode propagation constants, λ is the wavelength.
Use simple words to describe the differences between the two situations.
94 5 General Introduction to Coupled-Mode Theory

(b) For the case of two synchronous waveguides, sketch the two lowest SYSTEM
modes. Write down the expression for each of the two propagation constants
for the two modes.
7. Given two simple slab waveguides as below:
(a) Assume d1 = d2 = d. Calculate d so that each slab waveguide has a single
TE mode that has a normalized frequency of 20% below the m = 1 mode
cutoff. Assume λ = 1.55 µm.
(b) If guides are spaced to have a coupling coefficient of κ = 50 cm−1 , what
is the waveguide length L for the minimum power transfer from guide 1 to
guide 2? Assume guide 1 is excited.
(c) If guide 2 now has d2 = 0.7d, what is the length L for maximum transfer,
assuming κ stays the same?

References

Haus, H. A. (1984). Waves and fields in optoelectronics. Upper Saddle River: Prentice-Hall.
Hutcheson, L. D. (1987). Integrated optical circuits and components: Design and applications.
Optical engineering (Vol. 13, 417 p.) No individual items are abstracted in this volume. New
York: Marcel Dekker, Inc.
Kogelnik, H. (1988). Theory of optical waveguides. Guided-wave optoelectronics (pp. 7–88). Berlin:
Springer.
Marcatili, E. A. (1969). Dielectric rectangular waveguide and directional coupler for integrated
optics. Bell Labs Technical Journal, 48(7), 2071–2102.
Marcuse, D. (1974). Theory of dielectric optical waveguides (Google Scholar, pp. 181–193). New
York: Academic.
Marcuse, D. (1991). Theory of dielectric optical waveguides.
Nishihara, H., Haruna, M., & Suhara, T. (1989). Optical integrated circuits (Vol. 1). New York:
McGraw Hill Professional.
Saleh, B. E., Teich, M. C., & Saleh, B. E. (1991). Fundamentals of photonics (Vol. 22). New York:
Wiley.
Yariv, A. (1973). Coupled-mode theory for guided-wave optics. IEEE Journal of Quantum Elec-
tronics, 9(9), 919–933.
Chapter 6
Optical Couplers

Abstract The goal of this chapter is to examine in detail the practical side of inte-
grated optical couplers. Thus, for example, these couplers are fabricated of lithium
niobate via surface diffusion from a local titanium source or those fabricated of Si,
including Si wires. The chapter will also include consideration of the deviation of the
actual coupler geometry and materials components from that of the ideal waveguide.

6.1 Introduction

Optical couplers are one of the most important classes of integrated optical com-
ponents. These devices are used in directional routing of a light signal from one
waveguide to another or in dividing a signal between different lightguides. However,
even more important applications rely on the fact that the amount of coupling can
either be varied, as in various forms of electro-optical switches or on its sensitivity
to certain property of the optical signal, e.g., its wavelength, intensity, or tempo-
ral properties. The latter function is the basis of wavelength routers or nonlinear
switches.
In this chapter, we will discuss passive optical couplers. The discussion will
include a consideration of both conventional and adiabatic, or spatially varying,
couplers, as well as their practical implementation. This chapter also presents an
excellent opportunity to consider in more detail practical examples of the coupled-
mode theory in working device structures. Finally, this chapter is the first in this book
to introduce the realization of a practical integrated device, namely, different forms
of couplers, and thus we illustrate using a variety of couplers including coupled rings,
vertical-coupled laser sources, and power dividers. Note that numerical calculations
are also shown.

© Springer Nature Switzerland AG 2021 95

R. Osgood jr. and X. Meng, Principles of Photonic Integrated Circuits,
Graduate Texts in Physics,
https://doi.org/10.1007/978-3-030-65193-0_6
96 6 Optical Couplers

6.2 The Standard Two-Waveguide Directional Coupler

A backbone design step for integrated photonic circuits is the simple 2×2 coupler.
Our approach to obtaining an analytic solution for these devices follows closely that
of Nishihara et al. (1989). In essence, such a coupler consists of two waveguides—
either channel or slab—that are placed close enough so that the optical fields are
coupled from one waveguide to another by virtue of the evanescent field beyond the
confinement region. In general, the derivations assume weak coupling between the
individual waveguide modes and nearly identical waveguides. Deviations from these
assumptions will be discussed in Sect. 6.3.
The equations used for this discussion yield results identical to those in Chap. 5.
However, in this chapter, we chose an alternative approach based on using the primed
amplitudes, a1 and a2 , that is with the rapid variation due to the optical carrier
phase factored out, since this approach is common in the literature, see Nishihara
et al. (1989), for examples. Second, this approach also develops several practical
parameters such as that for field detuning and the separation of the two system
modes.
For simplicity’s sake, in our calculations, we assume that κ1,2 = κ∗1,2 = κ. If we
then factor out the rapidly varying portion of the envelope, i.e., a1 (z) = a1 (z)e− jβ1 z ,
we have

da1 (z)
= − jκa2 (z)e− j (β2 −β1 )z (6.1)
dz

da2 (z)
= − jκa1 (z)e j (β2 −β1 )z (6.2)
dz

We now seek the solutions for these equations by finding the normal modes of
the coupled system; that is, we wish to find the system modes of the combined two-
waveguide device. The eigenmodes may be solved for the general case of input to
both waveguides by assuming that the solutions have the form

a1 (z) = a1 e− jβc z e− j(z) (6.3)

a2 (z) = a2 e− jβc z e+ j(z) (6.4)


√
where a1,2 are constants and  ≡ (β2 − β1 )/2, βc = κ2 + 2 .
Insertion of these solutions into the coupled-mode equations (6.1) and (6.2) yields
the solutions for fields in this device. Note that the mode amplitude in waveguides 1
and 2 can then be written as the sum of two independent orthogonal modes, which
have even (e) or odd (o) character as indicated by their subscripts. Alternate versions
of these equations appear in the literature; see, for example, the text by Haus (1984)
or Chap. 5 of this book.
6.2 The Standard Two-Waveguide Directional Coupler 97

Finally, the general fields for the individual waveguide modes can thus be
expressed clearly and neatly as the sum of the system modes
 − jβe z  − jβo z
E 1 (x, y, z, t) = −[a1e e + a1o e ] f 1 (x, y)e jωt (6.5)

 − jβe z  − jβo z
E 2 (x, y, z, t) = −[a2e e + a2o e ] f 2 (x, y)e jωt , (6.6)

which have propagation constants

βe,o = βm ± βc (6.7)

where
βm = (β1 + β2 )/2 (6.8)

The symbols f 1,2 in (6.7) and (6.8) represent the normalized field distribution
in the individual guides. Notice that the expressions for βe,o are identical to the
eigenvalues of β found in the formal derivations in previous chapter. Since βc → κ
as the propagation constant of the two isolated waveguides β1 and β2 become equal,
the waveguide coupling eliminates the degeneracy in the β’s for the synchronized
waveguides βe and βo when coupling is present.

The amplitudes of the system modes, a1e and a1 0 , are determined by the boundary
conditions. Thus for light entering waveguide 1, i.e., a1 (0) = 1, a2 (0) = 0, at z, the
 
amplitudes for a1e and a1o may be found via substitution in (6.5) and (6.6) to, yielding
 
a1 (z) and a2 (z) be
 
 − jz 
a1 (z) = e cos βc z + j sin βc z (6.9)
βc
 
jκ
a2 (z) =e jz
− sin βc z (6.10)
βc

As was shown in Sect. 5.3.2, these initial conditions lead to solutions having a
sinusoidal power transfer along z between each of the individual waveguides,

|a1 (z)|2 κ2
= 1 − sin2 βc z (6.11)
|a1 (0)|2 βc2

|a2 (z)|2 κ2
= sin2 βc z (6.12)
|a2 (0)|2 βc2

and thus κ2 /βc2 is the maximum fraction that can be coupled from waveguide 1 to
waveguide 2. For example, if κ = 0, the optical wave sensibly remains in guide 1
(a1 = 1), for all z! Figure 6.1 shows a plot of the power in each of the two guides
98 6 Optical Couplers

Fig. 6.1 A plot of the power in each of the two waveguides for two values of 

for two example values of . Maximum power transfer occurs when z max ≡ L c =
π/(2βc ), or when L c = π/(βe − βo ).

6.3 Transfer Matrix for a Coupler

The equations obtained in the preceding section suggest a simplified formulation that
is useful for the overall design of a complex optical circuit, namely that of transfer
matrices. The transfer matrix of a device captures the effect of an ideal device on
its input waves. While this formulation is not intended to present a detailed realistic
electromagnetic simulation of the actual device, it does allow for the rapid design
of a large number of device elements via simple serial multiplication of the transfer
matrix of each element since the formulation does realize accurately the overall
properties of the device. Thus, while the limitation on transfer matrices is that such a
matrix only captures the overall behavior of a specific, generally idealized geometry;
it does not enable one to make a detailed exploration of the device geometry.
In this chapter, we use a typical 3 dB coupler as an example and show how to
determine its transfer matrix. For a two input device, the transfer matrix allows the
two inputs, a1i and a2i , to be presented in a compact, easily implemented vectorial
manner and related via the matrix to the outputs, a1o and a2o .
 o   i 
a1 u 11 u 12 a1
= (6.13)
a2o u 21 u 22 a2i

The matrix elements can be obtained by manipulating the solutions obtained in the
previous section for the case of arbitrary initial boundary conditions and by returning
to the original variables a1 (z) and a2 (z). It is then found that u 22 = u ∗11 , u 21 = −u ∗12 ,
and that

u 11 = cos(βc L) + j sin(βc L) (6.14)
βc
6.3 Transfer Matrix for a Coupler 99

and
κ
u 12 = j sin(βc L) (6.15)
βc

for a coupler of length L. As an example, if a1i = 1 and a2i = 0, the solutions for a1o and
a2o recover (6.11) and (6.12), once the transformation a1 → a1 and a2 → a2 is made.
This matrix allows the phases and amplitudes of both outputs of the ideal directional
coupler to be easily manipulated. Throughout the remainder of this chapter, transfer
matrices of several other simple devices will be introduced.

6.4 More Exact Treatment of Waveguide Couplers

The traditional coupled-mode theory for optical couplers is essentially a perturbation

theory. Its accuracy depends on weak waveguide coupling and near synchronization
of the individual waveguide propagating modes. It is possible, however, to use more
exact analyses, based on variational methods or a more exact treatment of the modes,
to see that this approach fails as the validation of the above conditions begin to degrade
as discussed by Hardy and Streifer (1985), Haus et al. (1987, 1989). Finally it is also
possible to use numerical methods to obtain an essentially exact treatment of the
coupler as long as there is also an exact understanding of the waveguide geometry
and materials.
Consider, for example, two nonidentical waveguides as depicted in Fig. 6.2a. The
fields of these guides have been calculated using the conventional coupled-mode
analysis given above and by a more exact theory, which deals more carefully with
radiative modes (Hardy and Streifer 1985). A comparison of these calculations is
given in Fig. 6.2b for the two lowest order system modes. Clearly, there is an important
discrepancy between the two calculations. A calculation of the system-mode prop-
agation constants of the coupled waveguides versus thickness also shows important
discrepancies. The values obtained with the more complete theory differ signifi-
cantly from those obtained with traditional coupled-mode theory at smaller spacing.
Note, however, that the difference in the propagation constants in both the more
sophisticated and the traditional theory is extremely small except at the very smallest
spacings. Radiative effects, which are not shown here, are, however, more accurately
calculated by the more exact theory.
An elegant variational theoretical approach has also been applied to this same
problem (Haus et al. 1987). The results of this treatment obtained similar numerical
results as that by Hardy and Streifer (1985). However, the variational treatment shows
clearly that the form of the traditional coupled-mode formalism can be recovered by
carefully reorganizing the system modes of two waveguides. This step is done by a
more complete expansion of the system modes in terms of the individual waveguide
modes. Further, as mentioned earlier, Haus et al. (1987) has pointed out that the
dominant effect on the modal propagation constants is a simple uniform shift in βo
100 6 Optical Couplers

Fig. 6.2 (Left) Sketch of two nonidentical waveguides. (Right) A comparison of the fields of these
guides calculated using the conventional coupled-mode analysis given above and by a more exact
theory [Adapted from (Hardy and Streifer 1985)]

Fig. 6.3 The effect of

uniform shift in βo and βe Exact
can be shown with several Snyder et al.
different versions of CMT with cross power
coupled-mode theory (CMT) Modified CMT with
[Adapted from (Haus et al. cross power
1989)]

and βe ; this shift is accurately calculated by standard coupled-mode theory. Thus the
coupling length in this case remains the same, since

βe − βo = 2κ12 (6.16)

as in conventional theory; see Fig. 6.3 for an example. Finally, an extension of weak
coupling theory has also been formulated, using the variational method, applied to
vector coupled-mode theory. Again, in this case, the coupling length is that given by
the first-order theory (Haus et al. 1989).
6.5 Three Examples of Actual Working Couplers 101

6.5 Three Examples of Actual Working Couplers

6.5.1 Lateral LiNbO3 Couplers—A Study of the Sensitivity to

Fabrication

In actuality, the fabrication of good couplers is often a difficult process. Part of the
difficulty is a result of the extreme sensitivity of directional couplers to the exact value
of the coupling coefficient, which makes them difficult to fabricate reproducibly by
standard lithographic techniques. In fact, even slight variations in etch depth and
lithographic dimensions can yield serious variation in the coupling length, since
the coupling coefficient depends exponentially on waveguide separation and on the
detailed profile of their evanescent fields, as given by γ. This is a particularly impor-
tant issue for diffused-waveguide-based couplers since the extent of the junction is
a result of both lithography and thermal diffusion.
A second issue, which was discussed in the previous section, is that the close
proximity of waveguides, particularly those with weak confinement, means that each
guide may perturb the mode of the other, or that inadvertently the guides coupling
may cross over from the weakly coupled limit, and becomes strongly coupled or even
a single waveguide. We will illustrate these considerations below, using the example
of graded-index (LiNbO3 ) guides. Similar studies have been done on high-index-
contrast semiconductor waveguides.
Graded-index-based couplers have been examined in detail, including theoretical
(WKB-based) and experimental studies for the case of Ti:LiNbO3 systems (Noda
et al. 1981). The goal in these studies was to understand clearly the dependence
of coupling length on the variation of waveguide parameters, the transition region
between weak and strong coupling, and the sensitivity to fabrication tolerances. In
addition, the perturbation to the waveguide parameters by the presence of neighboring
guide was also important.
Diffused waveguides clearly have a 3D character and thus their analysis can be
complex (see Fig. 6.4). However, not surprisingly, these couplers are most sensitive
to the variation in the y- (or in-plane) direction. For example, the coupling length of
two weakly coupled waveguides can be written as
λ
Lc = (6.17)
2(Ne − No )

Regions with indiffused Ti

Ti

LiNbO3

Fig. 6.4 Steps in forming Ti:LiNbO3 . Due to the three-dimensional diffusion and indiffused Ti, its
complex to analyze the characters such as refractive index and loss
102 6 Optical Couplers

Fig. 6.5 A plot of measured

variation of the refractive
index in the direction at the
surface for a LiNbO3 coupler

where Ne,o are the effective indices for the odd and even system modes. However,
in practice, it is found that the coupling length can be approximated by
λ
Lc = y y (6.18)
2(Ne − No )

i.e., using the effective indices for the y slab waveguide only.
The measured variation in the refractive index in the direction at the surface for
a LiNbO3 coupler is shown in Fig. 6.5, along with the physical parameters of one
example of this structure. In the figure, the location of the outer turning point for the
individual waveguide modes are shown in addition to the point at which light leaks
across the coupling region. It is obvious from the figure that the index profile of the
two guides, which are Gaussian in the isolated guide limit, significantly perturb each
other on their “inner” side.
The figure also can be used to quantify the transition between weak and strong
coupling. For the device to act as a weak coupler, light must be confined in the core
region of either of the two waveguides. Thus, using the symbols defined in Fig. 6.5,

n s + n > Ney , Noy > n s + n c (6.19)

y y y
Since Ne > No , the light is no longer weakly confined when Ne < n s + n c . In
this case, the device acts like a single waveguide.
Some of the issues associated with making such graded-index guides are illustrated
by a series of calculations on coupling performance (Noda et al. 1981) done using
index profiles, such as those shown in Fig. 6.5. One example is shown in Fig. 6.6,
which displays the coupling length, L c , versus spacing between waveguide centers,
h, for a series of values of n. The figure shows the extreme sensitivity of L c on n,
and, hence, on n. In fact, the slope of the linear region of the curve in the figure for
n = 0.006 indicates a value of d L c /dh of 6 mm/µm. At short coupling lengths,
the coupling becomes strong. This region is denoted by a dashed-dotted line. As
6.5 Three Examples of Actual Working Couplers 103

Fig. 6.6 A plot of the

relation between coupling
length L c and waveguide
spacing h, taking
refractive-index change n
as the varying parameter

Fig. 6.7 A plot of the

relation between coupling
length L c and
refractive-index change n,
taking waveguide spacing h
as the varying parameter

suggested by the discussion in the previous section, it is not possible to use a simple
effective-index approach to calculate the coupling in this region.
A second calculation shows the dependence of coupling length on index contrast,
n, for a series of fixed waveguide spacings. These results are given in Fig. 6.7.
Again, the region of strong coupling, which below the dot-dash line, is indicated
in the figure. Notice that the coupling length decreases to a very short length for
small n. While this can be desirable for certain applications, it comes as a result of
increased dimensional sensitivity. A related concern is wavelength sensitivity, which
results from the variation in effective index with wavelength. These calculations,
displayed in Fig. 6.8, show that the wavelength sensitivity of L c decreases as the
wavelength increases.
Although not shown here, experimental data have been compiled on Ti-diffused
couplers and generally show excellent agreement with the calculated response; that
104 6 Optical Couplers

Fig. 6.8 A plot of

wavelength dependence on
coupling length, taking
refractive-index change n
as the parameter

is, direct measurements of index contrast are in good agreement with those inferred
from the measurements of the coupling length versus waveguide spacing.
The results above show clearly the sensitivity of the coupling to fabrication errors
in waveguide spacing. An important consequence of this fabrication error is that it
can cause increased cross talk in the device. In this case, cross talk is calculated by the
ratio of the light-intensity output in guide 1 versus that in guide 2, where complete
crossover to guide 2 is desired. The tolerance in coupler length, L, required to
attain varying values of cross talk for different coupling lengths is shown in Fig. 6.9.
The plot shows that smallest crosstalk tolerance is encountered in the case of short
coupling lengths. Once the tolerance in L is obtained at a given L c , the tolerances in
waveguide spacing, n, etc., can be calculated using the standard waveguide equa-
tions. For example, a typical value for the tolerance in waveguide spacing to achieve
20 dB cross talk and waveguides with ∼ 13 µm separation, dx ≈ 5.0 m, n = 0.005,
and L ≈ 10 mm is 0.5% or 65 nm.

6.5.2 Lateral Si Wire Couplers

Several approaches have been developed to make a more reliable and robust coupler
technology. One obvious possibility is to make an active coupler by electro-optical
tuning or post-process trimming. Of course, this approach raises the complexity and
cost of the entire fabrication process and the final device. A second approach is to use
a vertical coupler; that is, separate the waveguide by epitaxial layers since such layers
can be grown with atomic precision and thus the coupling length can be fixed more
accurately than with lithography alone. Finally, of course, the increasing resolution
6.5 Three Examples of Actual Working Couplers 105

Fig. 6.9 A plot of normalized propagation constant difference vs coupling length at each value of
the crosstalk

Fig. 6.10 Top and cross-sectional views of a Si wire waveguide directional coupler

and fabrication tolerance available from the patterning tools for advanced Si ICs has
steadily improved lateral coupling position.
As mentioned earlier in this text, the Si wire is a basic device technology for
Si photonics and thus Si have been extensively studied. Thus, one investigation
examined the possibility of obtaining very compact couplers since the Si wire bending
radius can be very small, i.e., a several micrometers, due to their very large refractive
index contrast between the silicon core and the SiO2 cladding, which confines the
optical field within the device.
In another study (see Fig. 6.10), directional couplers were fabricated and their fun-
damental characteristics were measured. Thus, an SOI wafer was used, which had
a 0.3 µm thick crystalline Si layer on a 1 µm buried SiO2 layer on the Si substrate.
Standard experimental planar processing was used including electron beam lithogra-
phy for pattering and dry etching for pattern transfer. After patterning the waveguide,
the device was clad with 1 µm of SiO2 using chemical vapor deposition. Measure-
ments showed that the devices had micrometer-scale coupling lengths, (see Fig. 6.10)
typically ∼ 10 µm, allowing them to be used as compact power dividers or combin-
ers for optical communication signals. Because of the wavelength dependence of
106 6 Optical Couplers

the waveguide materials, the measured wavelength-dependent optical outputs from

the parallel and cross ports of the 800-µm-long device oscillated in magnitude with
a 2.5-nm wavelength period. This behavior can be useful in designing wavelength
MUX/DEMUX devices.

6.5.3 Vertical Couplers

The third coupler example is that of a basic vertical coupler using an on-chip Si wire
and a III-V hetero-materials laser source. The purpose of the device is to make an
electrically pumped light source on a silicon platform and to make the device more
manufacturable by only patterning on the Si wafer platform. In this process, both the
waveguide and the passive waveguide length are defined using the same etch step.
In addition, in this device, the laser cavity is defined in an in-plane direction solely
by the length of the silicon waveguide. During fabrication, it thus does not require
otherwise complex alignment in the III-V material. To summarize, this fabrication
approach allows the use of CMOS fabrication tools for the Si steps and the fabrication
of the optical gain step to be separately provided using the III-V materials. Bonding of
the two crystal types is carried out using a low-temperature oxygen plasma step. The
laser is electrically driven in the III-V materials, which in this case is AlGaInAs, and
the field is coupled into Si via the coupler evanescent wave. In the example in Fang
et al. (2006), the laser was CW and had an output power of 1.8 mW and a differential
quantum efficiency of ∼13%. Thus, in this case, vertical coupling allows the buried
Si waveguide device or array of devices to be excited from a surface-excited III-V
slab.

6.6 Adiabatic Couplers

While directional couplers are essential for many integrated-optics applications, the
“standard” design discussed above is sometimes particularly difficult to manufacture.
This difficulty stems from the fact that the coupling coefficient, κ, is exponential in
the waveguide separation or geometry. Thus, κ is sensitive to the effective index of
the coupling layer between the two guides and, hence, to fluctuations in lithography
and material composition; this fact is illustrated in Sect. 6.4 for Li N bO3 guides. As
a result, an alternate type of coupler has been developed, which uses a continuously
controlled change in waveguide parameters to make a device, which is more robust
to variations in device geometry. This device is termed an adiabatic coupler (see
Fig. 6.11).
As in a standard coupler, an adiabatic coupler switches light from waveguide 1 to
waveguide 2 as depicted in Fig. 6.11. While the geometry of the coupler changes in
the z-direction, at any given position, the coupler has a set of local normal or system
modes as shown in the inset of Fig. 6.11. These local normal modes are constituted
6.6 Adiabatic Couplers 107

Fig. 6.11 A sketch of an adiabatic coupler

such that they would be the modes of an infinitely long ideal waveguide structure,
having the same parameters as the guided-wave transition at this point. The coupler
is designed so that light in one system mode remains in that mode as it propagates
along the device; that is, the changes in the waveguide parameters must be sufficiently
mismatched in z that coupling between system modes is prevented. Finally, adiabatic
couplers designed in this way are less dimensionally sensitive than standard couplers,
but adiabatic couplers come at the cost of much greater length. Note that study of
these couplers also provides a gateway to understanding other forms of adiabatic
devices.
An example of an adiabatic coupler is shown in Fig. 6.11, which in this case is
tapered either in refractive index or in geometry. As the system modes propagate
through the device, they change shape due to the local modal coupling. At the end
of the device, the system modes evolve into the isolated-waveguide mode in one of
the two waveguides at the end of the device. For the coupler to work successfully,
no other system modes should be excited; the system modes should thus evolve
adiabatically through the device. To design such a device, we wish to minimize the
overlap between the system modes in any given differential element located along
the device. Thus, if we divide the tapered coupler into differential lengths, we can
define the overlap integral that describes coupling between two system modes,
 ∞
ci j = i0  j1 dx (6.20)
−∞

where x is the coordinate transverse to the local waveguide axis and i0  j1 are
the scalar fields of system-mode i entering (subscript 0) and system-mode j leaving
108 6 Optical Couplers

(subscript 1) the differential element. Thus we wish to design the device so that ci j
is everywhere minimized.

6.6.1 The Differential Coupled-Mode Equations

Because the geometry or effective index in such a device changes with length, it is not
possible to use the standard coupled-mode equations given in the earlier section to
analyze adiabatic devices. Instead, we must obtain coupled-mode equations that are
differential in nature. To obtain these equations, we consider a waveguide structure,
which undergoes a transition in one waveguide property, such as the waveguide width,
index, thickness or, in the case of two waveguides, separation. We then consider the
amplitude of the local normal modes as they propagate along this structure. As
these modes travel from one differential waveguide element to the next, the modes
couple weakly and thus respond to changes in response to the changing waveguide
parameters. If the differential change is slight, the waveguide modes will couple only
to neighboring modes. Using the electromagnetic boundary conditions for the optical
fields at such a transition, it can then be shown that for a differential change in one
geometric or material property (Burns and Milton 1990), designated generally by
δρ,

A j1 − A j0 = ci j Ai0 (6.21)

Ai1 − Ai0 = −ci j A j0 (6.22)

where ci j is the coupling coefficient for the transition δρ, and where A j0 designates
the jth local system mode at a position, 0, just before the position where the change
in δρ occurs. The same mode after the point of differential change would be A j1 .
Since the step is small, only a very small fraction of the original mode(s) is (are)
converted by the transition.
If we define a local propagation constant β j for the local normal mode, j, such
that A j = |A j |ex p[ j (β j z + φ j )], and set
ci j
Ci j = lim (6.23)
δρ→0 δρ

then (6.21) and (6.22) can be written as

dAj dρ
= Ci j Ai + jβ j A j (6.24)
dz dz

d Ai dρ
= −Ci j A j + jβi Ai (6.25)
dz dz
6.6 Adiabatic Couplers 109

These equations are similar to the usual coupled-mode equations except that they
allow coupling due to changes in the device structure along z to be considered. Note
that in the absence of coupling, i.e., Ci j = 0, the modes evolve purely according to
their propagation phase; that is,
  z 
ai, j ≡ Ai, j ex p − j βi, j dz  (6.26)
0

Substitution of this equation into (6.24) and (6.25) then gives

  z 
da j dρ
= Ci j ai exp j (βi − β j )dz  (6.27)
dz dz 0

  z 
dai dρ
= −Ci j a j exp − j (βi − β j )dz  (6.28)
dz dz 0

These equations can be solved analytically for certain coupler geometries. These
include the case where the change in the variable ρ along the waveguide axis, dρ/dz,
is such that
 
dρ βi j
=ζ (6.29)
dz Ci j

where ζ is a constant, and βi j = β1 − β j , the difference in the local modal prop-
agation constants. Note that βi j /Ci, j may vary with z. The general solutions to
these equations when ζ is a constant are given in Burns and Milton (1990). With
these solutions, it can be shown that the maximum power in mode j due to mode
conversion, P jmax = (a j a ∗j ), from mode i with an incident power of ai0 ai0
∗
is given
by

P max 4ζ 2
= 2 (6.30)
Pi0 4ζ + 1

for small coupling, i.e., ζ between the modes.

The mode conversion is thus controlled by the parameter ζ, and can be written
using (6.29), as
Ci j dρ
ζ= (6.31)
βi j dz

and is adiabatic if ζ
1 (small mode conversion), and abrupt if ζ > 1 (large mode
conversion). As a specific example, if the index changes linearly by n over a device
length L, then (6.31) can be written as
Ci j n
ζ≈ (6.32)
βi j L
110 6 Optical Couplers

Fig. 6.12 A sketch of an adiabatic coupler

A further interesting consequence of (6.47) is that, except for mode-interference

effects which oscillate, it can be shown that when the device parameters in the quantity
dρ/dz(Ci j /βi j ), that is ζ, are adjusted to be constant for a device, the resulting
device has the shortest possible device length for a given allowed amount of mode
conversion.

6.6.2 Adiabatic Couplers—An Example

In order to gain insight into the design of an adiabatic coupler, consider the specific
geometry of the coupler shown in Fig. 6.12. In this case, the coupling region has
a linear change in waveguide widths but retains a constant coupling coefficient.
The complete coupler is more complex than the coupling region, since it also must
incorporate input and output y-branches in order to interface with the remainder
of the circuit. These features are important and have been discussed at length in
Ramadan et al. (1998). However, for the sake of simplicity, only the isolated coupler
will be discussed here.
Before discussing the tapered adiabatic coupler, it is important to reformulate
some of the results for the fixed coupler that were obtained earlier in this chapter.
Consider the system modes for this simple fixed coupler, which is weakly coupled,
and shown in Fig. 6.12. The lowest order even and odd total system modes, e,o , of
this coupler with two waveguides having isolated lowest order modes a and b ,
and with waveguide a excited at the input to the device, can be written for the weakly
coupling limit as (see the discussion in Sect. 6.4, for example)
6.6 Adiabatic Couplers 111

e = aa φa + ab φb (6.33)

o = −aa φa + ab φb (6.34)

where aa and ab1 are the amplitudes of the lowest order isolated-waveguide modes
in each of the isolated guides, and φa,b are these waveguide modes containing their
lateral spatial distribution and their usual z-dependent phase factor. The system modes
for this case are now normalized, and thus aa2 + ab2 = 1, and locally orthogonal. In
addition, there is negligible overlap between the modes of the isolated waveguides.
The results in Sects. 6.1 or 6.4 can be used to show that
  1/2
1 χ
aa = 1+ 2 (6.35)
2 (χ + 1)1/2

  1/2
1 χ
ab = 1− 2 (6.36)
2 (χ + 1)1/2
or

aa
= (χ2 + 1)1/2 − χ (6.37)
ab

where we have explicitly chosen to write these expressions in terms of the ratio χ ≡
β/2κ. Here, β is again the mismatch in propagation constants of the lowest order
modes for the isolated guides. Using the prior results in this chapter along with the
quantity χ, it is readily shown that βi − β j = 2κ(χ2 + 1)1/2 . The local value of the
parameter, β/2κ, thus determines the modal transfer from one waveguide to another
in the device. If β/2κ → ∞, the light remains in guide 1; if, however, β/2κ,
the light is evenly divided in the two guides. This quantity, which is termed the
asynchronicity parameter, is thus very important for the design of adiabatic devices.
Returning now to the tapered coupler, the local coupling coefficient ci j be calcu-
lated for a linear variation of the propagation constants with length and a constant
coupling coefficient κi j = κ between the isolated waveguides. The essence of solv-
ing a differential coupled-mode problem is first determining the value of ci j . This is
typically done by using some form of model differential elements, since the system
modes can be coupled (Burns and Milton 1990). However, in the case of a weakly
coupled coupler, the system modes are analytic and the local coupling coefficient
can be written directly using
 ∞
ci j = i0  j1 dx (6.38)
−∞
112 6 Optical Couplers

where the subscripts 0, 1 again denote light, which enters or leaves the differential
element, respectively. Inserting the explicit form for  in (6.33) and (6.34) in (6.38),
we obtain

∞ ∞
ci j ≈ −ab,1 aa,0
 −∞ a0 a1 dx + ab, 0aa,1  ∞ −∞ b0 b1 dx + aa,1 aa,1
∞ (6.39)
−∞ a0 b1 dx − ab,0 ab , 1) −∞ a1 b0 dx

Using the fact that

 ∞
a0 ∗b1 dx ∼
=0 (6.40)
−∞

 ∞
a1 ∗b0 dx ∼
=0 (6.41)
−∞

and that
 ∞
∗
a0 a1 dx ∼
=1 (6.42)
−∞

 ∞
b0 ∗b1 dx ∼
=1 (6.43)
−∞

the following simple expression for ci j results:

ci j = ab,0 aa,1 − ab,1 aa,0 (6.44)

This equation can be further simplified by substituting in (6.35) and (6.36) to yield
f0 − f1
ci j = (6.45)
1+ f2

where

f ≡ −χ + 1 + χ2 (6.46)

In the adiabatic limit, the quantity f 0 − f 1 ≈ −(∂ f /∂ρ)ρ. Thus, the coupling coef-
ficient, Ci j = limρ→0 ci j /ρ, can be written using (6.45) as
1 ∂χ
Ci j = (6.47)
2(1 + χ2 ) ∂ρ

In this case, the changing waveguide variable, ρ, is the waveguide width. ρ = Wa −

Wb , where (Wa ,Wb ) are the coupler waveguide widths for guides a and b. Thus ∂χ/∂ρ
6.6 Adiabatic Couplers 113

is given by
βa = βa0 + α(Wa − Wa0 ) (6.48)

βb = βb0 + α(Wb − Wb0 ) (6.49)

where

βa0 − βb0
α= (6.50)
Wa0 − Wb0

This results in ∂β/∂W = α, where (βa , βb ) are the isolated-waveguide propagation

constants. Note that the superscript 0 in the above equations corresponds to the
value of the parameters at z = 0. Substituting Eqs. (6.49) and (6.50) into (6.48) and
substituting the result of ∂χ/∂ρ into (6.47) yields the following expression for Ci j :

1 χ0
Ci j = (6.51)
2W 0 χ2 + 1

where W 0 = Wa0 − Wb0 and χ0 = (βa0 − βb0 )/2κ. Thus χ0 is the asynchronicity at
the entrance to the coupler, and W 0 = Wa0 − Wb0 is the difference in waveguide
widths at the input of the coupler.
Recall also that the local propagation constant difference of the system modes
βi − β j can be written as

βi − β j = 2κ(1 + χ2 )1/2 (6.52)

and finally the rate of change of waveguide width, ∂W/∂z, for the coupler are given
by simple geometry as
∂W m(W1 − W2 )
= (6.53)
∂z L

where L is the length of the coupler, and where m = 1/2 for a 3 dB coupler, and
m = 1 for a full coupler.
Optimum adiabatic coupler design requires minimum coupling between system
modes in order to minimize power loss to the other unwanted system mode. Accord-
ing to (6.52), this requirement is satisfied for small values of asynchronicity χ0
1
at the input of the coupler, and hence throughout the entire coupler, since by the
nature of the coupler design, χ0 is the maximum value of χ along the coupler. In this
case, the coupling coefficient Ci j is approximately a constant given by

χ0
Ci j  (6.54)
2W
114 6 Optical Couplers

Fig. 6.13 A simulation of the fields in the waveguides of a coupler

and, furthermore, the difference in propagation constants of the system can be approx-
imated by
βi − β j  2κ (6.55)

This case corresponds to that discussed in conjunction with (6.45), and the coupled-
mode equations, (6.43) and (6.44), can be directly integrated to yield the ratio of the
unwanted-to-wanted system-mode power, q,
 2
mχ0
q sin2 (κL) (6.56)
2κL

Excellent agreement is obtained between the values of q calculated from (6.56) and
simulation done using numerical computations as illustrated in Fig. 6.13. For large
values of χ0 , the assumption of uniform propagation constant difference, i.e., (6.55),
is no longer valid and the oscillatory behavior in (6.56) is suppressed. However, as
shown in Fig. 6.13, the simulation results continue to follow the envelope of (6.56).
The above equation provides the basis for a useful design rule for the coupler length,
L, which, considering the envelope of (6.56), is given by

m χ0
L= √ (6.57)
2 q κ

for both types of couplers. For a given tolerance in the system-mode power ratio, the
required 3 dB coupler length is found to be half that of the full coupler. However, an
additional, even more stringent criterion pertains to 3 dB couplers when one considers
the balance of power in the isolated waveguides. The reader is referred to Ramadan
et al. (1998) for a more extensive discussion on this point.
6.7 Summary 115

6.7 Summary

This chapter has introduced the first of a series of passive components, i.e., the
direction coupler; other important passive devices will be presented in subsequent
chapters. These passive components are used in more complex devices such as filters,
delay lines, etc., later in the text. In addition, many of the design concepts presented
in this chapter such as adiabaticity, local normal modes, and the differential coupled-
mode equations will be used frequently in the chapters which follow. For example,
adiabaticity is an important component in the design of several forms of mode steering
devices. The details of adiabatic couplers are discussed in this chapter.
Finally, the discussion of passive couplers in this section will be crucial to under-
standing many forms of coupler-based optical modulators, which will be discussed
later in the text. In these devices, for example, electro-optical switching of “synchro-
nization” of a coupler can be used to make a high-performance waveguide switch.

References

Alferness, R., Schmidt, R., & Turner, E. (1979). Characteristics of ti-diffused lithium niobate optical
directional couplers. Applied Optics, 18(23), 4012–4016.
Burns, W. & Milton, A. (1990). Chap. 3, waveguide transitions and junctions. In T. Tamir (Ed.),
Guided wave optoelectronics
Fang, A. W., Park, H., Cohen, O., Jones, R., Paniccia, M. J., & Bowers, J. E. (2006). Electrically
pumped hybrid algainas-silicon evanescent laser. Optics Express, 14(20), 9203–9210.
Hardy, A., & Streifer, W. (1985). Coupled mode theory of parallel waveguides. Journal of Lightwave
Technology, 3(5), 1135–1146.
Haus, H., Huang, W., Kawakami, S., & Whitaker, N. (1987). Coupled-mode theory of optical
waveguides. Journal of Lightwave Technology, 5(1), 16–23.
Haus, H. A. (1984). Waves and fields in optoelectronics. Upper Saddle River: Prentice-Hall.
Haus, H. A., Huang, W.-P., & Snyder, A. W. (1989). Coupled-mode formulations. Optics Letters,
14(21), 1222–1224.
Kapon, E., Katz, J., Margalit, S., & Yariv, A. (1984a). Controlled fundamental supermode operation
of phase-locked arrays of gain-guided diode lasers. Applied Physics Letters, 45(6), 600–602.
Kapon, E., Katz, J., & Yariv, A. (1984b). Supermode analysis of phase-locked arrays of semicon-
ductor lasers. Optics Letters, 9(4), 125–127.
Louisell, W. (1955). Analysis of the single tapered mode coupler. Bell Labs Technical Journal,
34(4), 853–870.
März, R. (1995). Integrated optics: Design and modeling. Artech House on Demand.
Nishihara, H., Haruna, M., & Suhara, T. (1989). Optical integrated circuits (Vol. 1). New York:
McGraw Hill Professional.
Noda, J., Fukuma, M., & Mikami, O. (1981). Design calculations for directional couplers fabricated
by ti-diffused linbo 3 waveguides. Applied Optics, 20(13), 2284–2290.
Park, H., Fang, A. W., Kodama, S., & Bowers, J. E. (2005). Hybrid silicon evanescent laser fabricated
with a silicon waveguide and iii–v offset quantum wells. Optics Express, 13(23), 9460–9464.
116 6 Optical Couplers

Ramadan, T. A., Scarmozzino, R., & Osgood, R. M. (1998). Adiabatic couplers: Design rules and
optimization. Journal of Lightwave Technology, 16(2), 277.
Somekh, S., Garmire, E., Yariv, A., Garvin, H., & Hunsperger, R. (1973). Channel optical waveguide
directional couplers. Applied Physics Letters, 22(1), 46–47.
Yamada, H., Chu, T., Ishida, S., & Arakawa, Y. (2005). Optical directional coupler based on si-wire
waveguides. IEEE Photonics Technology Letters, 17(3), 585–587.
Chapter 7
Passive Waveguide Components

Abstract In this chapter, we focus our discussion of waveguide circuits by consid-

ering a wide variety of passive optical guided-wave components. These devices are
both important and difficult to design and fabricate and yet they play major roles
in many forms of waveguide circuits. For example, these devices include tapers,
Y-branches, bends, integrated mirrors, and mode conversion devices. Our intent in
this chapter is to develop an understanding of the devices and relevant engineering
formulae for their design.

7.1 Introduction

Most waveguide circuits rely on a variety of passive “components” such as those

shown in Fig. 7.1. These devices enable connecting guided-wave links and the tran-
sitions needed to make a complete circuit. They are surprisingly difficult to analyze,
as are many analog components in a microwave circuit. Generally, the most useful
and general approach for analytic studies is to use a formulation based on local nor-
mal modes. This analysis shows that if the change in geometry or material is abrupt,
new modes are generated. However, if the change is not abrupt, or “adiabatic,” cou-
pling between these modes does not occur; instead, the shape of each mode evolves
smoothly along the propagation path, and radiation from the transition is minimized.
In this chapter, we will discuss a series of important passive devices, such as
tapers, Y-branches, and bends. Our goal will be to develop both an understanding
of the devices and relevant engineering formulae and techniques needed for their
design. The chapter will also emphasize the loss of power and the coupling of modes
in common passive transitions. Designs which control the guided modes at the output
of this transition are crucial in many I/O applications. Spurious mode generation is
particularly severe if the waveguides are only “quasi-” single-mode, i.e., slightly
multimode. The use of such “quasi-” single-mode design is often required in real
devices because of the need to minimize waveguide loss. In such quasi-single-mode
© Springer Nature Switzerland AG 2021 117
R. Osgood jr. and X. Meng, Principles of Photonic Integrated Circuits,
Graduate Texts in Physics,
https://doi.org/10.1007/978-3-030-65193-0_7
118 7 Passive Waveguide Components

Fig. 7.1 Four of the most common passive components. These devices enable the connecting
guided-wave links and transitions needed to make a complete circuit

waveguides, transitions or changes in the waveguide geometry can cause generation

of higher order modes. As these modes move in and out of phase with each other
along the axis of a guided-wave section, the field amplitude in the guide will change
and give different lateral modal profiles.
Several approaches have been presented to analyze adiabatic transitions, includ-
ing mode matching at junctions and coupling of local normal modes. In the initial set
of devices discussed in this chapter, we will use each of these approaches in order to
understand the operation of such components. The results of the numerical compu-
tational methods, e.g., the beam propagation technique, will also be used to provide
“exact” solutions for comparison. Notice also in many places in this chapter, we have
chosen to analyze structures with low-refractive index. This condition is useful for
many material structures and often results in simple accurate approximate solutions
for design and analysis particularly for the important materials systems, e.g., LiNbO3
and SiO2 . A further advantage of examining the low-index structure is that it leads
to clearer insight into the operable guided-wave physics. However, it is important to
7.1 Introduction 119

note that in many devices involving semiconductors, the refractive index contrast is
high, e.g., Si wires, and in those cases it is important to use numerical methods for
simulation.

7.2 Simple Transitions in Waveguide Geometry

In order to introduce the central issues of designing waveguide components, we

will consider first, the simplest case: a transition between two ideal or “canonical”
waveguides of different geometries, such as different width and position. In the first
example, we will consider an abrupt transition between two semi-infinite, single-
mode, mode-mismatched waveguides. In this case, the transition leads only to power
loss. In the second example, we will consider a gradually tapered junction between
two ideal waveguides, with the input guide being a narrow, single-mode structure
and the output guide being a wider, possibly multimode structure. In this case, the
transition should be designed to suppress coupling to other modes in the transition
region. A poor design excites higher order modes or causes loss if the excited mode
is not guided but radiative.

7.2.1 Reflection and Transmission at An Transition Between

Two Waveguides

Consider an abrupt juncture between the two aligned, dissimilar waveguides as shown
in Fig. 7.2. Light incident from the left can excite an assortment of radiative and
guided modes (Marcuse 1991). The total transverse electric and magnetic fields of
these guides must satisfy the boundary conditions for continuity across the inter-
face between the two waveguides. Thus, if we assume here, for simplicity, two slab
waveguides, varying in the x-direction, and having normalized incident modes E i
and Hi , the boundary conditions at the interfacial junction of the two guides cause
the fields at the interface to be such that
 ∞  ∞
E in (x) + r E in (x) + rρ (β)E ρ (β, x)dβ = t E tm (x) + tρ (β)E ρ (β, x)dβ
0 0
  (7.1)
∞ ∞
Hin (x) + r Hin (x) + rρ (β)Hρ (β, x)dβ = t Htm (x) + tρ (β)Hρ (β, x)dβ
0 0
(7.2)
where i, r,, and t denote the incident, reflected, and transmitted normalized guided
modes, and ρ denotes the radiative modes. The quantities r and t denote the complex
reflectivity and transmissivity of the incident fields, respectively. Note that similar
mode matching analysis will also be used in Sects. 7.3.1 and 7.4.
Using the modal orthogonality condition
120 7 Passive Waveguide Components

Fig. 7.2 A plot showing an abrupt juncture between the two aligned, dissimilar waveguides. The
material quantities are those defined earlier in the book

 ∞
1 ωμ0
E n (x)E m (x)dx = δnm (7.3)
2 −∞ βn

where δnm is the Kronecker-delta symbol, it is possible to solve these two equations
for t and r , under the condition that we may neglect the radiative modes for small
waveguide mismatch. The solutions are then (Pollock 1995)
   ∞
βi βt 1
t= dx E in (x, 0) · E tm (x, 0) (7.4)
βi + βt ωμ0 −∞

and
βi − βt
r= (7.5)
βi + βt

Note that these equations reduce to the usual Fresnel equations as the field becomes
increasingly more plane-wave-like. It is important to emphasize that the radiative
fields have been neglected in this particular case.
The first equation shows that, neglecting radiation, the complex transmissivity
is given by the modal overlap of the modes in the two waveguides multiplied by
the complex Fresnel coefficient. Since the reflected light involves mode coupling
between identical modes, i.e., those in the same guides, the modal overlap is unity
and only the Fresnel reflection coefficient is important.
These simple expressions are important for a number of practical applications,
including the frequent practical situation of coupling between a fiber and an integrated
waveguide. A good example of such a practical application is the case of coupling
between an idealized Gaussian fiber mode with a diffused rectangular guide having a
Gaussian distribution in x and y (Burns and Hocker 1977). In that case, the expression
for the transmission reduces to
4
t =T   (7.6)
wx wy
a
+ a
wx a
+ a
wy
7.2 Simple Transitions in Waveguide Geometry 121

where wx and w y are the half-widths of the rectangular waveguide mode and a is
the mode radius of the circular Gaussian mode. This quantity on the right-hand side
reaches a maximum when a = (wx w y )1/2 . T is the usual Fresnel transmissivity (see
above)
N f Ng
T = (7.7)
N f + Ng

where N f,g is the effective index for the fiber and the guide.

7.2.2 Modal Coupling in Waveguide Transitions

In the discussion above, the transition region is a clearly defined plane. If guided
modes are excited in the second waveguide of the transition, they propagate in a
well-defined way because the waveguide (if canonical) is uniform. This situation
is not true in a tapered transition, in which modal coupling can occur throughout
the transition region. In fact, we have earlier seen an example of such distributed or
differential mode coupling in the case of the adiabatic coupler discussed in Chap. 6.
From a more general perspective, in this section, we introduce the central problem
of following the conversion of modes in a transition region and then designing the
transition to reduce undesired mode conversion to an acceptable level. Consider now
the design of a tapered or horn waveguide junction, such as that shown in Fig. 7.3.
Tapers such as this are extremely important for proper matching of waveguide modes
between a fiber and the optical input waveguide for a chip as well for changing the
mode size for two “on-chip” devices.
The calculation of mode conversion in a tapered device is most conveniently done
using the differential coupled-mode formulation, i.e, as shown in (6.43) and (6.44).
The crucial quantity that is needed for this calculation is the coupling coefficient
between the local system modes in the taper, ci j . To obtain this coupling coefficient,
it is first necessary to compute a general ci j for a channel waveguide with a differential
step discontinuity. This is done in the same manner as for the step discontinuity in
the coupler, discussed in Chap. 6, or in the different waveguides in (7.1) and (7.2),
namely, matching of the electromagnetic field at the discontinuity. In fact, the solution
for a taper is outlined in Burns and Milton (1990). In the case of a symmetrically
tapered waveguide of width w, the differential shown in Fig. 7.4 can then be used in

Fig. 7.3 A plot showing a

tapered or horn waveguide
juncture between the two
aligned, dissimilar
waveguides whose width are
W0 and W , respectively
122 7 Passive Waveguide Components

Fig. 7.4 A plot showing a

symmetrically tapered
waveguide of width w and
differential element δw

conjunction with the modal wavefunction to obtain a specific value of ci j , or for the
case of the differential coupled-mode equations, the form of ci j per differential, that
is Ci j . Formally, this reduces to solving equations which are similar to those given
in (7.1) and (7.2). For the case of a wide, well-confined waveguide of width w, i.e.,
well removed from cutoff, the result is

3 δw
ci j = − (7.8)
4 w
or
3
Ci j = − (7.9)
4w
where again, a well-confined mode is assumed.
Earlier in Chap. 6, we found that an analytic solution to the differential coupled-
mode equations may be found if the differential element in the varying waveguide
parameter, i.e., the width of the taper, w, in our case, is given by

dw βi j
=ζ (7.10)
dz Ci j

where ζ is a constant. This equation may be solved to determine the shape needed
for this analytic solution (Burns and Milton 1990).
The difference in propagation constants for the first two propagating modes, β0
and β2 in a planar waveguide may be obtained by simplification of the dispersion
relation for a slab waveguide of thickness d (see Appendix A); such a calculation
shows that β02 ≈ (2π)2 /β0 d 2 , and where in our problem d = w (Burns and Milton
1990). Then, (7.10) may be rewritten as

dw 8π λg
= −ζ (7.11)
dz 3 w

where λg = 2π/β0 , the wavelength of the phase front in the taper or horn for the
lowest order mode. This differential equation may be integrated to yield the parabolic
horn shown in Fig. 7.3,
7.2 Simple Transitions in Waveguide Geometry 123

Fig. 7.5 A calculation of

mode conversion versus
length. Notice that as the
taper length increases, the
power remains essentially
the same in a single mode

1
w = (w02 + 2αλg z) 2 (7.12)

where α = −8πζ/3, and w0 equals the width of the input waveguide. Thus, the
design of the taper involves determining a set of criteria, which “fixes” α for the
particular application of interest. For the case of a transition between a thin, single-
mode waveguide and a wider, somewhat multimode waveguide, and these design
criteria yield, for example, the acceptable upper limit on the mode conversion for a
specific device design.
Recall that in Chap. 6, a general expression for the maximum value of mode con-
version in (7.10) was obtained when solving the differential coupled-mode equations
under the condition of constant ζ. In particular, for small ζ, the maximum converted
power from the input power, to power in the jth mode is

P jmax 4ζ 2 (3α/4π)2
= = (7.13)
Pi0 4ζ + 1
2 (3α/4π)2 + 1

Because of the symmetry of the parabolic taper, the dominant power transfer is
x x
from the E 11 into the E 31 mode; transfer from an odd to even mode would give a
zero overlap integral. The parabolic shape is thus the most efficient since it maintains
constant coupling through the length of the horn. For a low value of mode conversion,
3α/4π  1. Thus once α (or ζ) is set, the specific horn shape is given by (7.12).
As an example, a horn having α = 1 will have a maximum mode conversion from
i → j of only ∼6%.
As in the case of the tapered coupler, it is possible to compute the actual value
of the conversion by referring back to the differential coupled-mode equations. A
calculation of mode conversion versus length is shown in Fig. 7.5. Notice that as
the taper length increases, the power remains essentially the same in a single mode.
124 7 Passive Waveguide Components

Fig. 7.6 A plot of mode

conversion for various
shaped horns

The oscillations seen in the exact solution to the problem result from mode beating
between the i and j system modes; that is, the quantity (Burns and Milton 1990)
 z
u= βi j dz  (7.14)
0

passes through successive values of π.

A plot of the mode conversion ζ for various shaped horns is shown in Fig. 7.6.
The horizontal axis is u, which is defined in (7.14). The parabolic shape has the
lowest mode conversion for the per device length. This result agrees with the earlier
statement in Chap. 6 that mode conversion is minimized when the parameter ζ is
constant.

7.3 Y-Branches

Y-branches are crucial elements in many integrated optical devices, including Mach–
Zehnder interferometers, power dividers, and light routing junctions. The “classic”
shape for this device is shown in Fig. 7.7. There are several different design issues
for Y-branches, and these cause performance trade offs, including coupling between
the exit waveguides in the junction region for small splitting angle, θ; radiative loss
at large θ; and device compactness or overall size, which also requires small θ. The
dependence of the modal behavior of the Y-branch on θ leads to the use of Y-branches
with relatively large θ being used as power dividers, while those with small θ are
used for mode routers. The next section will consider importance of modal overlap
in devices for each of these functions.
7.3 Y-Branches 125

Fig. 7.7 A sketch of a “classic” shape Y-branch

7.3.1 Mode Overlap to Analyze Y-Branches

One of the first concerns in designing a satisfactory Y-branch is the degree of mode
matching between the mode of the input guide and those of the two output waveg-
uides. This matching controls the excitation of the local modes of the transition.
To realize efficient mode matching, we must find the overlap integral between the
incident mode and the lossy radiative and the undesired higher order modes at the
junction. In performing this procedure, it is generally found that the mode overlap is
controlled by the tilt of wavefronts. Thus, large angles reduce this overlap, with the
unmatched field being lost into the substrate.
The procedure for mode matching of Y-branches involves a straightforward match-
ing of the scalar fields at z = 0, using the field spatial distribution. Several different
approximations to the overlap integral have been described in the literature for a
variety of Y-branch structures. In all cases, at small angles, the mode matching in the
Y- branch is good and thus the loss is low while at larger angles, the loss increases.
Consider now the mode matching at the initial junction of a simple “model” Y-
branch shown in Fig. 7.8 (Kuznetsov 1985). This particular Y-branch is selected for
discussion here because it has a simple analytic solution; having an analytic solution
makes it possible to understand clearly how the device impacts the interrelation of
field and geometry. Since the guides have a constant geometry in z away from the
transition, it is only necessary to do the mode matching between ideal guides at the
transition. This procedure allows a quantitative measure of how well a single mode
in the input waveguide drives the output modes in the two exit guides.
To calculate the mode matching in the Y-junction in Fig. 7.8, we can again apply
(7.4) to obtain the overlap integral of this junction. Recall that this overlap inte-
gral gives the complex transmission of the device. Note that it is assumed in this
calculation that only the lowest order TE mode is present in the slab waveguide.
This procedure allows us to determine the modal field in the input guide using the
126 7 Passive Waveguide Components

Fig. 7.8 A sketch of a simple “model” Y-branch with an analytical solution

well-known slab waveguide modes given earlier. Recall that the field distribution is
characterized by the transverse wavevector, kt , and the evanescent decay constant,
γ, which is determined from the following transcendental equation
 
1 kt
kt = tan−1 (7.15)
2a γ

and the relations

γ 2 = β 2 − n 2s k 2 (7.16)

and
β 2 = n 2f k 2 − kt2 (7.17)

and where the quantity 2a related to the waveguide width is defined in Fig. 7.8.
Inserting the expressions for the modal field in (7.4) then gives an analytic, but
still daunting, expression for the complex transmissivity:

e jνa 1 1
t=  e− j2γa sinc(ν − 2γ)a + e j2γa · sinc(ν + 2γ)a
1 2 2
1+
2kt a

e jν0 a
+ sinc(νa) + cos (2γa) 2
(2kt − jν)a
(7.18)
where ν = β sin α, and α is one half of the splitting angle.
A plot of this function is provided in Fig. 7.9. Notice as the branching angle in
the Y increases, the transmission of the Y-branch decreases. This decrease is due to
the increased coupling of the incident light into unguided or radiative modes. The
weak oscillations (solid line) are caused by the interference of the radiative light with
the guided light in the waveguide. Notice that the “simplified” analytic expression
works extremely well for small angles. At larger angles, radiation from the tilted
waveguides becomes increasingly important, and this regime is not handled well
7.3 Y-Branches 127

Fig. 7.9 A plot of the

relation between
transmission and α from
(7.18). As the branch angle
increases, the transmission
decreases

by the simple mode-overlap integral. However, several more exact approaches have
been made toward solving for the radiative loss in wider-angle Y-branches.
One approach, which is known from comparison with accurate computation to be
reliable, is the “volume current method” (VCM), which considers radiation from the
changing waveguide structure (Kuznetsov and Haus 1983). Along the propagation
distance, this method obtains the radiation field in the far field via the polarization
current in the waveguides. The radiation modes then act back on the Y-branch by
beating with the guided modes. The method uses the vector potential for radiation
field  
μ0 e− jk1 r 
Ar (
r) = ∂ν  J (r )e jk1 r̂ ·r (7.19)
4π r
ν

and inserts into this expression the polarization current from the waveguide modes.
The radiative power is then obtained from the form of the Poynting vector containing
the vector potential.
Application of this approach to the Y-branch yields the solid line shown in Fig. 7.9.
The weak oscillations shown in the solid-line plot are caused by interference of the
radiative modes with light in the waveguide. As seen in the figure, the power loss
computed simply from the quantity |t|2 , obtained from (7.18), agrees well with the
VCM method at small branch angles, but diverges and is lower at large angles.
This more rapid drop off in Y-branch transmission in the VCM calculation is due to
increasing coupling to the radiative modes in the spreading waveguides, that is, as
they decouple during the power dividing process. The figure also plots an “exact”
numerical computation, showing that the VCM picture accurately reflects the actual
loss in the Y.
128 7 Passive Waveguide Components

Fig. 7.10 A sketch of an “mode splitter” or “mode router” which routes selectively a certain input
mode into a given output port

7.3.2 Guided-Mode Coupling in Y-Branches: Mode Splitters

In the above section, two factors were considered in the design of Y-branches: mode
matching at the input and loss due to radiation from radiative modes in the diverging
guided-wave structure. In addition, it is also important to consider coupling between
guided modes after the transition region of the Y-branch. In particular, after the
entrance to the Y-branch is excited by the input mode, modes traveling along the exit
waveguides can couple with each other. This coupling can be useful or deleterious
depending on the application of the Y-branch.
An extremely important application of Y-branches, or other 1-to-N splitters, which
are designed to use mode coupling, is to route selectively a certain input mode into a
given output port. These are termed “mode splitters” or “mode routers.” An example
of this device is shown in Fig. 7.10. This device is an asymmetric Y-branch, designed
to separate one system mode from another, adiabatically. This capability, in turn,
allows selective routing of one mode of an input multimode wave to a specific output
waveguide arm. Such devices have been used , for example, in conjunction with
other passive elements to separate one polarization or one wavelength from another.
In these devices, mode steering is an important function, while mode conversion is
not desired.
Mode routers require that the Y-branch to be asymmetric either in real or effective
index or in geometry. To analyze mode coupling, consider, for simplicity, a Y- branch
which has a symmetric separation geometry for the two waveguide arms with dif-
ferent effective indices. Such a device has a changing geometry in z and thus mode
coupling must be analyzed using the differential coupled-mode equations introduced
in Chap. 5 and applied subsequently to treating mode coupling in tapers and adia-
batic couplers. To solve these equations, we must determine the differential coupling
coefficients for the system modes i and j of the waveguide branch as the waveguides
separate (Burns and Milton 1990). An appropriate differential, which when used in
a series of increments would lead to a symmetric Y-branch, is a “step” in the waveg-
uide spacing or separation, δw, shown in Fig. 7.11. The mode coupling coefficient
is then found through a procedure similar to that discussed earlier in this chapter,
and in detail in Chap. 6 for the case of the adiabatic coupler, namely, boundary-value
7.3 Y-Branches 129

Fig. 7.11 A sketch of a differential “step” in the waveguide spacing or separation

matching at the discontinuity. Again, recall that the equations for the system modes
in a canonical or ideal coupler are expressed in terms of the asynchronicity parameter
χ, see the discussion following (6.53). The modal coupling coefficient is then found
to be γχ
Ci j = (7.20)
2(χ2 + 1)
γχ
ci j ≡ Ci j δw = δw (7.21)
2(χ2 + 1)

where γ is defined by the fact that the local coupling between the waveguides sepa-
rated at their edges by w is given by κ = const × exp(−γw).
Once the expression for ci j is found, the differential coupled-mode equations can
then be solved to determine the extent to which modal power transfer occurs in the
Y-branch (Burns and Milton 1990). For example, in a mode splitter (see the following
paragraph), the output to the Y-branch junction is often designed such that when it
is excited with the symmetric, lowest order mode, light will be transferred only to
the guide with the highest effective index of refraction. This index is determined
either by the geometric or/and materials parameters of the waveguides. In this case,
modal separation will occur with low modal cross talk if mode conversion, that is ci j
remains small throughout the transition.
A computation of ci j for a Y-branch shows that there are two operational regimes
for this device: (1) power splitter or divider and (2) mode splitter. Each of these oper-
ational regimes is dominant at different values of angular separation and waveguide
asymmetry, a parameter termed χ. Basically, the difference between these regimes is
130 7 Passive Waveguide Components

Fig. 7.12 A sketch of a 1 × 2 splitter(left); the corresponding coupling coefficient ci j (right). Sig-
nificant coupling only occurs in the range of separation w < δwc . The corresponding δz c can be
determined as shown in the left figure

that in one case, the system modes remain in phase to allow mode conversion during
the splitting process; in the other, the modes are sufficiently asynchronous that no
conversion of the two excited system modes occur, and the power in the input mode
is divided according to a mode matching formulae such as that given in Sect. 7.3.1.
Note, of course, that significant conversion into radiative modes can still occur in
this regime.
Typically, mode conversion occurs dominantly in one spatial region of the device
(see Fig. 7.12). The factors controlling the transfer between the two modes are both
the magnitude of the peak ci j and its spatial overlap. These quantities depend on
the separating waveguide slope and the separation, and their effective index (or
real index) asymmetry as shown in Fig. 7.13. Thus, the localization of the mode
conversion allows us to apply a semiquantitative criterion for use when mode coupling
may be neglected. If δz c is the longitudinal distance, over which conversion occurs,
i.e., where ci j is largest, then when

(βi0 − β j1 )δz c  π/2 (7.22)

that is, the modes remain in phase, mode conversion can occur, and the power in the
input guide of the Y is split between the two output guides. If, on the other hand,

(βi0 − β j1 )δz c
π/2 (7.23)

then the modes will remain out of phase and thus coupling between the modes will
not occur. The modes then evolve naturally to permit mode splitting or separation.
Note that δz c and δwc (see Fig. 7.12) are related by the geometry of the Y-branch.
To obtain a more specific formula for the case of Y-branches, consider a symmetric
Y-branch, which has output waveguides that separate w(z) linearly with z (the region
of width δwc over which ci j is large, is related to δz c by δwc = θδz c ). Then, using
the expression for ci j (7.21) and the values for δβi j in terms of δβ, it can be shown
(see Appendix B) that
7.3 Y-Branches 131

Fig. 7.13 A plot of modes with different waveguide separation and their effective index

δβ
> 0.43 for a mode splitter (no mode conversion) (7.24)
θγ

δβ
< 0.43 for a power splitter (mode conversion) (7.25)
θγ

and where δβ is understood to be = 0. Here, δβ is the asynchronicity of the equivalent

isolated output waveguides, θ is the angular separation, and γ is again obtained from
the local waveguide coupling coefficient κ = const · exp −γw; each of the quantities
is important. In these equations
θγ ∼ δz γ (7.26)

where δz γ is the distance along for separation by approximately one γ.

The relation δβ/θγ > 0.43 is, thus, the criterion for an adiabatic splitter. If we
make use of the simple asymmetric structure such as that shown in Fig. 7.10, and
use the appropriate coupled-mode analysis, a representative plot of mode conversion
versus δβ/θγ can be obtained. This plot is shown in Fig. 7.14. Note the rapid drop
off in converted mode amplitude above a value of δβ/θγ ≈ 0.1; thus the criteria
given in (7.24) and (7.25) hold. These equations can then be used in the design of a
linear variation of θ distance separating asymmetric Y-branches for mode splitting
or power dividing by using them to find, say, θ if γ and δβ are known, etc. Analytic
solutions for shaped Y-branches are given in Burns and Milton (1990).

7.3.3 Advanced Y-Branch Design

The guided-wave physics discussed above suggest several approaches for improving
Y-branches. The first improvement is to adjust the geometry of the Y transition to
132 7 Passive Waveguide Components

Fig. 7.14 A representative

plot of mode conversion
versus δβ/θγ, demonstrating
that when δβ/θγ > 0.43, the
separating waveguide acts as
a mode splitter;<0.43, as a
power splitter

allow the mode from the input waveguide to broaden before coupling into the output
arms. This process is done simply by using the tapered geometry shown in Fig. 7.7.
An even more effective transition can be realized by adiabatically introducing the
dividing region through the use of index tapering in the central region of the Y-branch
(see Fig. 7.15). This tapered region allows a natural evolution of the symmetric system
mode in the transition region and thus suppresses formation of other higher order
system modes, even for relatively large splitting angles. Index tapering also allows
more efficient use of the splitting region by inserting it into the taper.
A second important improvement in design is the reduction of coupling to radiative
modes in the splitting region. This coupling to unguided modes can be problematic on
the basis of simple loss considerations or/and because it leads to unexpected variation
in the optical output with device dimensions. One approach to reducing loss has been
to decrease the guide index in the transition region (see Fig. 7.16). In a physical sense,
in the ray optics picture, this low-index region allows a greater modal angle for light
entering from the input guide, and hence reduces the phase front tilt for light entering
the branching region. A second approach uses coupling to the radiation modes to
occur but then adjusts the device dimensions to increase coupling of these modes
back into the device (Johnson and Leonberger 1983). In effect, the approach uses the
oscillations seen earlier in Fig. 7.9 to best advantage. This approach is identical to
the “coherent coupling” approach which will be described in Sect. 7.4 for waveguide
bends. Simulations by these same authors have shown that the approach can be used
to reduce loss by ∼0.5 − 1 dB in small angle LiNbO3 Y-branches. The same approach
can be used to reduce loss for two closely spaced Y-branches in series.
Finally, one approach, shown in Fig. 7.17, avoids modal splitting, and instead
uses two parallel linear couplers to split off light from a central feeder guide. The
device, thus, eliminates tilted-mode-front mismatch in the apex region of the Y,
and hence reduces radiation loss. The structure is in essence a form of the coupled
parallel waveguide arrays discussed in the previous chapter. From the viewpoint of
fabrication, the device trades off the difficulty of precisely fabricating the apex in
7.3 Y-Branches 133

Fig. 7.15 A sketch of a Y-branch with adiabatically introduced dividing region through the use of
index tapering in the central region

Fig. 7.16 A sketch of a Y-branch with lower index placed in the transition region to reduce the loss
134 7 Passive Waveguide Components

Fig. 7.17 The sketch of a triple guide coupler which avoids modal splitting, and instead uses two
parallel linear couplers to split off light from a central feeder guide

Fig. 7.18 A comparative

plot of the loss versus θ that
can be made for a standard Y
and a “triple guide” Y

a conventional Y against the difficulty in fabricating two precision couplers. The

elimination of mode tilting is attractive, but a careful analysis shows that even if
accurately fabricated, the device is on average not as good as a well-designed Y-
branch of comparable size. If an effective angle α is defined for the device by setting

α = tan−1 (D/L) (7.27)

where D is the separation of the outer guides of the coupler and L is the coupling
length, then a comparative plot of the loss versus α can be made for a standard Y and
a “triple guide” Y, with equivalent output waveguide spacing; see Fig. 7.17. Such a
plot is shown in Fig. 7.18. The large oscillations for the coupler Y are due to the fact
that its abrupt ends lead to severe interaction with radiative modes. The plot shows
that the coupler device is only as good as the standard Y for certain geometries and,
in general, it has a higher loss than the standard Y.
7.3 Y-Branches 135

7.3.4 Power Loss Due to Coherent Beam Combination and

Splitting in Single-Mode Waveguides

Y-branches are often used as phase-sensitive combining elements in optical Mach–

Zehnder interferometers; see, for example, their use in modulators and switches in
Chaps. 12 and 13, respectively. This application shows that in at least some appli-
cations, modal coherence must be considered in the design of Y-branches. Further,
these coherent phenomena, when viewed from the point of view of mode splitting,
lead to some initially unexpected phenomena for optical loss in the branches.
Consider first the two lowest order modes incident on a symmetric Y-branch,
depicted in Fig. 7.18. If the modes arrive in phase, the lowest order mode splits into
two equal in-phase components, while the antisymmetric mode splits into an in-
phase and an out-of-phase mode. Superposition shows that this leads to one excited
waveguide and one unexcited waveguide.
If the waveguide is now reversed so as to be used as a beam combiner, the same
combinations of system modes are also important. Consider first exciting only one
arm of the beam combiner. If the guide is a single-mode guide, feeding one of the
input arms of the Y-branch with the lowest order mode having power P excites the
transition with both the symmetric and antisymmetric waveguide modes each having
one half of the power, or P/2. Since the antisymmetric mode does not propagate,
the power is automatically cut in half in such a combiner. This 3 dB power loss is
thus inherent for a device having single-mode waveguides. A similar inherent loss
occurs when combining with N × 1 combiners.
On the other hand, if both arms of the combiner are excited, the output then
depends on the phase of the two excited modes. Using the arguments presented in
the previous paragraph, feeding each arm with an in-phase lowest order mode will
excite one in-phase system mode in the converging arms of the Y-branch and one out-
of-phase mode. These two out-of-phase modes will, in turn, have a 180◦ difference
in phase from each other and cancel. However, the in-phase system modes will be
added coherently and the output power will be the sum of the two modes, or 2P. If
the two modes are out of phase at the input to the combiner, then the out-of-phase
system mode will be excited, and after it enters the single-mode output guide it will
be lost to radiation.

7.3.5 Example: Asymmetric Y-Branch for Mode Sorting

In efforts to realize high data-rate data transmission systems, various orthogonal (non-
coupling) parameters are used to provide additional channels and, thus, bandwidth;
examples of these parameters include wavelength, spatial modes, and polarization.
Recently Y-branch-like devices, fabricated in Si from SOI have been utilized to sort
and manipulate modal pathways for this application. Thus an asymmetric Y-junction,
in which each arm supports a mode with a different wavevector, has been designed
136 7 Passive Waveguide Components

Fig. 7.19 A sketch of an asymmetric y-branch which can be used for mode sorting

and demonstrated to act as a mode sorter (Riesen and Love 2012; Burns and Milton
1975). In such a mode sorter, the arm supporting the fundamental mode with the
smaller (larger) effective index, n eff , adiabatically excites the first odd (even) mode
of the Y-junction stem, and vice versa as shown in Fig. 7.19. This functionality enables
a pair of asymmetric Y-junctions to be used as a mux/demux in a multimode link.
The criterion for mode sorting, which has been derived from coupled-mode theory,
defines a mode conversion factor (MCF) based on Y-junction properties:

MC F = |β A − β B |/θγ AB (7.28)

where θ is the angle between the Y-junction arms, β A(B) is the wavevector of the fun-
damental mode supported by Arm-A (Arm-B), γ AB = 0.5[(β A + β B )2 − (2kn)2 ]1/2
is related to the evanescent decay constant of the two modes, k is the free-space
wavevector, and n is the cladding index of refraction (Riesen and Love 2012; Burns
and Milton 1975). As described earlier in this chapter, it can be shown that for MCF
>0.43 (MCF <0.43), an asymmetric Y-junction acts as a mode sorter (power divider)
(Riesen and Love 2012; Burns and Milton 1975).
These devices have been recently fabricated and tested. They are shown to work
as designed to selectively address individual modes in a multimode Si waveguide.
The frequency response of the mux/demux pair depends upon the length of the
multimode section as well as the Y-junction angle. The measured cross talk can be as
low as −30 dB, < −9dB over the C band, with insertion loss < 1.5dB (Driscoll et al.
2013). The measured outcoupled power for both Arm-A and Arm-B excitation as a
function of wavelength shows that the power exiting the cross port (Px ) is substantially
lower than that at the through port (Pt ), indicating that the device serves as a mode
mux/demux. The device operates equally well for both modes of the multimode link
since Px and Pt are nearly identical for both inputs.
In addition, the insertion loss for this device was <1.5 dB over the C band. This loss
is higher than the 0.2 dB estimated from 3D finite-difference time domain calculations
mostly due to optical scattering from fabrication imperfections, especially at the sharp
branch point, and propagation loss. Further reduction in loss is possible through an
optimized design, including a reduction in the critical dimension of the branch point
7.3 Y-Branches 137

(Zhang et al. 2013). The cross talk, defined as Px /Pt , was < −9 dB over the C band,
with a minimum of −30 dB near λ = 1580 nm. Finally an important aspect of this
device as well as its more advanced follow on devices is that they are fabricated using
Si processing tools and thus the fabrication quality is excellent.

7.4 Abrupt Angle Bends

An abrupt fixed-angle bend has similar properties to a Y-branch; in fact, it can be

viewed as having similar mode-matching properties as one half of a Y-branch. Again,
this device, shown in Fig. 7.20, may be analyzed by the two methods mentioned in
Sect. 7.3.1, i.e., mode matching at the junction and the VCM approach. As before,
for a judicious choice of geometry, the mode matching treatment gives an analytic
formula for the transmission, t, of the bend:

jνa
t= e
1
 sinc(νa) + 21 sinc(ν + 2κt )a + 21 sinc(ν − 2κt )a+
1+ γa

− jνa  (7.29)
e jνa
cos2 (κt a) + (2γ+
e
jν)a
+ (2γ− jν)a

where κt ≡ kt is the transverse wavevector, γ is the modal evanescent decay constant

outside of the waveguide, and ν = β sin α, where β is the propagation constant in the
waveguides; these definitions are essentially the same as those given earlier for the
case of the Y-branch. Note that in this case, the widths of the input and output arms
of the bend are the same, i.e., 2a. The plot of t versus bend angle, α, in Fig. 7.21,
shows that the transmission decreases with increasing bend angle due to the mode
mismatch of the tilted waves at the junction, just as seen earlier in a Y-branch.
The bend performance can also be analyzed by the volume current method (VCM).
This method can give an accurate estimate of radiation loss in the structure, as

Fig. 7.20 A sketch of an

abrupt fixed-angle bend
138 7 Passive Waveguide Components

Fig. 7.21 A plot of t versus

bend angle, α, showing that
the transmission decreases
with increasing bend angle

discussed earlier for Y-branches. In particular, the use of sharp bends leads to radiation
into unguided modes. This radiation can couple back into the waveguides and lead
to oscillation in a plot of the loss versus waveguide parameters due to interference
effects.
The coupling between guided and unguided modes can be manipulated to greatly
reduce loss in the bends. This approach, called “coherent coupling,” was first
described by Taylor (1974). The approach has been experimentally demonstrated
in a series of abrupt LiNbO3 bends and has been shown to yield significant improve-
ment in the overall loss of a circular curve consisting of a series of small angle
abrupt bends (Johnson and Leonberger 1983). The bend loss is due to constructive
interference between the many radiative modes and one guided mode. Constructive
interference occurs when the length, L between successive bends (see Fig. 7.22) is
such that
(2m + 1)
L= λ m = 0, 1, 2 . . . , (7.30)
2N
where N is the difference between the effective index of the guided mode and
the weighted effective index of the excited unguided modes. The oscillations in
transmission versus L for λ = 0.63 µm measured for the simple test bend structure
shown in Fig. 7.22 are shown in the same figure. More complicated structures have
been made, with loss reduction from 50 to 6dB reported, again for λ = 0.63 µm.

7.5 Circular Bends

7.5.1 Conformal Transformation of Bends into Straight

Waveguides

Circular or continuous bends are a major, recurring design element in PICs. These
bends can be viewed as a series of the bent waveguides discussed in the preceding
7.5 Circular Bends 139

Light output

2°
1°
Light input

Fig. 7.22 Left: A sketch of a bend consisting of a progressive series of small angle abrupt bends.
Right: A plot of transmission versus L for λ = 0.63 µm for the bend showing on the left

Fig. 7.23 A sketch of a

simple circular bend. Note
that the bend causes shifting
of a guided-wave mode
toward the outside wall of
the guide

section. Such a curved waveguide leads to a continuous series of infinitesimal mode

mismatches along the guide, thereby causing loss in the guide. The bends also cause
shifting of a guided-wave mode toward the outside wall of the guide, which enables
coupling to radiative modes, as depicted in Fig. 7.23. As mentioned in Chap. 2, this
radiative loss from a curved waveguide can be serious if the radius of the bend is
small and the index contrast is low. In fact, the dimensions of many PICs are often
controlled by the acceptable level of loss in small-radius bends.
There are a number of approaches to solving for bend loss and for the mode shape,
but use of a conformal transformation gives the clearest results (Heiblum and Harris
1975; Smit et al. 1993). This conformal transformation can be illustrated through
a calculation for the TE mode of a curved 2D waveguide, such as that sketched in
Fig. 7.24.
The transformation starts by considering the Helmholtz equation in cylindrical
coordinates:
[∇r,φ
2
+ k 2 (r, φ)](r, φ) = 0 (7.31)

where  
1 ∂ ∂ 1 ∂2
∇r,φ
2
= r + 2 2 (7.32)
r ∂r ∂r r ∂φ
140 7 Passive Waveguide Components

Fig. 7.24 A sketch of a

conformal transformation for
the TE mode of a curved 2D
waveguide

and (r, φ) denotes the scalar-field of a waveguide mode. Note for simplicity, here
we consider only uniform bends with no structural variation along φ; thus the index
can only be a constant or a function of r , n(r ). In this case, the modal field can be
separated into its angular and radius-dependent terms:

(r, φ) ≡ Uγ (r )eγφ φ (7.33)

where γφ is a (complex) propagation constant along the angular direction specified by

φ. As for all propagation constants, γφ may have a real or an imaginary component:

γφ = αφ + jβφ (7.34)

Thus in this case, the modal phase is constant for specific values of φ, as opposed
to specific values of z in a straight waveguide. In this expression, αφ is the angular
loss coefficient and βφ is the angular propagation constant; both are expressed in units
of (radians)−1 . Thus, the unit loss in dB/rad for a 90-degree bend with an angular
loss coefficient of αφ is −20 log(e−αφ /2 ).
It is possible to transform this bent waveguide into an equivalent straight waveg-
uide by adopting a conformal transformation of the lateral direction of the waveguide;
that is, a transformation, which maps r onto u,
u
r = Rt exp (7.35)
Rt

where Rt is an arbitrary reference radius, which is generally the radius, R, of the

bend. This transformation will result in a new equivalent-index profile and simplified
Helmholtz equation. Inserting this variable transformation for r into (7.30) yields
 
∂2
+ (k n t (u) − γt ) Ut (u) = 0
2 2 2
(7.36)
∂u 2
7.5 Circular Bends 141

Fig. 7.25 A plot of a transformed index profile and the mode profile for a uniform bend of radius
R

where
n t (u) = n(r (u)) · eu/Rt (7.37)

and
γt = γφ /Rt (7.38)

With (7.35), the wave equation for a straight waveguide is recovered with a new
effective index n(u), in the transformed Cartesian coordinates (u, ϕ). The propa-
gation coordinate is thus now ϕ, and the transformed lateral direction is u. This
equation can be solved to obtain both the transformed propagation constant, γt (or
transformed effective index), and transformed modal field, Ut . From the definition
of γφ and from (7.37), it follows that βφ = βt Rt and αφ = αt Rt .
There are several advantages of using the above conformal transformation. The
first is that after transformation, analytic or numerical calculations for linear guides
can be readily performed on the transformed structure. In addition, the waveguide
physics are more easily and intuitively understood using the transformed structure.
Figure 7.25 shows a transformed index profile and the mode profile for a uniform
bend of radius R.
For large bends, such that
u
1 (7.39)
R
the transformed index may be simplified. Then the exponential in (7.36) can be
approximated with a Taylor’s series to yield
142 7 Passive Waveguide Components

Fig. 7.26 A plot of a beam propagation numerical simulation for the equivalent straight waveguide,
showing this outward shifting of the modal profile as z increases

 u
n t (u) = n(r ) 1 + (7.40)
R
and where the approximation,

u = Rt ln(r/R) ≈ −R + r (7.41)

has also been used. This transformation index profile then becomes the original
profile, n(r ), multiplied by a linearly increasing index profile in going across the
waveguide (see Fig. 7.25).
In order to find the modal shape, it is first necessary to find the first transformed
modal shape, Ut (u), by solving for the equivalent straight waveguide. Then, using
the explicit r -dependent form of u in the expression for Ut , i.e.,

U = Ut (u(r )) (7.42)

the actual mode shape can be obtained. For example, when u/R  1, application of
(7.40) shows that the modal profile is simply translated linearly toward the outside of
the waveguide. A beam propagation numerical calculation for the equivalent straight,
given in Fig. 7.26, shows this outward shifting of the modal profile as z increases.
Referring again to Fig. 7.23, note that for a sufficiently tight radius, the field at the
inner edge of the waveguide becomes negligible, and this side of the waveguide does
not contribute to the modal lateral confinement. The guided mode is then classified as
7.5 Circular Bends 143

a whispering gallery mode. Waveguides have the lowest loss when they are operated
so as to have whispering gallery modes, since optical scattering from one of the
waveguide walls is eliminated. Thus, a guide of width w and radius R operates in a
whispering galley mode when
δn
w> R (7.43)
n + δn

where δn is the index contrast in the untransformed waveguide.

7.5.2 Improved Design of Curved-to-Straight Waveguide

Junctions

The tilting of the transformed index in curved bends can cause difficulties in the
design of waveguide devices. In particular, the mode shifts away from the center
of the waveguide toward the outer wall when a mode from a straight guide enters
a bent guide; this point can be seen in Fig. 7.26. Hence, the mode overlap between
the two single waveguides is not unity and, as a result, the transition will lead to
loss at the transition. In addition, if, as is common in many integrated systems,
quasi-single-mode waveguides are used in the optical circuit (see Chap. 3), then a
curved-to-straight waveguide transition can trigger excitation of the higher modes of
the waveguide. This excitation results from the fact that there is overlap between the
single-mode input and the higher modes of the guide due to the lateral shift in the
bend. One solution to this problem is to offset the center of the two waveguides at
the beginning of the transition, thus maximizing the overlap for the two regions. This
approach is shown in Fig. 7.27. Another approach to countering the mode shifting
problem is to “push” the mode back with a region of higher effective index, e.g., a
heavier Ti:doped region (see Fig. 7.26) or a more deeply etched region. This approach
is called a “Crown” waveguide.

Fig. 7.27 A sketch of an

improved design of
curved-to-straight waveguide
junction which offsets the
center of the two waveguides
at the beginning of the
transition to maximizing the
overlap for the two regions
144 7 Passive Waveguide Components

7.5.3 An Explicit Equation for Bend Loss

In many instances, the design of waveguide bends is done with simple loss calcu-
lations using approximate formulae. This approach is useful for low-index contrast
waveguides. For example, an explicit solution to the Helmholtz equation for a slab
waveguide containing a bend of radius R has been solved by Marcuse (1971). This
solution is obtained for a slab waveguide of width w, surrounded by an index n s .
The formula gives the linear loss coefficient (i.e., loss/length) of the modal field in
the circular bend as
 
2γ 2 kt2 exp(γw) −2γ 3
α exp R (7.44)
(Ne2f f − n 2s )k 2 β(2 + γw) 3β 2

for αR  R and (Neff − n s )  n s , i.e., low-index contrast. A normalized loss curve,

given, for example, in Marcatili (1969), can be constructed for various curve radii R
versus the waveguide normalized frequency. Note that low-loss bends require large
R and the highest possible n.

7.6 90◦ Waveguide Crossovers

Waveguide crossovers are critical for reducing overall device “footprints,” since they
are often required for folded device configurations. For example, such crossovers
have been successfully used in GaAs spiral waveguides to allow the exiting radial
guide to cross through the bounding azimuthal guides. Similar 90◦ waveguide
crossovers have been used to fold the structure of a long short-pulse optical delay
line in GaAs (Hu et al. 1998); in this case, folding provides efficient use of the chip
area by shortening the length of the device. More recently, it has been found that in
the case of Si photonics switching arrays that crossover loss is the major factor in
scaling up the number of switching nodes (Rumley et al. 2015; Stern et al. 2015).
There are generally two important performance factors for a crossover: the cross
talk between intersecting waveguides and the optical attenuation in each intersecting
waveguide.
Cross talk between two intersecting waveguides at a crossover has been inves-
tigated experimentally and theoretically. These investigations have shown that the
cross talk is significant only for a small crossing angle and decreases very quickly
at larger angles. For example, one set of measurements of intersecting Ti: LiNbO3
waveguides showed cross talk < − 35 dB for a crossing angle >5◦ (Bogert 1987).
This result is in agreement with an analysis by Agrawal et al. (1987), which showed
the maximum cross talk between two equal width, intersecting waveguides is negli-
gible for large crossing angles.
In fact, for 90◦ crossovers, the loss in such a crossover can be treated approximately
by simple diffraction theory. The mode in one of the crossing waveguides far from
the intersection has a field profile determined by its isolated cross-sectional structure.
7.6 90◦ Waveguide Crossovers 145

As the guided mode reaches the intersection, it enters a short length of an infinitely
wide slab, and diffraction occurs. This near-field diffracted light is collected by the
intersecting waveguide, causing loss and cross talk. Since the length of the slab region
is short, i.e., usually equal to the waveguide width, the diffractive loss is small. For
example, the diffractive losses in the GaAs spiral waveguides mentioned above were
as low as 0.012 dB for a crossover formed by 2 µm waveguides (Hayes and Yap 1993).
Optical simulations have been performed on 90◦ step-index waveguide crossovers to
estimate their loss. The simulation shows, for example, that for a ∼5 µm waveguide,
the diffractive loss was < 0.1 dB per crossover (Hu et al. 1998).
One potential practical problem with crossovers is patterning imperfections, e.g.,
the rounding at the corners, where two waveguides intersect. These imperfections
can occur either during lithographic patterning or pattern transfer, e.g., etching. For
example, rounded corners effectively make the diffraction section at a crossover
longer and wider, causing additional diffraction of the light and thus an increased
loss at the crossover. However, simulations have shown that for a short diffraction
length, the total scattering loss remains small; e.g., with length equal to twice the
waveguide width, the scattering is ∼0.1 dB (Hu et al. 1998).
More recently the growing interest in complex Si photonics systems has put high
demand on finding extremely low loss solutions for waveguide crossings (Rumley
et al. 2015). And in fact, the continued improvements in Si patterning techniques,
e.g., lithography and etching as well as in simulation tools have greatly reduced cross-
ing loss. For example, Ma et al. (2013) has demonstrated experimentally compact,
spectrally broadband Si crossings for use at 1550 and 1310 nm. Specifically CMOS-
compatible processes based on 248 nm optical lithography and single etch step were
used to fabricated the structures. Their characterization measurements used reliable
multiple-die measurement to obtain transmission insertion losses of 0.028±0.009
dB at 1550 nm and 0.017±0.005 dB at 1310 nm, with cross talk lower than −37 dB.

7.7 Turning Mirrors

Another approach to folding an optical path on a PIC chip is the use of minia-
ture flat turning mirrors. These mirrors are mounted in the end of waveguides and
tilted at angles beyond the critical angle; as a result, the ideal mirror reflectivity for
plane waves should be 100%. Turning mirrors are particularly important for use with
the weakly confining waveguides that are needed to mode match with input/output
optical-fiber connectors. For such weakly confining waveguides, it would otherwise
be necessary to use a relatively large radius of curvature bends to steer modes in the
optical current.
Turning mirrors have been investigated using a variety of numerical techniques
(Chung and Dagli 1991, 1995). These studies show that losses have three origins:
mirror displacement, angular misalignment, and surface roughness. Not surprisingly,
loss is also a function of the optical polarization, since the polarization determines
in part the optical phase at the air-semiconductor interface.
146 7 Passive Waveguide Components

Fig. 7.28 A plot of the

computed transmission
versus rotation angle of the
turning mirror for the
TM-like waveguide mode for
different waveguide width W

In addition, the effect of displacement error has also been computed for a rib
waveguide with a TE mode (corresponding to the TM mode of the slab used in the
effective index model). The results show that output intensity peaks at a displacement,
which is 0.1 µm before the ideal mirror position. This displacement is due to the fact
that for small beams, diffractive effects cause the beam phase front to be distorted
from that of an ideal plane wave. This displacement is smaller for the TM mode than
for the TE mode: 0.1 versus 0.02 µm, respectively. Note that for the TE mode, the
mirrors must be positioned accurately to avoid a significant loss penalty, particularly
for a small diameter beam. Thus a displacement error of 0.2 µm can lead to 5%
reflection loss for the TE-like mode at 1.55 µm.
Figure 7.28 shows the computed transmission for an error in the rotation angle
of the turning mirror for the TM-like waveguide mode. Although not shown in the
figure, the angular sensitivity of the TM and TE modes is approximately equal. Not
surprisingly, the sensitivity to mirror rotation is greatest for the wider waveguides,
due to the more plane-wave-like nature of the guided wave in a wide waveguide. The
investigation that led to Fig. 7.28 suggested that, in general, the error due to angular
misalignment was a less severe constraint than that due to mirror displacement.
Loss due to mirror roughness has been examined (Chung and Dagli 1995) by
assuming either random or sinusoidal mirror roughness in the mirror surface, e.g.,
d(z) = d0 + dm · random(z), where random(z) is obtained from a random number
generator. Regarding random roughness, the results of a calculation which examined
transmission versus dm (the random roughness amplitude) showed that the loss was
much more pronounced for the TE-like mode compared to the TM-like mode. This
result follows from the larger reflectivity across all angles for the TE versus the reflec-
tivity of the TM mode. In addition, it was found that the loss was approximately inde-
pendent of waveguide width. In contrast, when the roughness √ was examined under the
assumptions of sinusoidal roughness, i.e., d(z) = d0 + 2drms (sin 2πz/ ), where
 is the spatial period, the conclusions were different. In this case, it was found that
the loss increase was sensitive to waveguide width, although it remained insensitive
to mode polarization. Specifically, it is found that turning mirror loss is most pro-
nounced for roughness, having a period between 0.1 − 2.0 of the waveguide width.
7.7 Turning Mirrors 147

Above this period, the loss drops off exponentially with spatial period. These results
are in accord with the phase distortion of the reflected wave. In addition, loss increases
nonlinearly with the amplitude of the spatial period.
The above loss phenomena are important in the fabrication of turning mirrors.
Typically, these mirrors are etched by use of some form of ion etching in conjunction
with a robust surface mask. The loss mechanism, which is of highest concern, in
etched surfaces is that of the surface roughness. Generally, it is possible to character-
ize the etching by decomposing the surface roughness into its dominant sinusoidal
component via Fourier analysis; in one case, for example, reactive ion etching (RIE)
etching was shown to yield a Fourier component with ∼5 − 30 µm period. In prac-
tice, mirrors with ∼0.8 dB and 1 dB loss have been fabricated for TE- and TM-like
polarized modes. In general, mirrors are etched more deeply than the feeder waveg-
uides to enable high reflection over the entire vertical extent of the structure.

7.8 Transfer Matrices

In Chap. 6, we introduced the concept of transfer matrices for the specific case of cou-
plers. Such transfer matrices make design of many standard elements of an integrated
optical system relatively easy. The final section in this chapter provides the transfer
matrices of the most common passive elements. These elements, which relate output
to the input fields, are listed in Table 7.1. The use of these elements requires some
explanation. Specifically, in the case of a single port input or output, it is necessary
to fix the nonexistent port at a value of 0. Thus for a Y-branch 3 dB splitter with unity
input, the splitter output will be written as
    
a1o 1 10 1
=√ (7.45)
a2o 2 10 0

or  
1 1
=√
2 1

An excellent example of the use of component transfer matrices to obtain the

matrix for a more complex device is provided by the example of a simple Mach–
Zehnder. For such a device, we multiply the matrices starting with the input of the
device. For a Mach–Zehnder, shown in Fig. 7.29, the transfer matrix is
    
1 11 exp(iφ/2) 0 1 10
√ √ (7.46)
2 00 1 exp(−iφ/2) 2 10

or cos(φ/2). The text by März (1995) provides a more extensive discussion of

these matrices
148 7 Passive Waveguide Components

Fig. 7.29 A sketch of a Mach–Zehnder interferometer

Table 7.1 Transfer matrices for some of the multiport devices in this chapter
Component Matrices Description
√
Lossy waveguide  power loss, 

exp(iφ/2) 0
Phase shifter phase shift, φ, in a waveguide
1 exp(−iφ/2)

0 i
Crossing exchanges input and output
i 0 ports
√ 
x 0
Splitter √1
x+1
1 : x power splitter
1 0
√ 
x 1
Lossless combiner √1
x+1
x : 1 combiner
0 0
√ 
x 1
Lossy combiner √1 √
x+1 1 − x

7.9 Conclusion

This chapter has examined the analysis and design of many of the basic passive
components that are present in PICs. Their performance in terms of loss, mode
conversion, and cross talk is at the very heart of designing high-performance photonic
circuits. Thus, careful analysis of these components is an essential step in building the
more complex PICs described in succeeding chapters. Use of the transfer matrices
included at the end of this chapter simplify analysis of more complicated devices
by straightforward multiplication of a chain of matrices in the simple components
comprising the longer device. However, an accurate prediction of performance must
consider the nonideal performance of each component in the circuit; this performance
can only be obtained with an accurate numerical simulation package.
Appendix A 149

Appendix A

To calculate βi j between the ith and jth TM modes for a slab waveguide of width,
d, first write the usual waveguide expression for the propagation coefficient of mode
m,
k 2ym + βm2 = k 2 n 2f (7.47)

But if mode m is far from cutoff,

k ym ∼
= (m + 1)π/dem (7.48)

where dem is the effective waveguide width for mode m.

Then, using a Taylor series expansion,

(m + 1)2 π 2
βm ∼
= kn f − 2
(7.49)
2kn f dem

Notice that dem is also mode dependent. This can be simplified by noting that the
effective width of a waveguide is just approximately the actual width d for high index
contrast or to the next level of approximation

dem ∼
= deo (7.50)

 2 1/2
∼ 2 ns 1
=d+ (7.51)
k nf n 2f − n 2s

for a TM mode and where deo is the effective width of the lowest order mode.
Then,
π 2 (m 2 + 2m)
β0 − βm ≡ β0m = 2 (7.52)
d (β0 + βm )

so that β02 ≈ (4π 2 )/(d2 β0 ) = (2πλg )/d2 , where λg ≡ 2π/β0 and β0 is the propa-
gation constant of the lowest-order mode. Note also that away from cutoff, β0 + βm ≈
β0 . This derivation will also recur in Chap. 9, Imaging Devices. For a channel waveg-
uide, β02 ≈ β13 (Burns and Milton 1990).

Problems

1. (a) Calculate the lowest-order mode profile for a simple slab waveguide with
the following parameters:
n f = 3.4
n s = n c = 3.38
w = 4 µm
λ = 1.5 µm
150 7 Passive Waveguide Components

(b) Calculate the displacement of the peak position if the waveguides has a 1
cm radius bend in it. Use a simple tilted-axis approach to calculate.
(c) Estimate the loss in this waveguide using (7.44).
2. Design a parabolic horn waveguide such that the power loss is 20%. Start with
the waveguide in Problem. 1 and expand the lowest order mode by a factor of 1.5.
(Note: you need to describe the shape in width and calculate the length L of the
horn waveguide.)
3. Prove (7.43) with sketches, normalized quantities and appropriate assumptions.

δn
w> R
n s + δn

where δn is the index contrast, n s is the substract refractive index and R and w
are radius and width of the waveguide.
Hint: you could use transformation method or ray picture to prove this.
4. Calculate the transmission loss for the lowest mode of a GaAs waveguide butted up
against a 6-µm-single-mode-fiber. The GaAs waveguide is square with a 2 µm ×
2 µm Gaussian shape mode. Use the following parameters:
λ = 1.5 µm; n GaAs = 3.4; n SiO2 = 1.5
5. An excited waveguide feeds a second slab waveguide with a 1-meter radius of
curvature, but otherwise is the same as the straight waveguide, which feeds it.

(a) Qualitatively, sketch the mode amplitude versus transverse coordinate for
BOTH waveguides on the SAME graph; label all features on the curves.
Explicitly state which modal phenomena happen near the trasition region
between the two guides.
(b) Sketch the geometry of a common waveguide design change to the above
layout that is used to make higher-performance straight-bend connections.

6. Explain using words how you would design a low-loss waveguide mode expander,
that is to expand from width w to 5w. You should supplement your words with
simple equations.

References

Agrawal, N., McCaughan, L., & Seshadri, S. (1987). A multiple scattering interaction analysis of
intersecting waveguides. Journal of Applied Physics, 62(6), 2187–2193.
Benson, T. (1984). Etched-wall bent-guide structure for integrated optics in the iii–v semiconductors.
Journal of Lightwave Technology, 2(1), 31–34.
Bogaerts, W., Dumon, P., Van Thourhout, D., & Baets, R. (2007). Low-loss, low-cross-talk crossings
for silicon-on-insulator nanophotonic waveguides. Optics Letters, 32(19), 2801–2803.
Bogert, G. (1987). Ti: Linbo3 intersecting waveguides. Electronics Letters, 23(2), 72–73.
Burns, W., & Milton, A. (1975). Mode conversion in planar-dielectric separating waveguides. IEEE
Journal of Quantum Electronics, 11(1), 32–39.
References 151

Burns, W., & Milton, A. (1990). Chap. 3,“waveguide transitions and junctions”. In T. Tamir (Ed.),
Guided wave optoelectronics.
Burns, W. K., & Hocker, G. (1977). End fire coupling between optical fibers and diffused channel
waveguides. Applied Optics, 16(8), 2048–2050.
Chen, H., & Poon, A. W. (2006). Low-loss multimode-interference-based crossings for silicon wire
waveguides. IEEE Photonics Technology Letters, 18(21), 2260–2262.
Chu, F., & Liu, P.-L. (1991). Low-loss coherent-coupling y branches. Optics Letters, 16(5), 309–
311.
Chung, Y., & Dagli, N. (1991). Analysis of integrated optical corner reflectors using a finite-
difference beam propagation method. IEEE Photonics Technology Letters, 3(2), 150–152.
Chung, Y., & Dagli, N. (1995). Experimental and theoretical study of turning mirrors and beam
splitters with optimized waveguide structures. Optical and Quantum Electronics, 27(5), 395–403.
Driscoll, J. B., Grote, R. R., Souhan, B., Dadap, J. I., Lu, M., & Osgood, R. M. (2013). Asymmetric
y junctions in silicon waveguides for on-chip mode-division multiplexing. Optics Letters, 38(11),
1854–1856.
Hanaizumi, O., Miyagi, M., & Kawakami, S. (1985). Wide y-junctions with low losses in three-
dimensional dielectric optical waveguides. IEEE Journal of Quantum Electronics, 21(2), 168–
173.
Hayes, R. R., & Yap, D. (1993). Gaas spiral optical waveguides for delay-line applications. Journal
of Lightwave Technology, 11(3), 523–528.
Heiblum, M., & Harris, J. (1975). Analysis of curved optical waveguides by conformal transforma-
tion. IEEE Journal of Quantum Electronics, 11(2), 75–83.
Hu, M., Huang, J., Scarmozzino, R., Levy, M., & Osgood, R. (1997). A low-loss and compact
waveguide y-branch using refractive-index tapering. IEEE Photonics Technology Letters, 9(2),
203–205.
Hu, M., Huang, Z. J., Hall, K., Scarmozzino, R., & Osgood, R. M, Jr. (1998). An integrated two-stage
cascaded mach-zehnder device in gaas. Journal of Lightwave Technology, 16(8), 1447.
Izutsu, M., Nakai, Y., & Sueta, T. (1982). Operation mechanism of the single-mode optical-
waveguide y junction. Optics Letters, 7(3), 136–138.
Johnson, L., & Leonberger, F. (1983). Low-loss linbo 3 waveguide bends with coherent coupling.
Optics Letters, 8(2), 111–113.
Kuznetsov, M. (1985). Radiation loss in dielectric waveguide y-branch structures. Journal of Light-
wave Technology, 3(3), 674–677.
Kuznetsov, M., & Haus, H. (1983). Radiation loss in dielectric waveguide structures by the volume
current method. IEEE Journal of Quantum Electronics, 19(10), 1505–1514.
Ma, Y., Zhang, Y., Yang, S., Novack, A., Ding, R., Lim, A. E.-J., et al. (2013). Ultralow loss single
layer submicron silicon waveguide crossing for soi optical interconnect. Optics Express, 21(24),
29374–29382.
Marcatili, E. (1969). Bends in optical dielectric guides. Bell Labs Technical Journal, 48(7), 2103–
2132.
Marcuse, D. (1970). Radiation losses of tapered dielectric slab waveguides. Bell Labs Technical
Journal, 49(2), 273–290.
Marcuse, D. (1971). Bending losses of the asymmetric slab waveguide. Bell Labs Technical Journal,
50(8), 2551–2563.
Marcuse, D. (1991). Theory of dielectric optical waveguides (2nd ed.). San Diego: Academic.
März, R. (1995). Integrated optics: Design and modeling. Artech House on Demand.
Pollock, C. R. (1995). Fundamentals of optoelectronics.
Riesen, N., & Love, J. D. (2012). Design of mode-sorting asymmetric y-junctions. Applied Optics,
51(15), 2778–2783.
Rumley, S., Nikolova, D., Hendry, R., Li, Q., Calhoun, D., & Bergman, K. (2015). Silicon photonics
for exascale systems. Journal of Lightwave Technology, 33(3), 547–562.
Smit, M. K., Pennings, E. C., & Blok, H. (1993). A normalized approach to the design of low-loss
optical waveguide bends. Journal of Lightwave Technology, 11(11), 1737–1742.
152 7 Passive Waveguide Components

Stern, B., Zhu, X., Chen, C. P., Tzuang, L. D., Cardenas, J., Bergman, K., et al. (2015). On-chip
mode-division multiplexing switch. Optica, 2(6), 530–535.
Taylor, H. F. (1974). Power loss at directional change in dielectric waveguides. Applied Optics,
13(3), 642–647.
Zhang, Y., Yang, S., Lim, A. E.-J., Lo, G.-Q., Galland, C., Baehr-Jones, T., et al. (2013). A compact
and low loss y-junction for submicron silicon waveguide. Optics Express, 21(1), 1310–1316.
Chapter 8
Components for Polarization Control

Abstract Controlling the state of polarization for a propagating mode on an inte-

grated optical chip is a particularly challenging aspect of integrated optical engineer-
ing. A significant aspect of that challenge is the need for fabrication with several
different forms of materials. In fact in many cases the use of a hybrid approach is
required. In this chapter, a wide variety of approaches from simple metal over lay-
ers to complex layered structure, and carefully fabricated optical nanostructures are
examined for their use in polarization control.

8.1 Introduction

Controlling the state of polarization for guided light in a PIC is one of the most chal-
lenging aspects of integrated optical design. In some cases, for example, polarization
control is essential for the operation of certain devices within the optical “circuit.”
An excellent example is the need for a well-defined, specific polarization states in
order to obtain optical isolation using a nonreciprocal optical element. In other cases,
the need for polarization control is purely to ensure a predictable or balanced per-
formance in the optical circuits. For example, in many cases, fibers employed for
communication or sensing applications do not maintain a constant polarization. Thus
the input of an optical chip used to receive the output from the fiber may have an
uncertain polarization state. This lack of polarization control can then pose a seri-
ous problem for signal detection and amplification since many photodetector and
optical amplifiers based on semiconductors are polarization sensitive, i.e., they have
different responses for different polarization of the light. In such cases, a polariza-
tion diversity operation/network may be an effective solution. For this application,
devices for polarization manipulation are required for functions such as polarization
of the input optical signals, conversion and rotation of the polarization states, and
splitting/combination of optical signals with different degree of polarization.

© Springer Nature Switzerland AG 2021 153

R. Osgood jr. and X. Meng, Principles of Photonic Integrated Circuits,
Graduate Texts in Physics,
https://doi.org/10.1007/978-3-030-65193-0_8
154 8 Components for Polarization Control

In this chapter, we will discuss several different components for polarization

control. These components include polarizing elements, polarization rotators, and
polarization splitters. Unlike most of the components discussed up to this point,
devices for polarization include not only integrated devices, but also may require the
employment of hybrid integration techniques such as those used for vertical polariz-
ing plates. In practice, such hybrid integration is difficult and expensive, as the cost
advantages of full integration such as planar processing are lost. However, hybrid
technology often offers specialized advantages such as the more rapid implementa-
tion of certain new or advanced materials systems or easing the addressing of optical
requirement for an off-chip application.

8.2 Polarizing Elements

The most fundamental polarizing element is a simple polarizer, which selects from
the two possible orthogonal polarization states. A number of different realizations
of such a polarizing element have been reported in the literature; however, the main
approaches to achieve the polarization control can be divided into three classes:
plates of metallic thin films or other absorbing media inserted in a slot normal to
a waveguide axis; birefringent or absorbing thin films deposited on a waveguide
surface; and polarization routers which guide each of the two states into different
waveguides. These three classes will be described in the following subsections. As
will be shown below, the best polarizing devices have extinction ratios of ∼30 dB,
with insertion losses of ∼1 dB.

8.2.1 Thin Polarization Plates

An important commonly used approach to polarizing a guided mode is the insertion

of a simple polarizing plate in a direction normal to the guided-wave axis. These
thin-film polarizers have been used to change the polarization state and to control the
polarization in isolators and circulators. As will be mentioned below, this approach
using insertion of a plate is best considered to be a micro-optics or hybrid opti-
cal approach, although the dimensions are close to those seen in many integrated
optics devices. This approach is particularly attractive for waveguides with low-
index materials such as silica and large beam spot sizes ∼8 µm. These devices are
more problematical for materials with high indices (such as semiconductors) and/or
devices with small beam sizes because reflection and/or diffraction loss are typically
unacceptable.
There are several optical considerations, which must be borne in mind, in selecting
polarization plates. Typically, these plates are inserted into a circuit by cutting a slot
into a waveguide structure, generally using a thin diamond saw. The slot and the
plate introduce four facets in the waveguide which back-reflect, causing a loss, R:
8.2 Polarizing Elements 155
 2  2
nf −1 np − 1
R=
nf +1 np + 1

where n f is the usual index of refraction of the waveguide and n p is the index of the
plate. If index matching fluid can be used, the losses are substantially reduced.
In addition, because typically there is no provision for index guiding in the plate,
diffraction occurs after light leaves the input facet and propagates freely in the plate.
This diffracting beam expands and can then lead to coupling loss at the facet of the
output waveguide. If the waveguide is, say, rectangular in cross section with dimen-
sions d and w, the coupling efficiency over a slot of length L will be approximately

d 2 w2
η≈ (8.1)
1.4λ2 L 2
This loss can be substantial even for relatively modest values of L. For example, a
GaAs ridge waveguide with d = 6 µm and w = 5 µm will produce a 10.7% loss for
L = 50 µm (Sieger and Mizaikoff 2016).
The most obvious choice of material for a polarizer plate would be Polaroid-like
polymers, see Fig. 8.1, which have in fact been used for imaging devices as will be
discussed in Chap. 9. Unfortunately, these plates are generally relatively thick, and
thus lead to high diffractive losses. One example of a thinner polarization plate is
that made of lamipol. Lamipol is a film polarizer which uses an alternating series
of thin metal and SiO2 layers and is sliced into a plate in a plane perpendicular to
the plane of the SiO2 /Al layers (Sato et al. 1993). It thus resembles a miniature wire
polarizer such as that used for commercial infrared polarizers. Lamipol slices with
thicknesses between ∼10 and ∼30 µm have been fabricated (see Fig. 8.1), and an
extinction ratio of >40 dB for a 20 µm plate has been reported (Okuno et al. 1994).

Fig. 8.1 A sketch of a slice

of lamipol showing the
materials in the lamipol
156 8 Components for Polarization Control

PI waveplate UV-curable adhesive

SiO2 over-cladding SiO2-GeO2 core

SiO2 under-cladding

Si substrate

Fig. 8.2 A cross-sectional view of a TE/TM polarization mode converter with polymide half-
waveplate

8.2.2 Thin Absorbing Film on a Buried Waveguide

The simplest approach to making a waveguide polarizer is to deposit a metal film on

the “cover” surface of a waveguide, as shown in Fig. 8.2. Polarization of the guided
wave is then achieved through the strong differences in the absorption loss in the
metal for TE and TM components. This differential ratio of absorption in the metal
film follows from the fact that metal films respond with both real and imaginary parts
of the dielectric constant at optical frequencies:

metal = r + ji (8.2)

where both components are negative since light penetrates into the metal only as a
lossy, evanescent wave. The difference in absorption in a waveguide between the
two polarizations arises from the differences in boundary conditions of the two
polarization states. The waveguide equations for the two cases can be obtained in
a manner identical to that of the slab-waveguide problem in Chap. 3, by simply
inserting a complex index for the upper or lower dielectric layer. These equations
then give the complex propagation constant for light in a metal-clad waveguide of
thickness d. This complex propagation constant may be written as

k z = β − jα (8.3)

where α is the guided-wave absorption constant due to the presence of the metal film.
For the case of well-confined waveguide modes, i.e., those formed in a waveguide
with reasonably high-index contrast, it is possible to obtain approximate values for
both β and α. The simplest example considers the case of a metal film with a thick-
ness, d, such that d  1/α. Specifically, for β  α, and a relatively thick, strongly
8.2 Polarizing Elements 157

confined waveguide 1/γ  d, the propagation constants for the two polarizations
may be written as
 
1 (m + 1)π 2
βe ≈ kn f − (8.4)
2n f k d
 2
1 (m + 1)π
βm ≈ kn f − (8.5)
2n f k d

where the arctangent terms, such as those found in the normalized dispersion curves,
have been approximated as zero and is not seen in (8.4) and (8.5). The corresponding
absorption coefficients are then

−1 [(m + 1)π]2
αe ≈ ηi (8.6)
β E d 3 Re(γm )

and   
−1 [(m + 1)π]2 r i
αm ≈ ηi + ηr (8.7)
βm d 3 Re(γm ) n 2f r

where 
Re(γ) ≈ k n 2f − r (8.8)

i
ηr + jηi = 1+ j (8.9)
r − n 2f

and m is an integer.
Expressions containing higher order approximations for β and α may be found
in (Nishihara et al. 1989). Actual computations of metal-clad waveguide loss are
given in Fig. 8.3 for the case of a waveguide covered with an aluminum film and a
wavelength of 1.55 µm. Notice in the figure that TE light has a much smaller loss
than TM, namely, 0.6 dB/cm versus 36 dB/cm. Note that, even for the TE case, the
absorption loss in the metal-clad waveguide is not negligible. However, the use of a
thin dielectric buffer layer between the metal and guiding layers can further alleviate
the loss (see below). Note that in order to obtain the loss in d B/cm from α(μm −1 ),
it is necessary to multiply α by 87, 000.
As suggested above, the eigenvalue equations for TE and TM show that the relative
absorption constants for the two polarizations depend on the ratio of the dielectric
constants of the metal and guiding layers. Aluminum is one of the most practical
cladding materials, because it has a high ratio of the imaginary to real component
of the dielectric constant and it is used in standard Si fabrication. Also, for the
same reason, waveguides with lower index materials have a higher ratio of the two
dielectric coefficients.
158 8 Components for Polarization Control

Fig. 8.3 A plot of

computations of metal-clad
waveguide loss

The effects of the metal overlayer can be enhanced or attenuated by interposing

a dielectric between the overlayer and waveguide. For example, a dielectric layer
such as SiO2 or polymer is often interposed between the electrode and substrate in
semiconductor or LiNbO3 electro-optical devices so as to reduce absorption loss in
the metal. In fact, if the thickness of this dielectric layer is properly adjusted, the
selective loss of the TM mode can be greatly increased (Čtyrokỳ and Henning 1986)
so as to improve polarization selectivity. Finally, adjustment of the dielectric layer
also changes the shape and the nature of the waveguide mode.
The effects of a thin metal/dielectric/waveguide/substrate structure can be under-
stood by treating the optical response of this stack as that of a four-level dielectric
waveguide. Such an analysis shows that in addition to the usual dielectric slab-
waveguide modes obtained without the metal film, addition of the film allows the
formation of a surface plasmon mode, sometimes designated as a T M−1 mode. This
mode is important for a small thickness of the dielectric spacer layer. As the thick-
ness of this spacer layer, tsl , is increased, two effects are noticed. First, the loss of
the T M−1 mode decreases monatomically. However, in the case of other modes, as
tsl increases, the propagation loss first increases until a characteristic distance t p is
reached after which the modal loss then decreases. This value of tsl ≈ t p is approxi-
8.2 Polarizing Elements 159

Fig. 8.4 Plots of effects of a

buffer layer on the TM mode
in a metal-clad optical
waveguide using Ti-diffused
LiNbO3 C-plate

mately the same for all of the first few modes beyond T M−1 . Figure 8.4 shows this
behavior for a characteristic slab-waveguide stack for both the T M−1 and the TM
modes.
Second, the shape of the modes evolves and changes character as tsl increases.
The variation in modal shapes is shown for T M−1 and T M0 waveguides in Fig. 8.4
as tsl increases. Notice that as the transverse modal shapes evolve, they change into
the next-higher slab-waveguide modes. When tsl ≈ t p , this evolution is essentially
complete. An approximate value for t p can be obtained (Nishihara et al. 1989) for the
case of a relatively low-loss metal film, i.e., metal ≈ r , and for a waveguide away
from cutoff, i.e., Neff ≈ n f . In this case,
⎛
⎞

2  n2 − 
k n
tanh−1 ⎝ sl  2
r
tp ≈ 
f ⎠ (8.10)
n2 − n |r | n f − n 2b
f sl

Finally, there is a second approach to realizing a waveguide polarizer via the

use of a waveguide overlayer. This method uses birefringence cladding materials
to provide a mode-sensitive effective index to waveguide structure. For a guiding
layer with refractive index of n f , a birefringent cladding film with indices n ec and n m
c
in the cladding are chosen such that n ec < n f < n m c , or vice versa. In this way, one
mode will propagate within waveguide while the other mode will leak into cladding.
Typically, birefringent materials such as calcite (CaCO3 ) and polyimide have been
used to demonstrate the polarization functionality. Birefringence in overlayers has
160 8 Components for Polarization Control

been artificially created in other materials, e.g., disordered superlattices on semicon-

ductors and even through the use of standard waveguide fabrication techniques such
as Ti diffusion and annealed proton exchange in LiNbO3 .

8.3 Polarization Splitters

As mentioned earlier, a waveguide TE/TM splitter or router is a basic element in pho-

tonic integrated circuits that require polarization control, e.g., polarization-diversity
coherent optical receivers, or polarization shift keying. Several types of polariza-
tion splitters have been reported for the common material systems such as LiNbO3 ,
glass waveguides, and III–V semiconductors. These devices require the presence of
some form of waveguide birefringence. This birefringence can be obtained either
through the inherent birefringence in a 3D waveguide or through material properties.
For example, in the case of z-cut Ti-doped LiNbO3 , the two waveguide transverse
axes have refractive indices of n e and n o , respectively. While III–V semiconduc-
tors do not have an intrinsic material birefringence, polarization-dependent behavior
in III–V devices can be obtained with a birefringent passive overlayer, waveguide
birefringence, strained layers, or the electro-optic effect. For glass waveguides, bire-
fringence has been typically obtained through the addition of a thin high-index layer
such as Si3 N4 , TiO2 , and Si.
In addition to thin film materials methods, there are two general categories of
waveguide polarization routers or sorters. One type uses asymmetric Y -branches to
do mode sorting by choosing the index of one of the two output waveguides so as
to guide either the TE or the TM mode in that output arm. The other method uses
interference within either a Mach–Zehnder interferometer or a directional coupler in
order to accomplish polarization-sensitive manipulation of the phases of the modes
in each arm. The phase of these modes subsequently allows “polarization-tagged”
modal steering, typically in a symmetric Y -branch. This latter device type was the first
used in polarization routing; however, it is the more wavelength- and dimensional-
sensitive of the two types. In this section, we will describe several examples of each
of these two types.
Figure 8.5 shows the basic structure used to accomplish mode sorting in an asym-
metric Y -branch. The device consists of single-mode input and output waveguides
joined through a dual-mode transition region located just prior to the split in the
Y -branch. As described in Chap. 7, an asymmetric Y -branch, if designed as a mode
sorter, will cause the fundamental mode to be routed into the guide with the higher
β and the first higher order mode into the arm with the lower β. This mode sorting
occurs efficiently when the guide is designed such that
β
> 0.43 (8.11)
θγ
8.3 Polarization Splitters 161

Fig. 8.5 A schematic of an asymmetric Y -branch, which acts as a mode evolution type polarization
splitter on InGaAsP/InP

where the symbols are those defined for the same inequality in Chap. 7, i.e., the
adiabatic criteria discussed earlier. When used as a mode sorter, the asymmetric Y in
Fig. 8.5 is designed such that the propagation constant β for the fundamental mode
is associated with a different output waveguide for each of the lowest order TE and
the TM modes. Further, the dual-mode waveguide transition region is designed so
that only the fundamental modes of each polarization are excited. This requirement
is essential to keep the device polarization crosstalk low.
Several different versions of such a device have been successfully demonstrated.
For example, one device (Goto and Yip 1989) used a branching waveguide in LiNbO3 ,
in which one arm had its index formed by proton exchange, while the other was
formed by Ti in diffusion. The proton-exchange process created a strong waveguide
for the fundamental TM mode, while the Ti-diffused region was used to form the
waveguide for the TE mode; see the discussion in Chap. 2. In this device, extinction
ratios of 20 dB at λ = 0.633 µm were achieved for each polarization. A second device
design (van der Tol and Laarhuis 1991), also in LiNbO3 , used the fact that the ordinary
and extraordinary indices in z-LiNbO3 exhibits a reversal in relative magnitude as
the Ti concentration is increased. Thus a careful choice of the waveguide doping,
as controlled by the thickness and width of the deposited Ti metal strip prior to
diffusion in each of the arms, can cause the higher β to be in opposite arms for
each of the polarization modes. As described above, this allows separation of the
two polarizations from an input with arbitrary polarization. Simulations showed
that this device should have ∼30 dB extinction ratio for λ = 1.55 µm light and a
waveguide separation angle of 0.1◦ . A related device design, using ridge waveguide
birefringence, has been described and demonstrated in InGaAsP/InP (Van der Tol
et al. 1993) and shown to have a 20 dB splitting ratio at λ = 1.55 µm. This device
was ∼3 mm in length.
162 8 Components for Polarization Control

Fig. 8.6 A schematic of an adiabatic-tapered asymmetric Y -branch, which is designed to “sort”

the lowest order TM and TE modes from a mixed-polarization input signal

One of the highest performance polarization splitters has been made using mode
sorting in a silica-on-silicon structure (Shani et al. 1991). This device, depicted in
Fig. 8.6, used an adiabatic-tapered asymmetric Y -branch to “sort” the lowest order
TM and TE modes from a mixed-polarization input signal. The design used the
strong waveguide birefringence obtained with a Si3 N4 layer loaded on a p-glass silica
waveguide. In this structure, a correct choice of the nitride-film thickness caused its
effective index for the mode to be larger than that for the p-glass waveguide, while
the reverse was true for the TM mode. Thus the signal is routed through the nitride-
covered arm, while the TM mode propagates through the pure P-glass arm. Finally,
as shown in Fig. 8.6, after separation, the nitride film is tapered again so as to allow
rejoining the purely P-glass waveguide structure. In this glass waveguide, extinction
ratios of 20 − 34 dB were achieved with an insertion loss of ∼1.5 dB at λ = 1.55 µm.
Silica-on-silicon technology allows easy and loss-free integration of multiple
structures. Hence, it is also possible to fabricate a series of two asymmetric Y -
branch mode sorters; this integrated mode sorter device, shown also in Fig. 8.6,
allowed even higher extinction ratios to be obtained, namely, −35 to −45 dB, than
for a single device. These extinction ratios indicate that higher performance has thus
been achieved in glass waveguides than in the LiNbO3 devices discussed earlier.
However, this performance comes at a price since it requires that the device lengths
for glass waveguides be long, i.e., >14 mm. Thus the length scale of the integration
is very large.
It is also possible to make an active variant of the mode-sorting device. In one
example, made in AlGaAs/GaAs, the electro-optical effect was used to change the
phase in the arms of a Mach–Zehnder. Note that while the principle of electro-
optical phase retardation is described later in Chap. 12, we include a brief discussion
of electro-optical devices in this chapter in order to provide a complete discussion.
8.3 Polarization Splitters 163

Fig. 8.7 Top view of a Mach–Zehnder TM/TE splitter using height-tapered Y-branches. The insert
shows a height-tapered symmetric Y-branch used as the input of the interferometer [Adapted from
(Hu et al. 1997)]

Thus, depending on the applied voltage and the polarization state, the output of the
Mach–Zehnder would generate the lowest symmetric and antisymmetric waveguide
modes at the entrance of an asymmetric Y -branch (see Fig. 8.7) (Hu et al. 1997).
The Y -branches in this device used tapering of the waveguide thickness, and, hence,
refractive index, to ensure adiabatic transition regions. The electrode configuration
and the interferometer path lengths were chosen such that for zero applied voltage,
TM light was routed through one waveguide while, when the voltage was switched
on, only light was passed by the other waveguide. The device had a 20 dB extinction
ratio and could be used either in 1.3 and 1.55 µm bands, after readjustment of the
voltage.
The second type of polarization router relies on the use of polarization-dependent
modal interference. An example of an early device of this type is shown in Fig. 8.8. In
this structure, a single-mode input waveguide, containing a combination of TE- and
TM-polarized light, feeds a dual-mode mixing region. When this region is excited by
the input waveguide, equal intensities of the first symmetric and asymmetric modes
of the wider waveguide are created for each of the TE and TM waves, separately. This
mode mixture propagates through the wider waveguide. The length of the guide and
the modal birefringence are designed such that at the end of the device, the TE modes
have phases such that they will couple out on one of the output waveguides, while
the TM modes will have the relative phases to couple out on the other waveguide.
Several devices of those described above have been made in LiNbO3 that utilize
polarization-sensitive directional coupling. One of the first of these devices, shown in
Fig. 8.9, was designed using waveguide birefringence so as to have a length which was
exactly designed to form the cross and bar state for each of the two polarizations,
using the interference approach just described. This passive device was designed
for 0.78 µm; it was fabricated and found to have a ∼12 dB splitting ratio for both
164 8 Components for Polarization Control

Fig. 8.8 A schematic of an early polarization router which relies on the use of polarization-
dependent modal interference

Fig. 8.9 A Schematic of a TE–TM mode splitter fabricated by diffusing titanium into lithium
niobate

polarizations at a wavelength of 0.78 µm (Yap et al. 1984). The performance of the

device was limited by the difficulty of optimizing the design for both TE and TM
splitting; it also requires high fabrication tolerance. A related version of this device
has been discussed which utilizes the “birefringence” in the coupling coefficients
in a passive Ti: LiNbO3 waveguide directional coupler (Alferness and Buhl 1984).
This “birefringence” would enable the coupler to be made such that κT E L = π while
κT M L = π/2, and hence, allowing splitting of the polarizations.
A second design (Maruyama et al. 1995) used material birefringence in the waveg-
uide, obtained with the use of the two main different doping techniques, i.e., Ti
diffusion or annealed proton exchange (APE) to fabricate the waveguides on two
8.3 Polarization Splitters 165

Fig. 8.10 A Schematic of a TE–TM mode splitter using material birefringence in the waveguide,
obtained with the use of two different doping techniques [Adapted from (Maruyama et al. 1995)]

Fig. 8.11 A schematic of an active version of the interference-based asymmetric coupler splitter

different arms of the coupler. Specifically, in this case, the input waveguide of the
waveguide directional coupler and one of the coupler waveguide arms was formed
by Ti-diffusion, while the other output arm was formed by APE (see Fig. 8.10). The
device operated as a result of the fact that only a TE-like mode can propagate in
the APE waveguide (see Chap. 2) and thus only this mode can couple into the APE-
formed output waveguide (see Fig. 8.10). Note that the TM mode remained in the
original input waveguide. With this device, an extinction ratio of >12 dB, with an
insertion loss of 2.4 dB, was demonstrated for a coupler length of < 1mm and at
λ = 0.813 µm.
Active versions of the interference-based asymmetric coupler splitters have also
been demonstrated. An example by (Mikami 1980) is shown in Fig. 8.11. In this
device, with the voltage on, the electro-optical effect is used to make the two parallel
waveguides highly asymmetric for the TM light entering at the input. Thus transfer to
the adjacent guide does not occur. On the other hand, the index is not changed by the
166 8 Components for Polarization Control

applied voltage and thus coupling to the adjacent guide does occur. More complex
electrode geometries can be used which make the device less sensitive to the precise
voltage. For example, one device has used a Mach–Zehnder interferometer with
one arm covered by a split reversed −β electrode (Alferness and Buhl 1984). In
addition, the waveguides were fabricated so as to be strongly birefringent to TE and
TM polarization. This device achieved polarization cross talk of each polarization, at
1.33 µm of ∼30 dB for extinction of a single polarization and a cross talk of −24 dB
when both TE and TM were simultaneously extracted. The device was ∼1 cm in
length.
Finally, recently, more sophisticated interference-like optical structures have been
designed and tested for polarization routing. These have used phase manipulation
involving the multimode imaging devices discussed in Chap. 9. One design, for exam-
ple, used a polarization-sensitive interferometer coupled in and out by 3 dB multi-
mode imaging devices (Soldano et al. 1994). These devices are based on the fact that
the input phase of a 2 × 2 MMI imaging device (see Chap. 9) depends on the phase
of the input beams. This device was made in InGaAsP/InP and was measured to have
∼19 and 15 dB extinction ratios for TE and TM light at 1.5 µm, respectively.

8.4 Polarization Converters

A polarization converter rotates any input polarization to the specific desired output
polarization state. Generally, it is desirable to have 90◦ rotation so as to convert TE
modes to TM modes. One important application for polarization conversion devices is
in polarization diversity receivers, such as might be used in a coherent optical system.
Another application is to average the polarization response of a complex symmetric
waveguide path, such as that found in a PHASAR-based wavelength router, over the
two, TE and TM, polarization directions.
Both active and passive polarization converters have been reported. For example,
active devices for polarization mode conversion have been described based on electro-
optical, acousto-optical, and even magneto-optical tuning. However, active devices
are generally, by their very nature, more complicated and require external power
and stability control. Thus, a passive polarization converter is preferable, especially
when, as in most cases, a fixed degree of conversion is needed.
Thus far in this chapter, it has been sufficient to specify the polarization state as
either TE or TM, or more exactly, as TE- or TM-like. With polarization converters or
even more important, with polarization controllers, it is sometimes useful to adopt a
more complex notation. In this case, the complex normalized amplitudes are specified
by a Jones vector (Saleh and Teich 2007)
   
αT E cos θ
= (8.12)
αT M sin θe jφ
8.4 Polarization Converters 167

Fig. 8.12 A schematic of a passive, plate-type converter with an inserted half-wave plate made of
a thin polyimide sheet

which specifies both the relative phase, φ, of the two components and the angle of
the polarization vector, θ. For example, right circularly polarized light is given by
φ = π/2 and θ = π/2.

8.4.1 Passive, Plate-Type Converters Placed Normal to the

Waveguide Axis

One conceptually simple passive, micro-optics solution to polarization conversion

is to insert a half-wave plate into a slot within the waveguide circuit. This mounting
method is similar or even identical to that used for the polarization plates mentioned
earlier in this chapter. The inserted half-wave plates are made of polyimide sheets,
which can be as thin as 10 − 15 µm (see Fig. 8.12). These sheets enable one compo-
nent to be converted into another, say TE into TM without attenuation. This device
thus allows polarizations to be interchanged. Thus if inserted into the middle of a
symmetric device, such as a bend, each of the two original components will sample
identical path lengths.
In this approach, a short path length, within the plate and its slot, is essential to
reduce the diffraction losses as the waveguide mode propagates across the free-space
region within the slot and the plate. Several methods have been made to reduce the
diffraction losses. One approach uses a relatively thick half-wave plate coupled with
an expanded waveguide core to reduce diffractive loss. The second uses a thin plate
to reduce waveguide diffraction, in conjunction with a standard waveguide core.
168 8 Components for Polarization Control

Use of a half-wave plate technique is conceptually straightforward but it does

rely on hybrid integration, which can be difficult in practice. Mounting of the plates
requires a precision sawing procedure, which has been developed for silica, but
typically not for other waveguide material systems. The groove or slot can be as
small as ∼18 µm in width in silica waveguides. A total insertion loss of <0.26 dB at
1.55 µm with a proper index matching has been obtained in silica with no significant
extinction degradation, with a 14 µm thick plate; the modal conversion was 99%
(Inoue et al. 1997).

8.4.2 Integrated Passive Waveguide Converters

It is also possible to use waveguide devices to form passive integrated polarization

converters out of a standard isotropic waveguide medium such as GaAs, InP, and
Si each of which has a high index. In general, these devices operate by interference
between hybrid modes (containing both TE and TM components) of waveguides,
which are designed to have the desired hybrid-modal structure.
The first general class of devices uses asymmetric waveguide structures to form
hybrid modes. These hybrid modes can then interfere to cause an overall polarization
rotation. Consider three kinds of such asymmetric structures: asymmetric-loaded
waveguides, tilted-loaded rib waveguides and angle-sidewall waveguides; these are
investigated in (Sun et al. 2015; Heidrich et al. 1992; El-Refaei et al. 2004). All
three generally have a longitudinally periodic structure and multiple sections. The
transverse waveguide asymmetry between any two adjacent sections is designed
to be in opposite direction. Phase-matching and hence efficient polarization rotation
occurs when the interference length (or period) is such that the two hybrid modes have
π phase difference. This interference length, L π , is determined by the propagation
constants β1 and β2 of the two fundamental hybrid modes:
π
Lπ = (8.13)
β1 − β2

After one L π length, the lateral asymmetry is reversed to allow a second addition of
the hybrid modes. After several such periods, “pure” TE and “pure” TM modes are
then obtained.
The most readily understandable device of this type is that having a single section.
Consider the case of an angle-sidewall device with a 45◦ sidewall. It is possible, by
a careful use of crystallographic and anisotropic etching, to make such a passive
polarization converter design (Tzolov and Fontaine 1996) which has sufficiently
hybrid modes that it does not retain longitudinal periodicity. This device has recently
been demonstrated in GaAs (Huang et al. 2000). This design uses the structure of
an asymmetric, angle-sidewall waveguide such that its optical axes are rotated 45◦
with respect to that of a comparable symmetric step waveguide. Thus the modes of
the angle-sidewall waveguide are sufficiently hybrid that only one section is needed
8.4 Polarization Converters 169

Fig. 8.13 A sketch of an asymmetric, angle-sidewall waveguide (Huang et al. 2000)

for polarization rotation; its rotation per unit length is thus larger than that used in
the multiple-section designs.
The asymmetric waveguide structure of this single section device is shown in
Fig. 8.13. The optical axis of the device is at a 45◦ angle with respect to the fixed
coordinate system of the device (x,y in Fig. 8.13); thus the TE (x axis) and TM (y axis)
components of each of two zero-order modes will have comparable field amplitudes
and very similar field distributions. When linearly polarized, say, TE polarized light
is launched into this longitudinally invariant, single-mode waveguide, only the two
zero-order hybrid modes are excited and at each half beat-length, a 90◦ polarization
rotation is achieved; for this example, after a single section, the output would then
be TM polarized. The beat-length L π is determined from the propagation constants
β1 and β2 of the two zero-order hybrid modes:

λ
L π = π/(β1 − β2 ) = (8.14)
2Neff

where Neff is the difference between the effective modal index of these two hybrid
modes and λ is the operating wavelength. In this 90◦ polarization rotation case,
the converted power into TM-polarized light can be expressed as a function of the
interaction length, z, for TE input light:
 
z
PT M (z) = PT E (0) sin 2
(8.15)
Lπ

In addition, the percent conversion, η, is given by

170 8 Components for Polarization Control
 
PT M (z) π·z
η= = sin2 (8.16)
PT M (z) + PT E (z) 2L π

where PT M (z) and PT E (z) are the power in TM and TE polarization, respectively.
Equation 8.16 assumes that there is no differential loss between TE and TM modes.
Also as seen from (8.16), any output polarization state can be obtained by controlling
the interaction length, z, between the two fundamental modes.
As mentioned earlier in this section, multiple sections are needed for the more
general case when the modes are not hybrid enough to allow a rotation of 90◦ in one
period. Of course, in this case, the device length increases because more sections
are needed to achieve 100% polarization conversion. Multiple section polarization
converters also suffer period-to-period coupling losses at each junction.
A second type is based on the conversion of hybrid supermodes and can also be
done in a single short section. The device (Mertens et al. 1995, 1998) uses the fact
that in strip-loaded waveguides, the T M11 and te21 modes degenerate in β and can
thus couple for certain waveguide dimensions. This coupling allows the resulting
set of coupled modes to be excited by an input TM wave. The excited modes may
then interfere at specific lengths, again given by L π = π/(β2 − β1 ), so as to yield a
rotated polarization in the form of an output wave. This device works well in practice;
93% conversion has been obtained from a 250 µm-long InP device (Mertens et al.
1998). However, these devices are not single-mode devices, since they use conversion
between the T M11 and T E 21 modes. Thus, for TM to TE conversion, the output is in
the first-order mode. In order to obtain the single-mode output required for practical
applications, either a mode converter must be used or a monomode waveguide at the
output must be employed to collect a half of the first-order output TE mode. The
latter approach would result in a minimum of a 3 dB insertion loss. The design also
converts a single-mode TE input to TM polarization.

8.4.3 Active Polarization Converters

A more complex but active approach for polarization rotation has been demonstrated
using an electro-optical device. A sketch of this device is given in Fig. 8.14. The
device is a Ti-diffused waveguide on x-cut LiNbO3 , oriented along the z-direction.
The waveguide is covered with an interdigitated waveguide, with the field oriented
along the waveguide, but reversed over a distance of length . This field reversal field
produces a dielectric periodicity similar to that described earlier for the multisector
polarization rotator.
Because of the z-orientation of the applied electric field, E (0) the TM and TE
modes are coupled via the off-diagonal element of the electro-optical tensor, r51 .
The coupling coefficient in this case (see Chap. 12) is given by

π 3 V (0)
κ= n r51 (8.17)
λ s d
8.4 Polarization Converters 171

Fig. 8.14 A schematic of an active EO polarization rotator in which any output polarization state
can be obtained by controlling the interaction length, z, between the two fundamental modes

where d is the electrode spacing, V (o) the applied voltage, n s is the zero-bias sub-
strate refractive index, and  the overlap factor as given in Chaps. 12 and 13. The
interdigitated electrode grating also provides the phase matching for the TM and TE
modes which have different effective indices:
2π T M 2π
(Neff − Neff
TE
)− =0 (8.18)
λ 
An active polarization controller allows input light to be changed from an input
of arbitrary polarization to an output of a second arbitrary polarization at the output.
This capability is important in fiber-optic communication systems, which are not
completely polarization insensitive or which require a specific polarization, e.g.,
coherent communication systems.
This device has been made in LiNbO3 , so as to allow electro-optic phase control
and rotation of the polarization. The device is generally oriented in the z-direction
of the LiNbO3 slab so as to minimize any wavelength shifts. A sketch of this device
is given in Fig. 8.15.
The performance of this device can be analyzed by multiplying the Jones matrices
of each of the three segments shown in the figure together. A detailed discussion of this
approach is provided in (Alferness and Buhl 1981); here we will only provide a brief
qualitative description based on this more complete discussion in these references.
Specifically, the device uses an input phase shifter which is adjusted so that the
relative phase, φi , between the two polarizations is ±π/2 at the entrance to the
mode converter. This phase adjustment allows the TE–TM converter to achieve a
172 8 Components for Polarization Control

Fig. 8.15 The schematic of active polarization controller which allows input light to be changed
from an input of arbitrary polarization to an output of a second arbitrary polarization at the output
[Adapted from (Alferness and Buhl 1981)]

rotation of the input polarization that is linearly controlled by the coupling coefficient,
κ. i.e., θo − θi = ±κL 2 and hence the electro-optical voltage. The relative phase of
the output, φ, is then controlled by the output phase shifter. Thus if the phase
shift imparted by the output phase shifter is φs , the output phase is given by φ0 =
φs ± π/2 when φi = ±π/2.
The device has several useful modes of operation. For example, if the first phase
shifter is adjusted so that φi = −π/2 and the second shifter is adjusted to set
φ0 = 0, the device then acts as a polarization rotator, with the rotation controlled
by the voltage through κ. The device can also be used in a feedback-controlled mode
in conjunction with a polarization sensor. Operated in this form, the device can keep
the polarization fixed in an optical system.

8.5 Polarization Devices for Si Photonics

As it is clear from the earlier discussion in this chapter, maintaining or controlling

polarization is key to high-performance photonics systems and thus it is not surprising
that the increasing role of Si photonics in optical subsystems and systems has also
generated much interest in novel methods for polarization control via integrated Si
waveguide devices. Our intent in this section is to present several of these devices
and examine their performance parameters and operational methods. Some recent
polarizer research has been nicely summarized in an extensive article by (Dai et al.
2012) (see in addition (Wang and Dai 2008; Dai 2012)). Note that the reader should
keep in mind that the polarization properties of pure Si waveguide devices, which are
8.5 Polarization Devices for Si Photonics 173

attributable to silicon’s high-index contrast can also be found also in earlier devices
based on GaAs or other III–V materials.
One interesting approach to polarization in an integrated Si photonics which makes
use of the light bending radius that is enabled by Si’s high-index contrast. This
approach is shown in (Paredes et al. 2016); it uses a simple series of waveguide
bends to achieve high loss for TM light and thus making an integrated TE-pass
polarizer. The device was measured to have a 20 dB TM/TE extinction ratio over a
50 nm spectral bandwidth with a 1–2 dB insertion loss.
A second device type uses asymmetrical-coupling based on polarization beam
splitting (PBS) (Dai et al. 2012). One version of this device consists of a combination
of SOI nanowires to form the asymmetric coupler. It allows separation of TE- and
TM-polarized light within a very short propagation length. Using this approach,
a short (25 µm long) PBS was designed based on silicon-on-insulator nanowires.
Numerical simulations showed that the PBS had good fabrication tolerance and
a broadband (50 nm) response and an extinction ratio of 15 dB. The device was
designed for phase matching of any TM polarization light to dominate within a short
length determined by the strength of the coupling region. For the TE polarized wave,
the phase-matching condition was not fulfilled and hence the coupler was always in
the bar state. Polarization management devices have also been demonstrated on the
silicon nitride (Si3 N4 ) on silicon-on-insulator (SOI) platform (Sacher et al. 2014).
SiO2 -clad silicon nitride (Si3 N4 ) waveguides which are an alternative Si photonics
materials platform are sometimes preferred for polarization control since they are
insensitive to thermal drift (i.e., they have low thermo-optical effects) and they have
low waveguide loss (due to the low-index contrast of the nitride, which reduces
sidewall-roughness. In addition, Si3 N4 waveguides can be readily integrated onto
SOI photonic platforms or vice versa to achieve CMOS-compatible integration.
Polarization control in these devices includes a broadband polarization rotator-
splitter using a T M0 − T E 1 mode converter in a composite Si3 N4 -silicon waveguide
(Sacher et al. 2014). In one device, the measured polarization cross talk, insertion
loss, and polarization-dependent loss were less than ∼19 dB, 1.5 dB, and 1.0 dB,
respectively, over an 80 nm bandwidth. These same materials’ platform enabled a
polarization controller composed of polarization-rotator-splitters, multimode inter-
ference couplers, and thin film heaters to be fabricated and tested. In particular the
polarization rotator-splitter (PRS) and a polarization controller were designed for
the Si3 N4 -on-SOI integrated photonics platform. These polarization devices are of
the type needed for control of polarization in polarization diversity operation and in
polarization demultiplexers and multiplexers.

8.6 Conclusion

This chapter has described devices to be used for manipulation of polarization in inte-
grated photonic circuits. The devices use a wide range of materials and approaches,
including the recent emphasis on Si devices. Historically, however, the most advanced
174 8 Components for Polarization Control

devices are in the area of LiNbO3 for switch arrays. The approaches use passive
micro-optical insertion of half wave plates and integrated devices as well as active
polarization rotators and routers. While polarization control is crucial in many areas
of integrated optics, research continues on methods to improve device performance.

References

Albrecht, P., Hamacher, M., Heidrich, H., Hoffmann, D., Nolting, H., & Weinert, C. (1990). Te/tm
mode splitters on ingaasp/inp. IEEE Photonics Technology Letters, 2(2), 114–115.
Alferness, R., & Buhl, L. (1981). Waveguide electro-optic polarization transformer. Applied Physics
Letters, 38(9), 655–657.
Alferness, R.C., & Buhl, L. (1984). Low-cross-talk waveguide polarization multi-
plexer/demultiplexer for λ = 1.32 μm. Optics Letters, 9(4), 140–142.
Čtyrokỳ, J., & Henning, H.-J. (1986). Thin-film polariser for ti: Linbo3 waveguides at λ = 1.3 µm.
Electronics Letters, 22(14), 756–757.
Dai, D. (2012). Silicon polarization beam splitter based on an asymmetrical evanescent coupling
system with three optical waveguides. Journal of Lightwave Technology, 30(20), 3281–3287.
Dai, D., Bauters, J., & Bowers, J.E. (2012). Passive technologies for future large-scale photonic
integrated circuits on silicon: polarization handling, light non-reciprocity and loss reduction.
Light: Science & Applications, 1(3), e1.
El-Refaei, H., Yevick, D., & Jones, T. (2004). Slanted-rib waveguide ingaasp-inp polarization con-
verters. Journal of Lightwave Technology, 22(5), 1352.
Glance, B. (1987). Polarization independent coherent optical receiver. Journal of Lightwave Tech-
nology, 5(2), 274–276.
Goto, N., & Yip, G. L. (1989). A te-tm mode splitter in linbo/sub 3/by proton exchange and ti
diffusion. Journal of lightwave technology, 7(10), 1567–1574.
Hayakawa, T., Asakawa, S., & Kokubun, Y. (1997). Arrow-b type polarization splitter with asym-
metric y-branch fabricated by a self-alignment process. Journal of lightwave technology, 15(7),
1165–1170.
Heidrich, H., Albrecht, P., Hamacher, M., Nolting, H.-P., Schroeter-Janssen, H., & Weinert, C.
(1992). Passive mode converter with a periodically tilted inp/gainasp rib waveguide. IEEE Pho-
tonics Technology Letters, 4(1), 34–36.
Hu, M., Huang, J., Scarmozzino, R., Levy, M., & Osgood, R. (1997). Tunable mach-zehnder polar-
ization splitter using height-tapered y-branches. IEEE Photonics Technology Letters, 9(6), 773–
775.
Huang, Z., Scarmozzino, R., Nagy, G., Steel, J., & Osgood, R. (2000). Realization of a compact and
single-mode optical passive polarization converter. IEEE Photonics Technology Letters, 12(3),
317–319.
Inoue, Y., Ohmori, Y., Kawachi, M., Ando, S., Sawada, T., & Takahashi, H. (1994). Polarization
mode converter with polyimide half waveplate in silica-based planar lightwave circuits. IEEE
Photonics Technology Letters, 6(5), 626–628.
Inoue, Y., Takahashi, H., Ando, S., Sawada, T., Himeno, A., & Kawachi, M. (1997). Elimination
of polarization sensitivity in silica-based wavelength division multiplexer using a polyimide half
waveplate. Journal of Lightwave Technology, 15(10), 1947–1957.
Maruyama, H., Haruna, M., & Nishihara, H. (1995). Te-tm mode splitter using directional coupling
between heterogeneous waveguides in linbo/sub 3. Journal of Lightwave Technology, 13(7),
1550–1554.
Masuda, M., & Koyama, J. (1977). Effects of a buffer layer on tm modes in a metal-clad optical
waveguide using ti-diffused linbo 3 c-plate. Applied Optics, 16(11), 2994–3000.
References 175

Mertens, K., Opitz, B., Hovel, R., Heime, K., & Schmitt, H. (1998). First realized polarization
converter based on hybrid supermodes. IEEE Photonics Technology Letters, 10(3), 388–390.
Mertens, K., Scholl, B., & Schmitt, H. J. (1995). New highly efficient polarization converters based
on hybrid supermodes. Journal of Lightwave Technology, 13(10), 2087–2092.
Mikami, O. (1980). Linbo3 coupled-waveguided te/tm mode splitter. Applied Physics Letters, 36(7),
491–493.
Nishihara, H., Haruna, M., & Suhara, T. (1989). Optical integrated circuits (Vol. 1). New York:
McGraw Hill.
Okuno, M., Sugita, A., Jinguji, K., & Kawachi, M. (1994). Birefringence control of silica waveguides
on si and its application to a polarization-beam splitter/switch. Journal of Lightwave Technology,
12(4), 625–633.
Paredes, B., Zafar, H., Dahlem, M.S., & Khilo, A. (2016). Silicon photonic te polarizer using adia-
batic waveguide bends. In 2016 21st OptoElectronics and Communications Conference (OECC)
held jointly with 2016 international conference on Photonics in Switching (PS), pp. 1–3. Piscat-
away: IEEE.
Sacher, W. D., Huang, Y., Ding, L., Barwicz, T., Mikkelsen, J. C., Taylor, B. J., et al. (2014).
Polarization rotator-splitters and controllers in a si 3 n 4-on-soi integrated photonics platform.
Optics Express, 22(9), 11167–11174.
Saleh, B.E. & Teich, M.C. (2007). Fundamentals of photonics (pp. 260–269). Hoboken: Wiey.
Sato, T., Baba, K., Hirozawa, T., Shiraishi, K., & Kawakami, S. (1993). Fabrication techniques and
characteristics of al-sio/sub 2/laminated optical polarizers. IEEE Journal of Quantum Electronics,
29(1), 175–181.
Shani, Y., Alferness, R., Koch, T., Koren, U., Oron, M., Miller, B., et al. (1991). Polarization rotation
in asymmetric periodic loaded rib waveguides. Applied Physics Letters, 59(11), 1278–1280.
Sieger, M., & Mizaikoff, B. (2016). Optimizing the design of gaas/algaas thin-film waveguides for
integrated mid-infrared sensors. Photonics Research, 4(3), 106–110.
Soldano, L., De Vreede, A., Smit, M., Verbeek, B., Metaal, E., & Green, F. (1994). Mach-zehnder
interferometer polarization splitter in ingaasp/inp. IEEE Photonics Technology Letters, 6(3),
402–405.
Suematsu, Y., Hakuta, M., Furuya, K., Chiba, K., & Hasumi, R. (1972). Fundamental transverse
electric field (te0) mode selection for thin-film asymmetric light guides. Applied Physics Letters,
21(6), 291–293.
Sugimoto, N., Terui, H., Tate, A., Katoh, Y., Yamada, Y., Sugita, A., et al. (1996). A hybrid integrated
waveguide isolator on a silica-based planar lightwave circuit. Journal of Lightwave Technology,
14(11), 2537–2546.
Sun, Y., Xiong, Y., & Winnie, N. Y. (2015). Compact soi polarization rotator using asymmetric
periodic loaded waveguides. IEEE Photonics Journal, 8(1), 1–8.
Suzuki, Y., Iwamura, H., and Mikami, O. (1990). Te/tm mode selective channel waveguides in
gaas/alas superlattice fabricated by sio 2 cap disordering. IEICE Transactions (1976–1990),
73(1), 83–87.
Takahashi, H., Hibino, Y., & Nishi, I. (1992). Polarization-insensitive arrayed-waveguide grating
wavelength multiplexer on silicon. Optics Letters, 17(7), 499–501.
Tzolov, V. P., & Fontaine, M. (1996). A passive polarization converter free of longitudinally-periodic
structure. Optics Communications, 127(1–3), 7–13.
Uehara, S., Izawa, T., & Nakagome, H. (1974). Optical waveguiding polarizer. Applied Optics,
13(8), 1753–1754.
Van Der Tol, J., Hakimzadeh, F., Pedersen, J., Li, D., & Van Brug, H. (1995). A new short and
low-loss passive polarization converter on inp. IEEE Photonics Technology Letters, 7(1), 32–34.
Van der Tol, J., Pedersen, J., Metaal, E., Oei, Y., Van Brug, H., & Moerman, I. (1993). Mode
evolution type polarization splitter on ingaasp/inp. IEEE Photonics Technology Letters, 5(12),
1412–1414.
van der Tol, J. J., & Laarhuis, J. H. (1991). A polarization splitter on linbo/sub 3/using only titanium
diffusion. Journal of Lightwave Technology, 9(7), 879–886.
176 8 Components for Polarization Control

Veselka, J., & Bogert, G. (1987). Low-insertion-loss channel waveguides in linbo3 fabricated by
proton exchange. Electronics Letters, 23(6), 265–266.
Wang, Z., & Dai, D. (2008). Ultrasmall si-nanowire-based polarization rotator. JOSA B, 25(5),
747–753.
Wei, P.-K., & Wang, W.-S. (1994). A te-tm mode splitter on lithium niobate using ti, ni, and mgo
diffusions. IEEE Photonics Technology Letters, 6(2), 245–248.
Yap, D., Johnson, L., & Pratt, G, Jr. (1984). Passive ti: Linbo3 channel waveguide te-tm mode
splitter. Applied Physics Letters, 44(6), 583–585.
Chapter 9
Imaging Devices

Abstract This chapter provides a detailed discussion of two important devices for
using the reconstruction of a number of spatial modes to carry out spatial multiplexing
or demultiplexing. The small number of modes involved means that the device is
relatively efficient. The two devices considered in this chapter are the multimode
interference device and the star coupler.

9.1 Introduction

This chapter includes discussion of two important imaging devices: multimode inter-
ference devices and star couplers. Imaging, in this case, refers to reconstruction of
an object in the two-dimensional plane of a guided-wave slab by using guided-wave
modes. Multimode interference devices operate by the interference of a discrete num-
ber of guided modes while the star coupler operates by the excitation of a continuum
of “free-space” modes, with the image formed in the far field. Further, both devices
enable 1×N splitting or N×M interconnection, although in the case of the multimode
interference devices, these numbers, N and M, are small—say, 1 − 16—while for
star couplers, their magnitudes maybe 100 s! In addition to power division, these
devices may be used for a variety of other functions; these functions are extensive
and in many cases subtle. These functions, which rely on their efficient operation
and easy design, will be explained in this chapter and other places throughout the
text.

9.2 Multimode Interference Devices (MMIs)

9.2.1 Overview

Modal effects and manipulation have been shown to be an effective tool for lateral
imaging in waveguides. The devices, which use multimode interference (MMI), can

© Springer Nature Switzerland AG 2021 177

R. Osgood jr. and X. Meng, Principles of Photonic Integrated Circuits,
Graduate Texts in Physics,
https://doi.org/10.1007/978-3-030-65193-0_9
178 9 Imaging Devices

Fig. 9.1 A numerical

simulation of the imaging
process of multimode
waveguide

be used to achieve optical routing and coupling functions. MMI-based devices have
the advantages of large bandwidth, small device dimension, polarization insensitiv-
ity, and low optical loss. Their useful properties have led to the demonstration of
a wide variety of advanced applications such as coherent receivers, variable power
splitters, WDM routers, etc. The application to wavelength routers, which is of par-
ticular importance, will be discussed in Sect. 9.3. A general sketch of the multimode
waveguide and its imaging process is shown in Fig. 9.1.

9.2.2 Self-imaging Principle and Mode Propagation Analysis

The self-imaging properties of waveguide structures were first described in detail

by Bryngdahl (1973), Ulrich (1975). Modal imaging or “self-imaging” results from
the repeating of the input field along the propagation direction in the multimode
waveguide. Typically these MMI devices have a 2D character: the imaging guide has
a single mode in the vertical direction and multiple modes in the horizontal direction.
Thus, imaging involves only the lateral modes of the waveguide.
The most common approach to analyze a multimode waveguide uses a guided-
mode-propagation analysis (MPA) of a 2D step-index multimode waveguide (Ulrich
and Kamiya 1978). To use this approach, the dispersion relation of the MMI waveg-
uide must be determined. To obtain the dispersion relation, we first assume that the
physical width of the guide is W , and that the waveguide has a high-index contrast.
Such a waveguide will have n lateral modes with mode numbers m = 0, 1, ..., (n − 1)
as depicted in Fig. 9.2:
9.2 Multimode Interference Devices (MMIs) 179

Fig. 9.2 A multimode waveguide with five lateral modes

k 2ym + k z2 = k02 n 2f (9.1)

or
k 2ym + βm2 = k02 n 2f (9.2)

where k0 = 2π/λ0 and k ym = (m + 1)π/Wem . Further, βm and Wem are the propa-
gation constant and the effective mode width, respectively, for mode m. Since the
waveguide is assorted to have high contrast, Wem ≈ W , or to a somewhat better
approximation:
  21
λ0 1
Wem ≈ We0 =W+ (9.3)
π n f − n 2c
2

for TE modes away from cutoff. In what follows, we will adopt the shorthand notation
of We0 = We . For highly multimode waveguides and for the case k 2ym  k02 n 2f , i.e.,
when the only lower order modes are excited, (9.2) can be approximated using a
Taylor’s series as

(m + 1)2 π 2
βm ≈ k0 n f − (9.4)
2k0 n f We2

The propagation constant and, hence, the phase shift of mode m can be conveniently
referenced to that of m = 0; thus,
m(m + 2)π
β0 − βm ≈ (9.5)
3L π

where
180 9 Imaging Devices

π 4n f We2
Lπ = = (9.6)
β0 − β1 3λ0

Thus, L π is the modal beat length of the two lowest order modes.

At the input we assume only the guided, i.e., no radiative or leaky, modes of the
wide waveguide are excited

(y, 0) = cm m (y) (9.7)
m

where the overlap factor, cm , gives the amplitude for the modes which are excited:
   21
cm = (y, 0)m (y)/ m2 (y)dy (9.8)

Equation 9.8 is based on modal orthogonality.

Then, referring the phase to that of the m = 0 mode,


n−1
(y, z) = cm m (y) exp( j (β0 − βm )z) (9.9)
m=0

and/or at z = L:


n−1    
m(m + 2)π
(y, z) = cm m (y) exp j L . (9.10)
m=0
3L π

It will be seen below that, at certain intervals of the distances to L π , the output field
(y, z) will be a reproduction or self-image of the input field (y, 0). The quality of
images is determined by cm , the modal coefficients that determine modal weighting,
and the phase factor of that same mode.

9.2.3 Imaging Modalities: General, Restricted, and Overlap

Imaging

In this section, we examine conditions for forming images of the lateral modes in an
MMI device or waveguide region with the application being, in general, for power
splitting. There are several operational mechanisms for image formation that depend
on how the modes in the MMI region are excited by the input field and how the excited
modes interfere at the output end of the device. If all modes are excited, the image
mechanism is termed “general interference.” If only certain modes are excited, the
mechanism is termed as “restricted interference.” Both types of interference result in
9.2 Multimode Interference Devices (MMIs) 181

uniform power splitting. A third interference mechanism, called “overlap imaging,”

may be used to obtain nonuniform power splitting.
General Interference
A single image in an imaging waveguide having parameters L π and We will occur
at a length L, such that
   
m(m + 2)π
exp j L = (−1)m (9.11)
3L π

irrespective of the values of cn , i.e., the relative modal content excited by the input
waveguide. At this axial distance (z), all of the original modes will be in phase
again, and the image of the input waveguide will be reconstituted. The length, L, for
imaging in this case is

L = p(3L π ) (9.12)

where p = 0, 1, 2 . . . , and its value labels the image periodicity along z.

In addition, further examination of the equation for modal fields (9.9) shows that
at distances: p
L = (3L π ) (9.13)
2
where p = 1, 3, 5, . . . , there is a double image described by
 
3 1 + (− j) p 1 − (− j) p
 y, p L π = (y, 0) + (−y, 0) (9.14)
2 2 2

thus having periodic images at 3L/2, 9L/2, 15L/2 . . .. Such a scheme allows the
formation of a x 2 -fold splitter.
In general, N -fold multiple images can also be found. This form of imaging can
be shown clearly by realizing that the modal field is approximately sinusoidal in a
highly confined waveguide, i.e.,

m (y)  sin(k ym y) (9.15)

Then, if
p
L= (3L π ) (9.16)
N
where the integers p and N are such that p > 0 and N > 1, and p and N do not
have a common divisor,
182 9 Imaging Devices

N −1

(y, L) = c in (y − yq ) exp( jφq ) (9.17)
q=0

where q identifies each of N images along y,

yq = p(2q − N )We /N (9.18)

φq = p(2q − N )qπ/N (9.19)

and
1
|c| = √ (9.20)
N
√
Thus, at z = L, N images are formed with amplitude 1/ N and phase φq . In this
case, as shown by (9.18), the N images are not equally spaced, but are periodically
distributed laterally across the waveguide in uniform fractions of units of We .
Consider an N × N MMI coupler with p = 1, so as to have the shortest length
of the imaging region (Bachmann et al. 1994). In the device, waveguides of finite
width are used as inputs and outputs. These waveguide dimensions impose the spa-
tial restrictions on the light distribution. In particular, for properly designed MMI
couplers, different images of the input field at the output usually should not overlap.
For a given MMI effective width We and a chosen number of images, N , the input
waveguide can be placed at an arbitrary position which is defined by an additional
free parameter α and 0 < α < We /N . Figure 9.3 shows the situation for N even and
N odd, respectively. Inputs and outputs are numbered with indices i and j, respec-
tively. The resulting phase for imaging input i to output j can be given now in a very
compact and direct form: when i + j is even,

φi j = π + φ N −( j−i)/2 = φ0 + π + (π/4N )( j − 1)(2N + j − i) (9.21)

and when 1 + j is odd,

φi j = π + φ N −( j−i)/2 = φ0 + (π/4N )(i + j − 1)(2N + j − i + 1) (9.22)

where φ0 is a constant phase, explicitly given by

φ0 = −β0 L π /N − π/N − (π/4)(N − 1) (9.23)

Thus, all the output-phase relationships for a practical N × N MMI coupler can be
determined by Eqs. 9.21, 9.22, and 9.23. These relationships permit the straightfor-
ward application of MMI couplers in the more complex integrated optical devices
9.2 Multimode Interference Devices (MMIs) 183

Fig. 9.3 N × N MMI couplers with even N(top) and odd N(bottom) [Adapted from (Bachmann
et al. 1994)]

such as generalized Mach–Zehnder switches and phased arrays for wavelength

(de)multiplexing, which will be discussed in Sect. 9.2.5.
Restricted Interference
Thus far, in this discussion, no restrictions have been placed on the modal excitation;
in fact, by selecting the modes to be excited leads to restrictive or selective imaging
as explained earlier. In practice, this method is achieved, for example, by shortening
the imaging distance, and hence the device length.
One possible condition for restricted interference is paired interference in the
MMI. In this case, selective mode excitation of certain pairs of waveguides is used to
obtain a desirable interference criterion. Then, it can be shown that the phase factor
   
m(m + 2)π
exp j L
3L π

is periodic in L = pL π instead of 3 pL π , if cm = 0 for m = 2, 5, 8, . . ., that is, if only

waveguides 0 and 1, and 3 and 4, etc. are excited. When only these particular modes
are excited, single images of the input field (y, 0) are obtained at L = pL π . The
device length is then three times shorter than that for a general interference device.
Similarly, as in the case of the general interference device, N -fold images are found
at L = ( p/N )L π . Thus, twofold images will occur at pL π /2 for p = 1, 3, 5, . . .
184 9 Imaging Devices

The selected set of modes needed for paired imaging at L = pL π can be excited
by launching a symmetric even input field (y, 0) at y = ±We /6 (see Fig. 9.2). By
inspection, this choice of launching position eliminates the excitation of the m =
2, 5, 8, . . . mode, since the spatial overlap of these modes with the input waveguides
is zero. However, the number of input waveguides is limited to two in this case. Such
short imaging devices have been made with I n P waveguides, e.g., one device was
107 µm long, with 0.9 dB excess loss and −28 dB cross talk, and was designed as a
3 dB coupler (Spiekman et al. 1994).
A second restricted interference condition can also be obtained using symmetric
interference that is exciting only even symmetric modes. In this case, the phase factor
   
m(m + 2)π
exp j L
3L π

is periodic with L = p(3L π /4), when cm = 0 for the odd modes, m = 1, 3, 5, . . .

In symmetric excitation of MMIs, the input waveguide is always in the center of
the multimode waveguide and excited by only a symmetric mode of the input guide.
In this case, single images of the input field (y, 0) are obtained at L = p((3L π )/4),
a fact, which can readily be seen by inserting sample even values of m in the phase
factor above. Thus, N -fold images are at L = ( p/N )((3L π )/4). The device length
is, in principle, four times shorter than for a general interference device. N -fold
output images of the input field (y, 0) are symmetrically located transverse to the
multimode waveguide axis, that is, the y-axis, with equal spacing We /N .
Table 9.1 summarizes the various properties for selective excitation.
Overlap Imaging
In both cases, i.e., general and restricted interference, each of the N -fold output
images will have equal output intensity. However, it is possible to achieve nonuniform
power splitting by selecting a specific set of positions for access waveguides such
that images of unequal intensities are formed (Bachmann et al. 1995). In essence, this
approach makes use of the phase of the images and achieves different image inten-

Table 9.1 Summary of the characteristics of the general-, paired-, and symmetric-interference
multimode imaging schemes (after Soldano and Pennings 1995)
Interference General Paired Symmetric
mechanism
Inputs × Outputs N×N 2×N 1× N
First single image 3L π Lπ 3L π /4
distance
First N-fold image (3L π )/N L π /N (3L π )/4N
distance
Excitation requirement None cm = 0 for cm = 0 for
m = 2, 5, 8, . . . m = 1, 3, 5, . . .
Inputs locations Arbitrary y = ±We /6 y=0
9.2 Multimode Interference Devices (MMIs) 185

Fig. 9.4 An overlap-imaging MMI coupler

sities by causing spatial overlap of two images with different phases; this approach
is thus called overlap imaging. To operate a MMI device, it is necessary to make use
of the relative phase in each of the images, which are overlapped. While the output-
phase relationship for the general interference case is given in (9.21), the discussion
of the phase in images has not been otherwise addressed in this chapter.
The concept of overlap imaging can be illustrated for the case of general interfer-
ence in the generic MMI device shown in Fig. 9.4. When α = 0 or α = W/N , output
images merge together in pairs, thus an overlap-imaging MMI coupler is formed and
the number of images at the output is reduced. For practical, i.e., the short version of
the MMI coupler shown in Fig. 9.4, the possible input and output waveguide positions
are
W
yiin = i (9.24)
N
where i = 0, 1, 2, 3, . . . , N − 1, N ,
W
y out
j =W−j (9.25)
N
where j = 0, 1, 2, 3, . . . , N − 1, N , and i + j is an even integer. Channels at the
edge of the coupler are denoted by numbers in parenthesis. Formulae for phases
and intensities are obtained using (9.26). This equation shows the results obtained
from the general interference results by considering the interference between the two
outputs that are merged:
4  π π
ri2j = cos2 (N − 1)i −b (9.26)
N 2N 2
for
186 9 Imaging Devices
 π π
cos (N − j)i −b >0 (9.27)
2N 2

π π π
i j = −(i 2 + j 2 ) + j +b (9.28)
4N 2 2
and for  π π
cos (N − j)i −b <0 (9.29)
2N 2

π π π
i j = −(i 2 + j 2 ) + j +b +π (9.30)
4N 2 2
where i j and ri j are the phase and intensity at output channels j for input channel
i, respectively, and b = 0 for symmetric input, whereas b = 1 for antisymmetric
input. Output channels j with odd values of i + j always have zero light intensi-
ties. Uniform as well as nonuniform power splitting is then possible. For example,
nonuniform power splitting into two output channels can have the following possible
splitting ratios: 100 : 0, 85 : 15, 72 : 28, and 50 : 50.
For overlap imaging, when yiin = k(W/2), or yiin = k(W/3), i.e., the input posi-
tion is at either the center or one- or two-thirds of the MMI width, the output intensity
is uniform. These two cases are exactly the same as the paired and symmetric-
interference cases discussed earlier. Note, however, by considering them as special
cases for general interference achieved through overlap imaging, the output-phase
relationship can be easily derived. Moreover, overlap imaging also gives the output
results for both the symmetric and antisymmetric input fields, which will be useful
for mode-filtering application.

9.2.4 Properties of MMIs

Resolution of an MMI
The spatial resolution of MMIs is determined by several factors. First, if the number
of modes is m, the spatial resolution ρ is, by simple Fourier analysis, approximately
equal to
We
ρ (9.31)
m
since the high spatial frequency component is set approximately by the width of
nodes in the highest order lateral mode. This equation shows that the higher the
mode number, for a fixed We , the better the resolution. Thus, since the higher the
index contrast, the larger the number of modes supported in the imaging region, and
high-index contrast gives high resolution. In addition, the weighing of the amplitude
9.2 Multimode Interference Devices (MMIs) 187

Fig. 9.5 The weighing of the amplitude of each of the spatial harmonics can affect the spatial
resolution of MMI

of each of the spatial harmonics, i.e., the modal amplitudes, such as shown in Fig. 9.5,
can also affect the spatial resolution.
Tolerance
If w0 is the focused image width, then by analogy with Gaussian optics (Saleh and
Teich 1991), the Rayleigh range of an image is

πn f w02
δL  (9.32)
4λ
But since w0 ∝ ρ, it follows that w0 ∼ We /m. Thus, generally, the wider the access
waveguide, the better the tolerance in length. Also, from the expression for the
characteristic image length,

L π = 4n f We2 /3λ0 (9.33)

the tolerances related to fabrication parameters can be obtained by simple differen-

tiation of (9.16),
δL δWe δλ δn f
=2   (9.34)
L We λ nf

For example, for a 2 × 2 MMI coupler with L ∼ 0.25 cm, δL ∼ 15 µm (from 9.34),
and We ∼ 8 µm, δWe ∼ 0.25 µm (Pennings et al. 1991). Thus, in this case, the
device width and its tolerance are the major fabrication challenges in MMI devices.
188 9 Imaging Devices

Fig. 9.6 The output intensities in the two output waveguides as a function of length. In this example,
a 2 × 2 MMI coupler is used and compared to a 2 × 2 directional coupler

Balance and Phase Balance in MMI Splitters

One of the most important device applications of multimode imaging devices is their
use as 1 × N splitters and, in particular, for 1 × 2 splitting for use in Mach–Zehnder
interferometers, etc. In this application, MMI performance is relatively insensitive
to dimensional errors since the power in each arm is equally affected by a length
or width error, that is, the relative power in each of the arms is balanced. A similar
behavior is true for the relative phase balance of each arm. A good comparison of
the insensitivity of MMI to dimensional tolerance can be had by comparing their
length tolerance with those of a 3 dB directional coupler. In particular, in Fig. 9.6, the
output intensities in the two output waveguides for a typical 2 × 2 3 dB MMI coupler
and a typical 2 × 2 3 dB directional coupler are plotted as a function of the length
of their coupling region (Weinert and Agrawal 1995). At a length of L 3 dB, the two
coupler intensity curves versus length intersect with each other with the maximum
but opposite slope in this plot, whereas the intensity curves for each of the MMI
arms have well-defined overlapping maxima. For 3 dB coupling applications, this
behavior makes MMI couplers more fabrication tolerant than directional couplers.
Bandwidth
Equation 9.34 shows for a wavelength deviation of δλ0 from the design wavelength λ0
causes an image to be focused at L + δL instead of L, the original imaging length.
This defocusing sets the wavelength bandwidth of the device for a given level of
acceptable loss. To determine this bandwidth, λ, we again assume Gaussian beam-
like behavior of the image, having a waist, w0 , at the focal point. The relation between
transmission loss, T , at length L due to a wavelength deviation of δλ from the design
wavelength can be obtained by calculating the overlap integral of the image spot at
9.2 Multimode Interference Devices (MMIs) 189

length L with that at L + δL (Besse et al. 1994). This calculation yields the following
equation:

T = 1 + z 2 / 1 + 5z 2 + 4z 4 (9.35)

where
2λδL
z≈ (9.36)
πnw02

Using (9.34), this expression can be recast in terms of the wavelength bandwidth for
an acceptable loss, λ.

π w2
2|λ|  Z · · a N · 02 · λ (9.37)
4 W
In (9.37), N is the number of images, a is a integer whose value is between 1 and
4, depending on the different imaging mechanisms, W is the MMI width, w0 is the
input-field width. Thus, the optical bandwidth is related to the main design parameters
through a loss factor.

9.2.5 Materials

Soldano and Pennings (1995) have discussed and investigated the important advan-
tages of MMI devices. Others, such as Bryngdahl (1973), Ulrich (1975), have shown
how relatively complex functionalities can also be obtained for certain materials. In
many ways, these advantages derive from the advantages in fabrication and material
growths that have occurred over the same period of time.
MMIs have been fabricated from most of the standard integrated optical mate-
rials. These include the materials mentioned above, i.e., InP, GaAs-based semicon-
ductors and its alloys. Due to the advantages of low propagation as the index contrast
increases, compact devices requiring small bend radii have used, see for example,
the reference Levy et al. (1998). Following this same approach, more recently, there
have been demonstrations of the MMI devices that have used SOI materials. These
devices are described in the applications section.
The interest in constructing large photonics data systems in SOI wafers has encour-
aged understanding carefully any limitations on using the SOI system. Clearly its
high-index contrast is an important advantage for reducing the scale size of optical
circuits using such material. In addition, SOI has many of the processing and pat-
terning advantages of commercial silicon, but use of SOI introduces disadvantages
as well. One of these problems is the difficulties with tight bends and associated
dispersion control. This feature will be discussed in a section below.
190 9 Imaging Devices

Specifically for silicon-on-insulator (SOI) waveguides, high refractive-index-

contrast guiding layers and the sharp bend radius (Li and Henry 1996; Okamoto
et al. 1992) enable the device size to be reduced by many magnitudes (Soldano et al.
1992; Okamoto et al. 1992; Dragone 1988). However, device design, e.g., MMI cou-
plers, is more challenging. Thus, the Si guiding layers are more prone to phase errors
due to high-index contrast.
These phase errors cause, for example, the smaller silicon AWGs to have higher
cross talk compared to those of low-index contrast materials. Careful design and
fabrication allows reduction of the phase errors and the associated cross talk to an
acceptable level (Dragone 1988).
A recent device which shows the importance of solving this problem has been
described by Pathak et al. (2013). This devices is a compact (560 × 350µm) 12-
channel 400 GHz arrayed waveguide-grating wavelength demultiplexers (AWG) in
silicon. It has a flattened spectral response due to the use of an MMI. The most
important feature is the flattened spectral response, which is a result of an optimized
mode shaper, utilizing the multimode interference (MMI) coupler as the input of
the AWG. This use of the MMI approach in SOI has led to not only small size but
in addition high performance as well. In particular, a critical feature of an AWG is
the spectral response of its channel waveguide. In a standard AWG, this response is
Gaussian-like (see next chapter). But, in fact, for many applications, a flat spectral
response is needed. In response to this need, (Deri et al. 1992) proposed using a MMI
(in fact, in this case, an InP AWG) with the AWG to fulfill this requirement.

9.2.6 Applications

In this section, various applications of MMI couplers are summarized. A brief discus-
sion on the principle, performance, and development for each class of application is
also presented. Application of MMI-type devices to wavelength routing and splitting
is discussed in Chap. 11 and the corresponding references, respectively.
MMI 3 dB Splitters
The most basic class of applications of MMI devices are splitters or couplers. This
application is important and makes use of the fact that a well-designed MMI device is
efficient and relatively simple to fabricate. The “apex,” which is a problematic area
in conventional Y -branches, is not used in MMI splitters. Planar optical couplers
were the earliest applications for multimode imaging devices (Pennings et al. 1991;
Soldano et al. 1992; Jenkins et al. 1992; Heaton et al. 1992).
In addition, 3 dB MMI couplers can be designed to be extremely small. For exam-
ple, Spiekman et al. (1994) demonstrated a miniaturized ∼ 107 µm, MMI 3 dB cou-
pler in InP using deep etching. In this 3 dB coupler, restricted interference and tapered
input waveguides were used to achieve the short length and low loss. Further length
shortening reported by Levy et al. (1998) was realized by using a tapered MMI
9.2 Multimode Interference Devices (MMIs) 191

Fig. 9.7 Two comparable 3 dB tapered MMI structures:(top) The parabolically tapered device.
(bottom) The linearly tapered device [Adapted from (Levy et al. 1998)]

design (see Fig. 9.7). This tapered 3 dB MMI coupler can be designed to be as short
as ∼ 30 µm. A tapered geometry can also be applied to 4 × 4 MMI couplers.
More Complex MMI Power Splitters/Couplers
Large N (input/output waveguide number), N × N , or center-fed 1 × N MMI split-
ters/couplers have been designed and demonstrated. For example, Heaton et al. (1992)
demonstrated a 1 × 20 power splitter (W = 120µm, L = 2374µm) with a splitting
uniformity of ±4% on GaAs/ AlGaAs, while Rasmussen et al. (1995) fabricated
and measured the performance of a 1 × 64 MMI power splitter, using a planar silica
waveguide platform. Smaller values of N have been reported for I n P-based systems;
thus, a 4 × 4 device has been reported by Pennings et al. (1993). The performance
degradation of large N devices can ultimately be shown to be due to intrinsic phase
error in all MMI devices, including conventional MMI devices with deep-etched
structures. This important limitation is more severe for N × N devices than for
center-fed 1 × N devices. A detailed discussion of this point is presented in Huang
et al. (1998a), Huang et al. (1998b).
MMI devices can also be designed so as to achieve an arbitrary output splitting
ratio; this is possible by making simple changes in the devices geometry from those
used in conventional structures. One method uses a nonuniform butterfly MMI struc-
ture in conjunction with overlap imaging (Besse et al. 1996). As shown in Fig. 9.8, by
192 9 Imaging Devices

Fig. 9.8 The geometrical design of a butterfly MMI coupler. In the design, as shown by Besse, the
self-imaging properties remain with carefully adjusting the length of each section during the design
process [Adapted from (Besse et al. 1996)] Source https://ieeexplore.ieee.org/document/541220

Fig. 9.9 The geometrical

design of a bent MMI
coupler. 0◦ < θ < 1◦ (figure
not to scale). By changing
the bending angle between
two adjacent MMI 3 dB
couplers and, hence, the
modal path length at each of
the two couplers, a
predetermined splitting ratio
at the output can be achieved

changing the taper of the multimode section, the relative phase differences for imag-
ing points between the two sections can be altered, thus changing the splitting ratio.
Also, an arbitrary splitting ratio can be realized by using both the butterfly struc-
ture and an overlapping imaging mechanism. A second approach uses a bent MMI
structure (Levy et al. 1997). By changing the bending angle between two adjacent
MMI 3 dB couplers and, hence, the modal path length at each of the two couplers, a
predetermined splitting ratio at the output can be achieved (see Fig. 9.9).
9.2 Multimode Interference Devices (MMIs) 193

Fig. 9.10 Cross section of the fabricated waveguide geometry

Generalized Mach–Zehnder Interferometer

MMI couplers have been used for the splitting of phases in the propagating waves of
Mach–Zehnder interferometers due to their modal uniformity and phase stability. For
example, a passive polarization splitter made of two MMIs has been demonstrated
by Soldano et al. (1994), yielding strong extinction over 60 nm (Fig. 9.10). Similarly,
Bachmann et al. (1993) have shown that electro-optical Mach–Zehnder switches can
be made to have high extinction ratios in III–V materials. In all of these designs, the
choice of MMI couplers played a crucial role in attaining the wide bandwidth and
polarization-independent operation of the integrated interferometer.
As will be explained in more detail in Chap. 13, these Mach–Zehnder devices,
with thermo-optical or electro-optical control of the phases in one, have been used as
switching elements. For example, ban a silicon MMI switch based on thermo-optic
control of modal interference. The device is both compact and a high-performance sil-
icon MMI switch, which uses thermo-optic control of symmetric-interference modes
by heating of the mode-peak regions. The direct heater is formed with n-i-n-i-n resis-
tors whose regions are placed at the peak regions of the first two-folded image.
Two Wavelength-Channel (980/1550 nm, 1310/1550 nm) Multi/
Demultiplexer
Since the characteristic beat length L π in (9.6) for an MMI coupler is related to
wavelength, an MMI coupler can separate two very different wavelengths, λ1 and λ2 ,
if it is a bar coupler for one wavelength and a cross coupler for the other wavelength.
The length L c of such a demultiplexing coupler has to satisfy the following relation:

L c = p(3L π ) = ( p + q)(3L π ) (9.38)

194 9 Imaging Devices

where p is an integral number and q is an odd integer. When λ1 and λ2 are well
separated, a relatively small value of p will satisfy this equation. In this case, a
very compact dual channel wavelength (de)multiplexer can then be realized using a
MMI device. Such a device has been demonstrated for 980/1550 nm, pump/signal
wavelengths for a device of Si O2 with an extinction ratio of 18 dB, an insertion loss
of 0.5 dB, and a length of 458 µm (Paiam et al. 1995). A polarization-insensitive
design on this type of wavelength (de)multiplexer has also been realized (Paiam and
MacDonald 1998). A similar design for splitting 1.3/1.55 µm has also been designed
on a silica platform (Li and Henry 1996).
Other More Complex Applications
MMI couplers have been demonstrated for a variety of other novel applications.
These include the following: a mode converter and combiner, which uses overlap
imaging to convert between the first-order and fundamental mode (Leuthold et al.
1996); an efficient integrated optical mode-width expander, which is designed based
on the magnifying, self-imaging properties of tapered multimode waveguides. MMI
mirrors have also been used to provide a low loss, uniform-splitting-ratio stability,
and compact size for a square-ring-laser diode (Kim et al. 1997). MMIs have also
been used to broaden the spectral range of the front end of a coherent receiver (Deri
et al. 1992). In this case, the wavelength-insensitive behavior of the 3 dB coupler
in combination with the compact design of the photodetectors resulted in a broad
spectral operating range for the receiver.
Specifically for silicon-on-insulator (SOI) waveguides, high refractive-index-
contrast guiding layers and the sharp bend radius (Li and Henry 1996; Okamoto
et al. 1992) enable the device size to be reduced by many magnitude (Soldano et al.
1992; Okamoto et al. 1992; Dragone 1988). However, device design, itself, e.g., in
the MMI couplers, is more challenging. Thus, the Si guiding layers are more prone
to phase errors due to their very high-index contrast. These phase errors cause, for
example, the smaller silicon AWGs to have higher cross talk compared to those of
low-index contrast materials. Careful design and fabrication allow reduction of the
phase errors and the associated cross talk to an acceptable level (Dragone 1988).
A recent device, which shows the importance of solving this problem, has been
described by Pathak et al. (2013). This device is a compact (560 × 350 µm) 12-
channel 400 GHz arrayed waveguide-grating (AWG) wavelength demultiplexer in
silicon. This most important feature is a result of an optimized mode shaper, utilizing
the multimode interference (MMI) coupler as the input of the AWG. This use of the
MMI approach in SOI has led to not only small size but in other forms of high
performances as well, such as low insertion loss, etc. in comparison to that seen
in other standard AWGs. In particular, a critical feature of an AWG is the spectral
response of its channel waveguide. In a standard AWG, this response is Gaussian-
like (see next chapter). But, in fact, for many applications a flat spectral response is
needed. In response to this need, (Deri et al. 1992) proposed using a MMI (in fact,
in this case, an InP AWG) with the AWG to fulfill this requirement.
Other approaches to deal with high lateral refractive index contrast have also
been presented. For example, nanofabrication in the lateral cladding region, e.g.,
9.2 Multimode Interference Devices (MMIs) 195

fabrication of a subwavelength grating (SWG), of a MMI has been shown to lower

index contrast. This approach reduces the mode phase error and is also a single
(etch) process. Using a periodic lateral SWG, a 2×4 MMI one group (Ortega-Monux
et al. 2011) designed and fabricated this and used it in a coherent optical receiver.
Compared to MMI with a homogenous lateral cladding, this approach increases the
receiver bandwidth from 36 to 60 nm.

9.3 Star Couplers

Star couplers evenly distribute light from any one input guide to all output guides;
thus, in an N × N device, any of N -independent input ports may couple light to
any of N output ports. The device uses guided-wave optics for the I/O ports and
free-space optics in the power-distribution region. The central problem in designing
such a device is to have the power distribution be truly independent of which of
these ports are chosen for the input/output waveguide. Notice one important point
regarding star couplers: since the device works by power division, one encounters
an automatic, intrinsic “loss” of 1/N when using the device. Any signal in an input
point is divided by 1/N . Note that this loss is not encountered in MMI devices.
Nonetheless, because of their simplicity, star couplers are extremely important in
integrated optics. Recently, high-quality integrated star couplers have been developed
for communication applications (Cao et al. 2005). These devices use coupled single-
mode waveguides and careful free-space-optics design.
A drawing of a complete star coupler is shown in Fig. 9.11. The figure shows
the wider spacing of the input ports at the edge of the coupler layout; this feature
is needed for fiber pigtailing. To reduce the waveguide spacing requires bends in
the coupler waveguides. Also note the dummy waveguides on the coupler; their
function will be discussed below. A sketch of the central region of the optical layout
used for the design of the star couplers is given in Figs. 9.13 and 9.14, along with
a graphical definition of the symbols used (Dragone 1988). Note that the input and
output waveguides are spaced by a distance, a, from each other and the input and
output arrays have angular width of a and θ, respectively. The imaging of an input
waveguide is carried out by the beam undergoing Fraunhofer diffraction, which is
equivalent to a spatial Fourier transform of the dominant mode, (u), emitted from
each input port,

1 ∞
(w) = (u)e jwu du (9.39)
2 −∞

where the wavevector, w, normalized by the interwaveguide spacing, a, required for

diffraction of the waveguide into an angle, θ
, is

πa sin θ

w= (9.40)
λ
196 9 Imaging Devices

Fig. 9.11 A drawing of a complete star coupler. The figure shows the wider spacing of the input
ports at the edge of the coupler layout; this feature is needed for fiber pigtailing [Adapted from
(Okamoto et al. 1992)]

where θ
is defined as the angle from the axis of the device. The normalized distance
u is given by
x
u≈ (9.41)
2a
where x is the distance from the center of the waveguide receiver or radiating array.
Now assuming that the mode is approximately Gaussian, then the Fourier trans-
form of such a Gaussian-distribution feature is also a Gaussian. Thus, without fur-
ther modification, such a Gaussian input source will not produce the desired uniform
“rectangular” illumination of the output waveguides, that is, one seeks a distribution
such that (w) ∼ r ect (w2wα ), where wα = (πa sin α)/λ.
To obtain this rectangular illumination distribution, it is necessary to have a near-
field distribution radiating from each of the input array wavegudies with sidelobes on
the large central lobe. The Fourier transform of this sinc-like function is a rectangular
function. The sidelobes are generated by an input-waveguide coupling light into its
neighboring guides, which then radiates into free space along with the light in the
original input guide. The waveguide coupling is achieved by fanning out the waveg-
uides so as to achieve a closer spacing over the length of the waveguide leading into
the radiating end. Because of the importance of this coupling for obtaining a uniform
output aperture, it is essential to adjust carefully the interwaveguide coupling. For
the edge input guides, dummy waveguides are inserted to make this coupling close
to that in the central part of the array (Okamoto et al. 1992).
As an example of this approach, consider a beam-propagation-method calcula-
tion of a well-designed 8 × 8 star coupler, which in this case uses an input from
9.3 Star Couplers 197

Fig. 9.12 A sketch of the

central region of the star
couplers. Two linear arrays
are separated by free-space
region. Array elements are
located on two circles

centered at O and 0 with

coordinates θ, θ specified by
angular displacements from
axis. Marginal elements are
displaced from axis by α
[Adapted from (Dragone
1988)]

only the center waveguide of the radiating array; see Fig. 9.13. Observe that the
sine radiation pattern evolves into a rectangular distribution as the light propagates
toward the output guides (Okamoto et al. 1992). As an example of one well-defined
device, Fig. 9.14 shows schematic of 144 × 144 wavelength-insensitive star coupler
fabricated in Si O2 /Si. It had an excess loss of 2 dB, with a splitting uniformity of
σ = 1.47 dB. The device had 64 input guides, a taper length of L T = 2mm, and a
radius of Rs = 2.1mm, L f = 0.1mm, θ = 0.2◦ (Okamoto et al. 1992). The waveg-
uide coupler is shown in Fig. 9.12. Its waveguides are arranged on circular arcs to
enable good matching of the waveguide supermode, which has a curved wavefront
with the output guide. Thus, each guide is approximately one focal distance from the
other side.
Understanding of these devices thus requires examination of the optical physics
of coupled waveguides. This discussion is provided in the preceding chapter on
couplers. Recall that a waveguide array has a grating-like angular distribution, such
as shown in Fig. 9.15, for infinitely narrow sources. This grating distribution can
be broken into zones, called Brillouin zones in analogy with the Bloch function
198 9 Imaging Devices

Fig. 9.13 Waveform transients of optical power obtained by Beam Propagation Method simulation

Fig. 9.14 Schematic configuration of 144 × 144 star coupler [Adapted from (Okamoto et al. 1992)]

encountered in solid-state physics. For each supermode excited, the maximum has a
different angular position within each of the Brillouin zones. In general,
λ
|a sin θm − mλ| < (9.42)
2
where θm is the angle of the zone edge. Thus, for the m = 0 zone,
9.3 Star Couplers 199

Fig. 9.15 A waveguide array has a grating-like angular distribution

Fig. 9.16 Schematic configuration of a star coupler has a curved radius of R such that Na
R ≈ θB Z

ka sin θ0 < π (9.43)

The angle subtended by this is 2θ0 ≡ θ B Z ≈ λ/a Ns , where we have now explicitly
specified the dependence on substrate effective index, Ns .
The star coupler shown in Fig. 9.16 has a curved radius of R such that
Na
≈ θB Z (9.44)
R
where N is the number of waveguides. Typically, the angular aperture of the set of
output guides is somewhat smaller than θ B Z .
The curved arrangement of the output guides matches the curved diffracted wave-
front from any input guide in the far field, thus ensuring that the 1/N -divided guided
waves have a uniform phase at the entrance to the output guides. Note that typically,
there is a small, 1 ∼ 5%, displacement of the phase center from the edge of the input
guides.
One of the most important considerations in designing such a device is, in fact, the
taper of the waveguides as they approach the free-space region. One of the reasons
that this region is important is that, first, the guides must couple together to form
the ideal radiation supermode pattern. This process makes the illumination uniform
across the receiving array. Second, the waveguides must taper properly in the process
of merging. If the tapering process is not adiabatic, higher order local modes will be
excited. These higher order modes cause radiation into the second and third Brillouin
zones, i.e., beyond the angular acceptance of the receiving array. Finally, the taper
design must be adjusted so that it is as insensitive to wavelength as possible. In one
200 9 Imaging Devices

case, the star coupler was designed to work equally well at 1.3 and 1.5µm (Okamoto
et al. 1992).

9.4 Summary

This chapter has presented two important integrated devices, which operate by modal
excitation and which are generally oriented toward the goal of imaging through peri-
odic interference of modes. Both devices can be used for power splitting, although
many other applications have been conceived and demonstrated. The first, the mul-
timode imager, uses excitation of a finite width multimode waveguide in such a way
that the excited modes form a series of useful images. The advantage of this device
is that the transfer of power from the input waveguide to the output image or images,
and hence the output can be lossless. The devices based on this approach can be
extremely small, i.e., tens of (μm)2 . The second device, the star coupler, is based
on excitation of an effectively infinite series of “free-space” modes. In this case, a
uniform far-field pattern of the modes is employed across the output waveguides to
create a uniform power splitter. This device design has intrinsic loss. However, the
simplicity and robust nature of its design makes it a commonly used approach to
power distribution in PICs.

Problems

1. Note: use the following parameters for this problem:

n f = 3.4;
λ0 = 1.55 µm.

(a) Calculate the spatial resolution possible for 20 µm wide MMI which supports
30 modes.
(b) Alternatively, if the input port is roughly 6 µm wide, how many modes are
needed for a 30 µm guide for good resolution?
(c) It is possible to make a 200 µm long MMI with a 20 µm width. What is the
needed δL for a 0.1 µm lithography patterning system?

2. What is the first focal point for a 1 × 2 MMI splitter? Use the following param-
eters:
n f = 3.4
λ0 = 1.55 µm;
ωe f f = 15 µm.
3. If 100 mW power enters a high-quality star coupler. How much power leaves on
each port?
Problems 201

4. We make an MMI using a buried-waveguide geometry. In this case, n c ≈ n s . We

know free-space wavelength λ = 1.55 µm, n s = 3.35 and n f = 3.40.
(a) Design a single TE-mode buried-waveguide structure, in which d and w are
such that they are 20% below the m = 1 cutoff, i.e., w = 0.8wcutom=1
f f , etc.
(b) Design the shortest 1 × 2 MMI using general imaging with m max = 25, with
d the same as in (a) above. Specifically find
(1) The parameter L π and width W f s , assume that We ≈ W f s .
(2) The minimum length of the imaging region needed for 1 × 2 imaging.

References

Bachmann, M., Besse, P., & Melchior, H. (1995). Overlapping-image multimode interference cou-
plers with a reduced number of self-images for uniform and nonuniform power splitting. Applied
Optics, 34(30), 6898–6910.
Bachmann, M., Besse, P. A., & Melchior, H. (1994). General self-imaging properties in n× n
multimode interference couplers including phase relations. Applied Optics, 33(18), 3905–3911.
Bachmann, M., Smit, M., Besse, P., Melchior, L., et al. (1993). Polarization-insensitive low-voltage
optical waveguide switch using InGaAsP/lnP four-port mach-zehnder interferometer. In Optical
Fiber Communication Conference, page TuH3. Washington, D.C.: Optical Society of America.
Besse, P. A., Bachmann, M., Melchior, H., Soldano, L. B., & Smit, M. K. (1994). Optical bandwidth
and fabrication tolerances of multimode interference couplers. Journal of Lightwave Technology,
12(6), 1004–1009.
Besse, P. A., Gini, E., Bachmann, M., & Melchior, H. (1996). New 2/spl times/2 and 1/spl times/3
multimode interference couplers with free selection of power splitting ratios. Journal of Lightwave
Technology, 14(10), 2286–2293.
Bryngdahl, O. (1973). Image formation using self-imaging techniques. JOSA, 63(4), 416–419.
Cao, G., Dai, L., Wang, Y., Jiang, J., Yang, H., & Zhang, F. (2005). Compact integrated star coupler
on silicon-on-insulator. IEEE Photonics Technology Letters, 17(12), 2616–2618.
Deri, R., Pennings, E., Scherer, A., Gozdz, A., Caneau, C., Andreadakis, N., et al. (1992). Ultra-
compact monolithic integration of balanced, polarization diversity photodetectors for coherent
lightwave receivers. IEEE Photonics Technology Letters, 4(11), 1238–1240.
Dragone, C. (1988). Efficient n× n star coupler based on fourier optics. Electronics Letters, 24(15),
942–944.
Dragone, C. (1990). Optimum design of a planar array of tapered waveguides. JOSA A, 7(11),
2081–2093.
Heaton, J., Jenkins, R., Wight, D., Parker, J., Birbeck, J., & Hilton, K. (1992). Novel 1-to-n way
integrated optical beam splitters using symmetric mode mixing in gaas/algaas multimode waveg-
uides. Applied Physics Letters, 61(15), 1754–1756.
Huang, J., Hu, M., Fujita, J., Scarmozzino, R., & Osgood, R. (1998a). High-performance metal-
clad multimode interference devices for low-index-contrast material systems. IEEE Photonics
Technology Letters, 10(4), 561–563.
Huang, J., Scarmozzino, R., & Osgood, R. (1998b). A new design approach to large input/output
number multimode interference couplers and its application to low-crosstalk wdm routers. IEEE
Photonics Technology Letters, 10(9), 1292–1294.
Jenkins, R., Devereux, R., & Heaton, J. (1992). Waveguide beam splitters and recombiners based
on multimode propagation phenomena. Optics Letters, 17(14), 991–993.
Kim, H., Kwon, Y., & Hong, S. (1997). Square ring laser diode with mmi coupler cavity. IEEE
Photonics Technology Letters, 9(5), 584–586.
202 9 Imaging Devices

Leuthold, J., Hess, R., Eckner, J., Besse, P., & Melchior, H. (1996). Spatial mode filters realized
with multimode interference couplers. Optics Letters, 21(11), 836–838.
Levy, D. S., Li, Y., Scarmozzino, R., & Osgood, R. (1997). A multimode interference-based variable
power splitter in gaas-algaas. IEEE Photonics Technology Letters, 9(10), 1373–1375.
Levy, D. S., Scarmozzino, R., Li, Y. M., & Osgood, R. M. (1998). A new design for ultracompact
multimode interference-based 2 × 2 couplers. IEEE Photonics Technology Letters, 10(1), 96–98.
Li, Y. P., & Henry, C. (1996). Silica-based optical integrated circuits. IEE Proceedings-
Optoelectronics, 143(5), 263–280.
Okamoto, K., Okazaki, H., Ohmori, Y., & Kato, K. (1992). Fabrication of large scale integrated-optic
n* n star couplers. IEEE Photonics Technology Letters, 4(9), 1032–1035.
Ortega-Monux, A., Zavargo-Peche, L., Maese-Novo, A., Molina-Fernández, I., Halir, R.,
Wanguemert-Perez, J., et al. (2011). High-performance multimode interference coupler in sil-
icon waveguides with subwavelength structures. IEEE Photonics Technology Letters, 23(19),
1406–1408.
Paiam, M., Janz, C., MacDonald, R., & Broughton, J. (1995). Compact planar 980/1550-nm wave-
length multi/demultiplexer based on multimode interference. IEEE Photonics Technology Letters,
7(10), 1180–1182.
Paiam, M., & MacDonald, R. (1997). Polarisation-insensitive 980/1550 nm wavelength (de) mul-
tiplexer using mmi couplers. Electronics Letters, 33(14), 1219–1220.
Paiam, M., & MacDonald, R. (1998). A 12-channel phased-array wavelength multiplexer with
multimode interference couplers. IEEE Photonics Technology Letters, 10(2), 241–243.
Pathak, S., Vanslembrouck, M., Dumon, P., Van Thourhout, D., & Bogaerts, W. (2013). Optimized
silicon awg with flattened spectral response using an mmi aperture. Journal of Lightwave Tech-
nology, 31(1), 87–93.
Pennings, E., Deri, R., Bhat, R., Hayes, T., & Andreadakis, N. (1993). Ultracompact, all-passive
optical 90 degrees-hybrid on inp using self-imaging. IEEE Photonics Technology Letters, 5(6),
701–703.
Pennings, E., Deri, R., Scherer, A., Bhat, R., Hayes, T., Andreadakis, N., et al. (1991). Ultracompact,
low-loss directional couplers on inp based on self-imaging by multimode interference. Applied
Physics Letters, 59(16), 1926–1928.
Rasmussen, T., Rasmussen, J. K., & Povlsen, J. H. (1995). Design and performance evaluation of
1-by-64 multimode interference power splitter for optical communications. Journal of Lightwave
Technology, 13(10), 2069–2074.
Saleh, B. E., & Teich, M. C. (1991). Fundamentals of photonics. Hoboken: Wiley
Smit, M. K., & Van Dam, C. (1996). Phasar-based wdm-devices: Principles, design and applications.
IEEE Journal of Selected Topics in Quantum Electronics, 2(2), 236–250.
Soldano, L., De Vreede, A., Smit, M., Verbeek, B., Metaal, E., & Green, F. (1994). Mach-zehnder
interferometer polarization splitter in ingaasp/inp. IEEE Photonics Technology Letters, 6(3),
402–405.
Soldano, L. B., & Pennings, E. C. (1995). Optical multi-mode interference devices based on self-
imaging: Principles and applications. Journal of Lightwave Technology, 13(4), 615–627.
Soldano, L. B., Veerman, F. B., Smit, M. K., Verbeek, B. H., Dubost, A. H., & Pennings, E. C.
(1992). Planar monomode optical couplers based on multimode interference effects. Journal of
Lightwave Technology, 10(12), 1843–1850.
Spiekman, L., Oei, Y., Metaal, E., Green, F., Moerman, I., & Smit, M. (1994). Extremely small
multimode interference couplers and ultrashort bends on inp by deep etching. IEEE Photonics
Technology Letters, 6(8), 1008–1010.
Ulrich, R. (1975). Image formation by phase coincidences in optical waveguides. Optics Commu-
nications, 13(3), 259–264.
Ulrich, R., & Kamiya, T. (1978). Resolution of self-images in planar optical waveguides. JOSA,
68(5), 583–592.
References 203

Weinert, C., & Agrawal, N. (1995). Three-dimensional finite difference simulation of coupling
behavior and loss in multimode interference devices. IEEE Photonics Technology Letters, 7(5),
529–531.
Zucker, J., Jones, K., Chiu, T., Tell, B., & Brown-Goebeler, K. (1992). Strained quantum wells
for polarization-independent electrooptic waveguide switches. Journal of Lightwave Technology,
10(12), 1926–1930.
Chapter 10
Diffraction Gratings

Abstract A diffraction grating is an increasingly important component in integrated

optics. They are used in integrated optics for such applications as in and out coupling
for integrated photonics chips, for on-chip multiplexing and demultiplexing, on-
chip reflecting elements, and wavelength filters. Analysis of the propagating wave is
described and carried out efficiently by use of the coupled wave equations.

10.1 Introduction

Diffraction gratings are important components in several types of devices used in

PICs (Yariv 1997) including those shown in Fig. 10.1: distributed feedback (DFB)
lasers, Bragg filters (demultiplexing), channel-dropping filters, DFB reflectors, and
reflection (spectrometer) grating demultiplexers. We will discuss these applications
in later chapters. Here, we will concentrate on presenting the fundamentals of gratings
used as general optical elements.
This chapter will first use coupled-mode theory to analyze several typical con-
figurations for gratings in integrated optical devices. We then consider computation
of the coupling coefficients for gratings having different surface-relief structures.
Finally, we will discuss the use of gratings as reflectors or feedback elements.

10.2 Collinear Coupling

When a grating is patterned on waveguides or waveguiding lasers, the dielectric

perturbation of the grating causes coupling of the waveguide modes. The “generic”
configurations for such applications—discussed in Chap. 5—are shown in Fig. 10.2.
In the upper panel, the waveguide modes of the grating coupler travel in oppo-
site directions within the waveguide. Typically, the two modes which interact most
© Springer Nature Switzerland AG 2021 205
R. Osgood jr. and X. Meng, Principles of Photonic Integrated Circuits,
Graduate Texts in Physics,
https://doi.org/10.1007/978-3-030-65193-0_10
206 10 Diffraction Gratings

Fig. 10.1 The sketches of distributed feedback (DFB) lasers (top left), Bragg filters (demulti-
plexing) (bottom left), channel-dropping filters (top right), and reflection (spectrometer) grating
demultiplexers (bottom right)

strongly have the same transverse profile. Only their propagation constants are oppo-
site in sign. In the second panel, the guided waves travel in the same direction. Here,
the propagation constants of the two modes are generally different, and the grating
enables coupling of dissimilar nodes.
As discussed in Chap. 5, the grating allows efficient phase matching between
the guided modes for both codirectional and contradirectional coupling. Thus, if the
dominant Fourier component of the grating is characterized by a spatial wavenumber
of K = 2π/ , then the phase-matching conditions for each of the two cases shown
in Fig. 10.2 are

β1 = β2 + K (10.1)

where β1,2 is the propagation constant of the 1, 2 mode, where only first-order
coupling is assumed, i.e., if the order is q then K is replaced by q K .
This phase-matching condition causes two modes to dominate the equations. In
addition, it also allows us to neglect any coupling to the radiative modes of the guide.
Recall that the degree of resonance of a grating is conveniently quantified by the
detuning constant, given by
β1 − (β2 + K )
δ= (10.2)
2
where β1 , β2 are positive if they are oriented in the same direction.
Consider now the solution to the coupled-mode equations for the two cases of
contradirectional and codirectional traveling waves.
10.2 Collinear Coupling 207

Fig. 10.2 The generic configurations for a patterned grating causing coupling of the waveguide
modes. The left panel shows the situation when the waveguide modes of the grating coupler travel
in opposite directions within the waveguide. The right panel shows the situation when the guided
waves travel in the same direction

10.2.1 Contradirectional Waves of a Surface Grating

Contradirectional coupling allows a surface grating to be used as a wavelength-

selective reflecting element for integrated optics. As mentioned above, typically in
this case β1 = −β2 . We seek the amplitudes for a1 and a2 ; recall their definition as
given in (5.26) relative a1 and a2 . The boundary conditions are that for a grating of
length L, a1 (0) = 1 at z = 0, and a2 (L) = 0 at z = L, and thus only one wave is
incident on the grating. In this case, the solutions to the coupled-mode equations are
 
βd cosh[βd (z − L)] + jδ sinh[βd (z − L)]
a1 (z) = e− jδz (10.3)
βd cosh βd L − jδ sinh βd L
 
jκ sinh[βd (z − L)]
a2 (z) =e jδz
(10.4)
βd cosh βd L − jδ sinh βd L
√
where βd = κ2 − δ 2 . These amplitudes can be manipulated to give the z-dependent
transmission and reflection coefficients of the grating
 2
a1 (z) 1 + (κ/βd )2 sinh2 [βd (z − L)]
T = = (10.5)
a1 (0) 1 + (κ/βd )2 sinh2 (βd L)
 2
a2 (z) (κ/βd )2 sinh2 [βd (z − L)]
R= = (10.6)
a2 (0) 1 + (κ/βd )2 sinh2 (βd L)

Figure 10.3 shows a plot of the reflected and transmitted power versus αL. Note
that the reflectivity at z = 0 is given by
208 10 Diffraction Gratings

Fig. 10.3 A plot of the

reflected and transmitted
power versus αL. The solid
and dashed lines show two
different situations as shown
in the legends

(κ/βd )2 sinh2 βd L
R= (10.7)
1 + (κ/βd )2 sinh2 βd L

Similarly, the transmission of the grating is given by

T =1− R (10.8)

1 − tanh2 βd L
= (10.9)
1 + (δ 2 /βd2 ) tanh2 βd L

βd2
≈4 exp(−2βd L) (10.10)
|κ|2

Returning to reflectivity, it is useful to plot this quantity versus the detuning, δ,

assuming κ is approximately constant over this range, i.e., dκ/dδ ≈ 0. This assump-
tion is reasonable for frequencies near the relatively narrow stop band. This plot is
shown in Fig. 10.4. Notice that the distribution has a flat maximum for δ < κ, and
that there are specific zero points for the distribution. These zero points are a result of
the fact that the backward wave has a set of nodal points within the grating, including
the two at either end. In effect, the grating has become a resonant cavity very much
like a lumped Fabry–Perot cavity, except in the case of the grating. The reflection is
distributed “within” the cavity. Note that as the detuning is increased, the width of the
grating resonances in Fig. 10.4 increases due to the fact that the distributed grating
reflectivity, and hence its “Q” also decreases with detuning. Notice also that outside
of the stop bands, the propagation constant is complex and not purely imaginary, as
10.2 Collinear Coupling 209

Fig. 10.4 A plot of the reflected and transmitted power versus the detuning, δ assuming κ is
approximately constant over this range

it is within the stop band. A complex propagation constant allows for the existence
of propagating modes.

10.2.2 Codirectional Coupling

For coupling between modes propagating in the same direction, such as needed for
a mode converter, β1 > 0 and β2 > 0. In this case, a longer period grating than
considered above is used to phase match the two modes, since the phase-matching
criteria are given by the difference of the two propagation vectors.

The amplitudes for a1,2 are then

   
jδ
a1 (z) = exp(− jδz) cos κ + δ z + √
2 2 sin( κ + δ z)
2 2 (10.11)
κ2 + δ 2

jδ 
a2 (z) = exp(− jδz) − √ sin( κ2 + δ 2 z) (10.12)
κ2 + δ 2

These solutions show that the equation is identical to that of a directional coupler
except that the definition of δ now includes the grating period K as shown in (10.2).
As a result, the maximum power transfer to mode 2, P2max , is given by
1
P2max = (10.13)
1 + (δ 2 /κ2 )
210 10 Diffraction Gratings

and
√
|a2 (L)|2 sin2 ( κ2 + δ 2 L)
= (10.14)
|a1 (0)|2 1 + (δ 2 /κ2 )

that is, the transfer between the two modes is periodic in length just as in a directional
coupler!

10.3 Grating Dispersion

Gratings are often used to compensate for material dispersion, such as for non-
dispersion-compensated fiber, or to compress a wide bandwidth pulse. Thus, it is
important to consider explicitly the conditions for grating dispersion.
For a Bragg grating at center frequency, ω0 ,
π
β(ω0 ) = (or K ) (10.15)

Thus, if we expand the propagation constant around this frequency, we obtain
dβ
β = β(ω0 ) + (ω − ω0 ) (10.16)
dω

but dβ/dω = νg , and thus if β(ω) − β(ω0 ) = δ

ω − ω0
δ= (10.17)
νg

A plot of a typical grating dispersion is shown in Fig. 10.5.

The presence of a grating also affects the propagation of light in a more profound
manner. This phenomenon has already been shown in the case of coupled waveguides,
where it was shown that in the case of near-synchronous waveguides, an optical or
photonic “bandgap” was excited. In the region of optical frequencies, light does
not propagate. Similarly, for the case of grating near the zero-detuning point, the
frequencies of the two normal modes of the grating are different. This behavior is
illustrated in Fig. 10.6, which shows a plot of β versus ω/c for the two normal modes,
and which are approximately the forward and backward propagating waves. The two
propagation constants are

β1,2 = ± δ 2 − |κ|2 (10.18)

or at δ = 0,

β1 (≈ β2 ) = ± jκ (10.19)
10.3 Grating Dispersion 211

Fig. 10.5 A plot of grating dispersion from (10.15) and (10.16) with unperturbed propagation
constant β(ω) proportional to ω

Fig. 10.6 A plot of β versus

ω/c for the two normal
modes when grating is near
the zero-detuning point. The
frequencies of the two
normal modes of the grating
are different
212 10 Diffraction Gratings

Fig. 10.7 A physical picture of coplanar coupling for the Bragg condition. The upper panel shows
the transmission type while the bottom panel shows the reflection type

10.4 Coplanar Coupling

In many applications, the direction of propagation of the grating is not normal to the
grating grooves. However, if the grating is within the plane, it is termed “coplanar
coupling”; a physical picture of this case for the Bragg condition is shown in Fig. 10.7.
The Bragg condition for the lightwave now must be stated in vector form, i.e.,

βd = βi + q K (10.20)

where q, an integer, is the “order” for the diffracted ray, q = 0, ±1, ±2; K is the
grating period | K | = 2π/ ; and βd , βi is the diffracted, incident wavevector with a
magnitude
2Neff π
|βd | = |βi | = , (10.21)
λ
10.4 Coplanar Coupling 213

see Fig. 10.7 for an example. In this case, if the projection of a unit vector of the
grating vector, | K |, with (10.20) is computed, i.e., the dot product, then

K   
· βb = βa + q K (10.22)
| K |

2π Neff 2π 2π
cos(90 − θ) = − Neff cos(90 − θ) + a (10.23)
λ λ 
which can be written as
λ
2 sin θ = q (10.24)
Neff

More generally, it can be shown that for the case of coplanar coupling, it is
again possible to write a set of coupled-mode equations. However, in this case, the
equations are written in terms of the amplitudes of the incident and diffracted modes
I (z), D(z). The equations are
dI (z)
cos θd = − jκ∗ D(z) exp(− j zδz ) (10.25)
dz

dD(z)
cos θi = − jκI (z) exp(+ j zδz ) (10.26)
dz

and

2δz = βd cos θd − (βi cos θi + q K cos θ) (10.27)

and κ is the coupling coefficient between the incident and diffracted waves. The
angles θi,d , θ, and θ are the angles of incident, the diffracted, and the grating grooves,
shown in Fig. 10.7.
These equations can be used to obtain useful working equations for both trans-
mission and reflection gratings. For example, in the case of a diffraction grating of
length L, the efficiency of diffraction, η, is given by

cos θi |D(L)|2
η= (10.28)
cos θd |I (0)|2


sin2 δz2 |κ|2 / cos θd cos θi )L

= (10.29)
1 + cos θd cos θi δz2 /|κ|2

or, exactly at the Bragg condition, δz = 0,

214 10 Diffraction Gratings
 
|κ|L
η|δ z = sin2 √ (10.30)
cos θd cos θi

Clearly, the efficiency increases with both κ and L for the grating. Notice also that
the grating efficiency varies as a sin x/x behavior and with detuning δz , just as seen
for collinear coupling. Although we will discuss coplanar reflection gratings, similar
expressions can be derived in that case as well (Nishihara et al. 1989).

10.4.1 Coupling Coefficients for Gratings

In Chap. 5, we examined the coupling coefficient for TE–TE coupling with a shallow
grating structure. In this section, we provide simplified approximate formulae for
other formula of grating profiles or guided-wave polarization.
Previously, it was shown that the coupling coefficient for shallow sinusoidal grat-
ings on a single slab waveguide is operating in the fundamental waveguide modes,

πa (n f − Neff )
2 2
1
κT E−T E = (10.31)
2λ Neff d + (1/γ) + (1/δ)

where d is the waveguide film thickness, γ is the substrate decay constant, δ is the
cover decay constant, a is the groove depth, and λ is the free-space wavelength.
In addition, it can be shown that this expression applies equally to the case of TM
waves on a single-mode slab waveguide. This equation may be rewritten using an
even simpler equation:
λ
Lc = (10.32)
2Neff

πNeff
κ (10.33)
λ
where
1
Neff = (Neff (max) − Neff (min)) (10.34)
2
For the case of a shallow rectangular grating in which coupling for the T E 0 − T E 0
and the T M0 − T M0 modes occurs

sin(qcπ) (n f − Neff )
2 2
2π a
κT E−T E = (10.35)
λ d + (1/γ) + (1/δ) qπ Neff

and
2π a sin(qcπ) 2 2 Neff
κT M−T M = (n f − Neff ) (10.36)
λ d + (1/γ) + (1/δ) qπ qs
10.4 Coplanar Coupling 215

Fig. 10.8 The sketch of a

shallow rectangular grating

where the grating is as shown in Fig. 10.8, q is the diffraction order, c is the fraction
of surface in one period which is not covered by a groove, a is the depth of the
groove, and
 2  2   
Neff Neff n2 1 1
qs = + −1 / c + 4
nf ns 2 n 4c nf

As an example of coupling coefficients in actual devices, consider a typical value

of κ for distributed feedback in single- and double-heterostructure diode lasers.
In the case of GaAs, for example, for λ = 1.53 µm, A = 500 nm, T = 1.3 µm,
and Neff (λ = 1.3 µm) = 3.348, we obtain κ = 24.96 cm −1 from (10.31), and κ =
25.11 cm−1 from (10.32). Alternatively, in the case of a grating at an air interface
with Ga As, κ typically is between 50 and 100 cm−1 .

10.5 High-Q Bragg Grating Structures

A simple Bragg grating has a “stop band” in the vicinity of its Bragg wavelength. That
is, as shown in Fig. 10.9, on a linear plot of reflectivity versus wavelength or frequency
detuning from its Bragg value, a Bragg grating acts like a high reflectivity mirror.
In fact, Bragg mirrors are the most satisfactory method of making high reflectivity
structures on a planar surface. They can be made by lithographic means and can
incorporate a variety of useful optical features including wavelength chirping or
apodization and wave front curvature as was just mentioned above. In addition, it is
also possible to introduce a small change in this grating structure to make a high-Q
narrow passband transmission filter using Bragg gratings. This latter structure has a
very high quality factor, Q, and thus is an extremely narrow bandpass filter with low
loss.
216 10 Diffraction Gratings

Fig. 10.9 A linear plot of reflectivity versus wavelength or frequency detuning from its Bragg
value. Notice that the Bragg grating acts like a high reflectivity mirror

Ultimately, this filter will be found to be equivalent to the Fabry–Perot filter

encountered in bulk optics, except that in this case, the two mirrors are replaced
by distributed Bragg reflectors. Thus, by analogy with the Fabry–Perot device, it
would be expected that at certain resonances, the cavity formed by the two mirrors
would have a transmission maximum. In fact, as was pointed out above, such maxima
exist even in a simple Bragg resonator; note, however, that for a uniform diffraction
grating, the transmission resonances are not at δ = 0, i.e., the Bragg frequency of
the grating.
In order to understand why a resonance does not lie at the center of the stop band,
consider making such a two-grating mirror filter as short as possible by placing the
two mirrors adjacent to each other with only an infinitesimal spacing. In addition,
for simplicity, allow each grating to be of length L  κ−1 , or effectively infinite in
length. The center of the separated grating is at z = 0.
Now consider excitation of the infinite grating lying in the negative region at a
frequency such that δ κ and z = 0. Since the grating is infinite, the solutions for
excitation at center are given by a special case of the earlier more general grating
problem, which is obtained from the coupled-mode equations. For the grating enve-
lope amplitudes A1 (z), where as defined earlier in (5.43), a1 (z) = A1 (z) exp(− j K z).
Then,

A1 = A−
1 (0)e
+|κ|z
(10.37)

|κ| −
A2 = A (0)e|κ|z (10.38)
κ 1
Similarly, execution of the positive half plane z > 0, again at z = 0, yields
10.5 High-Q Bragg Grating Structures 217

Fig. 10.10 The sketch of two gratings spaced by a quarter-wave section, /2

A1 = A+
1 (0)e
−|κ|z
(10.39)

|κ| +
A2 = − A (0)e−|κ|z (10.40)
κ 1
By continuity, we expect the two solutions to match at z = 0. However, if the reflectiv-
ity or A2 /A1 is examined for the two regions, it is clear that the reflectivity polarities
differ: one is negative, one is positive. Clearly this is not the case for the two half
infinite gratings; in effect, the fields at the boundary are such that their coupled reflec-
tivity, r , is 180◦ out of phase. Thus, without a change in the structure, a standing
wave solution is not possible.
Consider now separating the two gratings by a small spacing. In particular, if a
quarter-wave section, i.e., /2 in length, is inserted into the structure (see Fig. 10.10),
the two reflected waves will then be matched in phase. The origin of this phase shift
can be seen by now including the rapidly varying part of the solution, exp(− j K z),
which is not included in the envelope terms, A1 and A2 , of the solution, see Chap. 5. If
an extra grating length of a z = /2 is included in the interface, the relative phase of
the two outward moving waves will change and cause the two reflection coefficients
to be in phase. Note that this term is not important if there is no “phase-slip” at z = 0
because it is identically null under that condition.
A plot of the envelope of the forward-moving wave is plotted in Fig. 10.11, again
for the case of δ κ. The envelope function given in Fig. 10.11 for δ κ shows
that most of the modal energy is stored in the field near the center of the grating. In
fact, at the edges of a grating with L κ  1, the field has decreased exponentially,
∼ e−|κ|L . As a result of this decrease, very little power “leaks” out of the grating as
it circulates back and forth near the grating center, that is, the grating structure has a
very high Q.
In fact, a coupled-mode analysis of the problem (see Haus et al. 1989) shows that
the modal amplitude at the grating edge, z = L, can be written as

A1 (L) = 2 A+
1 (0)e
−|κ|L
(10.41)

where again the solution is for optical frequencies at the center of the stop band.
Hence, the power lost out of the grating end is
218 10 Diffraction Gratings

Fig. 10.11 The envelop of the forward-moving wave for the case of δ κ

P2 = |A+ 2 −2|κ|L
1 (0)| e (10.42)

The energy stored within the grating, u, is

 ∞
|a1 |2 + |a2 |2 2|A+ |2
u≈ dz = (10.43)
∞ νg νg |κ|

where the solution is for optical frequencies at the center of the stop band. Thus, the
number of periods required for the stored energy to be lost from the cavity, or the
cavity Q esc due to this loss, is
πc +2|κ|L
Q esc = e (10.44)
νg λ|κ|

Clearly, also, by symmetry, identical power loss occurs through the other end of the
grating. This is equivalent to a decay time
νg |κ| +|κ|L
(τ )−1 = e (10.45)
2
for the cavity due to its finite length, where the fact that there are two grating ends is
included in τ . Notice that a longer L and larger κ decrease the decay time or increase
the cavity Q, where internal losses are assumed to be negligible.
The reflectivity of such a quarter-wave grating of length 2L for a wave moving
from the left can be calculated by employing a more powerful matrix approach
10.5 High-Q Bragg Grating Structures 219

Fig. 10.12 A plot of reflectivity versus wavelength or frequency detuning from its Bragg value of
a 2L-long quarter-wave grating

(McCall and Platzman 1985). The result of the calculation gives the reflectivity of
a wave entering the grating from z = −L to z = L as

| 4κδ
βd
|2 sinh4 βd L
R= (10.46)
[( βκd )4 − δ2
2
βd
2 cosh 2βd L]2 + δ 2 sinh2 2βd L

This reflectivity is plotted in Fig. 10.12. Notice that this grating provides a very
narrow passband at δ = 0. The narrowness of the passband is attributable to the high
reflectivity of the grating at δ = 0 and is thus related to the cavity lifetime τ .
The width of this passband can be found using the expression for reflectivity or
simply using the expression for Q derived above.

ω1/2 = 4νg |κ|e−2|κ|L (10.47)

or, using the normalized units discussed earlier,

δ1/2 = 4|κ|e−2|κ|L (10.48)

This passband is also shown in the reflectivity curve of Fig. 10.12. The 1/4λ-step
Bragg grating described here is at the heart of many Bragg laser optical cavities.
Its narrow passband allows light to leak out at specific wavelengths. Tuning such a
grating either by fabrication or with temperature allows on to make multi-wavelength
sources on the laser chip. In addition, Chap. ?? describes the use of these structures
in a very narrowband channel-dropping filter.
220 10 Diffraction Gratings

10.6 Applications of Bragg Gratings

A Bragg grating is a very basic functional component for integrated optical cir-
cuits. Bragg gratings have found wide applicability in filters, as wavelength selective
reflectors, dispersive elements, and in more complex devices. In this section, we
discuss a few representative devices made with Bragg reflectors; other examples are
distributed throughout the text.
Bragg gratings can be used to achieve simple optical functionality for manipu-
lating two-dimensional guided waves, such as lensing. These devices are desirable
because their fabrication, while exacting, is not as complex as that required to fab-
ricate a geodesic lens. In one example, focusing Bragg gratings were made using
electron-beam holography to have an efficiency of 40% (Nishihara et al. 1989). In
another example, a series of two Bragg gratings was used to make a two wavelength
demultiplexer in a As2 S3 , SiO2 PIC. In this device, 70% grating efficiencies were
obtained with −15 dB suppression of cross talk. The device used two tilted gratings
to make wavelength-selective beam splitters which focused light into two detectors
(Suhara et al. 1982).
More recently, Bragg grating structures have been used for polymer integrated
optical devices (Eldada and Shacklette 2000). In this case, the gratings are made by
printing through a lithographic mask to cross-link the polymer. The simplest device
used a planar Bragg grating, which was tuned using the thermo-optic effect to make
a wavelength-variable optical filter, 2 cm in length. By using care in designing the
structure, the wavelength tunability around 1.5 µm was ∼0.04 nm/◦ C. The device
had a bandwidth utilization of 0.92 for 75 GHz channel spacing. A second device
printed such a grating across the arms of a Mach–Zehnder containing either two 3 dB
couplers or two MMI couplers to form a tunable add/drop wavelength demultiplexer.
The device was 4 cm-long, with a 6 µm-core and 10 µm-thick cladding layer, with a
n = 0.5% in n = 1.5 polymer. The device output/input parts were spaced 250 µm
apart. The functionality of the devices is shown in Fig. 10.13. The insertion loss for
each device was 2.5 dB. These devices have been integrated to form a large-scale
switch array, see Chap. ??.
A simple wavelength-dependent reflecting filter can be made using the Bragg
reflectors, such as that described in this chapter; however, an improved version can
be made in conjunction with the use of a Mach–Zehnder interferometer. A device
based on this principle is shown in Fig. 10.14. The device uses two 3 dB couplers in
conjunction with a Mach–Zehnder interferometer. As a result, two passes through
the coupler causes the power to exit fully through the opposite output port. When a
DFB grating is inserted in the device, the reflected light is coupled to port 2, while
transmitted (out of band) light is switched to port 4. Since a wavelength can be added
on port 3, the device functions as an add/drop device. An integrated form of this
device has been made, but its performance was limited by polarization splitting of
the peak wavelength.
10.7 Bragg Diffraction and Raman–Nath Diffraction 221

Fig. 10.13 A sketch of the device printing a grating across the arms of a Mach–Zehnder containing
two 3 dB couplers or two MMI couplers to form a tunable add/drop wavelength demultiplexer

Fig. 10.14 A wavelength-dependent reflecting filter based on the Bragg reflectors [Adapted from
(Eldada and Shacklette 2000)]

10.7 Bragg Diffraction and Raman–Nath Diffraction

Thus far, in this chapter, we have only considered gratings which have a relatively
well-defined spatial periodicity in the z-direction. In this case, the grating wave
number is established over a long length. In general, coupling from gratings can be
characterized by a Q-parameter , defined as follows:

K2L
Q= (10.49)
β
222 10 Diffraction Gratings

Fig. 10.15 An example of the geometry of Raman–Nath diffraction for a transmission grating near
a normal angle of incidence

where K = | K | is the magnitude grating wavevector and L is the length of the

grating. Gratings in which Q  1, such as one with a long interaction length L,
are termed Bragg gratings; these are gratings we have considered thus far in this
chapter. However, it is also possible to have gratings such that the periodicity in the
direction of propagation is not well established, and Q ∼ 1. This form of diffraction
phenomenon is seen most commonly in the case of ultrasonic devices. But it is
also closely related to that seen for the diffraction phenomena in 2D spectrometers.
Diffraction with Q 1 is termed Raman–Nath diffraction.
Thus, a major difference between Bragg diffraction and Raman–Nath diffraction is
that for the Raman–Nath case, diffraction occurs in many angular diffraction orders,
determined by phase matching in the lateral direction, while for the Bragg case,
diffraction occurs in a few orders, as determined by phase matching in the direction
of propagation. The physical criterion differentiating the two cases is the relative
length of the grating region.
An example of the geometry of Raman–Nath diffraction for a transmission grating
is shown in Fig. 10.15 for near-normal angle incidence. For Raman–Nath diffraction,
the phase-matching condition is required

β sin θi + q K = β sin θq (10.50)

where is the angle of incidence relative to the z-direction, and q is the diffraction
order, i.e., q = 0, ±1, ±2, etc.
In the case of Raman–Nath diffraction, it is also possible to analyze the grat-
ing fields using the coupled-mode equations. In this case, however, the coupling is
between the same mode in different angular diffraction orders, say q → q + 1. In
the case of Q ∼ 1, diffraction occurs in many orders. The solution at the exit of the
10.7 Bragg Diffraction and Raman–Nath Diffraction 223

Fig. 10.16 Diffraction efficiency of Raman–Nath diffraction

grating region, i.e., z = L, is then found to be

 
sin(Qα/2)
Il = |φ|2 = Jl2 2κL (10.51)
Qα/2

where Jl is the ordinary Bessel function of order l. At normal incidence, θi = 0(α =

0), then ηq = Jq2 (2κL), see Fig. 10.16. Because Raman–Nath theory is not typically
used in integrated optics, except in the subfield of integrated acousto-optics, we shall
not pursue a longer discussion of it in this text. However, a more extensive discussion
can be found in Chap. 4 in Nishihara et al. (1989).

10.8 Wavelength Selecting via Reection from a Grating and

Grating Fabrication

Grating “spectrometers,” which use Raman–Nath diffraction, are often used in PICs
for wavelength selection. They operate in a manner which is very similar to that of a
large-scale commercial grating spectrometer, with the central differences being the
two-dimensional geometry and small size. A top view of a slab-waveguide spec-
trometer was shown in the fourth panel of Fig. 10.1. In this case, the spectrometer is
fed by a single-mode waveguide, which is allowed to propagate in free space before
impinging on the grating. Definition of the grating grooves is usually done via etching
and is a major fabrication challenge.
Although the grating can be analyzed formally by Raman–Nath coupled-mode
theory, it is readily analyzed here based on a simple diffraction method. That is, it
can be easily shown via a simple phase-front argument that the diffraction angle of
224 10 Diffraction Gratings

Fig. 10.17 A sketch of a

slab-waveguide
spectrometer. The incident
wave is at angle θi and the
diffracted wave is at angle
θ D . λ is the grating spatial
wavelength

the qth order. If an incident wave is at angle θi and the diffracted wave is at angle θd ,
then

q aλ
sin θd = sin θi + (10.52)
 Neff

where  is the grating spatial wavelength, see Fig. 10.17. The wave diffracted from
a reflection grating then has an amplitude pattern

sin2 [Neff Mπ
λ
(sin θd − sin θi )]
I (θ) = (10.53)
sin2 [Neff πλ (sin θd − sin θi )]

where M is the total number of grooves which are illuminated, and is thus dependent
on the cross section of the feeder waveguide end facet and its spacing from the
grating. Notice that this angular distribution is, thus, that of a sin N x/ sin x function,
which peaks at x = 0 and oscillates in the “wings.” The principal maximum occurs
at θ = θd . The intensity of the diffracted light falls off to zero when θ is increased
from θd by a value δθ such that
M| pi
Neff (sin θd − sin(θd + δθ)) = π (10.54)
λ
This spectrometer has an angular resolution of
λ
δθ = (10.55)
Neff M cos θd
10.8 Wavelength Selecting via Reection from a Grating and Grating Fabrication 225

Fig. 10.18 A sketch of a blazed grating. The incident wave is at angle θi and the diffracted wave
is at angle (θ B + θi ) where θ B is the blaze angle

Two spectral lines begin to be resolved when the maximum of one wavelength coin-
cides with the first zero of the other:

θd (λ1 + δλ) − θd (λ1 ) = δθ (10.56)

which yields
mδλ λ
q = q (10.57)
Neff  cos θd Neff M cos θd

or
δλ 1
= (10.58)
λ mM

The ratio δλ/λ is termed the chromatic resolving power of the grating. To
give a concrete example, consider illuminating a 100 µm region of a grating
with  = l µm and M = 100. For λ = 1.5 µm, and the m = 1 diffraction order,
δλ = 1500 nm/100 = 15 nm.
Typically, a high-efficiency grating may have its grooves contoured such that
the shape enhances a particular order while suppressing other diffraction orders.
Such a grating is called a “blazed” grating, and can be accomplished by a variety
of fabrication methods, see Sect. 10.8. To find the condition for the best blaze, for a
particular order, at a particular wavelength, the diffraction angle should be equal to
that corresponding to specular reflection from blaze surface (see Fig. 10.18):
226 10 Diffraction Gratings

q
sin θd = sin(θi + 2θb ) (10.59)

aλ
sin(θi + 2θb ) = sin θi + (10.60)
Neff 

Thus, for example, if θi = 0, λ = 1.55 µm and for  = 1 µm, then for a thick
multimode waveguide in Ga As, in which n f ≈ n Ga As , θb = 13.6◦ .
q=1

Because grating spectrometers entail a considerable amount of free-space imag-

ing, it is crucial to design the optical paths such that diffraction is minimized and
aberration is avoided. In addition, the use of curved surfaces allows for focusing
of free-space beams. An extensive discussion of these points is described in März
(1995). In general, however, typically the familiar Rowland mounting scheme is used
since it is a simple design and since it minimizes aberrations.
Grating spectrometers have been used for a number of PIC applications, including
optical chemical sensors and wavelength demultiplexing. The latter application was
of particular importance in early WDM systems. Multi-wavelength grating WDM
demultiplexers were demonstrated in I n P and SiO2 . These devices are illustrated
here by an SiO2 device (Tong 1998). This device was designed for two-dimensional
“free-space” manipulation of the light. The spectrometer used a Rowland layout
consisting of a large curved grating focused on a smaller curved region with a detector
spacing of 140 µm for having detection at each wavelength channel. The receiver
detected 32 separate channels. A primary concern in fabricating the grating was the
versatility and precision of the etched grating groove, 16 µm wide by 20 µm deep.
The difficulty in achieving the desired profile caused, in part, the 6–7 dB loss in the
device. In addition, residual stress in the device caused 0.5 × 10−4 birefringence
in the TE and TM modes. Despite these difficulties, the device was useful in a
commercial network system.
Fabrication is a major consideration in using gratings as optical elements in PICs.
Because of their size and dimensional tolerances, as well as the “analog” nature of
grating spatial profiles, fabrication can be exacting, and hence costly. Two different
methods of patterning are used in practice: lithographic printing of an electron-
beam-written master mask and direct laser holography. These patterns must then be
transferred by an etching step.
In-plane gratings are made using either of the above two methods (Nishihara et al.
1989). For example, for Bragg gratings, the period structure is often fabricated by
interfering two coherent laser beams, generated with a beam splitter, at the surface
of a photoresist-covered wafer. The field pattern of interfering waves, at wavelength
λ on a sample surface shown in Fig. 10.19, requires that the wavevector along the
surface, k x , is given by the angle of the two beams, θ,
2π
kx = sin θ (10.61)
λ
10.8 Wavelength Selecting via Reection from a Grating and Grating Fabrication 227

Fig. 10.19 A sketch showing fabrication of an in-plane grating. Two coherent laser beams interfere
at the surface of a photoresist-covered wafer. The field pattern of interfering waves can thus be printed
on the sample surface

Since the total field is proportional to the standing wave created, i.e., 2 cos k x x then
I ∼ cos k x x where  is the distance between two maxima, and
λ
= (10.62)
2 sin θ
WDM components have often been made from a grating fabricated on the side
of an etched feature (see the fourth panel in Fig. 10.1). In this case, the grating is
typically made in photoresisting using an e-beam-lithographic master, followed by
pattern transfer via dry etching. The controlling issue in making such a grating is the
quality of the etched features (see the discussions in the previous subsection).

10.9 Gratings for Output or Input Coupling with

Guided-Wave Structures

A recurring problem in integrated optics is the coupling of light into a waveguide.

One class of solutions simply couples in by “butting” a fiber guide against a waveg-
uide facet. This approach, which is discussed in Chap. 7, is important and successful
in many applications. However, in some cases such as that of very high-index waveg-
uides or rapid testing, the butt coupling method is not satisfactory. Instead, gratings
are more readily used to couple light into and out of the waveguide. Because of
their importance, grating couplers have received a considerable amount of emphasis
(Tamir and Peng 1977; Tamir 1975).
228 10 Diffraction Gratings

Consider a grating having a regular surface structure such as that depicted in

Fig. 10.20 or earlier in Fig. 10.8. This structure gives rise to a spatially varying index
or dielectric function, which can be expanded in a Fourier series:

(x, z) = q (x) exp(− jq K z) (10.63)
q

where q is the order of a harmonic. Each of these harmonic q components modu-
lates the guided wave and gives rise to a z-dependent optical mode with a propagation
constant given by
βq = Neff k + q K (10.64)

where Neff is the effective index of waveguide without the grating and where we
assume a shallow surface grating.
These harmonic optical modes may be guided or radiative modes depending on
their propagation constants. Thus, the qth optical mode will radiate when its propa-
gation constant, βq , is such that βq < n c k < n s k. Significant amounts of power will
be lost radiating to the gth mode if phase matching occurs in the z-direction. To have
phase matching for radiation in the cover or substrate requires that

n c,s k sin θqc,s = Neff k + q K (10.65)

Obviously, depending on the relative values of, n c,s , and Neff , several values of the
radiation angles θqc,s may be allowed. Notice also that in our previous discussion of
coupled-mode behavior in grating, we assumed that such radiative coupling is non-
existent or small. Notice, further, that the magnitude of this coupling depends on the
presence of special spatial harmonics in the surface dielectric of the Bragg grating.
The values of these harmonics vary with the profile and the depth of the grating and
are not important in, for example, a sinusoidal grating.
For a given grating period, several different radiative beam orders can exist simul-
taneously. For example, Fig. 10.20 depicts a grating for which two beams in the
q = −1 order couple with the propagating mode through one grating vector. In this
case, both a radiative (i.e., in air) and a substrate beam are present; in addition, more
complex multibeam coupling can be present.
The coupling of light into different radiative modes is important for two reasons.
First, radiative modes can obviously be a significant loss mechanism in a grating
designed for another purpose, say, as a reflector. In addition, radiation from gratings
is directly useful as a way of coupling light out of a photonic integrated circuit into
a free-space interconnect beam. Out coupling is best understood by using coupled-
mode theory, that is, using the same approach as discussed earlier in the chapter for
coupling between propagating modes.
The full coupled-mode derivation will not be discussed in this text. Basically, the
approach we do use is, however, very similar to that discussed earlier and includes a
phase term, which requires phase matching, and a coupling coefficient, κq , indexed to
a particular order, q, and to the modes having phase matching. These equations lead
10.9 Gratings for Output or Input Coupling with Guided-Wave Structures 229

Fig. 10.20 A sketch of a

grating for which two beams
in the q = 1 order couple
with the propagating mode
through one grating vector

to the following equation for the amplitude of the propagating mode in the grating
region, i.e., 0 < z < L:
da(z)
= −αa(z) (10.66)
dz

where α is a loss constant given by α = π|κτ |2 , and where the coupling coefficient is
summed over the order, q, and the two possible beam directions: cover or substrate.
Thus, as the propagating mode passes through the grating, it decays exponentially
along distance z:
a(z) = a(0) exp(−αz) (10.67)

Similarly, the total radiative power decays at the same rate. The power in the propa-
gating mode decays at twice this rate:

P(z) = P(0) exp(−2αz) (10.68)

230 10 Diffraction Gratings

Fig. 10.21 A sketch of an example using gratings to out-couple the beam from thin silicon waveg-
uide

Thus, α is a critical parameter in designing the coupler.

Input couplers are also extremely important for a surface grating, since they enable
coupling between an external beam, which is particularly useful for probing and
the optical circuity. Input coupling can be handled using a similar coupled-mode
approach. However, another approach has also been used, which is based on the appli-
cability of time reversal to problems in electromagnetics and optics. This approach
shows that the efficiency of output coupling can be written as

Pguided | τ (z)i(z)dz|2
η= = Pqi  2  (10.69)
Pinput τ (z)dz i 2 (z)dz

where τ (z) is the spatial distribution of a beam radiating out of the grating from
guided mode incident on the grating and i(z) is the actual profile of the beam being
coupled with the grating coupler. Note that this equation assumes that the coupler
has a power distribution of Pqi for each of its output beams, and summation over q
and i is understood.
The efficiency of the coupler clearly depends on the nature of the beam overlap.
Clearly the best overlap is achieved when the excitation beams have spatial distri-
butions, which are identical to the corresponding output beam for that mode in the
coupler. For a one beam coupler, all input power is in one beam and order and thus
Pqi = 1. If this input beam has a Gaussian distribution and the grating is long enough
to complete the power transfer ηmax ≈ 80%. In practice, the grating length, L, is
determined by the requirements that L  1/ατ . A thorough discussion of the prac-
tical aspects of input coupling, including beam shaping, experimental setup, etc. is
given by Nishihara et al. (1989).
Grating coupling is an important technology in many different PIC materials and
device designs. An excellent example of grating coupling is that used for silicon-
on-insulator devices by Ang et al. (2000). In this example, the high-index and thin
vertical dimension of the silicon waveguide made grating coupling the best choice
for out coupling of the beam. The scheme used the q = −1 order with a substrate
10.9 Gratings for Output or Input Coupling with Guided-Wave Structures 231

Fig. 10.22 A plot of theoretical perturbation output efficiency curve versus measured output effi-
ciency data points with various grating heights at Si film thickness of 1 µm

and cover beam; the cover beam was the desired beam. A picture of the device (Ang
et al. 2000) is shown in Fig. 10.21. The grating spacing, d, for phase matching is
given by
λ
d= (10.70)
Neff − sin φ−1

In the paper, d = 0.4 µm for a Si thickness of ∼1.1 µm. In this coupler, η ≡

Pc /(Pc + Pg ), and by careful adjustment of the height of the rectangular grating,
i.e., 0.15 µm, a maximum efficiency of 70% was obtained; see also an experimental
measurement of efficiency versus groove depth in Fig. 10.22. This adjustment was
necessary since a change of 0.15–0.22 µm caused η to drop to 30%.

10.10 Summary

Diffraction gratings are widely used passive components in the design of integrated
optical circuits. While they are “costly” to make, their wavelength selectivity makes
them attractive in many multi-wavelength devices. Further, because gratings can
manipulate the phase front of guided light wavelengths, they provide a convenient
approach to make two-dimensional optical components such as lenses or reflectors.
Without gratings, these devices would have to rely on components with very chal-
lenging fabrication procedures. A very important and a closely related use is that
of coupling light beams from one medium to another, from free spaces to guiding
medium, and from one guided medium to another. The wavelength selectivity of
232 10 Diffraction Gratings

grating has made them important components in separating and combining multiple
wavelengths. In this application, they are alternatives to the devices to be discussed
in the next chapter on wavelength filters.

Problems

1. Calculate reflectivity R versus detuning δ for a grating with κ = 200 cm−1 and
κL = 1/3. Plot R with δ from 0 to 30 κ.
2. Calculate the angle φc in Fig. 10.21 from the text, for λ = 1.3µm and grating
spacing d = 0.4µm.
3. Sketch what an integrated grating reflector looks like and describe the function
of it. Point out the main relation between the geometry and the performance.

References

Ang, T., Reed, G., Vonsovici, A., Evans, A., Routley, P., & Josey, M. (2000). Effects of grating
heights on highly efficient unibond soi waveguide grating couplers. IEEE Photonics Technology
Letters, 12(1), 59–61.
Eldada, L., & Shacklette, L. W. (2000). Advances in polymer integrated optics. IEEE Journal of
Selected Topics in Quantum Electronics, 6(1), 54–68.
Haus, H. A., Huang, W.-P., & Snyder, A. W. (1989). Coupled-mode formulations. Optics Letters,
14(21), 1222–1224.
März, R. (1995). Integrated optics: design and modeling. Artech House on Demand.
McCall, S., & Platzman, P. (1985). An optimized π/2 distributed feedback laser. IEEE Journal of
Quantum Electronics, 21(12), 1899–1904.
Nishihara, H., Haruna, M., & Suhara, T. (1989). Optical integrated circuits (pp. 7–8). New York.
Suhara, T., Handa, Y., Nishihara, H., & Koyama, J. (1982). Monolithic integrated microgratings
and photodiodes for wavelength demultiplexing. Applied Physics Letters, 40(2), 120–122.
TTamir, T. (1975). Beam and waveguide couplers. Integrated Optics (pp. 83–137). Berlin: Springer.
Tamir, T., & Peng, S.-T. (1977). Analysis and design of grating couplers. Applied Physics, 14(3),
235–254.
Tong, F. (1998). Multiwavelength receivers for wdm systems. IEEE Communications Magazine,
36(12), 42–49.
Yariv, A. (1997). Optical electronics in modern communications, oxford series in electrical and
computer engineering.
Chapter 11
Wavelength Filtering and Manipulation

Abstract The use of multiple wavelengths has many applications in integrated

optics. For example, these beams can be used for high data rate optical systems or for
contoured wavelength beams in a nonuniform amplifier. In this chapter, we consider
the functionality of wavelength filtering and manipulation. These are illustrated with
several approaches for filtering, separation, or manipulation of wavelengths using
different device types. More generally there are many wavelength filters, which
are widely used, such as phased-array waveguide routers, Mach–Zehnder interfer-
ometric devices, various asymmetric coupler-based devices, and free-space grating
demultiplexers. This chapter examines in detail three device types. One of these,
phased-array routers, are currently the most important commercial devices for wave-
length selection. However, the other two device types (coupler filters and 1/4 Bragg
channel-dropping filters) have also each been the subject of extensive investigation
of their properties. Each of these three device types illustrates one of the major device
techniques, which we will be presenting later in this chapter.

11.1 Introduction

Use of multiple-wavelength light beams allows an important extended capability

for designing integrated optical systems. This capability allows spatial or angular
separation of the light beam into its well-defined spectral components. For instance,
very coarse spectral division, such as used in a 1.3 µm/1.5 µm transceiver (men-
tioned earlier in Chap. 9), can be used to separate transmission from detection in
transceivers. This separation prevents signal “swamping” in an application where
large variation in signal intensities is present. In other instances, e.g., WDM com-
munication, multiple-wavelength signals allow expansion of the effective system
bandwidth or permit sophisticated signal routing.
In this chapter, we will discuss the important closely related functionality: wave-
length filtering and manipulation. Our short introductory section will be followed

© Springer Nature Switzerland AG 2021 233

R. Osgood jr. and X. Meng, Principles of Photonic Integrated Circuits,
Graduate Texts in Physics,
https://doi.org/10.1007/978-3-030-65193-0_11
234 11 Wavelength Filtering and Manipulation

by extensive illustrations of three different approaches for filtering, separating, or

manipulating wavelengths using specific device types.
More generally, wavelength filters, which are widely used, include phased-
array waveguide routers, Mach–Zehnder interferometric devices, various asymmetric
coupler-based devices, and free-space grating demultiplexers. The latter category is
discussed in a previous chapter on diffraction gratings and will not be discussed fur-
ther here. Of the remaining three device types, phased-array routers are currently the
most important commercial devices for wavelength selection. However, the other two
device types (coupler filters and 1/4λ Bragg channel-dropping filters) have also each
been the subject of extensive investigation of their properties. Each of these three
device types illustrates one of the major device techniques, which we will present
later in this chapter.

11.2 General Classes of Wavelength Filtering and

Manipulation

Selecting, separating, or interchanging wavelengths can be accomplished by a wide

variety of devices. However, it is possible to separate these operations into several
general classes (see Fig. 11.1). For example, in the case of wavelength filtering,
there are two very broad classes: in-line and channel-dropping filters. In-line
filters simply pass a single wavelength or wavelength band by means of reflection
or transmission from the device. Channel-dropping filters pick one wavelength, or
channel, out of a multi-wavelength stream and drop it into a different transmission
line; channel addition is the reverse process.
More complex wavelength manipulation is also often needed. For example, mul-
tiplexing or demultiplexing of multiple wavelengths to or from a bus line is common
in WDM applications. These operators can be distinguished from the functionality
used in a star coupler which simply separates a signal into many bus lines irrespec-
tive of wavelength. Alternately, wavelength reshuffling or routing is required, which
resembles in part mux/demux.

Fig. 11.1 General classes of

wavelength filtering and in-line
manipulation
Filtering
channel
Opcal Mulplex/dem
Manipulaon ulplex

reshuffling various
roung architectures
11.3 Asymmetric Coupler Filters 235

11.3 Asymmetric Coupler Filters

Many integrated coupler-based filters use asymmetric couplers as the basic functional
element (Venghaus et al. 1992). These devices have been realized with a wide variety
of designs. These include devices based on two nonsynchronous waveguides: on
symmetric waveguides with dispersive delay lines, usually Mach–Zehnder-like in
physical appearance and on grating-enabled copropagating couplers, examples of
each of these devices will be discussed here.
The essential idea of these devices is best described with a simple grating-mediated
coupler. In this case, two asymmetric waveguides with different propagation con-
stants β1 and β2 are coupled to permit power transfer in the copropagating direction.
The coupling is accomplished by using a grating on or near one of the waveguides
so as to “phase match” the power transfer. The spatial periodicity of the grating,
denoted by K , must be such that

β1 − β2 = K (11.1)

and the grating length must be such that it is one coupling length L c ,
π
Lc = (11.2)
2κ
where κ is the overall coupling constant between the two waveguides. This device
acts as a wavelength filter because both waveguides are dispersive: β1 (λ), β2 (λ).
Thus, the condition in (11.1) is exactly satisfied only at one resonant wavelength;
in actuality, of course, the relation holds over a narrow frequency band because the
finite length of the grating broadens the spectral response. This basic device can be
enhanced by fabricating it of electro-optical or semiconducting material (Alferness
et al. 1992) so that it may be tuned, and by symmetrically tapering the grating depth
or spacing so as to suppress sidelobes in its wavelength response (Sakata 1992).
The basic physical principal of the grating-assisted coupler is that the grating mod-
ulates the coupling between the two asymmetric (or nonsynchronous) waveguides at
a spatial frequency which exactly compensates for the dephasing of the propagating
waves in the two different waveguides. This same modulated-coupling technique is
realized through other means in the other coupler-based devices. For example, the
meander coupler shown in Fig. 11.2 uses a more macroscopic variation of the cou-
pling region, and was one early realizations of such a spatially varying coupler filter
(Bornholdt et al. 1990). This device uses two asymmetric waveguides with different
effective indices, realized through two different rib heights. As described above, the
power transfer at a wavelength in an asymmetric coupler dephases over a spatial
period length of ,
 −1/2
[β(λ)]2
(λ) = π κ2 (λ) + (11.3)
4
236 11 Wavelength Filtering and Manipulation

Fig. 11.2 The schematic top (a)

view (a) and cross section
(b) of a meander coupler
including definition of
geometrical device
parameters. Notice that the
vertical scale is expanded

(b)

To achieve phase matching of the two waveguides in the coupler then requires that
coupling be modulated with a spatial periodicity of (λc ). In the meander coupler,
phase matching is achieved by modulating (see below) the coupling constant between
the two waveguides. This modulation was first achieved by “meandering” one of the
waveguides toward and away from the second waveguide with a cos2 (πz/ (λc ))
spatial dependence. This device has been realized in the GaInAsP/InP materials sys-
tem; the device had a 10 − 15nm wavelength bandwidth at 1.3 µm (Bornholdt et al.
1990). Sidelobe suppression was later demonstrated by smoothly and symmetrically
varying the separation of the two coupled waveguides along the meander length,
that is, the cos2 (πz/ ) variation in κ had an overall symmetric taper imposed on it
(Venghaus et al. 1992).
Finally, another version of the coupler filter uses periodic coupling of symmetric
wavegudies, i.e., β1 = β2 , but achieves asymmetry in the coupler structure by placing
delay lines in one arm of the coupler. The wavelength dispersion in this device is
again a result of the dispersion properties of the waveguide itself. From a second
perspective, this device resembles a series of delay sections joined together, with
3 dB couplers (see Fig. 11.3) to form a series of Mach–Zehnder-like interferometers.
This structure is easily analyzed mathematically and it is a valuable device structure
to understand how filters can be built up. We will therefore discuss it here in more
detail (Kuznetsov 1994).
In essence, this latter device functions as follows: the phase difference between the
two arms of the device depends apparently on input wavelength and device geometry.
One can design wavelength-selective device to separate different wavelengths to
different output ports by a geometry design. For example, as shown in Fig. 11.3b,
output ports P1 and P2 have different wavelengths.
For an easier understanding, the basic operation of a single element of this device
as a wavelength filter is shown by the simplified interferometer-like device depicted
in Fig. 11.3a. It has two separate paths, with each having a different path length, L a
11.3 Asymmetric Coupler Filters 237

(a)

(b)

Fig. 11.3 a A sketch of a single delay section with a 3 dB coupler; b A sketch of a cascaded
Mach–Zehnder-like interferometers consists of a series of delay sections joined together, with 3 dB
couplers

and L b , and two 3d B couplers, which are assumed to be ideal and λ independent;
notice that the path length difference serves to make a delay line.
The transmission, ta , into one arm, a, of the single-element filter, is given by

ta = cos2 (φ/2) (11.4)

where
φ = β(L a − L b ) (11.5)

and β is the common propagation constant of the two waveguides. Since the prop-
agation constant for waveguides is typically dispersive, the phase factor varies with
λ. The operation of this device can be seen by considering the two coupled modes,
or supermodes, which are excited in the two coupler regions. The introduction of the
delay in one waveguide changes the phase of each of the supermodes, thus allowing
light to exit on either arm a or b, depending on the phase. The power into arm b is
given by tb = (1 − ta ).
However, this single device element displays only a cos2 response, and thus does
not have a sharp enough spectral response to act as a useful filter. Higher discrim-
ination or resolution can be achieved, however, by using a series of couplers, each
with its own delay line. This multiple-section device is designed such that at one
wavelength, the length difference, δL, in each section of the device is identical, i.e.,

mλr
δL = (11.6)
Neff (λr )

where m is the integral number of the order of interference of the device, Neff (λr )
is the effective index of the waveguide, and λr denotes the resonant wavelength
for which all the path lengths are identical. At the resonant wavelength, the light is
completely coupled into the cross state of each element as well as that of the full
238 11 Wavelength Filtering and Manipulation

device. Light, which is off-resonance, remains dominantly in the bar state, the exact
percentage of which depends on the number of stages, etc.
The interferometer of the device has a free spectral range, FFSR , which is given
by
c
FFSR = (11.7)
Neff (L a − L b )

Two wavelengths within the same free spectral range can be measured without ambi-
guity. Notice that these devices can be relatively long. For example, to achieve a
FFSR = 400 GHz at λ = 1.55 µm in I n P requires a device of ∼ 4.5 mm in length,
with a 2 mm radius-of-curvature bend (Kuznetsov 1994).
This behavior can be treated analytically. Consider the output of a sequence of
delay-line coupler units also shown in Fig. 11.3b, which act together as a cascaded
filter. To determine the output of the filter, it is necessary to write the transmission
matrix of the two elements—the coupler, [tC (L i )], and the delay-line section (behav-
an asymmetric Mach–Zehnder section), [t D ]. The total length
ing like  of the coupling
region, L i , must be specified at the resonant  wavelength, λ r , κ L i = π2, that
is, to have the bar state for λr requires that L i should be equal to one coupling
length. Then, if the input (i) and output (o) fields in the top waveguide, a, and that in
the bottom waveguide, b, are written as a vector using transfer matrices (see Chaps. 6
and 5),  o  i
a a
= [tC (L N +1 )] · · · [t D ][tC (L C )][t D ][tC (L 1 )] (11.8)
bo bi

where  
cos(κL i ) − j sin(κL i )
[tC (L i )] = (11.9)
− j sin(κL i ) cos(κL i )

and  
exp( jφ/2) 0
[t D ] = (11.10)
0 exp(− jφ/2)

where we have used the fact that the transfer matrix of the waveguide delay section
depends on the difference in path length or phase.
A multiple-section filter is shown in the solid-line curve in Fig. 11.4 for the case
of five sections. Note that the filter is peaked but with strong multiple sidelobes. This
distribution can be further sharpened by employing additional elements. The side-
lobes in the response curve result from small but finite coupling at other wavelengths.
The sidelobes can be supressed by “weighting” of each section. This adjustment of
the weighting may be done, for example, via the coupler length, L i , such that

Li 
ωi = ωi (11.11)
L total

where L total = L i . There are many possible weighting schemes to use for a filter,
just as is common practice, for example, in the design of electronic filters. Three
11.3 Asymmetric Coupler Filters 239

Fig. 11.4 A plot of N = 5

stage coupled
Mach–Zehnder (CMZ) filter
transmission curve for the
uniform (solid line), cosine
(a = 0.8) (dashed line), and
binomial (dotted line)
coupler weight distributions

possible schemes are as follows:

U ni f or m ωi = 1 (i = 1, 2, . . . , m + 1),
Cosine ωi = cos(πa((i − (m + 2)/2/m))),
m!
Binomial ωi =
(i − 1)!(m − i + 1)

Note the trade-off between the width of the central maximum in transmission and
the side lobe suppression among the various distributions. Thus, in going from the
uniform to the binomial distribution, the width of the central peak widens; however,
the sidelobe levels decrease in intensity, being −8 dB for the uniform distribution
and −47 dB for the binomial distribution.
In other optical filters, this cascaded coupler filter can be characterized by its
finesse, F, where
FFSR
F= (11.12)
ν
where ν is the width of the central transmission maximum, measured at a char-
acteristic level, denoted by T f . The finesse of a filter increases with the number of
stages. Thus, for a binomial distribution,
  
1 4 1
= ln (11.13)
F π2 m Tf

where, again, T f is the transmission level used to define the width of the central
lobe and m is the number of stages. In fact, the transmission function, Tbin , for a
binomially weighted series is TBin ≈ cos2m (φ/2), i.e., the response of the single
device raised to the mth power. Interestingly, use of a truncated binomial distribution,
i.e., dropping the first and last element, makes it easier to achieve higher finesse, since
240 11 Wavelength Filtering and Manipulation

Fig. 11.5 The layout of a resonant coupler device having seven directional coupler and six cascaded
Mach–Zehnder sections. Notice that the vertical scale is expanded by 50X

then it is found that the finesse depends linearly on the number of stages,

1
F∼ (11.14)
m
Generally, device loss and fabrication variability make it difficult to use truly
large numbers of stages, i.e., m > 8. An extensive theoretical discussion of apodized
delay-line coupler filters is presented in (Kuznetsov 1994).
Despite these difficulties, there are several examples of multiple-element delay-
line couplers which have been fabricated and tested. An excellent example is shown
in Fig. 11.5, which shows a six-element cascaded Mach–Zehnder filter (Yaffe et al.
1994). The large length of these filters (∼4 cm) makes it necessary to use only very
low loss optical materials, viz., SiO2 on Si or polymeric layers; the device in Fig. 11.5
uses SiO2 on Si.
The device in Fig. 11.5 also incorporates two important design elements. First, the
waveguides feeding the couplers are both curved away with equal radius to cancel
out any spurious coupling in this region, and, second, the waveguides in the actual
coupler region are tapered to enhance out-of-band rejection. The latter capability is
similar to the use of adiabatic tapering discussed in Chap. 5. The response of this
filter, using an inverse-cosine weighting and seven sections, is shown in Fig. 11.6,
along with the output calculated from the product of the cascaded transfer functions.
The measured total insertion loss at ∼1.65 µm was −0.25 dB.
The same technology as used for these multi-element filters has been applied
to other WDM PICs. For example, a 6 cm SiO2 on Si device (0.6mm width) has
been used for mux/demux of 1.30 and 1.55 µm signals (Li et al. 1995a). This device
used the wavelength layout geometry shown in Fig. 11.7. In addition, an application
designed to enable an erbium-doped fiber amplifier (EDFA) gain equalization has
also been described, which uses a similar filter design (Li et al. 1995b). This filter
was fabricated using SiO2 /Si technology and found to have an insertion loss of 1 dB
and an out-of-band wavelength rejection of 50 dB.
11.4 Devices Based on Diffraction Gratings 241

(a) (b)
Transmission [dB]

Wavelength [ ]

Fig. 11.6 A plot of the transmission curve of a a resonant coupler filter(RCF) with seven tapered
couplers: solid curve bar state, dashed curve cross state. b simulated RCF: dashed curve—perfect
couplers

Fig. 11.7 Layout of a 1.3/1.55 µm EDFA gain equalizer. Arrows represent input and output ports.
The total length of the device is 60 mm and the width is 0.6 mm. Notice that the vertical scale is
stretched 20 times the length

11.4 Devices Based on Diffraction Gratings

11.4.1 In-Line Filters Based on Bragg Gratings

Recall that in Chap. 10 several applications of Bragg gratings in wavelength filtering

were discussed. Two of these applications considered in-line filters. In the first, a
simple Bragg filter can be used to block a range of wavelengths. For a grating with
coupling coefficient κ, the bandwidth is proportionate to κ. On the other hand, if a
1/4 λ step is located in the midst of the grating, the grating will act as a narrowband

filter, e.g., only a simple telecommunications λ will be passed. More details on these
filters are given in Chap. 10.
242 11 Wavelength Filtering and Manipulation

Fig. 11.8 A sketch of a Bragg filter consists of two waveguides with a coupling coefficient κ. The
drop waveguide (upper) has a 1/4 λ step Bragg diffraction grating

11.4.2 A Bragg Filter for Channel Dropping

As explained in Chap. 10 on diffraction gratings, Bragg gratings containing a 1/4 λ

step, approximately in the middle of the grating, can be used as a very narrowband
wavelength filter. This same device can be combined with a waveguide coupler to
form a narrowband channel-dropping filter. This channel-dropping filter can be suffi-
ciently narrowband that it can be used to select out individual wavelengths in a dense
WDM signal, e.g., the device described by (Haus and Lai 1991) has a wavelength
bandwidth of ∼0.5Å. A similar type of grating, but without the 1/4 λ step, has been
described by (Kazarinov et al. 1987). It has a narrow bandwidth but requires a more
complicated geometry for the gratings.
In such a channel-dropping filter, a waveguide carries multiple wavelengths and a
resonant grating reflecting structure extracts or drops one of the wavelengths. Several
different and very advanced forms of this device have been described. However, here,
the device proposed by Haus and Lai (1991), shown in Fig. 11.8, will be discussed.
In the device, many wavelengths propagate along the bus waveguide but only one
wavelength couples resonantly to the drop port for detection or other use. The device
uses two waveguides, with a coupling coefficient κ, in conjunction with a 1/4 λ
step Bragg diffraction grating on one of the waveguides. The coupled structure has
been the subject of many theoretical studies and several experimental realizations
(Kazarinov et al. 1987; Levy et al. 1992). Recall that this basic grating structure was
discussed in detail in Chap. 10.
In order to illustrate the basic principles, we will examine only the simplest ver-
sion, namely, which is shown in Fig. 11.8. In this structure, the “bus” waveguide,
i.e., waveguide 1, will only couple to the drop waveguide, waveguide 2, if one of its
wavelengths lies within the passband of the 1/4 λ step grating arm, that is, into the
central resonance of the Bragg resonator. Wavelengths outside of this band, but still
in the stop band of the Bragg resonator, will be reflected and will not couple. The
expected narrow width of the central resonance of 1/4 λ step grating suggests that a
narrow filter can be made.
11.4 Devices Based on Diffraction Gratings 243

This structure may be analyzed numerically by solving the following coupled-

mode equations, which are written for the case of identical waveguides:

d A1
= − jδ A1 + κB1 − jμA2 (11.15)
dz

dB1
= jδ B1 + κA1 − jμB2 (11.16)
dz

d A2
= − jδ A2 − jμA1 (11.17)
dz

dB2
= jδ B2 + jμB1 (11.18)
dz

where A1,2 and B1,2 denote the forward and backward amplitudes of the optical
wave in the two waveguides shown in the figure, δ = (ω − ω0 )/νg is the detuning
parameter, ω0 is the frequency at band center, νg is the group velocity, μ is the
waveguide coupling coefficient, and κ is the grating coupling coefficient. Notice that
the detuning parameter, δ, is in the same form as in Sect. 5.3.2, and field amplitudes
are those without the rapid grating-induced variation (Haus and Lai 1991).
However, a simpler approach has been adopted by (Haus and Lai 1991) to deter-
mine the behavior of these filters, which uses only power-loss arguments. This anal-
ysis shows that filter bandwidth, ω f , and the stop band, ωs , are given as follows:

μ
ω f = 8νg κ| |2 (11.19)
κ
and
ωs = 2νg κ (11.20)

Clearly the two quantities are related by the ratio |μ/κ|2 ; in fact, these quantities
give the ratio of the power stored in the DFB resonator versus the power in the bus
waveguide. As an example of the results obtained from such an analysis, typical plots
of the output from such a filter are shown in Fig. 11.9 for the case of κ = 200 cm−1 ,
μ = 5 cm−1 , L 1a = 250 µm, L 1b = 150 µm, (Levy et al. 1992).
An analytic solution to the equations has been proposed by (Haus and Lai 1992).
Among other results, this solution showed that in a single-stage device, such as
the type shown in Fig. 11.10, a maximum of 30% power output coupling could be
achieved. As a result, more complex multistage devices have been proposed to drop
a higher fraction of the input power.
Several experimental versions of these channel-dropping filters have been made
including those in Ga As/AlGa As (Levy et al. 1992) and in SOI (Yu et al. 2009).
The results are very close to those predicted from the analysis given above. The
chief impediment to the practical implementation of this device is fabrication of
reproducible directional couplers.
244 11 Wavelength Filtering and Manipulation

Fig. 11.9 Theoretical

response of the
channel-dropping filter with
1/4 λ step grating. The inset is

the zoomed figure [Adapted

from (Levy et al. 1992)]
0.3Å

Fig. 11.10 The 3D design of Receiver

the Bragg grating for channel Arm
dropping [Adapted from
(Levy et al. 1992)] Coupling Transmission
Region Bus
Quarter-
wave step
Grang
(230nm)

Shallow Resonator
Trench Secon
Straight-through
Input Waveguide

11.5 Waveguide-Grating Router

11.5.1 Introduction

One very important and widely used waveguide routing device (Smit 1988), is a
PHASAR, or waveguide-grating router. Such a device makes use of calibrated delay
lines, which are formed via a curved, parallel array of waveguides. These delay lines
are “fed” via uniform illuminated light from the input waveguide, which has first been
expanded in angle via diffracting in a free-space region. After passage through the
delay lines, the outputs from the delay waveguides are focused into a set of specific
wavelength channels formed by the PHASAR output guides. A “waveguide-grating”
router can serve as a simple 1 × N demultiplexer (Smit 1988) or a N × N wavelength
router (Dragone 1991).
11.5 Waveguide-Grating Router 245

Input
aperture
Output
aperture FPR
Object
plane Image
FPR plane
Transmier
waveguide

Fig. 11.11 A sketch of a 1 × N and an N × N arrayed waveguide-grating (AWG) router

The basic device is shown in Fig. 11.11. It consists of two symmetrically placed
star couplers separated by an uncoupled waveguide array with a graduated path
difference between each waveguide. Because this graduated path difference leads
to a precise phase relationship between neighboring guides at the array output, it is
sometimes called a “waveguide grating” since a diffraction grating also has such a
wavelength-sensitive delay. If the device receives a specific input wavelength, then,
as a result of the optical imaging and phase delays, it steers this wavelength to a
specific output guide. The graduated path difference in the waveguide array causes
a tilting of the wavefront emerging from the grating array at an angle, which is
distinctive for each input wavelength.
246 11 Wavelength Filtering and Manipulation

Fig. 11.12 a The schematic of a waveguide with polyimide 1/2-wave plate in the middle of a
waveguide-grating array to equalize the path lengths for the TE-polarized and the TM-polarized
guided waves. b A detailed schematic of the slab waveguide [Adapted from (Takahashi et al. 1995)]

11.5.2 Device Structure

PHASARs have been made of a variety of materials, principally SiO2 on Si (Taka-

hashi et al. 1993) or InP (Smit and Van Dam 1996). The actual device, on which
Fig. 11.11 is based, is made of SiO2 on a Si platform. For comparison, a second design
for such a waveguide router is shown in Fig. 11.12. As discussed below, polarization
insensitivity is essential for realizing a useful router. Thus, in the router in Fig. 11.12,
a polyimide 1/2-wave plate can be included in the middle of a waveguide-grating
array to equalize the path lengths for the TE-polarized and the TM-polarized guided
waves (Takahashi et al. 1993).

11.5.3 Basic Functions of the Waveguide Array

A waveguide array provides two important functions for a PHASAR: focusing of

the light onto one of the receiver waveguides and providing angular dispersion for
wavelength routing. Consider first the focusing function.
Focusing is accomplished by reversing the optical transformation achieved when
the mode from the input waveguide diffracts into the free-space region and is received
at the waveguide array. Thus, a uniform phase distribution fed by the curved end face
of the waveguide array is the very reverse of the input section, and will thus yield a
focused beam. This phase front will have a uniform phase when the length difference,
11.5 Waveguide-Grating Router 247

Fig. 11.13 A measured

transmission spectrum of a
PHASAR (de)multiplexer
from central input port to
central output port

Transmission [dB]
(Takahashi et al. 1995)
Source https://ieeexplore.
ieee.org/document/372441

Wavelength [nm]

L, between adjacent waveguides, is equal to an integral number, m, of wavelengths

in the waveguide; thus
mλc
L = f
(11.21)
Ne f f

f
where λc is the central wavelength of the array and Ne f f is the effective index of the
waveguides. Further, in order to reduce overlap of multiple orders, it is necessary that
the waveguides be as closely spaced as possible, i.e., typically within one waveguide
width of each other. This condition is similar to the design principle that a diffraction
grating uses its lowest diffraction order when its groove spacing is small. The input
and output apertures are spaced at a distance Ra from the receiver waveguides.
In this case, using an analogy to a Rowland mounting for diffraction gratings, the
output/input waveguides must be on a circle of Ra /2. One of the most important
advantages of Rowland mounting is that its image does not have coma distortion. It
also can be designed to eliminate spherical aberration for a linear phase distribution
(März 1995). Figure 11.13 shows a measured transmission spectrum from central
input to central output.
Second, the waveguide array provides angular or spatial dispersion at each wave-
length. In order to discuss this dispersion, we use the simplified geometry shown
in Fig. 11.12. This chapter focuses on the essential physics of the device disper-
sion and thus ignores some design elements of the device, such as the waveguide
mutual coupling near the input/output aperture. As described in Chap. 9 in conjunc-
tion with star couplers, this coupling is important to provide uniform illumination of
the receiver waveguides and, thus, to obtain the best performance in each of these
imaging devices. From Fig. 11.12, the equivalent grating equation, for order m, of
the system is given by
248 11 Wavelength Filtering and Manipulation

Table 11.1 Typical quantities for a reported near-commercial waveguide-grating array (Takahashi
et al. 1993). The parameters apply to the device shown in Fig. 11.12. Typical parameters used for
PHASARs are InP, SiO2, and Si
Basic materials and optics Device parameter
parameters
λ = 1.5538 N 16
Nes f f = 1.4529 Ra 1 cm
f
Ne f f = 1.4513 L 130 µm
n g = 1.4752 da 25 µm
ωo = 4.5 µm m 118
dr 25 µm
ν 100 GHz (0.8nm at 1.55 µm)
FSR 1600 GHz
Cross talk −30 dB
Insertion loss 5 dB
FW H M 30 GHz (∼4 µm spot size)

f
s
Neff da sin θi + Neff L + Neff
s
da sin θo = mλ (11.22)

s f
where Neff and Neff are the effective indices of the free propagation region (FPR)
and the waveguide, respectively; θi and θo are angles of the diffracted wavefront at
the input and output slabs from the emitting and the receiving (output) waveguides,
respectively; da is the spacing of the array guides; and L is the length difference
between successive waveguides in the grating array. The angles θi,o are given as
follows:
i · dr
θi = (11.23)
Ra

o · dr
θo = (11.24)
Ra

where i, o is the number index of the input/output waveguides, measured from the
center of the receiver waveguide array for wavelength λ, and da is the waveguide
spacing as shown in the figure. Note that (11.21) assumes that these angles were zero.
Typical numbers for the parameters used in (11.22) are given in Table 11.1. Observe
f
that for the center wavelength of the router, λc , (11.22) reduces to Ne f f L = mλc ,
i.e., (11.21) holds.
Differentiating (11.22) gives the angular dispersion of the grating in the vicinity
of the center wavelength, where θi,o ≈ 0; thus
11.5 Waveguide-Grating Router 249

dθ mλ2 n g
=− f
(11.25)
dν s
Neff da cNeff

and where the group index, n g , of the waveguide is given by

f
f dNeff
n g = Neff − ·λ (11.26)
dλ
and c is the speed of light in vacuum. The quantity in (11.25) is easily converted into
the spatial dispersion of the device, dx/dν, near the center frequency by multiplying
that equation by Ra to give

dx Ra mλ2 n g
=− f
(11.27)
dν s
Neff da cNeff

The channel spacing is then given by multiplying dν/d x by dr , the spacing or pitch
of the receiver arrays, to yield

−1
dr mλ2 n g
ν = f
(11.28)
Ra s
Neff da cNeff

Figure 11.14 demonstrated an 18-channel AWG (de)multiplexer with channel

spacing of 200 GHz. The cross talk is −18 dB and the insertion loss is −6 dB.

Fig. 11.14 An 18-channel

AWG (de)multiplexer with
channel spacing of 200GHz. ~6dB ~18dB
The cross talk is -18dB and
the insertion loss is -6dB.
250 11 Wavelength Filtering and Manipulation

11.5.4 Free Spectral Range

In order for the device to operate as a useful filter, it is necessary for the router to
operate over the entire spectrum of interest within a single “diffraction” order, m.
Alternatively stated, the frequency or wavelength range of the device must be such
that the spectral response does not overlap on itself; this requirement is guided by
the free spectral range (FSR) of the device. If the channel spacing is ν and the
total channel number is N , the minimum useful free spectral range (FSR) is N ν.
Thus, when ν → ν + F S R,m → m + 1. At this frequency, Neff s
→ Neff s
+ Neffs
f f f s, f
and Neff → Neff + Neff , where the differences Neff = − ν (λ(d Ns, f /dλ)).
FSR

Substitution of these quantities into (11.22) gives a difference equation for FSR
as

 f

  
−1 1 f dN dN s
(F S R) = Neff − λ eff L + s
Neff − λ eff (da sin θi + da sin θo )
c dλ dλ
(11.29)
1 
≈ n g (L + da sin θi + da sin θo ) (11.30)
c

where c is the speed of light in vacuum. For the diagonal ports in the device (0, 0),
(1, −1), (2, −2), etc., the inverse of the free spectral range is given by (F S R)−1 =
n g L/c.

11.5.5 The Frequency-Dependent Transmission Function of

the Router, i.e., The PHASAR Frequency Response

The focal spot centered at each output port depends on the focusing properties of
the waveguide array. As described in Chap. 9, the imaging quality and transmission
can be enhanced by introducing a degree of mutual coupling between neighboring
waveguides. In this section, this coupling will be neglected and the waveguides in
the array and the two apertures will be assumed to be the same! In addition, use of
a Rowland optical layout geometry is assumed since its usage leads to high-quality
images. The focal spot size of the grating on the output port can be fit with a Gaussian
distribution, and the coupling of the input and output waveguides is then that of two
coupled Gaussian distributions of spot size ω0 , laterally separated by a displacement
x, i.e.,
 

x 2
T (δν) ∼ exp − (11.31)
ω0
11.5 Waveguide-Grating Router 251

Fig. 11.15 A plot of

measured performance of the
16-port PHASAR
wavelength demultiplexer
near a resonance peak
(Takahashi et al. 1995)
Source https://ieeexplore.
ieee.org/document/372441

The frequency-dependent transmission function of the focal spot thus depends on

the frequency difference, δν, from the center frequency.
 
δν dx 2
T (δν) ≈ exp − (11.32)
ω0 dν

where dν/dx is the spatial linear dispersion of the device and ω0 is the focal spot
size. As pointed out in Takahashi et al. (1995), this response√function is the same
as a Gaussian bandpass filter, which has a FWHM or ν = 2 ln 2ω0 (dν/dx). The
spectral response or resolution of the device is thus set by the imaging quality.
Typical numbers for device resolution are ∼10 s of GHzs. A plot of the calculated
and measured value for the 16-part device of Takahashi et al. (1995) is given in
Fig. 11.15. Since the frequency response is directly related to the focal spot size, ω0 ,
factors which give well-defined focal spot, such as using the Rowland mount, etc.,
are important for good frequency filtering as well. A typical first approximation,
described by Smit and Van Dam (1996), is that the focal spot size is equal √ to the
effective width of the modal field times a constant, ωe , and thus ω0 = ωe 2/π.

11.5.6 PHASAR Uniformity and Insertion Loss

An issue related to, but separate from, the resolution of each wavelength channel is
the overall wavelength response envelope of the device (Smit and Van Dam 1996).
A diagram showing an example of this response is given in Fig. 11.16. The diagram
shows both the overall insertion loss at the central wavelengths plus the envelope
function that causes nonuniformity of the wavelength channels of the device. Notice
that in this example a portion of the light passing through the device appears in orders
252 11 Wavelength Filtering and Manipulation

central inseron loss Lc

100%
order m order m-1 order m order m+1

non-uniformity Lu
intensity

far field
envelope
FSR

order m-1 order m+1

4 1 4 1 4 1
channel 4 1 4 1 4 1
image plane frequency

Fig. 11.16 A diagram showing an example of the overall wavelength response envelope of the
device. The diagram shows both the overall insertion loss at the central wavelength plus the envelope
function that causes nonuniformity of the wavelength channels of the device

different from m; the existence of these other orders results in the insertion loss for
the desired light in order m.
This envelope function of the wavelength channels can be found by considering
a Gaussian transformation to be

I (θ) = Io exp(−2θ2 /θ02 ) (11.33)

where θ0 = πλ/Neffs
ω0 . Note that
√ ω0 is related to the effective modal half width in
the waveguide, ωe , by ω0 = ωe 2/π. The value of 2ωe has been discussed earlier
for slab waveguides, where it was called teff . In the notation here, it is
 
ω 2
ωe = 1+ (11.34)
2 γ
 
ω 1
≈ 0.5 + (11.35)
2 V − 0.6

where ω is the geometric width of the waveguide and V is the normalized frequency
of the guide. Equation 11.35 was obtained by curve fitting the variation in γ with
the normalized frequency V .
This function leads to a nonuniformity L u , i.e., 10 log(Imaxθ /Icentral ), across the
full spectral response, as depicted in Fig. 11.16, which is equal to

L u ≈ 8.7θmax
2
/θ02 (11.36)
11.5 Waveguide-Grating Router 253

where θmax ≈ xmax /Ra and xmax is the transverse distance in the receiver array and L u
is in dB. The uniformity is thus ∝ N dr . Note that this uniformity results in additional
insertion loss for the outer channels compared to the central channel.
The loss of the central channel is due to diffraction into higher orders. This loss,
which is due to the presence of any outer m ± 1 order and has a relative magnitude
to that in the center can be obtained by realizing it is displaced in frequency by
ν ≈ ν F S R . Thus, the power in this order, say m + 1, will be

Pm+1 ∼ exp(−2θFSR
2
/θ02 ) (11.37)

where
1 dx
θFSR = νFSR (11.38)
Ra dν

If only the adjacent m ± 1 orders are considered, the power loss in dB can be
estimated to be (where we have assumed that the propagation loss is zero)

loss ≈ 17 exp(−2π 2 ω02 /da2 ) (11.39)

This loss includes that occurring at both the input and output stages and the fact
that light is lost into two orders. Notice that as mentioned earlier, da should be small
to minimize coupling into higher grating orders.

11.5.7 Channel Cross Talk

Cross talk is a major consideration in designing routers. It arises from fabrication

imperfections, such as from index fluctuations as well as from factors, which can be
controlled by good design, including, most obviously, the overlap of the modal fields
in adjacent receiver waveguides. This overlap is clearly a function of the receiver
waveguide spacing, dr , and the receiver modal width, ∼ωe . Second, truncation of the
array aperture as a result of an abrupt end on a waveguide introduces a diffraction
ripple on the image. This spatial imaging spreads light into orders other than m.
Similar effects can arise from modal conversion in the array waveguides. Finally,
curves in the array waveguides can also lead to conversion into higher modes. These
modes radiate power at higher angles and couple into the m + 1 and m − 1 modes.
The first two sources can be placed on more quantitative footing using an analysis
by Smit and Van Dam (1996). Consider first cross talk due to the overlap of fields in
the receiver waveguides. This overlap is a result of the lateral “tail” of the mode in the
input waveguide, which is being imaged into one receiver waveguide, overlapping
onto a neighboring input waveguide. Since the “tail” portion of this mode is not
well fit by a Gaussian distribution, unlike the central region of the mode, the overlap
integral must consider the actual shape of the waveguide mode as a function of V
254 11 Wavelength Filtering and Manipulation

and the normalized receiver waveguide spacing, dr /ω. The results of this calculation
give cross talk versus dr /ω for several values of the waveguide parameter V .
The second source of design-controllable cross talk is that due to the finite angu-
lar width of the array aperture, 2θa . At the input region of the aperture, this finite
width causes loss since a finite aperture does not intercept the entire radiation lobe,
characterized by θ0 , from the input waveguide. At the exit aperture of the delay array,
a finite width gives rise to sidelobes on the focused image of each receiver channel.
These sidelobes can then cause cross talk to their deviation from a simple single-lobe
image. This source of cross talk is clearly a function of the actual relative position
of the receiver waveguide. It is, however, relatively insensitive to the waveguide V .

11.5.8 PHASAR Polarization Properties

Birefringence in the waveguide router can lead to major degradation of the output. If
the router responds differently to the two polarizations, then the output, as shown in
Fig. 11.17, will exhibit a dual response at each of the two polarizations. This splitting,
ν pol , can be expressed as

TE
Neff − Neff
TM
νpol ≈ ν T E
(11.40)
ng

T E,T M
where it is understood that Neff refers to the effective index of the waveguide
for TE or TM polarization. In one example, i.e., I nGa As P − I n P waveguides, the
f
shift was ∼5 nm. Since this shift scales as Neff it is clearly expected to be low in
low-index materials.

TMm TEm TMm-1 TEm-1

PD:
Polarizaon
dispersion
PD
TMm
TEm
FSR

TMm-1
TEm-1

Fig. 11.17 A diagram showing the output if the router responds differently to the two polarizations.
It will exhibit a dual response at each of the two polarizations [Adapted from (Smit and Van Dam
1996)]
11.5 Waveguide-Grating Router 255

Several approaches have been adopted to reduce birefringence, including overlap-

ping the m and m + 1 orders for different polarizations, use of symmetrical waveg-
uides, insertion of λ/2 plate in the array section, and compensation of the dispersion
by surface loading of the array. Recall that a diagram showing the insertion of λ/2
plate was depicted in the router given in Fig. 11.12; in this case, the plate was a thick,
free standing polyimide film.

11.5.9 Design of a Multiplexer/Demultiplexer

Practical design of a PHASAR-type device can be illustrated by one specific approach

to the design of a wavelength demultiplexer/multiplexer, the “Delft” procedure (Smit
and Van Dam 1996). It is assumed that the basic waveguide structure is fixed by the
materials technology used: the waveguide can be described by its width, ω, and
lateral normalized frequency, V . Further, the device channel spacing, ν, and the
total number of channels are specified as a result of the system needs. Finally, the
waveguide spacing da is chosen to be as small as possible, consistent with main-
taining independent waveguides; thus again in this illustration, the coupling between
waveguides is ignored. A small spacing, i.e., da , is important since the spacing allows
a large proportion of the diffracted input beam to be captured by the array. Note that
the presence of collecting horns (or reverse tapers) can be used to increase collection.
The first design goal is to achieve an acceptable degree of receiver cross talk, which
is determined by the spacing of the receiving waveguides compared to their width.
This spacing is set by the channel spacing. This overlap spacing results from the finite
size of the input-waveguide image between neighboring waveguides. This overlap
can be adjusted via the relative receiver waveguide spacing, dr /ω. To determine
dr /ω, the cross talk is set by the material and fabrication tolerance, say −40 dB, and
then use the geometry in figure a along with our waveguide V to obtain dr /ω.
Next, the radius of the free-space propagating region must be selected. This dimen-
sion is found from the acceptable margin on uniformity, which in turn fixes θmax as
shown in (11.36). This expression gives the excess loss for the outermost waveg-
uides, which are at angle θmax , compared to that of central waveguide. Once this
nonuniformity is decided, θmax is specified, then using the basic geometric relation
θmax = N dr (2θmax ), i.e., the receiver waveguide spacing and the total number of
waveguides, the value Ra ≈ N dr (2θmax ) is also determined.
The array length difference, L, is then obtained via the full device dispersion
relation, which is equal to dr /ν, i.e., the choice of receiver waveguide spacing and
the fixed channel spacing. Specifically, the required L is obtained from the angular
f
dispersion (11.25) and the fact that Neff L = mλc , yielding this relation:

dr 1 n g L Ra
= (11.41)
ν νc Nesf f da
256 11 Wavelength Filtering and Manipulation

where use had been made of the fact that d x/dν = dr /ν. This expression can then
be manipulated to obtain an equation for L.
The angular half-width of the array aperture, θa , is determined by the fact that a
finite angular width causes sidelobes in the focused spot at the receiving aperture. The
choice of θa then fixes the number of array waveguides, Na , due to simple geometric
considerations,
Na = 2θa Ra /da + 1 (11.42)

Note that two general PHASAR layouts have been used. These are shown in
Figs. 11.11 and 11.12. Figure 11.11 is the most straightforward design since the bend
radius is constant throughout, while Fig. 11.12 requires less waveguide transitions.
These two designs have been used for SiO2 and III-V devices, respectively.
Finally, note that this design procedure is useful only for a simple 1 × N PHASAR.
Wavelength routers require a somewhat different approach; this different procedure
is given in the paper by Smit and Van Dam (1996).

11.5.10 More Advanced Design Features

Since the development of the routers, router design has been continually refined.
As a result, many new features have been realized. For example, chirping or even
double-chirping of the waveguide lengths in the router can be used to eliminate side
lobes (Doerr and Joyner 1997). The reader is referred to the excellent reviews by Smit
and Van Dam (1996) and by Takahashi et al. (1995) for more details and examples.

11.5.11 MMI Phased-Array Wavelength (De)multiplexer

PHASARs can also be made by replacing its star-coupler section with an MMI-based
coupler. These MMI-PHASARs have an inherently periodic spectral response and are
of a compact size for the typical number of channels. In addition, in comparison with
conventional star-coupler-based devices, these phasers have low insertion loss, better
uniformity, and simpler device structure. However, they have relatively high cross
talk as a result of their narrow low crosstalk window, and their wavelength response
cannot be further flattened. The crosstalk problem has prevented their application
to wavelength routers, especially when N , the channel number, is large. Recently,
an optimized index-contrast technique (Van Dam et al. 1995) has been proposed to
improve the crosstalk performance for N × N MMI-PHASARs, especially for large
N.
MMI-based PHASAR demultiplexers are particularly useful for multi-wavelength
lasers since in this case cross talk is not a central issue (see below). Thus far, 1 × 4,
4 × 4 and 1 × 5 MMI-PHASARs have been fabricated, all using high-index con-
trast deep-etched structures on InP (Smit and Van Dam 1996). MMI-PHASARs
11.5 Waveguide-Grating Router 257

Fig. 11.18 Basic functions of arrayed-waveguide N × N multiplexer. a Multiplexing; b Demulti-

plexing; c Add-drop multiplexing operation; d N × N interconnect

have also been operated either passively or dynamically. Electro-optically controlled

phase shifters can be added to each phase-arrayed waveguide to achieve different
wavelength routing permutations. This type of MMI-PHASARs can be used in a
reconfigurable add-drop (de)multiplexer (see Fig. 11.18) to greatly reduce the device
complexity, compared to that in star-coupler-based PHASARs, although again, the
issue of cross talk is important in this application. Other applications of arrayed-
waveguide multiplexers are given in Fig. 11.1. Finally, the reader is reminded that a
discussion of MMI-based PHASARs is included in Chap. 9, although the discussion
in that chapter has a different motivation.

11.5.12 Applications of PHASARs

Because of their very powerful wavelength multiplexing and demultiplexing func-

tionalities, PHASARs have become an extremely versatile component in integrated
optic multi-wavelength applications. These applications are so numerous that we
will only list a few examples in this section. Some of these applications are reviewed
by (Smit and Van Dam 1996).
As an example, a simple but important class of waveguide-grating router appli-
cations is in integrating an semiconductor amplifier with a grating section. This
combination can be used in conjunction with cleaved “chip” facets for an integrated
series of laser sources. These sources have well-defined, fixed frequencies as set by
the PHASER device. However, tuning of the laser source or sources can be achieved
258 11 Wavelength Filtering and Manipulation

Fig. 11.19 InP-based HR (98%)

integrated series of laser
sources [Adapted from
(Zirngibl et al. 1994)]

Waveguide
Grating HR (80%)

Free-Space
Region

Amplifier
Array

by selectively turning on or off the amplifier sections. An example of such a device

is shown in Fig. 11.19 (Zirngibl et al. 1994). The device was made of I n P-based
materials. It had four laser sections for CW emissions in the vicinity of ∼1.5 µm
into seven wavelength channels.

11.6 Summary

Wavelength manipulation, including filtering, is at the very heart of the utility of

integrated optics in WDM systems. This chapter has examined at examples of three
different technologies for accomplishing this filtering. Of these three device types,
the most important commercially is that of the wavelength router. Its ability to manip-
ulate and route multiple wavelengths has enabled a large number of crucial system
applications to be demonstrated. Routers have been shown to be important in either
Si O2 or III-V material platforms; the application of single-mode Si photonics to this
device is however more of a challenge due to the very strong waveguide dispersion
in Si wire devices.
In addition, each of the three device types given in this chapter illustrates a different
feature of designing PICs. These include the use of transverse matrix design for the
delay-line coupler filter, use of Bragg gratings for resonators and coupling in the 41 λ-
shift channel-dropping filter, and phase arrays and planar optics for routers. While
each of these design areas appears in many other integrated optical devices, their use
11.6 Summary 259

for wavelength filtering is central for one of the most important areas of integrated
devices.

Problems

1. (a) Calculate the channel uniformity for the PHASAR having the characteristics
of that in Table 11.1.
(b) Calculate the spatial dispersion near λc = 1.55 mm.
(c) Use your skill with waveguides to show how to obtain w0 . You will first
need to obtain the V of the waveguide. Again use the data in Table 11.1.
(d) Determine the loss of the device based on (11.39). How does it agree with
the 5dB value in Table 11.1?
(e) Derive (11.41).
2. Optical Modulator: You wish to build the following phase modulator in LiNbO3
using proton exchange:

(a) What is the crystal orientation for using r33 ?

(b) What is the width, w, for single-mode performance?
(c) What is the estimated  if electrode spacing equals 1.5 w?
(d) What is a typical Vπ L for phase modulation in LiNbO3?
(e) Calculate Vπ length if V = 10 V and f RF ∼ 100 MHz (do not use 4).
(f) What is C/L in pF/cm using graph in notes?

3. Use what you have learnt so far about PHASOR, given the following parameters
of a PHASAR in Silica, try to answer the questions.
λs = 1.5 µm
s
Neff = 1.4529
f
Neff = 1.4513
W0 = 4.5 µm
n g = 1.4752
dr = da = 30 µm
Ra = 2 cm

(a) What is the required L for diffraction order m = 150?

(b) What is the channel spacing ν?
(c) If 16 channels are needed, what will be the nonuniformity in d B?
(d) What is the rough power loss of this device if there is no propagation loss?

4. Derive (11.30), the expression for FSR of a PHASOR.

5. If you test a grating reflector, and find the coupling coefficient κ = 200 cm−1 .
Assume the group index of this grating is in the order of unity. Answer the
following questions:
260 11 Wavelength Filtering and Manipulation

Fig. 11.20 Cross-sectional

view of a modulator

(a) How long must the grating be to get a reflection rate R = 99% at the detuning
δ = 0?
(b) What is the stop band of the grating?

6. Derive the filter function as in Fig. (11.3) for a three-stage uniform filter. Assume
the input fields of the input ports are ai and bi , respectively, and the couplers are
50:50 couplers. Hint: use matrix representation of photonic devices.
7. Consider the PHASAR shown in figure. Assume the PHASAR is designed for the
following “mission”: channel spacing 100 GHz, λc = 1.55 µm and has the fol-
f
lowing parameters: Neff s
= 1.453, Neff = 1.451, w0 = 4.5 µm, N = 16, Ra =
1 cm, da = dr = 25 µm, ν = 100 GHz

(a) Calculate the spatial dispersion in µm/GHz used in this device.

(b) Estimate the fractional nonuniformity of the images on the receiver array,
that is, what approximately is Ioutermost /Icenteal . Notice the answer should
NOT be in dB.
(c) Write down as an equation the condition for L to have an image of order
m on the receiver array (Fig. 11.20).

References

Alferness, R., Buhl, L., Koren, U., Miller, B., Young, M., Koch, T., et al. (1992). Broadly tunable
ingaasp/inp buried rib waveguide vertical coupler filter. Applied physics letters, 60(8), 980–982.
Bornholdt, C., Kappe, F., Müller, R., Nolting, H.-P., Reier, F., Stenzel, R., et al. (1990). Meander
coupler, a novel wavelength division multiplexer/demultiplexer. Applied physics letters, 57(24),
2517–2519.
Chen, S., Fu, X., Wang, J., Shi, Y., He, S., & Dai, D. (2015). Compact dense wavelength-division (de)
multiplexer utilizing a bidirectional arrayed-waveguide grating integrated with a mach-zehnder
interferometer. Journal of Lightwave Technology, 33(11), 2279–2285.
Doerr, C., & Joyner, C. (1997). Double-chirping of the waveguide grating router. IEEE Photonics
Technology Letters, 9(6), 776–778.
Dragone, C. (1991). An n* n optical multiplexer using a planar arrangement of two star couplers.
IEEE Photonics Technology Letters, 3(9), 812–815.
Haus, H. A., & Lai, Y. (1991). Narrow-band distributed feedback reflector design. Journal of
lightwave technology, 9(6), 754–760.
Haus, H. A., & Lai, Y. (1992). Narrow-band optical channel-dropping filter. Journal of lightwave
technology, 10(1), 57–62.
References 261

Herben, C., Vreeburg, C., Leijtens, X., Blok, H., Groen, F., Moerman, I., et al. (1997). Chirping of an
mmi-phasar demultiplexer for application in multiwavelength lasers. IEEE Photonics Technology
Letters, 9(8), 1116–1118.
Kazarinov, R., Henry, C., & Olsson, N. (1987). Narrow-band resonant optical reflectors and resonant
optical transformers for laser stabilization and wavelength division multiplexing. IEEE journal
of quantum electronics, 23(9), 1419–1425.
Kuznetsov, M. (1994). Cascaded coupler mach-zehnder channel dropping filters for wavelength-
division-multiplexed optical systems. Journal of Lightwave Technology, 12(2), 226–230.
Levy, M., Eldada, L., Scarmozzino, R., Osgood, R., Lin, P., & Tong, F. (1992). Fabrication of
narrow-band channel-dropping filters. IEEE photonics technology letters, 4(12), 1378–1381.
Li, Y., Henry, C., Laskowski, E., Mak, C., & Yaffe, H. (1995a). Waveguide edfa gain equalisation
filter. Electronics Letters, 31(23), 2005–2006.
Li, Y., Henry, C., Laskowski, E., Yaffe, H., and Sweatt, R. (1995b). Monolithic optical waveguide
1.31/1.55/spl mu/m wdm with-50 db crosstalk over 100 nm bandwidth. Electronics Letters,
31(24):2100–2101.
Lierstuen, L., & Sudbo, A. (1995). 8-channel wavelength division multiplexer based on multimode
interference couplers. IEEE Photonics Technology Letters, 7(9), 1034–1036.
März, R. (1995). Integrated optics: design and modeling. artech house. Inc., Norwood, MA.
Sakata, H. (1992). Sidelobe suppression in grating-assisted wavelength-selective couplers. Optics
letters, 17(7), 463–465.
Smit, M. K. (1988). New focusing and dispersive planar component based on an optical phased
array. Electronics letters, 24(7), 385–386.
Smit, M. K., & Van Dam, C. (1996). Phasar-based wdm-devices: Principles, design and applications.
IEEE Journal of selected topics in quantum electronics, 2(2), 236–250.
Takahashi, H., Hibino, Y., Ohmori, Y., & Kawachi, M. (1993). Polarization-insensitive arrayed-
waveguide wavelength multiplexer with birefringence compensating film. IEEE photonics tech-
nology letters, 5(6), 707–709.
Takahashi, H., Oda, K., Toba, H., & Inoue, Y. (1995). Transmission characteristics of arrayed
waveguide n/spl times/n wavelength multiplexer. Journal of Lightwave Technology, 13(3), 447–
455.
Van Dam, C., Amersfoort, M., ten Kate, G., van Ham, F., Smit, M., Besse, P., Bachmann, M., and
Melchior, H. (1995). Novel inp-based phased-array wavelength demultiplexer using a generalized
mmi-mzi configuration. In Proc. 7th Eur. Conf on Int. Opt.(ECIO95), pages 275–278.
Venghaus, H., Bornholdt, C., Kappe, F., Nolting, H.-P., & Weinert, C. (1992). Meander-type wave-
length demultiplexer with weighted coupling. Applied physics letters, 61(17), 2018–2020.
Weber, J.-P. (1997). A new type of tunable demultiplexer using a multi-leg mach-zehnder interfer-
ometer. In Proceedings of European Conf. Integrated Optics (ECIO 1997, Stockholm, Sweden),
pages 260–263.
Yaffe, H. H., Henry, C. H., Serbin, M. R., & Cohen, L. G. (1994). Resonant couplers acting as add-
drop filters made with silica-on-silicon waveguide technology. Journal of lightwave technology,
12(6), 1010–1014.
Yi-Yan, A., Deri, R., Seto, M., & Hawkins, R. (1989). Gaas/gaalas asymmetric mach-zehnder
demultiplexer with reduced polarization dependence. IEEE Photonics Technology Letters, 1(4),
83–85.
Yu, H., Yu, J., Yu, Y., & Chen, S. (2009). Design and fabrication of a photonic crystal channel drop
filter based on an asymmetric silicon-on-insulator slab. Journal of nanoscience and nanotechnol-
ogy, 9(2), 974–977.
Zirngibl, M., & Joyner, C. (1994). 12 frequency wdm laser based on a transmissive waveguide
grating router. Electronics Letters, 30(9), 701–702.
Zirngibl, M., Joyner, C., & Glance, B. (1994). Digitally tunable channel dropping filter/equalizer
based on waveguide grating router and optical amplifier integration. IEEE Photonics Technology
Letters, 6(4), 513–515.
262 11 Wavelength Filtering and Manipulation

Zirngibl, M., Joyner, C., & Stulz, L. (1995). Wdm receiver by monolithic integration of an optical
preamplifier, waveguide grating router and photodiode array. Electronics Letters, 31(7), 581–582.
Chapter 12
Electro-Optical Modulators

Abstract Electro-optic modulators are based on the control of guided waves using
electro-optic variation of the phase or amplitude using an applied electric field. Dif-
ferent theoretical approaches can be used to describe electro-optic control includ-
ing coupled-mode theory. Phase retardation of the guided light enables polarization
splitting, optical switching, and wavelength-selective coupling. Two model–material
systems, which are extensively used for E/O devices: GaAs (or other III–V semi-
conductor material) and lithium niobate. In addition, secondary material systems are
mentioned throughout.

12.1 Introduction

The electro-optic effect can be used in optical waveguide devices to control the phase
or amplitude of a guided wave. Typically, these devices use this effect to control polar-
ization or interference of guided-light beams, with the most important application
being in integrated optical modulators. These devices operate on the electro-optical
coupling of TE and TM modes in a waveguide, or pure phase retardation of waveg-
uide modes. This coupling can be understood and treated via coupled-mode theory.
In addition, electro-optic control of phase retardation of the guided light enables
polarization splitting, optical switching, and wavelength-selective coupling.
The electro-optic effect results from a redistribution of bound charges in a dielec-
tric waveguide upon the application of a voltage across the optical guide (Saleh et al.
1991). Deformations of the lattice from the applied electric field also contribute to
the effect. These perturbations cause a change in the optical impermeability tensor
relating the displacement vector, D,  and the electric field, E,
 of the optical wave.
Equivalently, the application of an electric field causes a change in the index of
refraction of the crystal.
In this chapter, we will first briefly review the fundamentals of the electro-optic
effect and consider two model–material systems, which are used heavily for the
E/O devices: GaAs (or other III–V semiconductor material) and LiNbO3 . In addi-
© Springer Nature Switzerland AG 2021 263
R. Osgood jr. and X. Meng, Principles of Photonic Integrated Circuits,
Graduate Texts in Physics,
https://doi.org/10.1007/978-3-030-65193-0_12
264 12 Electro-Optical Modulators

Fig. 12.1 The x- and

y-components of the E-field
in a crystal with different
refractive indices

tion, electro-optical devices have been made in a variety of other material systems,
including polymers, etc.; these secondary materials are mentioned throughout. The
chapter will then present a detailed discussion of optical modulators.

12.2 An Overview of Electro-Optic Modulation

Consider first the application of an electric field to a bulk crystal which has an
anisotropic set of refractive indices. The application of the field can change the
magnitude and direction of the indices in the crystal. In subsequent sections given
below, we will explain the details of this change, including its dependence on the
direction of the applied external electric field, E 0 , and the crystal. If we assume
that the crystal has well-defined indices of refraction, say n x,y , for an optical field
polarized along with x̂ and ŷ, respectively, where x̂ and ŷ are termed the principal
axis, then if the optical electric field is not along x̂, or ŷ, its components along x̂ and
ŷ have different propagation velocities because of the different indices of refraction
in the crystal; see Fig. 12.1.
This change in the principal axes can be used to make a simple modulator. If the
input is linearly polarized at 45◦ with respect to the x̂-axis and the input optical field
amplitude is E 0 , then after traveling a distance L in the crystal, the optical electric
 will be given by
field, E,

E0  
E = √ x̂e−i λ n x L + ŷe−i λ n y L
2π 2π
(12.1)
2

The different relative phases in the x̂- and ŷ-directions show that the wave is ellipti-
cally polarized.
A polarization analyzer at 45◦ to the x̂ or ŷ axes will “pass” optical radiation as
follows:  
  1
| E 45◦ | = E · √ (x̂ + ŷ) (12.2)
2
12.2 An Overview of Electro-Optic Modulation 265

Fig. 12.2 The magnitudes

of components in x- and
y-directions are the same.
However, each of the two
views of different refractive
indices. Thus, the total
magnitude is related to the
propagation distance in
material

π π
= E 0 e−i λ (n x +n y )L cos (n x − n y )L (12.3)
λ
where it is noted that (12.2)√contains a dot product, see Fig. 12.2. The intensity (here
we ignore the prefactor 1/2 ε/μ) along this 45◦ direction is then
π
| E45◦ |2 = E 02 cos2 (n x − n y )L (12.4)
λ
For the linear (Pockels) electro-optic effect, n x − n y = α · E 0 , where α is a con-
stant and E 0 is the applied electric field. Thus, by varying the applied voltage, one
can modulate the transmitted guided-light intensity.

12.2.1 Basics of Propagation of Lightwaves in Anisotropic

Crystals

The electro-optical effect operates by the modification of a crystal’s optical properties

in the presence of an electric field. Thus the starting point in describing the electro-
optical effect is the propagation of light in a crystal in the absence of an external field
(Saleh et al. 1991). In a material medium such as an anisotropic crystal, the derivation
of the wave equation shows that light propagation is governed by the relation between
the two optical fields, D  and E.
 This relation is generally given by

E i = i−1
j Dj (12.5)

where the sum over repeated indices, e.g., the index in (12.5), is understood, and
where i−1j is the inverse dielectric tensor or impermeability tensor. If one chooses a
set of reference coordinate axes, which lie along a special set of axes, called principal
axes, the tensor is diagonal, i.e.,
266 12 Electro-Optical Modulators

Fig. 12.3 The index

ellipsoid of zincblende-type
materials (solid line). In this
case, the ellipsoid is reduced
to a sphere, which means the
indices of refraction are the
same no matter what
direction light comes from

⎡ ⎤
11 0 0
[] = ⎣ 0 22 0 ⎦ (12.6)
0 0 33

If E is polarized along such a principal axis, E is parallel to D.

 Light polarized along
any of these principal axes will propagate with an index of refraction given by n i ,
where i−1 = 1/n i2 .

12.2.2 Index Ellipsoid

We can capture the directional dependence of i−1 by using the index ellipsoid, which
is a geometric construct that allows one to find the indices of refraction and corre-
sponding polarization directions for an arbitrary direction of propagation (Saleh et al.
1991). This ellipse, shown in Fig. 12.4, is obtained from (12.4), Maxwell’s equation,
and the definition of the Poynting vector in the medium. Given a direction of propa-
gation, one uses the index ellipsoid to find the indices of refraction by determining
the ellipsoid’s semi-major and -minor axes. As shown in Fig. 12.3, these quantities
are obtained from the index ellipsoid and the plane normal to the direction of propa-
gation while passing through the center of the ellipsoid. The directions of these axes
define the normal modes of polarization for the directions corresponding to those
indices since a wave polarized along a principal axis will remain along that axis as
it propagates through the crystal. The index ellipsoid is defined as
12.2 An Overview of Electro-Optic Modulation 267

Fig. 12.4 An example of the index ellipsoid. The meaning of the index ellipsoid with dashed lines
will become clear in the following section, which describes the change of refractive index due to
an applied electrical field, the so-called electro-optical effect

Fig. 12.5 Demonstration of how to read out the indices of refraction from an index ellipsoid

i j xi x j ≡ 1 (12.7)

where again a sum over repeated indices is implied (Fig. 12.5).

For zincblende-type materials, such as GaAs (cubic 43 m), the crystal is isotropic.
In this case, the index ellipsoid in the absence of an applied voltage is

x 2 + y2 + z2
=1 (12.8)
n 20
268 12 Electro-Optical Modulators

Since the crystal is isotropic, the principal axes can be chosen arbitrarily; see Fig. 12.3.
However, in general, other crystal types or symmetries are not isotropic in index.
Thus, if x̂, ŷ, and ẑ are then principal axes, the index ellipsoid is then written as

x2 y2 z2
+ + =1 (12.9)
n 2x n 2y n 2z

For example, consider the case of LiNbO3 (triagonal 3 m); for this crystal, n x =
n y = n z , which is the condition for a uniaxial crystal. Typically, its ellipsoid is written
as
x2 y2 z2
2
+ 2 + 2 =1 (12.10)
n0 n0 ne

where n o and n e are termed the ordinary and extraordinary indices of refraction,
respectively. Thus, propagating along two principal axes, a lightwave experiences
the same index of refraction, while in the third, the index is different; see Fig. 12.3.
LiNbO3 has only single optical axis and with n o > n e , i.e., LiNbO3 is a negative
uniaxial crystal.

12.2.3 Dependence of the Electro-Optic Effect on Crystal

Symmetry

In the presence of an external field, E0 , the electro-optic effect changes the dielectric
tensor of a crystal in the presence of an external field, E0 . This change can be
conveniently expressed in terms of the index ellipsoid based on the changed principal
axes, which includes the fact that an electric field is present,
 
  
1 1 1
+ r E
1k k
0
x 2
+ + r E
2k k
0
y 2
+ + r 0
3k k z
E 2
n 2x n 2y n 2z
+2yzr4k E k0 + 2zxr5k E k0 + 2x yr6k E k0 = 1 (12.11)

where a sum over k = x, y, z is implied, and where Ek◦ = (E x0 , E y0 , E z0 ) is the applied

electric field. Note that at typical values of the electric field, the field-dependent terms
are much, much smaller than the static values, thus the field-induced changes can be
treated as perturbations. More generally, the electro-optic effect can be expressed as

δi−1
j = ri jk E k
0
(12.12)

with a sum over the repeated index k, then

δi j = −i  j ri jk E k0 (12.13)

12.2 An Overview of Electro-Optic Modulation 269

Fig. 12.6 A sketch of the GaAs crystal axes

Table 12.1 Optical electro-optical constants of LiNbO3 and GaAs

Crystals Refractive index Electro-optical Dielectric Transparency
coefficient constant window
LiNbO3 n = 2.286 r33 = 30.8 ⊥ = 43
(3 m) n = 2.269 r13 = 8.6  = 28
n = 2.237 r22 = 3.4 0.4−5 µm
n = 2.157 r51 = 28
(λ = 0.63 µm) (λ = 1 µm)
GaAs n = 3.5 r41 = 1.2  = 13.2 0.9 µm
(43 m) (λ = 0.9 µm) (λ = 0.9 µm)

The tensor, ri j , in (12.13), which gives the material response, can be redefined
by the noting symmetry of the indices i j = 11, 22, 33, 23(32), 13(31), 12(21) =
1, 2, 3, 4, 5, 6. Then setting i = 1, . . . , 6.
⎛ ⎞
r11 r12 r13
⎜ r21 r22 r23 ⎟
⎜ ⎟
⎜ r31 r32 r33 ⎟
rik = ⎜
⎜ r41
⎟ (12.14)
⎜ r42 r43 ⎟
⎟
⎝ r51 r52 r53 ⎠
r61 r62 r63

with 1 ≡ x, 2 ≡ y, and 3 ≡ z. This tensor is called the electro-optic tensor and its
form depends on the symmetry group, to which the crystal belongs. A tabulation of
the electro-optical tensor elements for the most common electro-optic materials is
given in Table 12.1. For example, one common class of electro-optical materials is
that of the zincblende semiconductor crystals (GaAs, CdTe, etc.), which are of the
cubic 43 m. In this case,
270 12 Electro-Optical Modulators
⎛ ⎞
0 0 0
⎜ 0 0 0 ⎟
⎜ ⎟
⎜ 0 0 0 ⎟
rik = ⎜
⎜ r41
⎟ (12.15)
⎜ 0 0 ⎟⎟
⎝ 0 r41 0 ⎠
0 0 r41

with x̂ = [100], ŷ = [010], and ẑ = [001]; see Fig. 12.6 for a sketch of the GaAs
crystal axes.
Mechanical stress in electro-optical crystals, due either to lattice mismatch
between cladding and waveguide epilayers or to the presence of electrodes, can
affect the electro-optical properties of that crystal. This stress reduces the symmetry
of the crystal. For example, for GaAs, stress along [0, 1, 1] causes the crystal to shift
from cubic (43 m) to orthorhombic (2 mm), while for stress along [1, 0, 0], the sym-
metry shifts into yet another symmetry tensor. Since different symmetry groups have
different forms for the electro-optic tensor, rik , the appropriate electro-optic tensor
should be used for each orientation of the stress field. For example, a GaAs crystal
which is stressed (uniaxial) along the [0, 1, 1] direction belongs to the orthorhombic
(2 mm) symmetry group, with
⎛ ⎞
0 0 r13
⎜ 0 0 r23 ⎟
⎜ ⎟
⎜ 0 0 r33 ⎟
rik = ⎜
⎜ 0
⎟ (12.16)
⎜ r42 0 ⎟⎟
⎝ r51 0 0 ⎠
0 0 0

with x̂ = [0, 1, 1], ŷ = [0, 1, 1], and ẑ = [1, 0, 0].

A second common class of electro-optical materials is that of the ferroelectric
or ferroelectric-like metal oxides, with LiNbO3 being the best example. LiNbO3 , as
mentioned above, is a negative uniaxial crystal containing ordinary n o and extraor-
dinary n e indices of refraction, i.e., n x = n y = n o , and n o > n e . In such a crystal,
ẑ is thus the optical axis, since light propagating along ẑ experiences an index of
refraction of n o , irrespective of the transverse polarization of the lightwave. The
electro-optic tensor for this crystal class, trigonal 3 m, is
⎛ ⎞
0 −r22 r13
⎜ 0 r22 r13 ⎟
⎜ ⎟
⎜ 0 0 r33 ⎟
⎜ ⎟ (12.17)
⎜ 0 r51 0 ⎟
⎜ ⎟
⎝ r51 0 0 ⎠
−r22 0 0

All three of the above examples show that typically, symmetry causes the electro-
optic tensor to be “sparse,” and many of the matrix elements are identical or zero.
12.2 An Overview of Electro-Optic Modulation 271

Thus, the calculation of the electro-optical shift can be far less formidable than the
full tensor in (12.14) would suggest at first glance.
With the above review in hand, this chapter will now concentrate on a discussion
of electro-optic modulation of guided waves. Two materials are emphasized, GaAs
and LiNbO3 ; the former is meant to be the only representative example of the III–V
crystals, which are as suggested above broadly used for different sets of electro-
optical devices.

12.3 Electro-Optic Modulation Illustrated for Two Crystal

Types

Consideration of (12.11) shows that the application of external electric fields can
have two effects on the propagation of light in an electro-optical crystal (Saleh et al.
1991). First, the field can change the refractive indices of the crystal and, second, the
field can alter the crystal’s principal axes. The two points are illustrated below using
our two material types.

12.3.1 III–V (Zincblende) Crystals

Consider an applied external electric field, E ◦ , along the [1, 0, 0] direction, which
is vertical in the sketch in Fig. 12.7 for a [1, 0, 0] crystal in unstressed zincblende
crystals. In this case, the new index ellipsoid is:

x 2 + y2 + z2
+ 2yzr41 E ◦ = 1 (12.18)
n 20

Note that the presence of a cross product term when the field is applied means that a
new set of principal axes must be chosen. These transformed axes will be designated
by a prime notation, e.g., y and z . The transformation in the alignment of the electric
field involves a rotation of the principal axes about the x-axis. The new principal axes
are in the [0, 1, 1] and [0, 1, 1] direction; see Fig. 12.7. These may be written

x = x (12.19)

1
y = √ (y − z ) (12.20)
2

1
z = √ (y + z ) (12.21)
2
272 12 Electro-Optical Modulators

(a) (b)

Fig. 12.7 a After applying an external electrical field along the [1, 0, 0] direction, the principle
axes x̂ and ŷ transform to x̂ and ŷ . b A sketch of the zincblende crystal

The new index ellipsoid then becomes

   
x 2 1 1
+ y 2 + r41 E + z
0 2
− r41 E = 1
0
(12.22)
n 2o n 20 n 20

Since generally 1/n 2o

r41 E ◦ , that is, the electro-optical index change is small, the
equation may be simplified using a series approximation
 −2
1 1
n 0 ∓ n 30 r41 E 0  2 ± r41 E 0 (12.23)
2 n0

and thus,
x 2 y 2 z 2
+ + =1 (12.24)
n 20 (n 0 − 21 n 30 r41 E 0 )2 (n 0 + 21 n 30 r41 E 0 )2

With the new principal axis, for light polarized at 45◦ to these axes and in the
plane of these axes, the component along [1, 0, 0] has index n 0 , while the component
along [0, 1, 0] has index n 0 − 1/2n 30 r41 E 0 . Notice that for GaAs, and other III–V
materials, the change in the orientation of the principal axes occurs in the presence of
any measurable field since it breaks the symmetry of an otherwise isotropic crystal.
Recall that using the intensity modulation scheme discussed earlier, the intensity,
I , of an optical wave varies with the distance, L, in an electro-optical crystal as
π 
I ∼ cos2 (n x − n z )L (12.25)
λ
Thus, for GaAs, the indices in (12.24) give
π 
I = cos2 n 30 r41 E 0 L (12.26)
2λ
12.3 Electro-Optic Modulation Illustrated for Two Crystal Types 273

Fig. 12.8 One commonly

used external-field
orientations for LiNbO3 .
z-axis is the optical axis, the
light is propagating along
x-direction. The external
field is along z-axis

As an example, if λ = 1.5 µm, tabulated data show that n 0 ∼ 3.35, and r41 ∼ 1.3 ×
10−12 m/V. Thus, for a device having 5 V applied across a 3 µm gap between two
electrodes, E 0 = 1.66 × 106 V/m, and L = 5 mm; then the argument in (12.26),
(π/2λ)n 30 r41 E 0 L, will be 0.43 rad = 24.4◦ , i.e., a very sizable phase shift.
In the above case, an applied E-field along [1, 0, 0] gives principal axes along
[1, 0, 0] and [0, 1, 1]. By contrast, an applied E-field along [0, 1, 1] will give prin-
cipal axes with indices n 0 + (1/2)n 30 r41 E 0 and n 0 − (1/2)n 30 r41 E 0 . For this crystal
orientation, one must launch polarized light along [0, 1, 1] or [1, 0, 0] to get modu-
lation.

12.3.2 LiNbO3 Crystals

A similar analysis as applied above can be applied to LiNbO3 and related crystals.
However, here for the sake of brevity, two specific, common external electric field
orientations will be examined, and only the final formulae will be stated here. In
the first case, shown in Fig. 12.8, the external field is aligned along the z-axis, i.e.,
only E z0 = 0 and the z-axis is the optical axis of the crystal. Notice also that the
direction of propagation of the light beam is along the x-axis. After the application
of an external field, the index ellipsoid becomes
   
1 1
+ r 13 E 0
z (x 2
+ y 2
) + r 33 E z z =1
0 2
(12.27)
n 2o n 2e

For small fields, this equation can be written in a way that shows that such an external
field retains the “field-free” principal axes but changes the magnitude of the indices
of refraction; that is,
274 12 Electro-Optical Modulators

x 2 + y2 z2
+ =1 (12.28)
n 2o (1 − 2 r13 n 20 E z0 ) n 2e (1 − 2 r33 n 2e E z0 )
1 1

The second term in both parentheses is the field-induced change in the indices. This
electro-optic orientation causes only a change in the index, but not in the principal
axes; it can thus be used to modulate the phase of the light beam.
The second case involves aligning the external field normal to the optical axis,
along the y-axis. Then,
     
1 1 1
− r22 E y x +
0 2
+ r22 E y y +
0 2
z 2 + 2r51 E y0 yz = 1 (12.29)
n 2o n 2o n 2e

This orientation causes a new set of principal axes to be formed; they are oriented
by a rotation around the x-axis and have indices
 
1 1 1
n = n o − n 3o r22 E y0 + n 3o (r51 E y0 )2
y − 2 (12.30)
2 n 2e no
 
1 1
n z = n e − n 3e (r51 E y0 )2 2
− 2 (12.31)
ne no

While these equations allow significant manipulation, in fact, all the terms con-
taining the electric field are very small and to a very good approximation:

n y ≈ n0 (12.32)

n z ≈ n e (12.33)

Thus, the primary effect of the external field is simply to rotate the principle axis.
This rotation introduces a small off-diagonal element in the dielectric tensor:

δ23 = −n 2e n 20 r51 E y0 (12.34)

Since it is off-diagonal, it couples the E z to E y fields. Hence, in this case, the appli-
cation of the external field modulates the light through polarization by coupling the
TE and TM modes. To obtain the modulation of the lightwave requires the use of
a polarizer after the beam exits the crystal. Further, since r51 is as large as r33 , the
extended field orientation can be used to make an efficient modulator.
12.4 Electro-Optic Modulators 275

12.4 Electro-Optic Modulators

In this section, the device aspects of electro-optic modulation in waveguide devices

will be discussed. Thus, the section will introduce the most common types of mod-
ulators and discuss some of the more important design parameters of modulators,
including electric and optical field overlap and the configuration of the modulator
⎡ ⎤⎡ ⎤
0 −r22 r13 0 0 0
⎢ 0 r22 r13 ⎥ ⎢ 0 0 0 ⎥
⎢ ⎥⎢ ⎥
⎢ 0 0 r33 ⎥ ⎢ 0 ⎥
⎢ ⎥⎢ 0 0 ⎥
⎢ 0 r51 0 ⎥⎢
⎥ 0 ⎥
⎢ ⎢ r41 0 ⎥
⎣ r51 0 0 ⎦⎣ 0 r41 0 ⎦
−r22 0 0 0 0 r41

LiNbO3 GaAs
T rigonal(3 m) Cubic(43 m)

electrodes. Our discussion will be presented in terms of LiNbO3 modulators, chiefly

because LiNbO3 devices are conceptually the simplest. Lithium niobate is also the
most widely engineered material for modulators because of its high performance
and low optical coupling and insertion loss; although other similar materials, such
as the tantalates (Kaminow 1965) also have important advantages. Finally, optical
switches are closely related in function to modulators; however, this topic is suf-
ficiently large and the device requirements are sufficiently different that it will be
discussed Chap. 13.

12.4.1 Layout of a Basic Modulator

Modulators are fabricated by placing electrodes on the surface of an electro-optical

waveguide structure. The design of these electrodes and the waveguide structure
for modulation can be sophisticated. However, there are a set of common parame-
ters, which apply to all types. Since the quantities, which characterize the electrode
arrangements, are important for our subsequent discussion, we will briefly describe
these fundamental aspects first. The layout of electrodes is critical since it determines
the modulator voltage and the efficiency of interaction between the applied field and
the optical beam. As we will see later in this chapter, electrode design can be differ-
ent significantly for high-speed modulators; see Sect. 12.4.3. Electrode placement in
LiNbO3 modulators is made easy by the fact that the crystal is an insulating dielec-
tric. Thus, any free charge present in semiconductor materials makes their design
much more complicated.
The two most common electrode configurations for a LiNbO3 modulator are
shown in Fig. 12.9. Roughly speaking, the electric field in the waveguide region
276 12 Electro-Optical Modulators

Fig. 12.9 Upper: The Electrodes

configuration for an external Waveguide
field that is horizontal E  L
Bottom: Two common
W
electrode and electric field
configurations for the
LiNbO3 modulator G

V
R

varies as V ◦ /d, where d is the electrode spacing and V ◦ is the applied voltage. In
fact, however, for a realistic waveguide design, the actual electric field distribution is
more complex and varies in amplitude and direction between the electrodes. Further,
the magnitude of the electro-optic effect depends on the overlap of the external field
distribution over the cross-sections of the optical field. As a result, the idealized field,
V ◦ /d, is modified by an empirical factor, , which accounts for the averaged overlap
between the optical and external electrical fields. As an example, consider the case of
an optical beam in y-cut LiNbO3 , which is aligned such that it propagates along the
y-axis and is polarized in the z-direction. If the applied electric field is dominantly
along the z-axis, then
 
1 V◦
n = n 3e r33  (12.35)
2 d

Then the overlap factor  is obtained from the simple but generally nonanalytic
integral:  0
d E (x, y)E ∗ dA
= ◦  (12.36)
V E 2 dA

where E ◦ (x, y) is the applied field, E is the optical field, and the integral is averaged
over the cross-sectional area. The factor of d/V ◦ in (12.36) normalizes the magnitude
of the applied field.
The electrode configuration for a modulator is often dependent on the practical
constraints of the device, as well as the desire to make use of the largest element in
the electro-optical tensor. Thus, a Ti-diffused LiNbO3 waveguide modulator in an
x- or y-cut crystal, such as that shown in Fig. 12.9, can make use of the large r33
coefficient in this material if electrodes are placed on either side of the waveguides
12.4 Electro-Optic Modulators 277

so as to have a dominantly horizontal field. Further, the fact that the electrode is
placed away from the waveguide such that optical losses in the metal electrodes are
negligible. The disadvantage of this placement is that a relatively large electrode
spacing is incurred thus increasing the voltage requirement. Turning now to z-cut
material, a vertical field must be achieved. This can be done by placing one electrode
directly on the waveguide and positioning two ground electrodes adjacent to the
waveguide. This approach requires a transparent dielectric under the metal electrodes
to reduce electrode loss; however, this layer also lowers the overlap integral. Despite
this reduction in overlap integral, the z-cut configuration is more commonly used
in practice. Finally, electrode design becomes more complex when designing for
high-speed modulators. For example, the width of the electrode, ω, is important in
setting the impedance and, hence, the bandwidth of the modulator; see Sect. 12.4.3.
Some of the issues in this case will be discussed below.

12.4.2 Modulator Types

12.4.2.1 Phase Modulation

Modulation of the phase of an optical beam is readily accomplished in waveguide

modulators. A simple phase modulator is also one of the building blocks for making
other forms of modulators. For example, when a phase modulator is inserted into the
arm of a Mach–Zehnder (MZ) interferometer, the resultant device forms one of the
most widely used intensity modulators.
Phase modulators operate by changing only the phase of the propagating guided-
light wave. If the LiNbO3 is y-cut and the electric field is aligned along the z-axis
(or optical axis), then the induced phase is
 
1
φz = kx n e − n 3e r33 E z0 (12.37)
2

where the propagation direction is along x. A similar expression is obtained if the

wafer is x-cut, z polarized, and propagating along y, except for the exchange of x
and y in (12.37).
If a phase modulator requires that β L = pπ, where the value of p depends on the
exact type of modulator (Alferness 1982), then (12.35) shows that the voltage–length
product for a modulator can be written as

pλd
VL = (12.38)
n 3r 
where the symbols are as described above. This equation suggests that to improve the
voltage–length product of the device and thus minimize device power and device size
(and cost), one should improve both the overlap integral, , and reduce the electrode
278 12 Electro-Optical Modulators

Fig. 12.10 Pairs of +

interferometeric modulators
are operated to cancel out the
harmonics of the usual
nonlinearities inherent in the
sinusoidal response

gap, d. Of course, these factors also control other performance considerations in the
design, including parameters such as insertion loss and bandwidth (see below). Each
of these parameters will be discussed below.
Finally, if the modulating electric field having an angular frequency ωm is denoted
as
E z0 ≡ E zm
0
sin ωm t (12.39)

then the modulated optical field will be

E z = E zi cos(ωt − φ0 + δm sin ωm t) (12.40)

where the depth of modulation, δm , is given by

π 
δm = n 3e r33 E zm
0
(12.41)
λ

φ0 = 2πkn e x, and E zi is the amplitude of the z-polarized optical field on the modula-
tor. The argument for E z can be expanded in terms of Bessel functions; the amplitude
of the first sideband is then given by

J1 (δm )/J0 (δm ) (12.42)

12.4.2.2 Interferometric Modulators

The MZ modulator, shown in Fig. 12.10 is one of the most important commer-
cial LiNbO3 waveguide modulators. In this modulator, a phase shifter is typically
12.4 Electro-Optic Modulators 279

mounted in each arm. This layout allows a positive phase shift in one arm, which is
balanced by the negative of exactly the same phase shift in the other, causing push–
pull mode modulation. In this case, the total phase shift, 2φ, between the two arms
is obtained from (12.35) to be thus

n 3e r33 L V
2φ = 2π (12.43)
λd
where L is the total electrode length and the same field directions are used as for
(12.35). Based on the value for V which sets the total phase shift between the arms
of π, we can define the quantity, Vπ , for this modulator:

λd
Vπ = (12.44)
2n 3e r33 L

Note that Vπ is an important operational parameter for modulators (see discussion in

a section below) and is dependent on both material and device parameters. It scales
inversely with length, but other factors such as bandwidth can also limit the full use
of this quantity.
In the instances where the modulators are used for analog applications, it is often
crucial to eliminate any nonlinearity in the modulator response. In one important case,
pairs of interferometric modulators are often operated so as to cancel out many of the
harmonics of the usual nonlinearities inherent in the sinusoidal response of a typical
MZ device. In this operational mode, they are also operated around the inflection
point of the sinusoid (see Fig. 12.10) and both optical and electrical parameters are
adjusted to effect the cancellation (Bridges and Schaffner 1995). Unfortunately, these
schemes often come with a cost of reduced bandwidth.
Interferometric modulators can also be used as switches; this type of operation
will be discussed in the subsequent chapter on switches. One important fact regarding
their use as switches
√ is that in comparison to directional coupler-based switches, they
have a factor of 3 lower voltage–length product.

12.4.2.3 Resonant-Cavity Modulators: Operating Principles and

Voltage Dependence

Since the phase shift of an interferometric modulator scales with its path length, it
is of interest to consider extending its “effective” length by adding a high-quality
Q-optical cavity. The high Q in the device also makes it possible for small input
fields to build up within the device. However, such a cavity also prolongs the photon
interaction time by trapping the light in the structure due to the use of resonance (e.g.,
a cavity with two high-reflectivity mirrors or a circular ring, etc.). Further, since the
Vπ of a ring modulator is inversely proportional to its length, use of a resonant cavity
would also reduce Vπ (Gheorma et al. 2000). Recently and importantly, these same
considerations have been shown to be operable and, in fact, highly developed in Si
280 12 Electro-Optical Modulators

Fig. 12.11 Ring and Fabry–Perot resonators. The principle of a ring resonator (both a single-bus
ring and an add-drop ring) can be illustrated by Fabry–Perot cavity

resonant rings (Xu et al. 2005). Due to the different types of resonant modulators, it
is important to develop an analysis, which allows comparison of the different types.
This analysis appears below.
Resonant-cavity modulators, in the form of Fabry–Perot (FP) cavities, were among
the first modulators used in light modulation experiments. Resonant devices are
attractive because they can have an extremely low Vπ as the signal voltage is varied
around a resonant voltage; however, as mentioned above, a low Vπ comes with a
reduced bandwidth. In addition, Si ring modulators also can have very low drive
power due to their small diameter and thin Si film thickness. This attribute is crucial
for data PICs in many data system applications.
The basic principle of such a resonator-based device is best illustrated by the
FP modulator (see Fig. 12.11), which is identical in function to an FP filter or laser
optical cavity. The modulator consists of an etalon of electro-optical material which
is highly reflective coated on both sides. The relative intensity of the transmitted light
through an FP cavity, fabricated from an electro-optic material, as a function of the
modulating voltage, V , is

I (V ) 1
= (12.45)
I0 1 + (4F /π ) sin (−πn 2 r L V /λd)
2 2 2

√
where F = π R/(1 − R) is the finesse of the cavity, L the length of the cavity,
d the distance between the electrodes,  the RF optical field overlap coefficient, r
the electro-optical coefficient, and λ the free-space wavelength. Note that the output
in (12.45) is not zero due to the finite F; thus, we cannot define a Vπ voltage in
12.4 Electro-Optic Modulators 281

the same manner as is used, for example, for MZ or cross-polarizer modulators.

Instead, an alternate definition can be employed as is, sometimes used for other
types of modulators. This definition is based on the slope of the variation of the
optical intensity with respect to voltage for the modulator at the half-transmission
point, i.e., the most efficient modulation bias point. This slope-based definition gives
a lower limit of the voltage either for digital or analog applications.
For the FP modulator, with a unit input intensity
√
dI F P πn 3r R L
≈ (12.46)
dV λ(1 − R)g

dI M Z
In comparison, the analogous quantity for an MZ modulator is |1/2 = −(π/2Vπ ).
dV
We can redefine VπM Z for an MZ using its slope
 −1
π dI M Z
VπM Z = |1/2 (12.47)
2 dV

This expression also allows one to define an equivalent VπF P for an FP modulator by
substituting ( dIdVF P )|1/2 to give

π 1 g λ(1 − R)
VπF P = = √ (12.48)
2 dI F P /dV |1/2 L 2n 3r  R

This definition makes it easy to compare the MZ modulator with an FP device.

Traveling-wave optical resonators (TWR) including ring, race-track, disk, or
sphere resonators, have also been considered for integrated optic applications. In
fact, an equivalence between the standing and traveling-wave resonator modulators
can readily be shown using the coupled-mode approach (see Gheorma 2002). An
example of this set of equations is for a reflection FP modulator, which appears in
the equations below with dielectric mirrors, within the stop band of the mirrors (also,
note that the indexes on both sides of the mirror are assumed equal,
    
 b1   t F P iκ   a1 
 =   (12.49)
 b2   iκ t F P   a2 

where r is the mirror reflectivity and t F P is the mirror transmission. In both of these
equations, b2 and a2 can, with generality, be related by a2 = Ae+θ b2 . A = e−αd + · · ·
represents the roundtrip return (including the attenuation due to transmission through
the second mirror for an FP), is the roundtrip phase change, and is an equivalent
exponential loss per pass.
282 12 Electro-Optical Modulators

12.4.2.4 Bandwidth

Device bandwidth is an important quality for most applications of electro-optical

modulators. It is limited in practice by the magnitude of the electro-optical effect,
by the lumped circuit elements of the modulator, and, at very high speeds, by phase
mismatch between the electrical and the optical waves.
In order to determine the limitation on the bandwidth of the device, it is necessary
to divide modulators, and their electrodes, into two classes. In the first, the electrode
length is much smaller than the wavelength of the applied field. Thus the harmonic
period of the field is much longer than the optical transit time, and the electrode is
describable by a lumped circuit model. In the second, the conditions are the very
opposite, and field propagation, or traveling-wave, effects must be considered in
the modulator design. A key difference between the two electrode types is that the
traveling wave type is fed by a transmission line and, hence, requires a characteristic
impedance termination at the other end. Both types require consideration of the
capacitance per length of the device.
For a dielectric waveguide, such as LiNbO3 , the key question of capacitance can
be conveniently found by numerical techniques or by conformal mapping. A plot of
the capacitance per length versus d/w, where w is the width of the electrode and d is
again the distance between the two electrodes, obtained by conformal mapping for
three common electrode types used in RF-driven systems is given in Fig. 12.12. In
all cases, the capacitance per length, C/L, is found to be proportional to eff , where
eff is an effective or average dielectric constant for the electrode structure and is
independent of length. Because of the importance of eff , it is clear that low dielectric
materials, such as polymers, should and do yield much faster modulators (Chen et al.
1997). In addition, the figure also shows that C/L scales approximately as (d/w)−1 .
Thus, as suggested in the discussion above, scaling down of the electrode area can
significantly increase the device capacitance and, hence, decrease its bandwidth.
For the first type of modulator, i.e., the lumped-circuit type, the highest frequency
response is, first, ensured by terminating the high-voltage electrode with 50  resistor.
The frequency response or bandwidth, ν, is then governed by the usual RC circuit
limitation, ν = 1/π RC. Because of this relation and the inverse dependence of
device capacitance on length, ν ∝ 1/L, for a fixed 50  termination resistance. As
a result, the product of bandwidth × length, such as that also shown in Fig. 12.12
provides one figure of merit for the modulator. For a particular modulator structure,
this value is adjusted by varying the electrode geometry through the ratio d/w.
Regardless of how low the capacitance is, however, ultimately the frequency cutoff
for the modulator is determined by the transit time, νt . This transit time limitation is
given by
c
νt = √ (12.50)
π eff L

For LiNbO3 , eff is such that

νt ≈ 2 GHz · cm (12.51)
12.4 Electro-Optic Modulators 283

(a) (b) d

RC bandwidth. length, Δf . L(GHz . cm)

w
Capacitance/length, C/L(pf/cm)

9 0 Waveguide
LiNbO3
8
0.5
7 d
1
w
6
5
1.5
4
3 2 d
w
2 2.5
1
0 0.5 1 1.5 2 2.5

Fig. 12.12 a A plot of the capacitance per length and RC bandwidth times length vs. d/w. The
figure shows that the capacitance per length decreases as d/w increases, proportional to (d/w)−1 .
b A sketch of three common used electrode configuration, corresponding to the curves in (a)

At high frequencies, the electrode structure must be considered to be part of the

transmission line. In this case, the bandwidth limitation of traveling modulators is
typically velocity mismatch between the microwave and the optical signals. Analysis
of the traveling wave problem shows that the mismatch between the optical and
o m
microwave effective index, Neff and Neff , respectively, controls the mismatch velocity
and leads to the following limitation on the useful bandwidth, ν:
 
2 c N o −1
ν L ≈ 1 − eff (12.52)
π Neff
m m
Neff

For LiNbO3 , (Neff

o
)/(Neff
m
) ≈ 0.5 and, thus, ν L ≈ 9.6 GHz · cm. Other effects, such
as voltage drop along the electrodes, due to microwave absorption, are also important.
The use of thick electro chemically deposited copper helps alleviate this problem
(Noguchi et al. 1995).
Consideration of (12.38) and (12.55) yields insight into a practical aspect of high-
speed dielectric modulators. Specifically, these equations show that both modulation
voltage and bandwidth scale as 1/L, where again we neglect any loss in the modulator
electrodes is neglected. As a result, there is an important tradeoff involving the device
length and the modulator bandwidth and voltage. Specifically, low Vπ can be achieved
by increasing the device length. However, this comes at the cost of lower bandwidth.
Similarly, high bandwidth and Vπ are found for a short device.
In fact, one figure of merit for characterizing a modulator is the ratio of Vπ to
bandwidth Vπ /ν, since it eliminates the scaling with length. In its most general
form, this ratio is a complex relationship involving the device geometric factors,
the microwave and optical dielectric constants, and the modal overlap function; see
the discussion by Alferness (1982) for details. As a particularly simple example,
284 12 Electro-Optical Modulators

consider the case of a traveling-wave modulator. In that case,

  m
Vπ pπλ d Neff − Neff
o
= (12.53)
ν 2c  n 3r

where again electrode loss has been neglected. This equation gives the basic scaling
laws for the figure of merit of the voltage/bandwidth ratio. A high-performance modu-
lator will minimize this ratio, since a low Vπ for high bandwidth is then desired. Thus,
matching of the effective index, increasing of the overlap integral, , and decreas-
ing of the electrode gap, d, all decrease the V /ν ratio. For the case of LiNbO3 ,
values of ∼0.3 V/GHz are typical in standard traveling-wave commercial modu-
lators. This value may be compared to V /ν ≈ 2 for lumped circuit modulators.
Unfortunately, (12.53) ignores another equally important aspect of modulator design:
electrode impedance, namely to eliminate unwanted voltage reflections from the
electrodes, traveling-wave modulators should have 50  characteristic impedance.
To achieve this electrode impedance often requires a modulator design, in which
other factors, e.g., mode overlap, are compromised. As a result, achieving the best
performance comes from a series of interrelated tradeoffs. Nonetheless, extremely
high-performance devices have been made by paying close attention to the most
crucial design limitations (Noguchi et al. 1998, 1995).

12.4.2.5 Very High-Speed Modulators

Careful materials and structural design can result in much higher speed MZ modu-
lators, that is those with bandwidths of 50−100 GHz. These modulators have been
made in LiNbO3 and in polymer materials; see also Sect. 12.5 for very high-speed
electro-absorption (EA) modulators in semiconductors.
A sketch of one such high-speed LiNbO3 modulator is shown in Fig. 12.13. This
MZ modulator has a 75 GHz 3 dB bandwidth at 1.5 µm (Noguchi et al. 1998).
The LiNbO3 crystal is z-cut so that the maximum electro-optical coefficient, r33 ,
is achieved with a vertical applied electric field. To prevent optical loss from the
waveguide into the electrodes, a “buffer” layer of SiO2 is applied between the elec-
trodes and the LiNbO3 .
This device uses several design improvements to achieve its high speed. As alluded
to above, the central design issue is to eliminate the index mismatch between the
optical and microwave effective index. Recall that in LiNbO3 , wave matching is
particularly serious because of the different values of the dielectric constant at these
two frequencies. The problem can be reduced by lowering the microwave effective
index by allowing more of the waveguide to be exposed to the air. This allows
more of the microwave field to leak into the air region, giving a lower effect index.
This effect is accomplished by forming an etched rib. In addition, the electrode
thickness is increased and adjusted so that more of the microwave field is in the metal,
thus reducing loss and inter-electrode capacitance. These two effects are shown in
Fig. 12.14. However, notice that the increase in bandwidth in this device comes with a
12.4 Electro-Optic Modulators 285

CPW Electrode
Ti-diffused Waveguide

Buffer Layer

CPW Electrode

Buffer Layer
z-cut LiNb O3
Substrate

Ti-diffused Waveguide

Fig. 12.13 An example of the high-speed LiNbO3 modulator, with a bandwidth of 75 GHz at
1.5 µm [Adapted from (Noguchi et al. 1998)]

price: the overlap between the optical and the microwave field is reduced by forcing
the microwave field into the air region due to the etched rib. The Vπ L product
of this modulator ∼10.2 V · cm is relatively high (Noguchi et al. 1998). The optical
bandwidth is 110 GHz; the value ofVπ ν ≈ 0.5 VGHz for a 2 cm-long device. Other
approaches such as the use of thin LiNbO3 slabs can potentially give high bandwidth
plus low voltage (Gheorma et al. 2000).
Of course, it is also possible to achieve closer velocity matching of the microwave
and optical waves by using a material with comparable, and typically lower, dielectric
constants at both optical and microwave frequencies, specifically polymers. There
have been several recent demonstrations of very high modulation bandwidth in MZ
modulators fabricated of low dielectric constants using polymer materials. The poly-
mer used in these devices was designed to have a low optical loss, which can otherwise
be significant at telecommunications wavelengths. The device was demonstrated at
1.3 µm and had Vπ ≈ 2.2 V · cm and a bandwidth of 110 GHz. The velocity matching
286 12 Electro-Optical Modulators

Fig. 12.14 A plot for the

optical 3-dB modulation 200
bandwidth versus ridge
height (at different electrode

Bandwidth [GHz]
thickness). A higher 100
bandwidth can be achieved
when the ridge is higher and
T = 10μm
the electrode is thicker 50
T = 5μm
20

10
0 0.5 1 1.5 2 2.5 3 3.5 4
Ridge Heigth [ μ m]

in this and in other polymer modulators is very good; numbers such as 5% velocity
mismatch are feasible (Dagli 1999; Chen et al. 1997). Additional research on impor-
tant engineering issues such as optical damage is now being done to make these
devices commercially viable.

12.4.2.6 A Second Look at the Speed of a Resonant Modulator

For an FP modulator, based on an electro-optical index change, there is a tradeoff

between bandwidth and voltage. Thus, the ratio of bandwidth/voltage is an even
more useful figure of merit for an electro-optic modulator. For an electro-optical FP
modulator, this ratio can be obtained from

f 3FdB
P
(cl) cn 2 f 
eq = 0.643 (12.54)
Vπ 2πλg

Note that the modulator length disappears in this ratio, and the ratio is also indepen-
dent of cavity Q or finesse (Gheorma et al. 2000).
We can compare a figure of merit to that for MZ modulators, both lumped-element
and traveling-wave configurations. For example, neglecting the RC constant, lumped-
element MZ modulators have the following figure of merit:

f 3FdB
P
(cl) 2.8cn 2 f r 
= (12.55)
Vπ πλg

For the same material and same electrode gap, this ratio is a factor ≥ 8 larger for the
MZ modulator than for the electro-optic FP modulators.
12.4 Electro-Optic Modulators 287

Fig. 12.15 A sketch of the

three-layer semiconductor
modulator V1 V2

P+
GaAs P+ P+ AlGaAs
n - GaAs
n+ GaAs
n+ GaAs

12.4.2.7 Semiconductor Electro-optical Modulators

Thus far, we have focused on the case of LiNbO3 modulators. However, modulation
in semiconductor materials is also important (Alferness et al. 2013). In semiconduc-
tors, thus, the existence of free carriers in semiconductor modulators can make the
application of the voltage more complex to analyze, but more recently, semiconduc-
tors have been used to make a series of highly effective ultrasmall modulators, which
are having a major impact on complex, dense high-data rate systems (Xu et al. 2005).
Generally, a semiconductor modulator structure consists of a three-layer diode-like
structure, as is shown in Fig. 12.15. This structure must be fully depleted by a strong
reverse bias, or the resulting optical modulation will be due to both the electro-optical
effect and free carrier absorption. The presence of free carriers does allow electrical
control of modulation as will be discussed below. However, do note that in some
cases free carriers can also introduce additional optical and microwave loss. Finally,
note that the standard electro-optical effect for crystals and semiconductors is gov-
erned by a tensor as explained earlier in this chapter. Free carrier effects are not taken
into consideration in this chapter.
Typically, GaAs or other III–V based devices use (001) wafers. For a vertical,
applied field, and in the absence of carrier effects, the index change is then in the
[0, 1, 1]or [0, 1, 1] direction. It can be written as

1 3 V◦
n = n r41  (12.56)
2 d
The electro-optic tensor for GaAs, for example, is small, but the large index of
refraction causes the effective electro-optic response to be about 1/5 of that of lithium
niobate. Adding a rib waveguide helps localize the electric field as well as providing
lateral optical confinement. In this structure, the voltage is applied to reverse bias of
the Schottky or p+in junction. The presence of heavily doped regions which have
overlap with the optical mode introduces loss into the waveguide, e.g., 4 dB/cm
for GaAs. Introduction of a local stripe of undoped dielectric under the waveguide
region has been found to reduce this loss by a factor of ∼1/2. Simple Schottky
barriers have also been used for electrodes on semiconductor modulators. However,
288 12 Electro-Optical Modulators

the presence of a metallic surface contiguous to the optical mode introduces both
loss and polarization effects. This difficulty can be alleviated again by placing a
local dielectric directly over the waveguide but allowing the metal to contact the
guide laterally on either side of the stripe.
Traveling-wave modulators can also be made in III–V materials. As shown earlier,
a key factor in realizing a high-bandwidth modulator is to have Neff o
≈ Neff
m
, since the
o −1
bandwidth is proportional to (1 − Neff /Neff ) . At microwave frequencies, Neff
m m
≈
3.6, compared to Neff = 3.4; thus, the phase matching is relatively close. However,
o

in the presence of real electrodes, the effective dielectric constant for the microwave
signal is reduced since the field is not confined to the semiconductor and emerges
into the air. In fact, for coplanar strip lines, this gives rise to the following effect
0
index, Neff
 
1 + sc 1/2
0
Neff = (12.57)
2

where sc is the dielectric constant of the semiconductor; this phenomena leads to a
∼40% index mismatch (Dagli 1999). Despite this, velocity matching is better in III–
V materials and can be achieved by using slow-wave structure, done with periodic
capacitive loading, or by using “undoped” GaAs epitaxial layers in conjunction with
slow-wave electrodes with local transverse phase reversal. Because of its conceptual
simplicity, we will describe the latter approach in somewhat more detail. In particular,
low-doped epitaxial layers behave as an insulating dielectric film and thus do not
introduce microwave losses. Carefully engineered segmented microwave electrodes
can allow push–pull operation, good microwave-optical wave velocity matching,
and a ∼50  characteristic impedance. In one case, a 1 cm long modulator had
Vπ = 17 and a ∼50 GHz bandwidth at 1.55 µm. While these are impressive operating
parameters, they still are not as good as those in, say, LiNbO3 (Spickermann et al.
1996).

12.4.3 Free-Carrier Modulations: Ultrasmall Si

Semiconductor Modulators

Modulator device size is a key property of complex photonic chips for data manipu-
lation. Silicon is preferred in these applications due to the ease of using standard Si
patterning tools. If a resonant strongly light-confining structure is used, it is possible
to have the ring respond to small changes in the Si refractive index and still have
a high-speed operation. Under these conditions, the modulator diameter is much
smaller than the previous Si designs. Earlier work had shown that it is possible to
fabricate ultrasmall modulator devices in high-index materials, e.g., III–V compound
semiconductors (Sadagopan 2004). Similarly, all optical silicon ring resonators were
shown to work with an optical control beam Almeida et al. (2004), however, full use-
able modulator functionality requires having an electronically driven structure. Such
12.4 Electro-Optic Modulators 289

SiO2
Si

n+ doped p+ n+

V
p+ doped Ring Resonator

Waveguide

Input Output

Fig. 12.16 A sketch of a micro-ring Silicon EO modulator based on free-carrier effect.

electronic driving including attainment of high-speed operation has been demon-

strated by Liu et al. (2004) based on a CMOS patterned chip. However, high spatial
device resolution was not reached due to the weak optical confinement (Liu et al.
2004) and in fact only relatively long device lengths of mm could be demonstrated
(Xu et al. 2005).
Light-confining resonating structures and its attendant strong confinement can
enhance the effect of refractive index change on the transmission response (Heebner
et al. 2004; Almeida et al. 2004). In fact, using the resulting strong confinement
enables a silicon electro-optical modulator of a few micrometers in the size to be
fabricated. The modulator consists of a ring resonator coupled to a single waveguide,
with the waveguide transmission depending sensitivity on the particular signal wave-
length being used. In particular, the transmission is reduced at wavelengths, which
are multiples of the ring circumference (Little et al. 1998; Almeida et al. 2004).
Figure 12.16(inset) shows a cross section of a strip waveguide. The thickness of the
slab thickness (50-nm-thick Si layer) is much less than the wavelength in the 1.5 mm
mode profile and thus closely approximates that of the silicon strip waveguide. The
device tunes the effective index of the ring waveguide so as to strongly modulate the
transmitted signal. Index modulation is via electron and hole injection.

12.5 Semiconductor Electro-Absorption Modulators

EA modulators (Dagli 1999) are a very different modulator type. They operate by
using an applied voltage to modulate the absorption coefficient in a semiconductor
290 12 Electro-Optical Modulators

Fig. 12.17 A sketch of the P – electrode Polymide

typical EA modulator

P+ – InAlAs
P – InAlAs
AR Coating
i - MQW
n - InAlAs n –InP
n – electrode

sample. EA modulators typically use the physical structure illustrated in Fig. 12.17.
The optically active portion of the modulator is in a composite region containing an
intrinsic semiconductor region incorporating multiple quantum wells. The contact to
this region is through a p-type upper region and an n-type lower region. The complete
semiconductor region is etched into a channel waveguide as shown in the figure. The
refractive indices in the n and p regions are lower than in the intrinsic MQW region
and thus light is confined in the optically active region.

The mechanism for absorption in the quantum wells is via excitonic bands, which
have a relatively sharp narrow line. When an external voltage is applied to this struc-
ture, the hole- and an electron-wavefunction overlap is reduced and the absorption
feature is reduced in strength and broadened. In effect, optical absorption in this
material is controlled by the application of an external voltage. Figure 12.19 shows
the absorption versus wavelength for typical EA material. Note that in this case, two
excitonic features are present because of different electron–hole band combinations.
Several different InP and GaAs, quarternary or ternary alloys are used for EA mod-
ulators. For example, for λ = 1.55 µm GaAsP or GaAlAs quarternary alloys have
been used. In the case of ternary alloys, opposing strained-layer regions have been
used to achieve thicker layers. The fact that relatively complex material structures are
used for modulators means that integrated structures, which generally have simple
passive waveguides, require careful etch and regrowth.
Optical absorption in these materials is shown in (Dagli 1999) for two wave-
lengths as a function of voltage. The voltage-dependent transmission (loss) T (V )0
for an E modulator can be written as

T (V 0 ) = C exp[−α(V )0 L] (12.58)

where C is a coupling coefficient between the input and output of the device and the
input and output waveguides, respectively,  is the overlap between the optical mode
and the quantum-well cross section, L is the semiconductor pass length, and α(V 0 )
is the voltage-dependent absorption coefficient. This equation allows the on/off ratio
to be written as
12.5 Semiconductor Electro-Absorption Modulators 291

α
10 log[T (V 0 )/T (0)] = L (12.59)
α

where L is the propagation loss in d B in the modulator in the V 0 = 0 state and

α = α(V ) − α(0). The second term α(0) is the background loss which can be of
order −3 dB. The term α/α is typically of order −3 dB, which implies a large
on/off ratio.
Electro-adsorption modulators can be extremely fast. In part, this is due to their
strong on/off ratio allowing relatively short devices to be made, i.e., 50−200 µm,
which have, therefore, low capacitance. Thus, “lumped circuit” devices typically can
be as fast as 50 GHz with a 2 V driving voltage. Traveling-wave modulators have also
been made via careful design so as to avoid microwave loss. As an example of the
traveling-wave devices, a 200 µm long 54 GHz device has been demonstrated with
a 3 V drive voltage and a 20 dB on/off ratio (Kawano et al. 1997).

12.6 Summary

This chapter has described the operation of the linear electro-optic effect and showed
how it may be used to make modulators in insulating and semiconducting crystals.
This material is then used to focus the remainder of the chapter on how modulators
can be designed to have the wide bandwidths needed for high-speed communications
networks or for analog optical links. To make the discussion of modulators complete,
a section on electro-adsorption modulators has been included, despite the fact that
EA devices are not based on the electro-optic effect.
In the following chapter, the discussion of electro-optical devices will be extended
to switches, which can also be operated via the thermo-optical effect. There are other
applications of electro-optic-driven index changes, including scanners or tunable
filters. And many of these convenience these devices are also mentioned in other
chapters.

Fig. 12.18 Cross-sectional view of a modulator

292 12 Electro-Optical Modulators

Problems

1. We are to investigate a modulator based on the LiNbO3 materials platform. The

external electrical field (NOT the propagating wave) is along the optic axis. The
optical field is TE polarized (along optic axis). The device is being modulated
using the Pockels coefficient r33 . The figure shows a cross section of one arm of
an MZI modulator that has a modulator:

(a) Assume the electrode spacing is d = 2 µm. Calculate the length of the device,
assuming the applied voltage is uses 10 V to obtain a π phase shift between
the two arms. Assume only one arm is being modulated. Do not calculate the
modal properties; simply assume d = 2 µm.
(b) Given the bandwidth-length product ν L = 10 GHz · cm. What is the figure
of merit Vπ /ν for this device?
(c) List two more ways to achieve phase modulation other than using electro-
optic effect, briefly talk about their advantages and disadvantages.

References

Alferness, R., Burns, W., Donelly, J., Kaminow, I., Kogelnik, H., Leonberger, F., et al. (2013).
Guided-wave optoelectronics (Vol. 26). Berlin: Springer Science & Business Media.
Alferness, R. C. (1982). Waveguide electrooptic modulators. IEEE Transactions on Microwave
Theory Techniques, 30, 1121–1137.
Almeida, V. R., Barrios, C. A., Panepucci, R. R., & Lipson, M. (2004). All-optical control of light
on a silicon chip. Nature, 431(7012), 1081.
Bridges, W. B., & Schaffner, J. H. (1995). Distortion in linearized electrooptic modulators. IEEE
Transactions on Microwave Theory and Techniques, 43(9), 2184–2197.
Burns, W., Milton, A., & Tamir, T. (1988). Guided wave optoelectronics.
Chen, D., Fetterman, H. R., Chen, A., Steier, W. H., Dalton, L. R., Wang, W., et al. (1997). Demon-
stration of 110 ghz electro-optic polymer modulators. Applied Physics Letters, 70(25), 3335–
3337.
Dagli, N. (1999). Wide-bandwidth lasers and modulators for rf photonics. IEEE Transactions on
Microwave Theory and Techniques, 47(7), 1151–1171.
Gheorma, I.-L. (2002). Fundamental limitations of optical resonator based high-speed eo modula-
tors. IEEE Photonics Technology Letters, 14(6), 795–797.
Gheorma, I.-L., Savi, P., & Osgood, R. (2000). Thin layer design of x-cut linbo 3 modulators. IEEE
Photonics Technology Letters, 12(12), 1618–1620.
Haus, H. A. (1984). Waves and fields in optoelectronics. Upper Saddle River: Prentice-Hall.
Heebner, J. E., Lepeshkin, N. N., Schweinsberg, A., Wicks, G., Boyd, R. W., Grover, R., et al.
(2004). Enhanced linear and nonlinear optical phase response of algaas microring resonators.
Optics Letters, 29(7), 769–771.
Kaminow, I. (1965). Microwave dielectric properties of n h 4 h 2 p o 4, and partially deuterated k
h 2 p o 4. Physical Review, 138(5A), A1539.
References 293

Kawano, K., Kohtoku, M., Ueki, M., Ito, T., Kondoh, S., Noguchi, Y., et al. (1997). Polarisation-
insensitive travelling-wave electrode electroabsorption (tw-ea) modulator with bandwidth over
50 ghz and driving voltage less than 2 v. Electronics Letters, 33(18), 1580–1581.
Little, B. E., Foresi, J., Steinmeyer, G., Thoen, E., Chu, S., Haus, H., et al. (1998). Ultra-compact
si-sio 2 microring resonator optical channel dropping filters. IEEE Photonics Technology Letters,
10(4), 549–551.
Liu, A., Jones, R., Liao, L., Samara-Rubio, D., Rubin, D., Cohen, O., et al. (2004). A high-speed
silicon optical modulator based on a metal-oxide-semiconductor capacitor. Nature, 427(6975),
615.
Noguchi, K., Mitomi, O., & Miyazawa, H. (1998). Millimeter-wave ti: Linbo/sub 3/optical modu-
lators. Journal of Lightwave Technology, 16(4), 615–619.
Noguchi, K., Mitomi, O., Miyazawa, H., & Seki, S. (1995). A broadband ti: Linbo/sub 3/optical
modulator with a ridge structure. Journal of Lightwave Technology, 13(6), 1164–1168.
Sadagopan, T., Choi, S. J., Dapkus, P. D., & Bond, A. E. (2004). Digest of the leos summer topical
meetings mc2-3.
Saleh, B. E., Teich, M. C., & Saleh, B. E. (1991). Fundamentals of photonics (Vol. 22). New York:
Wiley.
Spickermann, R., Sakamoto, S., Peters, M., & Dagli, N. (1996). Gaas/algaas travelling wave electro-
optic modulator with an electrical bandwidth> 40 ghz. Electronics Letters, 32(12), 1095–1096.
Wooten, E. L., Kissa, K. M., Yi-Yan, A., Murphy, E. J., Lafaw, D. A., Hallemeier, P. F., et al. (2000).
A review of lithium niobate modulators for fiber-optic communications systems. IEEE Journal
of Selected Topics in Quantum Electronics, 6(1), 69–82.
Xu, Q., Schmidt, B., Pradhan, S., & Lipson, M. (2005). Micrometre-scale silicon electro-optic
modulator. Nature, 435(7040), 325.
Yariv, A. (1973). Coupled-mode theory for guided-wave optics. IEEE Journal of Quantum Elec-
tronics, 9(9), 919–933.
Chapter 13
Integrated Optical Switches

Abstract This chapter presents one related modulator-like functionality; i.e., switch-
ing of a propagating light wave. This functionality is important for many applications,
including signal multiplexing in time and wavelength, signal routing, and signal
encoding. An effective photonic switch requires a high extinction ratio, low loss, and
a short response time. Switches are shown to be based on mechanical movement,
thermo-optic, electro-optic, or electron-plasma optical to change the refractive index.
Our discussion focuses on electro-optical switches since they were the most important
type for integrated switches particularly those for ultrafast lithium niobate crystals.
In addition, the chapter includes a short section on thermo-optical-based switches
as well. Finally, methods of improving switch performance are also presented in the
chapter.

13.1 Introduction

In the previous chapter on the fundamentals of electro-optical devices, our discussion

focused on optical modulators operating via the electro-optic effect (with a short
discussion on electro-absorption). In this chapter, we will focus specifically on one
important modulator function: switching. Switching is used for a number of important
applications, including signal routing, signal multiplexing in time and wavelength
domain, and signal encoding. Basically, in these applications, the photonic switch
turns a particular optical channel on or off. Thus, an effective device requires a high
extinction ratio, low loss, and a short response time that, is dependent on the interval
between bitstreams. Switches can use simple mechanical movement, thermo-optic,
electro-optic, or electron-plasma optical to change the refractive index. Our initial
discussion will focus on electro-optical switches since they played an essential role
in the evolution of integrated switching technology particularly those using LiNbO3 ;
however, we also include a brief section on thermo-optical switches.

© Springer Nature Switzerland AG 2021 295

R. Osgood jr. and X. Meng, Principles of Photonic Integrated Circuits,
Graduate Texts in Physics,
https://doi.org/10.1007/978-3-030-65193-0_13
296 13 Integrated Optical Switches

13.1.1 Electro-Optical Switches

As suggested in the previous chapter, most integrated electro-optical switches rely

on phase modulation in their waveguiding structure. These switches use an external
voltage to cause a shift in the material refractive index, thus the propagation constant
of the waveguide. Electro-optical switches are available in several well-studied forms:
directional couplers, intersecting waveguides, and interferometers (Alferness 1982;
Campbell et al. 1975; Soldano et al. 1994).
Despite the fact that these switches can be quite different in structure, their electro-
optical function can be expressed in equations that show their close similarity. The
common equations stem from the fact that most, but not all, switches rely on some
form of phases shifting in a waveguide. Thus, as shown in Chap. 12, the phase shifter,
φ, in one arm of the waveguide containing an electro-optical element of length L
and electro-optic coefficient, r , can be written quite generally as

VL
φ = β L = −πn 3r  (13.1)
dλ
where the symbols are defined, for example, as in the case of (12.35). Since the phase
shift is usually related to some multiple of π, empirically, one can then also write

|β L| = pπ (13.2)

where p is a real number with a characteristic value, which varies from

√ one modulator
type to another. For example, for a directional coupler switch p = 3, for a Mach–
Zehnder (MZ) p = 1, and for a polarization switch p = 1/2. Thus, (13.1) and (13.2)
gives the voltage–length product, as determined by electro-optics, as

pλd
Vπ L = (13.3)
n 3r 
This more general formulation of the phase shift (Haus 1984) allows one to compare
the performance of switches, which are based on different modulation types.

13.1.2 The β Directional Coupler Switch

One of the most basic electro-optical switches, shown in Fig. 13.1, uses a typi-
cally symmetric, directional coupler formed in electro-optical material. This device
was one of the earliest demonstrated electro-optical switches; both Ga As-based
Campbell et al. (1975) and LiNbO3 Papuchon et al. (1975) versions were fabricated.
In the device, the coupling length was adjusted so that in the absence of an applied
voltage, light was coupled, with coupling coefficient κ, into the adjacent waveguide,
a condition termed the cross (×) state in switches. Thus, the device length L is such
13.1 Introduction 297

Fig. 13.1 A sketch of the electro-optical switch that uses a symmetric, directional coupler. Applying
voltage on the electrodes changes the refractive index of the coupling region, thus the coupling ratio
sketch of the electro-optical switch that uses a symmetric, directional coupler.

that L = L c , or
π
κL = (13.4)
2
When a voltage is applied to the structure, a shift in the waveguide refractive index
is induced. This shift, reflected in the waveguide propagation constant, β, causes
the waveguides to be no longer phase-matched and light remains in the original
waveguide; hence, the switch is called a β switch. In principle, the application of
a voltage can also change the value of the coupling coefficient, κ; however, typically
this effect is minimal and not important for this type of switch.
Recall from (5.37) and (5.38) that for a symmetric directional coupler, the power
transfer efficiency, ηx , from an input waveguide to its cross waveguide is given by
 2
κ
ηx  sin2 βc L (13.5)
βc

and the efficiency of remaining in the input (or bar) waveguide is

 2
κ
η = 1 − ηx = 1 − sin2 βc L (13.6)
βc
√
where βc = κ2 + 2 and where  = β/2. Figure 13.2 gives a plot of the nor-
malized output of the switch, that is, its cross waveguide power versus β, and
hence applied voltage. From relations (13.5) and (13.6), it is apparent that complete
power transfer to the cross state, ηx = 1, can only occur if βc = κ; that is, when
 = 0 or when the voltage is off (see (13.1)). In addition, the cross state requires
that sin2 βc L must be unity, or βc L = (2n + 1)π/2, where n is an integer. However,
as is also shown in Fig. 13.2, when  = 0, the power transferred to the opposite
waveguide is reduced, and in fact for certain specific values of the voltage-controlled
298 13 Integrated Optical Switches

Fig. 13.2 A plot of

normalized output of the
switch versus the
propagation-constant
discrepancy in the two

Intensity modulation
waveguides due to the =
2
applied voltage

propagation-constant mismatch of the two waveguides, β, all entering power exits
through bar state, that is, on the input waveguide. Thus, as is shown in the figure, a
typical β switch is designed so that in its bar state the voltage is on, and in its cross
state the voltage is off. In electro-optical switches, the value of β is controlled with
an external voltage via the electro-optical effect applied to one arm of the directional
coupler; if the device is operating in push–pull, voltage is applied to each arm. As
suggested by the first two maxima in Fig. 13.2, the light exiting the device on the
cross-state waveguide has sidebands of decreasing magnitude at higher values of β,
i.e., external voltage. These sidebands, which lead to undesired leakage through an
imperfectly fabricated device in its nominally off state, may be decreased by using
electrode tapering Schmidt and Kogelnik (1976).
Switches can also be made for devices, which are multiples of the coupling length.
Accessing
√ the first maximum,
√ that is, switching to the bar state, requires β L =
3π, or p in (13.2) is 3. However, as a consequence of the voltage dependence of
(13.5), the switching voltage from the cross to the bar state increases with the number
m of coupling lengths, L = m L c , in the device. Specifically, if the device length, L,
is (2m + 1)π/2κ, i.e. such that with no voltage applied η = 0, then the voltage for
switching increases with m. This can be shown by noting that in each case after the
voltage is applied, switching from the cross to the bar state occurs when

βc L = ((2m + 1) + 1)π/2. (13.7)

√
Since βc = κ2 + 2 and  = β/2, the (13.7) can be written as
 1/2
β L = 2π (m + 1)2 − (m + 1/2) . (13.8)

Thus, the √longer the length, the higher the voltage, and in the limit of large m,
β L ≈ 2 mπ. A plot of intensity in the cross channel is shown in Fig. 13.3 for two
values of length L = π/2κ and 3π/2κ. Note that the voltage required for the cross
13.1 Introduction 299

Fig. 13.3 A plot of

normalized output in the
cross channel for two
different lengths, L = π/2κ =
2
and 3π/2κ

Intensity modulation
=
2

state, i.e., the first null point in the plot, increases with the length of the device, L,
as indicated by (13.8).
In designing β switches, including those having multiple electrodes, which
are discussed below, it is necessary to have a more methodical approach for the
dependence of the switch geometry on variations in L and voltage (Kogelnik and
Schmidt (1976)). A convenient approach to this problem, and an approach, which is
transferable to more complex β switches, is that provided by the use of the transfer
matrix for a directional coupler (here a switch) that was introduced in Chap. 6. Recall
that if such a switch is characterized by inputs in the two waveguides of a1i and a2i ,
respectively, then the output a1o , a2o , can be obtained via the transfer matrix U for a
coupler of length L as follows:
    
a1o u 11 − ju 12 a1i
= (13.9)
a2o − ju ∗12 u ∗11 a2i

where the matrix elements have been given earlier in (6.16) and (6.17). This matrix
is written for a positive voltage, u + , for the device shown in Fig. 13.1. The voltage
for the opposite polarity, u − , is identical to u + except that the diagonal elements are
reversed, i.e., u 11 → u ∗11 and u ∗11 → u 11 .
For a simple directional coupler switch of length L, one input, say, a2i , is always
zero. In addition, the device is designed such that in its “off” condition, i.e., β = 0
or  = 0, the device is in its cross state, that is a1o = 0 at z = L. Consideration of
(13.9) shows that these conditions require that u 11 = 0. Using the explicit form of
the matrix element given in (6.16) and (6.17) shows that u 11 = 0 only when

κL = (2m + 1)π/2 (13.10)

or
L
= 2m + 1 (13.11)
Lc
300 13 Integrated Optical Switches

Fig. 13.4 A plot of

operation points of the
simple β switch,
illustrating one of the two
limitations of discrete
operation points. This figure
shows that the operation
points on the β axis are
actually only the bar state,
cannot be compensated for
the length discrepancy.
Notice that here the
electro-optical effect is used
to change the refractive
index of the coupling region,
thus the wavevector. Symbol
HV means high voltage

i.e., the exact conditions specified in the earlier discussion in this section and where
m is again an integer, starting at m = 0. In a similar manner, it can be shown that
after application of an external voltage, the bar state, which requires a2o = 0, also
implies that u 12 = 0. This condition holds when

(κL)2 + [()(L)]2 = (mπ)2 (13.12)

or  2  2
L β L
+ = 4m 2 (13.13)
Lc π

This equation has multiple solutions, in contrast to that for the cross state. The two
conditions are displayed in Fig. 13.4 in a plot of L/L c versus β L/π for both bar
and cross states. In this diagram, switching occurs by moving horizontally, at a fixed
value of L/L c , on the diagram. Notice, in addition, the condition (13.11) corresponds
to a series of fixed points along the ordinate, regardless of the voltage. As a result,
the device can attain the cross state only if L is precisely L c ; adjustment in L/L c
via voltage trimming (through β) is impossible in the cross state. However, the
condition for the bar state, given in (13.13), may be satisfied for a set of lengths L
and voltage; thus, if L is not precisely L c , the bar state can be reached by voltage
trimming. The physical basis for this behavior is discussed earlier in this section
in conjunction with (13.5) and (13.6). As examples, the operating points for two
devices, each with different values of L/L c are indicated in the figure.
13.1 Introduction 301

13.1.3 Reversed β Switch

The simple β switch discussed above suffers from two important limitations. First,
in order to have low cross talk for the voltage off or cross state of the switch, the device
must be exactly one coupling length (or odd multiples of the coupling length). If the
length deviates from the distance of one coupling length, light will “leak” through,
reducing the on/off ratio, which will lead to serious degradation in the performance.
The discrete operating points on the plot of L/L c versus β (see Fig. 13.4), mean
that an error in the length, L, cannot be corrected with a corresponding change in β.
Second, due to the typical electrode spacings used in most switches, the coupling
length is generally small compared to the electrode length needed to get a sufficient
phase shift at an acceptable range in voltages. Thus, since L/L c 1, the switching
voltage and, hence, the voltage-speed product will then be unacceptably high.
In order to solve the above two limitations, an approach is used which divides an
electrode of length L/L c > 1, into N segments of equal length. Each of the lengths
is periodically biased so as to reverse the charge polarity on the electrodes. Hence,
this switch is called a “reverse β-switch.” The response of such a series of multiple
reversed devices can be obtained via multiplication of the transfer matrix of each
individual segment. This analysis shows that for a switch with a series of N sections
having periodic voltage reversal, the overall cross efficiency is given by

η N = sin2 κ N L (13.14)

where
N √
κN = sin−1 ηx (13.15)
L
where ηx is the efficiency of transfer into the cross state for a single element on length
L/N , given earlier by (13.5). Finally, note that the required β L for this switch gives
√
3π (see (13.20), even for L/L c > 1,) so long as the number of electrode segments
equals approximately the number of coupling lengths.
The simplest multiple electrode switch is the reversed β switch with an overall
length of L, shown in Fig. 13.5 Schmidt and Kogelnik (1976). This switch uses two
β switches of equal length ∼ L/2. Because of its simplicity, it can be used to
delineate clearly the advantages of reversed β switches. The switches are arranged
in a series with β of the first section put in series with—β of the second; the
shortest device, which can be voltage tuned in both cross and bar states, has L > L c
or L c /2 for each section.
The most important feature of such a reversed β switch is that it is insensitive
to the exact length of the electrode segments since each “state” of the switch can
be voltage-tuned for optimum performance. This length insensitivity can be shown
clearly by an analysis based on the transfer matrix approach introduced in the pre-
vious section. This approach can also be readily applied to a device having multiple
sections.
302 13 Integrated Optical Switches

+V −V

Fig. 13.5 A sketch of the “reversed β-switch.” Reversed voltages are applied on each half of the
coupling region

The overall transfer matrix for a two section device is the product of two matrices
for electrodes of length L/2 and opposite-biasing polarity. The product matrix is
then  t 
u 11 − ju t12
ut = u− · u +
= (13.16)
1 1 − ju ∗t ∗t
12 u 11

where u t11 = 1 − 2u 212 = 2u ∗11 u 12 and where the fact that u 11 u ∗11 = 1 has been used.
The values of these matrix elements can be obtained from (6.16) to (6.17) and the fact
that the length of the segment is L/2. It is interesting that the nature of the coupling
interactions does not cause the two reverse bias sections to “cancel out” their effect.
For a single input only in waveguide 1, i.e., a2i = 0, the total device will be in the
cross state when u t11 = 0, or when

2u 212 = 1 (13.17)

Substitution of (6.17) in this equation leads to a family of curves, which obey the
relation,
π κ βc L
sin2 = ( )2 sin2 (13.18)
4 βc 2

Note that this same result can be obtained by direct application of the coupled mode
equations Haus (1984). Unlike the case of a single section device which has only
isolated operating points in the cross state, irrespective of β, the two section device
allows tuning of β to compensate for an inexact value of L as long as L > L c .
In the bar state, u t12 = 0, ai0 = 0. This condition may be achieved either via
u 11 = 0, which implies β = 0 and

L
= 2(2n + 1) (13.19)
Lc

or by u 12 = 0, which implies
13.1 Introduction 303
 2  2
L β L
+ = (4n)2 (13.20)
Lc π

where n is an integer 0, 1, etc. The first condition leads to isolated points and is not
of interest, while the second leads to a set of curves (circular quadrants) on a plot
of L/L c versus β/π. The curves also show that in the range of 1 < L/L c < 3,
both the bar and cross states can be accessed by a simple adjustment of the voltage
to change β. For example, if 2 < L/L c < 3, the device is switched between cross
and bar states by, roughly, doubling β. Since operating curves, rather than points,
exist for both the bar and cross states, the device can, as mentioned above, be tuned
to compensate for inexact fabrication of the electrode lengths. Note also in each case,
the shift in β and, hence, the value of adjustment voltage is at a value comparable
to the operating voltages in a single section β switch and thus does not impose any
additional voltage penalty for using multiple electrodes.
A more intuitive understanding of the operation of a reverse β switch can
be obtained by considering the operation of this switch at the particular length of
∼ L/L c = 2 (Nishihara 1980) for this length. These conditions lead up to a ∼ 50%
power transfer at the end of the first section or subcoupler, i.e., at L/2. When this split
light beam enters the second subcoupler, it experiences the opposite phase mismatch,
and the original local mode is reconstructed at the output of the cross waveguide.
If the voltage for the bar state is now applied, the signal is switched
√ to the bar state
at the end of each subcoupler. Finally, notice that if L/L c = 2 and a voltage of
β = 2π is added, the polarity of the second section can be merely reversed and the
device will be switched to the on the state of uniform β switch. Finally, note that
reverse β switches have been made with extremely low cross talk. For example,
Bogert et al. (1986) reported a LiNbO3 device with 43 dB extinction ratio.

13.2 Interference-Based Switches

Switching can also be accomplished by using electro-optical phase shifting in one

or both arms of a “coupler-equipped” MZ interferometer. For this switch, the nec-
essary two input/output waveguides are provided by 3 dB directional couplers at
the input/output (see Fig. 13.6) Martin (1975), Ramaswamy et al. (1978). In this
device, the output is either bar or cross, depending on the β induced in the device.
The device is thus identical to similar structures used for wavelength filtering. The
device switching efficiency from input waveguide one to output waveguide two can
be shown to be
η = cos2 β L/2 (13.21)

Notice that the required switching

√ voltage in this case is β L = π (i.e., p = 1 in
(13.2)), and is thus less, by 3, than that required for the β directional coupler
switches described in the previous section. Also, both switch states are obtained
periodically as the voltage is varied.
304 13 Integrated Optical Switches

+V
3 dB
Coupler

Fig. 13.6 A sketch of the interference-based switch. This is a MZinterferometer consists of two
3 dB coupler and two balanced arms. The voltage is applied to introduce a phase shift. Note that in
this case, the phase shift needed to induce is β = π/L c , which is smaller than that in a directional
coupler case. Also note that by introducing asymmetrical couplers, this switch can also operate as
a mode sorter

The MZ interferometer is a related interferometric device using 3 dB Y-branch

inputs and outputs. This device does not act like a typical optical switch, since the
output is not directed into one of two output ports. However, it can function as a switch
in certain integrated circuit layouts by generating an on–off waveform. Because this
device is described in detail in Chap. 12, only a few brief comments will be added
here. First, for this device, as for a 3 dB-coupler interferometer, the output is fully
modulated when β L = π, i.e., again p = 1. Also, it is possible to make a true
switch from a MZ device by using an asymmetric branch at the output so as to form
a mode sorter. In a conventional device, without the mode sorter, the antisymmetric
mode, which is formed in the off, on-transmission, state of the modulator is radiated
into the substrate. With the mode sorter, this mode can be directed to one of the
output arms of the asymmetric Y-branch Izutsu et al. (1982).
Multimode interference devices have also been used as the splitting or “feeding”
elements in interferometric-based switches. Specifically, 1 × 2 dB MMI couplers
can be used in conventional MZ devices because of their good output uniformity
and stable relative phase. For example, considering one purely passive device, a pair
of 3 dB MMI couplers was used with a MZ structure to make a polarization splitter
with a 60 nm wavelength range Soldano et al. (1994). Using the same basic elements
plus electro-optic control, MZ interferometer switches have been demonstrated in
double heterostructure (DH) Sekiguchi et al. (2012) as well as in multiquantum well
(MQW) III–V materials Morl et al. (1998); these devices had extinction ratios of from
−10 to −19 dB. The MMI couplers were important in attaining wide bandwidth and
polarization-independent operation Bachmann et al. (1993), Zucker et al. (1992).
More complex MZ-like interferometer switches can also be made with MMIs. In
such an interferometer, the input signal is split with a 1 × N (or N × N ) MMI splitter,
fed into individually addressable waveguide phase shifters, and then recombined in
an N × N MMI coupler (see Fig. 13.7). By controlling the phase at the output of
each of the N shifters, the input signal(s) can be switched to any one (or certain
13.2 Interference-Based Switches 305

1 × or
× MMI splitter Phase shifters × MMI combiner

Fig. 13.7 A sketch of the MMI-based switch. It can be a 1 × N or a N × N switch, with the active
tuning capability by adding phase shifters in the middle of two MMIs

sets) of the N output waveguides, although independent control is not possible. The
use of MMI couplers for switching, including both the 1 × N and N × N devices,
for the splitter/combiner elements allow the realization of very compact integrated
multiway optical switches. For example, 10.6 mm long 1 × 10 and 13.1 mm long
10 × 10 switches have been made in a Ga As/AlGa As epitaxial layers Jenkins et al.
(1994). These switches showed ±9% switching uniformity, −10 dB maximum cross
talk, and ∼ 6 dB excess loss.

13.3 Modal Interference Switches

Modal interferometric switches rely on interference between two or more modes,

which are excited by an input waveguide. Because the input conversion process and
the interference do not cause any modal extinction, and hence power loss, these
switches can be, in principle, of high efficiency. In general, these devices have not
been investigated as extensively as the β switches described earlier.

13.3.1 Intersecting Waveguide Structures

Modal interference in a multimode waveguide has also been used as the basis for
switching. An intersecting waveguide switch is one of the earliest versions of these
devices Neyer et al. (1986). In this device, sketched in Fig. 13.8, light entering each
306 13 Integrated Optical Switches

Fig. 13.8 A sketch of a 2 × 2 intersecting waveguide that uses the modal interference to switch

one of the input waveguides excites the two lowest, symmetric and antisymmetric
system modes. In the normally off state of the two output waveguides, the symmetric
and antisymmetric lowest order modes interfere, with the relative phase depending
on the total path length in the mixing region. The modes then propagate within
the structure so as to form a single local mode, at one of the waveguide ports. When
voltage is applied, the modal propagation constants are changed by the electro-optical
effect to interfere in such a way that light exits on the other waveguide. If the path
lengths are properly fabricated, the device will switch from one output waveguide to
another, according to its voltage state, with low cross talk. This device has a voltage–
length requirement similar to that for directional coupler switches and has displayed
low cross talk, i.e. 30 dB Neyer et al. (1986), but the requirements on the device
geometry are demanding.

13.3.2 MMI Coupler Switches

Switching can also be realized by changing the modal phases in the “free-space”
region of an MMI coupler by varying its refractive index. These MMI switches
are simpler in structure than those based on phase shifters, which were discussed
above, and thus they require simpler controlling circuitry. To illustrate this device, a
1 × 3 switch is shown in Fig. 13.9. The device dimensions, which are defined in the
figure, are chosen based on self-imaging principles for each switching function. The
required index changes are made in the regions that are shaded. In a semiconductor
device, for example, these are realized by current injection, since injection of current
can alter the index of a semiconductor as was discussed in Chap. 2. The device in the
figure operates as follows: when n 1 = n 2 = n g , the device is a 1 × 3 symmetric MMI
splitter; when n 2 < n 1 = n g , it is a 2 × 2 MMI coupler; and when n 1 = n 2 < n g , it
is a 1 × 1 MMI coupler. A general 1 × N switch can be constructed using a similar
structure Zhao et al. (1998).
13.3 Modal Interference Switches 307

n1 Out1

D
In1 W Out 2
ng D
W/2

Out 3
n2
d

Fig. 13.9 A sketch of a 1 × 3 MMI-based reconfigurable switch. The refractive index of n 1 and
n 2 . When n 1 = n 2 = n g due to the “free-space length” is equal to the self-imaging distance, the
signal would come out from output port 2; when n 2 < n 1 = n g lower part can be ignored, the upper
part becomes a 2 × 2 MMI, the signal will come out from output port 1

13.3.3 Digital Optical Switch by Mode Sorting

The reverse β switch described earlier has several attractive features, including
relative insensitivity to voltage and electrode length. However, it does not clearly
operate as a threshold switch, such as is needed, for example, in digital electron-
ics; it is also sensitive to input polarization. Similarly, the intersecting waveguide
switch described above also does not operate as a threshold switch and, in fact, has
a sinusoidal response.
A truly digital 2 × 2 switch has been described Silberberg et al. (1987), which is
based on electro-optical control of mode sorting in two waveguides which intersect
at an angle θ. A schematic of this device is shown in Fig. 13.10 for use on z-cut
LiNbO3 . The device operates by switching the modes entering on either the wider
or narrower waveguide via control of the refractive index on the output arms of the
switch.
Specifically, the device geometry is such that if light enters on the wider guide, it
excites only the lower order mode of the junction. If it enters on the narrower guide,
it will excite the second-order mode of the junction region.
This mode sorting occurs if the transition is adiabatic, that is, if

β
θ (13.22)
γ

where γ is the modal decay constant in the cladding region. Recall that this condition
was discussed previously in Chap. 7; see (7.24). Note that since the requirement for
308 13 Integrated Optical Switches

Fig. 13.10 A sketch of a digital 2 × 2 switch, based on electro-optical control of mode sorting.
Note that the upper input waveguide is wider compared to the lower one

mode sorting is that of a simple inequality, it requires only that the angle be sufficiently
small.
Once these modes are excited, the light will be directed to a particular output
waveguide based on the effective index of that waveguide, which in this section
of the device is controlled by the biasing of the output waveguide. For example,
the fundamental mode is always directed to the waveguide with the highest index.
Reversing the bias reverses the illuminated output guide. With no bias, both output
guides are illuminated.
Note that this switch does not rely on modal interference but instead on bias-
induced symmetry breaking. The calculated switch output intensity in the two output
guides versus induced index is shown in Fig. 13.11. The figure shows a sharp step
response in the switching signal with small changes in the effective index, indicating
true threshold behavior, for their over biasing does not alter this state.
This digital switch has been fabricated in Ti-diffused, z-cut Li N bO3 . The device
had a switch voltage for −15 dB cross talk of ±45 V for the TE mode and ±15V for
the TM mode, and a wide wavelength response ∼1.3 − −1.5µm. While the crosstalk
values for this switch are practical, the voltage requirement is very high; this high
voltage has inhibited greater use of the device. Because of the requirement on adia-
baticity in the output arms, the device can be very long. More recently, digital optical
switches have also been made in polymeric materials Moosburger et al. (1996);
however, these use the thermo-optic effect and are thus discussed in Sect. 13.4.
13.4 Thermo-Optical Switches 309

Fig. 13.11 A plot of output

intensity in the two output
waveguides verses induced
index change. The response
is shaped, which means over
biasing does not change the

Light output
device performance
0.5

-0.002 0 0.002
Index difference

13.4 Thermo-Optical Switches

Thermo-optical-induced changes in the refractive index of materials can also be used

to drive switches. Although, in general, devices based on thermo-optical effects are
slow compared to electro-optical switches, their response times, which may be as
fast as ∼1µs, are short enough that the switches are useful for many applications.
These include reconfigurable PICs, such as delay lines or dispersion compensators,
or WDM routers. Thermo-optical switching is also extremely important in a practical
sense because it can be used to give some “active-optical” character to a PIC, which
is made of normally purely passive materials such as silica, Si, or polymer.
In principle, the operating principle and device layouts for thermo-optical switches
can be identical to many of those used in electro-optical devices, except that in
this case, a temperature-induced index change is used to provide the needed phase
change in the switch. In practice, however, a few simple designs are used most
frequently. They include MZ interferometers and digital optical switches. Thermo-
optical switches are useful in WDM applications such as add/drop and cross-connect
functions since they can switch signals without resorting to optical/electrical signal
conversion prior to switching. Notice that the system is transparent to the signal
format, either analog or digital.
A typical design for a MZ switch made of SiO2 -on-Si is shown in Fig. 13.12. It
consists of a switching unit having two 3 dB MZ devices, each with a heating unit
on one arm. For this silica system, the switch time was 2ms.
More recently, faster devices, with ∼1 − 5 µs response times, have been achieved
in the Si-on-SiO2 or SOI materials system. The silicon waveguide in this system
is thermally isolated during switching and is amenable to simple analytic thermal
analysis. In this text, we will use this materials system to demonstrate the design
considerate for thermo-optical switches. The phase shift in one arm of an interfer-
310 13 Integrated Optical Switches

Thermo-opc phase shier

(a)
Input 1 Output 1
off

Input 2 Output 2
Signal on

First MZI Second MZI

(b) N stages

1
1
2
2

N
N

Intersecon regions between stages

:Off :On

Fig. 13.12 a A sketch of typical MZ-based switch on the platform of S O I . Due to the relatively low
speed of thermal–optical modulation, the device speed is about 2 ms. b A sketch for the MZ-based
switch implementation of a N × N switch

ometer can be obtained using a modified analytical treatment of the heat flow. This
phase shift is  
2π dn
φ = T L H (13.23)
λ dT

where λ is the wavelength, dn/dT is the thermo-optic coefficient of silicon, T is

the change in the temperature, and L H is the heater length. For example, (13.23)
shows that a temperature change of 8◦ K is needed to achieve π phase shift in a
silicon interferometer, which has a thermo-optical coefficient of ∼1.8 × 10−4 K −1 , a
length of 700 µm, and λ = 1.5 µm. Additional useful expressions can also be easily
obtained. For example, if the heat flow is corrected for lateral spread, the switching
power can be written as
13.4 Thermo-Optical Switches 311
   −1
WH dn
Pπ = λκSiO2 + 0.88 (13.24)
tSiO2 dT

where κSiO2 = 1.4 W/cm · K is the thermal conductivity of SiO2 , tSiO2 is the thickness
of the SiO2 , and W H is the width of the heater. Using the Tπ = 8◦ K calculated in
(13.23), we find that the switching power if Pπ = 46 mW if the heater is 14 µm in
length and tSiO2 is 1 µm.
The temporal response of the thermo-optic switch can be obtained from an ana-
lytical expression, derived by Fischer et al. (1994), relating the cutoff frequency to
the switching power and heated area of such a device,
  
1 Pπ dn
f cutoff = (13.25)
πλρth A dT Si

where Pπ is the switching power, A is the heated cross-sectional area, λ is the

wavelength, and ρ and th are the density and specific heat of thermally grown SiO2 ,
respectively. This equation gives an estimated cutoff frequency, f cutoff = 49k H z,
corresponding to a rise time, τrise = 7 µm.
Thermo-optical switches have been made in other materials platforms. For exam-
ple, a polymer digital–optical switch, which was based on an asymmetric waveguide
cross, i.e., the digital optical switch discussed in Sect. 13.3, has been successfully
fabricated. The device had low insertion losses, < 3 dB, due to the use of an oversize
rib waveguide and an extinction ratio of > 20 dB with a 200 mW heater Moosburger
et al. (1996). In addition, devices based on total internal reflection have also been
described Diemeer et al. (1989).

13.5 Optical Switch Arrays

13.5.1 Electro-Optical Switching Arrays

Optical switching arrays in LiNbO3 , and more recently Si, comprise some of the
most complex working PICs made, and their fabrication has provided a good testbed
for examining various PIC technologies. These devices have arrays of directional
couplers, and generally have a straightforward cross–bar switch geometry, which
requires N 2 switches for an N × N switch. This switch geometry requires that these
devices be arranged in 2N − 1 stages.
312 13 Integrated Optical Switches

13.5.2 LiNbO3 Arrays

Historically, LiNbO3 arrays were the first successful array type to be considered for
commercial switching fabrics. These large arrays were the product of a long series
of advances by, for example, Bell Labs. To achieve low crosstalk for this applica-
tion, reverse β switches were generally used. When low index-contrast Ti-diffused
waveguides are used, bend radii in this material are relatively large. Waveguide loss
and crosstalk and operating voltages are the major operating parameters of concern.
The most complex array reported was an 8 × 8 switch, containing 64 directional
couplers Granestrand et al. (1986). Another 8 × 8 switch used a rearrangeable switch-
path layout of 28 elements to achieve a non-blocking layout. In this case, each switch
had a 27 V switching voltage, Vπ , with a 1.3 V standard deviation, and a total loss
of 5.5 dB at 1.3 µm Duthie and Wale (1988). The switches in this case were β
switches, with ∼4 mm long interaction lengths. The bends had 40 mm radii.
Smaller, i.e., 4 × 4 LiNbO3 arrays have been examined by many groups and thus
their design has been carefully examined. In one example, a 4 × 4 array at 1.3 µm was
made using reverse-β cross switches and uniform-β bar switches. In this case, the
operating voltages, Vπ , were 8 and 13 volts for the bar and cross state, respectively,
with a cross talk of −35 dB Bogert et al. (1986). This switch was mounted on a
silicon optical bench and pig-tailed to an optical fiber input/output array. In another
example, an 4 × 4 crossbar array had a Vπ of 15 V, a crosstalk value of 35 dB and a
path-dependent loss of ∼5 dB McCaughan and Bogert (1985).
Finally, there has been an increasing interest in semiconductor arrays; the initial
projects were made using III–V (Ga As/AlGa As) materials. For example, a 4 × 4
array was reported which used reversed β switches. Each switch required 22 volts
for switching and had a 0.5 dB path-dependent loss Komatsu et al. (1991). Since this
initial work the commercial and advanced development groups have focused on Si
materials; see the following section!

13.5.3 Si Arrays: Principally Mach–Zehnder-Based

The very large advance in large-scale Si photonics systems principally to handle

large-scale data streams has naturally given rise to research on Si switching arrays.
An excellent review of this work has been written by (Fang 2012; Chen 2015) The
technological challenge, which drives the remarkable growth of Si photonics, is the
need for expanding overall bandwidths of systems for integration of computing and/or
signal processing applications; these needs cannot be realized with the traditional
metallic interconnects used in the past or even the high-quality arrays using LiNbO3
discussed above. On the other hand, silicon photonics provides a low-cost approach
to high-data rate transmission by using well-established planar processing methods
to fabricate photonic integrated circuits. In addition, in certain designs, the use of Si
ring switches allows ultrasmall on-chip footprints.
13.5 Optical Switch Arrays 313

A discussion of the technology in switching arrays provides excellent examples

of what is possible with silicon photonics. Note that arrays, whether using LiNbO3
or III–Vs have in the past been the photonic-system types, which have paced the
development of large-scale photonic circuits; Si arrays now have become the same
pacing leader for large-scale photonic circuits. As a result, we devote the remainder
of this section to describing three examples of Si switching arrays; our focus in
this discussion will be MZ-based devices due to many recent applications. However,
some comments on the emerging use of ring-based arrays will also be included as
well.
For example, (Suzuki 2014) has fabricated a silicon photonics 8 × 8 non-blocking
optical switch based on double-MZ element switches. These double-MZ switches
consisted of a waveguide intersection and two asymmetric MZ switches. The design
allowed crosstalk suppression for a large range of spectral bandwidth. In particular,
the 88 switches exhibited an average fiber-to-fiber insertion loss of 11.2 dB as well as
−20 dB cross talk over a 30 nm spectral bandwidth. As an important system demon-
stration, the devices were used to show 32-Gbaud dual polarization, quadrature-
phase-shift keying, for four-channel wavelength-division-multiplexed signal trans-
mission.
As a second example, Zhang (2016) demonstrated an 8 × 8 × 40-Gbps fully
integrated silicon photonic network on a Si chip. This was an integrated pho-
tonic network-on-chip circuit and operated using wavelength division multiplexing
transceivers.
In a third example, another group has shown the operation of a large-scale low-
loss 88 silicon photonic module for a wide variety of add–drop multiplexers. The
switch was based on using the thermoelectric effect to carry on the needed device
phase shifting. The system incorporated an advanced silicon optical switch chip with
spot size converters. The devices resulted in polarization-insensitive and wavelength-
insensitive properties over several optical bands. The optical loss in the fabricated
devices was particularly low and in fact the modules had only 6 dB excess optical loss
for each of the 64 optical paths. These paths also had low polarization-dependent loss
and low cross talk. With these modules, a transponder aggregators (TPA) prototype
was fabricated having a 100-port optical switch subsystem.
Please note while the majority of this section has focused on MZ-based arrays,
there have been major and very attractive array designs that have used ring-based
switching arrays (Chen 2015). These arrays are of lower power and with small foot-
prints due to their small diameters (an advantage ultimately based on their high
index contrast when the Si is grown on SiO2 ). Discussions of the early design of
these arrays are provided in the references given here (Chen 2015).

13.5.4 Thermo-Optical Switching Arrays

Because thermo-optical switching uses standard passive-waveguide device material,

it can provides a low-cost, high-performance solution to making large-scale arrays.
Since thermo-optical switches can be made in several optimized passive-waveguide
314 13 Integrated Optical Switches

materials platforms, the arrays can have very low input and propagation loss even
on large substrates of these materials. Of course, the switching speed is limited to
the speeds in our discussion above on the device level aspects of on thermo-optical
switches.
To give one example of a thermo-switching array, consider the matrix–array design
shown in Fig. 13.12 for use with silica planar circuits (Miya 2000). The individual
switch unit in this case consists of two 3 dB-coupler MZ interferometers arranged in
series, with a waveguide intersection. Pairing the interferometers allows any leakage
in the first interferometer into an on-chip absorber. The design of the array matrix
results in path-independent loss and a shorter total loss than a conventional cross–
bar switch. This design was been fabricated into a 16 × 16 array. This large array
was measured to have a 6.6 dB average insertion loss, a 55 dB extinction ratio per
path, and a total electric operating power of 1.1 W. The switch time for each node
is ∼2 ms. Equally impressive arrays have been made in polymeric materials (Eldada
and Shacklette 2000).

13.6 Summary

This chapter has discussed one particularly important application area for large-scale
PICs, namely switching, typically with the goal of selecting optical paths. Electro-
optical switches were one early “contender” for this application because of high speed
and electronic control. However, the advantages of thermo-optic switches in low-cost
passive materials such as polymers or planar SiO2 have made this approach very
attractive. Large-scale arrays have been made of both of these types of switches. The
largest scale optical switch applications use very high I/O count arrays for switching
fiber links. In this case, high extinction ratios are very important. Both requirements
have driven interest in micromechanical-based (MEMS) switches. These are not
discussed in this book, since its focus is in-plane or waveguide devices. Finally,
this chapter has also presented devices fabricated from different materials including
polymers, SiO2 , LiNbO3 , and semiconductors—particularly Si. In fact, in the chapter,
several different Si device sections are presented, since the growth of interest in Si
switching systems has been significant. Si arrays and devices are easy to fabricate
and can be ultrasmall.

Problems

1. Assume single-mode LiNbO3 waveguide with width w≈6 µm, y-cut, directional
coupler.
Design a 1 cm-long β switch, assume free space wavelength λ = 1.55 µm,
electrode spacing d ≈ 2w and field overlap factor  ≈ 0.3, determine the voltage
needed to introduce a π-phase change Vπ .
Problems 315

2. Consider a β modulator with a coupling coefficient κ = 200 1

µm−1 , plot out
the power transfer ratio ηx vs. δ/κ, for κL = 5, where L is the length of coupling
region.
3. If L = 2 cm and R ≈ 60π, determine γ for a lumped circuit modulator and
γtransit .
4. Refer to the β switch in Fig. 13.1, answer the questions below:
a. If the coupling coefficient is κ, what is the minimum length needed for a
β switch?
b. Plot modulation intensity for the cross port of a β switch as a function
of β L/π for a device of length L, where L is the physical length of the
electrodes.

References

Alferness, R. C. (1982). Waveguide electrooptic modulators. IEEE Transactions on Microwave

Theory Techniques, 30, 1121–1137.
Bachmann, M., Smit, M., Besse, P., Melchior, L., et al. (1993). Polarization-insensitive low-voltage
optical waveguide switch using ingaasp/inp four-port mach-zehnder interferometer. In Optical
Fiber Communication Conference (pp. TuH3). Optical Society of America.
Bogert, G., Murphy, E., & Ku, R. (1986). Low crosstalk 4× 4 tilinbo 3 optical switch with per-
manently attached polarization maintaining fiber array. Journal of Lightwave Technology, 4(10),
1542–1545.
Campbell, J., Blum, F., Shaw, D., & Lawley, K. (1975). Gaas electro-optic directional-coupler
switch. Applied Physics Letters, 27(4), 202–205.
Chen, C. P., et al. (2015) Performing intelligent power distribution in a 4×4 silicon photonic switch
fabric. In 2015 IEEE Optical Interconnects Conference (OI). IEEE.
Diemeer, M., Brons, J., & Trommel, E. (1989). Polymeric optical waveguide switch using the
thermooptic effect. Journal of Lightwave Technology, 7(3), 449–453.
Duthie, P., & Wale, M. (1988). Rearrangeably nonblocking 8*8 guided wave optical switch. Elec-
tronics Letters, 24(10), 594–596.
Eldada, L., & Shacklette, L. W. (2000). Advances in polymer integrated optics. IEEE Journal of
Selected Topics in Quantum Electronics, 6(1), 54–68.
Fang, Z., & Zhao, C. Z. (2012) Recent progress in silicon photonics: a review. International Scholarly
Research Notices.
Fischer, U., Zinke, T., Schuppert, B., & Petermann, K. (1994). Singlemode optical switches based
on soi waveguides with large cross-section. Electronics Letters, 30(5), 406–408.
Granestrand, P., Stoltz, B., Thylen, L., Bergvall, K., Döldissen, W., Heinrich, H., et al. (1986).
Strictly nonblocking 8× 8 integrated optical switch matrix. Electronics Letters, 22(15), 816–818.
Haus, H. A. (1984). Waves and fields in optoelectronics. Prentice-Hall.
Izutsu, M., Enokihara, A., & Sueta, T. (1982). Optical-waveguide hybrid coupler. Optics Letters,
7(11), 549–551.
Jenkins, R., Heaton, J., Wight, D., Parker, J., Birbeck, J., Smith, G., et al. (1994). Novel 1× n and
n× n integrated optical switches using self-imaging multimode gaas/algaas waveguides. Applied
Physics Letters, 64(6), 684–686.
Kogelnik, H., & Schmidt, R. V. (1976). Switched directional couplers with alternating δβ. IEEE
Journal of Quantum Electronics, 12(7), 396–401.
316 13 Integrated Optical Switches

Komatsu, K. et al. (1991). 4* 4 gaas/algaas optical matrix switches with uniform device charac-
teristics using alternating delta beta electrooptic guided-wave directional couplers. Journal of
Lightwave Technology, 9(7), 871–878.
Martin, W. E. (1975). A new waveguide switch/modulator for integrated optics. Applied Physics
Letters, 26(10), 562–564.
McCaughan, L., & Bogert, G. (1985). 4× 4 ti: Linbo3 integrated-optical crossbar switch array.
Applied Physics Letters, 47(4), 348–350.
Miya, T. (2000). Silica-based planar lightwave circuits: Passive and thermally active devices. IEEE
Journal of Selected Topics in Quantum Electronics, 6(1), 38–45.
Moosburger, R., Fischbeck, G., Kostrzewa, C., & Petermann, K. (1996). Digital optical switch based
on’oversized’polymer rib waveguides. Electronics Letters, 32(6), 544–545.
Morl, L. et al. (1998). A travelling wave electrode mach-zehnder 40 gb/s demultiplexer based
on strain compensated gainas/alinas tunnelling barrier mqw structure. In Conference Proceed-
ings. 1998 International Conference on Indium Phosphide and Related Materials (Cat. No.
98CH36129) (pp. 403–406). IEEE.
Neyer, A., Mevenkamp, W., and Ctyroky, J. (1986). Single-mode ti: Linbo [sub] 3 [/sub] waveguide
crossings and switches: Design rules and applications. In Integrated Optical Circuit Engineering
III (vol. 651, pp. 169–176). International Society for Optics and Photonics.
Papuchon, M., Combemale, Y., Mathieu, X., Ostrowsky, D., Reiber, L., Roy, A., et al. (1975).
Electrically switched optical directional coupler: Cobra. Applied Physics Letters, 27(5), 289–
291.
Ramaswamy, V., Divino, M., & Standley, R. (1978). Balanced bridge modulator switch using ti-
diffused linbo3 strip waveguides. Applied Physics Letters, 32(10), 644–646.
Schmidt, R., & Kogelnik, H. (1976). Electro-optically switched coupler with stepped δβ reversal
using ti-diffused linbo3 waveguides. Applied Physics Letters, 28(9), 503–506.
Sekiguchi, S., Kurahashi, T., Zhu, L., Kawaguchi, K., & Morito, K. (2012). Compact and low power
operation optical switch using silicon-germanium/silicon hetero-structure waveguide. Optics
Express, 20(8), 8949–8958.
Silberberg, Y., Perlmutter, P., & Baran, J. (1987). Digital optical switch. Applied Physics Letters,
51(16), 1230–1232.
Soldano, L., De Vreede, A., Smit, M., Verbeek, B., Metaal, E., & Green, F. (1994). Mach-zehnder
interferometer polarization splitter in ingaasp/inp. IEEE Photonics Technology Letters, 6(3),
402–405.
Suzuki, K., Tanizawa, K., Matsukawa, T., Cong, G., Kim, S.H., Suda, S., Ohno, M., Chiba, T.,
Tadokoro, H., Yanagihara, M. and Igarashi, Y., Masahara, M., Namiki, S., & Kawashima, H.
(2014) Ultra-compact 8 × 8 strictly-non-blocking Si-wire PILOSS switch, Opt. Express 22, 3887–
3894 (2014)
Zhang, C., Zhang, S., Peters, J.D., & Bowers, J.E. (2016) 8 × 8 × 40Gbps fully integrated silicon
photonic network on chip. Optica 3, 785–786 (2016)
Zhao, P., Chrostowski, J., & Bock, W. J. (1998). Novel multimode coupler switch. Microwave and
Optical Technology Letters, 17(1), 1–7.
Zucker, J., Jones, K., Chiu, T., Tell, B., & Brown-Goebeler, K. (1992). Strained quantum wells
for polarization-independent electrooptic waveguide switches. Journal of Lightwave Technology,
10(12), 1926–1930.
Chapter 14
Numerical Methods

Abstract Due to the complex nature of the light–matter interactions and the ultra-
small scale of many photonic components, analytical solutions of the Maxwell’s
equations in most cases may not exist. Thus, experimental studies rely heavily on
numerical analysis to provide guidance both for the design of the photonic com-
ponents as well as for the interpretation of their performance prior to fabrications.
In most cases, one must first develop a quantitative theoretical description of the
photonic systems using advanced computational techniques, which requires solving
the corresponding partial differential equations numerically. In a broad sense, there
are two categories of modeling methods: finite-difference method (FDM) and finite-
element method (FEM), as well as two categories of equation-solving techniques:
frequency-domain solver, and time-domain simulations. In this section, we briefly
present an overview of the modeling methods and solving techniques.

14.1 Numerical Methods in Nanophotonics

14.1.1 Finite-Difference Versus Finite-Element

In mathematics modeling, FDM are the popular methods for solving differential
equations by approximating them with difference equations, and then use a finite-
difference grid to approximate the derivatives. Due to the simple discretization pro-
cess, the development time for FDM is very short and it is easily understandable
and directly follows from the differential equations. The stability and dispersion and
inhomogeneous characteristics also follow from a simple, intuitive understanding of
the updating procedure. However, the orthogonal grid structure of the FDM implies
that the edges of structures modeling have “stair-step” edges, which can become
problematic for curved surfaces to achieve sufficient accuracy. Special treatments
have been developed to overcome this limitation. These include the nonuniform
grids; however, other methods, such as FEM, are generally better suited for complex
irregular geometries.
On the other hand, the FEM, in general, subdivides a large problem into smaller,
simpler parts that are called finite elements, which are based on triangular or tetrahe-
© Springer Nature Switzerland AG 2021 317
R. Osgood jr. and X. Meng, Principles of Photonic Integrated Circuits,
Graduate Texts in Physics,
https://doi.org/10.1007/978-3-030-65193-0_14
318 14 Numerical Methods

dral sub-regions. The simple equations that model the sub-regions are then assembled
into a larger system of equations that models the entire problem. FEM then minimize
an associated error function to approximate a solution of the system. Note that devel-
oping FEM is not as straightforward as FDM. For example, creating the numerical
grid along for FEM could require an entire software package, and understanding of
the discretization procedure can be quite convoluted.

14.1.2 Time-Domain Versus Frequency-Domain

The solution based on the time-domain can be computed by time stepping, whereas
the same problem in the frequency-domain can be solved only through a linear system
of equations. Using the time-domain solver, the time step, at which we advance the
solution, is limited by the spatial dimensions. Thus, for simulations with large spaces,
the simulation is very computationally expensive. On the other hand, frequency-
domain solvers generally require linear algebra or matrix inversions, and thus, there
is an inherent limit to the size of the simulation, especially for large three dimensional
problems.
The most popular numerical methods for solving Maxwell’s equations or the
wave equation are the combinations of the discretization methods and solving tech-
niques, as illustrated in Fig. 14.1, namely, Finite-Difference Time-Domain (FDTD),
Finite-Difference Frequency-Domain (FDFD), Finite-Element Frequency-Domain
(FEFD), and Finite-Element Time-Domain (FETD). In this chapter, we will provide
a brief overview of each major technique used in photonics design along with an
illustration of each of their use.

Fig. 14.1 Different

numerical methods for
solving the electromagnetic Time Domain
problems
FDTD TDFEM

Finite-Difference Finite-Element

FDFD FEM

Frequency Domain
14.2 The Finite-Difference Time-Domain Method 319

14.2 The Finite-Difference Time-Domain Method

FDTD is a well-known numerical technique in electrodynamics to compute

Maxwell’s equations. It translates the differential form of Maxwell’s equations into
difference equations that can be solved numerically. However, before the 1990s,
the FDTD method was limited by the need to discretize the simulation space on
sub-wavelength scales, with relatively small time steps. Thus, at that time, a typical
photonics problem would require a computer memory that would exceed the tech-
nology limits at that time. However, since the 1990s, the FDTD has become more
computationally affordable with large increases in computer memory and speed.
There are several advantages of FDTD. First, the method is accurate and robust,
such that approximations are minimized and detailed solutions are provided with an
accuracy determined by the grid resolution. Second, the method naturally includes
effects such as polarization, dispersion, and nonlinearities. Furthermore, FDTD is
able to readily calculate the full-wave response, which includes the transient behavior
of an electromagnetic system.
In this section, we first introduce Yee’s unique, yet powerful, algorithm for solving
Maxwell’s equations. Using a 1D example, we demonstrate the basic principle and
formulation of the FDTD method for the analysis of electrodynamic problems. We
then discuss stability analysis, boundary conditions, and the extensions to the analysis
of 2D/3D problems.

14.2.1 Yee’s Algorithm

Yee’s algorithm, introduced in 1966, established a set of finite-difference equations

for the time-dependent Maxwell’s curl equations system (Yee 1966). In this algo-
rithm, the continuous derivatives in space and time are approximated to second-order
accuracy with two-point centered difference forms. The resulting finite-difference
equations are solved via leapfrog stepping, that is, the electric field vector compo-
nents in the modeled space are solved at a given instant in time; then the magnetic
field vector components in the same spatial volume are solved at the next instant in
time using the previously stored electric field data. This process is repeating until the
desired transient or steady-state electrodynamic behavior is fully evolved.
The fundamental unit of the 3D grid, known as the Yee lattice, is shown in Fig. 14.2,
which discretizes and solves the six components of E and H fields that satisfy the
six-coupled scalar Maxwell curl equations in free space:
320 14 Numerical Methods

Fig. 14.2 Position of (i-1,j+1,k+1)

electric and magnetic vector
components in a 3D HZ
staggered unit cell known as
EX
Yee lattice. The vectors are
placed at the point in the
mesh at which they are
defined and stored
HY
EZ

HX
z
y
(i,j,k) EY x

   
∂ Ex 1 ∂ Hz ∂ Hy ∂ Hx 1 ∂ Ey ∂ Ez
= − = −
∂t 0 ∂ y ∂z ∂t μ0 ∂z ∂y
   
∂ Ey 1 ∂ Hx ∂ Hz ∂ Hy 1 ∂ Ez ∂ Ex
= − = − (14.1)
∂t 0 ∂z ∂x ∂t μ0 ∂x ∂z
   
∂ Ez 1 ∂ Hy ∂ Hx ∂ Hz 1 ∂ Ex ∂ Ey
= − = −
∂t 0 ∂x ∂y ∂t μ0 ∂ y ∂x

Rather than solving for the electric field alone with a wave equation, the Yee
algorithm solves the coupled Maxwell’s curl equations directly. In which way, both
electric and magnetic material properties can be easily modeled. This is especially
convenient when modeling the inhomogeneous materials and a full-wave response
of a dispersive medium.
Here, we demonstrate an example of using the FDM to calculate the fundamental
TE mode profile of a channel waveguide. In detail, we illustrate how the computing
grid size can affect the calculated mode profile. Figure 14.3a shows the structure
we calculated with a Si channel waveguide sitting on top of the SiO2 substrate. The
channel waveguide is 400 nm in width and 200 nm in height. Figure 14.3b, c, d shows
the calculated TE mode profile with different grid sizes of 50 nm, 20 nm, and 5 nm,
respectively.

14.2.2 FDTD: 1D Example

In this section, we illustrate the basic implementation of FDTD method for a 1D

case, including details on the discretization of Maxwell’s equations. To solve a spe-
cific problem of pulse propagation, we demonstrate a 1D model with details on the
discretization of Maxwell’s equation. Taking the advantages of the simplicity of 1D
14.2 The Finite-Difference Time-Domain Method 321

1.0
(a) 200nm (b)
Air

0.8
Si

SiO2 0.6

(c) (d) 0.4

0.2

0.0

Fig. 14.3 Demonstration of the calculated mode profile with different computing grid size: a The
structure of a Si channel waveguide (400 nm in width and 200 nm in height) on top of the SiO2
substrate. The fundamental TE mode profile was calculated using different grid size of b 50 nm,
c 20 nm, and d 5 nm. In all cases, λ = 1550 nm

example, we also illustrate the stability criterion for the FDTD simulations. This
discretized equation and stability criterion can be easily expanded to both 2D and
3D models.
In one dimension, the medium extends to infinity in the y-direction and z-direction.
∂
This translational symmetry then leads to ∂x = ∂∂y = 0. For a free-space situation,
Maxwell’s curl equations take the form:
∂ Hz 1 ∂ Ey ∂ Ey 1 ∂ Hz
=− and =− (14.2)
∂t μ0 ∂x ∂t 0 ∂x

In order to introduce 1D Yee discretization, we use the following notation:

E y (m · x, n · t) ≡ (E y )nm

(14.3)
Hz (m · x, n · t) ≡ (Hz )nm

where m and n are the index of the spatial and temporal grids, respectively. Using
the central difference approximation with second-order accuracy for space and time
derivatives, the equations can be discretized as

1 (Hz )m+ 21 − (Hz )m− 21

n+ 21 n− 1 n n
(E y )m − (E y )m 2
=−
t 0 x
(14.4)
(Hz )n+1
m+ 21
− (H ) n
z m+ 1
n+ 21 n+ 1
1 (E y )m+1 − (E y )m 2
2
=−
t μ0 x
322 14 Numerical Methods

Fig. 14.4 Visual illustration

of the numerical t
dependencies in the 1D
FDTD method HZ
n+1
EY
n+1⁄2

n-1⁄2
m-1 m m+1 m+2 x

In a practical sense, for simplicity of the calculations, physical constants are

omitted from mathematical expressions. For example, the implementation of FDTD
√
uses the speed of light c, which is exactly dimensionless 1, i.e., c = 1/ 0 μ0 ≡ 1. By
redefining the electric√field as a typical approach in many theoretical electromagnetic
computations, E y ≡ 0 /μ0 E y (14.4) can be expressed as

n+ 21 n− 21 t  
(E y )m = (E y )m − (Hz )nm+ 1 − (Hz )nm− 1
x 2 2

t   (14.5)
n+ 21 n+ 21
(Hz )n+1 = (Hz )nm+ 1 − (E y )m+1 − (E y )m
m+ 21 2 x
Figure 14.4 is the visual illustration of (14.5), which indicates the numerical depen-
dencies in the 1D FDTD formulation. The value of a field at any point is determined
by three previous values: two from the neighbors of opposite field at the previous
half time step; one from the same position at the previous one time step.
Note that this discretization procedure can be easily extended into 2D and 3D space
(Wartak 2013). However, for example, if 3D problem would require N grid cells in
each dimension, the total grid cells are N 3 . With a minimum of six fields to compute in
double precisions, it can easily take gigabytes of memory with billions of operations.
Thus, FDTD is a computationally intensive method. However, advances in CPU
speed and memory and the emergence of inexpensive parallel systems with parallel
computing technology enable a full 3D FDTD simulation without any constraints.
Despite the requirements for spatial grid size in order to maintain the numerical
accuracy, the time step must be small enough so that it satisfies the Courant condition
(Press 1996) in order to achieve a convergence while solving partial differential
equations numerically. A detailed mathematical discussion can be found in (Taflove
and Hagness 2005). The physical stability criterion is that the speed of numerical
propagation should not exceed the physical speed. Thus, the lattice speed x/t
must be less than the physical velocity v p in the medium with refractive index of n,
where v p = c/n. A summary of stability criteria for various dimensions is presented
in Table 14.1.
14.2 The Finite-Difference Time-Domain Method 323

Table 14.1 Stability criterion for FDTD

Dimensionality Criterion
1D v p t ≤ x
 − 1
2
2D v p t ≤ x1 2 + 1
y 2
 − 1
2
3D v p t ≤ x1 2 + 1
y 2
+ 1
z 2

14.2.3 Boundary Conditions

One of the major challenges of using FDTD method for solving unbounded electro-
magnetic problems is to employ a finite computational domain. Thus, our simulation
domain needs to be terminated with proper boundary conditions. This boundary ter-
mination can be accomplished by introducing an artificial layer to enclose the domain
of interest. However, in order to duplicate the original open-space environment, the
artificial boundary layer has to treat the field incident on top of the layer, so as to
eliminate the artificially reflected fields. There are several approaches to achieve this
implementing a mathematical boundary condition (i.e., absorbing boundary condi-
tion) or a fictitious absorbing material layer (perfect matching layer).

14.2.3.1 Absorbing Boundary Condition

An absorbing boundary condition (ABC) is a mathematical technique such that it acts

to estimate the missing field outside the FDTD domain, and thus emulates an infinite
space. This is normally done by assuming an incident plane wave. Unfortunately,
in many cases, the incident wave at the boundary is usually not a plane wave with
a well-defined angle of incidence. Thus, ABC is a general approximation, which
reflects some of the waves back into the FDTD space. An advanced treatment is
available, such as arbitrary wave can be decomposed into many place waves incident
(Jin 2011).

14.2.3.2 Perfectly Matched Layer

To correctly model the boundary condition, one can define artificial thin layers as
absorbers solely for simulation purposes. A popular absorber model has been pro-
posed by Berenger (1994) for the FDTD simulation, which is named as the perfectly
matched layer (PML). Regardless of the frequency, polarization, and angle of inci-
dence of a plane wave incident upon its interface, PML creates no reflected fields.
As an example to illustrate the idea of PML, the modified source-free Maxwell’s
curl equation for electric field is shown below:
324 14 Numerical Methods

∇s × E = − jωμH (14.6)

where ∇s is defined by
1 ∂ 1 ∂ 1 ∂
∇s = x̂ + ŷ + ẑ (14.7)
sx ∂x sy ∂ y sz ∂z

and ∇s can be considered as the standard ∇ operator in Cartesian coordinates whose

x-, y-, and z-axes are stretched by a complex numbers of sx , s y , and sz , respectively.
Throughout the simulation domain, the complex diagonal tensor is the identity tensor,
but inside the PML, it has the following form (Gedney 1996):
⎡ s y sz ⎤
0 0
⎢ sx ⎥
↔ ⎢ sz s x ⎥
s =⎢
⎢
0 0 ⎥
⎥ (14.8)
⎣ sy
sx s y ⎦
0 0
sz

Note that in a computation, when a material property changes abruptly and the
spatial discretization is not sufficiently dense, undesirable numerical reflections may
occur. One approach to avoiding this problem is to vary the material parameters
smoothly within the PML (Chew and Jin 1996), thus, we have
α − L 
sx,y,z = 1 − j δx,y,z (14.9)
L
where δx,y,z is the loss tangent in dimension x, y and z, α is the distance from the edge,
and L is the thickness of the PML, which is terminated at the simulation domain edge
with a perfect electrical conductor (PEC) boundary condition. Figure 14.5 illustrates
the effect of the PML for a point source in free space. The calculated spatial profile
is shown at time t = 5t0 , t = 7t0 , and t = 10t0 , respectively. As we can see, for the
simulation domain with PML, the spherical wave generated by the point source is
perfectly absorbed at the simulation domain boundary, while for the case without
the PML, the reflection from the boundary significantly interfaces with the spherical
wave generated from the point source.

14.3 Finite-Difference Frequency-Domain (FDFD)

As we discussed earlier, time-domain methods such as FDTD are also extremely

useful for transient behavior analysis. However, when looking into a steady-state
solution at a single frequency, the time-domain method is relatively time-consuming.
Instead, the frequency-domain method, such as the FDFD is highly applicable, since
it maintains the finite-difference spatial features, but removes time stepping (Kunz
14.3 Finite-Difference Frequency-Domain (FDFD) 325

t =5t0 t =7t0 t =10t0

1.0
PML

0.0

without
PML -1.0

Fig. 14.5 Visual illustration of dipole source in free space at time t = 5t0 , t = 7t0 , and t = 10t0 ,
with and without PML, respectively

and Luebbers 1993). The steady-state solution is found at a single frequency through
a matrix inversion process. In addition, the FDFD method has the advantage of
treating dispersive materials. In FDTD, implementing dispersive materials requires
either convolution terms or auxiliary equations, but in FDFD, only one simple set of
values of material properties are needed at the frequency of interest.

14.3.1 FDFD Using the Wave Equations

In this section, by taking the advantages of the simplicity of 1D example, we briefly

illustrate the frequency-domain representation of the wave equation and the possible
implementations through Maxwell’s equations. Thus, in practice, FDFD normally
utilizes the frequency domain wave equation for its setup (Joannopoulos et al. 2011).
The wave equation has a more compact form and does not require interleaving or a
Yee lattice. Meanwhile, only solving one wave equation is necessary; after solving
for E, for example, H can be calculated directly follows from the frequency-domain
Maxwell’s equations.
In the source-free space, the wave equations can be simplified to

∇ 2 E + k 2 E = 0 and ∇ 2 H + k 2 H = 0 (14.10)
√
where k = ω μ is the wavevector. To illustrate how these equations are solved, we
will consider the 1D example with a current source term Jz for generality
326 14 Numerical Methods

∂2 Ez
+ k 2 E z = jωμJz (14.11)
∂x 2
We proceed to discretize the above equation using a second-order centered difference:

(E z )m+1 − 2(E z )m + (E z )m−1

+ k 2 (E z )m = jωμ(Jz )m (14.12)
(x)2

Note that, using the notation defined in (14.3), we need to discretize in space as there
is no time dependence. Thus, there is no superscript n, compared with the notion
used in FDTD. The wavevector k is a complex constant. This difference equation
can be rearranged as

(E z )m−1 + A(E z )m + (E z )m+1 = b(Jz )m (14.13)

where a = [k 2 (x)2 − 2] and b = jωμ(x)2 . The entire system can then be repre-
sented in a tridiagonal matrix:
⎛ ⎞⎛ ⎞ ⎛ ⎞
a 1 0 0 ··· (E z )0 (Jz )0
⎜ ⎟⎜ ⎟ ⎜ ⎟
⎜1 a 1 0 · · ·⎟ ⎜ (E z )1 ⎟ ⎜ (Jz )1 ⎟
⎜ ⎟⎜ ⎟ ⎜ ⎟
⎜0 1 a · · ·⎟ ⎜ ⎟ ⎜ (Jz )2 ⎟
⎜ 1 ⎟ ⎜ (E z )2 ⎟ ⎜ ⎟
⎜ . ⎟⎜ . ⎟ = b⎜ . ⎟ (14.14)
⎜ .. ⎟ ⎜ .
· · ·⎟ ⎜ . ⎟ ⎟ ⎜ . ⎟
⎜ ⎜ . ⎟
⎜ ⎟⎜ ⎟ ⎜ ⎟
⎝· · · 0 1 a 1 ⎠ ⎝(E z )l−1 ⎠ ⎝(Jz )l−1 ⎠
··· 0 0 1 a (E z )l (Jz )l
        
M [E z ] [Jz ]

where the total number of spatial cells in our 1D system is l + 1. The solution of this
equation can be found by simple matrix inversion as

[E z ] = [M]−1 b[Jz ] (14.15)

Note that this 1D example of a homogeneous medium can be easily replaced with
inhomogeneous and frequency-dependent materials, by replacing a, b with am and
bm , where it takes consideration of the material constant such as , μ and σ for each
grid point. From this above example, we see that the solution from FDFD involves
taking an inverse of the matrix M, which can be easily done in 1D simulation. The
discretized FDFD equations in 2D and 3D follow straightforwardly from the 1D
example. However, in a higher dimensional setup, this matrix M can be very large,
thus advanced linear algebra techniques are required for efficient calculation. In
general, the Laplace matrices and the Kronecker products are introduced to assist
the FDFD setup in 2D and 3D models. Detailed discussion can be found in Inan and
Marshall (2011).
14.3 Finite-Difference Frequency-Domain (FDFD) 327

Finally, in the 2D and 3D formulation of this material, directly solving the wave
equation of E field may suffer from a problem with the current source. This is due to
the assumption that the divergence of E and H are both equal to zero. Clearly, this
is not valid at the position of the source. For this reason, it is common to solve the
wave equation of the H field, which can easily include the curl J component due to
the source, while ∇ · H = 0 still holds.

14.3.2 FDFD from Maxwell’s Equations

The FDFD method can also be formulated using Maxwell’s equations. Here, the
frequency-domain Maxwell’s equations are as follows:

∇ × E = − jωμH − M
(14.16)
∇ × H = jωE + J

In the 1D case, we consider an x-directed propagation with E field polarized

along y direction, we have

∂ Ey
= − jωμHz − M
∂x (14.17)
∂ Hz
= jωE y + J
∂x
Discretizing using the leapfrog method, we have the finite difference equations:

(E y )m+1 − (E y )m
= − jωμ(Hz )m+ 21 − M
x
(14.18)
(Hz )m+ 21 − (Hz )m− 21
= jω(E y )m + J
x
The matrix system can then be formulated into [M][F] = [S], where the column
vector [F] includes each of the E components followed by each of the H components
and [S] is the source of the system. This linear system can be rearranged as follows:

ah (Hz )m+ 21 + (E y )m+1 − (E y )m = −x M

(14.19)
ae (E y )m + (Hz )m+ 21 − (Hz )m− 21 = x J

Note that in this setup based on Maxwell’s equations, the matrix M, which must
be inverted, is no longer tridiagonal; this means the inversion process will be more
computationally intensive. In addition, the vector of field values [F] will double in
length compared with the wave equation setup (14.14), since we are simultaneously
solving for E and H. Thus, the doubling of the field vector results in a quadrupling
328 14 Numerical Methods

of the matrix M, which increase the computational cost considerably. Therefore, the
FDFD method from the wave equation, in general, is more attractive than the FDFD
method from Maxwell’s equations.

14.4 Beam Propagation Method

Among the many numerical methods available for modeling optical propagation in
integrated and fiber-optic photonic devices, the Beam Propagation Method (BPM) is
the most commonly used technique for larger photonic systems. BPM is an approx-
imation technique for simulating the propagation of light in slowly varying optical
waveguides (Obayya 2011). It solves the well-known parabolic or paraxial approx-
imation of the Helmholtz equation. There are several reasons for using the BPM
over other numerical methods. First, it is a conceptually straightforward technique
and is easily implemented even in three dimensions. Second, it is a very efficient
method with an optimal computational complexity, i.e., the computational effort is
proportional to the number of grid points used in the simulation. Overall, the BPM is
a very flexible method and requires less intensive computing power compared with
other methods such as FDTD.

14.4.1 Paraxial Formulation

In this section, we demonstrate the simplest version of BPM, where one assumes
scalar electric field E with paraxial approximations. These treatments restrict its
applicability to the fields propagating at small angles with respect to the axis of the
waveguide (Saleh et al. 1991). We define this axis as z axis. To illustrate the method,
we start from the monochromatic wave equation. Assuming a scalar field, φ, and
paraxiality, the wave equation is written in the form of the Helmholtz equation,

∂2φ ∂2φ ∂2φ

+ + 2 + k 2 (x, y, z)φ = 0 (14.20)
∂x 2 ∂ y2 ∂z

where the spatially varying wavenumber is k(x, y, z) = ko n(x, y, z), and the free-
space wavenumber is ko = 2πλ. The refractive index n(x, y, z) solely defines the
geometry of the problem. Considering that the most rapid variation in the field φ
is the phase variation due to propagation predominantly along the z direction, it is
beneficial to factor out this rapid variation by introducing a slowly varying field u,

φ(x, y, z) = u(x, y, z)eiβz (14.21)

where β is a free parameter called the reference wavenumber and is frequently

expressed in terms of a reference refractive index, n 0 , via β = k0 n 0 . Here, n 0 can
14.4 Beam Propagation Method 329

be the refractive index of the substrate or cladding. Substituting (14.21) into (14.20)
gives the equation for the envelope of the field:

∂2u ∂u ∂2u ∂2u

+ 2iβ + + + (k 2 − β 2 )u = 0 (14.22)
∂z 2 ∂z ∂ y2 ∂x 2

By assuming that the variation of u with z is sufficiently slow such that

 ∂2u   ∂u 
   
 2   2β  (14.23)
∂z ∂z

the above equation reduces to

 
∂u i ∂2u ∂2u
= + + (k 2 − β 2 )u (14.24)
∂z 2β ∂x 2 ∂ y2

which is known as a Fresnel or paraxial equation. This approximation eliminates

of the second-order derivative term in z reduces the second-order boundary value
problem to a first-order initial value problem, which can be solved by simple inte-
gration of the above equation along the propagation direction z. In addition, the
efficiency is enhanced by the fact that the longitudinal grid can be much coarser than
the wavelength for many problems.

14.4.2 Finite-Difference BPM

Here, we present the detailed implementation of BPM based on FDM (Scarmozzino

et al. 2000; Scarmozzino and Osgood 1991). The above differential equation can be
numerically integrated in the forward z direction using the Crank–Nicolson scheme,
which is a finite-difference approach and is the most widely used. In this numerical
scheme, the field in the transverse x–y plane is denoted as discrete points on a grid,
and at discrete points along the longitudinal propagation direction z. Given the field
at one z plane, the field at the next z plane can be determined. The stepping process
is repeated to account for the propagation throughout the structure. Assuming a 2D
BPM case, if we let u im denote the field at the transverse grid point i and longitudinal
plane m and assume the the grid points and planes are equally spaced by x and z
apart. Thus, in the Crank–Nicolson scheme, (14.24) is represented at the midplane
between the known plane m and the unknown plane m + 1 as follows:
  m+1
u im+1 − u im i δ2 ui + u im
= + (n 2 2
k − β 2
) (14.25)
z 2β x 2 0
2

where δ 2 is the second-order difference operator, δ 2 u i = [u (i+1) + u (i−1) − 2u i ], and

n is the averaged refractive index between the two planes. The above equation can be
330 14 Numerical Methods

rearranged into the form of a standard tridiagonal matrix equation for the unknown
field u in the plane (n + 1) in terms of known quantities, resulting in
n+1
au i−1 + bu in+1 + cu i+1
n+1
=d (14.26)

where the expressions for the coefficients a, b, c, and d above are readily derived
and can be found in Chung and Dagli (1990).
Boundary conditions. Since the field can only be represented on a finite compu-
tational domain, the above equation requires an appropriated boundary condition,
which completes the system of equations. Transparent boundary condition (TBC) is
commonly used, which assumes the field behaves as an outgoing plane wave near the
boundary. The TBC is very effective in terminating undesired reflections and details
on implementations are given in Hadley (1992b).

14.4.3 BPM Expansions of the Method

There are a few limitations on the traditional BPM that are based on the paraxial
approximations. For example, the fields must propagate primarily along the z axis,
i.e., fields are paraxial and limited to a small angular spread in wavenumber. This
places a restriction on geometries with large and abrupt perturbations along the z-axis.
Also, the gradient of the refractive index must be small. In addition, the elimination of
the second-order derivative term eliminates the possibility of a backward propagating
wave solution; thus, devices relying on large-angle reflections cannot be modeled. In
this section, we briefly introduce a few techniques, that is, to eliminate or significantly
relax these limitations.

14.4.3.1 Wide-Angle BPM

The physical limitation of the above BPM approach results from the parabolic approx-
imation to Helmholtz equation, which implies a paraxiality condition on the primary
direction of propagation. This restriction and the related restrictions on index contrast
can be addressed using the wide-angle BPM, which is an approach to incorporate the
effect of the second-order derivative term that was neglected in the basic BPM, thus
a more accurate approximation to Helmholtz equation. The most popular formula-
tion is based on Padé approximants (Hadley 1992c). In general, larger angles, higher
index contrast, and more complex mode interference can be analyzed in both guided-
wave and free-space problems as the Padé order increases. Detailed discussion for
using this technique can be found in Hadley (1992a).
In addition. various bidirectional BPM techniques have been considered to include
the backward traveling wave, with most focusing on the reflection of a wave along z-
direction. For example, the guided-wave propagation can be divided into regions that
are uniformed along z. Both forward and backward waves can exist at any point along
14.4 Beam Propagation Method 331

the structure. The essential idea is to employ a transfer matrix M  which describes the
entire structure that is composed of propagation and interface matrices (Kaczmarski
and Lagasse 1988; Rao et al. 1999).

14.4.3.2 Full-Vector BPM

The basic BPM approach discussed above results from the assumption of scalar
waves, which prevents the polarization effects from being considered. This limitation
can be overcome through a full-vector BPM technique, which is to recognize the
electric field as a vector and solving from the vector wave equation rather than the
scalar Helmholtz equation. This approach can be found with more details in Huang
and Xu (1993), Xu et al. (1994).

14.5 Finite-Element Methods

The FEM was originally developed for mechanical and structural analyses in the
1950s. It became popular in solving the vector electromagnetic problems after an
important breakthrough occurred in the 1980s with the development of an edge-based
vector element (Nédélec 1980; Barton and Cendes 1987). There is a major difference
between the FDM and the FEM. From the discussions earlier in this chapter, we note
that, in principles, FDM finds an approximation to the differential operators, and
then use these difference equations to solve for the fields at each grid.
In the sections below, we illustrate the basic principle of the FEM by briefly
introducing the methods for solving the boundary-value problems in mathemati-
cal modeling. Then we present the formulation procedure of the FEM to solve the
electromagnetic problems in the frequency-domain.

14.5.1 Boundary-Value Problems

Boundary-value problems have long been a major topic in mathematical modeling. A

typical boundary-value problem can be defined by a governing differential equation
in a domain , with boundary conditions specified on the boundary that encloses the
domain: Lψ = f , where L is a differential operator, f is the source function, and
ψ is the unknown quantity. In electromagnetics, the form of the governing differ-
ential equation ranges from a simple Poisson equation to complicated vector wave
equations.
To solving the boundary-value problems, various approximate methods have been
developed, and among them, the Ritz and Galerkin’s methods have been used most
widely (Axelsson and Barker 2001). The Ritz method is a direct method to find an
approximate solution for boundary-value problems. It is a variational method which
332 14 Numerical Methods

starts from the variational representation, which is referred to as functional, of the

boundary-value problem. The minimum of the functional corresponds to the gov-
erning differential equation under the given boundary conditions. The approximate
solution is then obtained by minimizing the functional with respect to its variables.
On the other hand, Galerkin’s method belongs to the family of weighted residual
methods, which start directly from the partial differential equation of the boundary-
value problem and seek the solution by weighting the residual of the differential
equation. In this method, a continuous operator problem is converted into a discrete
problem, then characterizes the discrete space with a finite set of basis functions, i.e.,
the weighting function, used for the expansion of the approximate solution. A brief
review of the Ritz and Galerkin’s methods and a detailed illustration of their solution
procedures to a simple boundary-value problem can be found in Jin (2015).

14.5.2 Implementation of FEM

In general, it is a very challenging step in the Ritz and Galerkin’s methods to find a trial
function defined over the entire solution domain, which is capable of representing
the true solution to the problem. This is particularly true for 2D and 3D problems.
To simplify the problem, we can divide the entire domain into small subdomains
and employ the trial functions defined over each subdomain. These trial functions
are usually in a much simpler form compared with original equations in the entire
domain since the subdomains are small. Therefore, the principle of the FEM is
to replace an entire continuous domain by a number of subdomains, in which the
unknown function is represented by simple interpolation functions with unknown
coefficients. A system of equations is then obtained by applying the Ritz variational
or Galerkin’s procedure and the solution of the boundary-value problem is achieved
by solving the system of equations. The basic steps for a finite element method can
be summarized as follows:

• Domain discretization. The first and perhaps the most important step in the FEM is
to discretize the domain over which the solution is desired. An effective discretiza-
tion with proper numbering for each element can significantly affect the computing
time, memory usage, as well as accuracy of the numerical results (Jin 2011). Note
that the linear line segments, triangles, and tetrahedral are the most frequently
used subdomain elements for 1D, 2D, and 3D modeling, due to their simplicity
and suitability for domains with arbitrary shape and volume. We demonstrate two
examples in Fig. 14.6 showing the finite element discretization of a 2D and a 3D
domains.
• Select interpolation functions. The approximation of the unknown solution is
assumed to take a specific functional form over each small element. In general,
the interpolation is usually selected to be the linear (first-order polynomial) or
parabolic (second-order) functions. These functions are then matched to the adja-
cent cells to ensure continuity across the cell boundaries
14.5 Finite-Element Methods 333

Fig. 14.6 Visual illustration of the finite element mesh grid for 2D (left) and 3D (right) object

• Formulate a system of equations. In this step, each elemental equation can be

formulated using either the Ritz variational or Galerkin’s method. The system of
equations can then be set up by summing the elemental equations over the entire
domain. The boundary conditions are then imposed to obtain the final form of
the system of equations. This system includes information about boundaries and
sources, also have the added constraint of continuity across element boundaries.
• Solve the system of equations. The resultant system, in general, has one of the
following two forms:
[M][φ] = [S] (14.27)

or
[A][φ] = λ[B][φ] (14.28)

In electromagnetics, (14.27) is corresponding to the wave equations with the known

vector [S] as the source. Equation 14.28 represents the eigenvalue systems that
associated with source-free problems. In this case, the source vector [S] vanishes
and the matrix [K ] can be written as [A] − λ[B], where λ denotes the unknown
eigenvalues. Similar to FDFD, solving these systems becomes purely a linear
algebra problem and the truncation the infinitely large solution domain into a
finite computation domain is taking cared by setting up an artificial mesh layer
with either ABC or PML.

14.6 Summary

In this chapter, we have briefly described the basic principle and formulation of a few
major computational methods for the numerical analysis of electromagnetic fields
334 14 Numerical Methods

for photonics applications. These include the FDTD method, FDFD method, beam
propagation method, and the FEM. We started the discussion with the construction
of the finite differencing formulae and demonstrated their applications in solving
1D Maxwell’s equations. Taking the advantage of the simplicity of 1D problems,
we then discussed the working principles, stability criterion, and boundary condi-
tions in the time-domain simulation. Further, the FDM was extended to frequency-
domain, where the linear system of equations can be formulated. In addition, we
have illustrated the basic principle and steps of FEM with possible applications to
electromagnetic problems. These four methods are chosen because they represent
the fundamental and popular approaches for numerical analysis of photonics engi-
neering design. The reader is also encouraged to consult more advanced books and
references listed in the end of this chapter, for a more comprehensive understanding
of these methods with a variety of advanced treatments and applications.

References

Axelsson, O., & Barker, V. A. (2001). Finite element solution of boundary value problems: Theory
and computation. Philadelphia: SIAM.
Barton, M., & Cendes, Z. (1987). New vector finite elements for three-dimensional magnetic field
computation. Journal of Applied Physics, 61(8), 3919–3921.
Berenger, J.-P. (1994). A perfectly matched layer for the absorption of electromagnetic waves.
Journal of Computational Physics, 114(2), 185–200.
Chew, W., & Jin, J. (1996). Perfectly matched layers in the discretized space: An analysis and
optimization. Electromagnetics, 16(4), 325–340.
Chung, Y., & Dagli, N. (1990). An assessment of finite difference beam propagation method. IEEE
Journal of Quantum Electronics, 26(8), 1335–1339.
Gedney, S. D. (1996). An anisotropic perfectly matched layer-absorbing medium for the truncation
of FDTD lattices. IEEE Transactions on Antennas and Propagation, 44(12), 1630–1639.
Hadley, G. R. (1992a). Multistep method for wide-angle beam propagation. Optics Letters, 17(24),
1743–1745.
Hadley, G. R. (1992b). Transparent boundary condition for the beam propagation method. IEEE
Journal of Quantum Electronics, 28(1), 363–370.
Hadley, G. R. (1992c). Wide-angle beam propagation using padé approximant operators. Optics
Letters, 17(20), 1426–1428.
Huang, W., & Xu, C. (1993). Simulation of three-dimensional optical waveguides by a full-vector
beam propagation method. IEEE Journal of Quantum Electronics, 29(10), 2639–2649.
Inan, U. S., & Marshall, R. A. (2011). Numerical electromagnetics: The FDTD method. Cambridge:
Cambridge University Press.
Jin, J.-M. (2011). Theory and computation of electromagnetic fields. Hoboken: Wiley.
Jin, J.-M. (2015). The finite element method in electromagnetics. Hoboken: Wiley.
Joannopoulos, J. D., Johnson, S. G., Winn, J. N., & Meade, R. D. (2011). Photonic crystals: Molding
the flow of light. Princeton: Princeton University Press.
Kaczmarski, P., & Lagasse, P. (1988). Bidirectional beam propagation method. Electronics Letters,
24(11), 675–676.
Kunz, K. S., & Luebbers, R. J. (1993). The finite difference time domain method for electromagnetics.
Boca Raton: CRC Press.
Nédélec, J.-C. (1980). Mixed finite elements in r3. Numerische Mathematik, 35(3), 315–341.
Obayya, S. (2011). Computational photonics. Hoboken: Wiley.
References 335

Press, W. H. (1996). FORTRAN numerical recipes: Numerical recipes in FORTRAN 90. Cambridge:
Cambridge University Press.
Rao, H., Scarmozzino, R., & Osgood, R. M. (1999). A bidirectional beam propagation method for
multiple dielectric interfaces. IEEE Photonics Technology Letters, 11(7), 830–832.
Saleh, B. E., Teich, M. C., & Saleh, B. E. (1991). Fundamentals of photonics (Vol. 22). New York:
Wiley.
Scarmozzino, R., Gopinath, A., Pregla, R., & Helfert, S. (2000). Numerical techniques for modeling
guided-wave photonic devices. IEEE Journal of Selected Topics in Quantum Electronics, 6(1),
150–162.
Scarmozzino, R., & Osgood, R. (1991). Comparison of finite-difference and Fourier-transform
solutions of the parabolic wave equation with emphasis on integrated-optics applications. JOSA
A, 8(5), 724–731.
Taflove, A., & Hagness, S. C. (2005). Computational electrodynamics. Boston: Artech House.
Wartak, M. S. (2013). Computational photonics: An introduction with MATLAB. Cambridge: Cam-
bridge University Press.
Xu, C., Huang, W., Chrostowski, J., & Chaudhuri, S. (1994). A full-vectorial beam propagation
method for anisotropic waveguides. Journal of Lightwave Technology, 12(11), 1926–1931.
Yee, K. (1966). Numerical solution of initial boundary value problems involving Maxwell’s equa-
tions in isotropic media. IEEE Transactions on Antennas and Propagation, 14(3), 302–307.
Appendix
Problem Set Solutions

Chapter 1

Chapter 2

1. According to ITU-T G.652, the FWHM of the mode is 8.6 ∼ 9.5 µm. Its mode
distribution is approximately Gaussian (in fact, it is in the shape of J0 Bessel
function).
2. According to batop.de
λ/µm x n
1.3 0.085 0.04
0.9 0.045 0.04

3. Consider at point (x, y) = (0, 0), index change can be obtained by

n 0 ≈ b · C0

where
2 τ w
C0 = √ erf
πd 2d

and
(λ) = 0.552 + 0.065/λ2
= 0.552 + 0.065/1.32
= 0.59

Plug in b and C0 , we have

1.18 τ  w 
n 0 = √ erf
π d 2d
© Springer Nature Switzerland AG 2021 337
R. Osgood jr. and X. Meng, Principles of Photonic Integrated Circuits,
Graduate Texts in Physics,
https://doi.org/10.1007/978-3-030-65193-0
338 Appendix: Problem Set Solutions

w
assume  1, we have
2d
1.18 τ
n 0 = √
π d

Plug in n 0 = 10−3 and τ = 900 Å, we have

d = 59.9 µm
√
Remind that d = 2 Dt and D = D0 exp(−T0 /T ), thus,

d2 (59.9 × 10−4 )2
t= = −4
4D0 exp(−T0 /T ) 4 × 2.5 × 10 × exp(−2.5 × 104 /1323.15)
= 5761640 s = 1600 h

We will need 67 days to obtain an index change of 0.001!

4. Helmholtz equation:
∇ 2 E x + (ko n x )2 E x = 0
√ √
In homogeneous material, we have n x = n = ε/ε0 and ko = ω/c = ε0 μ0 ω,
where ε0 is the vacuum permittivity. Thus,

∂2 
E x + (nko )2 E x = 0
∂z 2

The solution to the equation above is simply

E x = Ae− j (nko )z + Be j (nko )z

where the propagation in the direction of +z requires that B = 0. Considering

the wave having a frequency of ωo , the full solution for E x is
√
E x = Ae j[ωo (t− εμ0 z)]

5. (a) Lithorgraphy.
(b) Diffusion.

Chapter 3

1. From batop.de, we have n f (λo = 0.84 µm) = 3.65, n c (λo = 0.84 µm) = 3.59.
Thus,
Appendix: Problem Set Solutions 339

V = ko d n 2f − n 2s
2π  2
= d n f − n 2s
λo
2 × 3.14 
= × 0.2 × 3.652 − 3.592
0.84
≈ 0.985

Remind that phase equation for waveguide mode

 
√ b b+a
V 1 − b = mπ + arctan + arctan
1−b 1−b

where m = 0 for TE0 mode and a = 0 for symmetric waveguide.

Plug V in the equation above and solve for b yields b ≈ 0.185. Therefore,

n eff = b(n 2f − n 2c ) + n 2c ≈ 3.60

Thus, 
γ = ko n 2eff − n 2c ≈ 2.12 × 106 m−1

2. (a) The diffusion coefficient can be calculated by

D = D0 exp(−T0 /T ) ≈ 7.4 × 10−13 cm2 /s

and therefore the diffusion distance is

 
d = 2 Dt0 = 2 7.4 × 10−17 × 6 × 3600 ≈ 2.53 µm

The concentration at y = 0.6 µm is then

 2

2 τ y 2 0.08 0.62
C(y) = √ exp − 2 = √ × × exp − ≈ 0.034
πd d 3.14 2.53 2.532

Therefore, the index change would be

n(λ) = b(λ)C(y) = (0.552 + 0.065/λ2 ) × 0.034 = 0.0188 + 0.00221/λ2

(b) At λ = 1.3 µm, b = 0.552 + 0.065/1.32 = 0.59

Thus, the index change of the material along y is

n(y) = b(λ)C(y) = 0.0211 exp(−y 2 /d 2 )

340 Appendix: Problem Set Solutions

and
n s = 2.2204

n f = n s + n(0) = 2.2204 + 0.0211 = 2.2415

For graded index guide, the phase equation becomes

√  3
Vcg,m = 2π m +
4
where
2π  2
Vd = d n f − n 2s = 3.752
λ
Thus,
Vd 3
m ≤ √ − = 0.747
2π 4

which means that only 1 mode (m = 0) is supported.

3. Given that x = 0.9, λ = 0.84 µm:

n f = n(GaAs) = 3.645
n s = n(Ga0.9 Al0.1 As) = 3.590
n c = n(air ) = 1

**Sketch of mode profile


V = ko d n 2f − n 2s ≈ 0.94

V = 0.94 < π tells us that onlythe fundamental  mode is supported.

√ b b+a
by solving V 1 − b = arctan + arctan we can obtain b, thus,
1−b 1−b
we have n eff , and furthermore, β, also including δ, γ.

1 1
Teff = + +d
γ δ

γ −1
Therefore, the percentage is
γ −1 + d + δ −1
4. xpressing the relations of δ, γ, and κ in terms of β and κ, i.e., (3.12), and inserting
them into (3.16), we obtain an eigenvalue equation for the allowed βs.
5. The simplest way to do this would be to use the normalized frequency V . Accord-
ing to the definition of V , it implies how many modes a specific waveguide geom-
etry can support. In this case, according to (3.19), we obtain
Appendix: Problem Set Solutions 341

2πd  2
V = n f − n 2s = 45.5 (S.3.1)
λ
Use (3.27), then gives us the highest order mode index m
 
V
m= = 14 (S.3.2)
π

where the symbol “[a]” means to get the largest integer not larger than a. Recall
the total mode number is m + 1, so the total allowed mode number is 15.
Another approach is using the physical sense of ray picture of wave propagation
in waveguides, as shown below.
At the core and cladding interface, the light must have a total internal reflection,
which means

n f sin θ > n s (S.3.3)

where θ is the angle between the light and the interface, as shown in the picture
below. We then obtain the following relation using (S.3.3):
2πn f
κ= cos θ
λ
2π  2
(S.3.4)
< n f − n 2s = 3.03 µm−1
λ
We know that there should be a wave in the transverse direction with integer
number of periods, as shown in the following picture (Fig. A.1).
So the largest order mode this waveguide can support is
 
3.03 µm−1 × 15 µm
m= = 14 (S.3.5)
π

Again, we get the same result. Note that the form of κd is the same as normalized
frequency V , which should remind us of what the physical meaning of V is.

Fig. A.1 Ray picture of wave propagation in a waveguide

342 Appendix: Problem Set Solutions

6. According to (3.20),

κ
−1 κ
κw = (m + 1)π − tan − tan−1 (S.3.6)
γ δ

where the parameters are defined as in textbook. In the case where n f  n c ,

which means a good mode confinement, we obtain

κ π
κ
tan−1 = tan−1 = (S.3.7)
γ δ 2

Plug this into (S.3.6), it turns out that

mπ
κ= (S.3.8)
w
Then plugging this into (3.13) yields

2
2πn f  mπ 2
βm = − (S.3.9)
λ0 w

7. (a) This is a symmetric waveguide with a = 0, (3.27) yields

Vcm = mπ (S.3.10)

So, for m = 1, the cutoff normalized frequency is Vc(1) = π. Plugging this

into (3.19) yields

V (1) λ
ddesign = 0.5dcutof f = 0.5 c = 4.84 µm (S.3.11)
2π n 2f − n 2s

(b) Calculate b first. From (a), we have V = π/2 = 1.57, from the normalized
diagram, we obtain b ≈ 0.3. Plugging this into (3.24) gives

Nef f = b(n 2f − n 2s ) + n 2s = 1.5006 (S.3.12)

So, β = Nef f k0 = 6.29 µm−1 .

(c) According to (3.13),

δ=γ= β 2 − (n s · k0 )2 = 0.29 µm−1 (S.3.13)
Appendix: Problem Set Solutions 343

1
The decay length is L decay = = 3.45 µm. Note this is a quite large decay
δ
length, as we can see in this case, the mode is not well confined in the waveg-
uide.
8. Consider low index contrast, the TM modes can be treated as TE modes in effective
index method. First calculate the asymmetry parameter a:

n 2s − n 2c
a= = 76.88 ≈ ∞ (S.3.14)
n 2f − n 2s

Plug this into (3.27), which yields

√ 3
Vcm = mπ + tan−1 a= π (S.3.15)
2
Then we could easily get the desired thickness d

λ
d=  · 0.9Vcm = 1.83 µm (S.3.16)
2π n 2f − n 2s

where the normalized frequency V = 1.5π × 0.9 = 4.24. According to the nor-
malized parameter diagram, we have b = 0.64. Plug into (3.24), we obtain

ne f f 1 = b(n 2f − n 2s ) + n 2s = 3.393 (S.3.17)

Then the effective waveguide structure is shown below. To get the 10% reduction
to the cutoff frequency, V = 0.9π, similar to the previous procedure, we get

0.9π · λ
w=  = 1.517 µm (S.3.18)
2π n 2e f f 1 − n 2s

Also, we obtain from the normalized parameter diagram that

b = 0.6 (S.3.19)

So, eventually 
ne f f = b(n 2e f f 1 − n 2s ) + n 2s = 3.388 (S.3.20)
344 Appendix: Problem Set Solutions

Chapter 4

1. Step one: we divide the waveguide section along y into three regions
(Figs. A.2 and A.3):

Fig. A.2 Define the waveguide regions

Fig. A.3 Calculated the effective index for each region

In Region I:
 2π  2
VI = ko d n 2f − n 2s = d n f − n 2s = 1.26d
λo

n 4f n 2s − n 2c 3.412 − 12
a= = 3.414
× ≈ 2.1 × 104
n 4c n 2f − n 2s 3.412 − 3.42

By solving the phase equation,

 
√ b b+a
V 1 − b = mπ + arctan + arctan
1−b 1−b

and plug in b = 0, m = 1 we obtain that

√ 3
V = π + arctan a≈ π
2
Appendix: Problem Set Solutions 345

V
Therefore, dc = ≈ 3.74 µm.
1.26
ddesign = 0.8dc ≈ 3 µm

VI = 1.26ddesign = 3.78

Plugging VI in the phase equation yields

b I ≈ 0.58

Therefore, 
n eff,1 = n 2s + b I (n 2f − n 2s ) ≈ 3.4

In Regions II and III,

 2π  2
VI I = VI I I = ko h n 2f − n 2s = d n f − n 2s = 1.26 h
λo

Similarly, we have b I I = b I I I ≈ 0.28 and n eff,2 = n eff,3 = 3.40.

Step two: where a = 0, Vc,1 = π. By solving for wc from
2π  2
Vc,1 = wc n eff,1 − n 2eff,2 = 0.69wc ,
λo
π
we obtain that wdesign = 0.8 × = 3.64 µm.
0.69
2. Similarly, we can divide the waveguide into three parts (Fig. A.4):
Step one, in Region I,
 2π  2
VI = ko t n 2f − n 2s = t n f − n 2s = 6.27t
λo

Fig. A.4 Define the waveguide regions

346 Appendix: Problem Set Solutions

Fig. A.5 Calculated the effective index for each region

n 4f n 2s − n 2c 1.964 1.472 − 1.392

a= = × ≈ 0.54
nc n f − ns
4 2 2 1.394 1.962 − 1.472

√
Vc,I = π + arctan a ≈ 3.78

3.78
t=
× 0.8 = 0.48 µm
6.27

Solve for b yields b = 0.597, thus n eff = n 2s + b(n 2f − n 2s ) = 1.78
Step two (Fig. A.5):
Note that we have a = 0, Vc = π. Therefore,

2π  2
Vc = w n f − n 2s = 4.84w
λo

0.8Vc 0.8π
w= = = 0.52 µm
4.84 4.84
3. g
VI
dx =
2π  2
n f x − n 2s
λ
g
g 7√ 0.8VI
where VI,c = 2π. Thus dx =  ≈ 6.28 µm
4 2π
n 2f x − n 2s
λ
The phase equation for graded index guides is
 xt     21  
g
d x x 3
2V f −b d − π = 2mπ
0 d d 2
Appendix: Problem Set Solutions 347

By solving the equation, we obtain that b ≈ 0.29. You can use numerical method
or graphical method to solve for b. Then,

n eff = n 2s + b(n 2f − n 2s ) ≈ 2.2206

2g 11
Therefore, VI I = 0.8 × 3 = 1.504.
8

1.504
dy = = 6.96 µm.
2π √
2.22062 − 2.222
λo

4. 
n eff = n 2s + b(n 2f − n 2s )
n 
f
= ns + b (n f − n s ) + . . .
ns (4.21)
≈ n s + b(n f − n s )
Q.E.D.

Chapter 5

1.
j2γk x2
κ12 = e−γ(h−d)
β(k x2 + γ 2 )(2/γ)
⎧ 
⎪
⎪ k = k n 2f − n 2eff
⎪
⎨
x o

⎪
⎪ γ = k o n 2eff − n 2s
⎪
⎩
β = ko n eff

For V being 20% below cutoff for the m = 1 mode, V = 0.8π.


ko d n 2f − n 2s = 0.8π

Solving the above equation for d yields d ≈ 1.16 µm. Also by solving the normal-
ized phase equation, we can obtain b ≈ 0.554. Therefore, n eff =

n 2s + b(n 2f − n 2s ) ≈ 3.32.
Plug n eff into the equations above for k x , γ, we can obtain h by
348 Appendix: Problem Set Solutions

1 κ12 β(k x2 + γ 2 )(2/γ)

h = − log + d ≈ 2.73 µm
γ 2γk x2

π
where κ12 = .
2L c
2. (a)
√ 3 6
Similar to Problem 1: V = 0.8(π + arctan a) ≈ 0.8 × π = π. Also note
 2 5
that V = ko t n 2f − n 2s yields that t ≈ 9.4 µm

Solve for b yields that b I ≈ 0.293, and n eff,I = n 2s + b(n 2f − n 2s ) ≈ 1.52788.
(b)

V = 0.8π = ko w n 2eff,I − n 2s

Solve for w yields that w ≈ 11.57 µm.

(c)
Plugging in V = 0.8π, a = 0 into normalized phase equation, we can obtain that
b I I = 0.554, thus n eff,II ≈ 1.52749, and
⎧ 
⎪
⎪ k = k o n f − n eff ≈ 1.40 × 10 m
2 2 5 −1
⎪
⎨
x

⎪ γ = ko n 2eff − n 2s ≈ 1.56 × 105 m−1
⎪
⎪
⎩
β = ko n eff ≈ 6.19 × 106 m−1

Thus,
1 κ12 β(k x2 + γ 2 )(2/γ)
h = − log + w ≈ 14.94 µm
γ 2γk x2

|a1 (z)|2
3. For a codirectional coupler, the power transfer is =
|a1 (0)|2
 κ 2
1− sin2 (βc z), where
βc
  β − β 2 √
2 1 13
βc = κ2 + = κ
2 2

Maximum power transfer is

 κ 2  2 2 9
1− =1− √ =
βc 13 13

4. Equation (5.42):
Appendix: Problem Set Solutions 349

π d n f − n eff
2 2
κ= j
λ Teff n 2eff

2
From Problem 1, we have γ = 1.55 × 106 m−1 , thus, Teff = d + ≈ 2.45 µm.
γ
Therefore,
κλTeff n 2eff
d = ≈ 217.7 nm
π(n 2f − n 2eff )

5. (a) i. If β1 = β2 , the power transfer between the two waveguides becomes

perfectly matched with detuning δ = 0, according to (5.18), we obtain

P1 = P cos2 κz
(5.22)
P2 = P sin2 κz

where P is the incident power from waveguide 1, κ is the coupling

coefficient.
ii. If β1  β2 , recall when the two waveguides have a detuning δ, the
power transfers more frequently with less power evolved in this pro-
cess. According to (5.18), the power from waveguides 1 and 2 can be
expressed as ⎧
⎪ 2 
⎪
⎨ P1 = P 2 cos 2
κ2 + 2 z
κ + 2
 (5.23)
⎪
⎩ P2 = P κ
2
⎪
sin 2
κ 2 + 2 z
κ2 + 2

where  = β1 −β2
2
is the propagation constant difference in the two waveg-
uides.
In the limit where β1  β2 , we get  ≈ β1 /2. In a waveguide, κ
β.
Plug these into (5.23), we obtain P1 = P and P2 = 0.
(b) According to (5.15),

 β1 + β2 β1 − β2 2
β1,2 = ± − κ2 (5.24)
2 2

where β1,2 and β1,2 are the modes in a single waveguide and the coupled
mode of the system, respectively. Here, “+” for the even and “−” for the odd
system mode (Figs. A.6 and A.7).
6. (a)
350 Appendix: Problem Set Solutions

Fig. A.6 Power transfer versus propagation length

Fig. A.7 Even and odd

modes in directional coupler

λ 0.8Vc1
d= 
2π n 2 − n 2
f s

1.55 0.8π (5.25)

= √
2π 3.42 − 3.3872
= 2.087 µm

(b) According to (5.18), β = κ in this case,

π π
L min = = = 0.0314 cm = 31.4 µm (5.26)
2κ 2 × 50 cm−1

For all possible length, we could express it as L = (2m + 1)L min , where m
is an integer number.
Appendix: Problem Set Solutions 351

(c) When d = 2.087 µm, V = 0.8π, from the normalized parameter diagram,
we obtain that b ≈ 0.55. Plugging this into (3.24) yields

n e f f = 0.55(3.42 − 3.3872 ) + 3.3872 = 3.394 (5.27)

So the propagation constant for waveguide 1 is

2π
β1 = n e f f = 13.758 µm−1 (5.28)
λ
Similarly, we can obtain propagation constant β2 of waveguide 2 as
2π 
β2 = · b2 (n 2f − n 2s ) = 13.74 µm−1 (5.29)
λ
In order to use (5.18), we determine βc first

β2 − β1 2
βc = κ + 2 = 0.064 cm−1 (5.30)
2

According to (5.18), we then obtain

π π
L min = = = 0.025 cm = 25 µm (5.31)
2βc 2 × 64 cm−1

So all the lengths that have a maximum power transfer can be expressed as
L  = (2m + 1)L min , where m is an integer. Also note that in such case, the
 2
power that can be transferred reduces by a factor of βκc ≈ 0.61 due to the
asymmetry structure of such directional coupler.

Chapter 6

Chapter 7

1. (a) First, we determine n eff using normalized parameter

2π  2 2π 
V = w n f − n 2s = × 4 µm × 3.42 − 3.382 = 6.17 (7.32)
λ 1.5 µm

In this case, a = 0 as n s = n c . From the normalized parameter diagram, we

can get b ≈ 0.85. We then obtain

n eff = n 2s + b(n 2f − n 2s ) = 3.397 (7.33)
352 Appendix: Problem Set Solutions

According to (3.30), the field has the form of

⎧ −δx
⎪
⎪ Ae ,x > 0
⎪
⎪

⎪
⎨ δ
A cos κx − sin κx , −d ≤ x < 0
κ (7.34)
⎪
⎪

⎪
⎪ δ
⎪
⎩ A cos κd + sin κd eγ(x+d) , x < −d
κ

where ⎧
⎪ 2π  2
⎨κ = n f − n 2eff = 0.598 µm−1
λ
 (7.35)
⎪
⎩ γ = δ = 2π n 2 − n 2 = 1.422 µm−1
λ eff s,c

According to (3.33),

4κ2 ωμ0 P
A2 =
(7.36)
1 1
|β|(κ2 + δ 2 ) d + +
γ δ

Thus, we obtain that A = 3.51V / µm.

(b) According to (7.41),
u = Rt ln(r/R) ≈ −R + r (7.37)

(c) According to (7.44), we obtain



2γt22 exp(γw) 2γ 3
α≈ exp − 2R = 1.22 × 10−42 µm−1
(Neff − n 2s )k 2 β(2 + γw)
2 3β
(7.38)
Note this is a quite small value for the loss constant α.
2. In this case, the mode is well confined to the waveguide. So the width of waveguide
is 4 × 1.5 µm = 6 µm. According to (7.13),

(3α/4π)2
= 0.2 ⇒ α = 1.5π (7.39)
(3α/4π)2 + 1

Also we know
2π
λg = = 0.44 µm (7.40)
β0

Thus, according to (7.12), we obtain

1/2
4
w = 16 + π × 0.44 × L =6 ⇒ L = 12.24 µm (7.41)
3
Appendix: Problem Set Solutions 353

3. Here are several approaches to this; here, we show one using the ray optics picture.
Refer to the picture below, the whispering gallery mode (WGM) appears when
the light can only hit on one side of the ring waveguide.
According to the geometry relation, the following condition needs to be satisfied
when there is a WGM:

w > R − R sin θ (7.42)

Plug in the following expression

β n eff
sin θ = = (7.43)
k nf

One obtains
n f − n eff
w> R (7.44)
nf

Since δn = n f − n s > n f − n eff , we obtain

δn n f − n eff
> (7.45)
n s + δn nf

For all the mode in the ring waveguide to be WGM, we need to make the highest
order mode to satisfy (7.44). The higher order mode will have a refractive index
more close to n s , which means a strict condition for all the modes are WGM’s
should be
δn
w> R (7.46)
n s + δn

4. According to (7.6),

4
t=  T
wx wy
a
+ a
wx a
+ a
wy
N f Ng 4
=    (7.47)
N f + Ng wx + a wy
+ a
a wx a wy

= 0.382
⇒ t = 0.146
2

It yields the loss l in dB is

l = 10 log10 (0.146) = −8.36 dB (7.48)

5. (a) The modes are shown in Fig. A.9 with label.

354 Appendix: Problem Set Solutions

Note that at the transition region, the mode of straight waveguide needs to
couple to the mode of curved waveguide. But these two have a mode shape
mismatch, which leads to a loss.
(b) Usually, we will have a displacement to allow better mode matching, such as
in Fig. 7.27.
6. Instead of an abrupt transition design, we would prefer to use an adiabatic design,
which means to change gradually from one waveguide to the other. Intuitively,
we can understand that this will give more “time” and “space” for the modes to
evolve from mode in waveguide 1 to that in waveguide 2 (Fig. A.8).
Let’s first consider the difference between the above two structures. According to
(7.6), we have the transmission rate t1,2 of cases 1 and 2 are

Fig. A.8 Whispering gallery

modes

Fig. A.9 Mode shapes in

straight and curved
waveguides
Appendix: Problem Set Solutions 355
⎧
⎪ 2
⎪
⎨ t1 = = 0.38
w/5w + 5w/w
(7.49)
⎪
⎪ 2 2
⎩ t2 = = 0.53 > t1
w/3w + 3w/w 3w/5w + 5w/3w

This provides us a special case, where we add a small intermediate section in

between that can increase the transmission rate.
Now, let’s consider in a more general way, where we have two waveguides to
connect, with width w1 and w2 . We compare the case where we abruptly connect
them together and put a waveguide with width w1 + w in between.
In the abrupt case, we have

2
t1 = (7.50)
w1 /w2 + w2 /w1

In the case with one transition waveguide in between, we have

2 2
t2 = · (7.51)
w1 + w w1 w1 + w w2
+ +
w1 w1 + w w2 w1 + w

We want to prove: ∀w1 , w2 , w satisfying 0 < w1 < w2 and 0 < w < w2 −

w1 , it holds that t1 < t2 . As t1 and t2 are positive. Consider

t1 2(w1 /w2 + w2 /w1 )

=

t2 w1 + w w2 w1 + w w1
+ +
w2 w1 + w w1 w1 + w
2 2
= 
 
2  + 
2  

w1 + w 2 w2 w1 w1 + w 2
+1 +1 +1 +1
w1 w1 + w w1 + w w2
(7.52)

We change the variable and define f = t1 /t2 to have the following form:
t1 2 2
f (A, B) = = + (7.53)
t2 (A + 1)(B −1 + 1) (A−1 + 1)(B + 1)

where ⎧

⎪
⎪ w1 + w 2
⎪
⎨ A = >1
w1
⎪
2 (7.54)
⎪
⎪ w2
⎩B = >1
w1 + w

We only need to prove f < 1. Take the partial derivative of f with respect to A
∂f 1 1− B
= <0 (7.55)
∂A (A + 1)2 B + 1
356 Appendix: Problem Set Solutions

Fig. A.10 Abrupt (left) and adiabatic (right) transition

Due to the rotational symmetry of (7.53), we know ∂ f /∂ B < 0 for A > 1. So

the maximum value of f
f max < f (1, 1) = 1 (7.56)

Q.E.D.
Note that we have proved the adiabatic case is always better than the abrupt one.
And to make the argument even stronger, the more intermediate sections, the better
the performance, because we can always use the conclusion at each transition and
insert a new waveguide in between. In the infinite transition limit, we will get
a continuously changing transition such as described in the textbook. Here, we
won’t discuss which shape is the best, but readers should be able to explore in a
similar fashion (Fig. A.10).

Chapter 8

Chapter 9

1. (a) According to (9.31), the spatial resolution ρ is approximately

We 20 µm
ρ≈ = = 0.67 µm (9.57)
m 30
(b) According to 9.34, the tolerance of the length can be obtained by

δL δWe
=2
L We
(9.58)
δWe 0.1
⇒ δL = 2L = 2 × 200 × = 2 µm
We 20

2. According to Table 9.1, for a symmetric MMI, the first N -fold image distance is
3L π /4N . For a 1 × 2 MMI (where N = 2), one obtains
Appendix: Problem Set Solutions 357

3L π 4n f We2
L= = = 246.77 µm (9.59)
8 8λ0

3. This depends on how many ports there are. Consider we have N output ports in
total, also assume the outputs are uniform, then ideally we should have

Ptotal
P= (9.60)
N
at each output port. However, due to the nonuniform performance of a star coupler
in general, the outer ports will have a large loss, causing a decrease in the output
power.
4. (a) In this case, a = 0, we obtain
Vc1 = π (9.61)

Plug this into (3.19), and we get

0.8Vc1 λ
d=  = 1.07 µm (9.62)
2π n 2f − n 2s

According to the normalized parameter diagram, b = 0.55. Plug this into

(3.24) 
n e f f 1 = b(n 2f − n 2s ) + n 2s = 3.378 (9.63)

Then the waveguide is reduced to the following effective structure.

Again, we use the same technique to calculate the width of this waveguide,
where V2 = 0.8π. We obtain

0.8V2 λ
w=  = 1.43 µm (9.64)
2π n 2e f f 1 − n 2s

To conclude 
d = 1.07 µm
(9.65)
w = 1.43 µm

(b) i. If the wavevector in the transverse direction is κ, according to the maxi-

mum mode number m = 25, we obtain:

κmax w f s ≈ (25 + 1)π (9.66)


Also, we have κmax = 2π
λ
n 2f − n 2s = 1.76. These two equations yield
358 Appendix: Problem Set Solutions

26π
wfs ≈ = 46.43 µm (9.67)
κmax

And we ≈ w f s , according to (9.6), we get

4n f We2
Lπ = = 6264.18 µm ≈ 0.63 cm (9.68)
3λ0

ii. According to (9.13), for a double image to occur, the minimum length is
given by
1
L min = × 3L π ≈ 0.95 cm (9.69)
2

Chapter 10

1. The reflectivity at z = 0 is given by

(κ/βd )2 sinh2 βd L
R=
1 + (κ/βd )2 sinh2 βd L

where 
βd = κ2 − δ 2

Use the relationship above and plot for R versus δ.

2.
λ
d=
n eff − sin φc

By solving the phase equation, we can obtain that n eff ≈ 3.433. Therefore,

λ
φc = arcsin n eff − ≈ 10.54◦
d

Chapter 11

1. (a) According to (10.7),

sinh2 (κL)
R= = 0.99 (11.70)
1 + sinh2 (κL)

Solve for κ, it yields

Appendix: Problem Set Solutions 359

5.29
L= = 0.0215 cm = 21.5 µm (11.71)
200 cm−1
(b) According to (11.20),

ωs = 2vg κ ≈ 1.2 × 1013 Hz (11.72)

2. According to (11.8), one can use transfer matrix to evaluate the system parameters
by cascade the transfer matrix of each components
 o  i
a a
= [t c (L 4 )] · · · [t ][t
D c (L 2 )][t ][t
D c (L 1 )] (11.73)
bo bi

where in this case, the input vector, the transfer matrices [t D ] and [tc (L i )] are
given by  i  i
a a
=
bi 0
√  
2 1 −i
[t c (L i )] = (11.74)
2 −i 1
 
exp(iπ/4) 0
[t D ] =
0 exp(−iπ/4)

Plug these expressions into (11.92), one obtains

 o √  
a 2 i (1 + i)
= a (11.75)
bo 2 −(1 + i)

3. (a) According to (11.22), assuming we are calculating the central wavelength, it

yields
mλs 150 × 1.5 µm
L = f = = 155.03 µm (11.76)
Neff 1.4513

(b) According to (11.28), we obtain

 −1
dr mλ2 n g
ν = f
Ra s
Neff da cNeff

−1
30 µm 150 × (1.5 µm)2 × 1.4752 (11.77)
=
2 cm 1.4513 × 30 µm × 3 × 108 × 1.4529
= 5.717 × 1012 Hz
= 517 GHz

(c) According to (11.36),

Lu = 8.7θmax
2
/θ02 (11.78)
360 Appendix: Problem Set Solutions

where ⎧
⎪ 8 × 30 µm
⎪
⎨ θmax = xmax /Ra = = 1.2
2 cm
πλ π × 1.5 µm (11.79)
⎪
⎪
⎩ θ0 = s = = 0.72
Neff w0 1.4529 × 4.5 µm

Thus,
2
1.2
Lu = 8.7 × = 24.2 dB (11.80)
0.72

(d) According to (11.39), we obtain the loss l is

2π 2 4.52
l = 17 exp(−2π 2
w02 /da2 ) = 17 exp − = 10.9 dB (11.81)
302

4. The diffraction equation to get start is

f
s
Neff da (sin θi + θo ) + Neff L = mλ (11.82)

According to the text description above (11.29), we then obtain another expression
for ν  = ν + FSR as below.

 f


s
FSR d Neff f FSR d Neff FSR
s
Neff − λ (sin θi + sin θo )da + Neff − λ L = 1 − (m + 1)λ
ν dλ ν dλ ν
(11.83)

Note that the diffraction order becomes m  = m + 1 and the wavelength λ =

λ(1 − FSR
ν
). Substrate (11.82) from (11.83), it yields

s f

FSR d Neff FSR d Neff FSR FSR

λ(sin θi + sin θo )da + λL = mλ − 1 − λ
ν dλ ν dλ ν ν
(11.84)
We plug (11.82) into the right side of this equation and arrange the terms, it yields

 f


s
d Neff d Neff
FSR FSR f FSR
s
Neff −λ (sin θi + sin θo )da + Neff − λ L = 1 − λ (11.85)
ν dλ ν dλ ν

Note that ⎧
⎪ c
⎪
⎨ λ= ν
f,s (11.86)
⎪
⎪ dN
⎩ n g = Neff
f,s
− λ eff
dλ

We then simplify (11.85) as c

FSR · n g (sin θi + sin θo )da + FSR · n g L = (ν − FSR) ≈ c (11.87)
ν
Appendix: Problem Set Solutions 361

Note that usually the operation frequency ν is much larger than FSR, the right
side is reduced to c. Thus,

1
(FSR)−1 = [n g (L + da sin θi + dr sin θo )] (11.88)
c
5. 1. According to (10.7),
sinh2 (κL)
R= = 0.99 (11.89)
1 + sinh2 (κL)

Solve for κ, it yields

5.29
L= = 0.0215 cm = 21.5 µm (11.90)
200 cm−1
2. According to (11.20),

ωs = 2vg κ ≈ 1.2 × 1013 Hz (11.91)

6. According to (11.8),
 o  i
a a
= [t c (L 4 )] · · · [t D ][t c (L 2 )][t D ][t c (L1)] (11.92)
bo bi

where in this case, the matrices [t D ] and [tc (L i )] are given by

√  
2 1 −i
[tc (L i )] =
2 −i 1
 
exp(iπ/4) 0
[t D ] = (11.93)
0 exp(−iπ/4)
 i  i
a a
=
bi 0

Plug these expressions into (11.92), one obtains

 o √  
a 2 i (1 + i)
= a (11.94)
bo 2 −(1 + i)
362 Appendix: Problem Set Solutions

Chapter 12

1. (a) According to (12.44),

λd 1.55 µm · 2 µm
L= = = 689 µm
2
2n e r33 Vπ 2 × 0.75 × 2.1352 × 30.8 pm/V × 10 V
(12.95)
(b) Notice that

Vπ L 10 V · 6890 µm
Vπ /ν = = = 0.0689 V/GHz (12.96)
ν L 10 GHz · cm
(c) You could for example use Thermal Optics or Acoustic Optics Effect.

Chapter 13

1. According to (13.3), one obtains

√
pλd 3 × 1.55 µm × 2 × 6 µm
Vπ = 3 = = 2.9 × 103 V (13.97)
n r L 2.286 × 30.8 pm/V × 0.3 × 1 cm
3

2. One need to plot the cross-power-transfer efficiency ηx versus δ/κ as shown in

Fig. A.11 according to (13.5)

2
κ
ηx = sin2 βc L
βc
(13.98)
1 
= sin 2
( (δ/κ)2 + 1 · κL)
(δ/κ)2 + 1

Note we can see an envelope which is decreasing as the δ/κ increases. This
emphasizes the effect when the two waveguides are detuned with each other,
meaning they have different propagation constants β’s, less power transfer is
allowed.
3. 1. According to (5.19), the minimum length is
π
L min = (13.99)
2κ
Notice this is different from the reversed-β switch, which at V = 0 is at bar
state.
2. Refer to Fig. 13.2, which plot for (13.5)
Appendix: Problem Set Solutions 363

Fig. A.11 Power transfer

ratio versus detuning

2
κ
ηx = sin2 βc L (13.100)
βc

Note that at V = 0, βc = κ. With the increase of voltage, βc increases, which

results in the power allowed to transfer decreases by a factor of (κ/βc )2 and the
transfer period (the voltage needed to transfer from one maximum to another)
decreases (Fig. A.11).
Index

A Bragg filter for channel dropping, 242

Abrupt angle bends, 137 Bragg filters, 205
Absorbing boundary condition, 323 Bragg laser optical cavities, 219
Active EO polarization rotator, 171 Brillouin zones, 197
Active polarization converters, 170
Adiabatic couplers, 106, 107, 110
Adiabatic criteria, 161 C
Adiabatic design, 8 Canonical structures, 81
Adiabaticity, 115 Cascaded coupler filter, 239
Advanced Y-branch design, 131 C-band, 26
Amorphous Si, 24 Channel cross talk, 253
Analog components, 117 Channel-dropping filters, 205, 234, 242
Anisotropic crystals, 265 Channel guides, 58
Applications of Bragg gratings, 220 Channel waveguides, 65
Applications of PHASARs, 257 Chromatic dispersion, 25
Arrayed-waveguide grating (AWG), 194 Chromatic resolving power, 225
Asymmetric coupler filters, 235 Circular bends, 138
Asymmetric Y-branch for mode sorting, 135 Codirectional coupling, 87
Asynchronicity, 131 Coherent communication systems, 171
Coherent coupling, 132, 138
Coherent receivers, 178
B Components for polarization control, 153
Bandwidth, 188 Computational design, 8
Beam propagation method, 328 Confinement factor, 43
Beam propagation technique, 118 Conformal transformation, 139
Beat length, 180 Conformal transformation of bends, 138
Bend radius, 5 Coplanar coupling, 212
Bends, 117 Counterpropagating waves, 86
Birefringence cladding materials, 159 Coupled-mode theory, 79
Birefringent crystal, 19 Coupler-mediated devices, 8
Bloch function, 197 Coupling coefficients, 80, 89, 91, 121
Boundary conditions, 35, 323 Coupling coefficients for gratings, 214
Boundary-value problems, 331 Coupling of modes, 79
BPM expansions of the method, 330 Courant condition, 322
Bragg diffraction, 221 Criterion for an adiabatic splitter, 131

© Springer Nature Switzerland AG 2021 365

R. Osgood jr. and X. Meng, Principles of Photonic Integrated Circuits,
Graduate Texts in Physics,
https://doi.org/10.1007/978-3-030-65193-0
366 Index

Cross talk, 144 Finite-difference time-domain (FDTD), 320

Crown waveguide, 143 Finite-difference time-domain (FDTD)
Crystal orientation, 25 computations, 26
Current driven switches, 24 Finite-difference time-domain method, 319
Finite-element, 317
Finite element methods, 331
D Flame hydrolysis, 13
β directional coupler switch, The, 296 Fourier component of the grating, 206
β switch, 297 Fraunhofer diffraction, 195
Design of single-mode channel waveguides, Free-carrier absorption, 23
71 Free-carrier modulations, 288
DFB reflectors, 205
Free dangling bonds, 28
Dielectric slab waveguide, 31
Free spectral range, 250
Differential coupled mode equations, 108,
Full-vector BPM, 331
129
Diffraction gratings, 205
Diffused waveguides, 49
Diffusion coefficient, 20 G
Digital optical switch, 307 Galerkin’s method, 332
Directional routing, 95 General coupled-mode formulation, 80
Dispersive properties of waveguides, 45 General interference, 181
Distributed Bragg reflectors, 216 Generalized Mach–Zehnder MMI interfer-
Distributed feedback (DFB) lasers, 205 ometer, 193
Doping-induced changes in semiconductors, Geometrical dispersion, 25
23 Glass, 8
Glass densification, 15
Graded-index channel waveguides, 72
E
Graded-index gradient, 49
Effective-index method, 65
Effective wave velocity, 35 Graded-index structures, 58
Effective width of the mode, 43 Grating dispersion, 210
Effective width of the waveguide, 43 Grating-mediated devices, 8
Efficient grating couplers, 7 Group velocity, 46
Electrode gap, 284 Guided modes, 32
Electro-optical effect, 162 Guided-mode-propagation analysis (MPA),
Electro-optical modulators, 263 178
Electro-optical switches, 296 Guided-wave loss, 32
Electro-optic modulators, 275
Electro-optic tensor, 269
Electro-optical switching arrays, 311
H
Etch-induced sidewall roughness, 26
Helmholtz equation, 35
Explicit equation for bend loss, 144
High modulation rates, 4
High pressure oxidation of a silicon sub-
F strate, 13
Fabrication of gratings, 226 High-Q Bragg grating structures, 215
Fabry–Perot cavity, 208 High refractive contrast, 24
FDFD from Maxwell’s equations, 327 High thermal conductivity of the silicon, 13
Ferroelectric-like metal oxides, 270 HiPOx, 14
Finesse of a filter, 239 Horn waveguide junction, 121
Finite-difference, 317 Humidity resistance, 17
Finite-difference BPM, 329 Hybrid integration, 154
Finite-difference frequency-domain, 324 Hybrid nature, 59
Finite-difference methods, 317 Hybrid optics, 154
Index 367

I M
III-V, 8, 11 Mach–Zehnder interferometer, 124, 220
III-V materials, 22 Mach–Zehnder-like interferometers, 236
Imaging devices, 177 Mach Zehnder modulator, 278
Imaging modalities, 180 Marcatili’s method, 62
Impermeability tensor, 265 Meander coupler, 236
Implementation of FEM, 332 Metal organic chemical vapor deposition
In-coupling to a waveguide using a grating (MOCVD), 13
coupler, 227 Method of field shadows, 61
Index ellipsoid, 266 Micro-optics, 154
InP photonics, 23 Micro-ring resonators, 7
Integrated optical switches, 295 Microwave circuits, 11
Integrated passive waveguide converters, Mid-IR wavelengths, 7
168 Minimum radius of bends, 12
Integrated star couplers, 195 MMI-based switches, 304
Interference-based asymmetric coupler MMI-coupler switches, 306
splitter, 165 MMI phased-array wavelength
Interference-based switches, 303 (De)multiplexer, 256
Intersecting waveguide structures, 305 MMI 3 dB splitters, 190
Ion implantation, 49 Modal confinement, 12
Modal cutoff, 41
Mode beating, 123
Mode confinement, 13
J Mode converter, 209
Jones matrices, 171 Mode-dependent group velocity, 39
Jones vector, 166 Mode evolution type polarization splitter,
161
Mode matching, 13
Mode routers, 128
L Mode splitters, 128
Lamipol, 155 Molding or embossing, 18
Large-scale and high-performance inte- Moore’s Law, 3
grated optics for communications and More exact treatment of waveguide couplers,
computer interconnects, 24 99
Laser transmitter, 3 Multimode imaging devices, 166
Lateral LiNbO3 Couplers, 101 Multimode interference, 177
Lateral phase shift, 39 Multimode interference devices, 177
Lateral Si wire couplers, 104 Multiple-section, 238
Layout of a basic modulator, 275 Mux/demux, 8
Light routing junction, 124
Lightwave polarization, 32
LiNbO3 , 5, 8, 11 N
LiNbO3 -based technology, 19 Narrowband channel-dropping filter, 219
LiNbO3 crystals, 273 90◦ waveguide crossovers, 144
LiNbO3 surface, 22 Nomenclature for modes, 59
LiNbO3 waveguide, 22 Nonlinear-optical integrated photonic
Linear absorption coefficient, 12 devices, 8
Local normal modes, 115 Nonuniform butterfly MMI, 191
Low-energy modulators, 24 Nonuniform power splitting, 181, 186
Low pressure chemical deposition Normalized frequency, 38
(LPCVD), 13 Normalized length, 88
Lumped circuit, 282 Numerical aperture, 38
Lumped-circuit type, 282 Numerical computational methods, 118
368 Index

Numerical design, 8 Post-exchange annealing, 22

Numerical methods in nanophotonics, 317 Power divider, 124
Numerical solutions, 52 Power in coupled systems, 84
Numerical techniques, 61 Power loss due to coherent beam combina-
tion, 135
Principal axes, 265, 271
O Propagation axis, 36
1/4λ-step Bragg grating, 219 Propagation constants for the coupled
Optical axis, 273 modes, 83
Optical couplers, 4, 95 Propagation of lightwaves, 265
Optical loss due to background doping, 23 Proton-exchange guides, 22
Optical packaging, 13 P-TEOS, 15
Optical switch arrays, 311
Opto-electronic integrated circuit, 3
Ormalized waveguide index, 40 Q
Overlap factor, 276 Quality factor, 215
Overlap imaging, 184, 186 Quarter-wave section, 217
“Quasi-” single-mode, 117

P
Packing density, 26 R
Padé approximants, 330 Radiation modes, 36, 47
Paired interference in the MMI, 183 Radiative effects, 99
Parabolic horn, 122 Radiative loss, 12
Paraxial equation, 329 Raman–Nath diffraction, 221, 222
Partial thermal decomposition of polymers, Random mirror roughness, 146
17 Ray optics, 33
Passive components, 115 Rayleigh range, 187
Passive waveguide components, 117 Reflection, 119
Patterning techniques, 17 Reflection (spectrometer) grating demulti-
Perfectly matched layer, 323 plexers, 205
Perturbation theory, 70 Refractive index dispersion, 17
PHASAR-based wavelength router, 166 Residual stress anisotropy, 14
PHASAR frequency response, The, 250 Resolution of an MMI, 186
PHASAR insertion loss, 251 Resonant-cavity modulators, 279
PHASAR polarization properties, 254 Restricted interference, 183
PHASAR uniformity, 251 Reversed β Switch, 301
Phase-delay routers, 4 RIE etching, 18
Phase modulators, 277 Ritz method, 331
Photonic “bandgap”, 210 Rowland mounting, 247
Photonic integrated circuits, 1, 4
PIC materials systems, 12
Planar processing, 1 S
Polarization converters, 166 Scalar-wave equation, 70
Polarization devices for Si photonics, 172 Separation of variables, 62
Polarization diversity, 153 Shallow sinusoidal gratings, 214
Polarization routers, 160 Si, 11
Polarization-sensitive manipulation, 160 Si-based opto-electronic, 5
Polarization splitters, 162 Si integrated photonic circuits, 5
Polarizing elements, 154 Silicon MMI switch, 193
Polaroid-like polymers, 155 Silicon optical bench, 13
Polymeric PIC, 15 Silicon-on-Insulator (SOI), 6
Polymeric waveguide materials, 15 Simple transitions in waveguide geometry,
Polymers, 11 119
Index 369

Si3 N4 , 13 Traveling-wave, 282

Si nanowires, 24 Turning mirrors, 145
Single-crystal Si, 24 Two-dimensional diffusion profile, 20
Sinusoidal mirror roughness, 146 Two-dimensional routing, 32
SiO2 , 11 Two-waveguide directional coupler, 96
SiO2 -based materials system, 13
SiO2 :P, 13
Si optical bench technology, 13 U
Si photonics, 6 Ultrafast thermo-optical switches, 24
Si wires, 24 Ultrahigh-speed data processing, 7
Slab waveguide, 31 Ultralow energy-per-bit modulators, 7
Spatial Fourier transform, 195 Ultrasmall Si Semiconductor Modulators,
Spatial periodic structure, 86 288
Spectral bandwidth, 4 Undesired optical absorption, 16
Spectral reach, 7 Undoped SiO2 , 13
Stability, 4
Stability criterion, 322
Star couplers, 177, 195 V
Substrate radiation modes, 48 Variable power splitters, 178
Subwavelength grating (SWG), 195 VCM method, 127
Surface diffraction grating, 86 Vertical couplers, 106
Surface plasmon, 158 Vertical polarizing plates, 154
Symmetric slab, 44 Very high speed modulators, 284
Symmetric waveguide, 41 V-groove technology, 13
Synchronous, 79 Volatile LiO2 , 20
Synchronous waveguides, 85 Volume current method, 127
System modal fields, 79

T W
Tansformed index profile, 141 Wall-scattering loss, 58
Tapered 3 dB MMI coupler, 191 Wave equation for a slab waveguide, 34
Tapered waveguide junction, 121 Waveguide array, 246
Tapers, 117, 121 Waveguide asymmetry, 40
Temperature-dependent refractive index, 17 Waveguide cutoff, 38
TEOS, 15 Waveguide-grating router, 244
Thermo-optical switches, 309 Waveguides, 4
Thermo-optical switching arrays, 313 Waveguiding structures, 8
Thin polarization plates, 154 Wavelength filtering, 8, 233
Three-dimensional waveguide, 57 WDM routers, 178
Ti-diffused waveguides, 19 Wide-angle BPM, 330
Tight folding of Si PICs, 26
TM modes, 45
Total internal reflection, 33 Y
Transfer matrices, 147 Y-branches, 117, 124
Transfer matrix for a coupler, 98 Yee discretization, 321
Transmission, 119 Yee lattice, 319
Transmission matrix, 238 Yee’s algorithm, 319
Transparent boundary condition, 330
Transverse confinement, 36
Transverse electric, 36 Z
Transverse magnetic, 36 Zincblende semiconductor crystals, 269