Engineering aspects and applications of the new

Raman instrumentation
G.D. Pitt, D.N. Batchelder, R. Bennett, R.W. Bormett, I.P. Hayward, B.J.E. Smith, K.P.J. Williams, Y.Y. Yang,
K.J. Baldwin and S. Webster

Abstract: Raman instrumentation design has improved radically in efficiency and ease of use over
the past 15 years. New technologies made this possible with introduction of the first commercial
high-efficiency (high speed) Raman systems incorporating low power lasers. Systems were
introduced for the first time on production and process lines. Higher efficiency systems have
continued to evolve, as a result of novel engineering solutions. Raman-near field optical systems
and Raman-SEM (scanning electron microscopy) combinations have created a new area of
nanoscale spectroscopic measurements. New application areas are: (a) semiconductor processing;
(b) pharmaceutical drug processing; (c) gemology; (d) narcotic and explosives forensic detection; (e)
coatings on computer hard disks and read heads; (f) endoscopic and Raman detection of
oesophageal and other types of cancer. Comparisons are made with competing optical and
alternative instrumentation techniques for general Raman work and the applications.

1 Introduction The ratio of the inelastic (Raman) scattered intensity to

that of the Rayleigh is typically 10
6 for pure liquids, while
The original experiment conducted by Raman in 1928 to for micrometre sized particles in a powder it may be as low
detect optical inelastic scattering was remarkably simple [1]. as 10
12. The separation of the two types of scattered light
Subsequent instrumentation, which made use of this effect for spectral analysis remains challenging. Useful reviews on
for the investigation of the structural properties of solids, the early history of the Raman effect can be found in
liquids and gases, was cumbersome and difficult to operate. Schrader [2], Long [3] and Venkataraman [4].
A step-change occurred around 1990 in the optical and The introduction of this apparently simple experiment
mechanical design of such equipment with the emergence of into the sphere of commercial instrumentation had to await
new detection, filtering and laser devices. As a result the new the development of sufficiently intense and monochromatic
systems reverted to the more simple concepts inherent in the optical sources (i.e. lasers) and more sensitive detectors (e.g.
original Raman experiment. photomultipliers). This first step occurred during the 1960s
The first Raman experiment is shown schematically in [5]. Difficulty in eliminating the intense Rayleigh radiation at
Fig. 1. The intense Indian sunlight provided a sufficiently the detector still remained, however. It was possible to base
powerful beam for the effect to be observed. The broadband the design around standard double and triple monochro-
radiation was passed through an initial filter to produce a mators, where multiple gratings and long path lengths
near ‘monochromatic’ beam which, in turn, was passed into (typically 1 metre plus) provided the necessary dispersion to
the liquid benzene sample. The difficulty in detecting the separate the weak Raman scattering from the Rayleigh
predicted very weak inelastic (and frequency shifted) scattering [6]. The radiation intensity of the Raman spectrum
scattering from the intense elastic (Rayleigh) scattering was still too low for easy measurement, owing to the
was eliminated by inserting a second optical filter before the relatively poor throughput of such systems; high power
scattered light struck the ‘detector’. Initially this was the lasers (typically 1 W) were therefore a necessary and
human eye but was later replaced by a photosensitive plate. expensive part of the solution. Robust samples that were
not degraded by the extremely high power densities were
also required. Rosasco [7] wrote a state of the art review in
r IEE, 2005
1980, with particular reference to Raman microscopy.
IEE Proceedings online no. 20050015
The low intensities available for detection also demanded
doi:10.1049/ip-smt:20050015
stringent precautions to reduce any background radiation.
Paper first received 29th March 2005
Alignment stability problems originating from thermal
G.D. Pitt was with Renishaw plc, Old Town, Wotton-under-Edge, Glos. GL12
7DW, and is now a consultant on a new product design and marketing
variations and mechanical movement of the delicate system
development slits made it necessary to use large anti-vibration tables for
R. Bennett, R.W. Bormett, I.P. Hayward, B.J.E. Smith, K.P.J. Williams and mounting the boxed monochromators and launch optics
Y.Y. Yang are with Renishaw plc from the laser. These factors, plus the necessity for large
G.D. Pitt is also with the School of Physics and Astronomy, University of optical gratings and the maintenance of the water-cooled
Leeds, Leeds LS2 9JT, UK and the Department of Physics, H.H. Wills Physics lasers etc. meant that Raman work remained a highly
Laboratory, University of Bristol, Bristol BS8 1TL, UK
D.N. Batchelder is with the School of Physics and Astronomy, University of
specialised academic activity.
Leeds, Leeds LS2 9JT, UK The Raman systems developed during the 1990s
K.J. Baldwin and S. Webster are with Avacta, Leeds University, Leeds LS2 9JT, revolutionised the acceptance of Raman both as a research
UK tool, and for repetitive and robustness requirements
E-mail: [email protected] in industry. Many applications have emerged where

IEE Proc.-Sci. Meas. Technol., Vol. 152, No. 6, November 2005 241
 scattering ment environment with (among other factors) a further
liquid reduction in size, portability, use of fibre optics and
sunlight
(white) no moving parts.
violet The third was to use best engineering practice to improve
the sensitivity and precision of measurement using largely
violet
filter
Rayleigh- Raman- mechanical and electrical engineering expertise.
scattered scattered
light light It is now opportune to review and compare these new
violet green
developments. Understandably there has been an inbred
green filter
conservatism in the outlook of many Raman researchers
who have extensive experience of traditional monochroma-
green
tor systems and a mistrust of novel designs. Therefore an
attempt is made to explain the differences in design and the
observer subsequent new applications that have emerged. It is also
useful to analyse the advantages and disadvantages against
Fig. 1 Schematic diagram of Raman’s original experiment alternative techniques for materials’ analysis, e.g. infrared
The incident sunlight is filtered, leaving ‘monochromatic’ radiation to
enter the scattering liquid sample. A second filter before the detector
absorption and Fourier Transform infrared and Raman
removes the Rayleigh scattered light techniques.
Interfacing of Raman systems to other instrumentation
has also been a major feature of recent work. For example,
the technique would previously have been considered one major advantage of the new technology over well-
inadequate. Batchelder [8] and Carrabba et al. [9] instigated established IR absorption methods is the use of a
the combination of several newly developed components, microscope with spatial resolutions of less than 1 mm using
making for more compact systems with an optical visible excitation, on relatively easily prepared samples. This
throughput increased to typically 30% compared with a compares with spatial resolutions of 5–10 mm using
throughput of o3% for a triple monochromator. The first microscope based IR techniques. Prior to 1990, microscopes
commercial development of this new generation of instru- were attached to Raman systems [6] using standard
ments originated from collaboration between Renishaw [10] monochromators without optimisation of the optical
and the University of Leeds [11, 12] (earlier work had been design. These proved to be reasonably successful, albeit
carried out at Queen Mary and Westfield College, extremely cumbersome and expensive to start up and
University of London). Others followed this lead. maintain. They were suitable for use only in advanced
The critical components for this advance were: research laboratories. The University of Leeds/Renishaw
a. New multilayer dielectric and holographic dichroic filters. approach began with the premise that the optical design
Initially these were developed for high specification defence should start at the microscope and work backwards
applications; through the Raman system. This led, for example, to
the employment of infinity-corrected optics in the micro-
b. Highly sensitive charge coupled devices (CCD) detectors. scope, resulting in greater flexibility for linking to the
These silicon detector arrays, with up to 80% quantum spectrometer and permitting the insertion of additional
efficiency, superseded the sensitive (but relatively low sampling options.
efficiency) photomultipliers. They also enabled improved
signal processing techniques to be employed for spectral
analysis; 1.1 Further developments
c. The availability of low cost PCs for spectrometer and In this review we discuss the design for the basic Raman
data collection and processing; microscope highlighting the benefits over low optical
throughput systems. Subsequent work has concentrated
d. The possibility of using low cost, low maintenance lasers. on extending the specification range, while retaining system
The advantages of these new designs are listed in more ease of use and operational flexibility:
detail in Section 3, but incorporation of these sub-systems
into the new Raman spectrometers produced a reduction in a Laser excitation range. The wavelength range of the
size, increased robustness, and removed the need for excitation lasers has been extended, i.e. use of UV down to
expensive vibration isolation tables. As a result of the 229 nm (and potentially lower), and semiconductor lasers
increased optical throughput, lower power lasers (e.g. up to 830 nm. Much of this work has been promoted by the
1–10 mW) could now be used (including small semiconduc- possibility of using lower power lasers than hitherto, and a
tor devices), further reducing cost and maintenance. From new range of lasers has therefore become available for
the 1990s onwards, there has been a consolidation of these Raman applications. One advantage of using both higher
new techniques with system research and development and lower wavelength lasers is that the Raman spectrum for
branching in three directions. some materials can be more easily distinguished from
The first was to extend the specification envelope of the high broadband emissions such as fluorescence, a major problem
throughput system: in Raman spectral interpretation. This aspect will be
covered in the following theoretical Section 2, and some
1. Lasers of both shorter (UV) and longer (near IR) of the new applications that have emerged as a result will be
wavelengths were introduced as sources; described in Section 9.
2. The limit of low frequency detection was lowered; b Near the laser line measurement. Enhancing the ability of
3. Increased automation, robustness, speed of data acquisi- the system to exclude Rayleigh scattering has improved the
tion and analysis were incorporated using fundamental range of Raman spectral frequencies measured, particularly
mechanical and engineering practices. in the region of the frequency of the excitation laser.
Initially, filter based systems were able to give results to
The second was to adjust the new designs such that 4200 cm
1 from the laser line frequency. This limit has
the system could be interfaced with the industrial measure- been reduced with further development to approximately
242 IEE Proc.-Sci. Meas. Technol., Vol. 152, No. 6, November 2005
10 cm
1 for visible lasers, and 200 cm
1 (down from that the performance of this new method is at least as good
500 cm
1) for UV lasers. as the classical method, and offers improved alignment and
c Interfacing to other techniques. Developing novel fibre performance stability for extended (time) x–y and depth (z)
optic probes for linking Raman systems into other mapping. It can be an advantage that some confocality
instrumentation, e.g. scanning electron microscopes performance (i.e. spatial resolution) can be controlled
(SEMs), in vivo biomedical measurement and process simply by a software change. Several applications will be
control applications, have required considerable engineering described where the properties of inclusions in supporting
input, for example the development of high-vacuum seals matrices can be determined non-destructively. Subsurface
and high-precision manipulation and alignment techniques defects and local strain can also be measured. The
for Raman-SEM. For process control there are high- combination of these novel optical developments, and
pressure constraints for pipe insertion, as well as window higher resolution stages and mounts (d. above), enables
cleanliness and the possible presence of aggressive materials point-by-point mapping to spatial resolutions of better
at high temperatures. For in vivo work the diameter of the than 1 mm.
probe had to be minimised. h Further improvements in spatial resolution. Extension of
d Improved stability and high precision movement subsystems. the ability of the new Raman systems to measure the
The requirement for maximum robustness and ease of use spectral properties of a nanoscale sample requires still
has led to new mechanical designs of rotary stages for the further improvement in spatial resolution. Recent data
scanning gratings and mirror holders. High-precision will be given from the combination of Raman and scanning
sample holding stages for automated x–y point-by-point near-field optical techniques using both apertured and
scanning, for example, are now being increasingly used. apertureless probes. Measurement times for point-by-point
Encoder feedback systems combined with stepper motors can be lengthy at present, but it is envisaged that
are an improvement over earlier lead screw stages that were with further innovation faster measurements below 25 nm
simply driven to a point. Automated measurements are now will become routine. Improved mechanical stability for
routine over several hours, or even days. high resolution moving stages is essential for such applica-
e Flexibility in system control using CCD cameras. The use tions.
of a CCD camera linked to a PC, using a USB connection, i 2-D Raman imaging. In h. above, point-by-point scanning
gives added flexibility to automatically adjust the movement using automated moving stages has been briefly mentioned.
of critical components in the optical path, allowing Further techniques can be employed (including line
changeover of input laser wavelength. For the first time, a scanning) to provide a highly accurate picture of the
flexible Raman research and development system has been material content of a surface. There are occasions however
produced that requires no optical component tweaking. when the dispersion of segregated phases in a given matrix
Many of the mechanical design solutions that provide a cost should ideally be identified more rapidly than by mapping.
benefit and system stability derive from the latest manu- The use of filters in the spectrograph design radically
facturing technology. increased the system optical throughput; therefore it has
f ‘No moving parts’ Raman systems. A number of recent become possible to defocus the sample beam to cover a
process measuring Raman systems have no moving parts given sample area (e.g. 100 mm diameter and greater) and
and may also use multi-channel detection on the CCD. still have sufficient radiation at the detector for meaningful
Smaller portable units are now feasible, particularly with the Raman measurements to be made. The CCD can be used as
availability of frequency-stabilised semiconductor lasers as a spatially measurable 2-D detector area, and with special
the optical source. The forerunners of this approach are bandpass filters added at appropriate inclination (with
described in the publications of Carrabba et al. [9] and angular control) before the detector, a fast scan of critical
Yang et al. [13]. The latter group used a holographic material frequencies can be made such that different phases
transmission grating as well as holographic filters [14], and or segregated inclusions can be quickly (e.g. 10 s) identified.
several successful applications in process control using fibre Estimates can then be made of their respective content (at
optic probes have been addressed. The Renishaw/University least on the surface). One of the first published examples of
of Leeds approach has been to preferentially use multilayer direct 2-D Raman imaging (as distinct from mapping or
dielectric filters and reflective gratings to counter the line scanning) was for the compositional measurement of
humidity and environmental swings occurring in process polymer blends [17]. A more recent review by Schlucker
and industrial applications that lead to holographic et al. [18] summarises the relative advantages of imaging
component deterioration. In some applications the spectro- and mapping.
scopic specification can be relaxed (e.g. system resolution j Raman libraries. The output spectrum from the Raman
requirements) and the more robust multilayer dielectrics spectrometer effectively provides a ‘fingerprint’ of the
become the obvious solution, particularly when UV laser material and its structure. The surge in data emerging from
excitation is preferred [15]. these new developments, plus great improvements in signal-
g Confocal Raman microscopy. The integration of a to-noise for the data, has meant that practical validated
microscope with Raman systems opened up the possibility libraries of spectra for dispersive Raman systems can
of using confocal techniques. In this case, the sample now be assembled. Renishaw disseminated the first
volume is defined by using an aperture or apertures to block commercial libraries (with appropriate sample qualification)
out-of-focus or off-axis light originating from outside the for minerals, polymers and other organic and inorganic
selected sampling volume. Classical confocal techniques [16] material groupings. Thus the major problem of Raman
use small pinholes as the apertures; for highly accurate spectral interpretation has been greatly reduced, and
work, these call for continual realignment or refocusing due systems can now be technician–operated for routine
to thermal shifts over time. The new technique to be analysis. The spectrum can be acquired and directly
described here uses only one real aperture in the form of the compared in the software against the library data, and the
spectrograph entrance slit, and one orthogonal virtual slit result/identification obtained. As a result, it is believed that,
where the height of the captured image on the CCD can be eventually, Raman systems will become as ubiquitous as
set to filter spatially the scattered light. Results will show standard IR absorption instruments in most testing

IEE Proc.-Sci. Meas. Technol., Vol. 152, No. 6, November 2005 243
laboratories. Further comparisons of these two techniques used, spectra can be obtained within cells without damage
are made in Section 2.1.1. to the cell wall. Examples will be given whereby Raman
techniques are now being investigated to combat cancer,
and to control the quality (and possible patent violations) of
1.2 Further methods for enhanced Raman
drug composition. Many biomedical samples will be
spectral sensitivity
immersed in or contain water and these can be easily
This review is largely concerned with the development of
measured while IR absorption measurements are very
instrumentation for new dispersive Raman systems with
difficult.
high optical throughput, but there are other Raman
methods for enhancing signal-to-noise and it is appropriate ii. Semiconductors and hard disk quality control. Raman
to include brief descriptions and comparisons for both methods are now used on-line for strain measurement on
Fourier Transform Raman and resonance Raman (includ- integrated circuit devices, as well as determination of the
ing surface enhanced effects). The design of the former is disorder content in silicon substrates and diamond-like
based on similar interferometer designs to those used in carbon coatings. These advances have arisen from the
FTIR (Fourier Transform IR) systems [19]. The advantages increased speed of measurement, e.g. a point measurement
and disadvantages, relative to the new dispersive Raman on silicon can be obtained in o0.1 s, whilst with line-
techniques described here, will be described in Section 2.2. scanning techniques, large wafer areas can now be covered
FT Raman has found a niche application area for the within a favourable timescale for wafer production.
analysis of certain polymers, where shorter-wavelength Similarly, the non-destructive identification of defects and
dispersive Raman can run into difficulties differentiating particles on the material surfaces can be achieved with
the Raman signal from background fluorescence effects. FT integration to programmed microscope stages and scanning
Raman partially solves this problem by employing high- electron microscopes. Such results have a major impact on
powered Nd-YAG lasers emitting at 1.06 mm, and cooled device yields and the economics of the process line.
IR detectors. iii. Gems and minerals. The latest Raman systems are now
Resonantly enhanced Raman techniques take advantage the standard analytical tool for quality and process control
of the dramatically enhanced Raman signal that occurs by producers of synthetic industrial diamonds. Synthetic
when the wavelength of the incident laser coincides with an gem quality diamonds can now be distinguished from
optical absorption arising from fundamental internal natural diamond. This is a critical measurement directly
excitations in the material. Under these circumstances, the affecting the value of the diamond. Raman is now the
intensity of Raman scattering can increase by a factor of accepted technique for the detection of flaws filled by
106. The new optical design developments described here unscrupulous dealers. Treatments by oil and resin, selected
have enabled lower power (and lower cost) UV lasers to to avoid detection by standard gemology inspection devices,
become candidates for use with materials that are are immediately revealed non-destructively, and all major
transparent at visible wavelengths. national gem laboratories now have Raman systems. The
Further signal improvement can be obtained using Renishaw library of Raman gem spectra provides rapid
surface enhanced resonance Raman scattering (SERRS) identification of gems and minerals, backed up by samples
techniques. Over the years these have proved difficult to use with provenance. This has also been used for the study of
in a quantifiable way because of a lack of understanding inclusions within minerals; the new confocal techniques
of the bonding interactions occurring at the interface of map the internal strains within the crystal to provide
the sample material and the substrate, e.g. Ag, Au, Cu. information on the geothermal history of rock formations.
Results can also vary, depending upon the surface rough- It is now used to identify precursors indicating the presence
ness of a metallic substrate or the consistency of a colloidal of deposits of economically important minerals (e.g.
substrate, for example. However, recent meetings [20] have diamond).
indicated that most of the polarised views relating to the iv. Portable and on-line systems. It is a useful guide to a
theoretical background are coming together. A brief technology coming of age when previously apparently
description of some results will therefore be included in complex instruments can be packaged in a suitcase suitable
the Applications Section 9, with some discussion on the for air transport and collected from the carousel. The
implications for instrumentation design. It is believed that suitcase is opened at the site of investigation and the system
this could become an important application area in future. works. The cost of the system then falls to a level acceptable
For example, it has been demonstrated that novel SERRS to the wider market. This is now the case, and examples will
techniques can detect certain drugs and explosives in be given for on-line process control in polymer production
remarkably low quantities, and there are DNA related and use in several forensic applications. For example, the
applications. rapid and unequivocal detection of drugs (non-destruc-
tively) within their transparent packages means that
1.3 Raman results and applications evidence is not contaminated and measurements are
The results presented in this review are of two types: (a) speeded up. These are important economic factors for the
examples illustrating the specification improvements in the police. Fibre optic probes have also been developed to
new techniques compared to earlier methods, and (b) results enable on-site detection and remote inspection in otherwise
obtained from new fields of research, and industrial and inaccessible volumes.
process control areas that did not hitherto use Raman
techniques. Typical examples of the latter are:
Recent analyses of the growth in usage of general
i. Biomedical and pharmaceutical. Advantages over the instrumentation techniques confirm the rapid market
standard IR techniques commonly used in these areas, are penetration of Raman systems [21]. This review will attempt
that minimal sample surface preparation is now required to describe the fundamental background and practical
(i.e. no polishing or vacuum), and improved high spatial reasons for this trend. It will involve some comparisons with
resolution can distinguish between incipient and excipient in competing Raman instrumentation designs, as well as
powdered compacts. The measurement is wholly non- describing the limitations of the alternative analytical
destructive. Owing to the low laser powers that can now be techniques.
244 IEE Proc.-Sci. Meas. Technol., Vol. 152, No. 6, November 2005
2 Background theory and comparison data and light at the frequency nL7nV (Raman scattering), where
nV is the frequency of molecular vibration. The two Raman
This section explains with a minimum amount of theory the frequencies are called Stokes, nS ¼ nL
nV, and Anti-Stokes,
decisions that need to be made for the engineering design of nAS ¼ nL+nV.
Raman instrumentation and for given applications. It The quantum mechanical picture of Raman scattering by
provides a listing of the basic theoretical relationships and a diatomic molecule produces qualitatively similar results,
also comparisons with other optical analytical methods, e.g. but there are also important differences. The quantisation of
IR absorption, and those other effects which can cause the electronic and vibrational levels is illustrated in Fig. 2.
problems in the spectral interpretation of Raman data, e.g. Also shown in the Figure are the electronic transitions
photoluminescence and fluorescence. corresponding to Rayleigh, Stokes Raman and Anti-Stokes
Numerous theoretical Raman reviews deal with the Raman scattering. In normal Raman scattering, the
fundamental optical/phonon and quantum interactions excitation of the electron is to a ‘virtual’ electronic level
within materials. Herzberg [22] made an early contribution (in resonance Raman scattering the excitation is to the real
to defining Raman theory, but there are several later books electronic excited state). The virtual level lies at an energy
on the topic [2, 23]. DE below the excited state; the uncertainty principle
The new developments outlined in Section 1 are based on (DEDtBh), however, allows the electron to reside in the
the marriage of recently available technologies, each of real excited state for a time DtBh/DE. This is important for
which has brought about changes in our understanding of Rayleigh and Raman scattering as the creation of the
the trade-offs required for the selection of the different electronic dipole requires electronic transitions to excited
subsystems. Thus, while the flexibility in design has states that have wave functions of different spatial
improved, the system selection process has become increas- symmetry in the ground and excited states.
ingly difficult for those who struggled in the past with
difficult to operate systems, and consequently adopted a
3
conservative approach. It is believed that the necessary 2 upper (excited)
1
theory can be condensed, sufficient for design and ‘fitness 0 electronic state
for purpose’ decisions to be made. v
v0 −∆
v +∆
v0 v0

2.1 The Raman effect and spectra hv0

3
2.1.1 Scattering mechanism, and compari- 2 lower (ground)
1
sons with IR absorption spectroscopy: Stu- Rayleigh Raman Raman Raman Raman
0 electronic state
dents usually find that the physical mechanisms associated (Stokes) (Anti-Stokes) (res.) (pre-res.)
with IR absorption are simpler to understand than those of
Raman scattering. Using an ionic diatomic molecule as an Fig. 2 Energy state representations of Stokes, Anti-Stokes,
example, and taking a classical viewpoint, an IR beam preresonance and resonance Raman scattering
directed onto the molecule will be absorbed when the
photon energy of the IR beam is equal to that of the energy In four of the types of scattering illustrated in Fig. 2 the
of oscillation. For more complex molecules the number of molecule starts and finishes in the electronic ground state. In
vibrational frequencies rises in approximate proportion to Stokes Raman scattering, the molecule starts in the
the number of atoms in the molecule. The electromagnetic vibrational ground state but transfer of energy from the
field drives the ions in the molecule into oscillation and inelastically scattered photon raises the molecule to its first
energy is absorbed. For Rayleigh scattering (in this case the vibrationally excited state. In contrast, in Anti-Stokes
atoms of the molecule do not need to be charged), it is the Raman scattering, the molecule starts in a vibrational
electronic cloud associated with the molecule that is driven excited state and returns to the vibrational ground state by
by the electromagnetic field of the illuminating laser. This transferring energy to the photon. The consequence of this
motion produces an oscillating electric dipole that is about difference is revealed by the unequal intensities of the Stokes
1 nm long. This is analogous to the oscillating electric dipole and Anti-Stokes bands IS and IAS, and their dependence on
produced, for example, by a signal generator on a 100 m absolute temperature, T. The Boltzmann factor describes
long wire used for radio broadcast. In both cases the wire the relative number of molecules in the ground (NV ¼ 0) and
acts as an antenna, or nano-antenna, emitting electromag- excited (NV ¼ 1) vibrational states and, hence, the relative
netic radiation at the driving frequency, whether it is a laser intensities of the Stokes and Anti-Stokes Raman bands:
or signal generator. In both laser-induced Rayleigh scatter-
ing and single frequency radio broadcasting, the only IAS =IS ¼ NV¼1 =NV¼0 ¼ e
ðhn v=kT Þ ð1Þ
information in the emitted electromagnetic radiation is the h and k are the Planck and Boltzmann constants
presence of the antenna and the frequency of excitation. respectively.
Information can be introduced into the emitted electro- Providing a molecule retains its identity when in
magnetic field by amplitude modulation (AM) of the condensed phases, e.g. if is bound by van der Waals forces,
oscillating dipole on the antenna. In the case of the radio the Raman spectra of its liquid, amorphous and crystalline
broadcast, a second voltage of a lower frequency modulates forms will be similar to that of the dilute gas. The
the output of the signal generator. For the molecule, the vibrational frequencies and bandwidths of a molecule are
vibration of the atoms causes the amplitude modulation; modified by the interactions with neighbouring molecules,
when the atoms are stretched, the dipole is larger than when but these changes are usually sufficiently small for the
they are compressed. In both cases the amplitude modula- molecular fingerprint to remain recognisable. Some selec-
tion causes the antenna to produce sidebands in the tion rules may be altered and the Raman scattering
electromagnetic radiation either side of the driving associated with rotational motion will not be present in
frequency and separated from it by the frequency of the the condensed phases. In the liquid and amorphous phases,
amplitude modulation. Thus the molecular nano-antenna the Raman bandwidth is often considerably larger,
emits light at the laser frequency nL (Rayleigh scattering) B10 cm
1–50 cm
1, than for dilute gases or crystalline

IEE Proc.-Sci. Meas. Technol., Vol. 152, No. 6, November 2005 245
forms where it may be as small as B2 cm
1. The intensity In Fig. 3 there is symmetry in the position of the bands
of Rayleigh scattering at a highly reflecting surface is B106 about the excitation frequency, but the intensities are higher
greater than that of the inelastic Raman scattering. Rosasco for Stokes spectra, confirming the earlier description. Most
[7] has estimated that the Raman/Rayleigh ratio for a 2 mm published Raman spectra involve solely the more intense
sphere of optically uniform polyethylene in vacuum is Stokes part of the spectrum, but it is now possible with the
10
12. This presents a serious problem for differentiation of new CCD detectors to scan both Stokes and Anti-Stokes
Raman signals in the design of any detection scheme. spectra simultaneously, without saturation of the detector
Several solutions are offered in the new system designs for by the intense laser Rayleigh line (nL). This offers the
disentangling the weak and specific Raman excitations from possibility of non-contact thermal measurement of sub-
the intense Rayleigh scattering centred at nL. strates and surfaces during semiconductor production
To illustrate the relative Stokes and Anti-Stokes Raman processing and crystal growth by the analysis of the
frequency shifts, we examine a full Raman spectrum differences in intensity, and using a calibrated relationship
obtained from a spectrometer. Figure 3 shows typical data with absolute temperature (T), i.e. (1).
for silicon. The Raman spectrum consists of a series of The positions for the second order bands for silicon in
vibrational bands, frequency shifted from that of the Fig. 3 are displaced from the classical mechanical Hooke’s
incoming radiation. The horizontal axis of the spectrum is law approximation of the 2o1 position (i.e. 2  520 cm
1).
given in wavenumbers (cm
1) with the zero/centre point The second order band is broadened and centred near
(nL) taken at the wavelength of the incident radiation (in this 980 cm
1. Many theoretical and detailed experimental
case for an argon ion (Ar+) laser emitting at 514.5 nm papers have discussed the further vibrational perturbations
or, equivalently, an absolute wavenumber value of for silicon that could be responsible for this effect. This
19 436 cm
1). The relationship between absolute wavenum- detail is unnecessary here, other than to appreciate that the
ber (n) and wavelength (l) is: simple classical Hooke’s law does not always hold in
Raman spectra owing to additional third order and higher
1=l ¼
n ð2aÞ order effects. Further information on these effects can be
The normal relationship between wavelength and frequency derived from high-resolution work on line shapes (e.g.
width, Gaussian, Lorentzian, etc.) and more extensive
also holds:
analysis of the frequencies of the bands in a given spectrum,
1=l ¼ n=c ð2bÞ for comparison with expected frequencies derived from
crystal lattice calculations. Thus, in a research laboratory,
where c is the velocity of light (2.99793  1010 cm/s), n is the there will always be a need for ultra-high resolution Raman
frequency of the scattered light in units of s
1 (Hz) and the systems to act as a basis for such theoretical calculations. At
wavelength l is in cm. the practical level however, once the spectrum of a validated
material has been obtained, this ‘fingerprint’ is sufficient for
5 00 000 all subsequent identification of the material.
A comparison can be made between Raman and IR
4 00 000 absorption techniques. Both originate from the molecular
3 00 000
vibrational frequencies within the material. For non-
symmetric materials the frequencies of some Raman and
2 00 000 IR bands can match almost exactly. However, there is no
direct relationship between the effects giving rise to the band
1 00 000 amplitudes (absorbance in the case of IR, and scattering
0
intensity in the case of Raman). There are cases where a
Raman band or absorption band can be present in one or
− 600 − 400 − 200 0 200 400 600 800 1000
other of the spectra. For molecules having a centre of
Raman shift / cm−1
symmetry there is no agreement between the values for the
Fig. 3 Raman spectrum for silicon showing the first orders for both band positions and all vibrations symmetrical to the centre
Stokes and Anti-Stokes scattering are IR inactive. This is due to certain polarisation and
The Stokes scattering is more intense than Anti-Stokes, and both are symmetry effects (not described here), which can apply in
approximately symmetrical either side of the zero wavenumber shift both cases. The Raman effect arises as an inelastic event
that corresponds to the frequency of the incident laser line resulting from a change in the induced dipole (p), and hence
a change in polarisation, while the IR absorption process
involves a change in the permanent molecular dipole with
A calculation of absolute wavenumber value for the bond length and angle changes occurring during vibration.
520 cm
1 Raman shifted line in silicon gives 18 916 cm
1 For Raman measurements on crystalline solids, the
(19 436 cm
1–520 cm
1) and an equivalent wavelength of intensities of the Raman bands will depend upon the
528.6 nm. To detect the equivalent band using IR absorp- polarisations of the incident laser beam and the collected
tion, we need detectors that operate at the far end of the scattered light. Most Raman systems will incorporate
mid-infrared, as the wavelength corresponding to an polarisers that can be rotated, and sample holders that
absolute frequency of 520 cm
1 is 19.23 mm. The Raman provide angular adjustment. In Raman microscopy, most
measurement can be made with the highly sensitive silicon measurements are made in back-reflection but other
CCD detector. possibilities exist to excite and detect from different
Comparisons between a spectrum generated on a directions. Recent on-line measurements for drawn polymer
wavelength scale and the Raman frequency shifted scale fibres use polarisations parallel and perpendicular to the
are easily made. For combined Raman/photoluminescence fibre chains to provide quality control of the orientation and
systems (for example, measurements on wide energy gap crystallinity [24].
semiconductors such as silicon carbide or diamond), the The intensity differences for IR absorption and Raman
axis scales are calculated automatically by the software to band intensities can be explained by the differences in
give displays of spectra in either cm
1 or nm. measurement technique. For IR, the quality of the
246 IEE Proc.-Sci. Meas. Technol., Vol. 152, No. 6, November 2005
absorption measurement is dependent on the sample been available since the 1980s. Various books and reviews
thickness and orientation of the sample to the beam; for have been published (e.g. [19]). Many of the advantages and
solid samples the surface should be well polished and non- disadvantages described for IR systems can also be
scattering. The Raman spectral intensity is directly propor- translated to FT Raman since both use near-IR detectors.
tional to the intensity of the incident radiation scattered FT Raman requires the use of a costly, high power (e.g.
from the sampling volume, and hence to the number of 1 W) Nd-YAG solid-state laser with excitation at 1.064 mm.
molecules. Detection problems exist with the presence of water (since
IR spectrometers require detectors that are sensitive to typical spectra experience interference from water absorp-
wavelengths beyond 1500 nm. Results in this region can be tion bands), and spatial resolutions are limited to 5–10 mm.
badly affected by the strong absorption of water at 3500 nm It will be seen in Section 2.3 that fluorescence in some
and even with good optical design the spatial resolution for samples can produce an intense broad-band emission on
IR microscope work is difficult to improve below 10 mm. excitation by visible and UV lasers. For FT Raman, with
However the new Raman microscopes can operate with infrared excitation, fluorescence effects can often be
samples under water, or index matching oils using confocal reduced. The niche application of specific polymer research
optics, without specific sample polishing preparation. With and quality control will remain. The important relationship
UV and visible incident radiation, focal spot sizes below for Raman scattered intensity (I) and frequency shift (n)
1 mm are possible and spatial resolution mapping is now comes from classical theory. The total intensity (I) of a
commonplace. Table 1 extends this summary of the Raman band scattered over a solid angle (4p) by one
advantages and disadvantages of the two techniques. The randomly orientated molecule excited from an initial state m
availability of complementary data makes their routine use to a final state n is given by the equation:
in the laboratory highly desirable. A combined Raman-IR
system, engineered to obtain the best of both worlds, will be 
2
described later (Section 6.1). I / I0 n4S aij mn
ð3Þ

2.2 Intensity variations, and a comparison where I0 is the irradiance of the incident laser light in
of dispersive Raman versus FT Raman Wcm
1, nS is the frequency of the Raman scattered light
The designs described in Section 3 focus largely on the new and aij is the ijth element of the scattering tensor. The
dispersive Raman systems, but FT Raman systems have equation can be reduced to the simple explanation that the

Table 1: Comparisons of Raman spectroscopy and infrared absorption spectroscopy – practical aspects

Infrared spectroscopy Raman spectroscopy

Sampling usually requires a prepared and polished sample Sample is normally by reflective scatter, with little sample preparation
required
Detection of absorption requires IR sensitive point and linear Dispersive Raman uses highly sensitive silicon detectors i.e. 2-D CCD
detectors, i.e. Ge, InGaAs, InSb, HgCdTe that need LN2 cooling. cameras with a detection range to 1 mm. Thermoelectric cooling to
FTIR systems are available
701C. FT-Raman systems are available
Infrared cannot be easily focused below 5 mm. The focus Raman microscopes give spatial resolutions below 1 mm, and smaller
degrades further in the deeper IR. Sample inhomogeneities are samples can be detected. FT-Raman resolves spatially to 5–10 mm.
averaged because of the spot size Sample homogeneity can be averaged by automatically defocusing the
beam at the sample
Sampling is usually via a macrocell, with sample holding mounts Macrocell facilities are available. The new design has an option to use a
small macrocell and accessories to be fitted to the microscope
Spectra are affected by water absorption Not affected by water, and is excellent for macro or micro measure-
ments on surfaces under water, or aqueous solutions. With resonant
Raman, low concentrations of species can be detected in solution
Difficult to make measurements on monolayers, unless by total Monolayer measurements can be made, as well as following surface
internal reflection or reflective absorption. Both require either reactions on electrodes. With the use of new low power lasers the risk
careful surface preparation or vacuum conditions of burning has diminished
Not affected by background fluorescence Fluorescent effects can now be minimised by the use of UV and near-
IR. FT-Raman uses a 1.064 mm excitation and is still better for avoidance
of fluorescence
Difficult to obtain spectra below 600 cm
1 Spectra can be obtained from 10–15,000 cm
1. The top of the spectral
range is dependent on the excitation wavelength, and the limit of
silicon detectors
High transmission optical fibres beyond 2 mm are expensive and Easily connected by focusing into high transmissive silica fibres for
less robust remote applications over 100 m. Multiplexing to several remote
detectors is possible
Requires that the molecule absorbs the photon Raman interactions are effectively instantaneous, since scattering is
inelastic. With new systems fast reactions can be followed
Relative inflexibility in sample mounting for environmental Focused beam allows easy attachment of temperature/pressure
measurements. High temperature measurements are affected accessories (and others). High temperature measurements to 1,5001C
by black-body radiation. Opaque materials are difficult to plus are possible. Black and opaque materials can be measured
measure
Weak homopolar band sensitivity Relatively intense homopolar band identification e.g. C–C, S–S

IEE Proc.-Sci. Meas. Technol., Vol. 152, No. 6, November 2005 247
intensity of a Raman band is directly proportional to the now allows the electron to spend longer in the excited state
power density of the laser at the sample. (see Section 2.1). With a further increase in energy, the
As the laser excitation wavelength moves further to the photon is in resonance with the electronic excited state, and
infrared (longer wavelength, lower wavenumber) the Ra- the magnitude of the induced dipole may be greatly
man intensity decreases (the n4 relationship). Thus, excita- enhanced. In this case intense resonance Raman bands
tion by the Ar+ laser line at 514 nm has 16 times greater are obtained. Higher energy lasers in the UV have been
output intensity when compared with excitation by a Nd- used to provide enhanced Raman outputs from molecular
YAG at 1.06 mm, assuming that the input intensities from components in biomaterials and polymers containing a
the lasers and the scattering volumes in the sample material double bond, while materials with extended electronic
are the same. Since FT Raman systems use longer conjugation (e.g. carotenoids, polydiacetylene) can exhibit
wavelength lasers and specialised IR detectors that are resonance using visible excitation [12].
typically 100 times less sensitive than the silicon CCD Resonance Raman techniques have recently been used to
preferred for dispersive Raman. FT Raman systems are 2–3 detect chromophore tags that bond to specific molecules.
orders of magnitude poorer in terms of sensitivity. Table 2 When the very high intensity of resonance Raman is
summarises the comparison of dispersive Raman with FT combined with surface enhancement effects, detection limits
Raman. below 10
10 molar concentrations (M) then become
Dispersive systems are therefore faster, and lower laser possible [29]. An application for detection of drugs and
powers can be used with less risk of sample damage. explosives will be discussed in Section 9.5.5.
Importantly, the costs and maintenance of the lower power Fluorescence photon emission usually occurs when the
lasers now used for dispersive Raman are reduced. The excited molecule is in the lowest vibrational state. The
improved speed of measurement means that fast scanning process is relatively slow (410
9 s) compared with Raman
of sample areas for quality control has become possible. scattering (o10
11 s), owing to the time required for any
Non-contact strain measurement over integrated circuit vibrational energy in the electronic excited state to decay. At
chips is now an accepted technique [25–28]. The strain at the room temperature, the fluorescence band may be relatively
surface (tensile and compressive) can be mapped in broad as several ground state recombinations may be
semiconductors (i.e. by measurement of the band position involved. The speed of the process also suggests a method
of the 520 cm
1 silicon band which will change with lattice for distinguishing Raman scattering from fluorescence; a
constant). Similarly, the degree of crystalline disorder in a high speed pulsed laser can be used to excite Raman and the
sample is indicated by the half-width of Raman bands, since detector time-gated to exclude the fluorescence band [30].
a high degree of crystallinity will give a narrow band. Some Intense fluorescence bands can swamp the sharper but
examples of strain measurement in semiconductors are less intense Raman bands, and this is often a major
given in Section 9.7.1. drawback for interpretation of potential Raman spectra in
certain materials. Figure 4 shows the spectrum of a lesser
2.3 Resonance Raman and fluorescence quality industrial diamond film where the background non-
Figure 2 shows schematically the photon-molecular inter- diamond carbon content produces an intense broad
actions for resonance Raman. Pre-resonance can occur photoluminescence band from graphitic excitations, with
when the incident radiation energy is just below that the small specific Raman band just detectable at 1332 cm
1.
required to create an electronic excitation, and is almost Also shown is the Raman spectrum for a pure diamond
sufficient for absorption to occur. The uncertainty principle film, where the background is flat and the Raman band is

Table 2: Comparison between new dispersive Raman spectrometer designs and FT-Raman

Specifications Dispersive Raman FT Raman

Laser options 244–830 nm (UV-near IR) 1.064 mm Nd-YAG (on 99% of systems)
Laser power 0.2 mW–1 kW, air cooled 41 kW typical, water-cooled
Detectors 2-D CCD silicon arrays Point or linear IR detectors
Detector sensitivity Excellent. The CCD is the detector of choice 2-3 orders worse. Little flexibility in signal
and offers great flexibility in signal processing manipulation
Noise limitation Signal shot noise limited Detector noise limited
Wavenumber coverage approx. 10 cm
1–9,000 cm
1 50 cm
1–3650 cm
1, fixed
Aqueous samples No problem with water or solvents Self-absorption, plus laser absorption
Microscope spatial resolution o1 mm 45 mm
Spectral resolution 1 cm
1 (standard) 0.1 cm
1 (special) 0.5 cm
1
Fluorescence suppression Good at 780–830 nm (laser diodes); UV options Better at 1.064 mm. No UV options for resonant
avoid fluorescence enhancements
Resonant Raman and surface Both possible giving very high sensitivity Resonant Raman for only a few materials at
enhancement 1.064 mm
Raman imaging Direct imaging and mapping (point mapping, Direct imaging not possible
line-focus imaging, true Raman imaging)
4
n scattering enhancement  3.5 1
Elevated temperatures No problems. Measurements have been made Limited to o2501C
above 15001C
Black samples No problem Intense thermal emissions usually observed

248 IEE Proc.-Sci. Meas. Technol., Vol. 152, No. 6, November 2005
 of end-to-end system test results is shown to illustrate the
advances made in Raman system specifications. Finally,
designs are illustrated for 2-D and 3-D Raman mapping,
and direct imaging. Results in this Section are restricted to
a those that demonstrate the specifications.

3.1 Laser delivery

Delhaye and Dhamelincourt [6] described the use of lasers
for Raman microscopy in 1975. The lasers were coupled
b with standard double and triple spectrographs and then
900 1000 1100 1200 1300 1400 1500 1600 1700 linked to a microscope. Therefore the system optical
Raman shift /cm−1
throughputs were inherently low, and high power CW
lasers were required. Nevertheless, with patient application,
Fig. 4 Raman spectra for a pure diamond film and a diamond film research laboratories were able to obtain significant
containing a higher graphitic content data from samples with strong Raman cross-sections.
The pure diamond film (b) shows a sharp band at 1332 cm
1, while the High power (1 W–5 W) water-cooled (Ar+) ion lasers were
impure film (a) shows a much reduced (in intensity) diamond peak
normally used. These are expensive and have high
above background luminescence. Early films often shared smaller
diamond peaks and higher background effects than the results above.
maintenance/replacement costs. They also require a supply
The spectra are vertically separated to more easily observe the of cooling-water, which can be problematic. A comprehen-
difference sive Raman laboratory might consist of an optical table
supporting several large lasers, with additional optical tables
for mounting the launch optics and spectrographs. This
well defined. In Section 3, it will be shown that the use of assembly constitutes a complicated and sensitive installation.
new laser diode sources (having wavelengths near 780– The advent of the more advanced spectrograph designs
840 nm) combined with new spectrometer designs has after 1990, with higher optical throughputs, allowed the use
provided a solution with improved discrimination of of lower power lasers (typically 10–50 mW) than hitherto.
Raman bands from broad luminescence and fluorescence Several previously unconsidered lasers have now become
interference. This has proved to be particularly useful for feasible. Examples are air-cooled Ar+ (lines at 488 and
new biochemical applications (see Section 9.1) although 514 nm) and HeNe (line at 633 nm) lasers that are lower
interestingly, in the case of diamond films, the reverse is true cost and smaller in size. A further development has been the
(see Section 9.3). increasing use of near-infrared diode lasers (e.g. 780–
830 nm) with the benefit of even smaller size.
2.4 Raman and photoluminescence Initial efforts to combine commercial diode laser sources
Unlike the Raman and fluorescence processes that originate to a Raman application were fairly disastrous. The output
from light interactions involving the sample vibrational often failed catastrophically, with several ageing mechan-
structure, photoluminescence (PL) is a radiative recombina- isms having come into play. Air-cooled semiconductor
tion effect of carriers near the band edge or impurity levels sources of this type are particularly attractive for industrial
within the energy band-gap. PL is now used as a standard applications requiring rugged, portable systems. However,
analytical processing technique for rapidly determining the the support costs during this period were high. It proved
uniformity of impurity concentrations across a semicon- necessary for Renishaw to develop its own unit and Fig. 5
ductor wafer. It also provides information on band-gap
variations across the alloy range for compound semicon-
ductors, and hence uniformity of alloy composition. PL in
diamond is important for the identification of those defects
that produce specific energy states within diamond and
other larger band-gap semiconductor and insulator materi-
als. The ability to measure both effects simultaneously is
proving to be extremely useful as a diagnostic tool. New
synchronously-controlled wavelength/frequency scans
coupled with novel interpretation of CCD chip technology
have become available, such that combined Raman/PL
spectra are providing a new level of quality-control data.

3 Raman instrumentation design and new

developments

In this Section, we describe the simple Raman system,

broken down into three basic elements:
i. Laser source and launch optics
ii. Spectrometer and integration with a microscope
iii. Detector/information processing, and display

Fig. 5 Photograph of the new stabilised semiconductor laser source

Comparisons are again made following a more detailed for spectroscopy
description of the new developments. With the combination Emission can be at 785 nm or 830 nm in the near-infrared, and output
of mechanical, optical, electro-optic and signal processing powers extend to 300 mW. The size of the enclosure is
design, the discussion has a clear engineering bias. A series 205 mm  135 mm  70 mm

IEE Proc.-Sci. Meas. Technol., Vol. 152, No. 6, November 2005 249
shows a photograph of the new solid-state laser developed [32, 33]. They are large and relatively costly but have several
in 2002 that overcame these quality problems. Some useful emission lines in the UV from typically 229 nm to
applications in industrial process control (usually involving 264 nm, with power levels above 5 mW. At first sight, these
fibre optic delivery) require a higher power output (e.g. powers might appear to be low but the n4 relationship
300 mW). It is relatively easy to reduce the power from such means that 1 mW at 244 nm is equivalent to 16 mW at
a diode unit for more standard spectroscopic uses. 488 nm. Hence, sensitivity is not a problem for a spectro-
Key design aspects were based on similar criteria to those meter using suitably coated UV optics.
specified for stabilised laser diodes in optical fibre Tunable CW lasers such as Ti:sapphire, excimer and dye
interferometric sensors [31] but with some critical differences lasers are used for scanning the excitonic states near the
in the beam profile and stability to make them suitable for band edge in 3–5 semiconductors and their alloys, and for
spectroscopic work. An external cavity has been used to time-resolved Raman work. These lasers are relatively
stabilise frequency (o1 cm
1) and mode hopping, together expensive and require regular maintenance. Band-edge
with a further external filter for suppression of side bands, scanning is highly specific research work. The continuously
necessary for artefact-free spectroscopy. The full sub-system varying wavelength of the input beam is not compatible
optical design required accurate component positioning to with the more wavelength-specific designs of the filter-based
eliminate destabilising back-reflections. Additional optical systems (see Section 3.2.1). Additional monochromator
components are used to optimise the beam for Raman optics would be required prior to the Raman spectrometer
spectroscopy. Control of the beam profile is also important stages to provide the appropriate variable wavelength
for those applications where efficient launching into fibre scanning input.
optics is specified, i.e. for portable and remote applications. The new design uses special laser guiding optics, with
Table 3 is a summary of the laser specifications. All laser novel automated alignment and focussing optics for lasers
operations and the subsequent packaging must satisfy the covering the UV (from 229 nm) through to a semiconductor
required safety specifications. For example, the diode lasers laser operating at B830 nm. For the first time, a fully
and fibre optics must include appropriate safety switches automated system has been designed for high throughput
and audible operational warning signals. Raman to include UV to IR excitations. Up to three lasers
can be mounted on a single baseplate with a mechanically
Table 3: General specifications for high power 785 and stable honeycomb structure. The various excitation wave-
830 nm laser diodes for spectroscopy lengths are switched and controlled from the PC. Figure 6 is
the photograph of a typical three-laser installation at the
General specification Value rear of the spectrometer, with guiding optics and precision
kinematic mounts for the lasers. Immediately prior to the
Wavelengths 785 nm (70.25 nm) spectrometer are neutral density and plasma line rejection
830 nm (70.25 nm) filters that are individually positioned in and out of the
Frequency stabilitya,b,c o1 cm
1 beam. Thus, 4 filters give 16 power levels etc. An alternative
Repeatabilitya,c
o1 cm
1 option is the provision of a remote shutter between the laser
Absolute powerd Z300 mW and the Raman microscope. This gives added flexibility
when measuring samples prone to laser damage, e.g.
Power stability +1%
bleaching or photochemical degradation of biological
SMSR
24 dB (0.4%) samples.
Beam sized (nominal) 6 mm  6 mm
Beam divergence o4 mrad
Pointing stability o0.5 mrad
Polarisation state Linear, vertical (751)
Polarisation purity 95%
Fibre coupling performance 45 mW–240 mW
(25 mm–100 mm)
Warm-up time 30 min
Laser safety, alarm, cut-out, EN60825, CDRH, 21CFR1040
interlock
Fibre-optic coupler C-mount
Dimensions 205 mm  135 mm  70 mm
Weight 2 kg
Laser cooling Air cooled
a
FWHM (full width half maximum)
b
721C over 16 h
c
After 30 min warm-up
d
Measured at output aperture Fig. 6 Photograph of the rear of the spectrometer enclosure and
microscope showing the mounting facilities for three lasers
Light enters the spectrometer enclosure at the back left. It is possible to
switch between lasers automatically. There is also the facility for a fibre
The initial failure of the first commercial attempt using optic input. The unit can be bench mounted, and does not necessarily
semiconductor lasers for Raman applications caused some require a vibration isolated table
users to revert to the diode pumped frequency doubled Nd-
YAG laser that gives 50 mW at 532 nm. This type of laser is
relatively expensive but reliable. A further laser launch option exists for fibre optic links to
With the higher throughputs for the new Raman systems, the spectrometer. Fibre optic links for the UV and infrared
increasingly frequency doubled Ar+ lasers have been used are not deemed to be sufficiently transparent.
250 IEE Proc.-Sci. Meas. Technol., Vol. 152, No. 6, November 2005
 For completeness, the standard optical source used for
slit
FT Raman systems is the Nd-YAG laser operating at
1.064 mm [34, 35] and, at the research level, diode lasers
operating at 1.339 mm have been proposed [36]. grating

slit
3.2 Raman spectrometer design
grating

3.2.1 Traditional dispersive spectrometers: slit

Figure 7 shows a schematic diagram of a triple grating
reflective Czerny-Turner spectrometer, traditionally used to grating
record Raman spectra. This essentially comprises two
stages: the first has two gratings, focussing optics and three
adjustable slits to filter out the Rayleigh scattered light; the
second stage comprises focussing optics, a second grating to
disperse the Raman light followed by a detector to record Fig. 7 Schematic ray diagram of a triple grating reflective
the spectrum. Optical focal lengths are typically 1 m. Prior Czerney-Turner spectrometer
to 1990, the alternative to the microscope was the addition This design formed the basis of the early Raman spectrometers
of a macro chamber and the preferred detector was a cooled
photomultiplier. Longer focal length and multiple grating
spectrometers (2–5 m) have been used for Brillouin scatter- provides a summary of the two approaches and several of
ing to give data o1 cm
1 to the laser line. As described the advantages and disadvantages are described.
earlier, double and triple units have been largely superseded
by the lower cost, filter based and more user-friendly 3.2.2 Filter based dispersive Raman micro-
spectrometers; however for some highly specific research scopes: The next two Sections concentrate on the new
applications there are arguments in their favour. Table 4 microscope developments and ongoing improvements. The

Table 4: Comparison of high throughput Raman spectrometers with double or triple grating spectrometers

Filter-based dispersive Raman Multiple grating Raman

Smaller bench mounted footprint Requires high stability vibration isolation tables. Expensive space
costs
Optical throughput 430% Throughput 1–10% depending on number of gratings
Fast measurement cycle Slower
Auto-alignment, highly stable over time Limited automation and needs regular alignment checks
Stigmatic on-axis design gives diffraction-limited image on slits Off-axis; leads to non-uniform imaging and calibration drift over
extended measurement times
Resolution 1 cm
1 across the whole range; 0.1 cm
1 option available 0.1 cm
1 available but at the cost of narrow slit width, slow
measurements and system drift
Small CCD camera directly fitted (with electronics) to the spectrometer Normally uses large camera with separate shutter and electonics
Electronic Peltier cooling to
701C used as standard Normally offered with LN2 cooled camera (expensive). Lengthy
warm-up times
Dark current o0.005 e
/pixel/s (MPP) Read out noise o10e
Dark current 0.0005 e
/pixel/s (non-MPP) Readout noise
Stray light rejection better than 1014 o10 e
. Stray light rejection comparable
Image of dispersion is confined to a low number of pixels height. No automated flexibility for binning options
Variable binning factor to enhance S/N or reduce exposure time
Synchronous scan allows 410,000 cm
1 in one scan, and retains Step and stitch used as standard for large spectral range.
1 cm
1 resolution across the range with high optical throughput. Resolution across the range varies
Step-and-stitch can be used as an option
Statistically fewer cosmic rays Worse
Direct and proven direct (true) 2-D imaging Not really available
Operates with many options for excitation wavelength from UV to near Can be used with variable wavelength input, with added grating
IR, with full automation on switching. Retains high optical throughput stage before the spectrometer. Optical throughput drops for UV
at all wavelengths systems
Near the laser line filter can be used to 10 cm
1, and high optical Measurements possible to 5 cm
1 but at the cost of long
throughput is retained measurement time and poor signal levels
Automatic and internal traceable calibrations External calibration sources required
Fully safety interlocked Class1 enclosure option Not available
Can be used for both research and QA/QC laboratories Only now used for highly specific research projects
Lower initial cost, and also lower costs for installation and High costs for purchase, installation and maintenance
maintenance
Spectrometer design optimised for use with microscope Design not optimised for microscope addition, with added loss of
throughput

IEE Proc.-Sci. Meas. Technol., Vol. 152, No. 6, November 2005 251
basic optical ray diagram of the spectroscopy system is viewing are fitted. The signal from the latter is sent to the
shown in Figs. 8a and 8b. Below, we summarise the more computer and control software, enabling comparisons of
detailed design description of Baldwin et al. [37] for the white light and Raman images to be displayed simulta-
microscope integration and confocal design. At the early neously. The development of the control software and
design stage, [38, 39], a selection of target specifications information display is described in Section 7.
(Table 5) was compiled to meet the majority of require- Firstly, we describe the passage of the input laser beam
ments. that is eventually focussed at the sample placed under the
microscope. The parallel laser beam (typically 1 mm
diameter) enters the instrument enclosure at bottom right
microscope tube
electronic
and passes through a narrow line band pass filter to remove
objective lens
camera any non-lasing emissions (Fig. 9). Figure 10 shows the light
path through an adjustable beam expander where the
expansion is the ratio of the focal lengths of the opposed
microscope objectives. A pinhole spatial filter can be placed
at the mutual focus of the microscope objectives to improve
Fo D Ft
the spatial properties of the laser beam. The use of an
a automated adjustable beam expander at this point enhances
flexibility of the system for those applications where the
beam needs to be defocussed at the sample, e.g. for larger
dispersing
slit element
area sampling (with commensurate power density reduc-
microscope tube
objective L1 L2 lens
electronic tion), signal averaging, and imaging (described later). The
camera expanded beam is reflected via mirrors M1 and M2 on to a
holographic or multilayer dielectric notch or edge filter
inclined at a deliberately small angle of incidence of
approximately 101. Polarising optics can be placed before
D1 D2
microscope
Fo F1 F2 Ft
objective NF1 NF2
b

Fig. 8
a Schematic diagram of an optical microscope with an infinity-
corrected microscope objective of focal length Fo and a back aperture
D. The tube lens of focal length Ft focuses the light on to the CCD
detector spectrometer
b The optical microscope following the design of Fig. 8a with a M2
dispersing spectrometer shown schematically in the optical path. The
spectrum is spread across the CCD detector. In the final design the
beam expander
dispersing element consists of a prism/reflective grating arrangement and
(see Fig. 10) spatial filter

M1

Table 5: Target specification for the Raman spectrometer

development
Fig. 9 Schematic diagram for introduction of the laser beam into
Spatial resolution o1 mm the microscope and to reject the Rayleigh light from the spectrometer
Spectral resolution o1 cm
1 The elements NF1 and NF2 are notch or edge filters to reject the laser
High optical throughput to the detector 430% light; NF1 also acts as a mirror, as do M1 and M2, at the laser
wavelength
Measurement near the laser line o50 cm
1
Versatile design for future upgrades and ease of use
Robust, and with increased automation

diffraction
grating
holorgaphic CCD camera
microscope
filters slit
All the specification goals were achieved, albeit with some
compromises in component selection. It was decided to
develop further modules to enhance the specifications.

A. Integration with the microscope: A Leica research-grade sample

(DLML) infinity corrected microscope is used. The
microscope has excellent mechanical and thermal stability beam expander laser
(a specialist temperature stabilisation system further im-
proves stability of the full optical design). A 5-position Fig. 10 Schematic diagram of the full enclosure for the spectro-
nosepiece allows various objectives to be incorporated meter securely fixed to the microscope
(including high NA, high magnification and long focal The ray diagram shows the beam expansion, transfer to the
length optics as required). A field of view of up to 25 mm microscope, filtering of the Rayleigh light, and dispersion across the
diameter is available, and both binocular and video camera detector

252 IEE Proc.-Sci. Meas. Technol., Vol. 152, No. 6, November 2005
or after the beam expander, and the reflective design is such collimating lens after the entrance slit and the focussing lens
that any perturbations of the input polarisation to the before the detector. Magnetic and kinematically locating
sample are minimised (refer to Section 4 dealing with system mounts have been designed for manual and automated
validation). grating changes. Gratings with groove densities from 600–
The flatness quality specified for the mirror and filter 3 600 grooves/mm can be selected for varying spectral
optics is such that the parallel beam entering the microscope resolutions and are suitable for different applications and
can be focussed to a diffraction-limited spot at the sample exciting wavelengths.
(o1 mm for visible radiation). For maximum intensity and Referring back to the optical path across the top of the
optimum spatial resolution at the sample the expanded spectrometer box (Fig. 10), an intermediate image is formed
beam should be such that it just fills the back aperture of the at the conjugate point between lenses L1 and L2, where an
microscope objective. adjustable slit (Fig. 8b) is placed. An interesting comparison
The light returning from the sample comprises both the with this approach can be made with the simple prism
Raman and Rayleigh scattered light, and is passed again to spectrometer originally used by Raman [1], illustrating
the filter arrangement shown in Figs. 9 and 10. The 451 again that the preferred measurement technique has come
beamsplitter component at the top of the microscope full circle. The chief difference is that the more recent
column is adjustable for alignment and set-up purposes. In developments have used a CCD camera as the detector
order to permit the use of different excitation wavelength instead of Raman’s photographic plate. A methodology for
lasers, a further automated feature was included to calculating the optimum light collection efficiency, and for
substitute/rotate the correctly coated component into the obtaining the best spatial and spectral resolution using
beam path. This decision was made to eliminate manual different gratings and the CCD camera chip, is described
tweaking within the spectrometer box for performance below.
optimisation. This type of opto-mechanical component Polarisation analysis of the collected Raman light can be
exploited the unique precision mechanical engineering achieved by placing a linear polariser before L1 for light
expertise available at Renishaw. Adjustments to cover the polarised perpendicular to the lines in the grating, giving
different focal lengths, as well as the optimised anti- optimum diffraction efficiency of the grating (Fig. 8b).
reflection coatings for the system operating in both the Orthogonal polarisation can be measured with optimum
UV (e.g. HeCd at 325 nm) and visible/near-IR (at 785 nm), efficiency by placing a half-wave plate in front of the
called for considerable ray trace modelling in parallel with polariser to rotate the plane of polarisation by 901.
the automated opto-mechanical design for component The overall design is such that optical throughput from
holders/mounts. sample to CCD is very high, i.e. 30%, which can be
The integration of the spectrometer with the microscope compared with typically o3% for the traditional earlier
maintains an optimum on-axis and stigmatic design [38, 39]. dispersive systems. To achieve the correct selection of anti-
The spatial resolution of the image is limited therefore only reflection coatings, enhanced-reflectivity mirrors and high
by diffraction effects. High quality achromatic lenses were efficiency diffraction gratings were called for. This necessi-
selected, and the paraxial rays alone are used to maintain tated the setting up of rigorous performance quality control
the stigmatic properties of the microscope. Figure 11 shows and test specification systems for outside supply of
a photograph of a motorised and computer-controlled components.
rotating mounting stage capable of holding back-to-back B. Filter modules for rejecting Rayleigh scattering: Referring
reflective diffraction gratings. This is placed between the to Figs. 9 and 10, we have described the optical paths to
the sample on the microscope, and then back through
the dispersive grating design to the detector/camera. In
both cases there are specific interactions with the filter
module (top left of the instrument enclosure). A number of
filter types may be used to remove the Rayleigh scattered
light.
Commonly holographic filters have been used in Raman
spectrographs over the past 15 years. Originally the
technology was developed from work on ‘head-up’ displays
in aerospace applications [8, 9, 13, 14, 40, 41]. It was shown
that notch filters could reject the unwanted elastically
scattered light to a level at which the Raman spectra could
be detected. The first devices were capable of tuning to
efficiently transmit the Raman light to wavenumbers as low
as 200 cm
1 from the laser line. Since notch filters are
transmissive either side of the blocking band, both Stokes
and Anti-Stokes bands can be observed simultaneously.
Edge filters allow transmission of the Stokes bands only.
A holographic filter is manufactured by creating a
volume phase hologram within the matrix of a photo-
sensitive gelatin. Two coherent laser beams record a
periodic interference pattern in the gelatin, which gives an
approximation to a sinusoidal variation in refractive index
through the thickness of the polymer. The period of this
spatial modulation is selected such that the unwanted laser
wavelength is partially reflected by successive modulations
Fig. 11 Photograph of the kinematically mounted and ‘back-to- in refractive index. The component is manufactured in the
back’ gratings form of a film sandwiched between glass or quartz plates.
These enable the lines per mm to be altered automatically Highly stringent specifications were applied to the filters in

IEE Proc.-Sci. Meas. Technol., Vol. 152, No. 6, November 2005 253
order to minimise fluorescence and ripple effects (see Unfortunately, holographic devices cannot efficiently
below). The efficiency of transmission depends on the filter at UV excitation wavelengths below 350 nm, and
number of modulations (i.e. the thickness of the film) and multilayer dielectric edge filters are preferred [15, 4]. These
on uniformity of the modulation spacing. Figure 12 shows are more robust and more stable under variable environ-
schematically an idealised picture of the parameters used to mental conditions than the holographic filters, and comprise
define the efficiency of notch and edge filters. In practice, layers of HfO2 and SiO2. There are trade-offs in use; the
there are ripple effects at the edges of the filter profile that Renishaw policy has been to install holographic filters for
can cause some problems in establishing a flat instrument use in environmentally controlled laboratories and relatively
response. Following their early introduction, higher optical beneficial process environments, where data can be config-
densities (ODs) up to 6 (transmittance factor is 10
6) were ured to within 50 cm
1 of the laser line. Multilayer dielectric
later developed. Selected dual high performance filters have edge filters are used for UV work and for process units
been mounted and orientated to give Raman data for some installed in hot and high humidity environments. In this
materials to 30 cm
1 from the laser line, although 50 cm
1 is way, the filter lifetimes are maximised. The technology for
generally quoted. The combination and their accurate dielectric edge filters is continually improving, and more
alignment can reduce the ripple effects at the filter edges. At recently oxide filters with little or no edge ripple have
Raman frequencies beyond 400 cm
1, the transmission is made it possible (at 785 nm excitation) to obtain measure-
typically 490% with a band-to-band fluctuation of ments within 100 cm
1 of the laser line. This means that
approximately 5% within the operating wavelength. in hot and humid countries (e.g. Taiwan, Singapore) the
user has a further option. For most process control
applications, wavenumber measurements below 300 cm
1
> 90% are not required. On reliability and cost grounds there is
therefore a strong case for preferring multilayer compo-
nents. Rugate dielectric filters [43] are also improving and
T may enjoy wider use in future.
It is noted that a transmissive volume-phase holographic
diffraction grating has also been incorporated with holo-
graphic filters to successfully make a Raman spectrograph
> OD 5.5 [14], with a link via fibre optic probe to the remote process
1/ λ line. The difficulties described above for the operation of
~50 cm−1 holographic components in humid environments and with
a UV excitation also hold for this design. Comparisons of
system designs will be drawn later (Section 8) when various
industrial applications with ‘no moving parts’ Raman
> 90% ~5%
systems are described.
C. The CCD detector: Cooled charge coupled devices
(CCDs) were initially developed for astronomy research; the
T first reported application for Raman spectroscopy was in
June 1987 [8]. Talmi [44] had earlier reviewed their use in
more general spectroscopy, and another review of the
subject is contained within Sweedler et al. [45].
> OD 4.5 The original photographic plate used by Raman was able
1/ λ to record a wide spectral range, but the analysis (optical
~200 cm−1 densitometry) and data collection were very slow. CCD
b arrays subsequently superseded the use of cooled photo-
multipliers that could only collect one single point of the
Fig. 12 spectrum at a time. After a description of the critical modes
a Notch filters and their critical design parameters
of interrogating a CCD array chip and the benefits gained,
b Edge filters and their critical design parameters
Not shown for the idealised notch filter profile is the ripple effect that
a number of results will be given to illustrate how the use of
may occur at the edges. This can be partially eliminated by appropriate CCD detectors adds to the flexibility available to Raman
angle tuning and selection. More recent developments have further spectroscopy users.
reduced ripple effects for dielectric edge filters A CCD two-dimensional array comprises a large number
of small (10–20 mm) individual square silicon detectors.
Referring back to Fig. 3, the spectrum is spread along the x-
As indicated above and shown in Fig. 9, with the two axis and each pixel relates to a different frequency. A
filters in tandem, the total OD at the primary laser quantum-efficiency greater than 20% is achieved for
wavelength can be raised to 12. The first filter F1 has the detection of visible radiation from 500–900 nm.
dual function of reflecting the laser beam into the There are several decisions to be made in the selection of
microscope and then rejecting the Rayleigh radiation from an appropriate CCD package, e.g. large chips are available
the backscattered light. The second filter is used purely to with a commensurately large number of pixels, or there is
remove the Rayleigh scattering and transmit the Raman the option of cooling the CCD package down to liquid
light. With this dual filter arrangement, more than 80% of nitrogen temperatures to potentially improve signal-to-
the Raman radiation is passed. The angle tuning optimises noise. It will be seen that the simplistic notions that ‘bigger
Raman transmission and also retains polarisation informa- is better’ and ‘cooler is better’ are not wholly relevant, and
tion. A further option is to add a mirror beamsplitter before in fact they can have considerable negative cost and
the dual holographic filters (e.g. for specialist resonance maintenance implications.
Raman measurement of wide band gap semiconductors). Around 1990, the available CCD detector packages
This leads to some loss of some system efficiency, but the involved large enclosures to include the internal vacuum
low wavenumber cut-off can be reduced to 30 cm
1. chamber containing the mounted CCD chip. A further
254 IEE Proc.-Sci. Meas. Technol., Vol. 152, No. 6, November 2005
remote box would contain the processing and interfacing 50
electronics. Given that the Raman technique requires the deep depletion

measurement of low-level signals, the added box allowed for

40
prudent screening of the electronics against interference. It
was decided that the new Raman system design should aim

quantum efficiency / %
for a detector package reduced in size and mounted directly 30
on the spectrometer. The CCD package should also enclose standard
the electronics, with special attention paid to avoiding
interference. The priority of this opto-mechanical design 20
was to create a CCD camera to meet the optimum system
performance requirements. It was not the objective to build UV coated
a camera with outstanding line-by-line specifications, many 10
items of which would be redundant within the system.
Comparisons will be made below for alternative designs
0
where high specification, free standing and sometimes 200 300 400 500 600 700 800 900 1000 1100
overly expensive CCD units have been added to a Raman wavelength / nm
system without the commensurate system analysis, resulting
in performance degradation. The end result in our design Fig. 13 Typical quantum efficiency curves for standard deep
has been the production of a small low-weight unit, with depletion and UV coated detector chips
The respective coatings improve efficiencies in the infrared and UV
thermoelectric (Peltier effect) cooling, giving a high total
system specification. A liquid nitrogen cooled solution has
proved to be unnecessary for the vast majority of
applications. near-IR visible UV
To consider the detection mechanism, the CCD is seen as
a 2-D patterned array of electrodes that provide local
potential wells below the surface, attracting the electrons
generated as a result of the incoming radiation (photons).
electrode structures
The potential on these electrodes can be controlled and used
to shift the photogenerated electrons along the chip and depleted silicon (well structures)
allow them to be counted. The number of electrons
generated in each pixel can be accurately recorded via the depleted silicon
control electronics and computer to give the Raman
spectrum. The initial chip selected had 578  385 pixels,
bulk silicon
and a pixel size 22 mm  22 mm. A simple procedure to track
a typical series of acquisition steps in a Raman measure-
ment is as follows:
Fig. 14 Schematic diagram of light impinging on a standard CCD
– The shutter within the detector box is opened for chip
exposure to start; in the new Renishaw design this is a Black arrows indicate charge transfer
mechanical motorised vane controlled by the electronics.
– Charge accumulates below the various detector elements.
– Shutter closes. structure. For IR detection (c) the longer wavelengths
penetrate deeper, and thus the well structure is designed
– Potential differences within the elements transfer the such that the high resistivity silicon substrate (deep
accumulated charge laterally from element to element to the depletion) captures these deeper electrons rather than lose
edge of the CCD chip where the charge packet is removed them to random diffusion in the silicon.
by an A/D converter with 16 bit resolution.
– The sequence is repeated. Several additional geometries and electronic designs can
be specified taking the above three structures as the basis.
The minimum shutter open time is 10 ms, although the With AIMO (advanced inverted mode operation) devices
readout speeds can be set for (a) low noise mode (33 kHz) the surface structure is designed to reduce charge injection
and (b) high speed mode (200 kHz). Cool-down time for the from surface states into the CCD and hence provide dark
camera is short (typically 30 min) compared with consider- current reduction. The chip can be mounted for front
ably longer for LN2 cooled CCD or photomultiplier illumination, i.e. through the electrode structure, and back-
detectors. illuminated (thinned) structures where the transmission loss
in the electrode layer is reduced. The cost of the final chip
Spectral response: To optimise detection of the spectrum in selection will depend on chip size, coating, thinning and
the UV and IR, there are three basic options which are at pixel quality (i.e. number of dead pixels). The new system
the basis of CCD chip design. Figure 13 shows typical design called for chips having the minimum density of
spectral quantum efficiency curves plotted against wave- defective pixels.
length and Fig. 14 is a schematic representation to show the Latest developments include on-chip impact ionisation
photoelectron effects for the different spectral regions. The CCDs. These use a high voltage in the output register to
structure of the silicon below the electrodes is shown. The accelerate the electrons to the point where they amplify the
potential well for visible detection (a) is in the depleted number of electrons at the end of the register relative to the
region. For UV detection (b), the radiation does not start. The readout noise achieved can be o1e
. On-chip
naturally penetrate the surface, and therefore a special UV multiplication has many advantages over traditional
sensitive coating is applied to convert the radiation to the intensifier options in that it has full CCD sensitivity and
visible region and hence a signal enters the depleted well speed. Limitations are more to do with issues of multiplying

IEE Proc.-Sci. Meas. Technol., Vol. 152, No. 6, November 2005 255
up spontaneous charge (dark current etc.) along with the Chip size/more pixels: Large chips are available with a larger
signal. The drives for such chips require 60 V applied to the number of pixels, e.g. 2048  2048, and reduced pixel size.
phases of the multiple register, and so the electronics This costly solution has been selected by certain users for
required are not simple. For single shot, low signal their system on the basis that ‘bigger is better’, but an
measurements these newer CCDs appear to have definite appropriate full system performance analysis would show
theoretical advantages, but for longer integration times the this is not so. The new system design described here has an
advantages diminish. These chips have originated from image size of approximately 15 mm at the slit, and then
night vision applications and so the output amplifiers are approximately 20 mm (the pixel size) on the CCD surface
more suited to high-speed readout rather than slow scan plane. This geometry was produced to obtain ‘near critical
scientific/Raman work, but this could change with new sampling’ with maximum signal in each pixel and hence
designs if the demand is there. better signal-to-noise. Changing the pixel size gives a non-
Figure 13 emphasises the enhancement in efficiency using optimum trade-off between signal-to-noise and spectral
a deep depletion chip relative to a standard chip. The resolution.
high-end infrared wavelength limit for a CCD silicon
detector is typically just above 1.0 mm. The UV coating To explain this we need to describe the methodology
enhances detection in the UV region. The lowest excitation of pixel ‘binning’, a process to reduce the readout noise.
wavelength to date for this Raman design has been As described earlier, the spectral information is recorded
209 nm. along a defined set of pixels in the x-direction, and the
spatial information is recorded in the y-direction parallel to
Low temperature options: The preferred design for the new the slit, initially set at a height of 2 pixels. With the correct
system was based on a vacuum-sealed cavity, with the chip sequence engine to control events on the chip, various steps
mounted on a stack of four thermoelectric coolers to give can be taken to optimise the signal-to-noise for a given

701C. Since Raman measurements require detection of requirement. Thus, if the spatial requirement is not high
very low light levels, the spectrometer optical design and then the y-pixels can be added together on chip and read as
camera must minimise background radiation and also a single pixel to improve signal-to-noise. Similarly, if the
incorporate low noise processing and readout electronics. Raman spectrum has a low intensity, then the spectral
Before deciding on the appropriate chip design temperature, resolution can be relaxed, and nominated pixels in the x-
it was necessary to consider the noise sources and their axis can be added on chip (binned) to improve signal-to-
impact during system operation. noise. Logically therefore, with smaller pixels binning is
necessary to get an optimum signal. This negates the
selection of a chip with increased pixels. These actions can
Shot noise is intrinsic to the statistical nature of optical be instigated via interrogation/option boxes from the
photons. Another noise source is the so-called ‘dark current computer.
charge’; this is the spontaneous thermally generated signal With bigger chips and more pixels, a larger spectral
from the CCD when not illuminated. The stigmatic (on- coverage can be obtained in one single static exposure.
axis) spectrometer images the spectrum across the plane of This is a clear advantage. However, the larger chip can
the chip, and this is contained within 1–2 pixels (in the y- lead to an inherent loss in spectral resolution at the extremes
axis) at the detector. The dark current in each spectral pixel of the range and introduce an extended non-linearity
(for an MPP CCD type) is typically 0.05 e
/pixel/s. The into the calibration. With the design described here, a
RMS noise on the dark current is the square root of the resolution of 1 cm
1 is quoted across the whole range
total accumulated dark current, and for this to be observed (with excitation at 785 nm). If a wide range shot is required
it must be greater than the readout noise, i.e. approx. for a particular application, then a lower dispersion
7 e
+signal noise. The dark current induced noise is not grating may often be the better option within the new
significant until the exposure time is greater than 980 s. The system. With the larger chips a significant lens distortion for
new Raman system design has a high optical throughput the image on the CCD can be introduced. The confocality
and most measurements take less than 60 s exposures. With (see below) of such a system is further compromised as a
a lower temperature LN2 camera, the dark current is several result of the optical distortion near the edges of the static
orders of magnitude lower, but over the same period of spectrum.
exposure there are sufficient cosmic ray events to negate any Since large chip cameras tend to be specified as stand-
advantage. Thus, the LN2 camera may appear superficially alone units, the above problems relating to Raman
to offer an improved dark current specification, but when detection tend not to be quantified in the Raman system
taking the overall system performance into account there is specification. A larger/heavier mechanical housing is
no real advantage. Further disadvantages of the LN2 option required, together with a larger and a slower shutter, so
are linked to the larger size of the unit and the problematical minimum exposure times increase. There is a risk of stray
supply of liquid nitrogen, together with the operating light ingress when the large shutter is placed outside the
lifetime compared with the Peltier cooled option. The failure camera housing. Cost and reliability (maintenance) become
mechanism for the Peltier camera generally originates in a an issue particularly in more robust applications.
loss of vacuum through the seal. However, the new vacuum
seal has been designed to last for typically 8 years, and Interfacing and sequence engine: The new CCD camera
the design is such that repair is quick, with replacement was designed to carry an on-board sequence engine to
‘plug and play’. A further advance has been the novel control the detector and also to trigger external events, e.g.
introduction of a USB interface to the CCD with important from external and linked equipment. The engine reacts to
automatic set-up and system self-checking sequences commands sent via a specially designed USB interface (the
(see Interfacing and sequence engine, below). To illustrate first for any commercial Raman instrument). Commands
the overall system performance for low noise detection, exist to:
the measurement of the very weak 4th order silicon band
is described later in Section 4.4 dealing with Validation  Control detector parameters (readout speed, on-chip
tests. binning, gain, readout areas, chip temperature etc.);
256 IEE Proc.-Sci. Meas. Technol., Vol. 152, No. 6, November 2005
 Control the digital input and output lines of the user very large spectral ranges, including combinations of
parallel interface. Raman and photoluminescence spectra. Figure 16 shows
the result for a spectrum of irradiated SiC, [47], where the
The instrument can be controlled from program objects large spectral range covered is 10 000 cm
1.
written in scripting language, Microsoft Visual Basic or Figures 11 and 17 combine to illustrate the operation of
Microsoft Visual C++. The incorporation of the USB this approach. The grating has been mounted on a stage
interface means that it can be simply plugged into most with a closed loop integral rotary encoder to give a
PCs, without the need for additional interface cards. In positioning repeatability better than 0.15 cm
1. It was
earlier Raman system designs, the interfacing was a necessary to build in-house mirror and optical mounts to
relatively messy business requiring on-board settings.
Now, the peripherals are automatically detected and the
software is configured appropriately. This is an important
factor for servicing and upgrading in the field.
excitation
Extended spectral range and synchronous detection: The CCD 488 nm
camera has the ability to record a large spectral range
simultaneously when the grating is in a fixed position. In a
single exposure up to 400–1000 cm
1 can be covered
depending on the laser wavelength and line density of the
diffraction grating. In the traditional spectrometer the grating
is first rotated to the correct position then fixed while an
exposure is made, and so on in steps to cover a larger range.
The spectral sections can then be ‘stitched’ (joined) together
using the software to show a continuous spectrum. However
there are disadvantages to the stitching method:

 Vignetting will introduce artefacts into the spectrum;

 If the CCD has non-uniformities due to charge leakage, 450 500 550 600 650 700 750 800 850 900
charge trapping, or surface dust particles for example, wavelength / nm
erroneous bands and dips are then introduced into each
Fig. 16 Synchroscan Raman (and PL) spectrum of electron-
sub-spectrum;
irradiated SiC from below 450 nm to almost 1 000 nm (a spectral
 If the material has a time-variable fluorescent back- range of over 10 000 cm
1)
ground that decreases (quenches) with time, then artificial
steps can be introduced into the spectra if they are acquired ensure mechanical stability, principally for cost reasons, and
at increasing times after initial laser exposure. because commercially available units did not meet the
required performance specification. Figure 17 describes the
Figures 15a–d show schematically the introduction of these synchronous control of the charge across the detector pixels,
various errors. and the measurement of the accumulated charge at the edge
A novel solution has been to continuously rotate (move of the array [46]. The technique also has the advantage that
in very small steps) the grating during the exposure. The a high spectral resolution can be maintained across the
charges at the pixels are shifted synchronously along the whole spectrum, unlike other designs that use a large chip to
rows [46]. The result is that spectra can be obtained free of acquire a wide spectral range.
artefacts, and it has now become possible to measure over A relatively standard method of CCD operation occurs
join
when the image illumination is scanned across the CCD
synchronously with the clocking of the active area
electrodes, i.e. in time to the readout. This is called time
delay and integration (TDI). The synchronous scanning
described in Fig. 17 is more about driving a spectrum in
synchronisation with the charge transfer, i.e. actively
synchronising a mechanical motion with the timing of the
CCD transfers rather than relying on the image drift due to
a b relative motion. The TDI concept has been generally
applied to unbinned images rather than binned spectra.
join join Discontinuities within the output spectrum can be
reduced using scanning multi-channel techniques [48, 49].
"hot" pixel
Spectral segments on either side of a discontinuity point are
combined with some software averaging. This approach is
more appropriate for ‘no moving part’ systems (described in
Section 8) where the multi-channel options of the CCD
array are exploited.
dropout

c d Infrared detection: Although infrared detectors are not

specified in the standard design, it is desirable to discuss this
Fig. 15 Schematic representation of problems associated with the
‘step and stitch’ methods used for wide spectral range scanning option briefly. Cooled germanium (Ge) single detectors, or
a Schematic of a true spectrum with no glitches InGaAs linear arrays can be used for FT Raman
b Vignetting artefacts measurements, and it is also possible to use them in
c Hot pixel and dropout artefacts dispersive Raman systems. However there is no equivalent
d Luminescence quenching artefact to the higher sensitivity (by two orders of magnitude) and

IEE Proc.-Sci. Meas. Technol., Vol. 152, No. 6, November 2005 257
 second
detector

kinematic
detector slit locators
pixels

RenCam
CCD detector

grating scans charge is

spectrum across moved across
detector in a detector in time
series of steps with spectrum movement
mirror
(attached to
second detector
housing)

Fig. 18 Fitting a second detector

An inclined mirror diverts the beam through approximately 901 to the
kinematically mounted second detector
accumulated
charge is
microscope
measured when objective L1 aperture
it reaches the
edge of the detector, a b
and the point plotted

τa τb
dl

F0 F1

Fig. 19 Schematic diagram of the scattering volume in the sample

defined by the confocal aperture for diffraction-limited optics
Fig. 17 Detector readout sequence for charge transfer during a
Synchroscan
calculations have attempted to quantify the errors caused by
variations in refraction of the light within the sample for
subsurface measurement. Distortion of the spherical
lower noise of 2-D silicon CCD arrays for visible wavefronts at a plane surface produces primarily an
wavelengths. Dispersive systems combining the outputs elongation of the confocal volume. This is over and above
from both a CCD with photomultiplier or near-infrared the potential distortions originating from optical compo-
detectors and measuring to 1.8 mm have been supplied for nent aberrations in the microscope and collection optics.
specialist applications, e.g. for the detection of the higher The unknown aspects of this problem mean that the best
intensity PL originating from shallow impurity excitations figure of merit for any system design has to be based on the
in semiconductors. Figure 18 shows the mounting position end-to-end instrument test results for a given sample.
of the second detector. A kinematically located mirror
arrangement is inserted within the space before the CCD Everall et al. [51] have described work on the use of a
detector. A slit is provided for point detectors, e.g. cooled single-mode optical fibre as the ‘aperture’. In this case, the
germanium, but is not required for array detectors. The unit spectrometer’s optics are matched to the optical fibre rather
can be switched over between the CCD and the added than to the microscope. It will be seen in the following
detector without need for realignment. discussion, that it is becoming increasingly important to
D. Confocal design: It is a major advantage of Raman include some control and aperture adjustment in defining
microscopy that it is possible to detect non-destructively confocality. The optical fibre is a novel approach for remote
sub-micron sized sample volumes within a matrix of confocal measurement, and the single-mode fibre core
surrounding material, often without any sample prepara- diameter will be limited to a specific laser wavelength.
tion. The ‘classical’ confocal system design had a pinhole Adjustment for more general measurements covering UV
(aperture) located at the back focal plane of the microscope and near infrared excitation would involve fibre changes.
to provide a given depth resolution along the optical axis. There is also additional light loss at connections when
At the basic theoretical level it is necessary to define the two optical fibres are used.
important optical volumes: i.e. the focal volume defined by The overall system design lends itself to a new and novel
the focus of the laser beam, and the scattering volume confocal design which is incorporated as standard. The
defined by the microscope collection optics (Fig. 19). result has been a reduction in alignment problems inherent
Ideally, for the highest efficiency, both volumes should be in the classical technique. A further advantage is that it
the same size and superimposed. The theory and calcula- allows for flexible and automated adjustment for both the
tions to achieve this confocal performance for a stigmatic degree of confocality and the signal output. Figures 20a and
Raman microscope have been described earlier [37]. This 20b give a simplified view of a classical confocal design. The
Reference added the effect of the scattering volume to the ray diagram shows that only rays from point a will pass
calculations of the laser focal volume by Everall [50]. These through the pinhole (aperture), and those from point a0 will
258 IEE Proc.-Sci. Meas. Technol., Vol. 152, No. 6, November 2005
 aperture

Si Raman band intensity / arbitrary units

×50 objective
a a’

×100 objective

a’
0

−10 −5 0 5 10
distance from best focus / µm
a
Fig. 21 Confocal system test using a polished silicon sample
The intensity of the silicon 520 cm
1 band has been followed while the
b microscope has been driven vertically through the beam focus at the
surface. Results are shown for two microscope objective lenses for
Fig. 20 comparison
a Illustration of the transmission of rays from a point a0 in the focal
plane with blocking of rays from a point a0 outside the focal plane by
the confocal aperture develop a sufficiently stable and robust confocal system that
b Definition of a fully confocal aperture by selection of the pixels on can be used by technician level operators for routine
the CCD. Light from the centre of the slit at point a will be recorded, measurements. The benefits of the improved spatial
whereas that at point a0 , higher in the slit will not resolution have proved useful (among other applications)
for detection of drug particles and the excipient within
tablets and mixtures for pharmaceuticals, inclusions in gems
be largely blocked. A slit of the same width as the circular and geological sections, subsurface defect identification,
aperture provides effectively the same spatial filtering [37]. polymer laminates, sectioned biomaterials, and subsurface
The ray diagram given in Fig. 20b shows that the light from strain measurement in transparent materials. For forensic
points a and a0 in the slit will be focussed onto different applications, it is important to measure samples through the
rows on the CCD. This approach gives maximum flexibility polymer package without contaminating the evidence (e.g.
in blocking the off-axis and out-of-focus light, in that the illegal drug samples).
width of the slit can be controlled (by means of software Figure 22 shows a series of scanned depth measurements
and hardware) from the computer, together with the active through a polymer laminate that has been produced for
area on the CCD (software). The degree of confocality will packaging applications. The specifications for these lami-
depend on the collection optics: the highest is obtained nates have stringent requirements relative to thickness and
when high-magnification objectives and small aperture sizes chemical properties. The manufacturer requires a quality
are used. control method to verify these properties. The laminate
It can be argued that the slit approach provides a square consists of alternating layers of two polymer types,
‘confocality’ as distinct from the circular confocality of the approximately 10 mm thick. It is seen that the different
classical method, and hence is not truly confocal. However, layers are easily detected.
the stability of the Renishaw design is such that for real Two further measurements on a sample consisting of a
system operation there are major benefits. This is particu- block of polyethylene covered by a 2 mm layer of
larly true when automated extended (time) scans of a polypropylene are shown in Fig. 23. Spectrum (b) was
sample in 2-D are carried out: mechanical drift of the taken with the microscope focussed into the polyethylene
classical pinhole, an added element in between the using a  100 microscope objective and an active height of
microscope and the spectrometer, is the dominant source
of error. System tests using both methods have been based
on the practical response to a well-characterised sample
under calibrated illumination. Measurements on a clean
and smooth silicon surface can be made. The silicon is
highly absorbing at the incident visible radiation, and the
intensity of the Raman scattered 520 cm
1 silicon band is
measured as a function of the vertical distance from the
position of best focus. The quality of ‘confocality’ is
assessed by the efficiency of the light rejection for out-of-
focus light. Figure 21 shows the plot of typical results for
step scans through the focus using two objectives. The more
strongly confocal the system, the narrower is the band. The
results give FWHM (full-width at half-maximum) values of:
– 2 mm with a  100 (NA 0.95) objective
– 5 mm with a  50 (NA 0.80) objective
Fig. 22 Confocal depth profile of a polymer laminate consisting of
These values match published values for classical confocal alternate layers of polystyrene (PS) and polymethylmethacrylate
systems [52]. (PMMA)
Further examples of confocal operation are given in The data are presented courtesy of Dr Lav"en, Technische Universiteit,
Section 9 (Applications). It has proven to be important to Eindhoven, Netherlands, who provided the polymer laminate sample

IEE Proc.-Sci. Meas. Technol., Vol. 152, No. 6, November 2005 259
 100 ‘imaging’ where an area of the sample surface is illuminated
90 and a direct Raman image of the area is obtained. The
80 former is a relatively slow process, requiring high stability of
70 the mechanical stage and accurate positional location of the
60
4 pixel steps. It is necessary to provide automated mechanical
50 movement in x and y as well as control of the z-axis, to
40 maintain the incident beam focus at the surface.
30 Confocal point mapping and line mapping system designs: It
20 was important to select the correct mechanical stage to be
10 36 pixel located on the microscope. It was found that commercially
0 available stages generally used lead-screws with no posi-
1000 1100 1200 1300 1400 1500 tional feedback control loop, or overly expensive and very
Raman shift / cm−1 heavy units with controllers. As a result, a Prior [51] (UK)
stage was modified to incorporate Renishaw linear scale
Fig. 23 Confocal measurements made at the surface of a
polyethylene/polypropylene (PE/PP) laminate, showing the effect encoders. These are now capable of giving positional
of pixel binning repeatability better than 0.3 mm, while the step-length is
With increased pixel binning (pixel height 36) there is more evidence of 0.1 mm. Comparisons (see Section 4) with un-encoded lead-
the PP substrate, as shown by the increased appearance of screw devices confirmed the improvement in precision.
characteristic PP bands near 1150 cm
1 and 1325 cm
1. For a pixel Lead-screws are still generally used on other available
height of 4, only the PE peaks are seen Raman systems. The new design allows the user to return
with confidence to a given position using the keyboard. The
dimensions of the mapped region are limited by the stage
36 pixel rows. Weak Raman bands from the polyethylene size of 100 mm  75 mm  50 mm. Other design options
substrate are seen at 1147 cm
1 and 1327 cm
1. Spectrum have included automated rotating (y, F) stages.
(a) was taken with a reduction in height of the active area to
4 rows after the relevant entry from the computer. With the A typical map is obtained by selecting a spectral range
higher resolution, the polypropylene signature has disap- containing the critical bands for the materials to be detected
peared, without any appreciable loss in intensity of the and then sequentially taking the spectra point-by-point over
polypropylene band intensities. Figure 24 shows a series of the surface of the sample. The frequencies (wavenumbers)
spectra with the focus set at the top surface of the of the critical bands, their intensities and half-widths are
polyethylene film, a  50 objective and reductions in slit recorded using the curve-fitting software. The display can
width. It is seen that as the width decreases, so the provide 2-D intensity representations that correspond to the
contribution from the underlying polypropylene decreases. presence of a given material. It is then possible to quantify
the chemical distribution and composition of the different
phases. The resolution of the spectra remains at o1 cm
1,
100 and this enables accurate changes in band position to be
90 tracked across an area and hence also gives a picture of
80 strain effects. The half-width of critical bands is also a
140 µm measure of the degree of disorder in the material, and thus
70
60
maps can be displayed for strain, disorder and composition
60 µm from the same data set.
50
Collaborative work with Steele et al. [54] (Univ. of Bath,
40 UK) has provided an excellent example of step-by-step
30 35 µm mapping. An asthmatic inhaler mixture was deposited
20 in vitro and spectra were obtained in 1 mm steps for 20 s
10
15 µm
each, using a 785 nm diode laser as the excitation source.
0 The combination asthma therapy consisted of salbutamol
1000 1100 1200 1300 1400 1500 bronchodilator and beclomethasone dipropionate (BDP)
Raman shift / cm−1 corticosteroid. The separate spectra taken from isolated
samples and a map of the combined deposition are shown
Fig. 24 Confocal measurements made at the PE surface of a
polyethylene/polypropylene (PE/PP) laminate, showing the effect of
in Figs. 25a and 25b. The result confirmed the expected
slit width particle size distributions for the two components. Any
As the slit width decreases, the vertical spatial resolution improves and chemical changes caused by drug/drug interactions lead to
there is less evidence of the PP substrate layer the presence of new Raman bands or band frequency shifts.
Since none were observed, no drug/drug interactions had
occurred, making for ultimate efficiency in use. Further
examples of drug and pharmaceutical applications are given
These measurements illustrate the considerable flexibility later (Section 9.6).
(adjustments available to the user) afforded by the new To obtain mapping spectra at a faster rate, it is possible
design compared with the more inflexible classical ap- to use a line-scan illumination. Figure 26 shows the new
proach. The point is further made that the system design design that fortuitously benefited from the presence of the
involves no compromise in performance, covering UV to intermediate slit in the spectrometer. The image of the slit is
near-infrared excitation with the new approach to auto- transposed as an illuminated line onto the CCD detector.
mated alignment and adjustment of focal lengths. Thus individual spectra can be collected simultaneously
along a line length up to 1 mm, depending on the lens
E. Mapping and imaging: There is often confusion over objective specified. Note that the line-scan system has no
the terms ‘mapping’, i.e. building up a 2-D or 3-D picture moving optical components. A cylindrical lens can be used
by a series of stepped point-by-point measurements, and as a further option.
260 IEE Proc.-Sci. Meas. Technol., Vol. 152, No. 6, November 2005
 compared with the spot-focus. There are requirements for
fast quality control measurements over extended areas, e.g.
semiconductor wafers. Figure 27 shows a scanned line map
for the polymer component in filled wood.
relative intensity

salbutamol
20 µm

BDP

1200 1300 1400 1500 1600

Raman shift / cm−1
a

BDP

salbutamol
Fig. 27 Line focus image of polymer filled wood
50 µm The sample was taken from a piece of Hoop Pine impregnated with
hydroxymethyl methacrylate and subsequently polymerised by ex-
b posure to Co60 gamma rays. The colour image of the distribution of
the polymer component is displayed superimposed on the white light
Fig. 25 Mapping tests on asthma aerosol deposits microscope image. The result is provided courtesy of P.M. Fredericks
a Raman spectra of BDP and Salbutamol, the chief constituents of the and L. Chen, Queensland University of Technology, Australia
spray
b 2-D Raman map of the deposit with the darker areas representing
the Salbutamol
Alternatives to the above line-scan methods have been
proposed for other available Raman systems. One of these
[55] expands the beam by two moving scanning mirrors.
detector The design is complex and there is a reduced light
throughput (as a result of the additional optical compo-
diffraction
grating etc.
nents) and the necessary alignment of the line-scan moving
mirrors means that the system errors and movement control
are critical. Other methods [56, 57] have also been proposed
slit
over time, and a review of point, line and global imaging
Rayleigh has been published by Markwort et al. [55].
filters etc.
True (global) imaging: At the start of the original
development of the new spectrometer featured here, the
designers were imbued with the benefits of solutions using
filters. These were successfully included for the first time to
give a novel direct Raman imaging solution [58]. This
sample proved to be possible after finding that a series of tilted
band-pass filters could be combined in a rotating and tilting
wheel such that the majority of Raman spectra, caused by
Fig. 26 Schematic diagram of the line-scan design visible laser excitation and within the range of the silicon
The orientation of the slit relative to the detector array and the CCD detector, could be covered by programming the tilt
scanning movement on the stage are shown. A cylindrical lens is and rotation of the filter wheel. Figure 28 is a schematic ray
another practical option diagram of the global imaging solution showing the
complete spectrometer enclosure. It is important for direct
imaging that the system optical throughput is maximised
The line-scan facility effectively reduces the power density since the beam is expanded (with commensurate power
at the sample, and can thus be used on samples that might density reduction) over the sample surface. The beam
be easily damaged. Referring back to (3) we have shown expander at the base of the spectrometer enclosure carries
that the intensity of a Raman band is directly proportional out this function, and the resulting expanded beam passes
to the power density of the laser on the sample. Typically, to the sample on the microscope via the normal path. After
for a focussed 1 mm spot from a 514 nm laser, the power passing back through the notch filter arrangement, the
density at the sample is approximately 5 mW mm
2, while Raman information is bent through the ‘system imaging’
for a typical line illumination of 60 mm  1 mm the mean path situated below the grating stage used for spectral
power density is approx. 0.08 mW mm
2. Thus the line- analysis. It is then filtered and finally imaged on the CCD.
focus can give a 60  shorter mapping exposure time Figure 29 shows a photograph of the filter wheel containing

IEE Proc.-Sci. Meas. Technol., Vol. 152, No. 6, November 2005 261
 beam beam
deviation deviation
mirror mirror

imaging
filters

laser
defocus

Fig. 28 Schematic ray diagram of the configuration of the system

for ‘real’ global imaging within the spectrometer enclosure

Fig. 30 A 3-D truncated intensity image of the 520 cm
1 silicon

band across a diameter of 500 mm
Background intensity corrections have been applied at the CCD. The
intensity profile over the image area is remarkably uniform and the
noise at the top of the truncated intensity profile is small

background intensity profile over the irradiated area

(100 mm diameter) of the silicon test wafer. The result
shows a remarkably flat and uniform profile to give
confidence in the data.
The acquisition of data via global imaging is extremely
fast compared to mapping techniques. Figure 31 shows data
for an area of diamond film collected after mapping with
1 mm steps and also by direct imaging. The latter took a
mere 30 s against 8 h for mapping. However, mapped data
offer more flexibility for analysis, e.g. band values and
shifts, and line shape or width changes. It is also more
difficult in global imaging to separate multi-component
phases when the critical bands are close (o5 cm
1),
although the software and subtraction techniques are
designed to give remarkably good data with practice.
Fig. 29 Photograph of a set of 20 cm
1 band-pass filters for Essentially the two methods are both powerful and
imaging in the range 459 nm to 671 nm
These are suitable for imaging Raman bands produced by 488 nm and 200 30k
514 nm excitation
150 25k

20k
the range of dielectric bandpass filters, which are rotated, 100
and tilted about an axis perpendicular to the beam direction 15k
(controlled from the computer). The filters are angled-tuned 50
10k
to transmit specified frequencies corresponding to a Raman
0 5k
band of interest. a b

A typical measurement sequence begins with determining 1000

the Raman spectrum of the phase to be detected, and then
800
selecting the band that can be deemed to be specific to that
material. The computer controlled tilted filter is set so that 600
the bandpass filter transmits the selected band and a
400
measurement is made. A second image is then obtained at a
frequency value away from the selected band and with a 200
low Raman signal; this is then subtracted from the first
0
image to give a final image with background subtraction. c
Typically the multilayer dielectric filters have a 10–
20 cm
1 bandpass. The spatial resolution remains at the Fig. 31 A comparison between true Raman imaging and point
system value of 1 mm, and areas of up to 1.5 mm diameter mapping across an area (40 mm  40 mm) of non-uniform diamond
can be imaged. Further automatic compensation processing film
at the CCD provides background equalisation over the a White-light image
b Point-mapped image
irradiated pixel area using a calibrated uniform beam and c True global image of the same area
this gives an additional correction for pixel response. To It is seen that the black and lighter areas correspond for the mapped
check that the appropriate corrections have been applied, and global images. The light areas are diamond, using the 1332 cm
1
periodic testing of the uniformity of response across the diamond band for imaging, and the dark areas comprise graphitic
illuminated area can be made on a uniform polished silicon material. The true global image took 30 s to obtain, and the point
wafer. Figure 30 is a 3-D slice through a typically corrected mapping took 8 h

262 IEE Proc.-Sci. Meas. Technol., Vol. 152, No. 6, November 2005
complementary, in that imaging gives a fast overview to
show that phases can be distinguished separately over an
area, while the more laborious mapping can be used as a
follow up with its greater flexibility in data analysis. It is
noted that mapping and line-scan methods retain the

intensity / arbitrary
confocal facility, and hence depth measurement is possible.
An alternative global imaging technique has been
proposed whereby the input laser light is first passed
through a multimode optical fibre [55]. This has two
functions: (a) to transform the input Gaussian output of the
laser into a rectangular function at the output of the fibre,
and (b) the fibre is modulated to scramble both the speckle
interference from the fibre and polarisation of the laser
light. The beam is expanded after the fibre, and results have
been shown for 60 mm diameter areas. Unfortunately, the
420 430 440 450 460 470 480 490
use of the fibre introduces additional light loss and the
limited published results have tended to relate to intense Raman shift / cm−1

Raman scatterers and polymer test beads. Multi-component Fig. 32 514 nm excitation Raman spectrum of CCl4, showing high
systems increase the potential for alignment drift. resolution of the four peaks
Further systems have been designed to combine Raman
spectroscopy with IR digital imaging, following the initial
system development at the National Institute of Health,
Bethesda, Maryland, USA, by Lewis, Colarusso and Levin, For the standard system, the resolution is always quoted
and subsequently by Spectral Dimensions Inc., USA [59] for a given excitation wavelength, and the number of lines
using a liquid crystal tuneable filter (LCTF). It is possible to on the grating. Figure 33 shows the 514 nm excitation
carry out high speed scanning over a broad spectral range. comparison spectra for CCl4 at room temperature using
For Raman digital imaging, the system optical through- 1800 and 3000 lines/mm gratings, and in +1 and
1 orders
put is reduced and a high-power laser must be used owing respectively. The higher resolution is clearly seen for the
to system throughput losses caused by the additional higher number of lines/mm grating. A further result has
components and the use of an acousto optic tuneable filter been obtained for the Raman spectrum of CCl4 with
(AOTF) or the (LCTF) [60]. Applications tend to be highly different settings of the input slit. The intensity can be
specific given the equipment cost (including LCTF and increased 7-fold, with an associated decrease in resolution
laser). Treado and Nelson [61] have published a review on of  5. The automation of the entrance slit, as well as
this subject. automatic switching of the gratings on their kinematically
located mounts, provides flexibility for adjusting the
resolution/intensity requirements for optimum measure-
4 System validation
ment speed.
Absolute frequency and wavelength calibration are
In this Section we shall describe various system validation
provided automatically by the software, using internal
tools, and certain critical measurements to demonstrate the
optical reference sources, e.g. neon. The well-defined
efficiency of the new design in relation to the target
calibration emission lines across the spectral range to be
specifications. As emphasised earlier a total system
calibrated are automatically recognised. Figures 34a and
specification was sought. This philosophy was a new
34b show the spectra of Ne emission lines with curve fitting.
approach for the design of Raman instrumentation. The
With 514 nm excitation and a grating of 3000 lines/mm,
results described here are therefore more system end-to-end
the fit gives a full-width-half-maximum (FWHM) of
tests rather than specific sub-component data. Further
0.5 cm
1. With 244 nm excitation (UV) and a grating of
comparisons are then made with the earlier multi-grating
systems.

4.1 System spectral resolution

Figure 32 shows data for the isotopic splitting of carbon
tetrachloride (CCl4), which is the accepted standard high
resolution
resolution test. The three critical bands between 460 and (3000 lines mm −1, −1 order)
intensity / arbitrary

470 cm
1 are successfully resolved. The spectral resolution

performance is defined by the ratio of trough-to-baseline to
band-to-baseline. This ratio will not be zero for samples at
room temperature because of the natural line width of the
bands. The data in this Figure give a ratio of 46%, normal
indicating a resolution approaching the best obtainable. resolution
(1800 lines mm −1, +1 order)
Theoretical calculations suggest that the best possible ratio
lies between 33% and 55%. The quoted system resolution
of 1 cm
1 is achieved for excitation at 783 nm with the
standard system, or using a 3600 lines/mm grating with a
420 430 440 450 460 470 480 490
514 nm laser. This resolution is maintained across the whole
Raman shift /cm −1
spectral range. A still higher-resolution option has also been
designed to give resolutions as low as 0.1 cm
1 and based Fig. 33 Raman spectra of CCl4 at high resolution (3000 line-
on a reflective Fabry-Perot cavity. This has found limited s mm
1 grating) and at normal resolution (1800 lines mm
1).
applications, largely for research purposes. Excitation at 514 nm

IEE Proc.-Sci. Meas. Technol., Vol. 152, No. 6, November 2005 263
 In Section 2, Fig. 3, we showed a simple spectrum of
silicon at room temperature using visible laser excitation
and a notch filter. Both Stokes and Anti-Stokes Raman
spectra are obtained. With an edge filter, only the Stokes
spectrum would be observed.
intensity / arbitrary

To compete at the specification level (for proximity to the

laser line) with respect to traditional multi-grating spectro-
meters, it was necessary to design a new filter that would
give comparable data but retain the high light throughput
for the system. The device also had to be capable of being
used with UV excitation. Therefore a novel high throughput
filter was developed to replace and fit into that part of the
enclosure normally occupied by the global imaging filter
unit. It had to be self-contained and operate as a plug-and-
18 512 18 514 18 516 18 518 18 520 18 522 play unit.
frequency / cm−1 The design was based on the original idea of Dunstan
a and Frogley [64] (Queen Mary College, Univ. of London,
UK). It allows the same NExT (near-excitation tuneable)
filter design to support several excitation wavelengths. This
means that the wavelength can be tuned to obtain the best
Raman output (i.e. signal/noise and reduction in back-
ground signal) for a given sample at low wavenumbers, by
automatically using the best filter option to match the laser
intensity / arbitrary

emission wavelength. This approach gives the user maxi-

mum signal output with either the normal holographic or
dielectric filters, coupled with a rapid computer controlled
switch to the NExT filter as required. The automated
system alignments are such that all measurements can be
taken at the same focussed point on the sample.
Several general applications for the NExT filter are
briefly noted:

39 380 39 400 39 420 39 440

– Low energy conformal changes such as hindered rotations
and structural isomerism in biological samples and poly-
frequency / cm−1
morphism in pharmaceuticals;
b
– Lattice modes in crystals (e.g. acoustic modes in polymers
Fig. 34 Calibration accuracy and resolution can be checked using that allow measurement of crystallite size for lamellae, and
an in-built Ne emission source superlattice folding modes in semiconductor device material
a Spectra obtained with 1800 lines mm
1 gratings that provide information on heterostructure layer spacing;
b Spectra obtained with 3000 lines mm
1 gratings
Excitation at 514 nm – Vibrations of compounds containing heavy atoms such as
the heavy halides used in incandescent lamps.
– Rotational behaviour of gases (e.g. leading to bond length
determination).
3 600 lines/mm, the fit gives an FWHM of 5.8 cm
1. The Ne
emission lines are also used (see the following description of The most frequently used alternative technologies for the
system stability) to give data on system accuracy and above applications are far-IR absorption and inelastic
stability over time, and provide traceability back to the neutron scattering. These are limited on grounds of their
design of the rotating grating stage. Background intensity high cost and a lack of sensitivity for the limited range of
corrections are possible following insertion of a calibrated atoms that can be covered. Furthermore, they do not have
white light source or selected glass plates. To our knowledge the high spatial resolution (o1 mm) offered by Raman
this was the first such background radiation calibration microscopy.
work for a dispersive Raman system. It was carried out in The new design involved the incorporation of a small
collaboration with the National Institute of Standards, double subtractive spectrometer into the enclosure. The
USA [60, 61]. target specification requirements were as follows (to be
compared with the standard spectrometer target specifica-
tion in Table 5):
4.2 Near-excitation line spectroscopy
With regard to filters, in the standard system design
holographic super-notch filters for the visible region will – The filter bandpass to have a wide range of typically
extend data collection to less than 50 cm
1 from the laser 1 000 cm
1 eliminating the necessity for the ‘step and stitch’
line, although with accurate alignment results can be method to join spectral regions, as used in standard double
obtained on good Raman scatters down to approximately or triple spectrometers;
30 cm
1. Dielectric edge filters for visible and near-infrared – The resolution to be maintained at better than 1 cm
1;
lasers will work to 80–150 cm
1 depending on the laser
– The lower wavenumber limit to be better than 5 cm
1
used. Holographic filters will not operate in the UV;
from the laser line;
therefore dielectric filters are used for a 325 nm laser
permitting Raman readings to within 350 cm
1 of the laser – The laser rejection at 50 cm
1 to be better than 10
13;
line (244 nm laser: 500 cm
1). – Optical throughput to be better than 30%.
264 IEE Proc.-Sci. Meas. Technol., Vol. 152, No. 6, November 2005
 The improvement over traditional methods (see Table 4)
is self-evident. In summary, the new filter aims to combine
the broad bandpass of holographic and dielectric filters with
high resolution and tunability at low wavenumbers.
Following the decision to use the Dunstan/Frogley [64]
concept, further engineering design was necessary, coupled
with correct selection of optical components. The design
also had to comply with the basic system approach to
maximise optical throughput whilst maintaining system
resolution. All aspects of the above target specification were
achieved.
To illustrate the effectiveness of the NExT design, Fig. 35
shows the spectrum for HgCl2 using visible excitation [65].
The Raman band below 10 cm
1 is clearly seen. Further
spectral data for an InAs/GaSb superlattice structure show
clean and distinct folded acoustic phonon bands (Fig. 36),
5 10 15 20 25 30 35 40
where the lowest band is at 8 cm
1. The sample was kindly
Raman shift / cm−1
supplied by Philip Klipstein (Clarendon Laboratory,
Oxford) whose earlier publication [66] described the Fig. 36 Raman spectrum of an InSb/GaSb superlattice structure
spectrum obtained using a triple spectrometer. UV excita- at room temperature using the NExT filter
tion on a TiO2 sample shows both Stokes and Anti-Stokes Bands are obtained below 10 cm
1 with a high optical throughput
emissions close to the laser line (Fig. 37). retained. Excitation at 514 nm. The measurements were made on a
sample supplied by P. Klipstein, Clarendon Laboratory, University of
Oxford, UK

50 100 150 200 250 300 350

Raman shift / cm−1
−500 0 500
Fig. 35 Raman spectrum of HgCl2 using the NExT filter to obtain Raman shift / cm −1
bands near the laser line
Results courtesy of R. Devonshire and R. Forrest, University of Fig. 37 Room temperature Raman spectrum of titanium dioxide
Sheffield, UK [65] using UV excitation and the NExT filter to show Stokes and anti-
Stokes Raman shifts close to the excitation line

4.3 Extending the excitation wavelength

Development of new laser systems and the ability of such fluorescence band and clearer Raman features. For many
high throughput Raman systems to allow the use of lower biomedical samples a move to 830 nm excitation improves
output UV lasers and automated switching between lasers matters further (see Section 9.1). Figure 39b shows how the
has been discussed above. UV excitation spectrum at 214 nm has reduced fluorescence,
Fluorescence effects are found chiefly for UV and visible but has also been used to measure the spectra of molecular
excitation where the Raman signal may be submerged monolayers without destroying the sample [67]. The
behind the broad and intense fluorescence bands. This is intensities from the 5-layer sample are  5 those of the
not an exclusive statement however and it has proved single layer and this confirms the system’s ability to measure
very helpful if the excitation can either be moved to the monolayers.
IR or further to the UV. The danger of UV excitation is
that sample material can burn even at an apparently 4.4 System stability and accuracy
low intensity. Figure 38 illustrates how excitation wave- A series of improvements over traditional instrumentation
length selection can be used to overcome the fluorescence designs have been made using a strong mechanical design
problem, taking as an example spectra from a conducting input on all components and enclosure strength members.
polymer. The novel grating stage design for the synchronous scan
Figure 39a shows the spectra of dentine (calcified tissue facility has incorporated the patented Renishaw flexible
from a tooth) with visible (514 nm) and IR (782 nm) tape scale attached (wrapped) around the outer rim of the
excitation. The latter result shows a reduction in the rotary element with the static read head facing the scale.

IEE Proc.-Sci. Meas. Technol., Vol. 152, No. 6, November 2005 265
 visible excitation
8000

633 nm
HeNe 4000

0
near-infrared excitation

780 nm 8000
laser
514 nm diode
Ar + 4000

500 1000 1500 2000 2500

20 000 18 000 16 000 14 000 12 000 10 000
Raman shift / cm −1
absolute wavenumber / cm−1
a
Fig. 38 A series of spectral measurements, from visible laser
excitation to near-infrared excitation on a conducting polymer
sample
Fluorescence may dominate the spectrum for some materials in the
five monolayer
visible region, but with near-infrared excitation the sharp Raman 150 ×0.2
bands can become evident. This is a typical example; the shape of the (offset for clarity)
curve can be shifted by using a UV or infrared laser, and optimising
according to the material response
100

This reverses the traditional stage design, where the rotary C H O

16 33
encoder is attached to the rotating spindle at the centre of 50 OH
the stage. The new Renishaw closed-loop design provides single monolayer
the highest precision presently available for a scanning
grating system. Tests were carried out to determine the
system accuracy and repeatability (precision). The (abso- 0
1000 1200 1400 1600 1800
lute) frequencies of 4 neon lines at 14220, 14430, 14885, and
14971 cm
1 from the standard calibration lamp were Raman shift / cm −1
measured and compared with the system frequency b
determined by curve fitting as a function of: Fig. 39
a Spectra from dentine (calcified tissue from a tooth) using visible and
1. Position of the measured line on the CCD; near-IR excitation. The lower near-IR spectrum has greatly reduced
2. The number of grating movements; fluorescence (luminescence) and shows Raman features that are not
seen in the visible spectrum above
3. Normal room environmental variations over one week. b UV 244 nm excitation and resonance enhancement of organic
monolayers. The top spectrum corresponds to five monolayers with
Figure 40 shows the frequencies of the 4 neon lines as they the intensity scaled by  0.2 to allow direct comparison with the single
monolayer result below
are stepped across the CCD. Each data point represents a
measured frequency error (based on Renishaw’s calibration)
versus the pixel position of the line at the time of 800
measurement. Typically bands are stepped across at 14 220 cm −1
25 cm
1 increments for calibration purposes. The result 14 430 cm −1
600
shows that independent of the rotation position of the 14 885 cm −1
grating the spectrometer accuracy is 0.2 cm
1. A full 14 971 cm −1
counts

calibration test sequence uses 27 neon reference lines, and 400

shows an accuracy of 0.2 cm
1 for the spectral region

covered by HeNe laser excitation. 200
Grating movement stability over a normal working day,
without any special room environmental control, and using 0
the 4 neon lines, shows that all the errors track and the
typical repeatability is better than 0.05 cm
1. Mechanical 15 000 14 800 14 600 14 400 14 200 14 000
drift in movement of components in the presence of wavenumber / cm−1
temperature variations has also been investigated. Further
tests using the repeatability of the silicon 520 cm
1 band Fig. 40 Calibration data using four sharp neon lines which cover
the spectral range
give similar results.
A further extension of this calibration method is to use a higher density
Figure 41 shows the spectrum for silicon, where the 4th of neon lines (e.g. 21 lines)
order band is shown. This result is almost impossible to
demonstrate using a multi-grating system owing to low
throughput and movement of the various slit and grating new design. To obtain the 4th order band typically takes
stage combinations. However the measurement and display 20 min of excitation at 5 mW power at the sample. The
of the silicon 4th order band is routinely obtained using the CCD is counting at the rate of 0.5 photons s
1, yet the
266 IEE Proc.-Sci. Meas. Technol., Vol. 152, No. 6, November 2005
6000 properties of the sample. A beautiful example of high-
stability 3-D mapping on diamond films (with continuous
5500 3rd order Si N
2 measurements taking approx. 2–3 days) is given in Section
5000 9.4.1.
4500
4.5 Polarisation
4000
O Measurements using polarised radiation to the sample are
2 4th order Si
3500 useful for determination of crystal symmetry and material
orientation [68]. The microscope will provide polarised
3000
optics for viewing the sample with white light, but for
1200 1400 1600 1800 2000 2200 2400
Raman measurements polarising optics have been designed
Raman shift /cm−1 to be placed in either the input beam (to the sample) or the
Fig. 41 Silicon spectrum showing the weak 4th-order bands near Raman scattered beam. There are theoretical calculations of
1900 cm
1 anticipated ratios for parallel and perpendicular polarisa-
The stronger 1st to 3rd-order lines are off the intensity scale. At the tion output intensities for different materials. The common
low intensity levels being detected, sharp lines due to oxygen and proof test is to observe the parallel and perpendicular
nitrogen in the air are also seen polarisation data for the CCl4 bands near 459 cm
1. A
typical result shows that the intensity scattered for the
perpendicular/parallel exposure is effectively zero; the
314 cm
1 has a depolarisation ratio of 0.75.
There are cases where the optimum measurement regime
calls for a scrambled input polarisation, e.g. for the analysis
and identification of minerals, where there may be several
different crystal structures and orientations within the same
sample. In this case a scrambler can be inserted and the
library will therefore chiefly consist of polarisation-
scrambled spectra to provide a more comprehensive
database.

5 Accessories and special designs

Special optical designs have been required to extend further

the instrument capability, and the range of applications.
Several examples are described. Applications where the
Fig. 42 Photograph of the automated microscope stage with a Raman system has been engineered to interface with
Renishaw linear encoder attached to improve control feedback, alternative commercial instruments, e.g. scanning electron
accuracy and repeatability microscopes (SEMs), are covered in Section 6.

5.1 Macro-sampling on the microscope

stability is such that the accumulated signal is clearly stage
obtained above the noise. A sampling set was designed with a microscope adaptor to
The new closed loop x–y stage for sample mounting on turn the collection lens through 901. A small base plate
the microscope is shown in the photograph (Fig. 42). placed on the microscope stage supports an array of holders
Renishaw’s linear tape scale and read head are shown at the for powders, liquids within vials/cuvettes and capillaries,
top. Tests were carried out in tandem with a stage having and microscope slides. Since the confocality of the system is
no encoder enabled to represent lead screw devices. The test maintained, measurement of bulk samples inside various
comprised measurement of the Raman spectrum of an types of container is easily made without resorting to the
approx. 1 mm silicon particle with a 1 mm laser spot and with full macro chamber design (described later).
a backlash of 10 mm enabled. The stage was moved away
and then returned to the particle to repeat the Raman 5.1.1 Free space Raman microscope: An addi-
measurement (32 times in each direction). The encoded tional microscope is added in series with the standard
stage returned to the particle every time. The recorded microscope and stage (Fig. 43). The stage is omitted for this
positions were determined after curve-fitting the silicon second microscope. A volume below the mounted micro-
520 cm
1 band. The results showed errors of 2 mm for the scope is left free to accommodate large samples (e.g. clothing
unencoded stage and 0.3 mm for the encoded version. With for forensic examination), which can be orientated by
a higher precision encoder fitted, 0.1 mm step and 0.3 mm various clamping arrangements to cover specific configura-
repeatability can now be quoted. This includes any thermal tions. The spatial resolution of o1 mm is maintained. The
drift effects over the measurement period in normal additional microscope is fixed to the standard microscope
operating environments. It is important to achieve a with the first turret beamsplitter moved in or out such that
specification at this level to meet the latest quality assurance either microscope can be used as required. The spectrometer
requirements, where sub-micrometre defects and particles beam to the microscope is infinity corrected to allow such
are increasingly analysed. There is a continuing drive to extended beam paths and hence improve operational
improve this further. flexibility. An extended honeycomb structure base plate is
The z-axis positional resolution is provided by the required in order to ensure the necessary stability.
available microscope stepped movement for fine focussing,
i.e. typically 0.005 to 0.01 mm. This resolution should not be 5.1.2 Inverted Raman microscope, and a
confused with the system’s vertical spatial resolution, which solution for multiwell analysis: Inverted micro-
is defined by the system confocality and the relevant optical scopes are used chiefly in the pharmaceutical and life

IEE Proc.-Sci. Meas. Technol., Vol. 152, No. 6, November 2005 267
 et al. [72] for in vitro and in vivo biomedical work. Such
systems were not designed for microscopy and confocal
operation, although a particularly novel use of the fibre
optic core as one of the apertures in a confocal system was
described by Everall et al. [51] to link a process Raman
spectrometer design to a microscope. This gave some
flexibility of use in that the microscope could be remotely
mounted relative to the spectrometer, but there is a
significant loss in optical throughput and hence the speed
of measurement decreases. Also the fibre confocal solution
lacks flexibility when compared with adjustable slit designs.

5.2 Fibre-optic probe heads

The new process measuring systems were designed to
incorporate fibre-optic probe heads which would withstand
potentially adverse and hazardous working environments.
Typical applications for the former systems were on
Fig. 43 Photograph of the free-space microscope in series with the production lines for quality assurance of materials and
standard microscope components, and also in-process monitoring. Examples in
The additional microscope allows the user to place larger samples Section 9.2 describe the use of probes to examine old
below the objective. Typical uses are for the forensic examination of masters and paintings/frescoes in churches and museums.
shoes and clothes The probes have also been linked into research projects,
using both the process instrument and/or the microscope
spectrometer, e.g. for crystal growth monitoring, and
sciences application areas. A large series of multiwell plates catalyst and in situ corrosion studies. Figures 44a–c show
can be simply located and registered on top of the photographs of the three probes engineered to date. All
inspection station or on the stage under the microscope. three include the standard Rayleigh line filtering arrange-
The stage can be programmed to move and measure rapidly ment in the head, but the selection of holographic or
the array of wells in a repeating step and repeat fashion. dielectric filters will depend on the cut-off required and the
This approach has been welcomed as a time-saver for operating environment. The common features are:
measurement of a large number of samples where there are
small volumes of liquid, and also for locating larger
– Rugged design to withstand typical drop testing etc.;
samples. Furthermore, it also allows for manual or
automated experiments/tests to be performed where – Flexible armoured optical fibre sheathing;
reagents are added in situ. The Leicat DM IRM/IRB – Operating wavelengths incorporate the usual visible
inverted microscope has been used for these applications. operating laser wavelengths, i.e. 514, 532, 633, 785 nm;
The multiple-well software runs as a procedure on – Optical fibre cable lengths up to 100 m;
Renishaw’s WIRE 2t instrument control and analysis
application suite. The software presents the user with a – Fitted with different objective collection lenses, including
graphical representation of the multiple-well plate or array. long working distance components (e.g. working distances
There is a series of prompts for the user to set up acquisition up to 500 mm);
parameters, e.g. acquisition time and spectral range. The – Multiplexing facilities for process analysers where data
user is able to observe results and progress in real time. The can be acquired simultaneously from multiple probes;
software also supports the Focus Trackt system that – Special adjustable tables and other mounts for fixing the
automatically adjusts the height of the sample stage to probe head in the monitoring position.
maintain focus and cover for non/uniform filling of the
cells, providing optimum signal data collection quality.
The spectra from each well are collected, and the data The most flexible probe (RP20V) can include a video
analysed in terms of the band intensity and chemical camera for direct sample viewing using the Raman
identification etc. Results are then displayed in graphical collection optic. The dimensions are 377  89  64 mm3,
form or as a colour image superimposed on a template of with a weight of 3200 g. It is used optimally to give both
the array. An example of a typical WIRE 2t screenshot is
shown in Fig. 55 later, dealing with e-beam lithography
measurements.

5.1.3 Fibre optic links and accessories: For

those occasions when remote measurements are required, or
the spectrometer has to be linked into another instrument,
e.g. a scanning electron microscope, it has been necessary to
design and manufacture specialist fibre-optic probe heads.
Connections are made to both the laser input to the
spectrometer box and the output. The first fibre optic probe Fig. 44 Photographs of fibre-optic probe heads designed for
Raman spectroscopy
heads for Raman spectrometers were proposed by Owen a RP10 is a small probe for remote and portable applications
et al. [69] and Carrabba et al. [9]. Both types were initially b RP20V is a high performance probe head with a video viewer. The
designed for process applications and the detection of user can view an image of the sample and then manipulate the position
noxious materials in hazardous environments [70]. A further of the incident beam to the appropriate point
development has been the miniature fibre heads prepared by c RP30 is an immersion probe for process applications using an
Visionex, Inc. [71] (USA), and used initially by Puppels InPhotonics Inc. design

268 IEE Proc.-Sci. Meas. Technol., Vol. 152, No. 6, November 2005
white light images from the camera and accurate spatial 5.4.1 Electrochemical cell: These cells are used for
resolution analysis. in situ corrosion studies, e.g. to measure cyclic voltamma-
The RP10 is much smaller (140  45  24 mm3), lighter grams and to detect the corresponding chemical changes.
and portable (700 g), albeit without the video camera. We More recently they have been employed to electrochemi-
have found that RP10 has gained a reputation for use in the cally produce a surface enhanced Raman scattering (SERS)
art and gemstone worlds when it is mounted on manually surface and to record their efficiency in giving enhanced
adjusted tripod stands. The laser input fibre has a core Raman signals [72]. Electrochemically generated intermedi-
diameter of 50 mm and the output fibre of 62.5 mm. ate products have been observed for catalyst studies [73, 74].
Polarisation optic options have been designed for both Figure 45 shows a photograph of the standard cell that is
RP10 and RP20V. The objective and singlet lenses are supplied with a calomel mercury/mercurous chloride
interchangeable. electrode. The distance from working electrode to viewing
The spectrometer exit port can be split such that up to window is adjustable, and the mount is compatible with the
three fibre-optic probes can be attached where multiple Renishaw macro-sampling set.
sensors are required. A computer-controlled stage mounted
within the spectrometer body focuses the laser on to the
appropriate fibre in the fibre mount.
The RP30 immersion probe has some additional features
and is based on an InPhotonics [73] design. Apart from a
small outer diameter, it has miniature dielectric filters, and
100 mm (input) and 200 mm (output) fibre core diameters
along with:

– Long sampling head for in-reactor use, i.e. 25 mm long

and 9.5 mm diameter;
– Chemically inert head with a sapphire window for high
transmission in the visible;
– Stainless steel body;
– The head can withstand pressures up to 20 MPa, and
temperatures up to 5001C.

5.3 Fibre coupled microscopes and macro

chambers
The same spectrometer, designed initially for direct attach-
ment to the Leica research microscope, has also been linked
to it via fibre optics using a multi-fibre output for different
excitation wavelengths.
For analysis of bulk samples not requiring microscope
analysis, a specialist light-excluding enclosure has been Fig. 45 Photograph of the electrochemical cell and holder
This is placed on the microscope stage to allow the operator to view in-
constructed with an internal x–y–z stage for sample
situ corrosion and electrolytic processes
positioning. There is optical fibre coupling from the
enclosure both to the laser and to the Raman microscope.
Alternatively, the specially designed Raman process spectro- 5.4.2 Humidity control cell: Changes in humidity
meter can be used instead of the microscope spectrometer. can affect the structure and properties of catalysts,
The macro chamber enclosure is comparable in user semiconductors, microbiological cells and pharmaceuticals
requirements to the IR absorption instrument sample etc. Chemical and physical structure, as well as reaction
chamber that is used for routine measurements in most rates and degradation processes can all be affected by
polymer and pharmaceutical quality and test laboratories. It changes in humidity. A development for the microscope in
is a promising sign that Raman systems are now collaboration with Surface Measurement Systems, London,
increasingly being used in this way, and have been installed UK, has resulted in a unit for direct mounting on the
for quality assurance procedures. microscope stage. This has proved of considerable use for
pharmaceutical applications where moisture absorption and
5.4 Environmental sample cells the presence of specific hydrated states is critical in defining
Three examples are included here. The electrochemical cell the optimum takeup of a drug. A typical test can comprise
designed in-house was the result of several user collabora- following the spectral changes as the percentage humidity is
tions. The pressure and temperature generating systems ramped up in stepped stages and then reversed. Significant
were based on available commercial devices and could be changes in the spectra relate to alterations in inter- and
mounted directly on the microscope stage. Some effort was intra-molecular interactions as the number of water
required on optical design and electronic interfacing with molecules in the crystal structure is increased.
the software of the control system, and also to ensure that The temperature stability is 70.11C over a range of 101C
the accessories could be locked-down and were stable. to 401C, where the difference between ambient and
Descriptions are kept to a minimum, but it is worth noting programmed temperature is 721C and assuming an
that most users will require at least one of these accessories. ambient temperature of 251C. The humidity stability is
It is only with the advent of high-speed Raman measure- 71.5% RH, over a range 0% to 95% RH. The cell can use
ment techniques that this type of in situ measurement can be an ultra-long working distance objective lens, and standard
made with changing environments. microscope slides are used as the sample mounting.

IEE Proc.-Sci. Meas. Technol., Vol. 152, No. 6, November 2005 269
Temperature and humidity around the sample are con- minimised. Standard ruby line fluorescence pressure cali-
trolled by an integrated Peltier heat pump, circulating fluid bration is used.
pump, and water reservoir. The sample window is double The evaluation by Baldwin et al. [37] at Leeds University,
glazed to prevent condensation. An organic solvent is UK has characterised the effects of strain spatially (laterally
available for working with solvents other than water. There and vertically) within the diamond anvils themselves, using
is an interface to a dedicated digital process controller for polarised light analysis of the diamond Raman band at
both manual and programmed humidity profiles. The 1332 cm
1. Further design and interfacing work has been
software allows further programmable functionality with carried out with the group led by Dunstan at Queen Mary
data acquisition and playback facilities. College, Univ. of London, UK [80], primarily for high-
pressure photoluminescence work.
5.4.3 High- and low-temperature sample
cells: The design of the microscope stage allows the 6 Raman systems combined with alternative
insertion of options for hot–cold stages Linkam UK [77]. technologies
Experience has been gained with the THMS 600 (tempera-
ture range
1961C to 6001C) and TS1500 (temperature 6.1 Raman and IR absorption
range ambient to 15001C). These cells are used to observe Table 1 summarises the benefits of both Raman and IR
phase, morphology, polymorphism and sample degradation absorption from a theoretical design perspective. For
changes. Catalyst and corrosion research in situ are frequent routine sample and material analysis in QA and test
applications. The design consists of a flat internal sample laboratories, it was concluded that the two methods are
chamber placed on a thermally stable block that is clamped compatible, since there are certain materials that are not
directly to the microscope sub-stage. For high temperature Raman active, but are IR active and vice-versa. Analysis
work the chamber can be rendered gas tight for atmospheric also indicates that the most practical IR solution in
control, and the stage body is water cooled for work over combination with the Raman spectrometer for optimum
3001C. Alternatively, coolant can be circulated through the ease of use and sample preparation is an ATR (attenuated
low temperature block. A quartz window of 0.17 mm total reflection) FT-IR system. This is commonly used with
thickness interfaces to the microscope optics, and a range of either a diamond contact ATR or a diffuse reflectance
condenser and objective lenses can be specified. x–y objective. A modular FT-IR spectrometer has been
manipulators provide additional positional control. Ther- designed to fit above the infinity corrected Leicat DMLM
mal control (which is directly interfaced to the Raman microscope of the Raman spectrometer. Figure 46 shows
system control software and display) originates from a the IlluminatIRt module supplied by SensIR Technologies
100 ohm platinum resistor connected close to the sample, Inc., USA [81] mounted on the Raman microscope. The
and gives a reproducibility o0.11C. The sample chamber inset shows a contact ATR objective. The sample can be
itself is typically 8 mmf  2.5 mm high. viewed, and the Raman spectra collected, either through the
For very low temperature work, a collaborative effort IR system objectives or through the high quality objectives
was made with Oxford Instruments plc, UK [78] to include of the Raman microscope. There are simple automated
their continuous flow cryostats. Applications have been switching instructions between video view, IR and Raman
largely for low temperature measurement of semiconductor data acquisition.
Raman and photoluminescence properties, to determine The earlier discussion on the selection rules operating for
critical alloy compositions and device efficiency. Examples both IR and Raman vibrations described how the spectra
are given later for GaN and alloys. can be mutually complementary. An example of this can be
The cryostats are clamped and supported to the
microscope sub-stage, and connections made to the liquid
nitrogen or liquid helium supplies for the ‘MicrostatNs and
MicrostatHes’ units respectively. The former operates from
77 K to 500 K, while the latter has a range from 2.2 K to
500 K (and can also be used with liquid nitrogen and
helium). They are used with long working distance
objectives (o2 mm LN2, and o3 mm He). Controls are
again linked to the overall Raman system computer control.

5.4.4 High-pressure cell: A recent review of the use

and design of a high-pressure diamond anvil cell (DAC)
coupled with the present Raman spectrometer has been
published by Baldwin et al. [35]. A DAC provided by
Diacell Products, UK [79] has been included as an
accessory. Pressures up to 50 Gpa (500 kbar) can be
achieved, and an additional ring heater permits measure-
ments to approximately 7001C. The sample volume is small,
typically 100 mm diameter and 50 mm high. The sample is
placed in the cell in a metal gasket with a pressure medium
of oil or organic solvent that is transparent to the
measurement in progress, or with a background spectrum
that can be easily subtracted. The diamond anvils can be
adjusted to ensure the anvil faces are parallel. The confocal
scattering volume defined by the microscope lenses and slit Fig. 46 The infrared module mounted within the enclosure of the
widths is smaller than the cell volume, with the result that Raman microscope
the Raman scattering from the diamond anvils themselves is Inset: Contact ATR objective

270 IEE Proc.-Sci. Meas. Technol., Vol. 152, No. 6, November 2005
 1800 1850 1900 1950 2000 2050 2100 2150 2200
wavenumber / cm−1

Fig. 47 Raman (bottom) and FT-IR (top) spectra of

K4Fe(CN)6 500 1000 1500 2000 2500 3000 3500
Additional bands are seen for the Raman spectrum
wavenumber / cm−1

demonstrated for the complex K4Fe(CN)6 where theoretical Fig. 48 Raman (bottom) and IR (top) spectra of a white paint
flake to illustrate the additional information to be gained by
calculations predict three Raman active vibrations and one combining techniques (in this case a forensic application)
active IR vibration. The result illustrated in Fig. 47 proves The IR spectrum is dominated by bands related to the organic fillers
this to be the case. and lacquers. The Raman gives easy identification of inorganic
The advantages and disadvantages listed in Table 1 are pigments
all valid for the combined system, but the benefit of having
one instrument to cover all possibilities is an attractive
proposition in terms of cost, space saving and ease of use, measurement modes: secondary electron emission (SEI) and
with a single training course for both techniques. The two backscattered electron imaging (BEI). Usually the linked
chief application areas are forensics and pharmaceuticals, Raman-SEM is asked to operate with the former mode,
although others are emerging. Figure 48 shows both since the benefits of BEI are realised at higher cost and
Raman and IR spectra of a white paint flake, where the complexity, and require an additional detector. Energy
presence of different bands enhances the potential identifi- dispersive X-ray (EDX) analytical detectors are fitted to the
cation of the sample, e.g. from a car body involved in a road SEM and yield information on the elements present in the
accident. A further example of paint identification for image. EDX is insensitive to the lighter elements (i.e.
multiple layers is given in the later forensics discussion. sodium and below), and therefore measurement and
identification of organic compounds are not possible.
Raman on the other hand can detect most materials from
6.2 Raman and scanning electron their structure and there is an extensive database of spectra
microscopy (SEM) for organics. The spatial resolution of the SEM is
It has been a technological goal for some time to merge the
approximately 3–4 orders of magnitude higher (better) than
complementary analytical benefits of Raman and SEM.
Raman. Table 6 summarises a number of advantages for
The problem has centred on optical input and alignment,
the two techniques.
with precision manipulation of the Raman excitation laser
Operation consists of targeting features on the SEM
beam, into the vacuum chamber of the SEM. Precision
image with the visible laser spot. The resulting data
engineering design has now enabled the micron scale laser
comprise:
spot to be projected on to the surface of the sample visible
in the SEM image. 1. EDX spectra;
SEMs have recently extended their measurement cap-
2. Raman spectra from identified points on the image;
abilities, with enhanced ability to operate under poorer
vacuums than previously, and with the addition of field 3. Material content maps;
emission electron guns (FEGs). There are two principal 4. The white light image.

Table 6: Advantages of combined Raman–SEM

SEM Raman

Depth of field very good and retained at high magnifications. Raman adds information to SEM
The contrast mechanisms can easily distinguish optically EDX gives elemental information only, while Raman provides
identical or similar materials structural, chemical and physical information
EDX is poor for light elements. Raman is sensitive to light element
physics and chemistry
BEI indicates mean atomic number. Raman gives chemical composi-
tion
Spatial resolution-typically 3–4 orders better than Raman Raman can target features shown on the SEM image with the laser
microscopy spot, and identify them rapidly and unambiguously. Also EDX spectra
can be acquired without moving the sample or collection optics
Cathodoluminescence (CL) measurement requires an addi- CL and photoluminescence (PL) both use the same collection optics as
tional accessory the Raman. They give data on impurities in semiconductors

IEE Proc.-Sci. Meas. Technol., Vol. 152, No. 6, November 2005 271
All this is done without movement of the sample or
collection optics. It is also possible at the same position to
obtain PL and CL (cathodoluminescence) spectra. PL uses
the laser as the excitation source, while CL uses the electron
beam from the SEM. With the Renishaw extended scan
facility the full spectral ranges for all optical emissions can
be captured without discontinuities. PL and CL will give
added electronic carrier and impurity concentration, and
strain information on the sample (e.g. semiconductors), as
distinct from the Raman data that primarily give structural
information. The design specification for the optical input
and alignment is set out in Table 7.

Table 7: Target specifications for a combined Raman-SEM a

system

Mechanical
Repeatability o1 mm for X, Y and Z
Stability (over 8 h) o2 mm for X, Y and Z
Insertion/retraction time o5 s per 100 mm
Intermediate retraction position 10 mm to 20 mm (o1 s)
Laser excitation sources
Visible 514 nm, 532 nm, 633 nm
Near IR 785 nm, 830 nm
Two as standard, and PL can be obtained for each wavelength
Raman shifta o200 cm
1 to 3600 cm
1
PL rangea 200 nm to 1 mm b
b
CL range 200 nm to 1 mm
Spatial resolutiona o2 mm
a
wavelength dependent
b
grating dependent

Applications cover several important industrial require-

ments, plus forensic detection of drugs, explosives, gun-shot
residues, fibres and pigments. Figure 49 shows the SEM
image of a material surface (a). The EDX detection
technique indicates (b and c) the presence of silicon and
carbon, but the distribution bears little relationship to the
morphology shown in (a). From the Raman spectra
obtained (d), the dark and light regions are identified as c
diamond (C) and b-SiC respectively [82]. This confirms the
power of the combined technologies. Figure 50 shows the
photograph of an SEM linked to the Raman structural diamond
chemical analyser (SCA). β-SiC

6.3 Raman and sub-micron scanning probe

microscopy (SPM)
Improvement of spatial resolution formed part of the
continuing effort to extend the Raman instrumentation
specification envelope for the new high throughput Raman
systems. This was largely driven by the desire to reduce the
feature size in silicon and III–IV semiconductor integrated
circuits. Quality assurance instrumentation demands in
semiconductor production, to detect and identify defects
400 600 800 1000 1200 1400 1600 1800
and extraneous particles directly affecting device perfor-
Raman shift / cm−1
mance and production yield, are extending down to the
d
nanometre range. It is a bonus that Raman also provides
data on residual stress. This Section describes the combina- Fig. 49 Results obtained from a combined Raman-SEM config-
tion of Raman with scanning probe technologies, and some uration
results on silicon defects. The work has consisted of two a SEM image showing the morphology of the sample
steps: b, c, 2-D analyses by EDX that indicate elemental C and Si distributed
throughout the sample
– Evaluation of the system limitations using a laboratory d Raman spectroscopy allows the dark and light regions in the SEM
rig. image to be identified as diamond and b-SiC respectively

272 IEE Proc.-Sci. Meas. Technol., Vol. 152, No. 6, November 2005
 placed very near (e.g. Bl/20) to the sample and collecting
the scattered light with conventional microscope optics or,
conversely, illuminating the sample using conventional
optics and collecting Raman scattered light through the
sub-micron aperture. This probe can be raster scanned over
the surface while maintaining the near-field condition to
generate a mapped image. The critical design components
are therefore the manufacture of a suitable aperture and the
control of the position of the aperture relative to the surface.
– Several publications have described the preparation of
tapered aluminium-coated fibre-optic probes [84, 85]. The
optical transmission decreases exponentially with aperture
diameter. Typical transmission efficiencies are given in
brackets for given aperture diameters, i.e. 50 nm (B10
7),
100 nm (B10
4), 200 nm (B10
2). There are design aspects
Fig. 50 Photograph of an SEM with the SCA interface unit
The interface has been designed for attachment to several SEM of the fibre tip that can critically affect efficiency, not least
machines of which are the self-heating effects that can affect the
metallic coating of the fibre. Judicious fibre selection is
necessary to minimise background Raman and fluorescence
– Incorporation of an engineered combined atomic force signals from the fibre itself.
microscope (AFM) and SPM with the Raman microscope – Ideally the probe tip should travel at a distance of approx.
and stage. 30 nm above the surface (i.e. much less than l/2, for visible
excitation). Usually, this is achieved using ‘shear-force’
techniques, where the probe is dithered laterally by a few
nanometres at its resonant frequency by a small piezo-
6.3.1 Laboratory experiments and system electric transducer. As the probe approaches the surface,
limitations: Construction of the first effective Raman/ surface/probe interactions cause the amplitude of the dither
SPM was described by Webster et al. in 1998 [83], and a motion to decrease; this change can be used to give
further review of progress was included in Baldwin et al. feedback control for maintaining the probe tip-sample
[37]. Figure 51 is a schematic 3-D representation of the separation over the surface topography. It then allows the
experimental design. The critical target design parameters topographic image of the surface to be correlated with the
resulted from the following analysis: spectroscopic features. Several options for monitoring the
– The standard Raman microscope test system described in amplitude are possible. In this case a laser diode/split
this review detects scattered radiation in the ‘far field’. The detector scheme was selected.
spatial resolution of the Raman microscope is thus – To generate 2-D Raman NSOM maps, the lateral
diffraction limited to approx. l/2 (half the wavelength) of positioning at the sample is also critical, given that the light
the exciting light. collection times per point are considerably longer than for
– In NSOM, the diffraction limit can be overcome by normal Raman detection. As an example, if an NSOM
illuminating the sample through a sub-micron aperture image of 25 pixels square is to be obtained (625 spectra),
with 60 s exposure per point, then it takes 10 h to complete
the process. For an aperture of 100 nm, it is desirable to
limit any sample positional drift to o10–20 nm over the
period of the test. The selection of the stage design is critical
and good positional feedback control is needed. Materials
selection for these stage components proved to be crucial in
order to minimise thermal drift.
– The preferred solution for delivery and collection of
radiation from the sample was illumination through the
aperture, with efficient far-field collection of the scattered
light by the microscope. Several options were examined, but
the selection rested ultimately on an analysis of the possible
collection efficiencies of samples having a high opacity, and
the numerical aperture (NA) of the microscope.

Most results obtained were in the backscatter configura-

tion at 601 to the probe axis, and related to opaque
materials such as silicon. The test rig was also used for
measurements in transmission.
Comparisons of signals obtained from a silicon surface
using the fibre optic aperture versus normal detection by the
conventional microscope, give a Raman NSOM count rate
of B0.05 counts/s/ mW versus B1 counts/s/ mW. The dis-
crepancy can be attributed to the different NAs involved in
Fig. 51 Schematic 3-D diagram of the experimental NSOM the two schemes, and occlusion of the scattering volume by
design, with LED detection of the NSOM probe for dither control the probe. Images over a scratch of 2 mm width and 1.6 mm
The excitation optics and the return Raman beam back to the depth in a silicon surface are shown in Fig. 52. The 3-D
spectrometer are shown image (lower image) shows the dimensional deformation of

IEE Proc.-Sci. Meas. Technol., Vol. 152, No. 6, November 2005 273
 intensity of Si 520 cm−1 Si peak Raman shift required data can be presented by defined line scans across
Raman peak (stress) the scratch.
The general nature of this result and its interpretation
have been subsequently confirmed using a combination of
AFM and Raman measurements to investigate a nano-
indentation in silicon by Gogotsi et al. [86] and later by
Demangeot et al. [87]. The initial results are shown in
Figs. 53a and 53b. A defined cross-section is shown across a
14 mm  14 mm AFM image where the height profile of the
nanoindent is shown. Raman measurements were made at
the points indicated, and the far-field spectra collected in
three spectral regions below 540 cm
1. Additional Raman
bands emerge for points in the region of deformation, and
with wavenumbers lower than 500 cm
1. These are due to
the creation of different phases of silicon. The Si (1)
520 cm
1 band shifts with the residual stress, and broadens
height (µm) near the centre, in agreement with the earlier result above.
1.6 Research continues on the definition of the different silicon
phases.
0
– ‘Apertureless’ NSOM is an alternative to the apertured
NSOM described above. In this technique the sample is
dis
ta
nc
illuminated using conventional far-field optics but light is
e scattered from the near-field of the sample by a metal probe,
(µ
m
) usually an AFM tip, as it is scanned over the surface.
10
Coating the tip with a material such as silver, which may be
Fig. 52 NSOM-Raman plots of a scratch in a silicon wafer used to induce SERS, and maintaining it close to the focus
The 3-D image (lower) shows the topography of the scratch, and the of the incident laser light on the sample, can be used to
maps above show the intensity (left) and peak shift (stress) for the Si enhance the Raman signal. Work by Hartshuh et al. [88]
520 cm
1 line. The data can be directly converted to stress values. The and Wang et al. [89] (from the present group) have
band half-width data also provide information on the degree of
demonstrated excellent spectra and maps with a spatial
disorder
resolution of approximately 25 nm on carbon nanotubes
and C60 respectively. Subsequent information on a
controllable method for the preparation of the metallised
the silicon in the region of the scratch. A series of spectra probes for efficient scanning near-field optical microscopy
containing the silicon 520 cm
1 Raman band, across and has been given by Wang et al. [90]. The method, a
around the scratch, is shown. It is seen that the intensity of modification of the mirror reaction method, has produced
the band decreases towards the centre (black). The band controllable silver particles of the order of 80 nm diameter
shift data can also be converted to values equivalent to for apparent optimisation of the tip enhanced Raman
residual stresses, i.e. compressive and tensile (top right) signal, and confirming a resolution of 24 nm, this time
based on the conversion assumption [27] of 0.5 GPa cm
1. with carbon nanotube samples. The size of the particle
In the centre of the scratch no curve fitting was possible due can be controlled by the mirror reaction time. Emery
to the line broadening as measured by band half width, and et al. [91] showed that the surface enhanced Raman
an associated decrease in intensity. It was concluded that spectrum can be obtained from a single molecule. The
this effect is due to conversion to amorphous and other possible theoretical improvement calculated by Wessel [92]
phases of silicon, i.e. increased crystalline disorder. If is 8 orders.

Fig. 53 AFM-Raman combination used for measuring across a nanoindentation in silicon

a Optical image of a nanoindentation
b Intensity profile of the Si 520 cm
1 line along the dotted line over the nanoindentation
Raman spectra recovered from points (A, B, etc.) along the same line show that various disordered phases of silicon are found following the
differential stress that has been applied. Data courtesy of Y. Gogotsi, Dept. of Materials Engineering, Drexel University, USA, and R. Noons, R.
Forrest, S. Ward and R. Devonshire, Dept. of Chemistry, University of Sheffield, UK

274 IEE Proc.-Sci. Meas. Technol., Vol. 152, No. 6, November 2005
6.3.2 Engineered Raman system with NSOM Software integrated with the Raman microscope has
and AFM: The above tests confirmed that the combina- been developed [94]. The SPM requires software for the
tions of Raman, NSOM and SPM (e.g. AFM) could be probe positional controller and also the added specialised
useful, but that it would be necessary to rationalise the real-time image display, image acquisition (up to 8
components for the NSOM/AFM part. Ideally, the sub- channels) and analysis. The combined software gives a
system should be capable of mounting on the standard 3-D rendering of Raman, AFM and NSOM.
Raman microscope sub-stage, and have the necessary
stability for nanometre scale measurement. Accordingly it
was decided to integrate the Nanonics (Israel) [93] NSOM/ 7 Software, controls and spectral libraries
AFM 100 Confocalt. Figure 54a shows a photograph of
the system mounted on the Raman microscope and Fig. 54b 7.1 Software and controls
the Nanonics cantilevered fibre optic tip. Once the first high-throughput Raman systems had been
Some of the design features are briefly summarised: constructed, the emphasis was then placed on the
incorporation of additional hardware and software controls
– The cantilevered optical fibre is held between the objective for improved automation (e.g. lens focussing, grating
lens and the sample without obstructing the gathering of movement, auto focus and automated calibration). This
far-field data. For NSOM investigations the laser excitation also included the development of the interfacing for
is transmitted through the sub-wavelength fibre optic probe. mapping and imaging. Data acquisition and signal proces-
The probe moves over the surface of the sample with the sing from the CCD called for the inclusion of various novel
options of contact, non-contact or intermittent contact, and commands for the new confocal and synchronous scanning
the feedback mechanism is optical beam deflection. techniques. Linkage of the Raman system with automated
stages, and combination with other measurement techni-
– The scanning stage has a low mass (75 g) and will give a ques, required the development of an advanced and flexible
positional spatial resolution of o1 nm (x–y–z). The scan interface.
ranges are typically 70 mm (x–y–z). There is also piezo- An indication of typical screen displays for different parts
actuated coarse positioning. of the WIREt (Windowss-based Raman environment)
– The NSOM resolution depends on the aperture size of the system in operation is probably the optimum way to
probe tip: as low as 50 nm can be achieved. For AFM describe the development. This is the first time that many of
measurements, the z-resolution is o1 nm, while the x–y these facilities have been applied to dispersive Raman
resolution depends on the radius of curvature of the tip systems.
(Nanonics claim the best achieved is o1 nm). The same Figure 55 shows the display for Raman mapping of
configuration is also used for apertureless tips with a small an e-beam lithography structure, with the white light image
silver particle attached to obtain resonance enhancement (video), cross-hairs with Raman spectra shown along the
(see Section 6.3.1 above). x-hair axis. There are scalable panels with multiple panel
viewing. Fully integrated software for hardware control
with selection of the light path (e.g. spectral analysis or
imaging), together with the various exciting laser and
grating options have also been developed.
The system and accessories can be controlled with
user-defined settings. The calibration sequences run auto-
matically. The x–y–z mapping stage displacement limits
can be set and step sizes defined. The system can then
be set to run automatically for several hours to complete
the map.
The system configuration details are listed, offering
a range of options and settings. A control bar provides
access to further levels of control. The instrument commu-
nication and queue allows for automatic sequential running
of tests and data collection. The chief benefits are that
previously time-consuming measurements can now be taken
unattended or be carried out by relatively unskilled
personnel.
Software interfaces have been developed to facilitate
interfacing of the Raman system to other subsystems. The
latest Microsofts Windowss platform has been used,
which is compatible with the appropriate quality standard
21 CFR Part 11. There are also additional software
standards for compatible interfacing to specific industry
requirements (e.g. pharmaceuticals). Full spectra analytical
and image manipulation tools are provided with a link to
Galactic Industries Inc. GRAMSt software. A further
chemometrics package is available (PLS-IQ). Shaver [95]
has published a review of chemometrics related to Raman.
Section 8 describes the design of hardware and software
Fig. 54 for process applications. Custom PASSt software (Process
a Photograph of the combined Nanonics NSOM/AFM mounted on Application Spectroscopy Suite) has been further developed
the Raman microscope stage for automation, interfacing to sample handling, and
b A close-up of a cantilevered fibre-optic tip compatibility with the Raman process units to be described.

IEE Proc.-Sci. Meas. Technol., Vol. 152, No. 6, November 2005 275
 800
−40
750
−20
700

0 650

20 600

40 550
− 20 −10 0 10 20
−40 −20 0 20 40 60 Y
700
700
600
650
500
600
400
550
300
500
200
450
100
400
350 0
−10 0 10 20 300 400 500 600 700 800
x

Fig. 55 Typical screen dump for Raman measurements showing multiple panes with spectral analysis, imaging and positional control of the
beam
The plots show the intensity of the 520 cm
1 silicon Raman line in x and y (scales in microns) for an e-beam lithography test structure

7.2 Spectral libraries The database for polymeric materials includes rubbers
The first specialised Raman spectral databases have been and polymers, plus many of their variants. Part of the
produced for this new generation of dispersive spectro- polymer library has been developed in collaboration with
meters. They have made it possible to further de-skill the Queensland University of Technology, Australia. The
Raman operation. In the past, databases were unreliable analytical printout includes the sample description, e.g.
owing to the variability in spectra obtained (e.g. signal/ generic name, trade name, cross-section, manufacturer, fibre
noise, background) by the earlier low throughput multi- form and appearance, and diameter.
grating systems. These Raman systems were also not Figure 56 shows a typical output after running the
appropriate for technician-operation and routine materials database search. The upper spectrum is the result from the
identification. The new specification followed from the sample and the lower spectrum is the top match from
following broad requirements: evaluation of the whole spectrum. Therefore the user can
make a visual assessment of the match. At the bottom of
1. Multiple search algorithms for sample identification. the printout is a list of the likely ‘hits’ in order of fit, i.e. a
2. Multi-database search capability where databases can be type of quality index.
grouped and searched together to give extensive coverage. A minerals library based on internet-submitted data has
3. Operable with the existing Raman software analysis also been published on the web by George Rossman,
software. California Inst. of Tech., Pasadena, CA, USA [96].
4. The need to allow for development of customer databases
in conjunction with relevant Renishaw databases.
8 Raman for process applications

The spectral databases were generated from a range of 8.1 Design aspects
fully traceable samples including gemstones, oxides and The latest generation of Raman spectrometers has opened
other inorganic materials and other related materials such up a new range of process and on-line applications. Many
as polymers. The information stored with each spectrum of the early developments in this area were instigated by
includes the chemical formulation, together with related users of the Kaiser Optical Systems Inc. spectrometer which
compounds. relies on holographic filtering of the laser line, and uses a
The forensic database includes drugs, e.g. amphetamines, single component [14, 97], comprising two transmission
diamorphine, cocaine, heroin etc. (see also Section 9.5). It gratings, to split the output to the detector such that
also includes drug precursors to provide identification of the different spectral ranges are stacked above each other on the
manufacture, e.g. acetic anhydride, ephedrine. An explo- CCD. Figure 57 schematically illustrates an analogous
sives library has been generated in collaboration with the system used by the present group using a simple transmis-
Police Scientific Development Branch (UK) to include sion grating. The simultaneous measurement of these
DNT, MNT, PETN, RDX, TNT. Data on fibres and spectral parts can then be recombined to give the whole
paints are continually being added. spectrum. Carrabba et al. [9] also proposed using an echelle
276 IEE Proc.-Sci. Meas. Technol., Vol. 152, No. 6, November 2005
 counts

500 600 700 800 900 1000 1100 1200 1300 1400 1500 1600 1700 1800 1900 2000
Raman shift/cm−1

Fig. 56 Typical result for checking the observed spectrum versus the stored library spectrum
A ‘hit list’ (quality index) for identification is displayed together with the material formulae and other materials information. These data are taken
from the polymer library

spectrograph to split the spectra in a similar fashion, and and operate in static mode over the required, and more
used the system for non-laboratory applications. The chief limited, spectral range.
benefit of this approach was that the systems had no An alternative method of stacking spectral segments has
moving parts, and could cover the full spectral range been proposed (Smith et al. [98]) which uses dielectric filters
simultaneously. These techniques do not necessarily fully to break up the output light into several beams prior to a
overcome the step-and-stitch discontinuity errors that can reflective or transmission diffractive grating (Fig. 58). This
occur as each range is overlapped in the signal processing. was labelled the ‘sandwich scope’. Precise mechanical
The synchronous scan method used in the new design of alignment in system set-up of the filters allows the diffracted
spectrometer highlighted in this review (see Section 3.2.2) outputs additionally to provide a series of stacked spectral
certainly overcomes this problem and retains the spectral parts at the detector. The number of spectral segments can
resolution across the whole range, but it is slower if a wide be selected depending on the spectral range required. This
spectral range is required. Generally speaking, process development originated from concern over the longevity of
applications require a faster update rate (e.g. seconds for the holographic transmission gratings in common industrial
spectra) and a poorer spectral resolution (e.g. 10 cm
1 and process environments. The echelle solution required too
higher) than that provided by laboratory instruments. For many optical components for alignment, and was expensive
industrial applications some users nevertheless still prefer to produce.
this latter scanning solution, and tune the grating to the All of these systems are linked via fibre optics to probe
required range during installation trials and then lock down heads that include the laser line filtering optics. Some degree

edge
filter B CCD
CCD camera
lens detector
diffraction grating
edge
filter A

collimating lens CCD lens

slit / pin hole transmission
grating
light source dispersed
spectra
Fig. 57 A schematic ray diagram showing a method for stacking a mirror
series of spectra at the CCD using a transmission grating
input
This type of approach was successfully used in the industrial
beam
environment following the work of Battey et al. [14] using a edge filters
transmission holographic grating and holographic notch filters. In
this design dielectric edge filters are used in conjunction with a simpler Fig. 58 A schematic ray diagram using a reflecting grating after
transmission grating Smith et al. [96], with the extended spectra stacked at the CCD

IEE Proc.-Sci. Meas. Technol., Vol. 152, No. 6, November 2005 277
of multiplexing is possible with multi-option laser inputs via The above units also offer the data handling and
fibre optic links, and for up to 4 remote fibre-optic Raman processing functions available in the laboratory instrument,
probes including linkage to a remote microscope (Kaiser plus the WiRE Process Application Spectroscopy Suite
Optical), and 8 for the Renishaw design. A probe head and (PASSt). Essentially PASSt customises the WiRE soft-
fibre-optic link to the laser have been designed in ware to operate in a particular user defined running mode.
collaboration with the UK police, for the remote detection This ensures that the sequence of operation, data and result
of explosives [99]. saving is automatic and PASS/FAIL criteria can be set. In
Therefore in summary, the strategy evolved into the summary:
development of three process instrumentation systems:
Analysis and tolerancing
(a) A cut-down and ruggedised version of the laboratory
instrument, to retain the same high resolution, but allow the 1. Up to 32 separate analysis values may be defined from:
user to set the necessary limited spectral range for the a. Data (intensity or integral) under data/background
process and then operate with the moveable grating stage
b. Curve-fit (any curve-fit parameter)
locked down in static mode (System 100). The instrument is
over-specified for most process applications, but in the early c. Similarity to a given spectrum
stages of process investigations this flexibility has proved d. A combination (+
* C) of any two analysis values
useful. This unit and those below can also carry the auto- 2. Upper/lower tolerance limits may be set for each value
calibration and background response corrections.
3. User-defined tolerance failure messages and help files
(b) A series of sandwich scope designs with 2 or 4 diffracted
beams stacked at the CCD depending on the spectral ranges 4. Action on a tolerance failure may be set to none or:
required. Typical resolutions of 2 cm
1 are achieved for a a. Warn user (pause program)
four-stack system. b. Treat as error (halt program)
(c) A minimalist spectrometer with a single fixed grating and
the required spectral range illuminating the CCD in a single Output
line. Both the grating and the CCD are optimised for the
application. This was intended for high speed/low resolution 1. Analysis values may be sent to: display; printer; com port
work (above 5 cm
1), and ideal for the most popular 2. Spectral data to *.SPC (user prompt for filename if
operating region (i.e. 400–2000 cm
1). The system could required)
also be placed inside a small suitcase (Fig. 59) for those
3. Spectral and analysis files may contain results of one run;
users requiring easy portability, e.g. detection of drugs at
one sequence; every sequence; or all results of one user
remote locations. With the advent of higher power (200–
300 mW) and reliable near-infrared diode laser sources, the 4. Support for extra output text files e.g.
size and weight of these systems has shrunk considerably. – Analysis result file (*.TXT): measured analysis values
The importance of near-infrared lasers for the reduction of – Result log file (*.LOG): overall PASS/FAIL results
background fluorescence in process streams has been
confirmed. – Tolerance error log (*.ERR): results for each failure

There are the usual password protection and operator ID

entry facilities. Several separate tracks on the CCD may be
defined, and these are used (depending on requirements) for
calibration spectra, background subtraction, and instru-
ment response correction. Work continues on updating the
software.

8.2 Typical examples

Process measurements tend to fit into two categories: (a)
where the Raman is used to confirm that a process is
completed and that it should stop, and (b) where the
selected quality index of a product is continuously
monitored to maintain the economic process yield. Brief
examples are given here and also in Section 9.6.
(a) End-stop measurements for glyconate production: Figure
60a is a schematic diagram of a process reactor to carry out
glyconate polymerisation. Initial tests were run with the exit
beam from the fibre optic probe focussed through a window
which was pressure sealed at the top. In production a
second probe was also used from below in similar fashion.
The design of the windows and seals, and procedures for
maintenance are not discussed here. Figure 60b shows the
Raman spectra for a mixture of vinyl acetate and versatic
acid vinyl ester. Results are taken on heating, cooling and
the different rates of heating. The abrupt change in polymer
structure is clearly seen by the changes in band intensities.
Fig. 59 Photograph of a small Raman suitcase spectrometer In both these applications a Raman System 100 was used.
coupled to a fibre-optic probe for remote and portable measurements. During the initial evaluations a wider spectral range was
The suitcase size is approximately 50 mm  50 mm covered, and once the critical (more limited) spectral
278 IEE Proc.-Sci. Meas. Technol., Vol. 152, No. 6, November 2005
 drawn fibre
fibre optic probe

nozzle

1200

1000

800

Fig. 60 600
a Schematic diagram of a high temperature reactor vessel with
windows for insertion of light from Raman probes in the top and
bottom 400
b Raman spectra from the glyconate process vessel over time.
Excitation was at 785 nm to avoid any fluorescence effects. During the 1100 1200 1300 1400 1500
evaluation period measurements were made for both heating and delta wavenumbers, cm−1
cooling cycles b
Data courtesy of Wacker Chemie, Germany
Fig. 61
a Schematic of the fibre optic probe and the excitation beam (633 nm)
fingerprint region was determined, the grating was locked focused on the nylon fibre just above the nozzle and during drawing
from the melt
down.
b Raman spectrum of the nylon fibre while it is being drawn at 100 m/
(b) On-line measurement: Measurements of nylon fibre min. Typical update rate was 5 s
characteristics were taken while the fibre was being drawn. Data courtesy of DuPont, Gloucester, UK
The set-up is schematically illustrated in Fig. 61a. The beam
from the mounted fibre optic probe was focussed on to the
fibre being drawn at the rate of 100 metres per minute and instrumentation design that has contributed to wider
higher. The crystallinity of the fibre is monitored primarily acceptance of Raman techniques.
by evaluation of critical band half-widths. The stability of
the data under such conditions is remarkably good and 9.1 Biomedical applications
Fig. 61b shows clearly defined bands that can be rapidly
processed for changes in both intensity and half-width.
9.1.1 Detection of cancer: internal (oesopha-
gus, colon etc.): The in vivo detection of cancer and
the precursor stages for cancer within internal tracts using
9 Applications Raman spectroscopy has now reached the stage where
in vivo clinical trials are a real possibility. An early review of
The range of applications has grown in tandem with the the biomedical applications of Raman spectroscopy was
new instrumentation. It is impossible to cover them all published by Mahadevan-Jansen and Richards-Kortum
within the scope of this review. Apart from referencing [100]. Subsequently several groups have continued to
several reviews for each application area, the discussions develop this field of application, among them the spectro-
will focus on those projects where some kind of benefit to scopy group led by Feld at MIT [101–103]. More recent
the user has been demonstrated. The examples will largely reviews highlighting the transfer to the clinical arena have
be taken from collaborations with other groups and been published by Mantsch et al. [104] and Puppels et al.
industries. They also illustrate particular aspects of the [72].

IEE Proc.-Sci. Meas. Technol., Vol. 152, No. 6, November 2005 279
 There has been collaboration with many groups, but changes associated with the onset of dysplasia. The spectra
attention is focussed here on the work of G. Puppels have relatively broad features, and there is therefore a
(Erasmus Univ. of Rotterdam, Netherlands), and H. Barr degree of reliance on the efficiency of the subsequent data
and N. Stone (Cranfield Univ. and Gloucester Royal analysis to extract useful information. The data processing
Hospital, UK). Work has concentrated on the detection of techniques employed show good correlation for the
the pre-cancerous phases of the oesophagus and develop- percentage of enlarged nuclei to be found in those samples
ment of the Raman system to permit measurements using a with dysplasia. As a diagnostic tool light scattering has an
conventional endoscope. Comparable results have further advantage on the grounds of hardware simplicity, e.g. a low
been reported (albeit with different morphological changes) cost xenon lamp light source can be used rather than the
for cancer of the larynx, colon, stomach, bladder and stabilised laser required for Raman spectroscopy.
prostate [105, 106]. c. Raman: The Raman spectra obtained from oesophageal
Adenocarcinoma of the gastro-intestinal tract is one of epithelium exhibit a series of relatively distinct bands,
the most rapidly increasing forms of cancer in the UK and compared with the two methods discussed above. The
USA. The World Health Organisation has estimated there Raman spectra obtained by Puppels et al. [72] show a
are around 410 000 cases of oesophageal cancer, 87 000 new wealth of information using an Enviva Biomedical Raman
cases of stomach cancer and 940 000 new cases of colorectal Probe [71] (Cirrex Corp, USA). This comprises a single
cancer each year. Regular surveillance of risk populations is central laser delivery fibre of 400 mm core diameter and
therefore carried out to detect early signs of the disease. The seven surrounding collecting fibres of 300 mm diameter. The
critical early stages of morphological changes in the NA of the collection fibres is 0.22 and the rigid end tip of
epithelium of the oesophagus are difficult to detect even the probe is 1.4 mm in diameter and 40 mm long. The
with regular endoscopy, excision of tissue, and subsequent central fibre contains a narrow band pass dielectric filter
analysis in the pathology laboratory. The normal micro- approximately 25 mm from the fibre end-face. This
scopy techniques used on the samples do not provide transmits the laser light but blocks any Raman signal from
unequivocal diagnosis, and sample preparation (histology) the fibre itself. The seven collection fibres contain dielectric
may introduce delays. Equally an endoscopy examination long pass filters, with an OD of 3 at the laser wavelength,
cannot be guaranteed to obtain the critical samples from and a transmittance of Raman scattered radiation of
identified high-risk areas, and so a ‘random selection’ typically 90% above 300 cm
1 Stokes shift from the laser
process adds a further unknown dimension to the attempted line. The laser delivery fibre is optically insulated from the
diagnosis. Speed of diagnosis is crucial to survival rates and collection fibres to eliminate cross-talk. Ultimately this sort
is directly related to the subsequent cost of treatment. of probe should be regarded as a low cost throwaway unit,
The ideal solution is to combine within an endoscope a since there is gradual erosion over time of filter performance
more certain detection technique that requires no tissue due to varying environmental conditions. This causes
removal and real-time diagnosis. Recent developments have problems for extended research use, and entails repeated
reached the stage where reliable statistics for the different calibration procedures.
stages of oesophageal cancer have been obtained from
pathology samples using Raman spectroscopy. Endoscope
design has now also advanced to the stage where Raman The Raman technique gives real-time point (as distinct
fibre optic probes can be coupled to local confocal from larger area) sampling. This is not ideal, and therefore
microscopes at the endoscope tip (e.g. as developed at the some effort is being applied to combine the above methods,
Fraunhofer Institute for Silicon Technology, Itzehoe, or to combine them with other imaging technologies, e.g.
Germany [107]). optical coherence tomography.
The endoscope is an ideal conduit through which to With regard to the detection of the different stages of
perform in-vivo measurements using a range of optical development of dysplasia, the Raman method appears to be
techniques. The most important of these are briefly the most specific of the techniques described above.
reviewed to allow comparison with Raman spectroscopy: Analysis of data for fluorescence and diffuse reflectance
(for example) can be fitted to a 3-group model, i.e. normal,
a. Fluorescence: The spectral features generated by fluores- pre-cancer, cancer, with outcomes of approx. 90–95%
cence spectroscopy are generally broad, which limits the sensitivity and comparable specificity percentages. The data
capacity to discriminate structural changes in tissue. The from Stone et al. [105] for oesophageal cancer have been
UV incident radiation used to excite the fluorescence fitted to a more detailed 8-group model. Figure 62 shows
penetrates only the outer surface of the epithelium, and Raman spectra from selected oesophageal tissues exhibiting
can therefore fail to identify some of the useful diagnostics different stages of cancer development. The results show
in deeper tissue structure. Additional ‘background’ fluor- subtle spectral changes. Assignments can be made for the
escent effects can also confuse the spectral analysis. major vibrational modes in the oesophageal samples, and
However with careful data processing and specialised specific diagnostic algorithms using the line-shapes, posi-
algorithms it has been shown possible to detect squamous tions (wavenumbers), intensities etc., of the Raman bands
cell carcinoma and adenocarcinoma of the oesophagus [108, have been employed. Figure 63 shows plots of the linear
109]. discriminatory function weights for the 8-group model for
b. Light scattering: Reflectance spectroscopy, using a variant each spectrum when tested against the model using a cross-
on Mie scattering theory for analysis of the spectra, has validation process. A 3-group model is shown in Fig. 64.
been successfully tested using a specialised fibre optic probe The published cross-validation results for the 3-group
[110]. This consists of a central white light delivery probe model are: sensitivities between 84 and 97%, and specificity
surrounded by 6 surrounding fibres to collect the reflected between 93 and 99%.
light. The angled fibre tip is shaped to allow the tissue to be Raman is now gaining acceptance as a valid method of
directly contacted (e.g. contact to colon polyps and providing fast diagnostic data for in vitro work and it is
oesophageal tissue). Mie scattering of the illuminating light already of benefit to the research pathologist. For real-time
provides a spectrum related to the proportion of enlarged diagnosis, a combination of Raman techniques with a fast
nuclei and the aggregation of nuclei, i.e. morphological imaging technology would be useful, ideally with both
280 IEE Proc.-Sci. Meas. Technol., Vol. 152, No. 6, November 2005
fitting within an endoscope, although this will have cost group to attempt to classify the risk element in terms of the
implications. The data analysis techniques are clearly critical percentage of larger nucleotides (and hence morphological
to improved diagnosis. The novel approach of the MIT changes) is an interesting route to follow, while the more
discriminating spectral input obtained from Raman data
0.8 has allowed a more standard spectral analysis package [106]
normal
cardiac Barrett's to be employed.
0.7 fundic Barrett's
intestinal Barrett's
low grade Dysplasia
0.6 high grade Dysplasia
9.1.2 Application to skin: The volume of literature
in this area is expanding rapidly and is impossible to cover
intensity/a.u.

0.5 fully here. A number of different optical diagnostic and

imaging techniques have been deployed in the past to study
0.4
skin-related disease. More recent reviews including Raman
0.3 techniques have been published by Stringer (Leeds Uni-
versity, UK) [111] and Choo-Smith et al. [112] Caspers et al.
0.2 adenocarcinoma
squamous Dysplasia
[113] describe a dedicated confocal Raman system for
squamous cell carcinoma in vivo measurements on the concentration of water,
0.1
400 600 800 1000 1200 1400 1600 1800
constituents of natural moisturisers and compounds applied
to the skin. This approach assists clinicians in the field of
Raman shift/cm−1
personal care products.
Fig. 62 Raman spectra from 8 different stages of cancer As with oesophageal cancer described above, a real-time
progression in epithelial tissue [103] in situ analytical technique will help to reduce the need for
Data courtesy of Nick Stone and Hugh Barr of Gloucester Royal biopsy, reduce the time for diagnosis, and also assist in
Hospital, UK delineating the spatial limits of the tumour. Changes in

12 8
normal squamous mucosa
cardiac Barrett's
10 6
fundic Barrett's
intestinal Metaplasia
8 high grade Dysplasia 4
adenocarcinoma
6 squamous Dysplasia 2
squamous cell carcinoma

4 0
LD 2

LD 3

2 −2

0 −4

−2 −6

−4 −8

−6 −10
−6 −1 4 9 −6 −1 4 9 14
LD 1 LD 1

4 8

6
2
4

0 2

0
LD 4

LD 3

−2
−2

−4 −4

−6
−6
−8

−8 −10
−6 −1 4 9 14 −6 −1 4 9 14
LD 1 LD 2

Fig. 63 Statistical grouping for 8 stages of cancer development for oesophageal epithelial tissue [103]
Data courtesy of Nick Stone and Hugh Barr of Gloucester Royal Hospital, UK

IEE Proc.-Sci. Meas. Technol., Vol. 152, No. 6, November 2005 281
 6 Raman measurements on qualified samples using 830 nm
normal squamous mucosa laser excitation and found they could readily classify the
5 Barrett's oesophagus plaque into three groups, i.e. non-atherosclerotic, non-
Dysplasia & cancer calcified plaque, and calcified plaque. They found that the
4
Raman spectra could be modelled as a superposition of
3 spectra of cholesterol, esterified cholesterol, calcium hydro-
xy-apatite, b-carotene, triglycerides and de-lipidised artery.
2 The processing algorithms were sufficient to give quantita-
tive information on the relative fractions of these compo-
LD 2

1
nents to within a few percent. Subsequent work has added
0
to information about the changes in chemical composition
as the disease advances.
−1 Puppels and co-workers [114] have carried out in vivo
Raman measurements on atherosclerotic plaques. To do
−2 this they used Envivat probes, with both end-on and side
viewing. After catheterisation the probe was guided to the
−3
target points using fluoroscopy. The end-on probe collected
−4
spectra from the outer and inner aortic arch, while the side
−6 −4 −2 0 2 4 6 viewing probe took Raman spectra from the femoral artery,
LD 1
carotid artery, iliac artery and the inner arch of the aorta. A
typical measurement took 10 s with the fibre optic probe
Fig. 64 Statistical grouping for a 3-stage model for the develop- connected to a Renishaw spectrometer. The results were
ment of oesophageal cancer [103] consistent with those of Deinum et al. [113] above, showing
Data courtesy of Nick Stone and Hugh Barr of Gloucester Royal that in vivo identification of lipids, cholesterol, and
Hospital, UK
calcification is indeed possible.
Certain improvements are needed in the robustness of the
fibre optic probes. However, the small stable 830–850 nm
biological markers found in proteins, lipids and nucleic diode laser has improved spectral definition relative to
acids can indicate the onset of neoplasia. One advantage of background fluorescence effects, and increased penetration
investigating skin using optical techniques is that the probe of this excitation wavelength below the vessel wall provides
or microscope can be presented directly to the patient, e.g. further information on the state of the deposits. It is
using the inverted microscope configuration [72]. While estimated that with further spectrometer optimisation,
Raman spectroscopy has been shown to be excellent in signal collection times below 1 s should be feasible.
identifying the specific stage of cancer development, the
limited spatial coverage is a disadvantage. Work is therefore
now proceeding to integrate both Raman and imaging 9.1.4 UV Raman and DNA bases; analysis of
technologies. Recent advances in the field of terahertz nucleotides and aromatic amino acids: New
reflection data (terahertz pulsed imaging: TPI) appear to low power UV lasers have become available for use with the
overcome problems in sample penetration and limitations in high throughput Raman spectrometers. The frequency
the field of view for laser confocal microscopy and optical tripled Nd-YAG laser line at 244 nm has been used to
coherence tomography [108]. Both TPI and Raman can be obtain spectra of DNA bases. Figure 65 shows several
designed to fit within the same optical configuration. Raman spectra of DNA (Sands [118]). Since the excitation
In addition to diagnosing skin cancer, Raman spectro- line is sufficiently close to the first absorption band of the
scopy has been used to quantify photosensitiser in blood nucleic acids and nucleotides, the spectra are resonantly
samples [111]. The key advantage here is that little sample enhanced, with excellent signal-to-noise and a flat back-
preparation is required, eliminating laborious biochemical ground. Low fluorescence was observed. The use of
extraction procedures and the potential for human error. resonance Raman and low power incident radiation (e.g.
The group at Leeds University [114, 115] has used tetra- 1 mW) to achieve enhanced signals has been discussed in
sulphonated zinc phthalocyanine, one of several compounds Section 2.3.
now used in photodynamic therapy (PDT). A high Earlier UV Raman work on similar samples, with
correlation (99%) has been shown between the predicted traditional multigrating spectrometers and high power
and actual concentration based on measurements on the lasers, was carried out by Asher et al. [32, 119, 120]. This
747 cm
1 ts-ZnPc Raman line. Exposure times were pioneering work has provided important pointers to the
typically 30–90 s, and background fluorescence spectra were more practical applications that have followed. More
subtracted. recently Boustany et al. [120] have used a 251 nm excitation
line to assess the potential of UV resonance Raman for the
9.1.3 Coronary artery atherosclerosis: With spatial analysis of nucleotide distribution and amino acid-
the rising number of patients suffering from heart disease, to-nucleotide ratio in normal and neoplastic colon tissue.
increased attention has focussed on the ‘plaque’ deposits The advantage of the UV excitation in this case was that
attached to the artery walls. The state of the plaque is only the surface neoplastic changes in the epithelial surface
related to the cholesterol content of the blood, lipid of the colon mucosa were examined and not the deeply
deposits, and smaller amounts of collagen and glycosami- invasive parts of the tumour. The changes observed in the
noglycans. The plaque composition is important as it relates spectra result, in part, from changes in DNA metabolisms,
to the possibility that the lesion will rupture, i.e. when it and DNA content in the tumours. Analysis of the spectra
becomes more brittle. The stenosis (constriction) can then has again required specialised algorithms to obtain reliable
grow rapidly leading to a significant heart attack. There is a assessments of the ratios. The likelihood of this Raman
need to obtain information on the chemical composition of technique becoming more than a research tool is limited by
lesions in vivo. Deinum et al. [116] (MIT) carried out in vitro the cost of the UV laser, the added safety requirements, and
282 IEE Proc.-Sci. Meas. Technol., Vol. 152, No. 6, November 2005
 adenine to their battery of analytical tools. Several review papers on
cytosine this topic have been published by a group led by Clark at
guanine
thymine
University College, London, UK [123, 124]. Clark has
emphasised the fact that the increased usefulness of Raman
intensity/arb.units

has come about as a result of the new developments in

instrumentation design. From the perspective of art
restoration, it is important to identify the composition of
an original pigment since application of a different
composition, even though it may superficially appear
correct, may lead to ageing reactions and subsequent
cosmetic deterioration. The restorer can thus select an
400 600 800 1000 1200 1400 1600 1800 appropriate remedy to halt any ongoing degradation. The
Raman shift /cm−1 degradation products themselves can also be identified.
The high spatial resolution (o1 mm) offered by Raman
adenine
microscopy is a major benefit compared with other
cytosine techniques, particularly as it also gives molecular composi-
guanine tion rather than elemental data. Portable Raman units and
intensity/arb.units

thymine
fibre optic probes can be employed and the probe
manipulated over the painting/manuscript surface with
video viewer if required. This flexibility is particularly
relevant if the artwork is fragile or is situated in a remote
location or is too large to move. It is often preferable to take
the Raman instrument to the object to eliminate handling
and reduce the possibility of ‘disasters’ during transport to
400 600 800 1000 1200 1400 1600 1800 2000 2200
the laboratory. Figure 68 shows a 3-D motorised tripod
Raman shift /cm−1 mount for a fibre optic probe being used to examine a
Fig. 65 UV excitation (244 nm) Raman spectra of several DNA religious artifact. There is increasing use of the stabilised
bases near-IR lasers as the excitation source for this work because
The spectra (bottom) are the result of resonance enhancement and of their improved portability, and their enhanced ability to
exhibit no fluorescence [115], i.e. flat baseline reduce background fluorescence.
The spectra from visible excitation (top) are given for comparison As part of their programme Clark’s group [125] compiled
libraries of pigment materials spectra. This has proved to be
complementary to the Renishaw minerals and inorganic
materials libraries available on the present Raman system.
its unacceptability for in vivo work owing to the carcino- The measurement of specific pigments can lead to
genic effect of the UV radiation. detection of fakes. For example certain Egyptian papyri
purporting to date from the 13th to 1st centuries BC were
9.1.5 Raman measurements inside live cells: shown to be illuminated with mineral pigments which did
It has been important to prove that measurements on live not exist prior to the 19th century, e.g. (critical dates in
cells can be carried out non-invasively and with no damage brackets) Phthalocyanine blue (1935); ultramarine blue, the
now that low power lasers can be used. The initial test synthetic form of lazurite (1828); and synthetic anatase,
involved the use of direct global (‘true’) imaging to examine which is a form of titanium oxide rarely found in nature
single bovine ovary cells within two hours of removal from (post 1923) [124]. There were also other indications that the
the carcass [122]. Figure 66 shows white light images (left) of papyri were forgeries based on the way the pigments had
tissue samples, where the cells were taken from the cow’s been laid down.
ovary. A confocal Raman spectrum from the inside of a A more recent and possibly more controversial measure-
cell, shown in Fig. 67, indicates the presence of b-carotene, ment has been carried out in collaboration between
which is a strong Raman scatterer. The exposure for the Renishaw, Yale Univ. USA, and University College,
spectrum took 5 s. The most intense band at 1525 cm
1 London, where pigments from the Vinland Map (VM)
comes from the C ¼ C stretching mode of the b-carotene. and Tartar Relation (TR) have been examined by Brown
The beam was then expanded and the 1525 cm
1 band and Clark [125]. The map was first presented for validation
was selected to pass through the global imaging band-pass in 1957. It is a world map showing Iceland, Greenland
filter. The b-carotene areas are highlighted in Fig. 66 (right) and the Americas and had been bound into the Tartar
where the Raman images of corpus luteum cells of different Relation, detailing the expedition of Friar John de Plano
ages (approximately 12 mm diameter) are shown. Both Carpini to the Mongols in 1245–47 (dated to the 15th
images have flat-field corrections applied. It is seen that the century) with a modern binding. The purported provenance
concentration of b-carotene within the older cell is more indicated that the Vikings had reached America before
intense, while for the younger cell (taken at the beginning of Columbus.
the cycle) there is a more uniform spread. Each direct global Early tests on the map used polarised light microscopy,
image took only 20 s to obtain. This result has shown that X-ray diffraction, and SEM/energy-dispersive X-ray analy-
significant work can be carried out within cells to track sis by McCrone Associates Inc. [126, 127]. They found the
abnormal development. presence of synthetic anatase in the yellow lines of the map.
Transmission electron microscopy measurements found the
9.2 Pigments, old masters, art fraud crystals to be relatively rounded and with a narrow particle
There is increasing interest in identifying pigments on size distribution characteristic of synthetic calcined anatase,
paintings, manuscripts, ceramics and papyri. This is useful [124], a material not available prior to 1923. Nevertheless
for improved dating, conservation, authentication and other measurements still left certain aspects of the map’s
restoration. Numerous art museums have added Raman provenance open to discussion [128].

IEE Proc.-Sci. Meas. Technol., Vol. 152, No. 6, November 2005 283
 old cell

10 µm

young cell

white light Raman+ background

(1522 cm−1)

Fig. 66 White light (left) and Raman images (right) for corpus luteum cells taken from a cow’s ovary at different stages of the cycle
The Raman images compare the intensity of the 1522 cm
1 band with the background subtracted by setting the bandpass filter away from the
line. The top line refers to an ‘old’ cell showing a high concentration of b-carotene within the cell

Areas examined comprised the regions of black ink that

were identified as carbon from the characteristic Raman
1522

10 000
spectrum, and yellow fluorescent lines from which Raman
1156

8000 bands belonging to anatase were evident. It is conjectured

intensity

that the artist first drew yellow lines, and later traced over
6000 them with carbon-black ink. Figure 70 shows a close-up of
a map detail of black ink with yellow ‘under-paint’. The
4000
Raman results reinforced the conclusion that the map was a
forgery.
2000
A final example has involved collaboration between the
1000 1200 1400 1600 Univ. of Florence (Department di Ottica) and Renishaw
Raman shift/cm−1 [129]. The oldest signed masterpiece is an 11th century
wooden cross with a painting of Christ. A number of paint
Fig. 67 The Raman spectrum for b-carotene from within the flakes were removed for non-destructive testing under the
corpus luteum cell taken from a cow’s ovary prepared by enzymic
Raman microscope situated in the Firenze Art Restoration
digestion
The 1522 cm
1 band has then been used for direct real 2-D imaging of
Centre. Figure 71 shows the Raman spectrum from a paint
the distribution of b-carotene within cells, taken at different points of flake compared with a spectrum for lapus lazuli. The
the ovarian cycle pigments used in the masterpiece (now very faded after
Background flourescence can be subtracted using the image taken at years of hanging in a church exposed to the environmental
1609 cm
1 ambient) were identified. The art preservation group at
Firenze were then able to recreate painted sections of the
The Beinecke Rare Book and Manuscript library made original, in particular the border decoration.
the Vinland Map available for analysis in Oct./Nov. 2000.
The map is faded, but the parchment is in a clean condition. 9.3 Gems, minerals and diamond
It was decided that owing to the rarity of the object it
should be presented to a tripod-mounted fibre-optic Raman 9.3.1 Overview and gemological laborator-
probe, with integral video, mounted to the side to avoid ies: Most of the major national gemological and miner-
accidental damage. This configuration allowed for easy alogical research laboratories throughout the world now
positional adjustment and the critical areas could also be employ modern Raman microscopes for identification and
viewed on the remote screen. Figure 69 shows the certification of gems and minerals. This has come about
measurement set-up and the map. The probe was linked from the growing realisation that Raman is a rapid, high
to the Renishaw RA 100 portable spectrometer. spatial resolution, non-destructive method of measurement
284 IEE Proc.-Sci. Meas. Technol., Vol. 152, No. 6, November 2005
Fig. 68 Photograph of the examination of a wall-mounted Fig. 69 Photograph of the set-up to measure the Vinland Map at
reliquary using a portable fibre-optic Raman probe Yale University
A similar mounting arrangement is used for examination of paintings The map is placed in front of a fibre-optic probe which also has a
video camera fitted. We gratefully thank the Beinecke Rare Book and
Manuscript Library of Yale University for allowing us to photograph
the Vinland Map
requiring no sample preparation. Typically the spectrum of
a gemstone can often take less than one second to record.
The chief measurement decisions to be made are simply the
methods of mounting the sample, and whether to use
polarised light. The reduced cost of the instrumentation and
the recent establishment of the spectral library were also
significant factors. Useful reviews on the increased use of
Raman spectroscopy in gemology have been published by
the group led by H.anni (SSEF, Swiss Gemological
Institute) [130] and Bloomfield [131].
In order to determine provenance and establish the
genuine nature of gems on ancient encrusted reliquaries, in
situ tests have been conducted using the hand-held Raman
fibre optic probe. Figure 71 is a photograph of a gem-
encrusted reliquary being examined using a fibre optic
Raman probe. An investigation by H.anni et al. [132] on two
historical objects from Basel Cathedral (The Reliquary
Cross and Dorothy Monstrance), however, called for the Fig. 70 Detail of a region of the Vinland Map showing black ink
removal of the complete artefact from its location. Each with yellow ‘under-paint’
stone was aligned separately under the microscope. Many The yellow regions contain spectra that correspond to the presence of
of the mounted gems were found to be varieties of quartz, anatase, a synthetic form of TiO2 that was not in use before 1923
either as polished single pieces or as doublets. Evidence of
dried cement was also found, as well as coloured glass,
peridot, sapphire, garnet, spinel and turquoise.
The development of the first commercial and compu- XRD confirmation. The library continues to expand. A
terised gemstone and minerals spectral library for dispersive typical library search and screen display showing the
Raman was based on reference spectra from original probability index for identification has been shown in
samples supplied from the Nelson [133] collection. All Fig. 56. The printout also contains data on chemical
samples have comprehensive descriptions, traceability and composition.

IEE Proc.-Sci. Meas. Technol., Vol. 152, No. 6, November 2005 285
 (often from an identified inclusion) when issuing quality
certificates of faceted stones. As far as is known this is not
possible with any other technique.

9.3.2 Zircons: A series of tests has been carried out on

several zircon crystals supplied by Nelson [133]. Figure 72
shows a series of spectra from these samples. The zircons
were obtained from Sri Lanka and Burma. Zircon (ZrSiO4)

500 1000 1500

Raman shift /cm−1

Fig. 71 Spectrum of a small flake of paint taken from the earliest

signed masterpiece, an 11th century wooden cross with a painting of
Christ by Maestro Gugliemo
A high degree of fluorescence is observed. This can be background
subtracted to emphasise the peaks, but the raw data are sufficient to
obtain a match with the standard pigment lapus-lazuli from the stored
library data (lower spectrum). Investigations courtesy of D. Bertani,
Dept. di Ottica, University of Firenze, Italy 200 400 600 800 1000 1200
Raman shift /cm−1
a
Some published examples taken with the new system by
gemstone laboratories include:
(a) John Koivula of the GIA Gem Trade Laboratory,
Carlsbad, California found some inclusions of a single
intense green mineral in samples of pyrope from Amarillo,
Texas [134]. No such inclusions had been found hitherto.
The confocal facility of the Raman microscope was used to
obtain the Raman spectrum from the inclusion, and the
library immediately identified it as diopside. Subsequently,
the sample was heated to produce stress to cause a fracture,
and the inclusion was exposed at the surface. Further
analysis confirmed the initial identification. Generally it is
noted that the evaluation of inclusions in a host material
is economically highly significant, and this type of datum is
used in exploration for gas, oil, other mineral resources and
200 400 600 800 1000 1200
gem-bearing host materials. Global (‘true’) Raman imaging
−1
techniques on samples are also now being used to determine Raman shift /cm
the extent of disseminated pyrite deposits in quartz veins, b
for example.
(b) Two further examples from the GIA:
– Inclusions of zircon were found in jadeite for the first time
(1998) in coincidence with the discovery of a zircon
inclusion in a jadeite cabochon at the Centre for Gemstone
Testing in Bangkok [135].
– An 8 ct ‘ideocrase’ from alluvial gem deposits in Tanzania
was examined (1998). The stone showed some refractive
index and birefringent properties that could be interpreted
as ideocrase. However the Raman spectrum caused surprise
when it matched the reference library spectrum for
sapphirine. This possibility had not been considered because
the size was almost three times that of any chrome bearing
sapphire or one of this colour. XRD confirmed the
identification. However it would not have been possible to 200 400 600 800 1000 1200
identify the calcium-magnesium aluminosilicate by XRD Raman shift /cm−1
alone. It was concluded that this was the largest sapphirine c
reported to date and the first to have come from Tunduru,
Fig. 72 Fluorescence spectra of three zircon crystals
Tanzania [136]. Band half-widths are shown that vary with the structural ageing
The results illustrate how use of the Raman technique is history of the gem owing to the presence of U4+ ions. See text for
now able to help in attributing the origin of a mined gem description. Results courtesy of Jamie Nelson [130]

286 IEE Proc.-Sci. Meas. Technol., Vol. 152, No. 6, November 2005
can retain uranium ions, up to about 1.0 weight % U3O8, 9.3.3 Red spinels: Spinels normally display identical
where it is presumed to replace Zr4+ ions. When present, and well-structured, narrow fluorescence bands. This is to
U4+ undergoes spontaneous decay emitting high-energy be expected for natural crystals, since the longer formation
alpha particles, which can cause intense local damage and time allows for better crystalline ordering. Synthetic
displace a considerable fraction of the ions from their Vermeuil (flame-fusion process) and flux-grown spinels
normal positions (metamict). Calculations can be made on show similar spectra, but with wider line-widths (Fig. 73).
the assumption that 160 MeV of kinetic energy is liberated The natural spinel shows several narrow satellite bands
in each fission event of U235 and U238. Since this internal attached to the five main bands. They are not visible in the
bombardment has been continuing for approximately 700 synthetic material.
million years, it is hardly surprising that the original
tetragonal structure has become distorted. A younger stone 9.3.4 Treatment of gemstones by filling: The
from Late Cretaceous rocks in Burma (age about 70 million rapid detection of deliberately filled gem defects, using
years) shows little sign of damage, and the band-widths disguised refractive index matching oils and resins, is of
compare well with those of zircons with only a few U4+ major concern to the gem trade. Defects should be disclosed
ions in their structure. when the stones are traded, but unfortunately this disclosure
Three faceted stones were examined, all having approxi- rule is not always followed. Their treatment is now so
mately the same U4+ content. The normal density Burmese sophisticated that standard gemology microscopy techni-
(young) zircon shows the usual well-defined lines, but the ques are inadequate. Raman has now become a standard
low-density disordered Sri Lanka zircon has wider bands. screening method to identify those gems that have been
The third zircon is an ‘A’-type anomalous Sri Lanka gem treated in order to increase their value. Raman spectroscopy
which has a strong, and rarely seen, spectrum. All the is often able to fingerprint the exact oil or resin, allowing
results can be correlated to their geothermal history, and not only its presence, but also its origin to be determined.
correspond to the location of the deposit. The high spatial resolution (o1 mm) of the Raman
microscope is ideal for this application and the Raman
signals are so high that the global imaging facility is often
used to obtain a 2-D picture of the filled defect. The
spectrometer is also used in the Far East to detect imitations
of amber, e.g. Tay et al. [137] (Far East Gemological
Laboratory, Singapore), and pearl source conditions (e.g.
sea water or fresh water cultured), Zhang [138], Yunnan
Technical Univ., Kunming, China.

9.3.5 Inclusions: The minerals spectral library also

includes liquids and gases commonly found in inclusions.
Reference spectra have been supplied by the Wuhan
University Minerals Laboratory [139], following Raman
measurements carried out in China, using standardised
procedures. The interest here extends to oil and gas
exploration. Figure 74 shows a white light photograph of
inclusions in a quartz crystal. This was part of a study to
examine the geothermal history of a series of quartz
samples. Inside the inclusions, gas and liquid phases can be
clearly seen. The spatial resolution of the confocal technique
is such that each phase can be detected separately. As a
result, the crystalline formation history under varying T/P
conditions has been interpreted.
Izraeli et al. [140] have used the presence of olivine
inclusions in natural diamond (from the Udachnaya mine in
Siberia) to determine the Raman barometry of diamond
formation. Confocal Raman measurements were made on a
specific olivine inclusion, and also by scanning the diamond
band at 1332 cm
1 in the region of the inclusion. The
internal compressive stress in the olivine was calculated
by a measure of the olivine line shift relative to the
atmospheric pressure line position at 856 cm
1. Similarly
the internal stress within the diamond itself was plotted as a
function of the distance from the inclusion interface. The
result was fitted to a linear dependence of the internal
pressure on the source temperature. It has been proposed
that this method could present an alternative type of
geobarometer. By combining these data with temperature
estimates based on the nitrogen concentration aggregation
state from two of the diamonds, it was possible to obtain
the source pressure for the Udachnaya diamonds as
4.4–5.2 GPa at 12001C.
Fig. 73 Fluorescence spectra of natural and synthetic red spinels
The synthetic spinel has wider bands. Results courtesy of Jamie Nelson 9.3.6 Treated diamonds: In 1999 Lazare Kaplan
[130] International (LKI) and General Electric announced a new

IEE Proc.-Sci. Meas. Technol., Vol. 152, No. 6, November 2005 287
 CO 2
N2 CH 4

1500 2000 2500 3000 3500 1500 2000 2500 3000 3500
Raman shift, cm−1 Raman shift, cm−1

quartz H 2O

1500 2000 2500 3000 3500

Raman shift, cm−1

Fig. 74 Spectra showing inclusions in quartz

Inclusions in quartz may contain solid, liquid and gas phases. Spectra have been taken from the different regions showing evidence of H2O, N2, CO2
and CH4. This is an excellent example of the use of confocality in Raman measurement. Such measurements are used for analysis of rock bearing
minerals in oil and gas exploration

process for irreversible improvement in the colour and SSEF SWISS GEMMOLOGICAL INSTITUTE
brightness (i.e. brown to colourless; a result of the Renishaw Raman Systems, luminescence spectrum
destruction of vacancies and interstitials) of 1% of the 5000
[N-V]-
diamonds on the market. These are known as Pegasus 637 nm
[N-V]˚
diamonds after the LKI subsidiary Pegasus Overseas Ltd. 4000 575 nm
(POL). They have been sold with appropriate markings i.e. liquid nitrogen temperature
counts

GE POL, but several unscrupulous dealers have subse- 3000

quently removed them. This could have a serious effect on 2000
the diamond market as a whole, particularly when there are
further announcements from other manufacturers of new 1000 575 nm 637 nm
‘gem quality’ synthetics. Some publications have indicated room temperature
that there is no way of detecting the difference between 0
1000 2000 3000 4000 5000
these treated/synthetic diamonds and natural diamonds.
Raman shift, cm−1
H.anni’s group at SSEF [141, 143] carried out a series of
Raman microscope tests at low temperatures (LN2) on both Fig. 75 Raman/PL measurements on diamond and a high
types of gems and found differences in the spectra. The pressure/high temperature (HPHT) treated GE-POL diamond
relative intensities of fluorescent bands at 575 nm and The treated diamond shows the reversal in intensity of the 575 nm and
637 nm, due to nitrogen-vacancy defects, were reversed for 637 nm bands. Data courtesy of H.A. H.anni of the SSEF
treated and untreated brown diamond: see Figs. 75a and Gemmological Institute, Switzerland
75b.
Smith et al. [144] of the Gubelin Gem Lab in Switzerland
have also identified weak UV excited photoluminescence response. This further confirms the approach of the present
bands that exhibit differences in intensities from treatment group in combining Raman-PL with SEM-CL techniques
due to effects on H3, H4 and N3 point defects. It has also (described in Section 6.2). However the cost of a UV laser
been found that this new UV-PL technique is a more and the combination with an SEM will be beyond the reach
sensitive measure of the presence of nitrogen impurities than of most smaller gemstone evaluation facilities. There are
traditional methods, i.e. IR spectroscopy. This exercise has treated diamonds from other sources coming on to the
proved to be an excellent demonstration of the efficient use market with fewer controls than those offered by the GE-
of the synchronous scan facility in the spectrometer to cover LKI route.
the full spectrum range required. Consequently there remains further work in this area to
The Gubelin Lab’s work also extended to the examina- be done on simplifying the technology and reducing its cost.
tion of CL measurements on diamonds before and after the Measurements are required on a greater number of samples,
high pressure high temperature (HPHT) treatment by GE, to obtain a more confident statistical base, than those
and found further significant differences in spectral carried out in the publications to date, to confirm the
288 IEE Proc.-Sci. Meas. Technol., Vol. 152, No. 6, November 2005
validity of the use of focussing on a particular spectral
feature to determine the provenance of a questioned
diamond.
It is noted that the Raman spectrum of synthetic
moissanite (silicon carbide), which can be cut and mounted
as a substitute for diamond, is radically different from that
of diamond. Moissanite is also being marketed as a
substitute for diamond through reputable and less reputable
sources, and there are concerns in the gem trade that
batches of diamonds might be ‘salted’ with moissanites. In
this case the differences in the visible laser excited Raman
spectra give a helpfully clear and definitive answer on the
crystalline structure of the gem.

9.3.7 Detection of primordial material: In a

January 2004 the NASA Stardust spacecraft passed within
147 miles of comet Wild 2’s nucleus [145]. As it passed, a
tray of aerogel particle capture cells picked up many
thousands of small particles. The capsule with the samples
will return to Earth in January 2006 [146]. It is hoped that
the particles will provide an insight into the origin of the
comet and the formation of the solar system. The capture
efficiency of the aerogel grid was evaluated by using capture
cells to the Mir space platform, and bringing back
primordial dust particles. The aerogel material itself
comprises a silica based porous material that is 1000  less
dense than glass. The particles entering the cell are gradually
brought to a halt, leaving a clear entry path (rather like a
bubble chamber picture), with the particle lying at the end
of the track. Raman was used to track the polymerisation of
orthosilicate sol gel processes, [147], but it was also
necessary to probe the areogel panel non-destructively and b
identify the embedded particles (typically 1 mm or more in
size). Fig. 76 Pictorial description of confocal Raman measurements on
trapped primordial material within aerogel panels
Figure 76 gives a combined picture story of the collection The Raman library was used to identify this particle as peridot
and confocal Raman detection process. Samples from Mir (below). The particles may be as small as o1 mm. Results courtesy of
were successfully measured in collaboration with Peter Tsou Peter Tsou, JPL, NASA, Pasadena, CA, USA
[148] (NASA JPL, Pasadena, CA, USA, 1997), and then
compared with the Renishaw minerals spectral library. It is
seen that peridot (the gem name for olivine) was found in
this sample, together with other minerals and carbon. nescence effects from disordered diamond and graphitic
Olivine minerals are the chief constituents of the earth’s regions in deposited diamond films (Fig. 4). This was an
upper mantle. The carbon spectra exhibited the character- important result, and was instrumental in progressing the
istic double Raman bands near 1560 cm
1 and 1330 cm
1 design of the new Raman systems. Around 1990 the
(see also Section 9.4), leading to speculation that perhaps development of diamond films by plasma deposition (as
nanocrystalline diamond might be found. Some organic distinct from the traditional high pressure and high
material was also found, but whether this came from earth temperature methods) became a focus for many industrial
bound contamination was not resolved. The Raman system groups. The films could be used for laser optic windows,
installed at NASA in 1999 has also included the UV bearing surfaces, heat sinks for chips and components in
excitation option in order to give signal enhancement and electronic circuits, and potentially high temperature electro-
possibly detect signs of life (i.e. organics). A paper on the nic devices. The quality control procedures in force at that
simulation of Raman detection of particulates in aerogel time to follow the growth of a film tended to be lengthy and
panels has been given by Burchell et al. [149]. laborious, involving both X-ray and Raman (using the
traditional multi-grating systems). The results took several
9.4 Different forms of carbon and their uses days to obtain, and changes in the processing as a result of
Figure 77 shows four typical spectra of carbon, including this feedback were therefore very slow. With the advent of
diamond. The clear Raman spectral differences that exist the high efficiency Raman systems, the result could be fed
indicate how useful the Raman technique can be in the back to the processing engineers within seconds. The new
detection of the carbon form for both fundamental low powered lasers operating in the near infrared that
investigations and industrial uses. The SEM, for example, became available, and well-suited to CCD detectors, also
will indicate only that elemental carbon is present in a gave a better insight into the background non-diamond
sample, while Raman can distinguish the different forms of content of the film, since it was found that with increased
carbon and also map and image them. wavelength (e.g. moving from 514 nm to 633 nm and
further) the non-diamond background signal increased
9.4.1 Industrial diamond films: In Section 2.3 we [150]. Apocryphal stories subsequently emerged of certain
used the spectra of diamond films to illustrate the difficulty groups who ignored the new Raman technology and
in distinguishing the diamond 1332 cm
1 Raman band continued to believe they had high quality pure diamond
from the background fluorescence and broad photolumi- films, with a low non-diamond background. They had

IEE Proc.-Sci. Meas. Technol., Vol. 152, No. 6, November 2005 289
 carbon fibre diamond content with the highest sensitivity. As a result the
quality of their diamond films improved rapidly over a two

1585.54
to three year period and the growth processing was
1000 transferred to production. Raman systems have subse-
quently been incorporated into manufacturers’ standard
Intensity

QA procedures.

1359.41
500
Figure 78a is a photograph of a single platelet within a
diamond film. The theoretical growth profile of a single
crystal below the surface is calculated to be an inverted
triangle (b). As a research project on diamond growth,
0 confocal Raman measurements were made in 3-D. The
500 1000 1500
Raman line half-width (indicative of structural disorder),
Raman shift /cm−1 wavenumber position (indicative of strain) and intensity
graphite
were mapped. The result shown in Fig. 78c is of the 3-D
line-width plot that gives a very good approximation to the

1589.25
5000 proposed subsurface triangular growth pattern for well-
ordered diamond [151]. Although this measurement cannot
4000 1362.19 be classed in the quality assurance category, it illustrates the
importance of providing a highly stable system that is able
Intensity

3000
to perform lengthy multi-point measurements over extended
2000 periods. Data collection in this case took 2 days. 3-D
subsurface strain measurements and profiles were also
1000
obtained [152].
0
500 1000 1500 9.4.2 Diamond-like-carbon coatings (DLC),
Raman shift /cm−1
and computer hard disk and read head
quality control: DLC coatings are now routinely used
C60 Buckminsterfullerene on hard disks for computers and as passive corrosion
10 000 resistant coatings of magnetic media. They are used on both
1464.63

the disk and the read-head to provide high wear resistance,

8000
hardness and low friction. As the requirement for the
6000
information density on the disk has increased with time (at
Intensity

present 100 Gbits in
2 has been demonstrated), the thick-

4000 ness of the deposited coatings has continually decreased, to
1423.78

1568.12

the extent that it is now below 5 nm (and will reduce in the

492.297
268.732

2000 shorter term to 1–2 nm; this is only approximately 7 atomic

layers thick). The ride height of the read head over the disk
0 has also been reduced. Even at this small film thickness
500 1000 1500 there is sufficient Raman signal, and it is possible to
Raman shift /cm−1 measure the bonding strength of the film. More extended
reviews on this topic have been published by Lauer [153]
diamond
20 000 and Whitley [154], in which the bulk of the data comes from
collaborative work with Renishaw.
1331.74

Figure 79 shows a typical spectrum obtained from a

15 000
DLC film of approximately 10 nm thickness. Two broad
bands are seen, near 1360 cm–1 and 1560 cm–1. These
Intensity

10 000 correspond to the diamond-like (D) and graphite-like (G)

contributions to the bonding respectively. Theoretical
5000 analyses related to physical hardness and wear criteria
allow us to interpret the spectrum in terms of band
intensities, positions and band half-widths, and then relate
0
1000 1200 1400 1600
them to wear. Comprehensive and systematic measurements
of these parameters led to a series of simple plotted
Raman shift /cm−1
relationships for films of varying thickness and process
Fig. 77 Four spectra of different polymorphs of carbon parameters, to provide optimisation of the deposition
Note the characteristic double peaks near 1360 cm
1 and 1590 cm
1 processes. Measurements were carried out in-house by the
for carbon fibre and graphite. These are also found in diamond-like several read-head and disk manufacturers using standard
carbon although the bands are broader. In freshly cleaved graphite, the accelerated tribological wear and friction (stop-start) life
1360 cm
1 band tends to be missing. The spectra for Buckminsterful- testing techniques on production samples, to eventually
lerenes and carbon nanotubes give characteristically narrow bands
correlate Raman spectral features (engineering diagrams) to
wear resistance. Although the theoretical relationships
linking wear resistance to bond strength may be relatively
stayed with the traditional diamond Raman measurement, complex, the engineering optimisation was made on the
which meant the use of the higher-powered Ar+ laser basis of practical empirical testing. The film should have
operating at 514 nm, resulting in less sensitivity to film appropriate hardness, low friction and high elasticity. Other
imperfections. Most astute groups quickly took up the use results to be obtained from analysis of the spectrum were
of the new Raman system design and extended the laser the thickness of the film itself (from the intensity values),
tests to use 785 nm laser diodes in order to quantify the non- and nitrogen bonding and concentration from analysis of
290 IEE Proc.-Sci. Meas. Technol., Vol. 152, No. 6, November 2005
 1 µm
a ~150

µm

[001]
[ T11] [1T1]

[110]

D
~ 60 µm

C ratio of areas of graphitic to diamond peaks

B >1 0
A substrate low quality diamond high quality diamond
b c

Fig. 78
a Scanning electron micrograph of the surface of a diamond film grown at NRL by Jim Butler. The large square faceted diamond crystal (centre) has
been used for this test. The film was thinned from the front and back leaving a square defect-free region associated with the (001) growth sector
surrounded by defective (111) growth zones
b A schematic representation of the growth regions for such a facet which was thinned to the level shown below the original surface
c Raman 3-D map of the FWHM (full width half maximum) of the first order diamond Raman band at 1332 cm
1, and extending below the surface
of the section shown. Dark regions correspond to high quality diamond. The result shows that the predicted growth profile is correct

example, Bing et al. [155] established that the intensity of the

CN band near 2200 cm
1 could be related to the percent
nitrogen concentration and hence gave a method for
monitoring the nitrogen concentration in the sputtering
D band
gas. Several other relationships were determined, e.g. G-
band position versus nitrogen concentration by other
G band groups. Gopinathan et al. of ReadRite Corporation, USA
[156] have also provided data showing a linear relation
between G-band intensity and film thickness between 2 and
ID 13 nm. Useful engineering relationships covering critical
spectral parameters have been built up by the industry to
IG assist in monitoring the deposition processes (Table 8).
CN band
As the film thickness approaches 1 nm, both a-CN and a-
CH films exhibit pin-holes and the protection against
corrosion and wear is degraded. It is possible that
hydrogen-free tetrahedral amorphous carbon (ta-C) will
give the necessary inertness and smoothness. The sliding
1000 1200 1400 1600 1800 2000 2200
mechanism of the read-head over the disk causes numerous
Raman shift /cm−1 tribological interactions with significant morphological
changes. It has also proved useful to analyse these changes
Fig. 79 Raman spectrum of a diamond-like carbon (DLC) film
The characteristic D and G bands are evident, and the data can be
together with the wear debris following testing under
curve-fitted to give the useful engineering design data indicated in varying environmental conditions.
Table 8 Renishaw Raman systems are now used by most hard
disk and read-head manufacturers world-wide as their
standard quality assurance tool, and for batch analysis. The
the CN band, which is situated near 2200 cm
1. The ‘hot measurement time of Raman has proved to be much faster
topic’ in the industry around 1995 was the use of (e.g. 20 s) in terms of supplying the sputtering deposition
amorphous nitrogenated carbon films (a-C:N), since they process engineers with feedback (e.g. to optimise the film
appeared to have chemical and wear properties better than thickness and annealing processes) than the earlier alter-
those of hydrogenated carbon, including an enhanced natives of X-ray and the traditional contact start-stop
attraction to perfluorinated lubricants which are used to technique CSS. In the latter method the disk was placed in a
finally coat (1–2 monolayers) the bearing surfaces. For disk drive simulation rig and accelerated to flight conditions

IEE Proc.-Sci. Meas. Technol., Vol. 152, No. 6, November 2005 291
Table 8: Spectral correlations and engineering information (determined from the Raman spectrum, shown in Fig. 79)

Property of film to be measured Spectral parameter Relationship

Wear rate G position complex

Wear rate G width complex
Wear rate D/G ratio complex
Thickness G intensity relative to the base line approx. linear
Thickness G intensity approx. linear
Sputtering source health check G position when it has more than about 5 cm
1 change complex
Hydrogen content linear base line slope complex
Nitrogen content nitrogen band intensity/area at approximately 2200 cm
1, and a complex
further band near 600 cm
1

and then decelerated to rest. This procedure would be the first to show that UV Raman excitation (244 nm) could
repeated numerous times to correlate with common usage provide a means of separating the sp3 and sp2 bonding
in practice, e.g. humidity variations. The effect on wear contributions in the films. This was after comparative
resistance, friction etc. was monitored with tribological electron energy loss spectroscopy analysis (EELS) on the
equipment. Such a test sequence would typically take 3 test samples; the sp2/sp3 ratios were then correlated to film
days. When this delay is equated to the production rates of hardness and wear. Further analysis and interpretation by
disks and read-heads and their respective unit costs (at a Ferrari and Robertson [159, 160] has been based on data
rate of several thousand units per day and assumed cost of obtained with largely the same equipment. Much of the
approximately 8 Singapore $ per unit, say), the relative cost effort has concentrated on defining the bonding relation-
of a single Raman system is insignificant. ships for the respective intensities and positions of the G
Table 8 is a compilation of the extensive engineering and D bands, as well as the interpretation of an additional
information that can be gathered from the relatively simple band near 1150 cm
1 that has been found to scale with sp3
spectrum shown in Fig. 79. Analyses continue to follow the content. Further information on UV Raman spectroscopy
trends of the band parameters as functions of thickness, for the measurement of different types of carbon has been
hydrogenation and nitrogen content, as well as the expected given by Bormett (Renishaw) in the paper by Lauer [161].
wear and friction properties. The spectral analysis can be The use of UV lasers for these measurements is generally
carried out using PASS software, and a user-configured beyond the financial reach of smaller coating companies.
printout for the data gives the critical disk production Hence, UV Raman systems are usually situated at
parameters. Special disk handling units were designed in centralised locations, where time on the machine can be
order to rapidly and accurately present the surface of each accessed. These measurements are also proving useful in
sample disk to the laser beam. The Raman units were assessing the wear/hardness properties of thicker films for
required to meet Class 1 clean room operating specifica- bearing surfaces e.g. gears. Most hard disk manufacturers
tions. A sample disk takes approximately 20 s for each now use UV Raman to provide further quality-based
measurement, and initial tests were carried out to ensure information to enhance production.
that the measurement cycle caused no sample burning
within this time frame. To reduce the power density at the 9.4.3 Carbon fibres: Carbon fibres (e.g. mesophase
film surface, the user can use the ‘line-scan’ and defocusing pitch fibres) are now used in many applications as light
features of the instrument. A further solution is to spin the strength members in synthetic fibres and polymers. The
disk, which then gives a spatially averaged reading of the modulus of the fibre can be adjusted by adjusting the
film. This is not an option for small magnetic read-heads diameter, orientation, and embedded strain applied to the
(MR) where the intensities and focusing techniques must be fibre. Most carbon fibre manufacturers will now have access
carefully monitored for the measurement cycle. Spinning of to the new Raman systems for routine quality control
the disk or sample must become the preferred solution for checks on the manufacturing process to measure the shift of
UV excitation of the films where there is greater risk of the G band, which moves to lower frequency with tensile
burning. stress. Shifts as low as 0.4 cm
1 can be readily measured
Manufacturers also use Raman systems to detect defects using the curve fitting procedures available, and this is
and determine the composition of particulates found on the equivalent to a 150 Gpa tensile modulus. Polarised light
disk/read-head surface. These can then be related to Raman techniques are useful when evaluating orientation
cleanliness problems in the processing line. This investiga- effects and internal deformations of the fibre. The modulus
tive process can have a large beneficial impact on the value is also reflected in the half-width of the two Raman
production yield. bands near 1360 cm
1 and 1530 cm
1. Other quality control
Commercial razor blade edges with DLC coatings have uses are in the analysis of carbon fibre surface treatment,
been monitored using the described Raman equipment as characterisation of functional groups in stabilised pitch
the process feedback and quality assurance tool, and the fibres, and the profile of oxygen content across stabilised
application has been reported by Grill [157]. Other pitch fibres. Earlier X-ray diffraction analysis of carbon
applications for DLC films requiring quality control fibre matrix structures was often masked by the fibre
measurements are in insulating coatings, antireflective wear intensity profiles, and composites of mixed fibre and mixed
resistant coatings, and chemically passive coatings. composite types gave only area averages. The microprobe
Measurements on different samples of tetrahedral nature of the Raman technique gives individual fibre
amorphous carbon (ta-C) and hydrogenated ta-C films at characterisation.
Renishaw by the Cavendish Laboratory and the Engineer- Internal strain within the matrix is also measured by
ing Department, Cambridge, UK (Gilkes et al. [158]) were confocal measurement of the surrounding resin/polymer,
292 IEE Proc.-Sci. Meas. Technol., Vol. 152, No. 6, November 2005
with the positions of their specific Raman and fluorescence as nanoscale cables and wires, including replacements for
bands plotted as the spot focus approaches the critical fibre tungsten filaments in light bulbs, and semiconducting
surface interface. This also provides a method for electronic components; these will depend on their purity,
characterising surface oxidation and other attached com- diameter and chirality. Other potential applications are in
plexes. Further tests and quantification of interface effects nanoscale force sensors.
have been made by controlled tensile pulling of polymer Isolated nanotubes were found to exhibit resonance
fibres and composites, while being observed under the Raman spectra with a multiplicity of reasonably sharp
Raman microscope with the Raman excitation beam bands. It has become routine to carry out Raman analysis
focused at the surface of the fibre. Extensive work in this on a single SWNT. The 1-D nature of the structure has
area has been carried out and reviewed by Young and his provided an excellent test bench for theoretical modelling of
group at Manchester University, UK [162, 163]. the Raman excitations, and no doubt the theory will be
extended to other 1-D systems in future. Reference has been
9.4.4 Carbon nanotubes and fullerenes: These made to the major thrust of the present developments to
carbon structures have provided the basis for an array of measure Raman spectra at small spatial resolutions. This
academic research over the past decade. It is fair to was tackled by combining the Raman system with near-field
comment, however, that large-scale production and the optical microscopy (NSOM), the SEM and other scanning
industrial use of fullerenes are still awaited. Reviews on probe microscope (SPM) devices, with the development of
both topics by Dresselhaus et al. [164, 165] discuss the the Renishaw SCA system (see Sections 6.2 and 6.3).
background structural theory and band assignations for the Use of the apertureless NSOM with resonant enhance-
Raman spectra, together with details of intercalation effects ment is discussed in the results of Hartschuh et al. [88] and
and some of the potential applications. Particular emphasis Wang et al. [89], where an isolated SWNT has been mapped
is placed on the carbon nanotube work since the new using spatial resolutions close to 25 nm. In addition, a defect
Raman techniques for extending the spatial resolution for in the nanotube was also detected [88] (a critical step
measurements down to the nanometre region have been forward in showing that nanoscale impurities can be
particularly apposite. The Buckminsterfullerene discussion identified, in relation to the future purity requirements for
however is of limited interest here. Both carbon polytype carbon nanotubes for semiconductor devices). The standard
structures are highly specific and provide unique spectra optical microscope is not powerful enough to image/map a
with several sharp and readily identifiable bands. For single nanotube, but it is probable that the SPM/Raman
example see Fig. 77 where the spectral structure of C60 combination will be essential for rapid feedback on carbon
buckminsterfullerene shows sharply defined bands. The nanotube processing in future.
omens are therefore good for Raman spectroscopy to Carbon nanotubes are often found as a series of tangled
become the analytical tool of choice in the future, both at ropes and bundles. Raman spectra of bulk SWNTs can be
the research level and for routine analysis. easily measured using the Raman microscope. However
The two structures were discovered around 1990 more by detection of an isolated nanotube requires careful stepping
accident than as a result of directed research. Buckmin- (0.5 mm steps) of the controlled microscope stage, until the
sterfullerenes were found as a result of laser vaporisation characteristic resonant spectrum of an SWNT is located.
experiments on graphite to simulate the environment in a This can be a laborious process. The problem has been
red giant carbon star (Kroto [166]), although their existence solved by the use of the SEM-SCA combination, with the
had been earlier predicted. Iijima [167] discovered carbon SEM having sufficient resolution to enable guiding of the
nanotubes in the graphitic residues left from a carbon arc Raman spot to the required region of nanotubes shown on
discharge experiment. the SEM display. The following results were taken with an
SCA system attached to a LEO-1450VP SEM (see Section
Fullerenes: Fullerene C60 has a unique spherical structure. At 6.2 for the typical measurement cycle). Two samples were
varying temperatures and pressures several phase transitions analysed: a bulk sample consisting of a ‘matted’ mass of
have been followed using Raman [168]. At high pressures an nanotubes and a micro-scale sample of discrete nanotubes
amorphous phase is present, but on reversal to atmospheric aligned between two gold electrodes, 1 mm apart (Fig. 80).
pressure the fullerene structure (perhaps unpredictably) As a control, the bulk sample was also measured with the
re-emerges. The authors have carried out numerous microscope.
measurements over the past 10 years on behalf of several
groups on a range of other fullerenes and mixed fullerene
samples, e.g. C60/C70, but they are not reported here.
Carbon nanotubes: Carbon nanotubes are showing a better
prospect for widespread use. They have remarkable
mechanical and electronic properties, with a thermal
conductivity as high as diamond. Most attention has been
paid to single-wall carbon nanotubes (SWNTs) that can be
pictured as a graphite layer rolled up into a seamless
cylinder, with each end open or capped with a half-molecule
of fullerene. The points at which the rolled up graphite sheet
are connected define the tube’s diameter and chirality,
which in turn govern its properties. Tube diameters are
typically 0.5–3 nm, and can be of the order of 30 nm in
length. The earlier nanotubes were made up of several
concentric tubes, but the SWTNs have improved properties.
For example they are ten times stronger than steel and six
times lighter. It has been suggested that they could reduce
spacecraft weight by 50%. SWNTs can show metallic or Fig. 80 Image of discrete nanotubes suspended between gold
semiconducting electrical behaviour, with applications such electrodes within an SEM

IEE Proc.-Sci. Meas. Technol., Vol. 152, No. 6, November 2005 293
 633 nm spectrum has three bands corresponding to
G-band
diameters of 1.0 nm, 1.1 nm and 1.3 nm. Further analysis
of line shape, and the data at the different wavelengths,
gives additional information. The spectral features give the
following information:
G′-band (a) The frequency of the radial breathing mode (RBM),
usually in the region of 150 cm
1 to 300 cm
1, depends
directly on tube diameter [169]. For the bulk samples a
rapid determination of the tube diameter distribution can be
D-band
obtained.
RBM
a (b) The fine structure seen in the tangential G-band depends
on tube diameter and chirality. The most intense feature
near 1590 cm
1 can be used with (a) above to help to
identify individual metallic and semiconducting SWNTs.
b
Added shifts in frequency are observed with filled
nanotubes (e.g. with AgCl [170]).
1000 2000 3000 (c) The G-band is affected by the oxidation state of the
Raman shift /cm−1 nanotubes, and has been used to measure strain in SWNTs
embedded in a matrix. Measurements on SWNTs have
Fig. 81 Raman spectrum of bulk nanotubes situated inside the
UHV chamber of an SEM using 633 nm excitation been made using 785 nm excitation, by Young’s group at
a Using the SCA Manchester Univ., UK [162].
b Using conventional sampling under the optical microscope (d) The disorder induced D-band depends on nanotube
The results are closely identical; the vertical axis is shifted to give diameter and chirality, while the G-band is the second order
clarity overtone of the D-band, and unlike the D-band, its intensity
is not reduced in highly ordered crystals. The G-band is
used to provide information on isolated SWNTs.
Figure 81 shows the Raman spectra obtained from
within the UHV chamber of the SEM and with a laser It is concluded that these new combined Raman
excitation of 633 nm. The result shows excellent band techniques will be used with increasing frequency in future
resolution for the SEM-SCA combined system. With to control and monitor the physical properties of nano-
633 nm excitation, the bands are 3–4 times more intense tubes.
than for 532 nm excitation, which comes from a combina-
tion of the n4 effect and increased resonance effects. Figure 9.5 Forensic applications
82 shows measurements on a small number of well-defined Multiple Raman measurement units have now been
nanotubes on the micro-scale inside the SEM chamber, accepted in forensic science laboratories, in national law
compared with a result from the bulk sample at 532 nm. enforcement agencies, and by local police forces. The main
The latter bands are much wider due to inhomogeneities. criteria for selection of a Raman system from these
From analysis of the spectra using the variables and their institutions are many and varied, but the units tend to be
respective relationships listed below, the spectra indicate used in tandem with already installed infrared systems. The
tubes in the region of 0.9 to 1.3 nm. The more intense chief requirements are as follows:
1. The data printouts from the Raman test are always to be
G-band
considered as potential evidence in court. Therefore the
system printout must comprise the result, operator, full
sample particulars, date and time etc. There must be set
procedures for system backup and data retrieval, whilst the
paper used and inks should be suitable for storage.
D-band
2. All measurements and calibrations must satisfy full
traceability. For the present systems, lasers, wavelength
calibrations, background correction, displacement measure-
a
ment (e.g. rotary stages) etc. have all been measured with
NPL or NIST standards traceability. Moreover the soft-
ware is designed to meet 21CFR Part 11 quality standards
for additional traceability.
3. System start-up procedures are specifically designed and
b automated to cover the respective calibration sequences,
with printout. For forensic laboratories the written
procedures for each measurement type, e.g. fibres or drugs,
1000 1200 1400 1600 will require the test running of standard control samples
Raman shift /cm−1 before and after the critical measurements. Therefore the
software and printouts must be user friendly, with the
Fig. 82 Raman spectrum of carbon nanotubes acquired using capacity to accept all the in-house libraries and databases
532 nm excitation inside the SEM that may be required.
a Discrete
b Bulk 4. Portable systems must meet the appropriate environ-
Figures 83, 84 and 85 originate from work carried out on a Raman– mental specifications, e.g. vibration and drop test require-
SEM machine, courtesy of Jim Butler, NRL, Washington DC, USA ments. In some cases the instrument is housed in a
294 IEE Proc.-Sci. Meas. Technol., Vol. 152, No. 6, November 2005
specialised vehicle, and in others it can be portable and 30 000
housed in a small suitcase, with easily demountable fibre 25 000
optic connectors. This is frequently referred to as the in-

counts
20 000
flight and airport-carousel survival test.
15 000
5. For police work and scene of crime activities operation of
the portable or vehicle based system should ideally consist 10 000

of one large switch labelled ‘ON’ and ‘OFF’. This is to 500 1000 1500 2000
somewhat overstate the degree of simplicity required, but it Raman shift /cm−1
originates from a common theme of several discussions with
police authorities. The data from the portable units tend to Fig. 83 Raman spectra of lines taken from a suspect figure ‘4’ on
be referred back to centralised systems in the central an insurance claim document (right)
laboratories with regular calibration and validation proce- The original ‘1’ has been modified to a ‘4’. The top spectrum comes
dures. from the added ink

This list is not comprehensive, but the requirements serve

to illustrate the specialised nature of forensic and police 1. Figure 83 shows the spectra from a document used as
work. A critical result in the acceptance of a new analytical evidence in an insurance fraud. The result shows spectra
technique for each application comes when it is used in a taken from a ‘1’ modified to a ‘4’. The spectrum of the
court of law to secure a conviction or more frequently to original ink on the ‘1’ gave a Raman spectrum on a broad
provide good reason for further investigation. Over the past background fluorescence, while the added ink gave a broad
few years this has occurred frequently. This was not the case fluorescence with only weak Raman features. It was
while earlier unwieldy double or triple monochromator superfluous to manipulate the latter data to extract the
systems were being used. Raman features more clearly, since the difference in ink
A recent review of Raman work in the forensic field has spectra was so marked.
come from Smith et al. [29]. This concentrates largely on the 2. Figure 84 shows spectra taken from another suspect
specialised use of surface enhanced resonance Raman intersection on paper made with strokes from one or more
scattering (SERRS). There is a bias towards the author’s blue pen(s). It is impossible visually to determine which
interests in the examination of documents and inks as well stroke was made first. The high spatial resolution confocal
as detection of explosives. Of the numerous reviews on the facility on the microscope enables the subsurface region at
use of Raman over the past decade, few have dealt with the the intersection to be probed. The spectrum from the
more practical aspects of Raman for use in the field by surface reveals Raman bands and fluorescence characteristic
professionals. of (a), while from 5 mm below the surface (b) the Raman
features are similar but there are large changes in the shape
9.5.1 Inks, questioned documents and of the background fluorescence probably due to differences
fraud: The paper by Claybourn and Ansell [171] provides in the organic binders. It was concluded that the amend-
an excellent introduction to the potential problems in this ment of ‘1’ to a ‘4’ was made at a later date.
area, and the use of Raman in their resolution. This forms
part of a continuing series of joint measurement projects on
documents with expert witnesses and expert analysers. As a 6000
result, several real case examples can be described. Raman 5000 laser focused at
techniques appear to have solved some issues that proved top surface
4000
counts

difficult for earlier professional document examiners, i.e.

comparisons of black ballpoint inks and chronological 3000

sequencing of crossed lines. These basic problems for fraud 2000

laser focused 5 microns
into the intersection
investigators are exacerbated by the lack of reliability and 1000
non-reproducibility, as well as the destructive nature of 400 600 800 1000 1200 1400 1600 1800 2000
present testing, using more traditional methods.
For comparing inks, visible and infrared luminescence Fig. 84 Raman spectra taken through the intersection of a crossed
have tended to be used, but these methods have poor ink line
chemical selectivity as evidenced by their broad bands
compared with the many characteristic and sharp bands Diagnosis is sometimes not as simple as the cases given
found in Raman. Other methods have included such above might indicate, owing to the cross-interactions of the
destructive techniques as thin layer chromatography [172]. inks at the intersection, and there have been a small number
Inks age with time. This can be a function of ink type and of instances where some doubt exists over the interpreta-
environmental conditions. There are small changes in tion. Apart from the detection of forgery, measurements
Raman spectra due to slow photochemical changes in the have also been made to investigate the origin of obscene
ink and the paper (e.g. surface oxidation), as well as solvent letters and the altering of police statements. Testing has
loss and cross-linking of components. This area remains proved that in many cases fraud has not occurred. It is
complex, and it will be a fruitful research area over the next possible that in future Raman will be combined with
few years. The present dating methods rely on destructive microscopic 3-D mapping of crossed lines to provide added
analysis of the ink’s chemical composition. The more information.
limited range of commercial printing inks [173] and Several governmental forensic laboratories have now
standardised papers preferred for computer printouts may built up extensive libraries of ink spectra for all colours, but
be a more worthwhile area of investigation. with greater emphasis on those inks that are more difficult
Raman can give clear non-destructive evidence of line to distinguish, i.e. black and blue. Figure 85 shows spectra
sequencing and alterations to numbers, e.g. on a cheque of blue and black inks courtesy of Niu T.L., of the Police
with a different ink. Typical case examples using the present Ministry, PR China. The different inks have varying
Raman microscope system designs [172] are: formulations to meet internal specifications relating to

IEE Proc.-Sci. Meas. Technol., Vol. 152, No. 6, November 2005 295
 Fig. 86 Photograph of a 5-layer paint flake taken from a car to be
tested for the origins of the layers by Los Angeles Police Dept. USA

1800 1600 1400 1200 1000 800 600 400

Raman shift /cm−1 layer 3
layer 1
Fig. 85 Raman spectra for a series of blue and black inks Hit: rutile (synthetic) (rutile group)
Most forgeries involve these inks. Data courtesy of Niu T.L., Police (INORGAN.LIB #559)
Ministry, PRC

colour, colour density, viscosity, water-fastness, light-

fastness and fungal resistance and, not least, cost. The
chemical composition and organic content (dye/pigment)
therefore vary widely.
The authors are aware of several fraud and forgery cases
in the UK where the Raman data presented have been so
convincing that there was a change of plea. Inks from
banknotes have also been classified in various countries in
order to combat forgery.
500 1000 1500 2000
−1
9.5.2 Paint and pigments: The identification of Raman shift /cm
paint chips is valuable for linking tiny paint particles to a Fig. 87 Raman spectra of layers 1 and 3 from Fig. 86
vehicle in a road accident, for example. IR reflectance Results identify the paint colour and that the pigment has the rutile
spectroscopy was once the preferred analytical technique structure of TiO2. The data is compared with the library spectrum for
where the spectra were matched to the standard materials in rutile (below). Data courtesy of LAPD, USA
the literature. Unfortunately the spatial resolution is poor,
i.e. 10–20 mm, which is insufficient to resolve the separate
layers in a multi-layered paint sample. Analysis was A further, and very important, extension of this type of
therefore achieved by successive removal and analysis of forensic science is the identification of fibres (natural or
each layer; this procedure destroys the evidence. With the synthetic) discovered at the crime scene, and of the pigments
high spatial resolution of the microscope system the in these fibres. The combination of Raman-SEM will
different layers can be distinguished according to type and become increasingly useful here.
manufacturer.
The result described here originates from a Raman 9.5.3 Gunshot residues (GSRs): When a gun is
microscope installed at the Scientific Investigations Division fired, minute traces of gunshot residue, originating mainly
of the Los Angeles Police Department (LAPD), USA. They from the primers in most cartridges, will be deposited on the
were able to identify the constituents of a five-layer paint handler’s clothes and hands, particularly over distances less
chip in conjunction with the Renishaw inorganics spectral than 1 metre. These deposits, consisting of aspherical
library. The video image shown in Fig. 86 is a side view of particles, may be measured directly on the suspect’s clothing
the paint chip with scale bars to indicate dimensions; by the Raman microscope, or collected on specialised stubs
minimal sample preparation was required prior to examina- for SEM analysis. Residues can consist of gunpowder
tion. The layers are numbered 1–5 from the left, and are particles, carbonaceous soot as well as nitrate and nitrite
indicative of a car that has been repainted several times. compounds, and lead, barium and antimony compounds.
Figure 87 shows the spectra taken from layers 1 and 3 with The gunpowder itself can be single based (nitrocellulose),
the library ‘hit’ below indicating synthetic rutile (TiO2), a double based (with nitroglycerin), or triple based. The
white pigment. Layer 2 was found to be goethite; layer 4 was residue tends not to degrade with time and so a collection
molybdate orange, and layer 5 was a silicate-based overcoat. process of wiping the stub on the suspected areas is deemed
The whole procedure took 2 minutes with good quality hits to be satisfactory. However this subject area requires a high
from the library. All layers were related to manufactured degree of experience to fully understand and interpret the
paints with some information on the dates they were being data. There can be a wide range of possible compositions
used. Many police departments have now extended their through to the contamination of samples by blood at the
spectral libraries to include all manufacturers’ paints using crime scene etc. More recently, for example, the inclusion of
the system’s facilities. The original samples are retained in a lead in ammunition has been reduced for toxic reasons, and
suitable environment, and stored to ensure traceability. polymer jackets (e.g. with cobalt blue-cobalt aluminate)
296 IEE Proc.-Sci. Meas. Technol., Vol. 152, No. 6, November 2005
have also been designed to further reduce the dispersion of – With small portable Raman units and collection of data at
lead [174]. These developing aspects illustrate the experience the scene of the drug collection. A small fibre optic probe
needed for accurate interpretation. can be used.
McCrone Associates Inc., USA, have considerable – At a Raman microscope where the sample can be
experience in analysing GSRs, mainly using the SEM (for measured directly through a transparent evidence bag to
further information see their case histories [175]). They have avoid tampering with the sample or cross-contamination.
now added Raman to their battery of analytical techniques At some local police, state and forensic laboratories, large
to raise their identification level and determine the backlogs of seized samples awaiting examination have built
molecular species in the residues rather than just the up, since the normally used chromatography methods in
elemental content. They possess a large amount of historical general take too long.
information on the materials used by manufacturers of
– At national laboratory level where the more costly UV
ammunition.
Raman systems have been used to eliminate background
We have described earlier (Section 6.2) the combination
fluorescence effects and obtain clean resonant spectra. An
of Raman with SEM, and this is already proving the tool of
choice for identifying GSRs, either in linked format or as example of this is the work done by the Police Scientific
Development Branch, UK in collaboration with Renishaw
separate units with transference of samples. The EDX
and Leeds University [178]. This work also carried over to
spectra will typically identify the residue content to be
composed of different concentrations of lead, barium and the detection of explosives (see next Section), where the
more recently available (and lower cost) 830 nm near
antimony for example. The results from three particles
infrared lasers are also used to reduce fluorescence.
shown in Fig. 88 (J. Yarwood of Sheffield Hallam
University, UK [176]) show that they are associated with – Carrying out surface enhanced Raman scattering for the
barium oxide and barium carbonate. This compound detection of minute quantities of drugs and explosives using
identification provided additional information to the specialised material surfaces [29]. This work is still in its
elemental SEM data on the samples. infancy and will only be referred to briefly later. It could
have the potential to determine very low concentrations of
certain drugs quickly (within minutes) when in solution, e.g.
urine and blood samples.
– By direct Raman imaging over larger areas. Figure 89
A shows the white light image of an area of the suspect
material. It shows a mass of indistinguishable matter.
Taking images at the critical wavenumber for cocaine
hydrochloride, the regions of cocaine are shown as bright
spots, and the dark regions were identified as being largely
B
sugar. Following the rapid global image result, the different
regions can be explored in more detail.
– Some work by Bell et al. [179] has shown that the analysis
C time for measurement of ecstasy and related phenethyla-
mine tablets has been drastically reduced using Raman
1400 1200 1000 800 600 400 measurements. They used a laboratory test rig with a
Raman shift /cm−1
785 nm Ti-sapphire laser pumped by a 100 mW Ar+ laser.
From the data, they were able to show that fluorescence
Fig. 88 Raman spectra from three gun shot residue (GSR) was reduced and excellent spectra were obtained, e.g. with a
particles typically 2 min accumulation time. This increased the
Features associated with barium oxide (BaO) and barium carbonate throughput for measurement of seized samples to 20
(BaCO3) are exhibited. The features are sufficiently different for samples per hour compared with much longer times for
determination of the firearm to be made. With an SEM alone, this
earlier techniques. With the Raman method, there is an
degree of discrimination is not possible. Data courtesy of J. Yarwood,
Sheffield Hallam University, UK additional benefit of the retention of the sample for further
analysis if required owing to the non-destructive nature of

9.5.4 Drugs/narcotics: Forensic use of Raman tech-

niques has branched into several directions for this
application.
Identification of bulk contraband substances such as
amphetamines, cocaine and heroin is important for drug 500 µm
enforcement. Also of interest is the identification of
precursors and bonding agents in tablets. In some
methamphetamines synthesis may use a precursor such
as phenyl acetone, and others may use variants of
ephedrine. An awareness of these additional materials
including cutting agents for cocaine, for example, provides
intelligence on the manufacturer, the place of origin and
the transport route. The standard analytical techniques a b
employed to date comprise chromatography and capil- Fig. 89 White light and true Raman images of cocaine
lary electrophoresis [177]. In many respects Raman is The white light image (left) gives no indication of cocaine content. The
simpler and faster but it is likely that the analytical cocaine hydrochloride shows very clearly after Raman imaging (right).
techniques will be used in tandem. Measurements have Measured against the scale, the particles are estimated to be
been carried out: approximately 5 mm diameter, and are situated within a sugar matrix

IEE Proc.-Sci. Meas. Technol., Vol. 152, No. 6, November 2005 297
the measurement process. Unfortunately, they did not have RDX (cyclotrimethylene-trinitramine) and PETN (pentaer-
access to the improved 830 nm laser diodes that further ythrital-tetranitrate) in the ratio 60:40 respectively, and
reduce fluorescence. Nevertheless the spectra are sufficiently bound by an amorphous waxy material. Lewis et al. [182] in
rich in information that it is possible to obtain quantified collaboration with the FBI (USA), and using the present
data on the content, e.g. the excipient and different instrument, carried out a wide-ranging series of Raman
polymorphs and hydrated forms making up the drugs of measurements on many explosive species. Some fluores-
abuse. In Section 9.6 dealing with pharmaceuticals, there is cence was found from the binders, but the spectra were still
some discussion on the detection of different drug contents easily identifiable. They also carried out successful measure-
and their quantification using various mapping/imaging ments through transparent evidence bags. At 730 nm the
methods and specialised chemometrics software. The fluorescence was less and at 1 014 nm with FT-Raman it
problem here is effectively the same, with the objective of was still lower. A comment made at the time suggested that
improving sample throughput, and in this case using the a diode laser at 840–50 nm would be useful for this
highly specific drug content to identify the drug’s trace- application and this is now available. Spectra of the binders
ability of manufacture. With the new software develop- and other constituents, e.g. sugar, scouring powder, are also
ments for spectral analysis, and use of the 830 nm laser, the usefully determined for an evaluation of the origin of the
analysis throughput of such seized drugs will further explosive. The drugs and their cutting agents are combined
improve. with the binders and explosives into a single comprehensive
library.
As referred to before, in this field absolute rigour is A standard depletion series test for explosive contamina-
required, with clearly defined measurement procedures and tion of fingerprints was carried out [183]. The RDX
clean operating conditions to avoid cross-contamination contaminated finger was continually pressed 40 times on
effects. This has also involved the construction of sample to an uncontaminated surface (in this case an aluminium
handling methods for consistent presentation of the sample plate). The spectrum of RDX was still identifiable at the end
to the Raman instrumentation. of the exercise [184] and the concentration of the RDX at
this stage was estimated to be 1 picogram. To check the
9.5.5 Detection of explosives: In a continuing distribution of the RDX on the surface of the fingerprint,
series of joint projects involving Leeds University, Strath- direct Raman images were taken. Figure 90 shows the 2-D
clyde University and the PSDB (Home Office, UK), Raman image and white light image of the surface and a
measurements have been made on numerous explosives 2 mm diameter particle of RDX can be clearly seen. RDX
using remote detection methods through to highly sensitive and PETN have strong bands at 885 cm
1 and 874 cm
1
SERRS techniques. It is becoming increasingly important respectively, so for Semtex the 20 cm
1 imaging filter
to develop systems that will give reliable detection of bulk window can be set at 880 cm
1 for both. The image was
(and vapour phase) explosives at a contamination limited taken using 632.8 nm excitation in 20 s with background
level, and with low incidence of false alarms [180]. In this subtraction.
Section we shall concentrate on measurements of RDX, The variants in microscope stage available, with ample
PETN, and TNT (compositions explained below). space for locating samples, has allowed measurements to be
The requirement comes from the need to detect made on large sample surfaces, e.g. clothes and soles of
explosives in suitcases, packages, transport vehicles and shoes, with no special sample preparation.
aircraft. At present several methods are available for bulk Further measurements have been carried out using UV
materials, i.e. X-ray fluorescence spectroscopy, HPLC, and resonance Raman [33] on bulk samples. Low power
capillary electrophoresis that necessitate taking the sample excitation was used, and the bulk samples were spun under
to the machine for what is destructive analysis. At airports, the microscope in order to prevent burning. The technique
X-ray imaging has been used together with nuclear was investigated to develop a further option to avoid
techniques for the detection of the vapour phase. Since background spectra from grease etc. Resonance bands were
direct Raman laser excitation and measurement is not found with an enhancement of approx.  105 in intensity
possible through a suitcase, the application calls for a over previous work using visible/near IR lasers. Remote
method that detects the vapour phase. fibre-optic spectra using visible lasers [97] were successfully
Raman also finds a use in the measurement of bulk taken over a 100 m distance.
samples when, for example, a bomb making factory is being The vapour pressure of many explosives is relatively high.
evaluated. Containers, bottles and liquid explosives or Despite the material being enclosed in a suitcase for
active chemical reagents are ideally examined in situ and example, a suitable sniffer can take a sample, and the
non-destructively. Hence there has been some work on
remote detection using Raman and fibre-optic probes.
Differentiating between mixed containers of dangerous and
safe materials is difficult and time-consuming. Transporta-
tion of samples back to the laboratory may increase the risk
of cross-contamination. For example, RARDE (Royal substrate

Armaments Research and Development Establishment) at

Fort Halsted (Sevenoaks, Kent, UK). was found to have an
unwanted background level of contamination from Semtex.
RDX
This caused concern over identification of subsequent cases
[181].
Explosive materials such as TNT (trinitrotoluene), 100 µm
ammonium nitrate and sugar, C4 and Semtex have sharply
a b
defined spectra with many readily identifiable bands. There
is also little or no fluorescence background. The waxy Fig. 90 RDX Raman true global image taken from a fingerprint
binders used, or grease from fingerprints may cause a The RDX 10 mm particle is not distinguishable within the white light
problem in this respect. The constituents of Semtex are image (left)

298 IEE Proc.-Sci. Meas. Technol., Vol. 152, No. 6, November 2005
vapour transferred to a measurement cell. To measure the accommodate possible changes in material concentration
required levels above the contamination level, it is necessary variations from the surface measurement relative to bulk
to add all the Raman enhancement features and this concentration.
specifically means SERRS [29]. The goal of reducing TNT A general summary of activities is:
to a suitable derivative, possessing a chromophore that will
give resonance effects with visible excitation, has been 1. Analysis of drug mixtures, actives (incipients) and
achieved [185]. The samples were initially analysed in a excipients and their distributions;
single stage flow cell for mixing the analyte, sensitising silver 2. Identification of contaminants in the drug formulation;
colloid and aggregating agent. The intensities of the Raman 3. Characterisation of formulated materials and an under-
bands were found to be pH sensitive, and under optimum standing of the blending processes involved;
conditions the dye was detected at 1  10
9 mole. 4. Identification of minute quantities of polymorphs within
For RDX, the reduction process has proved to be more the formulation, for both in-house controls, and for
complex since RDX does not readily give good SERRS competitive and generic products to detect patent infringe-
spectra. After several attempts, the reduction to a ments;
specifically designed pyridine azo dye has shown promise
for detection at the 10 femtomole level [186]. The initial flow 5. In situ crystallisation/hydration/process. Test data using
cell work and use of colloids for explosive detection has the relative humidity Raman cell has proved useful here
progressed to the use of more consistent prepared surfaces (Section 5.4.2);
where a higher degree of control is now possible for 6. High throughput screening and dedicated well plate
uniformity and surface roughness, in addition to the use of analysis for drug discovery;
a chromatography technique to obtain the required 7. Analysis of drug tablets in finished form through the
molecular specificity. blister packs, and reducing the need to crush and press into
a KBr pellet for ATR (attenuated total reflectance) infrared
9.6 Pharmaceutical applications analysis.
In the recent 2000 review of the use of Raman in
pharmaceutical applications, Williams [187] gave a relatively
pessimistic picture of increased use of Raman in the 9.6.1 Goods-in verification of raw materials:
pharmaceutical industry at that time. However he was Prior to use all components of the formulation need to be
quite optimistic about the future. The explanation given rapidly checked. The capability of the new systems to be
here is that the primary Raman technology before 2000, used with fibre optics for remote detection, including through
and accepted by most pharmaceutical workers, was FT- the container or vessel wall, is ideally suited to input
Raman, because of its perceived superior performance with acceptance procedures. An example is the rapid checking of
respect to fluorescence. This fostered a natural conservatism various solvents through Winchester containers. However at
in the selection of new analytical Raman techniques by present most of the analysis of powders is still done by near
pharmaceutical analysts. The changes that have occurred in infrared analysis primarily on the grounds of cost.
dispersive systems since that time have been considerable,
with the emphasis on a combination of high spatial 9.6.2 Analysis of blended and tableted ma-
resolution, stability, speed of measurement, spectral terials: Earlier we described how Raman has been used
libraries, infrared and UV excitation lasers, and the ability for evaluating the uniformity and interactions (or lack of
to measure the contents of aqueous mixtures. Over the past them) in asthmatic inhaler mixtures [54]. Here we examine
few years a remarkable increase in the use of these new the analysis of blended and tableted materials. Raman offers
Raman systems for practical pharmaceutical use has been a method of analysing the homogeneity of the blending
observed. The examples given in this Section illustrate process at the micron level. Units have been used to monitor
several of the advantages. samples taken from different levels in the blender, but this in
The industry is properly surrounded by rigorous quality itself does not provide a rigorous analysis of the mixture:
standards. Work was needed to meet the IQ/OQ (installa- mapping must be employed. Analysis has been made on two
tion quality/operational quality) equipment standards for tablets, one from each manufacturing process [189]. The
the various companies by preparing the documentation to idea was to determine the uniformity of the spatial
fit within their internal operational systems. Furthermore distribution for the various ingredients.
the Federal Drugs Agency (FDA) has been encouraging Table 9 summarises the two different manufacturing
process analytical (PAT) initiatives to monitor the proces- processes for Paracetamol tablets. There are three main
sing of pharmaceutical technology products on-line. ingredients:
Raman has proved to be a major tool in this advance 1. Paracetamol–active constituent. The excipient is the
and systems are now used on-line at all parts of the bulking agent. Only micrograms of the active component
production process from speedy assessment of ‘goods in’ need be present in the tablet to be mixed with the inert (in
through to a final test of the tablet inside its package. Apart the pharmaceutical sense) excipient for handling purposes;
from the hardware developments, good chemometrics
software for the analysis of multi-component images is 2. Dibasic calcium phosphate dihydrate;
now routinely available to extract the desired quantitative 3. Magnesium stearate–used as a solid lubricant in
data from the Raman data obtained from the new systems. manufacture. It is important to understand its distribution
The enhanced system stability (low system drift) over the throughout the final product.
collection time of a typical image has allowed such analyses
to be increasingly valid. A more recent review [188] of The tablets were microtomed and the tablet core exposed.
Raman in the pharmaceutical industry emphasises this The critical Raman bands for each component were
point with a collection of examples where volume identified, and the integrated intensity for each band was
percentage accuracies for polymorph content and other used to obtain 2-D maps for the different components.
components have been measured from tablet mixtures Figure 91a shows the Raman spectrum of magnesium
inside and outside blister packs. The experienced user can stearate, with the critical band at 1287 cm
1 marked.

IEE Proc.-Sci. Meas. Technol., Vol. 152, No. 6, November 2005 299
Table 9: Summary of the manufacturing processes for the drug Paracetamols

Wet granulation process (WG) Direct compression (DC)

1. All ingredients added to granulator except magnesium stearate. 1. All ingredients added together except magnesium stearate.
2. Mixture blended and granules made by adding water. 2. Mixture blended.
3. Granules are dried. 3. Magnesium stearate added.
4. Magnesium stearate added and mixture blended. 4. Mixture blended and made into tablets
5. Mixture made into tablets.

300 µm molecular structure. Raman is the optimum technique to

measure these because of the highly specific and sharp
nature of the spectra: we can be looking for small changes
in spectra for low concentrations of constituent, e.g. the
800
examination of the different forms of hydrochloride
700 * dihydrate. Infrared absorption is problematic since the
Raman counts

600
500 water absorption relating to the hydrate obliterates critical
400
300
spectral features.
Polymorphism has become a key area in the protection of
300 µm

200
100
900 1000 1100 1200 1300 1400
patents by the major drug companies against the encroach-
Raman shift /cm−1 ment of generic drug manufacturers. A ruling following US
a litigation over the antidepressant Paxils appears to open
the door for increased generic manufacture, although the
300 µm
process is still open to appeal [190]. At the more general
level, all patents are listed in the ‘Orange Book’ published
by the Food and Drug Administration (FDA), and a
generic manufacturer must provide an appropriate patent
2000 certification involving a challenge to the Orange Book
1800
patents, i.e. that the listed patents are invalid and/or will not
Raman counts

1600
1400
1200
be infringed by the proposed generic drug. The brand
1000
800
manufacturer is then given a period of time to challenge the
certification. Polymorph patents and picket fences (several
300 µm

600 *
400
200 patents related to the core patent) are critical.
900 1000 1100 1200 1300 1400
Raman shift /cm−1
Polymorphs refer in the pharmaceutical context to other
b
molecular forms including hydrates, e.g. bound water and
solvates present in the crystalline structure, as well as small
Fig. 91 Combined images of the surface of Paracetamol tablets, differences in the structures of molecules of the same
with all main constituent images superimposed. The content of chemical composition. The former are referred to as
magnesium stearate over the surface can be variable. The images are pseudopolymorphs. In the past these were characterised
collected using point-by-point mapping by X-ray powder diffraction patterns and infrared absorp-
a Spectrum taken from a region rich in magnesium stearate. The
critical band used for mapping is the magnesium stearate band at
tion. Given the complexity of patent cases and the low
1287 cm
1 concentrations of materials that are being considered for
b Spectrum taken from a less magnesium stearate rich region. It is seen infringement or otherwise, it has become increasingly
that the other constituents comprising dibasic calcium phosphate important to carry out Raman measurements. The
dihydrate (985 cm
1) and Paracetomol – the excipient – (1168 cm
1) sophisticated brand manufacturer needs to be able to
have increased intensities in their respective bands compared with a identify the absence of the patented polymorph from the
above active pharmaceutical ingredient (API). The properties of
the drug must be tracked from the raw material to the
Figure 91b shows the spectrum taken from the mixture of pharmacy shelf. It will have been subject to various
dibasic calcium phosphate dihydrate with the band used for environmental conditions. In patent actions additional,
mapping marked. The maps for wet granulation (WG) and more specific, techniques were used than had earlier been
direct compression (DC) tablet mixtures are shown in employed, specifically the new Raman spectrometer design.
Figs. 92a and 92b, with the dissolution maps separated, and In Section 9.6.2, the example given required a lengthy
then combined. It is found that there is a clear difference in period for completion of the map or image in 2-D. Much of
magnesium stearate distribution, i.e. DC has many small the effort within the pharmaceutical industry is to
and distributed particles, while WG comprises a few large investigate methods for speeding up the measurement cycle
clumps. This result has confirmed the distributions [191]. It is now accepted that the stability and measurement
suggested by other experimental methods. sensitivities are adequate to give data that allow quantifica-
To collect the data, the laser power at the sample was tion of distributions in mixtures to low percentage content.
2 mW with a static grating of 1800 lines cm
1 centred at Thus line-scan and global imaging techniques are under
1130 cm
1. The areas for DC and WG were 1200  1200 constant evaluation. Recent tests on coated bead samples
and 1180  850 microns2 respectively, and the point-by- by the Pfizer Global Research and the University of
point measurement times were 165 and 133 h respectively, Greenwich groups [192] compared both univariate and
with step sizes between 6 and 20 mm. multivariate methods for analysis of spectra from the same
samples, and found that the multivariate approach
9.6.3 Detection of polymorphism: The efficacy efficiently reduced noise and produced good images
of the drug can depend on small changes or modifications in even in the presence of added noise in the original data.
300 IEE Proc.-Sci. Meas. Technol., Vol. 152, No. 6, November 2005
 DC results WG results
300 microns
300 microns

a b a b

2000
1800
800
1600 ∗
700
Raman counts

1400

Raman counts
1200 600
1000 500
800 400
∗
600 300
400
200
200
900 1000 1100 1200 1300 1400 100

wavenumber, cm−1 900 1000 1100 1200 1300 1400

wavenumber, cm−1
c d c d

Fig. 92 Raman imaging results obtained from two tablet manufacturing processes
A more detailed series of images for direct compression (DC) and wet granulation (WG) processes. All the main constituents have separate images
which can be superimposed
a Paracetamol distribution using spectral line at 1168 cm
1
b Dibasic calcium phosphate dihydrate using spectral line at 985 cm
1
c Magnesium stearate distribution using the spectral line at 1287 cm
1
For WG there are larger clumps of magnesium stearate. The DC process distributes the magnesium stearate more uniformly

Thus with the appropriate analysis the measurement time combinatorial chemistry for drug discovery and measure-
can be radically reduced. Although this is useful for ments of arrays of tablets through their blister packs. More
pharmaceutical applications, it is perhaps not as essential recently there have been uses of the technique with small
as the requirements for high-speed mapping/imaging in the microfabricated cells (lab-on-a-chip) where parallel synth-
biomedical and semiconductor spheres. In the former it is esis of new compounds can be carried out and also the
important to gather quantified distribution data on chemical reactions followed. More improvements in this
biochemical reactions as they change, while for the latter and related technologies can be expected in future. The
there are mainly commercial pressures on wafer throughput outputs can range from data on the reaction process to
to consider. For these reasons there have been discussions simple pass/fail criteria.
on the speed of measurement across silicon wafers and Use of combined systems, Raman-SEM and Raman-IR:
the analysis of biomedical samples. The ability to use Taking a tablet section (for example), the SEM shows the
binning procedures, and to adjust/optimise the optimum morphology and atomic number with image contrasts to
spectral resolution in the new systems gives more flexibility extremely high resolutions. The Raman then identifies the
in increasing measurement speeds to provide valid data. constituents in terms of their chemical composition and
Thus both the major manufacturers and generic structure. The use of an SEM and Raman in tandem (albeit
companies are installing numbers of Raman systems not necessarily combined) is a consistent theme (e.g. [54]) in
throughout their organisations for basic research, process several of the examples described earlier. A further
monitoring and quality control. development has been to extend measurement capabilities
to measure Raman properties under varying humidity (e.g.
9.6.4 Detection of inhomogeneities and con- using the relative humidity cell described in Section 5.4.2)
taminants: Foreign particles or unwanted accumula- and other environmental variations to replicate bio-reactive
tions of constituents can enter the manufacturing process effects and the storage lifetime of pharmaceutical products.
accidentally or from faulty chemistry. The Raman system The use of environmental SEM chambers combined with
quickly identifies rogue particles to o1 mm size, leading to Raman in future is an interesting proposition. The Chenery
an understanding of the failure point in the process. and Bowring [193] investigations are a good example of
how combined Raman-IR imaging measurements have
9.6.5 Use of other techniques related to provided critical molecular explanations for bonding
Raman: A number of additional methods used with the strengths and cell toxicity of tissue adhesives.
new systems are briefly mentioned.
9.7 Semiconductors
Multiwell analysis: Fig. 93 shows a typical screenshot of a There is enormous back-literature on the Raman effect in
continuing measurement cycle on an array of multiwells semiconductors with many major reviews and books on the
containing different mixtures of materials. The system topic [194–196]. Whilst the subject is covered from a
operation, software attributes, and user friendliness of the fundamental/research viewpoint [194] there have been
system have been briefly discussed in Section 5.1. The same attempts to describe more practical uses of the effect [195,
procedures can also be used for chemometrics analysis, 196]. Use of Raman systems for critical measurements to

IEE Proc.-Sci. Meas. Technol., Vol. 152, No. 6, November 2005 301
Fig. 93 A system screen shot showing a typical 96-well analysis in progress. The scanning of the multiwells is automated, with stored data for
all spectra
The spectrum shown on the right comes from the second well on the top row; it gives counts against cm
1

control or provide feedback to production processes 25 nm, but the data collection time is too long at present for
remains limited. The reasons for this are difficult to efficient production line operation. Further work will be
determine since semiconductor Raman spectra tend to be necessary to improve the response. The likely solution is
sharp and highly specific, and should therefore lend SERS (local field enhancement) coupled with novel
themselves to relatively easy analysis. It is suggested that positioning control of the Raman probe tip.
the chief reasons are that semiconductor processing At present there are several applications that have
technologies are extremely fast moving with feature sizes entered the production process in niche areas. Some of
reducing rapidly per year, and it has proved difficult for these will be described, with emphasis on the engineering
Raman technology and measurement resolution to keep integration for silicon fab process control, i.e. defect review.
pace, particularly when the earlier Raman system manu- The detection of strain in multilayer and quantum well
facturers had a scientific base rather than a larger scale devices reflects directly on device efficiency, and examples
manufacturing and supply base. To quote a well-known are given for silicon processed integrated circuits, and Si/Ge
Japanese semiconductor researcher, ‘The semiconductor alloy structures, where lattice mismatches at layer interfaces
industry is based on electrons and holes rather than and the corresponding effects of strain on conduction and
phonons’. This implies that a high level of education is valence bands need quantifying. Strain and alloy composi-
necessary to introduce Raman systems into critical parts of tion are also critical for GaN and alloys with InN and AlN.
the production line, and that the prime use of Raman is for In these cases we shall describe how the new developments
structural (including strain measurement) and chemical have made the use of Raman systems more effective in
identification. The new Raman system designs do incorpo- several sectors. Some technology developments and their
rate however the possibility to detect both phonon and validation are described first of all, to give some idea on
electron interactions. The combination of Raman and PL how they relate to recent semiconductor applications.
measurements in one scan, and also Raman – PL – CL
when linked to the SEM is now available. In these cases 9.7.1 Technology developments and results:
information can be obtained for impurity and carrier High stability and high resolution measurement of strain
concentrations. effects in silicon: We have demonstrated that the new
Currently all semiconductor companies have Raman Raman systems have improved the overall accuracies
systems specified as a necessary tool for their background and repeatability of measurement (Section 4.4) such that
research and design modelling checks as well as trouble- mapping of complex integrated circuit structures can
shooting for the identification of particulates and defects. now be carried out over extended periods. Taking
Taking this further, the consumer electronics feature size silicon, and concentrating on the 520 cm
1 band with
is rapidly approaching a 90 nm technology base. We visible laser excitation, a point measurement can be
compare this with the diffraction limited Raman spatial made in o0.1 s. The step time to the next measurement
resolution of 0.5 mm for the present Raman microscope. point can also be of the same order. Over larger areas,
The combination with NSOM, as described in Section 6.3.2, the line scan facility is faster. Both will use the new
can give Raman data with a spatial resolution down to autofocus facility to retain optimum signal. At each
302 IEE Proc.-Sci. Meas. Technol., Vol. 152, No. 6, November 2005
point within the line scan, the intensity, half-width and readings were taken at 520 cm
1 and the background
wavenumber are logged. The data are retained for subtracted at 620 cm
1 (i.e. away from the main band).
subsequent line shape and curve fitting analysis. Figure The light regions correspond to the silicon and the black
94a is an optical micrograph of an aluminium-patterned regions to the aluminium contacts. Figure 95a is an
circuit, with the Al test pad on a Si (001) wafer. The HWHM (half width half maximum) image of the
substrate consists of a 500 nm layer of SiO2 thermal pattern, with a specific measurement path defined by
oxide grown on the wafer. The following results were the dotted line. The lighter regions correspond to more
taken from one of the earliest applications of this disorder. Below Fig. 95b is a linear plot of the HWHM
technique by Ajito et al. [25]. Below Fig. 94b is the for the Si Raman band, showing that there is a sharp
Raman intensity image of the structure where the maximum at X corresponding to a disordered (da-
maged) region.
Figure 96a is the Raman strain image of the pattern.
Brighter areas correspond to positive shifts (compressive
stress) while dark areas relate to negative shifts (tensile
400
stress). Below Fig. 96b is a set of readings over a dotted line
across the pattern. Positive compressive stress peaks are
clearly seen at B and F. Subsequent AFM images revealed
that the Al layer was misaligned by approximately 500 nm
to the right causing the increased strain. This result
µm

200

400

µm

0
0 200 400
200
µm
a

400
0
0 200 400
µm
µm

1.80 2.00 2.20 2.40 2.60

HWHM/cm−1
200 a

3.0

X
2.8
HWHM/cm−1

2.6
0
0 200 400
µm 2.4

2.2
0 2000 4000 6000 8000 10 000
Raman intensity, counts/0.025s
2.0
b 0 100 200 300 400 500
distance, µm
Fig. 94 b
a White light image of an Al test circuit on a Si (100) wafer. The
dimensions are shown. The dark regions correspond to the bare Si Fig. 95
substrate and the light regions are the reflective regions of Al a Image of the variation in the HWHM of the Si 520 cm
1 peak in the
b Raman intensity image of Si taken over the same area as a same area as Fig. 97
The image has been processed after subtracting the background signal. b Cross-section profile
The spectral bandpass was set at 500–540 cm
1, and the background The sharp maximum at point X corresponds to the damaged region in
was taken at 600–640 cm
1 for subtraction Fig. 97a
Data courtesy of K. Ajito [25] Data courtesy of K. Ajito [25]

IEE Proc.-Sci. Meas. Technol., Vol. 152, No. 6, November 2005 303
 400
µm

200

0
0 200 400
µm

−0.20 0 0.20 0.40 0.60

relative Raman shift, cm−1
a

1.5 Rama 0.2

B
relative Raman shift, cm−1

n shift,

1.0
F 10
0
cm

A
−1

0.5 C D E 5
G
−0.2 0
0
− 10
−5 −5
0
−0.5 mic
0 100 200 300 400 500 ron 5
− 10
distance, µm 10
b b

Fig. 96 Fig. 97 Stress map in Si induced by LOCOS stripes orientated

a Raman strain image of the sample indicated by the relative shift of along (110)
the Si 520 cm
1 peak. The area of the image is the same as in Fig. 98a a White light image of the sample; dark regions correspond to oxide
Black regions correspond to the lighter Al regions in Fig. 98a and light regions correspond to bare silicon. The 20 mm  20 mm
Brighter areas correspond to positive peak shifts and compressive square inset is the investigated area
stress b Area map of the centre of the Stokes 520 cm
1 peak. The data are
b Cross-section profile taken at the same location as Fig. 99a courtesy of E. Bonera [187]
A positive maximum at point B occurs at the left-hand side edge of the
Al circuit after a small minimum at point A. Similar combined pairs of
maxima/minima are found on the wafer at points E and F. Data
courtesy of K. Ajito [25] Raman shifts. Unfortunately this argument does not take
into account the advantages of the higher efficiency of the
single monochromator systems, coupled with curve fitting
emphasises the potential use of combined Raman-AFM routines, and the ability to take readings whilst scanning
systems for higher spatial resolution analysis. The work of much faster.
Shen and co-workers [196] on specified test patterns, using The high efficiency system was used to test whether a
an apertureless NSOM on test samples to demonstrate faster speed of acquisition was possible without loss of
improved spatial resolution, is a good example. resolution. Measurements were made on samples having a
In Section 4.4 the stability (repeatability) of a typical test pattern of localised oxidation of silicon (LOCOS).
Raman system over several hours using the silicon 520 cm
1 Figure 97a is an optical image of an array of LOCOS lines
Raman line was found to be better than 0.05 cm
1, and the orientated along (110) in the silicon reference system. Light
accuracy of wavenumber calibration was 0.2 cm
1. Bonera areas correspond to bare silicon while the darker regions
et al. [198] built on these basic measurements to explore relate to those regions having a 0.5 mm thick layer of SiO2.
further the full system resolution and strain resolution using A line (point by point) was made in Fig. 97b showing the
silicon device structures. It has been argued by some Raman shift/strain using excitation at 633 nm. The max-
potential users that the higher resolution multiple grating imum stress at the centre of the active areas was calculated
systems, where system resolutions of 0.1 cm
1 and lower to be 150 MPa assuming a uniaxial stress along (110). The
can be obtained, are necessary for quantification of strain shifts were recorded after Lorentzian fitting of the 520 cm
1
304 IEE Proc.-Sci. Meas. Technol., Vol. 152, No. 6, November 2005
band. A theoretical analysis of the spectral curve fitting by setting the grating position at 0 cm
1, with no saturation
showed that the worst case 0.05 cm
1 error only appeared of the CCD detector. The 476.5 nm plasma line of the Ar+
when the S/N ratio for the band was less than approxi- laser was also followed Fig. 98d. It has been found that this
mately 20, which is well below its real experimental value. can change position in an unpredictable way. Plasma lines
This resolution is therefore comparable with that quoted for provide optical artifacts particularly when their optical path
multiple grating spectrometers. However the latter have involves reflection on to a non-uniform patterned surface
added disadvantages in that a narrow slit width must be such as a microelectronic device. The presence of such
used, and over the longer measurement times required they artifacts affects the position of the main silicon peak, and
are susceptible to further calibration drift due to thermal hence the reliability of measurement. Finally, Fig. 98e
and mechanical movement. shows a qualitative plot for edge induced stress (see the
Figure 98a shows the cross-section of a LOCOS Figure text for the tensor representation) based on Hu’s
configuration used to confirm the theory. The pattern model [199] for edge induced stress. It is clear that the ability
consisted of four 4.0 mm active area stripes of silicon to scan across the excitation laser line at zero Raman shift is
separated by three 0.25 mm oxide stripes. A 488 nm Ar+ a bonus. It provides a method to measure simultaneously
laser excitation was used, where the penetration skin depth the differences in intensity of equivalent Raman bands. As a
is approximately 0.48 mm. It is seen Fig. 98b that the result, non-contact measurement of semiconductor wafer
detectable stress occurs only at the interface between the surface temperature during processing can be performed
bare silicon and the larger edge semiplane of oxide. The using (1) from Section 2.
variation in position from point to point was found to be
better than 0.05 cm
1. 9.7.2 Line shape and mapping of amorphous
The equivalent Anti-Stokes peak was also followed and polycrystalline silicon using variable
Fig. 98c at the same time. With the filter based high excitation wavelengths: Amorphous-polycrystalline
efficiency systems, it is possible to simultaneously take silicon: There are several instances where various forms of
Stokes and Anti-Stokes measurements across the laser line, disordered silicon (e.g. amorphous to polycrystalline) have
been purposely manufactured. This paper has earlier
(Section 6.3) touched on disordered silicon phases when
discussing Raman-NSOM-AFM measurements [86, 87]. It
was found that differing disordered phases of silicon could
a be generated at the centre of nanoindentations and
0.4 scratches in the surface of crystalline silicon. There are
compelling economic reasons for the development of
B deposition technologies for amorphous silicon in the
0.2
manufacture of large area solar cells and flat panel displays
(FPDs). For active matrix liquid crystal displays the
0 deposited amorphous silicon is processed to eventually
make thin film transistors. The thin 50 nm amorphous layer
is deposited on a quartz or glass substrate, and a laser or
−0.2 A
thermal process is then used to anneal the film and convert
it to a uniform polycrystalline form. The spectra of the
−0.4 various forms of silicon from the amorphous to crystalline
b
1 state are now well established and require no further
relative shift, cm −1

comment here. The assignments of the spectral bands are

0 well established, for example References [200, 201]. For
strict control of the annealing process it has been necessary
c −1 to monitor the power levels of the laser in order to ensure
even and efficient crystalline conversion. This has also been
1
applied to laser or thermal annealing of surface damage
0
caused by ion implantation doping procedures e.g. boron.
d
Throughout these conversions (amorphous to polycrystal-
line) the changes in spectra can be relatively subtle but are
adequate for control purposes. The laser anneal power
intensity over the device surface area requires stringent
control to prevent ablation (power is too high) and non-
homogeneity (power is too low).
Figure 99 shows several spectra from 100 cm
1 to
900 cm
1, covering the critical region around the
e 520 cm
1 line for crystalline silicon (c-Si) [202]. Amorphous
20 30 40 50 silicon (a-Si) shows a reduction in frequency and intensity of
position, µm this peak with some broadening. The position of the
520 cm
1 band moves down to 475–480 cm
1. There is
Fig. 98
further structure near 190 cm–1, between 300 and 400 cm
1,
a Cross-section of the LOCOS sample
b Relative shift of the Stokes peak and at 600 cm
1. For polycrystalline silicon (p-Si), there are
c Relative shift of the simultaneously collected anti-Stokes peak smaller but still significant changes from c-Si, i.e. movement
d Relative shift of the simultaneously collected Ar+ plasma line of the peak slightly below 520 cm
1. There is some non-
e Qualitative plot of Hu’s [188] model for edge-induced stress for the symmetry of the band, with a shoulder at the base of the
syy0 (thick solid line), sz0 z0 (dotted line), and sy0 z0 (thin solid line) band around 490 cm
1, and small structures near 200 cm
1
components in the reference system. The representations x0 ¼ (110), and 600–605 cm
1. The additional bands (looking at the
y 0 ¼ ð110Þ, z0 ¼ (001) hold. Data courtesy of E. Bonera [187] transition in reverse) as the structure moves through

IEE Proc.-Sci. Meas. Technol., Vol. 152, No. 6, November 2005 305
 2500 [205]. This energy value is well above the E1 energy gap. The
c-Si scattered intensity decreases owing to the extremely small
2000 a-Si (as-deposited) skin depth and sampling volume. No evidence of the
1500
a-Si (recrystallised)
disorder bands below 500 cm
1 was observed, because the
counts

p-Si
mc-Si
prime excitation concerns only the allowed resonance band.
1500 Section 9.7.5 on SiGe alloys demonstrates that the ability to
measure solely the strain in the silicon cap layer, without
500
interference from the SiGe graded substrate layer, is an
0 extremely important characteristic.
300 400 500 600 700
Raman shift (∆ cm−1) Ion implantation and annealing: In the early paper by
Nakashima and Hangyo [196] there is some discussion on
Fig. 99 Composite plots of Raman spectra for the range of a-Si,
the characterisation of ion implantation and the laser/
p-Si and mc-Si transitions
Data courtesy of Hugh Gotts, Analytical Services Group, CA, USA thermal annealing processes that are in use. The laser is used
to anneal the surface damage caused by the implantation in
the polysilicon. Raman intensity profiles of laser recrystal-
lised polycrystalline islands were measured, using polarisa-
polycrystalline silicon to the amorphous state are the result tion measurements. From these measurements, the growth
of breakdown of crystalline order. The selection rules are of crystalline grains could be observed. The work was
broken with additional modes now allowed. As the originally carried out using lengthy point-mapping, over the
crystallite size of p-Si increases, the spectrum becomes less full area of the test island. In the Osaka laboratory the
easy to distinguish from c-Si. present authors carried out global imaging of these islands,
A 2-D Raman intensity profile at 520 cm
1 of the silicon and obtained 2-D images of the intensity profiles within
surface gives an excellent representation of the polycrystal 30 s, with clear indications of crystalline quality and grain
uniformity. Small changes in the intensity and line shape size. A line scan along the islands would have been just as
will also give a measure of the spatial size of the crystallites. effective to give the spectra at specific positions for
The line scan facility in the new Raman system provides the subsequent analysis.
means (a) to reduce the laser power in the Raman beam by
spreading it over a line and hence reducing the risk of Doping levels from ion implantation can be determined
burning during measurement, and (b) to increase the speed from analysis of the silicon spectrum from 200 cm
1 to
of scanning the wafer surface. The spectrum from each 1 mm 1000 cm
1 using non-contact Raman methods. There are
segment of the line is stored for subsequent detailed analysis several examples in the literature where Raman measure-
of the progress of the process, using different line shape and ments have been carried out by industrial research groups
curve fitting templates. on samples subjected to both low and higher implant doses,
When polycrystalline silicon (e.g. 50 nm) is deposited on followed by annealing processes. Most of the studies used
crystalline silicon, the Raman signal using Ar+ excitation at traditional double or triple monochromator systems. While
488 nm will give a spectrum primarily of c-Si, since the large changes in the spectra were evident, most of the
penetration (skin) depth probed is approximately 480 nm. published descriptions (for example Reference [206, 207])
The ability of this newly designed system to rapidly switch concern simply the details of the different phonons involved
excitation wavelengths allows measurements to be taken and the changes in the various bands (i.e. Raman shifts,
with UV excitation, e.g. at 364 nm for Ar+, where the intensities, half-widths). Several of the bands, particularly in
penetration depth (approximately 12 nm) is very small e.g. the 300 cm
1 to 400 cm
1 range and 600 cm
1 to 1100 cm
1
[203]. Sampling will involve only the surface layer. A further range, tend to be rather noisy owing to the poorer system
advantage of this excitation is to give resonance Raman efficiency, although the changes appear to be significant.
since this energy (3.4 eV) is close to the critical E1 energy Little attempt appears to have been made to date on
band gap of 3.72 eV in Si. However the early work of extensive curve fitting. However if these effects are to be
Renucci et al. [204] has indicated from analysis of the used for monitoring and control of doping and annealing
second order 2TO spectra near 980 cm
1, that the true processes further careful analysis is called for in order to
resonance energy gap condition is not E1 but either E0 obtain better quantification.
(4.1 eV) or E2 (4.3 eV). The cross-section of the Raman
signal from the 520 cm
1 band would be expected to be Nanocrystalline and porous silicon: The applicability of
higher under resonance or pre-resonance conditions, and Raman methods has been described for nanocrystalline
strain measurement is specific only to the surface p-Si layer. silicon [208] and silicon nanowires [209]. The References are
Calculations of the laser focusing and sampling volumes a selective representation of the type of work being carried
in silicon can be carried out in accordance with the earlier out. These measurements build on the earlier interpretations
discussion on confocal design of Baldwin et al. [37] and of the origins of the disordered silicon spectra. It is possible
Everall [50]. These confirm the spatial resolution of the that in future Raman may become a useful adjunct to the
measurement, and underline the importance of the UV production process of these new materials. It is worth
system design and specialised selection of optical compo- noting that the use of the extended synchronous scan
nents and coatings. Results from initial test devices often technique has proved useful in the continuous measurement
show unexpected compressive and tensile stress variations in of Raman and luminescence bands in porous silicon [210].
the vicinity/edges of metal (e.g. Al) and insulator (e.g. oxide Strain and phase differentiation in silicide formation: Raman
and nitride) overlays. These can be fitted to the appropriate has been used to control and monitor the annealing and
models with redesigns made to correct any problems caused phase differentiation of deposited metals and their resulting
by the earlier unquantified stress profiles. silicides (e.g. TiSi2, CoSi2) [202, 211–214] during the
The authors also made additional measurements on annealing processes. These materials are used for metallisa-
silicon using excitation by the HeCd line at 325 nm, in tion of gates, sources and drains, and as local interconnects
collaboration with Martin Kuball, Bristol University, UK in ULSI. It is important to follow the structural changes in
306 IEE Proc.-Sci. Meas. Technol., Vol. 152, No. 6, November 2005
the forms of silicide during the two anneals in order to IC522
GaSb 6.5 nm
determine uniformity and the degree of nucleation of the AISb 800 nm
higher temperature phase. TEM analysis is inadequate for GaAs GaSb 400 nm
GaAs substrate
such control owing to confusing similarities in the electron 47 min
diffraction patterns for both phases (e.g. the C49 and C54 40 min
phases of TiSi2). X-ray diffraction too can fail to detect the
presence of the intermediate orthorhombic base C49 phase 30 min

intensity, a.u.
GaSb
during annealing. The unequivocal Raman spectra provide 20 min
unique signatures for the critical phases. The fast line
mapping and/or real global imaging techniques rapidly 18 min

highlight non-uniformities and grain dispersion due to 10 min

uneven phase nucleation. Results have been obtained a-Sb 5 min
whereby the intensity of the silicon 520 cm
1 band from
below the silicide layer has been used to estimate film 2 min
thickness. The Raman data have also been correlated with GaSb AISb
before etching
the silicide sheet resistance and the degree of doping in the
silicon substrate. Systems have also been used for mapping 100 150 200 250 300 350 400 450
of the strain on the silicon (or polysilicon) substrate at the Raman shift, cm −1
edges of the deposition [214].
Fig. 100 Raman spectra for a sample of 0.8 mm AlSb sandwiched
between a 6.5 nm GaSb cap and a 0.4 mm GaSb buffer etched in
9.7.3 Surface cleaning and on-line etching photoresist developer MF319
See text for explanation
control: There are several examples in the literature Data courtesy of R A Stradling, Imperial College, Univ. of London,
where Raman systems have been used to monitor and UK
control specific etching and cleaning procedures in semi-
conductor manufacture. Pelletier et al. [215] have success-
fully used systems to monitor the composition of the SC-1
cleaning liquid used in wafer cleaning baths to eliminate by photolithography. From these results the etching times
extraneous particulates. There are advantages in the use of for the different layers were confirmed, and the method was
Raman since it gives highly specific analysis and can operate then found to be applicable to the other heterojunction
non-destructively in hazardous environments using fibre structures. The presence of the amorphous Sb (typically 30
optics. Various on-line etching methods have also been monolayers) originates from the breakdown of the AlSb
proposed where the confocal optics allow measurement of with some oxidation of the Al. A more detailed description
the semiconductor material surface to be made through the of the chemical processes involved in the etching of GaSb
etching medium, including plasma etching [216]. and AlSb has been described [217]. The unique feature of
As far as is known, the first demonstrated example of in Raman monitoring and control is that the etching process
situ real-time monitoring of the selective etching of can be stopped at any part of the operation. Additional
semiconductor structures is described in the work of Gatzke surface area and uniformity checks can be carried out.
et al. [217]. The technique was applied to GaSb/AlSb/InAs
heterostructures to allow accurate timing of the etching as 9.7.4 Contaminant detection: Wafer particle ana-
well as confirmation of the chemical composition of the lysis: Defect review stations for accurate classification of
exposed layer at any given time. These materials are of defects and contaminants on bare and processed wafers are
interest for resonant tunnelling diodes, high electron field a necessary feature on most production lines. Based on the
mobility transistors and quantum wires. A further example new developments, a Raman system has been successfully
of the importance of heterostructure identification is given coupled to a normal microscope for defect review purposes.
for Si/Ge alloys in Section 9.7.5. For GaSb/AlSb/InAs The system design meets clean room and operator laser
heterojunctions a specific etching requirement is to selec- safety standards, and is designed to download defect
tively remove the antimonides to allow contact to the InAs. coordinate files. A number of results and a brief system
Figure 100 describes the etching sequence on one of several description have been given by Ballast et al. (AMD) [211].
test heterostructures, with the compositions and thicknesses The value of this approach was highlighted by the ability of
shown at top right. The photoresist developer MF319 was the Raman system to identify contaminants that were not
used and the Raman spectra taken through the etchant. visible under normal microscopy at a delamination defect
Each exposure lasted 60 s and the Raman excitation was at site. It was found that the photoresist removal and wafer
514 nm (Ar+). Remarkably, it proved possible to obtain cleaning procedures prior to film deposition were inade-
clean spectra even though the agitation of the stirring can quate, leaving a residue that caused the delamination.
lead to random changes in the beam focus position. Systems have also been designed in collaboration with
Without agitation the etching came to a halt. JEOL (Japan) [218] to include standard positional KLA-
Starting from the bottom of Fig. 100, and the beginning Tencor coordinate files, and control software to drive the
of the etching process, the GaSb cap of 6.5 nm disappears 300 mm wafer stage. The microscope images the particle
within the first 2 min. The AlSb layer is visible in the initial and local metallisation to allow analysis of both the particle
spectrum prior to etching since the penetration depth in size and shape, as well as identification of the chemical
GaSb is 8 nm (at 514 nm excitation) and greater than the species and crystalline structure. This system has been
cap thickness. The broad band near 150 cm
1 was identified supplied to several leading commercial production lines,
as amorphous Sb. After 20 min this disappears and the and there are numerous examples of identification of critical
GaSb combined bands at 230 cm
1 and 240 cm
1 reappear. sub-micrometre and extraneous organic/inorganic particu-
After a further 20 min the LO phonon band characteristic lates. Their identification allows process engineers to trace
of the GaAs substrate appears. The measurement depths the source of the contamination and improve production
were confirmed with a Dektak stepper on samples patterned line yield. In one instance the JEOL Wafer Particle Analyser

IEE Proc.-Sci. Meas. Technol., Vol. 152, No. 6, November 2005 307
(WPA) [218] was applied to particulate spectra consisting of particulates, and does not offer rational interpretation of
carbon based atmospheric pollutants. The source was their origin.
identified as diesel particles from goods inward vehicle
exhaust. The clean room filtration was consequently
improved and the yield returned back to normal.
General cleanliness procedures have been continuously
improving on fab process lines. The integrated circuit size
has reduced to 90 nm and is likely to be scalable to 65 nm.
This means that the critical particulate size for detection will
fall below the horizon of the standard Raman optical
microscope. However in the larger scale IC requirements for
liquid crystal displays (LCD) and organic light emitting
diodes (OLED), the critical contaminant size is still
compatible with the optical microscope design and installa-
tions have continued. There are stories of lost production
amounting to more than the cost of a Raman system.
Raman has the added capability of identifying polymer
and organic contaminants. A semiconductor processing
spectral library contains many of the most commonly
used materials in semiconductor processing technology, and
this was partially generated in collaboration with Hugh
Gotts, Analytical Services Group, Santa Clara, CA, USA
[202].
SEM – structural chemical analyser (SCA): The recent
development of combining an SEM with a Raman analyser
helps to overcome the spatial resolution requirements. The
spatial resolution of the SEM is at the nanometre level, and
the advantage of Raman in identifying organics and
chemical structure is compatible with the SEM’s ability to
detect and image elemental composition. Further measure-
ment of PL and CL at the same point on the sample also
provides imaging impurity and carrier concentrations
particularly in larger band gap semiconductors such as
GaN and associated alloys (used for green light emitting
diodes and lasers: see Section 9.7.6). An SEM–SCA system
may also incorporate a cryogenic stage for liquid nitrogen
measurements during product development exercises.

Since small sub-micrometre particles may be damaged or

become charged by the electron beam, low accelerating
voltages are used to image the sample. If the particles
become charged with respect to the substrate, they may be
repelled and lost from analysis.
Figure 101a shows part of the electron image of a sample
taken from a batch of TTL integrated circuit wafers. Several
contaminants were introduced to illustrate the measurement
capabilities and to simulate defects that would produce
poor yield and device failures prior to passivation [219].
Figure 101b shows Raman spectra taken from the SCA.
Since the silicon spectrum is strong, this is subtracted from
the particulate spectra. The top spectrum corresponds to a
remnant of PMMA photoresist. The middle spectrum
reveals the polymer PTFE. Following analysis of the
morphology of the particle from the SEM image it is
interpreted as wear residue from a worn bearing. Simple
optical microscopy failed to detect the PMMA particle,
since the particle is optically transparent and the microscope
could not resolve sub-micrometre morphological detail. The
lower spectrum comprises a mixture of silicon carbide,
diamond and amorphous carbon. A small silicon line at
520 cm
1 deriving from the wafer is also present. The Fig. 101 Raman and SEM images of a TTL device with a
contaminated surface
particle may originate from the lapping process and suggests
a Secondary electron image of the TTL device and particles
that the post-process cleaning stage needs revising. Figure b Raman spectra of particles acquired using Renishaw’s SCA
101c is an examination of the latter particle using the c Typical EDS spectrum of particles
SEM–EDS combination. The Al and O signals originate The Raman measurement identifies organic compounds and the
from the metallisation and passivation. The SEM–EDS structures of diamond amorphous carbon and silicon carbide, while
method alone failed to fully identify and characterise the the EDS technique detects the elemental composition

308 IEE Proc.-Sci. Meas. Technol., Vol. 152, No. 6, November 2005
On-line growth monitoring: The ability to introduce Raman silicon substrate wafer. The lattice constant of pure Ge is
probes into UHV chambers has led to some activity in 4.2% larger than that of Si, but by gradually increasing the
designing systems for monitoring molecular beam epitaxy Ge content the lattice mismatch can be increased without
(MBE) or metal organic vapour phase epitaxy (MOVPE) the introduction of too many defects. Beyond critical
processes. Developments have been made on experimental thicknesses, misfit/threading dislocations form to relieve the
cluster tools for systems with JEOL (Japan) [218], where structure. Nevertheless the defect density has tended to be
Raman measurements can be made in a side cool-down too high, so a further SiGe layer is grown on top, with the
chamber. Other examples of earlier investigative materials Ge content and lattice parameter matched to the final
work of this type, largely for 3–5 or 2–6 semiconductors, are graded composition immediately below. This gives a fully
given [220, 221]. Growth rates and compositions of relaxed and relatively defect-free layer upon which the final
heterostructure layers can be monitored. Competing strained silicon cap layer is deposited. This final silicon cap
technologies are spectroscopic ellipsometry (SE) combined layer aligns itself with the larger spacing of the SiGe atoms
with laser light scattering (LLS) [222], and secondary ion below and is stretched to give the desired effects. With
mass spectroscopy (SIMS), although the latter is not used in appropriate control the degree of strain induced will depend
real time. The accuracies of thickness and composition on the Ge content. For s-SOI structures, a thin layer of
measurement will be discussed in the following Section. insulator is situated below the strained silicon. This can give
further advantages in terms of device speed and lower
power requirements.
9.7.5 Strained silicon, and SiGe alloys: the
next generation devices: The next generation of Polarisation Raman measurements on SiGe alloys for
silicon devices is making use of deliberately induced strain. composition and strain measurement: R.A. Stradling’s group
The International Technology Roadmap for Semiconduc- at Imperial College, University of London, [223, 224] was
tors has identified the development of three strained silicon encouraged to use their installed Raman system to examine
geometries: whether strain and compositions could be quantified from
the Raman spectra of a range of Si1
xGex epitaxial alloys
1. Bulk strained silicon, where the strain is induced (materials supplied by C. Pickering, QinetiQ, UK). They
following deposition of a cap layer of Si3N4. This improves found that the metrology for the determination of
the drive current in NMOS devices by 10% and leakage has composition from the first-order Si–Si, Ge–Ge and Si–Ge
been reduced such that 45 nm processing becomes possible. optical phonon modes in the alloy spectra suffered from
The development is economically beneficial in that it has distortion of the intensities and lineshapes by various
been based on present processing technologies and has subsidiary features in the region 340–500 cm
1 arising from
required only a small additional processing step. various disorder and clustering defects in the layers. This
2. Strained silicon on SiGe on insulator (SGOI). affected the ability to use true Lorentzian curve fitting. They
3. Strained silicon on insulator (s-SOI). demonstrated that the effects of these features could be
minimised using specific Raman polarisation measurements.
The latter two technologies require higher investment in For example the subtraction of the first-order forbidden
processing and control. The interest in strained silicon spectrum from the allowed Raman spectrum obtained with
devices has followed the realisation that specific controllable parallel-polarisation geometry gave less distorted Si–Si and
strains can split the degeneracies of the conduction and Ge–Ge mode bands. Also measurements in a cross-
valence bands. Electron and hole mobilities can be polarisation geometry caused the background to be
dramatically improved. Whilst this effect has been known suppressed. It is proposed therefore that the earlier
for 40 years, it has only now been taken up, as more radical equations based on the phonon band positions for bulk
processes are required to reduce feature sizes down below alloys may have additional errors for heterostructures that
65 nm and simultaneously improve low current leakage and are subject to strain effects [225]. This later analysis, based
speed. There have been various estimates of these improve- on integrated phonon intensities and line positions, per-
ments depending on the processing technologies employed. forms better in both strained-layer and bulk alloys.
For example Amber Wave, Texas Instruments and
International Sematech have demonstrated a 90% improve- This project has further confirmed the advantages of
ment in electron mobility with strained silicon on SGOI. using higher efficiency Raman systems with fast curve-
The result showed a considerable improvement in the fitting to give improved resolution of strain. The polarisa-
specification for HFET devices for RF applications (e.g. for tion options have also allowed the determination of
mobile phone transmitters and receivers). The main composition. Figure 102 describes the curve fitting of the
emphasis however remains on logic devices for the chief (a) unsubtracted and (b) subtracted spectra for x ¼ 0.141
device manufacturers, where the cost and complexity and x ¼ 0.217 samples. Note the complexity of the spectrum
(commercial risk) of the new processes are weighed against below 500 cm
1 in such alloys, which has led more
device performance improvements in the shorter and longer conservative workers to specify that traditional high
terms. resolution Raman systems (but with poor optical efficiency)
There have also been various proposals on the manu- should be used to ensure improved peak separation. Since
facture of SiGe for direct use in active devices. These range several bands are incorporated into the curve fitting and
from improved RF devices to optoelectronic waveguides. analytical solution this is unnecessary, and fast data
Currently there are large numbers of defects present in the processing with a higher signal/noise and stability system
alloys, but it is only a matter of time before further is preferred. First order derivative spectra can also assist in
improvements in growth techniques will occur. For control spectral interpretation.
of the processes, Raman is being increasingly used since it Fast Raman mapping techniques have also been used to
can measure both strain and Ge composition in the alloy. explain the characteristic cross-hatching surface morphol-
Preparation of a strained silicon cap layer on SiGe can be ogy of SiGe films. It was found that the cross-hatching was
achieved by depositing a series of thin graded (increasing Ge related not to compositional variations, but to stress
composition) SiGe epitaxial layers, on top of the bulk undulations within the SiGe layer [226].

IEE Proc.-Sci. Meas. Technol., Vol. 152, No. 6, November 2005 309
2000
a undoped
Si 0.7 Ge 0.3
E2 (high)
569 cm −1
1500 Raman spectrum
curve-fitted spectrum
E1 (TO)
1000 559 cm −1

A1 (LO)
A1 (TO) 733.4 cm −1
500

light intensity, arb. units

532 cm −1

E1 (LO)
0 742 cm −1
parallel
a
0 perpendicular
2000 b si-doped

1500

sapphire A 1g
sapphire E g
418 cm −1
742 cm −1
1000

parallel

perpendicular 0
500

400 600 800

0 Raman shift, cm −1

b
Fig. 103 Raman spectra of epitaxial GaN grown on sapphire by
MOCVD. Parallel and perpendicular detection geometries are used
Fig. 102 Polarised Raman spectra of Si0.7Ge0.3 corresponding to a Undoped GaN
z(x,x)z scattering geometry b Silicon doped GaN
a Curve fitting of the unsubtracted z(x,x) z With increased silicon doping the A1(LO) peak intensity decreases.
b Curve fitting of the subtracted z(x,x) z
z(x0 ,x0 ) z Small peaks due to the sapphire substrate can be minimised by
adjustment of confocal focusing

UV excitation and measurement of strain in the Si cap layer

of SGOI and s-SOI structures: Raman measurements on the interest here is the decrease in intensity and change in
thin strained silicon cap layer require the flexibility of rapid frequency of the A1(LO) line with increased doping. The E2
system switching to the UV laser option. With standard peak is affected only by stress (frequency shift) and
systems, and excitations at 488 nm and 514 nm, the spectra crystalline quality (band half-width), so there is the potential
obtained are largely of the SiGe layer(s) below. Use of the to separate the different effects of stress and free carrier
364 nm Ar+ laser or HeCd 325 nm excitations will only concentration in one series of spectra. Firstly we shall
sample the surface silicon layer, owing to the reduced skin describe how the effect arises.
depth. Polarisation Raman measurements will then give the
desired strain (uniaxial or biaxial), from the shift of the
520 cm
1 line, without subsurface interference. Longitudinal optical phonons interact strongly with the
free carrier plasma when both frequencies are close. The
coupling and creation of these modes (phonon-plasmon
9.7.6 Gallium nitride and alloys: This Section modes) can lead to upper (L+) and lower mode (L
)
describes a series of recent collaborative projects with the branches shifted to higher and lower frequencies respec-
group led by Martin Kuball at Bristol University, UK. tively. The LO phonon-plasmon effect was first demon-
GaN and alloys with InN and AlN have excited consider- strated in GaAs by Mooradian and Wright [227]. They
able interest for use as blue-green light emitters (lasers and produced carrier concentration versus frequency (cm
1)
diodes) and heterostructure field-effect transistors. The curves for the plasmon–LO phonon interaction covering
Raman techniques employed also relate to other 3–5 carrier concentrations from 1017 to 1019 cm
3. Subsequently
semiconductor materials. Yuasa et al. [228] extended this work to AlxGa1
xAs, and
carrier concentrations down towards 1016 cm
3. Kozawa
Carrier concentration measurements in GaN and other 3–5 et al. [229] observed the effect in GaN for carrier
materials: Figure 103 shows polarised spectra from doped concentrations over the range 1017 to 1019 cm
3. Subsequent
and undoped GaN epitaxial layers grown on a sapphire results were combined with the above data by Kuball [230].
substrate. The high purity GaN has an A1(LO) phonon Thus it is possible to map carrier concentrations over an
band at 733.4 cm
1. Small background peaks from sapphire area of the surface of a sample, using curve fitting to obtain
are observed, but can be reduced by improving the the peak positions for the lower intensity and broader LO
confocality such that the beam just touches the surface. phonon–plasmon modes. Extensive measurements have
The A1(TO), E1(TO) bands are only allowed if disorder- also been carried out on InAs samples with non-contact
activated, and probably originate from the high density of Raman 2-D carrier concentration mapping [231].
defects consistent with early GaN heteroepitaxy. The strong For silicon doped GaN, Ponce et al. [232] used the
E2 characteristic line of GaN is marked. The chief point of intensity of the A1(LO) phonon to directly image the carrier
310 IEE Proc.-Sci. Meas. Technol., Vol. 152, No. 6, November 2005
concentration across the area of an epitaxial layer grown on 20 mm to 50 mm was selected to specifically examine the
a sapphire substrate using polarised 514.5 nm Ar+ excita- effects of surface crystal structure on carrier concentration
tion. A hexagonal crystallite with dimensions of the order of distribution. Raman global imaging (see Section 3.2.2 E)
was used, with the A1(LO) band at 734 cm
1 selected for
the bandpass filter, and the background subtracted by
setting the bandpass filter off the main peak at 785 cm
1.
Although the A1(LO) band also shifts in frequency, it
remains within the bandpass of 20 cm
1.
Additional normalising and flat-field correction proce-
dures were carried out. The images are shown in Figs. 104a–c,
where it is seen (c) that at the centre of the hexagon there is
a high signal indicative of low carrier concentration values,
and at the facet edges the dark regions imply a high carrier
concentration. The results also enabled further interpreta-
tion to be made of the incorporation of silicon and oxygen
during growth.
Strain and annealing in GaN and alloys at high temperature:
High temperature annealing of 3–5 nitride heterostructure
device structures is necessary for the reduction in defects
after doping by ion implantation. It is also important to
activate Mg acceptors in MOCVD material, and create a
good ohmic contact. Both strain and carrier concentration
a
will change depending on the implantation and annealing
process. The former can be detected by following the shift
of the E2 phonon frequency of GaN at a rate of 2.9 cm
1/
GPa, while the carrier concentration is measured from the
A1(LO) frequency, as above. These measurements illustrate
the advantages of the system being able to switch rapidly
between UV and visible laser excitation.
Details of the test MBE heterostructure device structure
are given in Hayes et al. [233], and measurements were
taken after anneals from 800 to 13001C in different
atmospheres of nitrogen, oxygen and air for 20 min. Water
vapour was introduced during some anneals.
Figure 105 shows the frequency of the strain-sensitive E2
phonon of GaN after high temperature processing recorded
using 514 nm excitation (the inset shows the Raman
spectrum of the sample before annealing). It can be seen
that the shifts are largest for annealing in oxygen and water
5 µm

b 572

air E 2 GaN
Raman signal, a.u.

O2
N2
sapphire
571

A 1(LO) GaN
E2 frequency, cm −1

O2 + H 2 O

500 600 700 800

570
wavenumber, cm −1

569

before annealing
568
800 900 1000 1100 1200

c annealling temperature, ˚C

Fig. 104 Raman scattering direct images of a hexagonal feature on Fig. 105 Frequency of the strain-sensitive E2 phonon of GaN after
the epitaxial GaN surface grown on the (000-1) N-terminated face high-temperature processing in various annealing atmospheres, using
a Scattering at 735 cm
1 514 nm excitation
b Background for subtraction at 765 cm
1 The inset confirms the Raman spectrum of the sample before
c Flat-field image obtained by subtraction annealing

IEE Proc.-Sci. Meas. Technol., Vol. 152, No. 6, November 2005 311
vapour and air, while nitrogen has little effect. This suggests 220
that the incorporation of oxygen and/or possibly hydrogen 120 S
200

current/mA
100
is responsible for the strain. 180 80 G D
The authors, in earlier work [234], had showed that with 60

temperature rise/ ˚C
160 40
325 nm excitation on GaN, sharp resonance Raman peaks 20 S
of many orders could be detected. The samples were 140 0
0 20 40 60 80
supplied by the Nottingham University group, who later 120 source-drain voltage/V
published further data on the topic [235]. Kuball et al. [236] 100
took advantage of this result and, since shifts of higher 80
order peaks have a higher sensitivity, demonstrated a very

source

drain
gate
60
small (but significant) decrease of 1  1017 cm
3 in carrier
concentration after the 10001C anneal in oxygen by 40
measuring the Raman shift of the third order A1(LO) 20
phonon. Further information was also gained regarding the 0 1 2 3 4 5 6 7
position/ m
incorporation of oxygen into the lattice and its role in
producing strain, from analysis of the photoluminescence Fig. 106 AlGaN/GaN HFET device temperatures in the source-
spectrum captured in tandem with the Raman spectrum. drain opening, as a function of applied source-drain voltage at zero
A further advantage of having several resonance orders gate voltage
in a spectrum is that a selection can be made on the ’ 15 V(1.8 W), & 30 V(3.0 W), 45 V(3.75 W), J 55 V(3.9 W), 
optimum order for strain measurement and fluorescence. D 65 V(4.0 W), m 75 V(4.1 W). Heat sink temperatures were subtracted
can be avoided. The other option is to use a different from the experimental data. Inset (left) is the experimental I–V curve at
exciting wavelength. zero gate bias above the knee voltage, and inset (right) is a schematic
of the device layout with the region of the line-scan denoted. Data
Measurements on GaN and related alloy structures, e.g.
courtesy of Martin Kuball, Bristol Univ., UK
AlGaN and AlN, have proved useful in determining the
points at which degradation occurs during high temperature
annealing [230, 237]. Interpretation of the high temperature Figure 106 displays the measured temperature profiles of
spectra taken to 12001C enabled the intermediate structural the device at varying source-drain bias. The I–V curve
stages to be identified, and gave evidence of structural above the knee voltage is given in the inset. The temperature
degradation. This work was carried out using the 244 nm profiles show large electric fields near the gate contact
(UV) laser excitation line. The combination of this (Fig. 107). The errors associated with the three-dimensional
excitation line and other wavelengths has proved beneficial thermal finite difference simulations suffer from some lack
for probing layer interfaces, e.g. for measurements on of certainty in the values of the different parameters
confinement effects in quantum well structures used for
laser diodes [238], and self-assembled GaN quantum dots
[239]. 1.6
1.4 45V source-drain distance from surface
The alloy spectra are now sufficiently well defined to
electric field/MV cm −1

voltage
source

drain
1.2
allow accurate confirmation of the layer’s aluminium gate 38 nm
1.0
content based on definition of the relevant phonons across 60 nm
0.8
the alloy range (e.g. [240]). Kuball [230] has summarised the 83 nm
0.6
relevant phonons together with the selection rules for
0.4
different polarised Raman geometries.
0.2
Measurement of temperature and electric field distributions in 0
AlGaN/GaN heterostructure field-effect transistors −0.2
0 1 2 3 4 5 6 7
(HFETs), light emitting diodes and laser diodes: Raman is
position / m
now regularly used for the measurement of temperature and a
electric fields in the critical regions of active devices. The
detailed temperature profile in the source-drain opening of a 1.6
single finger HFET has been analysed by Raman spectro- 1.4 75V source-drain
electric field/MV cm −1

scopy, followed by temperature distribution profiles for voltage

source

drain

1.2
gate

multifinger transistors [241]. When self-heating in the device 1.0

occurs, cross-talk can occur between gate fingers to affect 0.8
device optimisation, reliability and performance. With a 0.6
spatial resolution of o1 mm, Raman measurements are an 0.4
improvement on other potential thermal measurement 0.2
techniques. For example the use of liquid crystals involves 0
a process that alters the device impedance, while infrared −0.2
methods have a spatial resolution of only 5–10 mm, 0 1 2 3 4 5 6 7
inadequate for many of today’s devices. The technique also position / m
b
tends to underestimate the actual temperatures.
The temperature can be determined from the Raman Fig. 107 Electric field distributions obtained by 2-D Monte Carlo
simulations for different depths from the device surface
shift of the GaN E2 (high) phonon frequency relative to
a 45 V source-drain voltage
zero bias applied. This was established earlier [242] as a b 75 V source-drain voltage
result of temperature/phonon frequency measurements. The true conversions in the calculations from temperature to electric
Frequency resolutions of the order of 0.1 cm
1 are quoted. field rely on the input of material parameters that remain uncertain.
This translates to a temperature resolution of 101C, which is The presence of a maximum in electric field below the drain region
more than adequate for this application. However work is agrees with the test data, but errors remain regarding the magnitudes.
continuing to improve this further. Data courtesy of Martin Kuball, Bristol Univ., UK

312 IEE Proc.-Sci. Meas. Technol., Vol. 152, No. 6, November 2005
required. This confirms the necessity for practical measure- Comparisons with competing alternative technologies as
ments on devices and packages for thermal management, well as earlier Raman system designs have been made. It
during the design-modelling phase. It is recognised from has been demonstrated that many of the perceived earlier
past experience with GaAs and alloys that high electric field shortcomings in the use of Raman as a technique have now
devices, including laser diodes, have failure mechanisms been overcome.
originating from, and exacerbated by, self-heating and hot Acceptance of the technique can be gauged from this
spots at critical points in the device structure. This work has highly selective set of examples to illustrate the new range in
extended to thermal management temperature testing of versatility:
diode mounting methods (e.g. flip chip mounting with heat
extraction through the gold bumps also assists light 1. Units are now on production lines providing critical
extraction for light emitters), and the associated package feedback control to the process, with information that
[243]. impacts directly on yield and profitability. They are capable
Full wafer analyses of HFETs show varying and specific of round-the-clock operation in rugged environments [246].
temperature distributions across the wafers [244]. The 2. Raman spectral information now has traceability back to
reasons for this are probably related to the defects, substrate national calibration laboratories. Measurements are ac-
homogeneity (e.g. SiC), and surface states. Ideally for cepted as valid evidence in legal cases, e.g. for drugs
manufacturing purposes the temperature distribution seizures, drug copying and patent cases filed by pharma-
should be uniform, but a practical select-on-test method ceutical companies, and fraudulent alteration of documents.
using Raman would be an alternative. This is also 3. Raman systems are now installed at the major gem
applicable to higher speed devices, solar cells, light emitting laboratories worldwide as the standard equipment for
diodes and laser diodes [245]. discrimination of natural from synthetic materials, e.g.
diamonds, and the assessment of treated gemstones, i.e.
with specialised fillers. Space programmes are incorporating
10 Conclusions and the future Raman measurements on minerals recovered into their
remote probes.
The technology for Raman instrumentation has improved
radically over the past 15 years. New design developments Applications are increasing rapidly as awareness of the
have been described enabling Raman techniques to be taken ease of use and minimal sample preparation becomes
out of the laboratory and introduced into more difficult apparent. Looking to the future, it is foreseen that there will
environments. Following on from the design breakthroughs be a continuous improvement in specification, plus an
made by the present group around 1990, the system increase in the potential use for measurements of nanoma-
specifications have been continuously improved. The terials, e.g. particles and defects on wafers and integrated
wavelength excitation range has increased, owing to the circuits that interfere with device operation and yield,
increased ability to use lower power lasers. The resulting carbon nanotubes, and thin molecular coatings. Accord-
flexibility has eliminated many of the fluorescence back- ingly the present group have successfully developed for the
ground problems that have plagued dispersive Raman in first time combined Raman–NSOM, and Raman–SEM
the past. Increased sophistication in the use of sensitive systems. The engineering involved in these design pro-
silicon CCDs has resulted in novel confocal and extended grammes to obtain stability in alignment and sample
wavelength scanning. Control of system optical compo- orientation for lengthy mapping measurements called for
nents, to adjust foci for different wavelength excitations, active control of stages and manipulatable holders. The
and slit manipulation have been coupled with selective positional repeatabilities must be at a level that is an order
binning at the CCD to give flexibility in increasing of magnitude better than the quoted measurement cap-
throughput, thereby optimising speed of measurement ability.
versus resolution for a given requirement. Improvements in signal output can be achieved from
The new design has been based on the latest engineering surface enhanced and resonance Raman effects. To some
inputs from mechanical, optoelectronic and controls extent in the past these techniques have been held back by
expertise, to produce systems with full automation in the lack of uniform surfaces for enhancement, and the non-
alignment, wavelength switching and calibration, and high availability of appropriate systems operating with low
stability. Several of the accepted ‘tried and tested’ mounts power UV lasers. These problems have now been overcome,
and stages which were used on earlier traditional Raman and there appears to be a growing consensus on the
systems were found to be inferior, and in-house develop- theoretical understanding of many surface enhanced effects.
ments were required, together with construction of the Initial results from the developments show the potential for
critical new infrared laser diode excitation sources. the detection of drugs and explosives in the vapour phase
The design philosophy moved away from the accepted down to 10
9 mol and lower.
mode of transferring to the customer costly science-based Biomedical applications will increase, with the potential
laboratory systems consisting of separate components for the highly specific Raman technique to be coupled with
assembled to make a working but non-optimised instru- larger area imaging technologies. Examples have been given
ment. A systems approach was adopted whereby the full of applications now moving towards clinical trials.
system specification overrode the specifications of the Examples of the increased use of dispersive Raman
individual components. The objective was to produce fast systems for semiconductors have been described, with the
high-resolution spectra with materials identification using chief focus on silicon and strained silicon devices. The other
new and original spectral libraries for dispersive Raman area of GaN and its alloys used for blue laser light emitters
systems. These libraries were developed to enable operation and high electric field devices has also been described.
by technicians in QA laboratories, on a remote process line,
or as a portable unit [246]. This was achieved, and Raman 11 Acknowledgments
instrumentation is now one of the fastest growing sectors in
worldwide analytical equipment markets for both research The authors thank all their many collaborators on jointly
and industrial applications. funded projects over the past ten years. They have

IEE Proc.-Sci. Meas. Technol., Vol. 152, No. 6, November 2005 313
attempted to include many of these acknowledgments in the 22 Herzberg, G.: ‘Molecular spectra and molecular structure: II.
text. They also thank a number of international partners infrared and Raman spectra of polyatomic molecules’ (Van
Nostrand Reinhold Company, 1945)
and friends, who have made helpful inputs to the design 23 Long, D.A.: ‘The Raman effect: a unified treatment of the theory of
specifications and the project priorities, in particular Shoji Raman scattering by molecules’ (John Wiley & Sons, Chichester,
Suruga of Tokyo Instruments Ltd. (Japan), Ichiro Miura, 2002)
24 Grasselli, J.G. and Bulkin, B.J. (Eds.).: ‘Analytical Raman spectro-
Renishaw KK (Japan) and S. Muraishi, JEOL (Japan) and scopy’ (Academic Press, New York, 1991)
his colleagues. Helpful collaborations and inputs have been 25 Ajito, K., Sukamoto, J.P.H., Nagahara., L.A., Hashimoto, K., and
Fujishima, A.: ‘Strain imaging analysis of Si using Raman
made with the late Tony Stradling, Imperial College, and microscopy’, J. Vac. Sci. Technol. A, Vac. Surf. Films, 1995, 13,
Robin Clark, University College, both University of pp. 1234–1238
London; Ewen Smith, Strathclyde University; Nick Stone 26 Bonera, E., Fanciulli, M., and Batchelder, D.N.: ‘Combining high
resolution and tensorial analysis in Raman stress measurements of
and Hugh Barr, the Gloucester Royal Hospital; Dick Lacey silicon’, J. Appl. Phys., 2003, 94, (4), pp. 2729–2740
PSDB (Home Office); John Steeds and Martin Kuball, 27 De Wolf, I.: ‘Micro-Raman spectroscopy to study local mechanical
Bristol University; Philip Klipstein, Clarendon Laboratory, stress in silicon integrated circuits’, Semicond. Sci. Technol., 1996, 11,
(2), pp. 139–154
Oxford University; Chris Pickering, Qinetiq; Zhang Shu- 28 De Wolf, I., Maes, H.E., and Jones, S.K.: ‘Stress measurements in
Lin, Beijing University; Zhang PX, Yunnan Technical silicon devices through Raman spectroscopy: bridging the gap
between theory and experiment’, J. Appl. Phys., 1996, 79, (9), pp.
University; David Lee, GSK; Peter Tsou, JPL, NASA; 7148–7156
Edgar Etz, NIST; Gerwin Puppels, Erasmus University and 29 Smith, W.E., Rodger, C., Dent, G., and White, P.C.: ‘Raman and
University Hospital, Rotterdam; Tim Prusnick and Steve surface enhanced resonance Raman scattering: applications in
forensic science’, in Lewis, I.R. and Edwards, H.G.M. (Eds.).:
Barnett, Renishaw Inc; Chunwei Cheng of Renishaw and ‘Handbook of Raman spectroscopy’ (Marcel Dekker, New York,
ex-Leeds University; Mike Claybourn and Pierre Graves, Basel, 2001), pp. 733–748
both ex-Renishaw. David McMurtry, Chairman and Chief 30 Matousek, P., Parker, A.W., Phillips, D., Scholes, G.D., Toner,
W.T., and Towrie, M.: ‘A picosecond time-resolved resonance
Executive of Renishaw plc, and the Renishaw Board have Raman study of S-1 cis-stilbene’, Chem. Phys. Lett., 1997, 278, (1–3),
strongly supported the continuing development from pp. 56–62
31 Pitt, G.D., Extance, P., Neat, R.C., Batchelder, D.N., Jones, R.E.,
inception. A number of students and collaborative projects Barnett, J.A., and Pratt, R.H.: ‘Optical fibre sensors’, IEE Rev. IEE
were part-funded by EPSRC, DTI and DOH. Proc., 1985, 132, pp. 214–248 (Pt. 2, J No 2)
32 Asher, S.A., Bormett, R.W., Chen, X.G., Lemmon, D.H., Cho, N.,
Peterson, P., Arrigoni, M., Spinelli, L., and Cannon, J.: ‘UV
resonance Raman-spectroscopy using a new CW laser source –
12 References convenience and experimental simplicity’, Appl. Spectrosc., 1993, 47,
(5), pp. 628–633
1 Raman, C.V., and Krishnan, K.S.: ‘A new type of secondary 33 Sands, H.S., Hayward, I.P., Kirkbride, T.E., Bennett, R., Lacey,
radiation’, Nature, 1928, 121, p. 501 R.J., and Batchelder, D.N.: ‘UV-excited resonance Raman spectro-
2 Schrader, B. (Ed.).: ‘Introduction to infrared and Raman spectro- scopy of narcotics and explosives’, J. Forensic Sciences, 1998, 43, (3),
scopy, methods and applications’ (VCH, Weinheim, 1995) pp. 509–513
3 Long, D.: ‘Early history of the Raman effect’, Int. Rev. Phys. Chem, 34 Hendra, P.J.: ‘Fourier transform Raman spectroscopy in pharma-
1988, 7, pp. 314–349 ceutical analysis and research’, Am. Lab., 1996, 28, p. 18
4 Venkataraman, G.: ‘Journey into light: life and science of C.V. 35 in Laserna, J.J. (Ed.).: ‘Fourier transform Raman spectroscopy,
Raman’, ISBN 81-85324-00-X,’ (Indian Academy of Sciences, 1988) modern techniques in Raman spectroscopy’ (John Wiley & Sons,
5 Townes, C.H.: ‘Optical and infrared masers’, J. Opt. Soc. Am., 1961, Chichester, 1996), pp. 73–108
51, (4), p. 471 36 Asselin, K.J., and Chase, B.: ‘FT-Raman spectroscopy at 1.339
6 Delhaye, M., and Dhamelincourt, P.: ‘Raman microprobe and micrometers’, Appl. Spectrosc., 1994, 48, (6), pp. 699–701
microscope with laser excitation’, J. Raman Spectrosc., 1975, 3, pp. 37 Baldwin, K.J., Batchelder, D.N., and Webster, S.: ‘Raman micro-
33–43 scopy: confocal and scanning near-field’ in Lewis, I.R. and Edwards,
7 Rosasco, GJ.: ‘Raman microprobe spectroscopy, Advances in H.G.M. (Eds.).: ‘Handbook of Raman spectroscopy: from the
infrared and Raman spectroscopy’, in Clark, R.J.H. and Hester, research laboratory to the process line’ (Marcel Dekker, New York,
R.E. (Eds.).: (Wiley Heydon, Chichester, 1980), pp. 223–282 Basel, 2001), pp. 145–189
8 Batchelder, D.N.: ‘Multichannel Raman spectroscopy with a cooled 38 Williams, K.P.J., Pitt, G.D., Smith, B.J.E., Whitley, A., Batchelder,
CCD imaging detector’, Eur. Spectrosc. News, 1988, 80 D.N., and Hayward, I.P.: ‘Use of a rapid scanning stigmatic Raman
9 Carrabba, M.M., Spencer, K.M., Rich, C., and Rauh, D.: ‘The imaging spectrograph in the industrial environment’, J. Raman
utilization of a holographic Bragg diffraction filter for Rayleigh line Spectrosc., 1994, 25, pp. 131–138
rejection in Raman spectroscopy’, Appl. Spectrosc., 1990, 44, pp. 39 Williams, K.P.J., Pitt, G.D., Batchelder, D.N., and Kip, B.J.:
1558–1561 ‘Confocal Raman microspectroscopy using a stigmatic spectro-
10 Pitt, G.D.: ‘Raman shifts into industry’, Phys. World., 1991, 4, (10), scopy and CCD detector’, Appl. Spectrosc., 1994, 28, (2),
pp. 19–20 pp. 232–235
11 Batchelder, D.N.: ‘Renishaw Transducer Systems Ltd and the 40 Schoen, C.L., Sharma, S.K., Helsley, C.E., and Owen, H.:
University of Leeds – collaboration leads to a new image’, Meas. Sci. ‘Performance of a holographic supernotch filter’, Appl. Spectrosc.,
Technol., 1992, 3, pp. 561–563 1993, 47, (3), pp. 305–308
12 Batchelder, D.N., Cheng, C., Muller, W., and Smith, B.J.E.: ‘A 41 Tedesco, J.M., Owen, H., Pallister, D.M., and Morris, M.D.:
compact Raman microprobe microscope – analysis of polydiacety- ‘Principles and spectroscopic applications of volume holographic
lene Langmuir-blodgett-films’, Makromol., Chem. Makromol. Symp., optics’, Anal. Chem., 1993, 65, pp. 441A–449A
1991, 46, p. 171 42 Barr Associates Inc, 2 Lyberty Way, Westford, MA, USA 01886
13 Yang, B., Morris, M.D., and Owen, H.: ‘Holographic notch filter for 43 Bovard, B.G.: ‘Rugate filter theory - an overview’, Appl. Opt., 1993,
low-wavenumber Stokes and Anti-Stokes Raman spectroscopy’, 32, pp. 5427–5442
Appl. Spectrosc., 1991, 45, (9), pp. 1533–1536 44 Talmi, Y.: ‘Optoelectronic image detectors in chemistry, an over-
14 Battey, D.E., Slater, J.B., Wludyka, R., Owen, H., Pallister, D.M., view’, ACS Symp. Ser., 1979, 102
and Morris, M.D.: ‘Axial transmissive f/1.8 imaging Raman 45 Sweedler, J.V., Ratzlaff, K.L. and Denton, B.M. (Eds.).: ‘Charge
spectrograph with volume-phase holographic filter and grating’, transfer devices in spectroscopy’ (John Wiley & Sons, Chichester,
Appl. Spectrosc., 1993, 47, (11), pp. 1913–1919 1994)
15 Jensen, T.R., Johnson, R.L., Ballou, J., Prohaska, W., and Main, 46 Dyer, C.D., and Smith, B.J.E.: ‘Application of continuous extended
S.E.: ‘Environmentally stable UV Raman edge filters’. Proc. 43rd scanning techniques to the simultaneous detection of Raman
Annual Technical Conf. 2000, Society of Vacuum Coaters, 2000, pp. scattering and photoluminescence from calcium disilicates using
1–5 visible and near-infrared excitation’, J. Raman Spectrosc., 1995, 26,
16 Wilson, T.: ‘Confocal microscopy’ (Academic Press, London, 1990) pp. 777–785
17 Garton, A., Batchelder, D.N., and Cheng, C.: ‘Polymer blends’, 47 Steeds, J.W., Evans, G.A., Danks, L.R., Furkert, S., Voegeli, W.,
Appl. Spectrosc., 1993, 47, pp. 922–927 Ismail, M.M., and Carosella, F.: ‘Transmission electron microscope
18 Schlucker, S., Schaeberle, M.D., Huffman, S.W., and Levin, I.W.: radiation damage of 4H and 6H SiC studied by photoluminescence
‘Raman microscopy: a comparison of point, line, and wide-field spectroscopy’, Diam. Relat. Mater., 2002, 11, (12), pp. 1923–1945
imaging methodologies’, Anal. Chem., 2003, 75, (16), pp. 4312–4318 48 Deckert, V., Liebler, W., Eck, R., and Kiefer, W.: ‘New device for
19 Chamberlain, J.: ‘The principles of interferometric spectroscopy’ Raman difference spectroscopy with multichannel and scanning
(Wiley-Interscience, Chichester, 1979) multichannel detection’, Appl. Spectrosc., 1997, 51, (7), pp. 939–943
20 ‘Progress in surface Raman spectroscopy theory and practice’, SRS- 49 Barbillat, J., Da Silva, E., and Roussel, B.: ‘Demonstration of low-
2000, Xiamen, Fujian, China, August 2000, pp. 14–17 frequency performance and microanalysis capability of multi-channel
21 ‘Overview of Raman-survey’, Analytical Consumer, 1998, 9, (1), Raman spectroscopy with near-infrared excitation’, J. Raman
pp. 2–12 Spectrosc., 1991, 22, pp. 383–391

314 IEE Proc.-Sci. Meas. Technol., Vol. 152, No. 6, November 2005
 50 Everall, N.: ‘Modelling and measuring the effect of refraction on the 83 Webster, S., Smith, D.A., and Batchelder, D.N.: ‘Raman microscopy
depth-resolution of confocal Raman microscopy’, Appl. Spectrosc., using a scanning near-field optical probe’, Vib. Spectrosc., 1998, 18,
2000, 54, pp. 773–782 pp. 51–59
51 Everall, N., Owen, H., and Slater, J.: ‘Performance analysis of an 84 Betzig, E., Trautman, J.K., Harris, T.D., Weiner, J.S., and Kostelak,
integrated process Raman analyzer using a multiplexed transmission R.L.: ‘Breaking the diffraction barrier – optical microscopy on a
holographic grating, CCD detection, and confocal fiberoptic nanometric scale’, Sci., 1991, 251, (5000), pp. 1468–1470
sampling’, Appl. Spectrosc., 1995, 49, (5), pp. 610–615 85 Betzig, E., Finn, P.L., and Weiner., J.S.: ‘Combined shear force and
52 Tabaksblat, R., Meier, R.J., and Kip, B.J.: ‘Confocal Raman near-field scanning optical microscopy’, Appl. Phys. Lett., 1992, 60,
microspectroscopy: theory and application to thin polymer samples’, (20), pp. 2484–2486
Appl. Spectrosc., 1992, 46, (1), pp. 60–67 86 Gogotsi, Y., Baek, C., and Kirscht, F.: ‘Raman microspectro-
53 Prior Scientific Instruments Ltd, 3-4 Fielding Industrial Estate, scopy study of processing-induced phase transformations and
Wilbraham Rd, Fulbourn, Cambridge, CB1 5ET, UK residual stress in silicon’, Semicond. Sci. Technol., 1999, 14, (11),
54 Steele, D.F., Young, P.M., Smith, T., Price, R., and Lewis, D.: ‘The pp. 1019–1029
potential use of Raman mapping to investigate in vitro deposition of 87 Demangeot, F., Puech, P., Paillard, V., Domnich, V., and Gogotsi,
combination pressurized metered-dose inhalers’, The AAPS Journal, Y.G.: ‘Spatial distribution of strain and phases in Si nano-
2004, 6, (4), pp. 1–4 (Article 35) indentation analysed by Raman mapping. gettering and defect
55 Markwort, L., Kip, B., da Silva, E., and Roussel, B.: ‘Raman engineering in semiconductor technology’, Solid State Phenom., 2002,
imaging of heterogeneous polymers: a comparison of global versus 82–84, pp. 777–782
point illumination’, Appl. Spectrosc., 1995, 49, (10), pp. 1411–1429 88 Hartschuh, A., Sanchez, E.J., Xie, X.S., and Novotny, L.: ‘High-
56 Bowden, M., Gardiner, D.J., Rice, G., and Gerrard, D.L.: ‘Line resolution near-field Raman microscopy of single-walled carbon
scanned micro Raman spectroscopy using a cooled CCD imaging nanotubes’, Phys. Rev. Lett., 2003, 90, (9), (article 095503)
detector’, J. Raman Spectrosc., 1990, 21, (1), pp. 37–41 89 Wang, J.J., Smith, D.A., Batchelder, D.N., Saito, Y., Kirkham, J.,
57 Barbillat, J., Dhamelincourt, P., Delhaye, M., and Da Silva, E.: Robinson, C., Baldwin, K., Li, G., and Bennett, R.: ‘Apertureless
‘Confocal imaging and fiberoptic remote sensing-a further step in near-field Raman spectroscopy’, J. Microsc., 2002, 210, (3),
molecular analysis’, J. Raman Spectrosc., 1994, 25, (1), pp. 3–11 pp. 330–333
58 Renishaw plc European Patent 0543578A1, (filed Nov.13, 1992; 90 Wang, J.J., Saito, Y., Smith, D.A., Batchelder, D.N., Kirkham, J.,
published May 26, 1993) and Robinson, C.: ‘A Controllable method for the preparation of
59 Spectral Dimensions, Inc, Suite 210, Olney, MD, 20832, USA metallised probes for efficient scanning near-field optical Raman
60 Morris, H.R., Hoyt, C.C., and Treado, P.: ‘Imaging spectrometers microscopy’, Appl. Phys. Letts., 2005, (to be published)
for fluorescence and Raman spectroscopy: acousto-optic and 91 Emery, S.R., Haskins, W.E., and Nie, S.M.: ‘Direct observation of
liquid crystal tuneable filters’, Appl. Spectrosc., 1994, 48, pp. 857–865 size-dependent optical enhancement in single metal nanoparticles’,
61 Treado, P.J., and Nelson, M.P.: ‘Raman imaging’ in Lewis, I.R. and J Am. Chem. Soc., 1998, 120, (31), pp. 8009–8010
Edwards, H.G.M. (Eds.).: ‘Handbook of Raman spectroscopy’ 92 Wessel, J.: ‘Surface enhanced optical microscopy’, J. Opt. Soc. Am.,
(Marcel Dekker, New York, Basel, 2001), pp. 191–249 1985, B 2, pp. 1538–1541
62 Etz, E.S.: ‘National Institute of Standards and Technology’, 100 93 Nanonics Imaging Ltd, Manhat Technical Park, Malcha, Jerusalem,
Bureau Drive, Gaithersburg, Maryland 20899, USA, private 91487, Israel
communications, 2000 94 Cavendish Instruments Limited, West Mount, 463 Glossop Rd,
63 Etz, E.S., Hurst, W.S., and Choquette, S.J.: ‘Raman intensity Broomhill, Sheffield, S10 2QD, UK
calibration with glass luminescence standards’, Inst. Phys., Conf. Ser., 95 Shaver, J.M.: ‘Chemometrics for Raman Spectroscopy’, in Lewis,
2000, pp. 121–122 I.R. and Edwards, H.G.M. (Eds.).: ‘Handbook of Raman spectro-
64 Dunstan, D.J., and Frogley, M.D.: ‘Double subtractive spectrometer scopy’ (Marcel Dekker, New York, Basel, 2001), pp. 275–306
as a tunable high-resolution broad-bandpass optical filter’, Rev. Sci. 96 Rossman, R.: ‘California Institute of Technology, Mineralogy
Instrum., 2002, 73, (11), pp. 3742–3746 Library’, 2000 (www.minerals.gps.caltech.edu)
65 Devonshire, R., Forrest, R.: High Temperature Science Laboratory, 97 Slater, J.B., Tedesco, J.M., Fairchild, R.C., and Lewis, I.R.: ‘Raman
Dept. of Chemistry, Univ. of Sheffield, UK; and also APL Eng. Spectroscopy and its adaptation to the industrial environment’,
Materials, Inc, Urbana, IL, USA for provision of materials in Lewis, I.R. and Edwards, H.G.M. (Eds.).: ‘Handbook of
66 Booker, G.R., Klipstein, P.C., Lakrimi, M., Lyapin, S., Mason, N.J., Raman spectroscopy’ (Marcel Dekker, New York, Basel, 2001),
Murgatroyd, I.J., Nicholas, R.J., Seong, T.-Y., Symons, D.M., pp. 41–144
Vaughan, T.A., and Walker, P.J.: ‘Growth of InAs/GaSb strained 98 Smith, B.J.E., Batchelder, D.N., and Baldwin, K.J.: ‘International
layer superlattices. 1’, J. Cryst. Growth, 1994, 145, pp. 778–785 Patent WO 9532408 (Spectroscopic apparatus)’, US5638173,
67 Prof. Zhao Bing, Institute of Theoretical Chemistry, Jilin University, (Spectrum splitting spectroscopic apparatus)
PRC, private communication, 1998 99 Hayward, I.P., Kirkbride, T.E., Batchelder, D.N., and Lacey, R.J.:
68 Bremard, C., Dhamelincourt, P., Laureyns, J., and Turrell, G.: ‘The ‘Use of a fibre optic probe for the detection of explosive materials by
effect of high-numerical aperture objectives on polarization measure- Raman spectroscopy’, J. Forensic Sciences, 1995, 40, (5), pp. 883–884
ments in micro-Raman spectroscopy’, Appl. Spectrosc., 1985, 39, pp. 100 Mahadevan-Jansen, A., and Richards-Kortum, R.: ‘Raman spectro-
1036–1039 scopy for the detection of cancer and precancers’, J. Biomed. Opt.,
69 Owen, H., Tedesco, J.B., and Slater, J.B.: ‘Remote optical 1996, 1, (1), pp. 31–70
measurement-probe’, US Patent 5377004, (filed 1993; issued 1994) 101 Feld, M.S.: ‘Light scattering for biomedical analysis and disease
70 Lewis, I.R., and Griffith, P.R.: ‘Raman spectrometry with fiber-optic diagnosis’. Proc. 8th European Conf. on Spectroscopy of Biological
sampling’, Appl. Spectrosc., 1996, 50, pp. 12A–30A Molecules’, Univ. of Twente, Netherlands, Aug 29–Sept 2, 1999,
71 Cirrex Corp, Atlanta, GA, USA. This company has taken over the pp. 531–533
original Visionex Inc 102 Feld, M.S., Manoharan, R., Salenius, J., Orenstein-Carndona, J.,
72 Puppels, G.J., Bakker-Schut, T.C., Caspers, P.J., Wolthuis, R., van Romer, T.J., Brennan, J.F., Dasari, R., and Want, Y.: ‘Detection
Aken, M., van der Laarse, A., Bruining, H.A., Buschman, H.P.J., and characterization of human tissue lesions with near-infrared
Shim, M.G., and Wilson, B.C.: ‘In-vivo Raman spectroscopy’, in Raman spectroscopy’, Proc. SPIE, 1995, 2388, pp. 99–104
Lewis, I.R. and Edwards, H.G.M. (Eds.).: ‘Handbook of Raman 103 Dasari, R.R., Motz, J.T., Haka, A.S., Hunter, M., Shafer-Peltier,
spectroscopy’ (Marcel Dekker, New York, Basel, 2001), pp. 549–574 K.E., and Feld, M.S.: ‘Raman spectroscopy of biological tissue’, in
73 InPhotonics, Inc., 111 Downey St, Norwood, MA 02062, USA Mink, J., Jalsovsky, G. and Keresztury, G. (Eds.).: ‘Proc. of 18th Int.
74 Saito, Y., Wang, J.J., Batchelder, D.N., and Smith, D.A.: Conf. on Raman spectroscopy’ (John Wiley & Sons, Chichester,
‘Simple chemical method for forming silver surfaces with controlled 25–30 Aug 2002), pp. 791–792
grain sizes for surface plasmon experiments’, Langmuir, 2003, 19, 104 Mantsch, H.H., Choo-Smith, L.P., and Shaw, R.A.: ‘Vibrational
pp. 6857–6861 spectroscopy and medicine: an alliance in the making’, Vib.
75 Sanchez, J.C., Lopez-Sanchez, J.A., Wells, R.P.K., Rhodes, C., and Spectrosc., 2002, 30, (1), pp. 31–42
Hutchings, G.J.: ‘Effect of group 13 compounds and bulky organic 105 Stone, N., Kendall, C., Shepherd, N., Crow, P., and Barr, H.: ‘Near-
oxygenates as structural promoters for selective oxidation of infrared Raman spectroscopy for the classification of epithelial pre-
n-butane with vanadium phosphorous oxide catalysts’, New J. cancers and cancers’, J. Raman Spectrosc., 2002, 33, pp. 564–573
Chem., 2001, 25, pp. 1528–1536 106 Kendall, C.A., Barr, H., Shepherd, N., and Stone, N.: ‘Optimum
76 Sanchez, F.J.C., Wells, R.P.K., Rhodes, C., Bartley, J.K., Kiely, C.J., procedure for construction of spectral classification algorithms for
and Hutchings, G.J.: ‘In-situ laser Raman spectroscopy studies of the medical diagnosis’, Proc. SPIE., 2002, 4614-25, pp. 152–158
transformations of VOPO4.0.5 H2O and VO2 P2O7’, J. Chem. Phys., 107 Fraunhofer Institute for Silicon Technology, Itzehoe, Germany,
2001, 3, pp. 4122–4128 Biophotonics Int. April/May 2000
77 Linkam Scientific Instruments Ltd, 8 Epsom Downs Metro Centre, 108 Kleinbaumm, D.G., Kupper, D.G., Muller, K.E., and Nuzan, A.:
Waterfield, Tadworth, KT20 5HT, UK ‘Applied regression analysis and other multivariable methods’, 3rd
78 Oxford Instruments plc, Old Station Way, Eynsham, Witney, Oxon, edn. (Duxbury Press, Pacific Grove, CA, 1998)
X29 4TL, UK 109 Van Houwelingen, J.S., and Le Cessu, S.: ‘Predictive value of
79 Diacell Products Ltd, 54 Ash Tree Rd, Oadby, Leicester, LE2 5TD, statistical models’, Statistics in Medicine, 1990, 9, pp. 1303–1325
UK 110 Wallace, M.B., Perelman, L.B., Backman, V., Crawford, J.M.,
80 Dunstan, D.J., and Spain, I.L.: ‘The technology of diamond anvil Fitzmaurice, M., Seiler, M., Badizadegan, K., Shields, S.J., Itzkan, I.,
high-pressure cells’, J. Phys E Sci. Instrum., 1989, 22, pp. 913–933 Dasari, R.R., Van Dam, J., and Feld, M.S.: ‘Endoscopic detection of
81 SensIR Technologies, Inc., 14 Commerce Drive, Danbury, CT, dysplasia in patients with Barretts Esophagus using light-scattering
CT06810, USA spectroscopy’, Gastroenterology, 2000, 119, (3), pp. 677–682
82 Butler, J.: ‘Naval Research Laboratories’, Washington, DC, USA, 111 Stringer, M.R.: ‘Non-invasive technologies for skin cancer diagnosis:
private communications, 2002 current status’, Medical Physics and Engineering, 2004, (in press)

IEE Proc.-Sci. Meas. Technol., Vol. 152, No. 6, November 2005 315
112 Choo-Smith, L.-P., Edwards, H.G.M., Endtz, H.P.H., Kros, J.M., 143 Hanni, H.A., Chalain, J.P., and Fritsch, E.: ‘Letters: new spectral
Barr, H., Bruining, H.A., and Puppels, G.I.: ‘Medical applications of evidence for GE POL diamond detection’, Gems and Gemology,
Raman spectroscopy: from proof of principle to clinical implementa- 2000, 36, (2), pp. 96–97
tion’, Biopolymers (Biospectroscopy), 2002, 67, (1), pp. 1–9 144 Smith, C.P., Bosshart, G., Ponahlo, J., Hammer, V.M.F., Klapper,
113 Caspers, P.J., Lucassen, G.W., Carter, E.A., Bruining, H.A., and H., and Schmetzer, K.: ‘GE POL diamonds: before and after’, Gems
Puppels, G.J.: ‘In vivo confocal Raman microscopy of the skin: non- and Gemology, 2000, 36, (3), pp. 192–215
invasive determination of molecular concentration profiles’, J. 145 Tsou, P., Brownlee, D.E., Anderson, J.D., Bhaskaran, S., Cheuvront,
Investigative Dermatology, 2001, 116, (3), pp. 434–442 A.R., Clark, B.C., Duxbury, T., Economou, T., Green, S.F., Hanner,
114 Freeman, T.L., Cope, S.E., Stringer, M.R., Cruse-Sawyer, J.E., M.S., H.orz, F., Kissel, J., McDonnell, J.A.M., Newburn Jr, Ryan,
Brown, S.B., Batchelder, D.N., and Birbeck, K.: ‘Raman spectro- R.L., Sandford, R.E., Sekanina, S.A., Silen, Z., Tuzzolino, J.,
scopy for the determination of photosensitiser localization in cells’, Vellinga, A.J., and Zolensky, J.M.: ‘STARDUST encounters Comet
J. Raman Spectrosc., 1997, 28, pp. 641–643 81P/Wild 2’, J. Geophys. Res., 2004, 109, (SRD 1-1-10)
115 Freeman, T.L., Cope, S.E., Stringer, M.R., Cruse-Sawyer, J.E., 146 Tsou, P., Brownlee, D.E., H.orz, F., Sandford, S.A., and Zolensky:
Brown, S.B., Batchelder, D.N., and Birbeck, K.: ‘Investigation of the ‘Wild 2 and interstellar sample collection and Earth return’,
subcellular localization of zinc phthalocyanines by Raman mapping’, J Geophys. Res., 2003, 108, (E10), (SRD 3-1-21)
Appl. Spectrosc., 1998, 52, (10), pp. 1257–1263 147 Mulder, C.A.M., and Damen, A.A.J.M.: ‘Raman analysis of the
116 Deinum, G., Rodriguez, D., R.omer, T.J., Fitzmaurice, M., Kramer, initial stages of the hydrolysis and polymerisation of tetraethylortho-
J.R., and Feld, M.S.: ‘Histological classification of Raman spectro- silicate’, J. Non-Cryst. Solids, 1987, 93, pp. 169–178
scopy of human coronary artery atherosclerosis using principal 148 Tsou, P.: ‘Mir sample return experiment’. NASA Mir First Quarterly
component analysis’, Appl. Spectrosc., 1999, 53, pp. 938–942 Report, JSC, 1997, pp. 6–13
117 Buschman, H.P., Marple, E.T., Wach, M.L., Bennett, R., Bakker- 149 Burchell, M.J., Creighton, J.A., and Kearsley, A.T.: ‘Identification
Schut, T.C., Bruining, H.A., Bruschke, A.V., van der Laarse, A., and of organic particles after capture in hypervelocity impacts on aerogel’,
Puppels, G.J.: ‘In vivo determination of the molecular composition of J. Raman Spectrosc., 2004, 35, pp. 249–253
the artery wall by intravascular Raman spectroscopy’, Anal. Chem., 150 Wagner, J., Wild, C., and Koidl, P.: ‘Resonance effects in Raman
2000, 72, (16), pp. 3771–3775 scattering from polycrystalline diamond films’, Appl. Phys. Lett.,
118 Sands, H.: ‘Resonance Raman Microscopy with visible and 1991, 59, (7), pp. 779–781
ultraviolet lasers’. PhD thesis, University of Leeds, UK 1998 151 Pickard, C.D.O., Davis, T.J., Wang, W.N., and Steeds., J.W.:
119 Asher, S.A.: ‘UV resonance Raman spectroscopy for analytical, ‘Mapping crystalline quality in diamond films by micro-Raman
physical, and biophysical chemistry’, Anal. Chem., Part 1, 1993, 65, spectroscopy’, Diam. Relat. Mater., 1998, 7, pp. 238–242
(2), pp. 59A–66A 152 Pickard, C.D.O., Davis, T.J., Gilmore, A., Wang, W.N., and Steeds,
120 Asher, S.A.: ‘UV resonance Raman spectroscopy for analytical, J.W.: ‘Calculation of the stress in large square facets of MPCVD
physical, and biophysical chemistry’, Anal. Chem. Part 2, 1993, 65, grown diamond from cathodoluminescence and Raman spectro-
(4), pp. 210A–210A scopy measurements, and comparison to stress predicted from finite
121 Boustany, N.N., Crawford, J.M., Manoharan, R., Dasari, R.R., and element models’, Diam. Relat. Mater., 1997, 6, pp. 1062–1066
Feld, M.S.: ‘Analysis of nucleotides and aromatic amino acids in 153 Lauer, J.L.: ‘Raman spectra of quasi - elemental carbon’, in Lewis,
normal and neoplastic colon mucosa by ultraviolet resonance I.R. and Edwards, H.G.M. (Eds.).: ‘Handbook of Raman spectro-
Raman spectroscopy’, Laboratory Investigation, 1999, 79, (10), scopy’ (Marcel Dekker, New York, Basel, 2001), pp. 863–918
pp. 1201–1214 154 Whitley, A.: ‘Raman spectroscopy and carbon overcoats’, in Lewis,
122 Arikan, S., Sands, H.S., Rodway., R.G., and Batchelder, D.N.: I.R. and Edwards, H.G.M. (Eds.).: ‘Handbook of Raman spectro-
‘Raman spectroscopy and imaging of b-carotene in live corpus scopy’ (Marcel Dekker, New York, Basel, 2001), pp. 975–998
luteum cells’, Animal Reproduction Science, 2002, 71, pp. 249–266 155 Bing, Z., Wei, B.O., Sullivan, D.J.D., and Gotts, H.E.: ‘A systematic
123 Smith, G.D., and Clark, R.J.H.: ‘Raman microscopy in art history study of a-CNx overcoat film for proximity recording’, IEEE, Trans.
and conservation science’, Reviews in Conservation, 2001, 2, pp. 93–106 Magn., 1997, 33, (5), pp. 3109–3111
124 Clark, R.J.H.: ‘Pigment identification by spectroscopic means: an 156 Gopinathan, N., Robinson, C., and Ryan, F.: ‘Characterization and
arts/science interface’, C.R. Chimie, 2002, 5, pp. 1–14 properties of diamond-like films for magnetic recording application’,
125 Brown, K.L., and Clark, R.J.H.: ‘Analysis of pigmentary materials Thin Solid Films, 1999, 355–356, pp. 401–405
on the Vinland Map and Tartar Relation by Raman microprobe 157 Grill, A.: ‘Diamond-like-carbon: state of the art’, Diam. Relat.
spectroscopy’, Anal. Chem., 2002, 74, (15), pp. 3658–3661 Mater., 1999, 8, pp. 428–434
126 McCrone, W.C.: ‘Chemical analytical study of the Vinland Map’. 158 Gilkes, K.W.R., Sands, H.S., Batchelder, D.N., Robertson, J., and
Report to Yale University Library, New Haven, CT, USA, 1974 Milne, W.I.: ‘Direct observation of sp3 bonding in tetrahedral
127 McCrone., W.C.: ‘The Vinland Map’, Anal. Chem., 1988, 60, pp. amorphous carbon using Raman spectroscopy’, Appl. Phys. Lett.,
1009–1018 1997, 70, pp. 1980–1982
128 Cahill, T.A., Kusko, B.H., Eldred, R.A., M.oller, G., Dutshke, D., 159 Ferrari, A.C., and Robertson, J.: ‘Interpretation of Raman spectra
and Wick, D.L.: ‘The Vinland Map revisited: new compositional of disordered and amorphous carbon’, Phys. Rev. B, 2000, 61,
evidence on its inks and parchment’, Anal. Chem., 1987, 59, pp. 95–107
pp. 829–833 160 Ferrari, A.C.: ‘Determination of bonding in diamond-like carbon by
129 Dr D. Bertani, Dept. di Ottica, Univ. of Firenze, Italy (private Raman spectroscopy’, Diam. Relat. Mater., 2002, 11, pp. 1053–1061
communications) 161 Bormett, R.W.: in Lewis, I.R. and Edwards, H.G.M. (Eds.).:
130 Kiefert, L., H.anni, H.A., and Ostertay, T.: ‘Raman spectroscopic ‘Handbook of Raman spectroscopy’ (Marcell Dekker, New York,
applications to gemmology’, in Lewis, I.R. and Edwards, H.G.M. Basel, 2001), pp. 863–918
(Eds.).: ‘Handbook of Raman Spectroscopy’ (Marcel Dekker, New 162 Young, R.J., Private communications
York, Basel, 2001), pp. 469–489 163 Andrews, M.C., Bannister, D.J., and Young, R.J.: ‘The interfacial
131 Bloomfield, M.: ‘Recent applications of microRaman and PL properties of aramid/epoxy model composites’, J. Mater. Sci., 1996,
spectroscopy to the analysis of geomaterials’, in Mink, J., Jalsovszky, 31, pp. 3893–3913
G. and Keresztury, G. (Eds.).: ‘Proc. 18th Int. Conf. on Raman 164 Dresselhaus, M.S., Dresselhaus, G., Pimenta, M.A., and Eklund,
spectroscopy’ Budapest, Hungary, 2002 (John Wiley & Sons, P.C.: ‘Raman scattering in carbon materials’, in Pelletier, M.J. (Ed.).:
Chichester), pp. 919–920 ‘Analytical applications of Raman spectroscopy’ (Blackwell Science,
132 H.anni, H.A., Schubiger, B., Kiefert, L., and Haberle, S.: ‘Raman Oxford, 1999), pp. 367–434
investigations on two historical objects from Basel Cathedral: the 165 Dresselhaus, M.S., Dresselhaus, G., and Eklund, P.C.: ‘Science of
Reliquary Cross and Dorothy Monstrance’, Gems and Gemology, fullerenes and carbon nanotubes’ (Academic Press, San Diego, 1996)
1998, 34, (2), pp. 102–125 166 Kroto, H.W.: ‘The charisma of C60 buckminsterfullerene’, MRS
133 Nelson, J.B.: ‘Nelson Gemmological Instruments’, 1 Lyndhurst Rd, Bull., 1994, 19, pp. 21–22
Hampstead, London NW3 5PX (private communications) 167 Iijima, S.: ‘Helical microtubes of graphitic carbon’, Nature, 1991, 354,
134 Koivula, J.I.: ‘Diopside inclusions in Arizona pyrope identified by pp. 56–58
Raman analysis’, Z. Dt. Gemmol. Ges., 1998, 47, (1), pp. 39–43 168 Rao, C.N.R., and Seshadri, R.: ‘Phase transitions, superconductivity,
135 Lab. Notes, Gems and Gemology, 1998, 26, p. 45 and ferromagnetism in fullerene systems’, MRS Bull., 1994, 18, pp.
136 Lab. Notes, Gems and Gemology, 1998, 26, pp. 132–133 28–30
137 Tay, T.S., Shen, Z.X., and Yee, S.L.: ‘On the identification of amber 169 Jorio, A., Souza Filho, A.G., Dresselhaus, G., Dresselhaus, M.S.,
and its imitations using Raman spectroscopy - preliminary results’, Swan, A.K., Unlu, M.S., Goldberg, B.B., Pimenta, M.A., Hafner,
Australian Gemmologist, 1998, 20, pp. 114–123 J.H., and Lieber, C.M.: ‘G-band resonant Raman study of 62
138 Zhang, P.X.: Yunnan Technical University, Kunming PRC (private isolated single-wall carbon nanotubes’, Phys. Rev. B, 2002, 65, (8),
communications) (article no. 155412)
139 Zhang Wen-Hua, H.: ‘School of Geology’, University of Geology, 170 Corio, P., Santos, P.S., Pimenta, M.A., and Dresselhaus, M.S.:
China, Wuhan (PRC private communications) ‘Characterization of single wall carbon nanotubes filled with silver
140 Izraeli, E.S., Harris, J.W., and Navon, O.: ‘Raman barometry chloride’, in Mink, J., Jalsovszky, G. and Keresztury, G. (Eds.).:
of diamond formation’, Earth Planet. Sci. Lett., 1999, 173, ‘Proc. 18th Int. Conf. on Raman spectroscopy, Budapest, Hungary,
pp. 351–360 25–30 Aug. 2002’ (John Wiley & Sons, Chichester), pp. 455–456
141 Chalain, J.P., Fritsch, E., and H.anni, H.A.: ‘Detection of GE POL 171 Claybourn, M., and Ansell, M.: ‘Using Raman spectroscopy to solve
diamonds: a first stage’, Revue de Gemmologie., 1999, 138/139, crime: inks, questioned documents and fraud’, Science and Justice,
pp. 30–33 2000, 40, (4), pp. 261–271
142 Chalain, J.P., Fritsch, E., and Hanni, H.A.: ‘Detection of GE POL 172 Laing, D.K., and Isaacs, M.D.J.: ‘The comparison of nanogram
diamonds: a second step’, Journal of Gemmology, 2000, 27, (2), quantities of ink using visible microspectrometry’, J. Forensic Science
pp. 73–78 Society, 1983, 23, pp. 147–154

316 IEE Proc.-Sci. Meas. Technol., Vol. 152, No. 6, November 2005
173 Savioli, S., Bersani, D., Latici, P.P., and Garofano, L.: ‘Identification 204 Renucci, J.B., Tyte, R.N., and Cardona, M.: ‘Resonant Raman
of inks and toners for printers: a forensic application of the micro- scattering in silicon’, Phys. Rev. B, 1975, 11, (10), pp. 3885–3895
Raman technique’, in Mink, J., Jalsovszky, G. and Keresztury, G. 205 Kuball M, Wills H.H.: Physics Laboratory, University of Bristol,
(Eds.).: ‘Proc. of 18th Int. Conf. on Raman spectroscopy, Budapest, Tyndalls Avenue, Bristol, UK BS8 1TL (private communications)
Hungary, 25–30 Aug. 2002’ (John Wiley & Sons, Chichester, 2002), 206 Tuschel, D.D., Lavine, J.P., and Russell, J.B.: ‘Raman spectroscopic
pp. 943–944 study of ion-implanted and annealed silicon’, Mater. Res. Soc. Symp.
174 Zona, C.A.: ‘The Analysis of Nyclad ammunition discharge residues Proc., 1996, 406, p. 549 554
using transmission electron microscopy and polarized light micro- 207 Tuschel, D.D., and Lavine, J.P.: ‘Raman spectroscopy of
scopy’, Microscope, 1996, 44, (1), pp. 11–14 ion-implanted silicon’, Mater. Res. Soc. Symp. Proc., 1997, 438,
175 McCrone Associates Inc., 850 Pasquinelli Drive, Westmont, IL pp. 143–148
60559 208 Ossadnik, Ch., Verprek, S., and Gregora, I.: ‘Applicability of Raman
176 Stich, S., Bard, D., Gros, L., Wenz, H.W., Yarwood, J., and scattering for the characterization of nanocrystalline silicon’, Thin
Williams, K.P.J.: ‘Raman microscopic identification of gunshot Solid Films, 1998, 333, pp. 1–4
residues’, J. Raman Spectrosc., 1998, 29, pp. 787–790 209 Li, B., Yu, D., and Zhang, S.-H.: ‘Raman spectral study of silicon
177 Lurie, I.S.: ‘Application of capillary electrophoresis to the analysis of nanowires’, Phys. Rev. B, 1999, 59, (30), pp. 1645–1648
seized drugs’, International Laboratory, 1996, 26, pp. 21–28 210 Moreno, J.D., Agull!o-Rueda, F., Montoya, E., Marcos, M.L.,
178 Lacey, R.J., Hayward, I.P., Sands, H.S., and Batchelder, D.N.: Gonz!alez-Velasco, J., Guerro-Lemus, R., and Martinez-Duart,
‘Characterisation and identification of contraband using UV J.M.: ‘Depth–resolved micro-Raman study of porous silicon
resonant Raman spectroscopy’, Proc. SPIE., 1996, 2937, pp. 100–104 at different oxidation states’, Appl. Phys. Lett., 1997, 71, (15),
179 Bell, S.E.J., Burns, D.T., Dennis, A.C., and Speer, J.S.: ‘Rapid pp. 2166–2168
analysis of ecstasy and related phenethylamines in seized tablets by 211 Ballast, L.K., Hossain, T.Z., and Campion, A.: ‘Raman spectroscopy
Raman spectroscopy’, The Analyst, 2000, 125, pp. 541–544 – a multifunctional analysis tool for microelectronics manufacturing’,
180 Lacey, R.J.: ‘Some advances in the use of Raman spectroscopy in Proc. SPIE, 2000, 4182, pp. 221–230
security screening applications’. Proc. ECOS’97, European Conf. 212 De Wolf, I., Howard, A., Lavers, A., Maex, K., and Maes, H.E.:
in Security Screening Applications, 1997, Vol. 437, pp. 10–12 ‘Local identification and mapping of the C49 and C54 titanium
181 Hamer, M.: ‘Forensic lab awash with Semtex’, New Scientist, 25 May phases in submicron structures by micro-Raman spectroscopy’, Appl.
1996, pp. 4–5 Phys. Lett., 1998, 70, pp. 2262–2264
182 Lewis, I.R., Daniel, N.W., Chaffin, N.C., Griffiths, P.R., and 213 Privitera, S., Quilici, S., La Via, F., Spinella, C., Meinardi, F., and
Tungol, M.W.: ‘Raman spectroscopic studies of explosive materials: Rimini, E.: ‘Kinetics of the C49-C54 transformation by micro-
towards a fieldable explosives detector’, Spectrochim. Acta, A., 1995, Raman imaging’, Microelectron. Eng., 2001, 55, pp. 109–114
51, pp. 1985–2000 214 Li, B., Huang, F., Zhang, S.-H., Gao, Y., and Zhang, L.: ‘Micro-
183 Neudorf, P., McCoveye, M.A., and Elias, L.: ‘Testing protocol for Raman spectroscopic study of two-dimensional stress distributions in
surfaces: sampling detectors’. Proc. 4th Int. Symp. on Analysis and poly-Si induced CoSi2 patterns, Semiconductor beam epitaxial
Detection of Explosives, Jerusalem, Sept 1992, pp. 337–343 growth’, J. Raman Spectrosc., 1996, 27, pp. 265–270
184 Cheng, C., Kirkbride, T.E., Batchelder, D.N., Lacey, R.C., and 215 Pelletier, M.J., Davis, K.L., and Carpio, R.A.: ‘Applications of
Sheldon, T.G.: ‘In situ detection and identification of trace explosives Raman spectroscopy for SC-1 processing bath concentration
by Raman microscopy’, J. Forensic Sciences, 1995, 40, (1), pp. 31–37 control’, Electrochem. Soc. Proc., 1995, 95, (2), pp. 282–293
185 McHugh, C.J., Keir, R., Graham, D., and Smith, W.E.: ‘Selective 216 Brault, P., Mathias., J., Laure, C., Ranson, P., and Texier, O.: ‘In situ
functionalisation of TNT for sensitive detection by SERRS’, Chem. Raman spectroscopy of silicon surfaces during SF6 plasma etching’,
Communications., 2002, pp. 580–581 J. Phys., Condens. Matter, 1994, 6, pp. L1–L6
186 McHugh, C.J., Smith, W.E., Lacey, R.J., and Graham, D.: ‘The first 217 Gatzke, C., Webb, S.J., Fobelets, K., and Stradling, R.A.: ‘In situ
controlled reduction of the high explosive RDX’, Chem. Commu- Raman spectroscopy of the selective etching of antimonides in
nications, 2002, pp. 2514–2515 GaSb/AlSb/InAs heterostructures’, Semicond. Sci. Technol., 1998, 13,
187 Williams, A.C.: ‘Some pharmaceutical applications of Raman pp. 399–403
spectroscopy’, in Lewis, I.R. and Edwards, H.G.M. (Eds.).: 218 JEOL Ltd, 1-2 Musahino 3 – Chome, Akishima, Tokyo, 196-8558
‘Handbook of Raman spectroscopy’ (Marcel Dekker, New York, Japan
Basel, 2001), pp. 575–592 219 The data were obtained from an installation at NRL, Washington
188 Giles, J.H., Shackman, J.G., and Denton, M.B.: ‘The impact of DC, USA, courtesy of Jim Butler
Raman-the awakening giant of spectroscopy-on the pharmaceutical 220 Wagner, V., Drews, D., Esser, N., Zahn, D.R.T., Geurts, J., and
industry’, American Pharmaceutical Review (Analysis and Control) Richter, W.: ‘Raman monitoring of semiconductor growth’, J. Appl.
Winter, 1999, pp. 44–51 Phys., 75, pp. 7330–7333
189 Thorley, F.C., Baldwin, K.J., Lee, D.C., Batchelder, D.N.: ‘FACCS, 221 Wagner, V., Richter, W., Geurts, J., Drews, D., and Zahn, D.R.T.:
Ft. Lauderdale, Florida, USA, Poster, https://faccs.org/stuff// ‘On-line Raman and molecular for alloy compositions based on
contentmgr//files/2973100e02f7d7ae9c45f194dba24c5e/pdf/final_ polarization measurements’, Semicond. Sci. Technol., 2003, 18,
program_2003 page 67 of this PDF, 2003 pp. 566–575
190 Sklar, S.H.: Paxils and polymorph patents: have things gone 222 Boher, P., Piel, J.P., Stehl!e, J.L., Pickering, C., and Tarnowska, A.:
from bad to worse for big pharma? Leydig and Mayer Ltd, News, ‘Feasibility and applicability of integrated metrology using
www.leydig.com/news/pli mar2003.asp.2003 spectroscopic ellipsometry in a cluster tool’, Proc. SPIE, 2001,
191 Brody, R., and Clark, D.: ‘Applications in product development, 4449, pp. 79–89
Raman spectroscopy’, Eur. Pharm. Rev., 2003, 8, pp. 31–37 223 Rath, S., Grigorescu, C., Hsieh, M.L., Voudouris, E., and Stradling,
$ si!c, S., Clark, D.A., Mitchell, J.C., and Snowden, M.J.: ‘A
192 Sa$ R.A.: ‘Polarization-dependent Raman spectroscopic protocols for
comparison of Raman chemical images produced by univariate and calibration of the alloy composition and strain in bulk and thin-film
multivariate data processing-a simulation with an example from Si1
xGex’, Semicond. Sci. Technol., 2000, 15, pp. L1–L5
pharmaceutical practice’, The Analyst, 2004, 129, pp. 1001–1007 224 Rath, S., Hsieh, M.L., Etchegoin, P., and Stradling, R.A.: ‘Alloy
193 Chenery, D., and Bowring, H.: ‘Infrared and Raman spectroscopic effects on the Raman spectra of Si1
xGex and calibration protocols’,
imaging in biosciences’, Spectrosc. Eur., 2003, 15, (40), pp. 8–14 Sci. Technol., 1998, 13, pp. 634–636
194 Cardona, M. and Guntherodt, G. (Eds.).: ‘Light scattering in solids 225 Tsang, J.C., Mooney, P.M., Dacol, F.H., and Chu, J.O.: ‘Measure-
1-6. Topics in applied physics’ (Springer-Verlag, Berlin, 1984) ments of alloy composition and strain in thin GexSi1-x layers’, J. Appl.
195 Pollak, F.H.: ‘Semiconductor characterization by Raman spectro- Phys., 1994, 75, pp. 8098–8108
scopy’, Microscience, microelectronics-applications, Materials and 226 Chen, H., Li, Y.K., Peng, C.S., Liu, H.F., Liu, Y.L., Huang, Q., and
Technology., 1985, 6, pp. 185–284 Zhou, J.M.: ‘Crosshatching on a SiGe film grown on a Si (001)
196 Nakashima, S., and Hangyo, M.: ‘Characterization of semiconductor substrate studied by Raman mapping and atomic force microscopy’,
materials by Raman microprobe’, IEEE J. Quantum Electron., 1989, Phys. Rev., B, 2002, (article 233303)
25, (5), pp. 965–975 227 Mooradian, A., and Wright, G.B.: ‘Observation of the interaction of
197 Sun, W.X., and Shen, Z.X.: ‘Apertureless near-field scanning Raman plasmons with longitudinal optical phonons in GaAs’, Phys. Rev.
microscopy using reflection scattering geometry’, Ultramicroscopy, Lett., 1966, 16, (22), pp. 999–1001
2003, 94, pp. 237–244 228 Yuasa, T., Naritsuka, S., Mannoh, M., Shinozaki, K., Yamanaka,
198 Bonera, E., Fanciulli, M., and Batchelder, D.N.: ‘Stress mapping in K., Nomura, Y., Mihara, M., and Ishii, M.: ‘Raman scattering from
silicon: advantages of using a Raman spectrometer with a single coupled plasmon-LO phonon modes in n-type AlxGa1
xAs’, Phys.
dispersive stage’, Appl. Spectrosc., 2002, 56, (5), pp. 560–563 Rev. B, 1986, 33, (2), pp. 1222–1232
199 Hu, S.M.: ‘Film-edge-induced stress in substrates’, J. Appl. Phys., 229 Kozawa, T., Kachi, T., Kano, H., Taga, Y., Hashimoto, M., Koide,
1979, 50, (7), pp. 4661–4666 N., and Manabe, K.: ‘Raman scattering from LO phonon-plasmon
200 Pollak, F.H., and Tsu, R.: ‘Raman characterization of semiconduc- coupled modes in gallium nitride’, J. Appl. Phys, 1994, 75, (2),
tors revisited’, Proc. SPIE, 1984, 452, pp. 26–43 pp. 1098–1101
201 Tsu, R., Gonzalez-Hernandez, J., Dohler, J., and Ovshinsky, S.R.: 230 Kuball, M.: ‘Raman spectroscopy of GaN, AlGaN and AlN for
‘Order parameters in a-Si’, Solid State Commun., 1983, 46, process and growth monitoring/control’, Surf. Interface Anal., 2001,
pp. 79–82 31, pp. 987–999
202 Gotts, H.: ‘Analytical Services Group’, Santa Clara, CA, USA 231 Stradling, R.A., unpublished work, 1997
(private communications) 232 Ponce, F.A., Steeds, J.W., Dyer, C.D., and Pitt, G.D.: ‘Direct
203 Paillard, V., Puech, P., Laguna, M.A., Temple-Boyer, P., Caussot, imaging of impurity-induced Raman scattering in GaN’, Appl. Phys.
B., Couderc, J.P., and Mauduit, B.: ‘Resonant Raman scattering in Lett., 1996, 69, pp. 2650–2652
polycrystalline silicon thin films’, Appl. Phys. Lett., 1998, 73, (12), 233 Hayes, J.M., Kuball, M., Bell, A., Harrison, I., Korakakis, D., and
pp. 1718–1720 Foxon, C.T.: ‘High-temperature processing of GaN: the influence of

IEE Proc.-Sci. Meas. Technol., Vol. 152, No. 6, November 2005 317
 the annealing ambient on strain’, Appl. Phys. Lett., 1999, 75, (14), Ga1
xAlxN solid solutions’, MRS Internet J Nitride Semicond.
pp. 2097–2099 Res., 2, 1997, (article 40)
234 Dyer, C.D., and Pitt, G.D.: unpublished measurements, 1995 241 Kuball, M., Rajasingam, S., Sarua, A., Uren, M.J., Martin, T.,
235 Dewsnip, D.J., Andrianov, A.V., Harrison, I., Lackliston, D.E., Hughes, B.T., Hilton, K.P., and Balmer, R.S.: ‘Measurement of
Orton, J.W., Morgan, J., Ren, G.B., Cheng, T.S., Hooper, S.E., and temperature distribution in multifinger AlGaN/GaN heterostructure
Foxon, C.T.: ‘Observation of resonant Raman lines during the field-effect transistors using micro-Raman spectroscopy’, Appl. Phys.
photoluminescence of doped GaN’, Semicond. Sci. Technol., 1997, Lett., 2003, 82, pp. 124–126
12, pp. 1–4 242 Rajasingam, S., Pomeroy, J.W., Kuball, M., Uren, M.J., Martin, T.,
236 Kuball, M., Hayes, J.M., Bell, A., Harrison, I., Korakakis, D., and Herbert, D.C., Hilton, K.P., and Balmer, R.S.: ‘Micro-Raman
Foxon, C.T.: ‘The influence of annealing ambient on strain and temperature measurements for electric field assessment in active
doping in GaN during high temperature processing’, Phys. Status AlGaN/GaN HFETs’, IEEE Trans. Electron. Devices, 2004, 25,
Solidi A, 1999, 176, pp. 759–762 pp. 456–458
237 Kuball, M., Demangeot, F., Frandon, J., Renucci, M.A., 243 Liu, M.S., Bursill, L.A., Prawer, S., Nugent, K.W., Tong, Y.Z., and
Sands, H., Batchelder, D.N., Clur, S., and Briot, O.: ‘Degradation Zhang, G.Y.: ‘Temperature dependence of Raman scattering in single
of AlGaN during high-temperature annealing monitored by crystal GaN films’, Appl. Phys. Lett., 1999, 74, (21), pp. 3125–3127
ultraviolet Raman scattering’, Appl. Phys. Lett., 1999, 74, (4), 244 Chitnis, A., Sun, J., Mandavilli, V., Pachipulusu, R., Wu, S.,
pp. 549–551 Gaevski, M., Adivarahan, V., Zhang, J.P., Khan, M.A., Sarua, A.,
238 Gleize, J., Demangeot, F., Frandon, J., Renucci, M.A., Kuball, M., and Kuball, M.: ‘Self-heating effects at high pump currents in deep
Grandjean, N., and Massies, J.: ‘Resonant Raman scattering in ultraviolet light-emitting diodes at 325 nm’, Appl. Phys. Lett., 2002,
(Al,Ga)N/GaN quantum well structures’, Thin Solid Films, 2000, 81, (18), pp. 3491–3493
364, pp. 156–160 245 Kuball, M., Pomeroy, J.W., Rajasingam, S., Sarua, A., Uren, M.J.,
239 Kuball, M., Gleize, J., Tanaka, S., and Aoyagi, Y.: ‘Resonant Martin, T., Lell, A., and H.arle, V.: ‘High spatial resolution micro-
Raman scattering on self-assembled GaN quantum dots’, Appl. Phys. Raman temperature measurements of nitride devices (FETs and light
Lett., 2001, 78, (7), pp. 987–989 emitters)’, Phys. Status Solidi, (in press)
240 Demangeot, F., Groenen, J., Frandon, J., Renucci, M.A., Briot, O., 246 Pitt, G.D.: ‘Metres to nanometres: measurements in clean and dirty
Ruffenach-Clur, S., and Aulombard, R.L.: ‘Raman study of environments’, Eng. Sci. Educ. J., 2001, 10, (6), pp. 225–236

318 IEE Proc.-Sci. Meas. Technol., Vol. 152, No. 6, November 2005