FOURIER TRANSFORM INFRARED SPECTROMETRY

James A. de Haseth
Department of Chemistry
The University of Alabama
University, Alabama 35486

I NTRODUCTI ON
Fourier Transform Infrared (FT-IR) Spectrometry has evolved
during the last quarter century fran a limited and specialized
technology to a widely accepted and powerful tool. Although FT -IR
was originally the province of astronomers it quickly found appl i-
cation in high-resolution spectroscopy and analytical chemistry.
This chapter will cover the basic aspects of Fourier Transform Infra-
red Spectrometry as well as explore some recent applications of the
technique. More in-depth treatments of the subject may be found
elsewhere.1,2
The Michelson Interferometer
The basic component of most Fourier Transform Infrared spectro-
meters is the Michelson interferometer. This is not the only inter-
ferometer used in FT-IR, but it is employed more often than other
designs. A treatment of many other interferometer deSigns is avail-
able. 3 The Michelson interferometer in a Fourier Transform Infrared
spectrometer replaces the monochromator in a dispersive instrument,
although the functions cannot be correlated. A monochomator divides
a continuous bandwidth into its component frequencies, whereas an
interferometer produces interference patterns of the bandwidth in a
precise and regulated manner. It should be noted that this type of
interferometer is not restricted to the infrared region and its use
can be extended to the visible and mill imeter regions of the
electromagnetic spectrum.

387
A. G. Marshall (ed.), Fourier, Hadamard, and Hilbert Transforms in Chemistry
© Springer Science+Business Media New York 1982
388 James A. de Haseth
A schematic of a Michelson interferometer as used in its most
common configuration is shown in Figure 1. The radiation paths have
been offset for clarity. It is preferable to consider the case in
which the incoming radiation is monochromatic before the general
polychromatic case is presented. The incoming radiation of intensity
I(vl} first strikes the beamsplitter which reflects some of the
radlation and ideally transmits the remainder. (An ideal beam-
splitter will not absorb or scatter any of the radiation.) If the
frequency of the monochromatic radiation is i1 wavenumbers, we may
consider the reflectance of the beamsplitter to be RV1 and the
transmittance to be Til, at frequency VI. Both the reflected and
transmitted beams, RUl and TUl, are directed towards mutually
perpendicular plane mlrrors w~ich return the beams to the beam-
splitter. As can be seen from Figure 1, a portion of the radiation
equal to Rp: 2 + Tv 2 is returned to the source and a portion equal
to 2Rv1TVl ls emitied perpendicular to the source. The magnitude of
the two portions is determined by the relative distances of the two
mirrors from the beamsplitter.
For monochromatic radiation the radiation from both arms of the
Michelson interferometer will be in phase at the beamsplitter if the
two mirrors are equidistant from the beamsplitter. In practice one
mirror is fixed, the other is variable in its distance from the beam-
splitter. Using the conventional symbol, 8, for the optical path
difference between the two mirrors, constructive interference of the
two beams from both arms will occur when they are in phase, or when

[1]

L Fixed Mirror

.......

I
Moving Mirror Rv, Ry,

Tv, Source
~ ::::-10---+-+----_*
~~----~------.
TVI Rij. + T~l

2 Rv,Ty,
Beomsplitter

Figure 1. Optical layout of a Michelson interferometer. The

beams in the two arms of the interferometer have been
offset for clarity.
Fourier Transform Infrared Spectrometry 389

Where n is an even integer. Destructive interference occurs when n

is an odd integer. The radiation passed by the interferometer and
directed perpendicular to the source, i.e. to the detector, has an
intensity or energy proportional to the optical path difference.
This intensity, I'(8)Det, can be shown to be equal to:
[2]

As energy must be conserved, the remainder of the total source

intensity I(v 1 ) is returned to the source. The intensity of this
radiation is:
I'(8)Src = (R v1 + TV1 )I( v 1) -2Rv1Til1I(v1)COS 2vV18 [3]
As noted above an ideal beamspl itter does not absorb or scatter
radiation, thus:
Rill + TV1 = 1 [4]

Using this relat ionship it can be verified easily that

[5]
It is interesting to note that a Michelson interferometer is most
efficient when Til = Rv = 0.50. The most intense signal is received
by the detector ;rten thJ radiation interferes constructively, i.e.
when n is an even integer in Equat ion 1, and then the intens ity at
detector is 1/2 I(vl)' This shows that a Michelson interferometer
can have a maximum efficiency of only 50 percent.
In the configuration shown in Figure 1 only the beam I'(8)Det
is measured. As can be seen from Equat ion 2, there are two part s to
I'(8)Det. One part is constant which is equal to 2RJ11TiilI(vl), and
may be considered a DC component. The other part, 2RillTlilI(1il) cos
2~1i18, is modulated or an AC component. The composite of the AC and
DC parts is called the interference record, whereas the AC component
alone is called the interferogram. Only the interferogram is
recorded; thus, Equation 2 becomes:
I( 8) Det = 2RiilTill I(i" 1 )cos ~vl 8 [6]
-If we let
Q(Vl) = 2RliJ.Till [7]
then Equation 6 becomes
[8]
390 James A. de Haseth
Equation 8 describes the situation when the source is monochro-
matic, yet FT-IR is most useful when the source is polychromatic. By
integrating over all possible values of Vwe obtain:

[9]

Infrared detectors do not measure the energy of the photons, but

rather the power (watts) of the radiation striking the detector. The
detector has an AC-coupled amplifier which converts the detector
response to a vo lt age, and thus it wi 11 have a response funct ion
R(l1. The detector output E(8)Oet is equal to:
E(8)Oet = R(;;}I(8}Oet (volts) [10]
From Equat ion 9

[11]

where
B(ii) = Q(v}R(li}I(OV)

Equation 11 indicates that E(8)Oet is the cosine (or real) Fourier

transform of B(v}. Unfortunately, this is rarely the case due to
the practical constraints of the apparatus. A frequency-dependent
phase error, 81/, is often present and affects the phase angle 2,"p8.
The source of this phase error can be misalignment of the interfero-
meter, dispersion by the beamsplitter, errors in data acquisition, or
caused by electronic filters in the detector amp 1ifier. If the phase
error is considered Equation 11 becomes:

E(8 }Oet = f:B(1i}COS(2Jrrl8+8v)dYi [12]

This is equivalent to
E( 8)Oet = f:B(v)e27riV8dii [13]

where i = Y-I. Equation 13 states that the interferogram, that is,

the electrical response at the detector E(8)Oet' is the complex
inverse Fourier transform of B(y). B(;;) is the spectrum of the
polychromatic source. This spectrum can be calculated from the
measured interferogram by performing the complex Fourier transform of
Equat ion 13, hence,
 =1:
Fourier Transform Infrared Spectrometry 391

8(ii) E(8) Dete-21Ti ii8d 8 [14]

An example of an interferogram is given in Figure 2. Although

this interferogran is nearly synmetrical a small amount of phase
error is present. The short leading side of the signal is used to
calcul ate the phase error and correct the spectrum. The methods of
FT-software are discussed elsewhere in this book.

SPECTRAL RESOLUTION AND APODIZATION

To obtain an accurate represent at ion of the spectrum 8( iT), the
interferogram E(8)Oet must be sampled for all possible values of 8,
according to Equation 14. In practice it is not possible to collect
all values of 8, that is, the maximlJll optical path difference (or
retardation) must be limted to some practical value, L. If inter-
ferogran data are collected in the interval of optical retardation
extending from -L to +L, then Equation 14 becomes

[15]

Figure 2. A typical interferogram of the mid-infrared region.

392 James A. de Haseth
The spectrum B(v) is affected by this change, consequently it must be
renamed B'(v). This is equivalent to multiplying E(8)Oet over the
original limits by a truncation function U(8) so that

[16]

where
1. 181<L
U(8) = { [17]
o. 181> L
Equation 16 states that the spectrum B'(~) is the complex Fourier
transfonn of the product of E('b }Oet and U(8). By recall ing that the
product of two funct ions in one domain (distance, 8) is the
convolution of the functions in the other domain (spectral, v) then
B'(v) = B(~)*S(v) [18]
where S(v) is the Fourier transfonn of U(8). Upon taking the Fourier
transform of U(8), the rectangular truncation function, one obtains:

S(~) = f\-21ri va d8 = 2L sinc(27riiL) [19]

-L
S(v) is known as the instrument line shape function (ILS) and is
illustrated in Figure 3. The effect of convolving the spectrum B(v)

ItS) B(Y)

I
j
-L o L
S

Figure 3. A schematic representation of a rectangular truncation

function and its Fourier transform, the sinc function.
Fourier Transform Infrared Spectrometry 393

with SOi} is to cause all the absorption or emission bands to modify

their profiles from the normal Gaussian or Voigt functions.
The problem with using a rectangular function to truncate the
interferogram is the production of the ringing or "feet" around the
infrared absorption or emission band. This ringing produces an error
in the spectrum; it affects the absolute absorption intensity of the
band itself, and the feet may interfere with the absorption intensity
of any closely spaced band.
The IlS not only distorts the true band shape in terms of the
"feet", but the magnitude of l determines the width of the IlS. It
can be shown that the full width at half height (FWHH) for a sinc
function is 1.21/l. Consequently as L becomes greater, the ILS
becomes narrower and the resolution of the spectrum increases, but is
of course limited to the inherent sample resolution.
As might be expected the rectangul ar truncation function may not
be desirable because it produces the feet around bands which incl ude
regions of negative signal. A method by which the effect of the feet
can be lessened is by apodization (from the Greek, meaning "without
feet"). If the rectangular truncation function is replaced by a
triangular function, as shown in Figure 4, the resulting line shape
is equivalent to a sinc 2 function. In this case there is no negative
region in the bandshape, and the relative amplitudes of the feet have
been reduced by a factor of four. Although these benefits are
readily apparent, the use of a triangular apodization function has

lIS) B (V)

-L o L o
E. ii

Figure 4. A triangular apodization function and its Fourier

transform, the sinc 2 function.
394 James A. de Haseth
other effects. The FWHH is increased to 1.79/l, which is consider-
ably wider than the IlS.
As was stated above the apodization functions affect the resol u-
tion of the spectrum. The measurement of resolution is rather
arbitrary, and various criteria have been used. For example,
resolution is often measured by the ability to separate bo closely-
spaced bands. In some doctrines resolution is not complete unless
the signal basel ine is recorded between bo bands. This is a rather
stringent criterion. Another that is often accepted is the Rayleigh
criterion. The Rayleigh criterion states that two bands are resolved
if the first zero val ue of one band is at the peak of the other band.
This al so assumes the magnitudes of the two bands are nearly
identical. When the two bands are recorded according to this
criterion, the signal between the two bands is approximately eighty
percent of the full band height. The Rayleigh criterion is
illustrated in Figure 5 for two bands with a sinc 2 function profile.
As can be seen the presence of two bands can be easily recognized.
When a rectangul ar truncat ion funct ion is used on the inter-
ferogram, the inherent resolution of the spectrum can be shown to be
1
Resolution (IlS) [20J
2l
On the other hand, it can be shown that if the Rayleigh criterion is
obeyed and a triangular apodization function is used the resolution
becomes
1
Resolution (triangular) [21J
L

ih (vo + 1.79/L)
Figure 5. The Rayleigh criterion for resol ut ion of two sinc 2
functions of equal amplitude.
Fourier Transform Infrared Spectrometry 395
By comparing Equations 20 and 21, the resolution decreases by a
factor of two in going from rectangular truncation to triangular
apodization.
Other apodization functions are possible, such as algebraic and
trigonometric funct ions. It has been sho\'KI that no matter what
apodization function is used, there is a trade-off between FWHH and
the rel at ive ampl itudes of the feet. 4 As a result there is no such
thing as a perfect apodization function. Line-widths cannot be kept
narrow and ringing eliminated simultaneously, hence compromises must
be made.

FT-IR DATA COLLECTION

Thus far descriptions of the basic optical components and
mathematical theory have been given, yet nothing has been mentioned
as to how the interferogram is acquired. Figure 2 is a single-scan
interferogrilll from an FT -IR and to all appearances is an analog
signal. As shown in Equation 13 the interferogram is indeed con-
tinuous. Due to practical constraints it is not possible to store
a continuous interferogram in a digital computer: the storage
allocat ions on even the 1argest systems would be quickly exceeded,
hence the interferogram must be sampled at discrete intervals. The
size of the interval must be carefully selected so as not to intro-
duce errors into the computed spectrum. The choice of sampling
interval is made according to the Nyquist criterion.
It is known that to avoid the phenomenon of spectral folding or
aliasing, one must sample at a rate of at least twice the maximllll
frequency in the bandwidth being measured. This is a criterion of
Nyquist's Sampling Theorem. In the mid-infrared region for example,
the region extends from approximately 4500 to 400 wavenumbers. In
this case, it is necessary to modul ate and record the signal fran the
inteferometer at a frequency of at least 9000 wavenumbers. It is not
always convenient to have a sampl ing interval exactly twice the
maximum frequency of the highest frequency in the bandwidth. As a
result, reference is generally made to a stable, very high frequency,
such as a single-mode He:Ne laser.
A single-mode He:Ne laser has an emission at 632.B nm (visible,
red) which corresponds to a frequency of -15,BOO wavenumbers. The
reference He:Ne laser beam is coupled to the moving mirror of the
interferometer by either its own beamsplitter in parallel with the
infrared beamsplitter, or directly through the infrared optics. The
laser radiation is detected separately from the infrared radiation
and is recorded as a cosine wave, as indicated by Equation 6.
According to the Nyquist sampling criterion, to correctly sample the
He:Ne signal data would have to be collected at twice its frequency,
396 James A. de Haseth
or at - 31,600 wavemlllbers. If the AC component of the laser radia-
tion is measured, a frequency of 31,600 wavenumbers is achieved by
sampling the signal every time the cosine wave equals zero. Conse-
quent 1y if we samp 1e the 1aser int erf erogr am at every second zero
value (or zero crossing), sampling is made at a frequency of -15,800
wavenumbers and the maximum frequency that can be faithfully recorded
according to the Nyquist sampling criterion is one-half that fre-
quency, or -7,900 wavenumbers. Using every second zero crossing is
called undersampling by a factor of two, or alternatively, using an
undersampling ratio of two. This ratio is most often used in mid-
infrared spectroscopy as it will conveniently cover the bandwidth
4500 to 400 wavenumbers.
Higher undersampling ratios may be used. For example, a ratio
of four would cover the range 3950 to 0 wavenumbers. This may be
convenient if the upper limit of the bandwidth can be restricted to
below 3950 wavenumbers. Of course, because of spectral aliasing,
the bandwidths, 7900 to 3950, 11850 to 7900 and 15800 to 11850 wave-
numbers will all be sampled accurately at the same time. Usually
only one of the potential bandwidths is wanted; hence, the appro-
priate optical and electronic filters mu~t be used to reject the
unwanted ~andwidths. By the use of optical and electronic filters
high undersampling ratios may be used to collect interferograms of
spectra of limited bandwidth. This is especially useful for high-
resolution studies where large interferograms are collected [Resolu-
tion «l-1, see Equations 20 and 21]. If high undersampling ratios
are used, interferogrc1l1 storage and data transformat ion time can be
minimized.
It is possible to apply the rapid-scanning Michelson interfero-
meter to far-infrared studies (below 400 wavenumbers). Although the
conversion of most commercial rapid-scanning FT-IR spectrometers to
the far-infrared region is a simple task (vide infra), many far-
infrared spectroscopists employ slow-scann~interferometers, with
which the data are collected over a relatively long time period for
each sampling interval. These spectrometers often use the fringes of
a Moire grat ing to provide the sampl ing interval. Such devices are
not sufficiently accurate to be used by rapid-scanning interfero-
meters, yet they are adequate for the relaxed tolerances of slow-
scanning instruments at long wavelengths.
At the high frequency end of the mid-infrared region is the
near-infrared. This can be adequately sampled with the He:Ne laser
system at the appropri ate undersampl ing rat io and by the use of the
correct filters. Rapid-scanning Michelson interferometers have been
extended to the visible and ultraviolet regions of the electromag-
netic spectrum. 5- 7 As was discussed above the He:Ne laser can be
used to sample frequencies accurately up to and including 15,800
wavenumbers, if an undersampl ing rat io of 1 is used (every zero
Fourier Transform Infrared Spectrometry 397

crossing). By chopping the He:Ne cosine signal into smaller seg-

ments, that is by sampling the max ima and minima of the signal and
half-way between zero and the maxima or minima as well as at the zero
crossings, the bandwidth can be extended to 63,200 wavenumbers (or as
high as to 158 nm).

INSTRUMENTATION
The function of the interferometer in a Fourier transform infra-
red spectrometer has been presented. An FT-IR spectrometer optical
layout is now described and information is provided for each element
of a typical spectrometer design. A schematic of a typical FT-IR
optical design is given in Figure 6.
Source and Associated Optics
In the mid-infrared region the source is generally a filament in
which a resistive wire is wrapped around a ceramic element. Another
o
E

K
Oireclion of
mirror travel
M

Figure 6. Schematic optical layout of FT~IR spectrometer. A =

Source, B = Off-axis paraboloid, C = Beamsplitter, D =
Fixed mirror, E = Movable mirror, F = Off-axis paraboloid,
G = Sample point, H = Off-axis ellipsoid, I = Infrared
detector, J = Visible beamsplitter, K = Fixed mirror, L =
He:Ne and white light sources, M= He:Ne and white light
detectors.
398 James A. de Haseth

coomon source is a globar which is a sil icon carbide rod electrically

heated to about 1300°C. The source is depicted at point A in the
schematic in Figure 6. The source radiation is collected by a colli-
mating mirror B.

In the above treatment of FT-IR the source has been considered

to· radiate fran a point. If the coll imator is perfect, the rays
entering the interferometer are parallel. This was assumed in the
treatment given above. In actual fact the source is not of point
dimensions but subtends a solid angle which we shall call.o.. In this
case the coll imator is unable to produce a parallel ray pattern;
hence, some rays oblique to the parallel path will have a larger
retardation than the on-axis rays. If the source sol id angle becomes
too great the path differences between the off-axis and on-axis rays
may be considerable and resolution will be lost. As the retardation
becomes larger each sampling interval has a higher probability of
having information about more than one path difference. It can be
shown 8 that the resolution is dependent upon the solid angle
subtended by the source by the relationship

Resolution = 'ji'fL/2.". [22]

where iii is the frequency (in wavenumbers) of the band whose resolu-
tion is being measured. As the resolution is frequency dependent, a
maximum allowable sol id angle for the source may be calcul ated by
using the maximum frequency in the spectrum to be measured. Although
the size of the source is fixed, the sol id angle it subtends may be
reduced by the use of an aperture. Many optics designs incorporate

Ellipsoidal Mirror
Mirror

Parallel Beam
Aperture to Interferometer
'---

Mirror --""'---;J-~ Mirror

Figure 7. Source optics with prOV1Slon for aperture. (As used in

Digilab R FTS Systems).
Fourier Transform Infrared Spectrometry 399
a focal point before a final coll imation mirror so that apertures may
be inserted for high-resolution studies. A schematic of such a
source opt ical design is presented in Figtq'e 7. In the schematic in
Figure 6 no such allowance is made, yet if an aperture is put at the
sampling point G it has the same effect as if it were placed before
the interferometer. The funct ion of the aperture is somewhat ak in
to slit width in a dispersive spectrometer.
The mid-infrared soures can be used in the far-infrared region
down to frequencies of approximately 100 wavenumbers. Below this
region a high-pressure mercury vapor lamp can be used, provided the
ultraviolet and visible frequencies are removed by the suitable
optical filter. This filter is usually carbon impregnated poly-
ethylene. In the near-infrared region a tungsten filament lamp
functions well as the source.
Beamspl itter
The beamsplitter is labeled C in Figure 6. The beamsplitter is
usually pl aced at 45 between the ho mutually perpendicul ar mirrors
0

in the interferometer. Beamsplitters must have low absorption pro-

perties and high reflectance and transmittance. The beamspl itter
material has a high refractive index in order to affect these
criteria. For the mid-infrared region the actual beamspl itter
material is either germanium (refractive index, n ~ 4.0) or silicon
(n ~ 3.5). As these beamsplitters are by necessity thin films, the
beamsplitter material must be deposited onto a suitable substrate, as
the material itself is too brittle to have any durabil ity as a self-
supporting film. The suitability of the substrate is determined by
its bandpass and its refractive index as well as the abil ity to
polish and coat the substrate. The refractive index of the substrate
should be as low as possible to minimize reflections fran the surface
and the substrate should be pol ished to one wavelength fl atness of
the shortest wavelength to be measured. These criteria are satisfied
by salts such as NaCl, CsI and KBr. Potassium bromide is the common
choice as it has a lower frequency cutoff than NaCl and is less
deliquescent. Even though CsI has a lower frequency cutoff than KBr,
it is too plastic to retain a high degree of flatness, hence the high
frequency limit is considerably lower than that of potassium bromide.
Because the beamspl itter is on a substrate which affects the path-
length in only one arm of the interferometer, a compensator plate of
the substrate materi al is pl aced in the other arm of the interfero-
meter. This will lessen the phase error. Near-infrared beamsplit-
ters have either a calcium fluoride or quartz substrate with a ferric
oxide beamsplitting film.
Far-infrared spectrometers usually use a self-supporting film of
polyethylene terphthalate, which has the trade name Mylar (or Melinex
in Europe). The Mylar sheet has sufficient mechanical strength not
400 James A. de Haseth
to require a substrate. The thickness of the sheet determines the
wavenumber range passed by the beamsp1itter, but these sheets are
usually between 50 and 5}Lm thick. A Mylar beiJllsp1itter has a
relatively low refractive index (n ~ 1.8) and the efficiency of the
beiJllsp1 itter is highly dependent upon the incident angle of the
radiation. Highest efficiencies are found when the incident angle is
greater than 75°.9 Other beamsp1 itters for the far-infrared are
wire meshes; whereas these have not been as widely used as the My1 ar
beiJllsp1 itters, at least one manufacturer supp1 ies a wire grid beam-
sp1 itter.
Interferometer Mirrors and Drive
The fixed and moving mirrors of the Michelson interferometer are
1abelled D and E, respectively, in Figure 6. The a1 ignment of the
plane mirrors must be stringently controlled to maintain the integ-
rityof the interferometer. If one mirror tilts slightly so as to
perturb the alignment, the beam that is reflected from that mirror
will no longer coincide with the beam that has been reflected from
the other mirror. That is, if mirror E tilted some small angle d8,
the beams transmitted by the beamsp1 itter would strike E at an angle
of incidence equal to d8 and have an equal angle of reflection. The
reflected beams would not be coincident with the corresponding beams
from D at the beamsp1itter. Beams from different areas of E will
travel different distances, possibly sufficiently different so that
they may not all interfer constructively or destructively with the
beams from D. In other words, the detector will record several dif-
ferent retardations at a single sampling interval. This effect is
simi1 ar to that of an extended source (vide supra) and will leeK!
to a degradation of spectral reso1ution:--uT course, as the retarda-
tion becomes larger, and thus the resolution greater. the tolerances
for d8 become smaller. One way to alleviate the problem is by the
use of retroreflectors, Which will always produce parallel input and
output beams .10
Various mirror drive systems have been used. For far-infrared
spectrometers where the wavelengths are long and the tolerance for
mirror tilt is large, simple drives such as lubricated pistons or
screw drives have been used. In the mid-infrared most moving
mirrors are supported on air bearings that are moved rapidly by a
so 1eno i d. Sc ill rates vary fran abo ut one mill imeter per second to
about 6 cm per second. These air bearing drives can maintain optical
alignment to produce spectral resolutions of approximately 0.02
wavenumbers. Traditionally, step-scan interferometers have been
used for higher resolutions with which the moving mirror has been
stopped at each samp1 ing position and real igned. The signal is
averaged over a relatively long time period at each position.
Recently, one commercial supplier of FT-IR spectrometers has started
providing systems in which the drive is rapidly driven by pulling a
carriage through a hollow tube. The carriage is supported by roller
Fourier Transform Infrared Spectrometry 401
bearings and the moving mirror allowed to tilt. The fixed mirror is
tilted in unison with moving mirror by employing a dynamic al ignment
system. Using this system resolutions as hi~h as 0.002 wavenumbers
in the mid-infrared have been demonstrated. 1
Sampl ing Optics
The output of the interferometer is collected and focused at the
sampling position. The focus is usually 3 to 12 mm in diameter, but
this value often is not determined by the focusing mirror (Figure
6F), but by the sol id angle subtended by the detector. Commercial
FT-IR systems have beam diameters of about 2.5 to 7.5 cm as the beam
emerges fran the interferometer. The sampl ing optics do not produce
a point focus, which is preferred as the sample will cover a finite
area. Figure 6 illustrates a single-beam spectrometer design, yet
instrllllents are available that have a "double-beam" option. Unlike
many dispersive systems an FT -IR beam is not chopped and rat ioed
through sample and reference cells. In a typical FT-IR experiment
the sample and reference are always separated in time, i.e. one is
collected, then the other; and, they are sometimes separated in
space. The sampling optics (labelled F, G, and H in Figure 6) may be
replaced by two parallel sets for the double beam option. Either
beam can be selected at the operator's discretion. One set holds the
reference, the other the sample. An example of such an optical
set-up is given in Figure 8. In this set-up, the two flipper mirrors
are coupled so that the beam is alternatively focused at the two
sampling points.

Sample chamber~

Parabalic Mirror I Mirror

Sample
: I

Reference :~:.. T~---\----l

; I I ..

Output Beam ~ii_~~~~~~~~ __ ::~

from
Interferameter Parabolic - - - I I Mirror
Mirror

Figure 8. Double-beam optics for sample and reference spectra

collection. (Digilab R FTS Systems.)
402 James A. de Haseth
In practice if a sample and reference are collected in a single
experiment, usually signal-averaging is alternated for a short time
between the sample and referenc~ to compensate for long term drifts
in the source or detector and associated electronics.
An interesting aspect of FT-IR is that it makes little differ-
ence where the sample is placed in the system. Equal results are
obtained most often by pl acing the sample before the interferometer.
(For example, the source and detector in Figure 6, A and I respec-
tively, could be interchanged.) Scattered radiation is not a problem
as it is in dispersive instruments, because scattered radiation
wi 11 be constant and only change the DC component of the interference
record. It is often preferable to place the sample in one position
or the other. Emission measurements require that the sample be the
source. Absorption studies of high temperature samples, such as
vapors emitted by molten salt cells, should have the samples placed
after the interferometer so that infrared radiation emitted by the
sample is not modulated.
Detector
The type of detector used in an FT-IR spectrometer is highly
dependent upon the bandwidth (i .e. the spectral frequencies), the
modulation rate of the interferometer, and the intensity of the.
radiant flux. Several types of detectors are used in the infrared
regions: photoconductive, photovoltaic, bolometers, pyroelectric and
Golay cells. A detailed discussion of detectors may be found else-
where. 12 In general, the photovoltaic and photoconductive detectors
can be used in the near- and mid-infrared regions as rapid response,
high sensitivity detectors. Usually the bandwidths are 1imited and
will not cover the total range passed by the beamspl itter. Examples
of such detectors are given in Table I. As can be seen from the
Table, these detectors are cooled. The value D~o ** corresponds to
the sensitivity of the detector and may be considered a figure of
merit associated with the detector, where ~o is the long wavelength
limit of the detector. This val~e is independent of detector field-
of-view and has the units cm Hz /2 ster l/2 /watt. The larger
the value of D~~, the more sensitive the detector.
Other detectors that are useful in the near- and mid-infrared
regions are bolometers and pyroelectric detectors. Both these
detectors have very 1arge bandwidths and can operate at room temper-
ature; however, they have 10n2.response times compared to the photo-
detectors and they have low D~:'s. Pyroelectric detectors are
useful in the far-infrared reg~on with rapid-scanning spectrometers
whereas Golay cell detectors are often used with slow scanning
far-infrared interferometers. These cell s are modul ated at or below
20 Hz.
 "o
I::
-s
.....
(t)
-s
Table 1. Infrar ed Detect or Chara cteris tics* -I
-s
III
~
VI
Approx imate Range Opera ting .......
o
~~ -s
3
Name (wavenumbers) Tempe rature (K) (cm Hz1/2 Sr1/2/ watt) ......
~
.......
-s
~
(t)
Trigly cine sulfat e 10-5000 300 a.
(pyro electr ic) 2 x 109 tn
-c
(t)
n
Golay cell 10-1000 c-i"
300 6 x 10 9 -s
o
3
Mercury Cadmi lJI1 700-3800 (t)
77 3.1 x 1010 c-i"
Tell uride -s
(photo condu ctive) '<

Indium Antimo nide 1850-6600 77

(photo vo 1ta ic) 6 x 1010

Therm istor 5000-10000 300

(bolom eter) 2 x 108

* This table is not exhaus tive and only a few of all possib
le detect ors are presen ted.

.j:::o
o
w
404 James A. de Haseth

At first inspection it may appear desirable to use the most

sensitive detector available, assuming the bandwidth and modulation
spectrometer criteria can be met. This may not necessarily be the
case because the spectrometer is constrained by the dynamic range of
the data acquisition system. If a broad band source is used, the
interferogram will have a large value at the centerburst (approxi-
mately the point of zero retardation, i.e. both interferometer
mirrors equidistant from the beamsplitter), assuming the phase error
is small. The wings of the interferogram will have small val ues
relative to the centerburst (see Figure 2). The detector responds to
the radiant flux by producing an analog signal which is digitized for
the data system. Analog-to-digital converters (ADC's) are usually 12
to 16 bits. A 15-bit ADC, for example, has a full dynamic range of
about 30,000. If the signal-to-noise ratio (SNR) of an interferogram
being measured is greater than 30,000 the noise level will be limited
by the ADC, rather than by the inherent noise level in the Signal or
detector. This SNR is not unreasonable for an interferogrilll,
espec ially for absorbance measurements.

The interferogrilll signals are signal-averaged by coadding the

digitized interferogram values. In order to avoid adding noise to
the signal-averaged data, it is often necessary to use computer word
lengths of at least 27 bits. 13
Sampl ing and Trigger Optics

It was expl ained above that data must be sampled at a rate at

least twice the frequency of the highest frequency to be recorded. A
convenient way to do this is measure the mirror movement, that is,
optical path difference, by the fringes fran a He:Ne laser beilll
passed through the interferometer. The measurement using the laser
bean can be accomplished in ho basic ways, directly through the
beamsplitter or remotely. As the beamsplitter material is often
opaque to visible He:Ne radiation, remote sensing is used as shown in
Figure 6. A laser source is situated at L and directed to a visible
radiation beamsplitter J which directs the radiation to the fixed
mirror K and to the mov ing mirror E. As E is sc anned a detector at M
records the He:Ne cosine wave and this signal can be used to measure
the optical retardation.

An alternative method is to use a beamspl itter that transmits

both infrared and visible radiation, such as Sb2S3 on a KBr sub-
strate. The only beamspl itter in the system would be that at
position C in Figure 6. The He:Ne bean could be passed through an
edge of the main beamspl itter and recorded physically adj acent to the
infrared radiation. Otherwise, small holes could be cut in mirrors B
and F so that a laser source could be put behind B and a visible
detector behind F. The laser bean is then coincident with the infra-
red beam and sampling errors due to mirror tilt are minimized. For
Fourier Transform Infrared Spectrometry 405

this alternate method, the beamsplitter does not have to be trans-

parent to visible radiation, only a small area need be.
In very high resolution spectrometers, such as the dynamic
alignment spectrometer mentioned above, three laser beams are run in
parallel to the infrared beam. These three laser beams are
positioned at the apices of an equilateral triangle. Laser fringe
measurements are taken on all three beams and any error, (that is, if
all the zero crossings are not simultaneous) is corrected by tilting
the fixed mirror. By this method, both mirror tilt and sampling
frequency measurement are accomplished.
It is evident there are two sources at position L and two
detectors at M in Figure 6. The second source is a broad-band white
light source and it has its detector at M. The function of the white
light is to produce an interferogram centerburst, offset to the
infrared interferogram centerburst. The white light centerburst,
arranged so that it occurs in the scan before the infrared center-
burst, serves as a trigger for the commencement of data collection.
As the moving mirror is scanned, the white light interferogram is
monitored. When the intensity of the interferogram reaches a preset
threshold (in the centerburst) data collection of the infrared signal
is begun on the next laser fringe. This is a reliable method to
assure that data collection is started at the same retardation each
scan, making coaddition for signal averaging possible. The white
light signal can be remote as shown, or run through an infrared
beamsplitter that will also transmit visible radiation. This case is
analogous to the He:Ne system.
Computer System
Commercial FT-IR spectrometers have dedicated computer systems
to collect, store, transform and output the data. These computer
systems range from limited-task microcomputers to versatile, powerful
minicomputers. Computer systems are general to all Fourier transform
techniques and will not be discussed in this chapter.

FT-IR ADVANTAGES
At this point the FT-IR experiment may appear to be a rather
difficult way to produce an infrared spectrum. Clearly, there must
be some advantages to the technique to justify its existence. There
are three distinct facets of FT-IR which make FT-IR superior to con-
ventional dispersive infrared spectroscopy.
406 James A. de Haseth
Fe11gett Advantage
The Fellgett or multiplex advantage deals with the fact that a
Fourier transform spectrometer records data from the entire spectral
region throughout the experiment. This is quite different to the case
with a di spers ive spectrometer, as the grat ing or pr i sm instrument
only measures a narrow bandwidth at any time. The measurement band-
width of the dispersive spectrometer is regulated by the instrument's
exit slit. This difference has important effects on the acquisition
of data.
It is well-known that the signa1-to-noise ratio (SNR) in many
data acquisition systems can be improved by signal-averaging. This is
accomplished by either sampling the signal for a long period of time
so that each datun is more accurately measured, or by rapidly
measuring the entire signal repeatedly and coadding the results. In
spectroscopy account must be taken of the number of spectral
elements, N, that are being collected. The number of spectral ele-
ments is simply the bandwidth divided by the resolution. In a
grating or prism spectrometer, when the noise is random and independ-
ent of signal strength, the SNR should decrease with sampling time,
t, but should also depend upon N, thus
SNRO a: (t/N)l/2 [23]
where the subscript 0 refers to a dispersive instrument. An inter-
ferometer obeys different rules because of the Fe11gett advantage.
If the noise is random and independent of signal strength, the SNR
for the interferometer (SNRI) decreases with samp1 ing time, but is
independent of N. As stated above, the entire bandwidth is being
sampled at all times, thus N plays no role in the signa1-to-noise
ratio, and
SNRI a: t 1/ 2 [24]
Taking the ratio of the two SNR's one has:
SNRI
- = Nl/2 [25]
SNRO
That is, in a fixed time an interferometer system will produce a
spectrum with a signa1-to-noise ratio greater thill that of a disper-
sive system by the square root of the number of spectral elements.
In a high resolution system this can be significant, on the order of
10 to 103• In the infrared region the noise is generally independ-
ent of signal strength so the advantage holds.
Fourier Transform Infrared Spectrometry 407

The multiplex advantage is not val id in the visible region of

the spectrum as the noise is signal strength dependent. The differ-
ence arises fran the fact that optical detectors are better than the
infrared detectors, hence the signal is source limited in the visible
region. It can be sho\'l1 that SNRO and SNRI are equal in the
visible region, hence there is no multiplex advantage for an inter-
ferometer system.
Jacquinot Advantage

An interferometer has no slits, so theoretically there is no

attenuation of the source intensity by the spectrometer. This is a
basis of the Jacquinot advantage (also called the throughput or
etendue advantage.) Jacquinot stated that the brightness of an image
is equal to the brightness of the source in a lossless optical
system. Therefore, a slitless Michelson interferometer assures that
all the information (radiant flux) will reach the detector at all
times, dependent upon the interference pattern prov ided by the inter-
ferometer. This is not so in a dispersive instrument because the
radiant flux is attenuated by the spectrometer exit slit. It can be
shown that the throughput for a Michelson interferometer is consider-
ably greater than that of a pri sm or grating instrument. Throughputs
for a Fourier Transform Infrared instrlJllent can be in excess of a
factor of 100 greater than those of dispersive instruments.

Unlike the Fellgett advantage, throughput is not dependent upon

the wavelength or detector used in the instrument. For this reason
interferometers enjoy the ability to record and detect spectra of
weakly emitting or weakly absorbing samples in either the optical or
infrared regions of the spectrum. This property has been heavily
exploited in the field of FT -IR.

Connes Accuracy
A third advantageous property of interferometers can be found in
the inherent spectroscopic accuracy of the technique. The use of a
He:Ne laser to measure the optical retardation has been discussed
above. If this laser is a single-mode laser, and thus monochromatic,
the optical retardation can be measured to a high degree of accuracy.
If the instrument is correctly manufactured data will be recorded in
a prec i se manner. Accur ate and prec i se measurement in the interfero-
gram yields accurate frequency assignment to the spectral val ues.
With most interferometers it is possible to assign spectral features
to better than 0.01 wavenumbers of the true position.
 408 James A. de Haseth
APPLICATIONS (f FT -IR
Fourier Transform Infrared Spectrometry has been applied to a
vast array of chemical problems. The majority of these problems are
beyond the scope of this review, consequently, this chapter will be
1imited to those techniques that have been developed or greatly
advanced within the last few years. Several recent reviews have
appeared that deal with the established techniques and the reader is
referred to these reviews for further information. 14 - 16
Infr ared spectroscopy often is app 1 i ed to absorpt ion stud ies of
organic and inorganic materials. In this area Fourier Transform
Infrared Spectrometry often has 1ittle to offer over and above con-
ventional dispersive spectroscopy, especially in the area of solid
and 1 iquid or condensed phase samples. FT -IR excels in the area of
high resolution spectrometry, therefore it is well applied to low
pressure gaseous samples or matrix isolated samples where the
inherent sample resolution may be exploited. Fourier Transform
Infrared Spectroscopy may be very useful for absorption studies of
-1 iquid and sol id samples if the samples are very minute, where the
throughput advantage can be used to obtain a suitable spectrum. FT-
IR is anployed often in qual ity assurance appl icat ions. In qual ity
assurance appl icat ions it is necessary to record spectra as rapidly
as possible, as there may be a great number of samples for analysis
in a short time period. FT -IR has the abil ity to record high signal-
to-noise ratio spectra rapidly. This is due to the Fellgett or
multiplex advantage. As was discussed above, a Fourier Transform
Infrared Spectrometer will produce a much higher signal-to-noise
ratio spectrum than will a dispersive spectrometer if the collection
conditions are the same for both instruments. Conversely, if a fixed
signal-to-noise ratio is sought in a spectrum then the FT -IR
spectrometer can produce this spectrum faster than an equivalent
dispersive spectrometer.
Those areas Wlich have received great attention in Fourier
Transform Infrared Spectrometry recently have been the chromato-
graphic infrared spectrometry hybrid systems. This includes gas
chromatography-infrared spectrometry and 1 iquid chromatography-
infrared spectrometry. Two other areas of current interest are
photoacoust ic-infrared spectroscopy and the use of infrared spectro-
scopy to determine impurities in semiconductor materials.
GC/FT -IR
The potential to interface a gas chromatograph to a rapid-
scanning infrared interferometer was recognized shortly after the
interferometers became avail able. The earl iest studies used low
resol ut ion interferometers.17 ,19 This work was quickly expanded to
laser-referenced interferometers capable of high-resolution
studies.1 9 ,20 The goal of this work was to collect the infrared
Fourier Transform Infrared Spectrometry 409
spectrlJll of gas chromatographic effluents as they eluted from the
column. Commercial FT-IR systems accomplished this by trapping
eluates in a small volume gas cell and signal-averaging a number of
scans to provide a good spectrum. This necessitated the shut-down of
the chromatographic process and the ensuing chromatographic degrada-
tion; or, the chromatographic process was allowed to continue, but
subsequent peaks to the one in the cell were discarded. Only a
fraction of all the chromatographic eluates in a single run could be
adequate ly invest i gated.
SpectroscopiC acquisition of dynamic GC eluates was realized in
1976 by Azarraga. 21 This is accompl ished by pass ing the chromato-
graphic effluent through an internally gold coated glass tube that is
coincident with the infrared beam. This tube or "light-pipe" has
dimensions of approximately 0.7 to 6.0 mm internal diameter and 5 to
125 em overall length. The manufacture of the light-pipe is a
strai~htforward procedure and can be carried out in the labora-
tory. 2 USing this method Azarraga was able to reduce the detec-
tion limits of GC eluates to a few hundred nanograms when capillary
WCOT columns were used. With this technique both packed and
capillary GC systems could be investigated. The volume of the light-
pipe must ,be optimized to provide the best signal from the GC eluate.
The optimum light-pipe parameters have been established by
Griffiths .23 .
The mode of data collection for GC/FT-IR spectra has been to
scan the effluent in the light-pipe at the rate of at least one scan
per second. As capillary GC peaks can be quite narrow, signal-
averaging is not always performed and each individual interferogram
is stored. Although the resolution of the spectrum is low (often 8
cm- 1) a single GC run can produce massive amounts of stored data.
Methods whereby the data handling could be streamlined to locate
interferograms of eluates were sought. Unlike gas chromatography/-
mass spectrometry no direct analog of the total ion current exists in
GClFT-IR. T\\O basic methods of locating eluate interferograms were
devised. One used the Gram-Schmidt vector orthogonalization
algorithm to reconstruct the chromatogram directly from the interfer-
ogram data;24 the other transformed a "short" interferogram to
calculate the absorbance within various user-defined spectral
windows.25 The use of such algorithms permitted the spectroscopist
to selectively store data on the systan disk so as not to exceed the
data storage capabilities of the instrument. These software develop-
ments led to the development of user-oriented packages that include
GC reconstruction, data reduction, Fourier transforms, and spectral
search systems .26
Instrumental development in GC/FT-IR has been toward the
improvement of sgectral detect ion 1imits. Griffiths has ut il ized
dual-beam FT -IR.27 Matching light-pipes are put in both beams of
the system, one containing the eluate, the other a reference gas. 28
410 James A. de Haseth
A block diagran of the optical layout is reproduced in Figure 9. The
advantage to such a technique is that the detector measures the
opt ical subtract ion signal of both beams of the interferometer. In
this situation a high-sensitivity infrared detector can be use to
full advantage without exceeding the dynamic range of the detector.
Griffiths was able to record spectra of GC eluates of 1 ng sample
size 29 (see Figure 10). Commercial systems in a single-beam FT-IR
configuration are now available where the detection limits are 5 to
50 ng per eluate from a capillary column.30 M example of a
spectrum from this system is given in Figure 11.

1------------- ---I
DIGILAB I

MODEL 296

IR
,
SOURCE+---

I --
OFF-AXIS I

PARABOLOI~AL
MIRROR I

,
I

,------ I
L____ _

~t-t-1-----;- MC T D E TE C TOR

L ________ J

Off . A X I S

ELLIPSOID

Figure 9. Dual-bean spectrometer optical layout as used for GC-IR

measurements. (Reproduced from reference 28, by permis-
sion of the American Chemical Society, copyright 1978).
Fourier Transform Infrared Spectrometry 411

Other approaches to the improvement of the GC/FT-IR technique

have been directed to the chromatography. One such study has been
the employment of two-dimensional chromatography to separate unre-
solved chromatographic peaks. 31 The system trapped the peaks of
interest in an activated-charcoal cold trap, then rechromatographed
the peaks on a different column. No loss of spectral sensitivity was
found. By another method, very low sensitivity eluates have been
chromatographically separated and each eluate frozen onto a cryogenic
mirror. 32 This allowed the spectroscopist to improve sensitivity by
signal-averaging the trapped eluate. Of course, the dynamic
advantages of GC/FT-IR were lost in favor of spectral sensitivity.
LC/FT -IR
The logical extension of GC/FT-IR is to apply that existing
technology to liquid chromatographic systems. This major difference
between the two technologies is the mobile phase, which is totally
transparent in GC but is never transparent in LC system. There is
always a high solvent:solute ratio in liquid chromatographic
effluents, often on the order of at least 1000:1. As a consequence
there are two contradicting criteria: the cell pathlength must be
increased to improve the absorbance of the solute, but the path-
length must be minimized to decrease the absorbance of the solvent.

w
(/')
0::

1800 1200 em- 1

Figure 10. GC/FT-IR spectra of anisole, collected on the dual-beam

spectrometer. A comparison is made between three solute
quantities A = lOOng, B = 5ng, C = 1ng. (Reproduced from
reference 28, by permission of the American Chemical
Society, copyright 1978).
412 James A. de Haseth
Pit early attempt to produce on-l ine LC/FT -IR spectra used a
0.030-mm pathlength flow-through cell connected to a liquid chroma-
tograph. 33 Spectra were collected every 0.5 sec. in an on-the-fly
recycl ing mode. The resulting spectra at first appearances were
solely of the solvent, but when the pure solvent spectrum was sub-
tracted, low signal-to-noise ratio spectra of the solutes were
evident. A spectrum of 2~g of p-nitrophenol with a signal-to-noise
ratio of 10:1 was shown. It was calculated the original mixture had
a solvent:solute ratio of 2500:1, the solvent being chloroform. This
detection limit is not adequate for all LC systems and overall is not
very practical.
Even though many solvents have regions which exhibit some high
transmittance, they do not transmit at the 100 percent level. Those
bands that are present in the solvent systems absorb very strongly.
If a low concentration solute is present, and it is a strong infrared
absorber, its spectrum may be accurately recorded in those regions
that are free of solvent bands. In those regions where the solvent
,. DISILAB frs-DO<

..

99.5

Figure 11. A capillary GC/FT-IR spectrum of 20 ng of caffeine.

(Spectrum courtesy of D. Kuehl, Digilab; to be submitted
for publication).
Fourier Transform Infrared Spectrometry 413
absorbs very strongly, the spectrun of the sol ute will not be
accurately recorded as the dynamic range of the detector at those
frequencies will be too low. In other words, very high signal-to-
noise ratios will be required to distinguish the solute features.
This is not possible with single-scan dynamic acquisition systems.
In those cases \\tlere a polar solvent is used, such as in aqueous
phase chromatographic systems, much of the bandwidth is obliterated
by solvent and the solute is difficult to detect.
A more recent attempt at LC/FT -IR was made by Vidrine and
Mattson 34 who coupled a gel permeat ion chromatograph to an FT -IR
spectrometer. As some instrumental advances had been made in
spectrometer hardware since the work of Kizer et al. 33 , primarily
in the area of detection, the sensitivities of the LC/FT-IR work was
improved. The cells were of flow-through design with pathlengths
varying from 200fLm to 11.2 mm. When tetrahydrofuran was used as the
solvent, detection limits of 100 ng were achieved, but spectra could
only be recorded in regions of good solvent transparency. Very few
flow-through cell system papers have been published, but exclusion
chromatography is often used because of the relatively low solvent:
solute ratios and the ability to use nonpolar solvents.

FOURIER GUYS

0.75
~

"a: 0.625
~
III
-'
C
a:
ti 0.5
III
A-
lii
~ 0.315
;:
c
-'
III
a: 0.25

o
.000 3600 3200 2800 2400 2000

Figure 12. Spectrun of Indophenol Blue dye by solvent elimination

LC/FT-IR method. 1 ~ of Indophenol Blue produced this
spectrum. (Reproduced fran reference 36, by permission
of Preston Publications, Inc., copyright 1979).
414 James A. de Haseth
Another method to obtain LC/FT -IR spectra is by solvent el imina-
tion. This has been accomplished by Kuehl and Griffiths 36 ,37 by
concentrat ing the LC effl uent onto s intered KCl powder, then evap-
orating the solvent by heating and inert gas flow. The spectra are
measured by diffuse reflectance spectroscopy.28 These workers
were ~le to separate stereoisomers by adsorption chromatography and
record the spectra. Figure 12 reproduces the spectrum of Indophenol
Blue dye obtained from the LC/FT-IR technique. The sensitivities
extended to the submicrogran sample sizes. The primary drawback of
such a system is that only organic, highly-volatile solvents could be
used. The systan could not function if the solute were evaporated
with the solvent, or if the solvent dissolved the KCl powder.
Griffiths has discussed the possibility of using dual-beam FT-IR
for liquid chromatography.39 This in part would reduce the dynamic
range problan by putting the LC effluent in one bean and a solvent
reference in the other. The dynamic range problem would not be
totally el iminated, however, as those regions of high transmittance
will still exhibit a high signal throughput. Other regions may be
essentially opaque, beyond the range of the ADC. Further problems
would occur with gradient el ut ion systems. It would be very
difficult to assure that the solvent make-up is identical in both
beams. For isocratic systems the reference cell could be fixed and
would not require flow-through capabil ity. The feasibil ity of the
dual-beam apparatus for low concentration solutions has been demon-
strated. 40 Spectra of 0.001% v/v anisole in CC14 and 5 ppm
t-butyl methacrylate in CC14 have been shown, but up to 10,000
scans of the interferometer were required.
FT -IR-PAS
Fourier Transform Infrared Photoacoustic Spectroscopy (FT-IR-
PAS) has been developed recently. In this technique the standard

IR Transparent
Window

Sample Chamber

Figure 13. Schematic of FT-IR-PAS sample cell.

Fourier Transform Infrared Spectrometry 415
infrared detector, usually a deuterated triglycine sulfate detector,
is replaced by a microphone. A simple schematic of a typical sample
cell is shown in Figure 13. Both the Fellgett and Jacquinot advan-
tages play important roles in this spectroscopy. The throughput or
Jacquinot advantage is necessary so that sufficient radiant flux will
reach the sample to be converted into heat by the sample. This in
turn can be detected as pressure waves by the detector-microphone.
Conventional spectrometers generally have too weak a radiant flux to
permit photoacoustic spectra being collected in the infrared region.
A Fourier Transform Infrared Spectrometer has a throughput advantage
of 50 times higher than a conventional dispersive infrared spectro-
meter. The multiplex or Fellgett advantage also improves the photo-
acoustic spectroscopic technique as high resolution PA spectra can be
collected. This is not to imply that infrared photoacoustic spectra
cannot be collected on a dispersive instrument, as this has been
demonstrated by Low and Parodi. 41 Low and Parodi were able to
produce low resolution photoacoustic infrared spectra of solids using
a single beam grating spectrometer.
The Michelson interferometer used in a Fourier Transform Infra-
red Spectrometer acts as a modulator of the infrared signal. The
detector receives the modulated signal in the audio frequency range.
The audio frequency is wavelength dependent and of course dependent
on the scan velocity of the moving mirror. For example, if the
mirror moves at the rate of 3 mm per second, the modulation frequency
of the helium neon laser wavelength, 632 nm, is 5 kHertz; on the
other hand, an infrared wavelength of 20~m (or frequency of 500
wavenumbers) has a modulation of only 316 Hertz. As a result, the
FT-IR-PAS spectrum is collected at a series of different modulation
frequencies. It has been shown 42 that signal intensity decreases
as the modulation frequency increases. As a result, the PA spectra
collected on a FT-IR spectrometer tend to have more intense signals
at lower frequencies. Regardless, FT-IR-PAS spectra can be corrected
by various techniques to produce accurate, ratioed spectra. Figure
14 illustrates the comparison between PA spectra and a transmission
spectrum of polymethylmethacrylate.
Although infrared photoacoustic spectra have been collected of
gaseous samples 43 ,44 the primary advantage appears to lie in the area
of solid and surface samples. Very little has been published thus
far in the area of FT-IR-PAS, however, several publications have
involved the measurement of solid samples~46,47 the measurement of
surface samples,48 biological materials,4~ and binary mixtures. 50
Silicon Impurity Determinations
The determination of impurities in silicon is important to the
semiconductor industry. If impurity determinations can be made in
silicon wafers prior to burning and etching of integrated circuits
(IC's) on those wafers, the rejection rate and long-term reliability
416 James A. de Haseth
of the IC devices can be greatly improved. This is of economic
benefit to the manufacturer and consumer al ike. Mlst manufacturers
produce single crystal sil icon from molten sil icon that is handled in
quartz crystal s and heated by graphite furnaces. Thi s is known as
the Czochlarski method. Oxygen and carbon find their way into the
molten silicon by various pathways, but originate in the quartz
crucibles and graphite heaters. These impurities change the electri-
cal properties of the silicon and can destroy the semiconductor
properties.

DIG1LAB FrS-1M)(

WA VE"IUI1BF.RS

Figure 14. Spectra of plexiglass films cast on AgCl. A = PA

spectrum of 0.3 JLm thick fi 1m; B = PA spectrum of 0.7 JLm
thick film; C = PA spectrum of 3.7 JLffl thick film; D = PA
spectrum of 13.8 JLm thick film; and, E = Transmission
spectrum of 13.8fLm thick film. Note saturation effects
in D. (Spectra courtesy of K. Krishnan, Digilab; to be
submitted for publication.)
Fourier Transform Infrared Spectrometry 417
There are various methods by which the impurities may be
quantified~ but dispersive infrared spectroscopy is often the method
of choice.~1,52 This method has serious shortcomings due to inher-
ent problems with dispersive instruments such as poor resolution and
poor wavenumber repeatability in the spectral regions of interest. A
further criterion is that the wafer under test must be matched with a
pure wafer of identical thickness. Only recently have FT-IR spectro-
scopic techniques been applied to the problem and considerable
success has been realized.
The carbon and oxygen content of silicon wafers can be deter-
mined at room temperature. 53 The procedure is to record the
spectrum of the sample wafer and ratio it against a background (empty
cell) spectrum. The resulting absorbance spectrum is stored
on the spectrometer data system. A similar absorbance spectrum of a
pure wafer is recorded. A spectral subtraction is performed between
the two spectra and only the impurity bands of the sample wafer
remain. The peak height can be correlated to impurity concentration.
This process can detect oxygen and carbon down to a few parts per
million. This method has been shown to be very reliable and can be
automated. 54
The impurity bands have a pronounced temperature dependence and
spectral sensitivity and can be improved by lowering the temperature
of the wafer to 20 K or be10w. 55 At least one system has been
developed that is totally automated: it introduces the sample, cools
it, records the spectrum and quantifies the impurities. 56 Such a
system has a sensitivity of carbon detection four times that of the
room temperature system. Cryogenic analysis can improve the oxygen
detection limits by a factor of twenty. Analysis at cryogenic
temperatures is not restricted to oxygen and carbon, but can also be
applied to dopants. These include, phosphorus, boron, antimony,
arsenic, aluminium, gallium and indium. Here detection limits can
exceed a few parts per billion. An example of a silicon wafer
spectrum is shown in Figure 15.
The strength of the Fourier transform method lies in the
throughput of the spectrometer that allows very minute quantities of
impurities to be detected. Furthermore, the absorbance spectral
subtraction can compensate for sample and reference thickness differ-
ences. The automated cryogenic system is capable of handling a
sample every few minutes. 56
 ..,..
-'
00

joo...l f
DJ:GJ:LI\B F JS--l'1>(

.I !~~il
.I I I'
~~~~~
sc

11"00 ,Jloo £I.e

., eIJl! 7100
'WAV.ENlJMBERS
sIw 4JW 3110 c.....
~
ro
Figure 15. FT-IR spectrum of a doped Si wafer at 20 Kelvin. Interstitial oxygen bands are at Vl

1130 cm- I and subst itution carbon bands are at 607 cm- I . (Spectrum courtesy of )::>

K. Krishnan, Digilab; to be submitted for publication). CL

ro
::c
OJ
Vl
ro
.......
::;-
Fourier Transform Infrared Spectrometry 419
REFERENCES
1. Griffiths, P. R., 1975, Chemical Infrared Fourier Transform
Spectroscopy, John Wiley &Sons, New York.
2. Bell, R. J., 1972, Introductory Fourier Transform Spectroscopy,
Academic Press, New York.
3. Steel, W. H.,1967, Interferometry, Cambridge University Press,
Cambridge.
4. Beer, R. and Norton, R. H., 1976, J. Opt. Soc. An., 66, 259.
5. Horlick, G. and Yuen, W. K., 1975, Anal. Chern., 47, 775A.
6. Yuen, W. K. and Horlick, G., 1977, Anal. Chern., 49, 1446.
7. Horlick, G. and Yuen, W. K., 1978, Appl. Spectrosc., 32, 38.
8. Ref. 2, Chapt. 11. -
9. Griffiths, P. R., 1978, "Fourier Transfrom Infrared Spectro-
scopy: Theory and Instrumentation" in Transform Techniques in
Chemistry, Griffiths, P. R., Ed., Plenum, New York, p. 127.
10. Ref. 2, Appendix C.
11. "Specifications of Bomem Model DA3.XX Interferometric Spectro-
photometers," Tech Note 18, Bomem, Inc., 910 Place Dufour,
Ville de Vanier, Quebec G1M 3B1, Canada.
12. Optical and Infrared Detectors, Keys, R. J., ed., Springer-
Verlag, Berlin, 1977.
13. Foskett, C. T., 1976, Appl. Spectrosc., 30, 531.
14. Griffiths, P. R., 1978, "Infrared FourierTransform Spectro-
metry: Appl icat ions to Analyt ical Chemsitry," in Transform
Techniques in Chemistry, Griffiths, P. R., Ed., Pelnum, New
York, Chapt. 6.
15. Fourier Transform Infrared spectroscogy: Application to
Chemical Systems, Ferraro, J. R. and asile, L. J., Eds.,
Academic Press, New York; Vol. I, 1978; Vol. II, 1979.
16. Analytical Applications of FT-IR to Molecular and Biological
Systems, Durig, J. R., Ed., D. Reidel Publishing Company,
Dordrect, Holland, 1980.
17. Low, M. J. D. and Freeman, S. K., 1967, Anal. Chern., 39, 194.
18. Low, M. J. D., 1968, Anal. Letters, 1, 819. -
19. Low, M. J. D., Mark, H. and Goodsel,-A. J., 1971, J. Paint
Technol., 43 (562), 49.
20. Kizer, K. C, 1973, An. Lab., 5 (6), 40.
21. Azarraga, L. V., 1976, Improved Sensitivity of On-the-Fly-GCIR
Spectroscopy, Paper No. 334, Pittsburgh Conf. Anal. Chern. Appl.
Spectrosc., Cleveland, OH.
22. Azarraga, L. V., 1980, Appl. Specrosc., 34, 224.
23. Griffiths, P. R., 1977, Appl. Spectrosc.--,31, 284.
24. de Haseth, J. A. and Isenhour, T. L., 1977-,Anal. Chern., 49,
1977 .
25. Coffey, P., Mattson, D. R. and Wright, J. C., 1978, An. Lab.,
10 (5), 126.
26. For example, Azarraga, L. V. and Hanna, D. A., GIFTS, Athens
GC/FT-IR Software User's Guide, U.S. EPA/ERL, Athern, Ga., 1979.
(Available from authors.).
420 James A. de Haseth
27. Kuehl, D. and Griffiths, P. R., 1978, Anal. Chem., 50, 418.
28. Gomez-Taylor, M. M. and Griffiths, P. R., 1978, Anal:'" Chem.,
50, 422.
29. KUehl, D., Kemeny, G. J. and Griffiths, P. R., 1980, Appl.
Spectrosc., 34, 222.
30. Kuehl, D., i981, Hardware Considerations for a Capillary
GC/FT-IR Interface, Paper No. 242, Pittsburgh Conf. Anal. Chem.
Appl. Spectrosc., Atlantic City, NJ.
31. Azarraga, L. V. and Potter, C. A., 1981, J. High Resol.
Chromatog. & Chromatog. Comm., 4, 61.
32. Reedy, G. T., Bowne, S. and Cunningham, P. T., 1979, Anal.
Chern., 51, 1535.
33. Kizer, ~ L., Mantz, A. W. and Bonar, L. C., 1975, Am. Lab.,
7 (5), 85.
34. Vidrine, D. W. and Mattson, D. R., 1978, Appl. Spectrosc., 32,
502.
35. e.g., Brown, R. S., Hausler, D. W., Taylor, L. T. and Carter,
R. C., 1981, Anal. Chem., 53, 197.
36. Kuehl, D. and Griffiths, P:-R., 1979, J. Chromatogr. Sci.,
17, 471.
37. Kuehl, D. T. and Griffiths, P. R., 1980, Anal. Chern., 52,
1394.
38. Fuller, M. P. and Griffiths, P. R., 1978, Anal. Chern., 50,
1906. --
39. Griffiths, P. R., 1978, "Infrared Transform Specrometry:
Applications to Analytical Chemistry," in Transform Techniques
in Chemistr5' Griffiths, P. R., Ed., Plenum, New York, p. 147.
40. Kemeny, G. . and Griffiths, P. R., 1980, Appl. Spectrosc.,
34, ·95.
41. Low, M. J. D. and Parodi, G. A., 1980, Appl. Spectrosc., 34, 76.
42. Black, R. E. and Wakefield II, T., 1979, Anal. Chem., 51,-;0.
43. Busse, G. and Bullemer, 1978, Infrared Phys., 18, 255.---
44. Busse, G. and Bullemer, 1978, Infrared Phys., 18, 631.
45. Rockley, M. G., 1979, Chern. Phys. Lett., 68, 4~.
46. Rockley, M. G., 1980, Appl. Spectrosc., 3~ 405.
47. Vidrine, D. W., 1980, Appl. Spectrosc, 3~ 314.
48. Rockley, M. G. and Devlin, J. P., 1980,~4, 407.
49. Rockley, M. G., Davis, D. M. and Richardson, H. H., 1980,
Science, 210, 918.
50. Rockley, ~G., Davis, D. M. and Richardson, H. H., 1981,
Appl. Spectrosc., 35, 185.
51. Annual Book of AS~Standards, part 43, p. 518, Standard
#F-121-76, 1977.
52. Ibid, p. 523, Standard #F-123-74.
53. Mead, D. G. and Lowry, S. R., 1980, Appl. Spectrosc., 34, 167.
54. Vidrine, D. W., 1980, Anal. Chern., 52,92. --
55. Mead, D. G., 1980, Appl. Spectrosc.:-34, 171.
56. Krishnan, K., "Measurement of Impurities in Silicon USing a
Conventional FT-IR System & the New Digilab SIM-lOO," Digilab,
Inc., 237 Putnam Ave., Cambridge, Ma. 02139.