Thin Solid Films, 113(1984)101-113

ELECTRONICSAND OPTICS 101

UNAMBIGUOUS DETERMINATION OF THICKNESS AND

DIELECTRIC FUNCTION OF THIN FILMS BY SPECTROSCOPIC
ELLIPSOMETRY

H. ARWIN* AND D. E. ASPNES

Eel1Loboratories, Murray Hill, NJ07974 (U.S.A.)
(Received June 14,1983; accepted December 23,1983)

We present a spectroscopic method of determining both thicknesses and

dielectric functions of thin films on previously characterized substrates. The method
requires that spectroscopic ellipsometric data be taken over an energy range where
the substrate has a sharp optical structure, e.g. a critical point. These data are used to
determine a “pseudodielectric” function for the film, which necessarily depends on
the assumed film thickness. We develop a general first-order theory which shows
that the correct thickness and dielectric function of the film are those for which the
substrate feature vanishes in the calculated pseudodielectric function. The theory is
also used to investigate sensitivity and applicability. We show that sensitivity can be
enhanced by differentiating the pseudodielectric function with respect to energy.
The method is demonstrated with several numerical and experimental examples,
which illustrate that monolayer thickness accuracy can be achieved.

1. INTRODUCTION

A major difficulty in determining the dielectric response s,, = =s,r+ iso of thin
films from ellipsometric data is the three unknowns at any wavelength, i.e. E,,, E,,~
and the film thickness d I. If the film is transparent, only two unknowns are involved,
but for very thin films the quantities determined are not ~,,i and d but the optical
thickness nod where n, is the index of refraction2.
One approach to resolving this problem is to obtain further data by varying a
system parameter. For example, the sample can be measured in different
ambients3-5, the film can be deposited in different thicknesses4*6 or on different
substrates or additional known films can be deposited on the sample’. However, the
dielectric properties of the film are not necessarily independent of such manipula-
tions. Alternatively, the angle of incidence 4 may be varied3,4**-‘2. However, in
addition to complicating the design of sample chambers, the variable 4 method
suffers from a lack of sensitivity for materials with large dielectric functions because
Snell’s law shows that propagation within the film is nearly normal to the surface

9 Permanent address: Laboratory of Applied Physics, Department of Physics and Measurement

Technology, Linkiiping University, S-58183 Linkiiping, Sweden.

0040-6090/84/$3.00 0 Elsevier Sequoia/Printed in The Netherlands

102 H. ARWIN, D. E. ASPNES

regardless of the value of 4. More seriously, if the film is very thin, the angle of
incidence enters the appropriate ellipsometric equations only trivially and no
independent conditions can be obtained from iti3.
Reflectance data have also been used to provide the missing information. Paik
and coworkers14*15 and Cahan and coworkers i6*r7 showed that a simultaneous
measurement of the change AR/R in reflectance resulting from the growth of a film
on a previously bare substrate could be combined with ellipsometric data to yield all
three unknowns. Alternatively, in some cases the reflectance was simply assumed to
be independent of the film thickness (AR/R = 0) to yield the necessary third
condition3~18-20. However, this is a somewhat dangerous assumption and, in any
case, computations involving reflectance do not always converge or lead to physical
solutions2’. If convergence does not occur, it is not clear whether the model or the
data are inadequate. Also, at present thin films must be grown in situ to obtain the
necessary accuracy in AR/R. Problems in precision can also arise if the dielectric
properties of the film are very similar to those of the substrate22.
Another approach is to vary the wavelength. If the film is known to be
transparent in a certain region of the spectrum, d can be determined there and used
as a system invariant to calculate the dielectric properties of the film in the region
where the film is absorbing23*24. While this can also be done at single wavelengths,
the advantage of spectroscopic measurements is that the assumptions about the
validity of the model and the value determined for the thickness can be checked by
looking for interference- or interface-related structure in the calculated dielectric
response of the film 25s26. If the film is sufficiently thick, the simple models can be
extended to obtain information about interface regions2sV26.
The purpose of this paper is to show that spectroscopic ellipsometry can also
provide accurate values of film thicknesses by requiring that substrate-related
optical structure in the calculated dielectric properties of films also vanish. This
approach has several advantages over previous methods. Much thinner films can be
investigated than is possible with interference-oscillation-dependent methods. No
special thicknesses or ambients are necessary. The method can also be used with
absorbing films. Finally, it yields some insight about the widths of film interfaces.
The principle and theory are given in Section 2 and several representative examples
are discussed in Section 3.

2. THEORY

2.1. General formulation

We suppose a three-phase system consisting of a substrate, film and ambient
whose dielectric functions are given by E, E, and E, respectively. We assume for
simplicity that the substrate and film are isotropic so that E and E, are scalar
quantities. The complex reflectance ratio of the system can be calculated by taking
the ratio ofthe Fresnel reflectance expressions for p- and s-polarized light’. Assuming
an ideal system, the complex reflectance ratio can be expressed formally as
P exp = P(E, co, &a, d 4 4) (1)
where d is the film thickness, ;1is the wavelength of the light and perp is the quantity
measured by the instrument. We assume that everything is known but E, and d.
The objective is to find E, without knowing a priori the value of d. If (d)
PROPERTIES OF THIN FILMS BY SPECTROSCOPIC ELLIPSOMETRY 103

represents a guess approximately equal to the true value d, eqn. (1) still can be solved
numerically for E,, but the value obtained will be a pseudodielectric function (a,).
This quantity can be related to the exact solution E, by expanding eqn. (1) to first
order in (d > - d, assuming this quantity to be small. We find that

(2)
which shows how an error in <d) affects the calculated spectrum <E,).
Now, although d is not known, its value can in principle be inferred because E,
must satisfy some rather general criteria. Obviously E, cannot be a function of the
angle of incidence or other physical dimensions of the system or of the substrate or
ambient dielectric functions, and its dependence on energy is presumably different
from that of the other parameters of the system. From the form of eqn. (2) it is clear
that (8,) will be generally independent of these parameters,only when (d) = d.
Only then does the correction term, which in principle is a function of all the
parameters of the system, vanish completely. The problem is therefore solved,
because if(d) = d, then (6,) = e,,.
Equation (2) is the formal first-order representation of the statement that a
correct solution to E, will show no system-related artifacts such as interference
oscillations23.26 or, as will be developed here, optical structure originating from the
dielectric response of the substrate. The above approach can clearly be generalized
to n-phase laminar systems, systems containing inhomogeneous media or any
physical system where measurable quantities can be modeled in terms of a few
parameters.
Precisely because it is simple, eqn. (2) provides new insight into the practical
aspects of determining both the thickness and the dielectric response of a film. The
most favorable cases obviously occur when the derivative ratio is large. If the ratio is
small, then the data may not be sufficiently accurate to allow a unique solution of d
and E, to be obtained. A special case of this type occurs if d and E, appear as a
separately factorable function f(d,E,) within p. Then eqn. (2) can be written

(E,) = .z-((d)-d)-
aflad (3)
af/h
and the remaining system parameters disappear regardless of the form of p. The
solution for d and E, must then depend on second-order effects. We shall show that
this is the situation for thin relatively unpolarizable films on highly polarizable
substrates but not for similar films on relatively unpolarizable substrates.
Finally, if substrate- or interference-related artifacts cannot be made to vanish
for any (d), this indicates that the term called E, in eqn. (2) must itself be a function of
the system parameters. This will occur if the film is not a completely independent
phase but contains some substrate material. This can also occur if the solution for E,
represented by eqn. (2) is not exact because the actual sample configuration is more
complicated than that assumed in the model. Examples of the latter situation
include the assumption of the isotropic three-phase model to describe a sample
where the interfaces are microscopically rough or otherwise graded or where the film
is anisotropic. The proper approach here is to use the more complicated model that
more accurately represents the sample when solving for a,.
104 H. ARWIN, D. E. ASPNES

2.2. Special cases

The mathematical formulation of the three-phase model’ is too complicated to
treat analytically. Therefore in this section we shall illustrate the above theory in the
special case where the film is thin enough for the first-order expansion in d/r2 to
apply. In addition to providing physical insight, the model is also of practical
importance in many systems. Thicker films will be discussed in Section 3, where
other applications of the above theory will be illustrated by numerical calculations.
Because spectroscopic ellipsometric data are generally represented in the thin
film limit in pseudodielectric function form, we use the first-order expansion*’
47&r, E(E- s&z, - 8,) s 112
(E) = E+T --sin*4 (4)
s,(s - 8,) ( s, >
where (E) is the experimentally determined quantity. Usir 2 eqn. (2) (it should be
noted that ap/ad = @p/a(&)) a(E)/ad etc.), we obtain
W-d %(Eo-E&---E,)
k,)a= W-~
E,E- E,*
(5)

We see that the sensitivity is low if E, is nearly equal to E, or Ebut that it can be very
large ifs, is near (8,s) I’* . If I& 1, (E, 1 B E, then eqn. (5) SimplifieS t0

(e,) = E.+((d)-d)?

In this limit the substrate dielectric properties appear explicitly in the calculated
pseudodielectric function of the overlayer. The sensitivity can be enhanced by taking
the Kth derivative of eqn. (6) with respect to the energy E:
a”(Eo> (d) -d a”E
=-aE”d (7)
aE”
where we have used the assumption that E, varies slowly with E. Therefore only when
(d) = d do the substrate features aKE/aECvanish in ak(&,)/aEk.
If I E I % )E, 1, )E,) then eqn. (4) simplifies to

(E) = E+?E~,* (8)

Here d and E, now factor as the function f(d, E,) = d(1 - E,/E,). By evaluating eqn. (2)
explicitly we obtain

(E,) = E,--(<d)-d)E”(~E-ea)
a
Equation (9) is a specific manifestation of eqn. (3): there is no substrate dependence
in (E,), and consequently the correct thickness cannot be found by minimizing
substrate features in (E,). We conclude, as earlier researchers have also done*, that
the thickness and dielectric function of the film cannot be separated in first order if
the film is very thin and the substrate is highly polarizable. However, the above
derivation shows that this restriction is only limited to situations where I EI % IQJ.
For many overlayer-substrate systems, both optical properties and film thickness
can be obtained.
PROPERnES OF THIN FILMS BY SPECTROSCOPIC ELLIPSOMETRY 105

3. APPLICATIONS

Since the sensitivity and ultimately the limits of the method depend on the
magnitude of the derivatives in eqn. (2), we give several examples showing the level of
accuracy attainable. We begin with three numerical examples covering both
transparent and absorbing overlayers as well as a four-phase system with an
intermediate rough region between the substrate and the film. We finish with two
experimental applications: amorphous tellurium (a-Te) on HgCdTe and poly-
pyrrole on gold.
All calculations were done in three- or four-phase models using the exact
Fresnel expressions for the complex reflectance ratio p = tan + exp(id) ‘. When the
dielectric function of a film was calculated for an assumed film thickness, the exact
Fresnel expressions were solved by iterative numerical inversion. Thus no math-
ematical approximations were used in any of the calculations. The film(s), substrate
and interfaces were all assumed to be ideal.
Dielectric function data used in the numerical computations were obtained
from the following sources. Crystalline silicon (c-Q) and a-Si data are from refs.
28 and 26 respectively, the SiO, data are from work by Malitson2’ and the
Hg, _,Cd,Te data for a nominal composition x = 0.29 are from ref. 24. Measure-
ments on thin films were done with a spectroscopic ellipsometer30*31 with an energy
range of 1.5-6.0 eV. The samples were cleaned and maintained in a dry N, ambient
in a windowless cell. An angle of incidence of 1.171 rad was used in all examples. The
polypyrrole films were formed by electropolymerization on evaporated gold films32.
Fabrication details and further results on polypyrrole will be presented in a future
publication.

3.1. Numerical examples

3.1.1. SiO, on c-Si
The complex reflectance ratio p was calculated from 1.5 to 6.0 eV for a 2 nm
Si02 overlayer on a c-Si substrate. Using these values of p as “experimental” data,
we solved the three-phase model numerically for the pseudodielectric function of the
overlayer with different estimates of(d). The second derivatives of (8,) with respect
to energy were then calculated to enhance the sensitivity to substrate-related
structure. In Fig. 1 we show some examples of derivative spectra near the 3.4 eV E,
critical point of c-Si.
This numerical example represents a severe test of sensitivity since it falls nearly
within the scope of eqn. (9). The main sources of uncertainty are the statistical
fluctuations in the data for c-Si, which generate noise in the higher derivative spectra
of (8,). Nevertheless, the thickness can be determined to within 0.02 run. Since the
film is transparent, we can also use the commonly applied criterion that the correct
film thickness is that for which the imaginary part of E, vanishes. Using this we also
find the correct value of d to within 0.02 nm.
3.1.2. a& on c-Si
In this example we study an overlayer which is everywhere absorbing in the
experimental energy range. Thus the criterion of finding a value of (d) for which
(8,) is purely real is not applicable. In the same manner as before we calculate an
experimental p spectrum for a 5 nm layer of a-Si on a c-Si substrate and then
106 H. ARWIN. D. E. ASPNES

I I I

3.0 3.0 3.4 3.6 3.8

ENERGY(eV)
Fig. 1. Second energy derivative of the imaginary part of (s,) for different estimates of the film thickness
(d) for data numerically calculated for 2 nm of SiOz on c-Si near the E, critical point of c-Si.

evaluate the film thickness from computations of (E,). In Fig. 2(a) we see that the
second derivative of (E,,~) for a-Si has a broad spectral dependence owing to its
dispersive properties. However, the sharp structure at 3.4 eV due to the c-Si
substrate is clearly distinguishable if a wrong thickness estimate is used in the
calculation. The correct thickness can readily be determined to well within atomic
dimensions. In Fig. 2(b) we show that attempts to determine the correct value of d by
direct observation of (E,) are at least one order of magnitude less sensitive and are
also subject to judgment as to which spectrum is best. The differentiation method is
more accurate because it accentuates the structure due to the substrate.
3.1.3. a-Si@nite interface/c-Si
Here we discuss a five-phase sample consisting of a c-Si substrate covered with
three layers. The first is a 1 nm film consisting of equal volume fractions of c-Si and
SiOt. The second is a 1 nm film consisting of equal volume fractions of SiOz and
a-Si. The third is a 3 run film of a-Si. The dielectric functions of the mixtures were
calculated in the effective medium approximation33. The structure is intended to
simulate a microscopically rough oxidized substrate upon which an a-Si film is
deposited. Assuming that we are not aware that the roughness and oxide are present,
we analyze the experimental p spectrum, calculated correctly with the five-phase
model, as if the three-phase model were applicable.
Spectra calculated for three different values of(d) are shown together with a
reference spectrum of a-Si in Fig. 3. In no case is it possible to eliminate either the El
or the 4.2 eV E2 spectral features of the substrate from (E,). (As usual, this is much
PROPERTIES OF THIN FILMS BY SPECTROSCOPIC ELLIPSOMETRY 107

I
2 3 4
ENERGY (6V)

I I I I

40

20 E 3

2 3 4 5 6
ENERGY (eV)

Fig. 2. (a) As Fig. 1 but for 5 nm of a-Si on c-Si; (b) imaginary part of (e,,) for the same thicknesses.

more apparent in numerically calculated derivatives.) Therefore we can conclude

that the term E, in eqn. (2), as calculated in the three-phase model, is still dependent
on the system parameters. That is, the three-phase model does not adequately
represent the sample. The substrate structure can be minimized for (d) = 4.5 nm,
but this thickness does not correspond to any combination of thicknesses used in the
model calculation. The results indicate also that finite interface widths may be
detected from this analysis.
Despite these complications, the gross behavior of (a,) shows surprisingly little
dependence on (d): the overlayer would be correctly identified from the spectral
dependence of (a,,) as a-Si although the apparent magnitudes are reduced by about
15%. If the warnings provided by the inability to eliminate artifacts in (s,) were
108 H. ARWIN, D. E. ASPNES

E (ev)
Fig. 3. As Fig. 2(b) but for 3 nm of a-Si on c-Si with an intermediate rough layer: ....., a-Si; -. -,
(d) = 4.0 nm;-, (d) = 4.5 nm; - - -, (d) = 5.0 nm. Details are given in the text.

ignored and the data were analyzed in the effective medium approximation, the film
density would be incorrectly underestimated by about 10%.

3.2. Experimental examples

3.2.1. a-Te on HgCdTe
If anodic oxides on HgCdTe are stripped with hydrochloric acid, a residue
precipitates on the surface 24. This residue has tellurium-like optical and chemical
properties and its thickness is in the range of 5%-10% of that of the original oxide. If
we simply calculate the second derivative of (E,) in the region of the El transition of
Hgo.,,Cd0.29Te at 2.37eV for various (d), then d can be determined to about
+O.l nm. This is sufficiently accurate for most purposes because most interfaces
cannot be assumed to be definable to less than 0.1 nm. However, if we smooth the
derivatives by reducing the high frequency noise, d can be determined to within a few
hundredths of a nanometer. This case is shown in Fig. 4(a) for a film 43.5 nm thick.
The distortion in the derivative spectra introduced by filtering is of no importance as
long as the main features are not completely smoothed out. Figure 4(b) shows that
the overlayer spectrum calculated for this film is absorbing everywhere in the
experimental energy range. Application of this approach to precipitated a-Te films
of thicknesses ranging from 1 to 5 nm has shown that the E spectra are the same
except for density effects.
3.2.2. Polypyrrole on gold
Polypyrrole is a conducting polymer32*34 which, like most other organic
materials, has a low polarizability in the visible and UV wavelength region.
Furthermore, it has broad adsorption bands in this energy range. The actual
dielectric response of polypyrrole depends strongly on its oxidation state which
readily can be changed by charge injection34-36.
We have applied our method to a thin polypyrrole film on a gold substrate. This
system represents a class of applications in which the dielectric functions of both
substrate and film are not significantly higher than that of the ambient. Gold has two
PROPERTIES OF THIN FILMS BY SPECTROSCOPIC ELLIPSOMETRY 109

suitable critical points in the energy range 1.5-6.0 eV: the d band to Fermi level
transition near 2.5 eV and the L transition near 3.6 eV. Depending on the type of
polymer or its oxidation state, either or both of these critical points may be used.
Our measurements were done on an air-exposed polypyrrole film.

0 4.35

I I I I
2.0 2.2 2.4 2.6 i
ENERGY (WI

I I I I

_-I
6
ENERGY (6’4)
(‘4
Fig. 4. (a) Second energy derivative of the imaginary part of (E,) for different estimates of the film
thickness <d) for ellipsometric data taken on a-Te on Hg,,,lCd 0,29Te near the E, critical point of the
substrate; @) dielectric function of a-Te calculated for <d) = 4.35 mn.

In the optical spectrum of this particular film, the tail of the free carrier
absorption extended above the gold critical point at 2.5 eV. In Fig. 5(a) we show the
second derivative of (a,) with respect to E for the pyrrole film for different values of
(d). We see immediately that the gold-derived spectral feature at 2.5 eV can be
minimized but cannot be completely removed. The most appropriate thickness is
110 H. ARWIN, D. E. ASPNES

22.OkO.5 nm. According to the previous section, these results suggest that the
sample configuration is more complicated than the model used to calculate (a,). It is
likely that the interface between the substrate and the film was roughened by the
polymerization process. We have verified this hypothesis by measuring gold surfaces
which were processed in the same way but with no pyrrole monomer in the solution.
If we ignore the roughness as a minor perturbation, we obtain the refractive index
N = n + ik = E “* shown in Fig. 5(b) for the pyrrole film. The results are in good
qualitative agriement with literature data34*35.
We also attempted a similar analysis on data from a poly(N-methyl pyrrole)
film that showed very little absorption below 3 eV. We could not find the thickness

2.0 2.2 2.4 2.6 2.6 3.0

ENERGY (6V)

f--zI-j2O

I I I 1 I I
2 3 4 5 6

(W ENERGY (6V)

Fig. 5. (a) As Fig. 4(a) but for polypyrrole on gold; (b) refractive index and absorption coefficient for
polypyrrole calculated for (d) = 22.0 nm.
PROPERTIES OF THIN FILMS BY SPECTROSCOPIC ELLIPSOMETRY 111

by analyzing derivatives near 2.5 eV. However, at the L transition near 3.6 eV the
thickness could be determined to within 0.2 nm.

4. DISCUSSION

We have described a spectroscopic method for determining the thickness and

dielectric function of a thin film on a substrate. The method can be applied directly
to existing data and does not require special ambients, deposition conditions or
sample configurations. However, it does require that the substrate have a distinct
optical feature somewhere in the experimentally accessible energy range. This
requirement is fulfilled by many semiconductors and metals. The method loses
sensitivity for very thin films on highly polarizable substrates, but this effect can be
entirely understood from the general theoretical formulation that we have
developed and verified with numerical calculations. In this case the first-order effect
vanishes and the sensitivity must come from second-order terms in the expansion.
The method is ultimately limited by the sample complexity and the accuracy
with which optical data can be measured. If there is a graded interface between the
film and the substrate, it may or may not be possible to reduce the amplitudes of the
substrate features in (E,) below that of the intrinsic uncertainty due to noise. Thus
the ability of the data to show whether the assumptions of the model used to
represent the sample are actually satisfied by the sample reduces ultimately to the
question of experimental sensitivity. It is usually true that the inability to eliminate
artifacts completely in a film spectrum is an indication that the model is too simple.
In the thin film configurations discussed here, such failures can be traced to finite
interface widths or to film inhomogeneities or anisotropies. For example, uniaxial
films have been treated to first order in d/l in ref. 13, and the improper assumption of
isotropy leads to an equation for the film dielectric function where those of both
substrate and ambient appear explicitly (see ref. 13, eqn. (4.2): the term e,.sleJass, in
this expression should be E,+ s1s3/a3s1).
The numerical example consisting of an a-5 layer on a thin SiOZ film on a
rough c-Si substrate illustrates a situation where the three-phase model is too simple
and the substrate structure in (E,) cannot be made to vanish for any (d). However,
model calculations for the c-Si/SiO,/a-Si system show that the substrate structure in
(E,) can be made nearly to vanish in a three-phase model if the interfaces are
assumed to be planar. Therefore, while the inability to eliminate artifacts shows that
an assumed model is too simple, the ability to eliminate artifacts to within
experimental accuracy may not be conclusive evidence that this model is adequate.
Of course, if independent measurements show that a sample is more complex,
difficulties can be avoided by using the more complicated model that better
represents the sample.
We have also found that the sensitivity to variations in (d) can be quite
dependent on specific values of d, E and E,, as indicated in the example of poly(N-
methyl pyrrole) on gold where analysis at 3.6 eV gave much better results than
analysis at 2.5 eV. These low sensitivity combinations may be due to a small
numerator in eqn. (2) or to an effective factoring of the derivative ratio as in eqn. (3).
Owing to the complexity of the reflectance equations, these low sensitivity
combinations are best found by numerical calculations and best eliminated by
112 H. ARWIN, D. E. ASPNES

changing wavelength, film thickness or substrate material in situations where this

can be done. Another difficulty can arise if the dielectric function of the overlayer has
sharp structure which overlaps that of the substrate. This is generally not a problem
as many films are amorphous and have optical parameters that vary slowly with
energy. However, some organic films, such as phthalocyanine3’ may have sharp
threshold energies. Here the calculated derivative spectra may be dominated by the
intrinsic optical properties of the film. The method was not applicable to an
absorbing 1,4,5,%naphthalenetetracarboxylic dianhydride film on a silicon sub-
strateJ8, although other substrates could have been used.
To utilize the method fully, it is also desirable to perform some type of noise
filtering either by smoothing the calculated derivatives or by Fourier filtering3’.
From Fig. 2 it is obvious that differentiating (E,) with respect to energy enhances
sensitivity. We have found empirically that a second-order derivative is generally
sufficient. Third derivatives require more smoothing of the data, which can be done
as long as the optical features of the substrate are not completely removed.

ACKNOWLEDGMENT8

One of us (H.A.) gratefully acknowledges partial financial support by grants

from the National Swedish Board for Technical Development and from the Swedish
National Science Research Council. We are also thankful to Ingemar Lundstrom
and Robert Bjorklund for collaboration on the polypyrrole experiments.

REFERENCES

1 R. M. A. Azxam and N. M. Bashara, Ellipsometry and Polarized Light, North-Holland, New York,
1977.
2 W. J. Plieth and K. Naegele, Surf. Sci., 64 (1977) 484.
3 F. Lukes, W. H. Knausenberger and K. Vedam, Surf. Sci., 16 (1969) 112.
4 S. S. So and K. Vedam, J. Opt. Sot. Am., 62 (1972) 16.
5 G. A. Egorova, N. S. Ivanova, E. V. Potapov and A. V. Rakov, Opt. Spektrosk., 36 (1974) 773.
6 G. Jungk, Phys. Status Sofidi A, 34 (1976) 69.
7 C. L. Nagendra and G. K. M. Thutupalli, Appl. Opt., 22 (1983) 587.
8 M. M. Ibraham and N. M. Bashara, J. Opt. Sot. Am., 61(1971) 1622.
9 P. Pieozzi, S. Santueei and A. Balzarotti, Surf Sci., 45 (1974) 227.
10 0. Hunderi, Surf. Sci., 61(1976) 515.
11 E. V. Krasil’nikova and V. N. Lebedeva, Sou. Pkys.-Tech. Phys., 26 (1981) 1119.
12 M. E. Pedinoff and 0. M. Stafsudd, Appl. Opt., 21(1982) 519.
13 M. J. Dignam, M. Moskovits and R. W. Stobie, J. Chem. Sot., Faraday Trans. ZZ,67 (587) (1971)
3306.
14 W.-K. Paik and J. O’M. Boekris, Surf. Sci., 28 (1971) 61.
15 S.H.Kim,W.-K.PaikandJ.O’M.Bockris,SurfSci.,33(1972)617.
16 B. D. Cahan, J. Horkans and E. Yeager, Surf. Sci., 37(1973) 559.
17 C.-T. Chen and B. D. Cahan, J. Electrochem. Sot., 129 (1982) 17.
18 K. Vedam, W. Knausenberger and F. Lukes, J. Opt. Sot. Am., 59 (1969) 64.
19 R. C. O’Handley, Surf. Sci., 46(1974) 24.
20 R. C. O’Handley and D. K. Burge., Surf Sci., 48 (1975) 214.
21 B. D. Cahan, Surf Sci., 56 (1976) 354.
22 G. Vuye and T. Lopez-Rios, Appl. Opt., 21(1982) 2968.
23 D. E. Aspnes, G. P. Schwartz, G. J. Gualtieri, A. A. Studna and B. Schwartz, J. Electrochem. Sot.,
128 (1981) 590.
PROPERTIES OF THIN FILMS BY SPECTROSCOPIC ELLIPSOMBTRY 113

24 H. Anvin and D. E. Aspnes, J. Appl. Phys., 54 (1983) 7132.

25 D. E. Aspnes and J. B. Theeten, J. Electrochem. Sot., 127(1980) 1359.
26 D. E. Aspnes, A. A. Studna and E. Kinsbron, Phys. Rev. B, 29 (1984) 768.
27 D. E. Aspnes, in B. 0. Seraphin (ed.), Optical Properties of Soliak: New Developments, North-
Holland, Amsterdam, 1976, p. 799.
28 D. E. Aspnes and A. A. Studna, Phys. Rev. B, 27 (1983) 985.
29 I. H. Malitson, J. Opt. Sot. Am., 55(1965) 1205.
30 D. E. Aspnes and A. A. Studna, Appl. Opt., 14 (1975) 220.
31 D. E. Aspnes and A. A. Studna, Rev. Sci. Instrum., 49(1978) 291.
32 K. K. Kanaaawa, A. F. Diaz, W. D. Gill, P. M. Grant, G. B. Street, G. P. Gaidini and J. F. Kwak,
Synth. Met., 1(1979-1980) 329.
33 D. A. G. Bruggeman, Ann. Phys. (Leipzig), 24 (1935) 636.
34 K. K. Kanazawa, A. F. Diaz, M. T. Krounbi and G. B. Street, Synth. Met., 4 (1981) 119.
35 W. R. Salaneak, R. Erlandsson, J. Prejaa, I. Lundstr(im, C. B. Duke and W. K. Ford, Pofym. Prepr.,
Am. Chem. Sot.. Div. Polym. Chem., 23 (1982) 120.
36 W. R. Salaneck, R. Erlandsson, J. Prejza, I. Lundstrijm and 0. InganPs, Synth. Met., 5 (1983) 125.
37 B. H. Schechtman and W. E. Spicer, J. Mol. Spectrosc., 33 (1970) 28.
38 P. H. Schmidt, D. C. Joy, M. L. Kaplan and W. L. Feldmann, Appl. Phys. Lett., 40 (1982) 93.
39 D. E. Aspnes, Surf Sci., I35 (1983) 284.