An infrared dielectric function model for amorphous solids

R. Brendel and D. Bormann Citation: J. Appl. Phys. 71, 1 (1992); doi: 10.1063/1.350737 View online: http://dx.doi.org/10.1063/1.350737 View Table of Contents: http://jap.aip.org/resource/1/JAPIAU/v71/i1 Published by the American Institute of Physics.

Related Articles
On the design of experiments for determining ternary mixture free energies from static light scattering data using a nonlinear partial differential equation J. Chem. Phys. 137, 034201 (2012) Counterion adsorption theory of dilute polyelectrolyte solutions: Apparent molecular weight, second virial coefficient, and intermolecular structure factor J. Chem. Phys. 137, 034902 (2012) A minimal model for excitons within time-dependent density-functional theory J. Chem. Phys. 137, 014513 (2012) Stimulated Mie scattering in nanocrystals suspension Appl. Phys. Lett. 101, 011110 (2012) Optical generation and detection of gigahertz-frequency longitudinal and shear acoustic waves in liquids: Theory and experiment J. Appl. Phys. 112, 013502 (2012)

Additional information on J. Appl. Phys.

Journal Homepage: http://jap.aip.org/ Journal Information: http://jap.aip.org/about/about_the_journal Top downloads: http://jap.aip.org/features/most_downloaded Information for Authors: http://jap.aip.org/authors

Downloaded 26 Jul 2012 to 130.237.29.138. Redistribution subject to AIP license or copyright; see http://jap.aip.org/about/rights_and_permissions

An infrared dielectric
FL Brendel

function

model for

amorphous

solids

Institut ftir Werkstofiissenschaften VT, Universitiit Erlangen-Niirnberg, W-8520 Erlangen, Federal Republic of Germany

Martensstrasse 7,

D. Bormann
Institut ftir Theoretische Physik, UniversitZt Heidelberg, Philosophenweg 19, W-6900 Heidelberg, Federal Republic of Germany

(Received 31 July 1991; accepted for publication 30 September 1991) For the modeling of infrared spectra it is a common approach to use a dielectric function that treats the vibrational modes as damped harmonic oscillators. This model was found to be rather crude for some applications to amorphous solids. A dielectric function model yielding a Gaussian shape of the absorption lines and satisfying KramersKronig relations is suggested.The model function is constructed by a convolution of a Gaussian function with the dielectric function of the damped harmonic oscillator model. An analytical solution of this integral is given. It is demonstrated that this model describes the spectra of thermally grown ultrathin (1.3 run) silicon oxide films, plasma-depositedsilicon films, plasma-depositedsilicon nitride films, and amorphous aluminum oxide films very well. The physical motivation of the dielectric function model suggestedis the randomnessof the vibrational frequenciesin an amorphous structure.

I. INTRODUCTION

The modeling of infrared spectra is a powerful tool for the quantitative interpretation of measured spectra.ls2 Modeling becomes essential if measurements are performed at oblique incidence and/or the films of interest are contained within a complex multilayer structure.ld3 The straightforward way to model thesespectra is to treat every vibrational mode as a damped harmonic oscillator. This approach is called the Drude model. It successfully describes the main features of the absorption lines: frequency, width, and oscillator strength. There is no free parameter to model the shape of the absorption lines. It is therefore not surprising that there may be a poor agreementbetween measured and calculated line shapes. This was demonstrated for ultrathin ( 1.3 nm) silicon oxide tilms within a metal-insulator-semiconductor (MIS) structure.3 The necessity to simulate the spectra of ultrathin silicon oxide films for their quantitative evaluation has lead us to develop a new model function which will be called the Gauss model throughout this work. The Gauss model has been applied successfullynot only to ultrathin silicon oxide tihns but also to amorphous solids such as silicon, silicon nitride, and aluminum oxide.
II. THEORY A. The dielectric function model

is the contribution of each vibrational mode k. Here 7 is the wave number, Ye is the resonancewave number, YPthe plasma wave number, and 7r the damping constant. E, is due to the electronic polarizability. It is a commonly applied and quite successfulmethod to decomposespectral features containing more than one vibrational mode by assigning a Gaussian line shape to every single mode (e.g., for a-SiN:H in Refs. 4 and 5). Frequencies,heights, and widths are then chosen so as to make these Gaussianssum up to the measuredabsorbance spectrum. From this we can learn that the line shapesof the absorbancespectra can often be described well with a Gaussian function. The dielectric function model should therefore also be related to a Gaussianfunction. In order to construct a dielectric function fulfilling Kramers-Kronig relations and yielding a Gaussian shape of the imaginary part we chose the approach

+CQ
-dx.

Sk

x2 - ? + i?&

(3)

The Drude model uses a dielectric function E of the form Em=E, where
xk(q
1

The number of harmonic oscillators per frequency interval is governed by a Gaussian distribution. Depending on the value of ?+/okk, the imaginary part X$ will vary in shape from purely Gaussian to purely Drude shape.The integral (3) can be solved analytically (see Appendix) : Xk=-s+exp

+ i

k=l

x,km,

(1)

[ exp( -

(ak;$)z)e,~;~)

=i$/(Gk

- 7 + ic$)

-,w)erfc

(*)

1,

(4)

(2)
0021-8979/92/130001-08$04.00 @ 1992 American Institute of Physics 1

J. Appl. Phys. 71 (I), 1 January 1992

Downloaded 26 Jul 2012 to 130.237.29.138. Redistribution subject to AIP license or copyright; see http://jap.aip.org/about/rights_and_permissions

TABLE I. Parameters for the modeling of the infrared spectra in Figs. l-4. Parameters that have been kept Exed during the iteration procedure are marked with an asterik. For the Gauss model a = dma which is the FWHM of a Gaussian curve with standard deviation (r has been listed instead of the standard deviation itself. See Eq. (3). The calculated spectra have been multiplied by a correction factor C. Mode number Sample A (Fig. 1): SiO, d= Drude: 1.3 nm* 1 2 Wave numbers [cm- 1 3 4 5 Ecc 1191 263 49 1194 268 85 4.4 1175* 2 153 1142 284 157 -c/ii/lo .-. ... ... ... ... ... ... ... ... ... ... ... ... ... c il.999

Gauss:

% s Gr
70 ?P is FT

790 325 96 783 359 223 9.4 465 376 126 481 411 201 Z/l0 2016 264 97 2013 209 88 12 866 601 57 849 580 120 10

1101 616 46 1090 584 100 8.3 852 1167 138 863 1060 197 Z/l0 2090* 88 61 2090* 103 76 Trl ... ... ... *.. ... ..1 ...

... ... .*. ... ... ... . ..

2198 215 122 2201 192 143 iVl0 ... ... ... ... ... ... ... ... .. . *.. ... ... ... ...

... ... ... ... ... ... . . ..

3325 188 118 3320 167 131 VlO . .. ... ,.. ... ... ... ... ... ... ... ... ... ... . ..

2.10*

2.10*

0.998

Sample B (Fig. 2): a-SiN:H, d = 0.78 pm* Drude:

TO P VT

3.68*

1*

Gauss:

FO % B TT

3.68*

1*

Sample C (Fig. 3): a-Si:H, d = 0.7 pm* Drude:

70 % Fr

11.2*

0.983

Gauss:

TO s d Gr

11.2*

0.991

Sample D (Fig. 4): AlO, d = 28 nm* Drude:

TO % Tr

2.9*

(see text)

Gauss:

70 % 3 ir,

2.9*

(see text)

with ak. = dw ._. and a: > 0. Here e&(x) = 1 ,- erf(x) is the complementary error function. This analytic solution is of interest not only for theoretical purposes but also for the evaluation of .%?k a computer if the on complex error function is available.
B. Modeling of the spectra

All spectra have been modeled using the algorithm of Ref. 6. This algorithm allows calculation of the transmission and reflection spectra of any multilayer structure with parallel interfaces and homogeneous isotropic dielectric properties in each film. It has the special advantage that thick films such as substratescan easily be treated as incoherent films. Thick films do not produce interference fringes if the spectral resolution is sufficiently low. In some casesa good fit of the measuredspectra was only possible by multiplying the calculated spectrum with a frequency independent factor C (see Table I). This factor accounts for base line shifts of unknown reasonsand should always be close to 1. Formal incidence transmission spectra (TO) and spectra measuredwith the technique of grazing external reflec2 J. Appl. Phys., Vol. 71, No. 1, 1 January 1992

.d c??

0.1-

500

. 1000
[cm-t]

1500

FIG. 1. Sample A: An 8(Y grazing internal reflectance (GIR) spectrum of an ultrathin silicon oxide film (1.3 nm) within an MIS diode. The light passes normally through the entrance and exit faces of the internal reflection element (IRE). Polarized light is used and the ratio of the reflectance of parallel polarized light R, to perpendicular polarized light R, is plotted: (-) measured, (- - -) fitted with Drude model, (. . . . . . . -) fitted with Gauss model. R. Brendel and D. Bormann 2

Downloaded 26 Jul 2012 to 130.237.29.138. Redistribution subject to AIP license or copyright; see http://jap.aip.org/about/rights_and_permissions

tion (GER, also named IRRAS) and grazing internal reflection3 (GIR) have been modeled. For the GIR spectra an internal reflection element (IRE) is used (see inset of Fig. 1). The algorithm6 used is not capable of taking into account the reflections at the entrance and exit slits of the internal reflection element. We therefore first calculated the reflectance RMIs of the metal-insulator-semiconductor (MIS) structure with the front half-space silicon and the back half-space aluminum. As for the TO spectra silicon was assumed to be nonabsorbing with an index of refraction of 3.42 and as for the GER spectrum the aluminum parameters for the Drude dielectric function model were taken from Ehrenreich.7 Due to the reflections at the entrance and exit slits there is a slight influence of the nonzero absorption of silicon (absorption constant a) on the spectrum. This was taken into account by replacing R,, withs 1 - (Rsie-dRMIs)]. I(1 - &i) 2e-dRMIs]/[ Here, Rsi is the reflectance of the air/silicon interface and d the length of the light path within the IRE. The silicon absorption constant (Y was determined from a normal incidence transmission spectrum of the IRE prior to film deposition. For the Z spectra the silicon substrate was treated as YI an incoherent film.
III. EXPERIMENT A. Samples

1.1

G . 5 0.6

sample

*IIt TVS ~Si B

1000

2000 V [cm-*]

3000

4000

FIG. 2. Sample B: Transmittance spectrum of an amorphous-plasmadeposited silicon nitride film (0.78 pm) on a silicon substrate. The ratio of the transmittance of the system film on substrate Tf,s divided by the transmittance of an uncoated substrate TX is plotted: (-) measured, (- - -) fitted with Drude model, (. . . . . . . .) fitted with Gauss model.

with the NH, flow set to zero. A film thickness of 0.7 pm and a refractive index of 3.35 was determined from the infrared transmission spectrum. Sample D is an aluminum oxide film on aluminum. First a l-pm aluminum film was electron-beam evaporated onto a silicon wafer. Then the aluminum film was heated for 10 min at about 400 C in air. Ellipsometrically we determined an oxide thickness of 28 nm and an index of refraction of 1.7.
B. Spectra

The Gauss model has been applied to the following four different materials. Sample A, a Si/1.3 nm SiO/Al metal-insulator-semiconductor (MIS) structure, was built on an 80 singlereflection IRE of 8.5X25X 1.5 mm3 size from 50-a cm, boron-doped, ( lCO)-oriented float zone (FZ) silicon. It was mechanically polished on the entrance, reflection, and exit faces. The damage layer from polishing was removed by chemically etching 5 pm of silicon. After etching of the natural oxide in diluted HF an Ohmic back contact was evaporated on the backside of the IRE. This contact is of no importance for the subject of this paper. Next an oxide f&n of 1.3 nm thickness was grown on the reflection face by heating in dry oxygen at 500 C for 10 min. The oxide thickness was measured ellipsometrically at 632.8 nm assuming an index of refraction of 1.45. Finally a 5+m Al film was electron-beam evaporated onto the thin oxide. Sample B, a silicon nitride film, was deposited onto a silicon substrate. The nitride film was prepared by-the reaction of NH3 and SiI& at 270 C in a glow discharge reactor (Plasmol, ASM) at a gas ilow ratio NH3/SiH4 = 7 and a total pressure of 1.13 mbar and a t-f power of 180 W m - The substrate was a 385~,um-thick, 5-a cm, bo. ron-doped, ( lOO)-oriented, both sides polished FZ silicon wafer. An index of refraction of 1.92 was determined ellipsometrically and a film thickness of 0.78 pm has been measured with a uv/visible spectral photometer. Sample C, an amorphous silicon film deposited onto silicon, was prepared under the conditions of sample B
3 J. Appl. Phys., Vol. 71, No. 1, 1 January 1992

All spectra were recorded with a commercial Fourier transform infrared spectral photometer (model FTIR1760, Perkin-Elmer) at room temperature with 16-cm - resolution and a deuterated triglycine sulphate detector. Sample A was measured with a suitable reflection unit placed in the light path. The light used is s or p polarized by a AgBr gold wire grid polarizer. The absorption of the p polarization is largely enhanced due to the high field in the thin film enforced by the Maxwell equations. The spolarized light is shorted at the Al film. It is hardly absorbed at all and serves as a reference measurement. 500 scans were taken at alternating polarization in the unpurged spectral photometer. Rp is the reflectance of the parallel-polarized light and R, that of the perpendicularpolarized light (see Fig. 1). For details of this measurement technique see Ref. 3. Samples B (50 scans, unpurged) and C (250 scans, Nz purged) have been measured in normal incidence transmission geometry and the spectra are presented in Figs. 2 and 3, respectively; the transmittance of the film on substrate structure is denoted by Tfjs and that of the uncoated substrate by T, Sample D (250 scans, unpurged) has been measured in the GER geometry (see insert Fig. 4) using a reflection unit (model 501, Spectra Tech). In Fig. 4, R/Rs denotes the reflectance spectrum of sample D after the division
R. Brendel and D. Bormann 3

Downloaded 26 Jul 2012 to 130.237.29.138. Redistribution subject to AIP license or copyright; see http://jap.aip.org/about/rights_and_permissions

sample C

0.89

1700

I 1900

2100

I 2300

V [cm-l]
FIG. 3. Sample C: Transmittance spectrum of a plasma deposited amorphous silicon film (0.7 pm) on a silicon substrate. The ratio of the transmittance of the system film on substrate T,, divided by the transmittance of a similar uncoated substrate T, is plotted: (-) measured, (- - -) fitted with Drude model, ( *..*****) fitted with Gauss model.

through the spectrum measuredprior to oxidation. The natural oxide thickness prior to oxidation was taken into account for the m o d e ling.
IV. RESULTS

The spectrum of sampleA (Fig. 1, solid line) shows a spectral contrast of 8 0 % although the SiO film is only 1.3 n m thick. This enormous sensitivity makes the G IR a method well suited to study M IS structures.A reduction of the size of the IRE by a factor of 2 and the use of an AgBr polarizer greatly reduced the noise and further increased the accessible frequency range over an earlier measurement.3 The spike at 608 cm- is attributed to noisedue to the absorptionin the silicon which has a sharp maximum at this wave number. A further improved signalto-noise ratio would have been possible by using a MeCdTe-detectorwhich was not available.

pz 0.9 . cz

0.8
sample D w---e

400

600

800 V

1000 [cm-l]

1200

1400

FIG. 4. Sample D: A 70 grazing external reflectance (GER) spectrum of an aluminum oxide film (28 nm) on aluminum. Polarized light is used and the ratio of R/R, after to R/R* prior to oxidation is plotted: (-) measured, (- - -) fitted with Drude model, (. . . . . . . . ) fitted with Gaussian model. 4 J. Appl. Phys., Vol. 71, No. 1, 1 January 1992

The broken line (Fig. 1) shows the result of a leastsquaresfit with the Drude m o d e l. The silicon oxide film has been m o d e led with three oscillators following Ref. 1. The quality of the fit is rather poor. For comparison,the dotted line shows the excellentfit achievedwith the Gauss m o d e l [seeEqs. ( 1) and (3)]. Therefore this m o d e l function greatly improves the capability to quantitatively interpret infrared spectra of ultrathin oxides within an M IS structure. The transmission spectrum of sample B (Fig. 2, solid line) is also fitted rather badly by the Drude m o d e l (broken line). In order to avoid a drifting of the resonance frequencyof m o d e 3 (Table I) to rather unrealistic values during the Drude fitting procedure, To3 was kept fixed at 1175cm- The Drude m o d e l was not able to describe . the spectrum in the frequencyregion of m o d e 3. The use of the Gauss m o d e l (dotted line) leads to a substantial improvement in this frequency region. In the frequency range from 1600 to 3000 cm - * the measurementis also m o d e led better by the Gauss m o d e l. During the fitting procedurethe parameter?iTk always been kept fixed to has one-tenth of the full width at half-maximum (FWHM) of the Gaussianfunction. For many different samplesit was found that Grk=Z;(/10 yields a good fit, which is only slightly improved by allowing ?T,kto vary independently. Therefore it is demonstratedthat a much better agreement with experiment can be obtained from the Gauss m o d e l while the number of free parametersis the sameas for the Drude m o d e l. The transmissionspectrum of sample C (Fig. 3 , solid line) is fitted with two vibrational modes.Si-Si-H3 tetrahedral bond contigurations can be ruled out due to the deposition temperatureof 270 CL Then we are left with the Si,-Si-H and Si,-Si-H, configuration. The stretching frequency of Si,-Si-Hz tetrahedrais known to be 2090 cm - i and was kept fixed for the fits. This was done because Si,-Si-Hz is containedonly in a small amount and the spectrum does not contain enough information to determine this frequency accurately. Again the quality of the Gauss fit is appreciablybetter than that of the Drude fit. F inally, the GER spectrum of sample D (Fig. 4, solid line) shows only one peak that is characteristic for amorphous aluminum oxide. The baselineof this spectrum is not flat. This is probably due to a roughness of the aluminum/oxide interface. To achieve satisfactory agreement with experiment,the spectrum m o d e ledas described in Sec.II B had to be m u ltiplied by a function linear in the wavelength rather than just by a frequency-independent factor C. This function was determined prior to the fitting procedureas to reproducethe baseline in the regions left and right of the absorption line. Due to the interface roughnessan error is introduced into the ellipsometrically determinedthicknesswhich is difficult to estimate. But this is of no relevancefor what we want to show here: Also, for aluminum oxide the Gauss m o d e l (dotted line) gives a much better description of the absorption lines shapethan the Drude m o d e l (broken line). The parametersof the fits have been listed in Table I.
R. Brendel and D. Bormann 4

Downloaded 26 Jul 2012 to 130.237.29.138. Redistribution subject to AIP license or copyright; see http://jap.aip.org/about/rights_and_permissions

V. DISCUSSION

VI. SUMMARY

The Gauss model was constructed from a phenomenological point of view. As a phenomenological model it might be tested for all kinds of materials. But all materials considered in the experiments described above are amorphous and there is a natural explanation why the Gauss model should be more suitable than the Drude model. In our model each vibrational mode is assumedto be made up out of a superposition of damped harmonic oscillators differing in frequency. In an amorphous solid the resonancefrequency of a vibrating unit is expected to follow a statistical distribution. This frequency will depend on a large number of unknown parameters describing the local environment of the vibrating unit. If each of these parameters again follows a statistical distribution the superposition of their effects, in the spirit of the central limit theorem, leads to a Gaussian distribution of the resonance frequency. The Gauss model (3) can be understood as a superposition of damped harmonic oscillators whose resonance frequencies follow a Gaussian distribution. In this interpretation the parameter Tr has the meaning of the average line width of a single vibrating unit. It is therefore interesting to compare the values of T7 obtained from the Gauss model fits in Fig. 1 to Fig. 4 with the linewidths that have been measured for isolated molecules or in ordered structures where the linewidth observed is equal to the linewidth of the individual vibrating unit. The linewidths of the vibrational modes of isolated SiOz molecule within an Ar matrix at low temperature have been measured to be about 5 cm - 1,12 value similar a to those of 7T = 4-9 cm - for sample A. The natural linewidth of a Si-N vibration can be estimated from the linewidths of the peaks in c-Si3Ne Both the spectra of (r- and &S&N., show a complex structure with many peaks of about 20 cm - linewidth. l3 This is a similar magnitude as the parameter 7r = 15-20 cm - used for the fit. The linewidth of Si-H stretching vibrations adsorbedin vacuum on c-Si has been found to be 6 cm - small.14This is the same order of magnitude as TT = 12 cm - obtained from the fit. For the Si-H bond it has been well established that there is a strong correlation between the bond length and the vibrational frequency. 5 oP With this relation we can follow from the infrared spectra that the variance in Si-H bond length is approximately 1%. This is just one example how the Gauss model may be used to extract additional information from the spectra. Isolated AlO molecules have also been synthesized within an argon matrix and show linewidths of about 5 cm - * for the stretching and bending modes.16The fit shown in Fig. 4 produced 10 cm- for the damping of the individual vibrating unit. Considering the fact that the quality of the fits deteriorates only slightly if V;, is smaller than its optimum value and considering that the linewidth of a vibrating unit in an amorphous structure should also be broadened for other than statistical reasons, the linewidths reported in the literature justify the model proposed.
5 J. Appl. Phys., Vol. 71, No. 1, 1 January 1992

A dielectric function model in the infrared frequency range has been proposed. The model function is a convolution integral of a Gaussian function with the dielectric function of a damped harmonic oscillator. This integral was solved analytically. The model was shown to be more appropriate for the quantitative evaluation of infrared spectra of amorphous solids than the Drude model. This was demonstrated for amorphous films of silicon oxide, silicon nitride, hydrogenated silicon, and aluminum oxide. The model is physically motivated by the randomness of vibrational frequencies of the individual oscillators in amorphous materials.
ACKNOWLEDGMENTS

R.B. would like to thank L. Ley and R. Hezel for discussing the model. Technical assistanceby A. Nisslbeck is gratefully acknowledged. This work was supported by the Bundesministerium fiir Forschung and Technologie of the Federal Republic of Germany.
APPENDIX

Omitting the index k, the integral to be solved is

exp

(Al) With a:= JT and

K(q):= li: exp( -&)&dx,

[we choose Im(a) > 0]

WI

(A3)

this can be rewritten as

X=&a

[K(xo a>- K(xo a> + I.

Now the integral K(q) is solved by first Fourier decomposing the exponential term and then applying the residue theorem. Depending on the sign of q = Im(q) the path integration has to be closed in the upper or lower half of the complex plane, respectively. We then obtain

xc+q[B( =iJS&O(

- q)O(k) - q)H(

- O(q)f3(

-k)]

- iq) - O(q>H(iq>], C-45)

with:
X(u):=c dkexp( -$+ku).

(A6)

The value of 8 is 1 for positive and 0 for negative arguments. Substituting x = ak - U/U and with the definition
R. Brendel and D. Bormann 5

Downloaded 26 Jul 2012 to 130.237.29.138. Redistribution subject to AIP license or copyright; see http://jap.aip.org/about/rights_and_permissions

erf(y):=(~)1 2J~

dxexp( -G)

(A71

[sgn(

-q>

erf(iq/a)].

(A91

of the error function, Eq. (A6) yields H(U)=&iexp($)[ 1 +erf(i)] . (A8) The function sgn is - 1 for negative and + 1 for positive arguments. If Eq. (A9) is finally substituted back into E!q. (A4) we obtain the result 1
X+BJexp( - 7; a)2) [ SgIl(d)
-fxf~'xo~O))] -exp( -'~~a'2)[sgn(-o") -efif'xo,u')]]

Substituting Eq. (A8) back into Eq. (A5) we obtain

=e;[

exp( - (;jxo 2)erfc(

( ix )

ew

- (;+ xo)i)edc( (a 2 ) (
I

I*

(AlO

Grosse, B. Harbecke, B. Heinz, and R. Meyer, Appl. Phys. A 39,257 P. (1986). *P. Grosse, Mikrochim. Acta (to be published). 3 R. Brendel, Appl. Phys. A 50, 587 (1990). 4M. Maeda and H. Nakamura, J. Appl. Phys. 58, 484 (1985). Bustarret, M. Bensouda, M. C Habrard, J. C. Bruyere, S. Poulin, E. and S. C. Guyrathi, Phys. Rev. B 38, 8171 (1988). Harbecke, Appl. Phys. B 39, 165 (1986). B. ID. Y. Smith, E. Shiles, and M. Inokuti, in Handbook of Optical Constants of Solids, edited by E. D. Palik (Academic, Orlando, FL., 1985), p. 394. N. J. Harrick, Internal Reflection Spectroscopy (Wiley Interscience, New York, 1967), p. 138.

Lucovsky, R. J. Nemanich, and J. C. Knights, Phys. Rev. B 19,2064 G. (1979). G. Lucovsky, Solid State Commun. 29, 571 (1979). P. Brilesch, R. Kijtz, H. Neff, and L. Pietronero, Phys. Rev. B 29,469l (i984). H. SchnBckel, Z. Anorg. Allg. Chem. 460, 37 (1980). 13E. Bustarret, Ph.D. thesis, University of Grenoble, France, 1988, p. 66. i4Y . J . Chabal, E. E. Chaban, and S. B. Christman, J. Electron Spectrosc. Relat. Phenom. 29, 35 (1983). A. L. Smith and N. C. Angelotti, Spectrochim. Acta 15, 412 (1959). 16D M. Makowieki, D. A. Lynch, Jr., and K. D. Carlson, J. Phys. Chem. 7i, 1963 (1971).

J. Appl. Phys., Vol. 71, No. 1, 1 January 1992

R. Brendel and D. Bormann

Downloaded 26 Jul 2012 to 130.237.29.138. Redistribution subject to AIP license or copyright; see http://jap.aip.org/about/rights_and_permissions