This document summarizes the measurement and modeling of the dielectric functions for In1-yAlyAs films on InP substrates with Al compositions (y) ranging from 0.305 to 0.635. Spectroscopic ellipsometry was used to measure the pseudo-dielectric functions over the 1.5-5.6 eV range. A critical point-based model was developed to parameterize the dielectric functions as a function of y. The model was then used in applications such as monitoring epitaxial layer uniformity and developing an etch process.
This document summarizes the measurement and modeling of the dielectric functions for In1-yAlyAs films on InP substrates with Al compositions (y) ranging from 0.305 to 0.635. Spectroscopic ellipsometry was used to measure the pseudo-dielectric functions over the 1.5-5.6 eV range. A critical point-based model was developed to parameterize the dielectric functions as a function of y. The model was then used in applications such as monitoring epitaxial layer uniformity and developing an etch process.
This document summarizes the measurement and modeling of the dielectric functions for In1-yAlyAs films on InP substrates with Al compositions (y) ranging from 0.305 to 0.635. Spectroscopic ellipsometry was used to measure the pseudo-dielectric functions over the 1.5-5.6 eV range. A critical point-based model was developed to parameterize the dielectric functions as a function of y. The model was then used in applications such as monitoring epitaxial layer uniformity and developing an etch process.
Dielectric function modeling for In Al As on InP 1yy y Leonard I. Kamlet, Fred L. Terry Jr. U Department of Electrical Engineering and Computer Science, 111 DTM Bldg., The Uniersity of Michigan, 2360 Bonisteel Bld., Ann Arbor, MI 48109-2108, USA Abstract . The pseudo-dielectric functions for In Al As on InP 0.305FyF0.635 were measured using spectroscopic ellipsometry 1yy y over the 1.55.6 eV photon energy range. The layer compositions were measured using double crystal X-ray diffraction. The pseudo-dielectric functions were t using a critical point-based approach, and a numerical dielectric function model was generated with the Al fraction, y, as the only free variable. The extracted optical responses were used in process monitoring applications. 1998 Elsevier Science S.A. Keywords: InAlAs dielectric functions; Dielectric function modeling; Process monitoring; Strained layer epitaxy 1. Introduction The lattice-matched In Al AsrIn Ga AsrInP 1yy y 1yx x materials system is used in a growing number of high frequency communication and optoelectronic inte- grated circuit applications. In some cases, small amounts of strain are intentionally introduced into the epitaxial layers to change electrical andror opti- cal properties. Also, variations in epitaxial growth systems will produce unintended compositional . changes and, therefore, changes in the lm strain . For In Al As on InP, there is only one composi- 1yy y . tion yf0.48 at which the two materials are lattice- matched. Thin epitaxial lms are under compressive . strain for In-rich lms y-0.48 , and under tensile . strain for Al-rich lms y)0.48 . The dielectric func- tions of the epitaxial lms are changed by both com- w x position and strain effects 13 . It has been well established that spectroscopic el- . lipsometry SE and normal-incidence spectral reec- . tometry SR can be used for highly accurate non-de- structive lm thickness measurements on multilayer structures provided that the dielectric functions of the U Corresponding author. Tel.: q1 734 7639764; fax: q1 734 9360347; e-mail: fredty@umich.edu materials in the layers are known. However, use of inaccurate dielectric function data in reection-based thickness measurements can lead to signicant thick- ness errors. Therefore, accurate dielectric functions . and a parametric model as a function of y for In Al As are needed for run-to-run epitaxial 1yy y growth control, reactive ion etch control, and other room-temperature process monitoring applications. The dielectric function of In Al As on InP has 0.52 0.48 been previously reported for the 2801900 nm wave- . w x length range ;0.65 to 4.43 eV 4 and the composi- tion-dependent refractive indices of In Al As on 1yy y GaAs have been reported at laser wavelengths of w x 632.8 nm and 1530 nm 5 . However, the dielectric functions of In Al As on InP in the near infrared 1yy y to near ultraviolet range have not been widely re- ported for non-lattice matched compositions. In this paper, we will present the pseudo-dielectric functions . for ve compositions ys0.305, 0.44, 0.48, 0.55, 0.635 of In Al As. We will discuss a composition-depen- 1yy y dent parameterization for the complex dielectric func- tion for these pseudo-dielectric functions. We de- scribe our parameterization scheme, which is an ex- tension of the critical point parabolic band method of w x Kim 6,7 . This is a phenomenological method which is based on single electron transitions in the semicon- 0040-6090r98r$19.00 1998 Elsevier Science S.A. All rights reserved . PI I S0 0 4 0 - 6 0 9 0 9 7 0 0 8 0 7 - 9 ( ) L.I. Kamlet, F.L. Terry Jr. rThin Solid Films 313314 1998 177182 178 ductor, and has been previously used for composition- dependent parameterization of the dielectric func- w x tions for lattice-matched materials systems 810 . In this paper, we will present a rst application of this method to strained layer materials. . We will also show results using the InAlAs data in the analysis of small spot spectral reectome- . try data in two process monitoring applications: i mapping of as-grown epitaxial layer uniformity; and, . ii the development of an electron cyclotron reso- . nance ECR reactive ion etch process for the base contact of an InAlAsrInGaAsrInP heterojunction . bipolar transistor HBT . 2. Experimental details . Five samples were grown on 001 Fe-doped InP substrates by solid-source molecular beam epitaxy . MBE in a Riber 2300. The layers were grown at a substrate temperature of 490C with an As pressure of 1.31.4=10 y5 torr. The samples were nominally
;10 A InGaAsrIn Al AsrInP samples with ys
1yy y 0.3050.635. The InGaAs capping layer was grown under ux conditions that produce lattice matched material in steady state growth. The capping layer was incorporated to suppress oxidation of the Al-contain- ing lm and possible optical complications due to oxidation-induced roughness. All target thicknesses were below the Matthews-Blakeslee critical thickness w x 11 for the given composition; however, in some cases this thickness was exceeded, creating the possibility of incoherent strain in those lms. Compositions were determined using double crystal . X-ray diffraction DCXRD . Rocking curves from the . 004 reection were obtained using a Rigaku diffrac- tometer using the Cu K wavelength. The lm com-
position was determined using a program based on
w x dynamical X-ray scattering theory 12 and Vegards law. Spectroscopic ellipsometry data were collected us- ing a Sopra GESP-5 system over the 1.55.6 eV range in increments of 0.01 eV. The instrument was oper- ated in the tracking analyzer mode. We used two separate prism-grating monochromators, one of which was optimized for UV operation. Data were collected on both monochromators between 2 and 3 eV and examined to ensure consistency of the two data sets. The monochromator change-over was done during the collection of data from each sample to avoid variations in sample alignment. Data that did not . agree to be better than 0.01 in both tan and . cos were recollected following realignment of the sample. SE data were collected at both 70 and 75 angles of incidence. We intended to collect data below the fundamental band edge, but the combined effects of reection from the back surface of the sample, the . well-known difculties of measuring at low tan conditions with rotating-polarizer ellipsometers, and the increased noise of the InGaAs near-infrared de- . tector vs. the UV-visible photomultiplier made these data unreliable. 3. Results and discussion Extraction of the pseudo-dielectric function from the ellipsometry data was performed using a two-step process. A complex harmonic oscillator approxima- . w x tion CHOA 13 with ve oscillators was used as a rst model for the InAlAs dielectric response. The thickness of the InAlAs layer, the InGaAs cap layer . plus thin native oxide over-layer , and a CHOA esti- mate of the InAlAs dielectric response were de- termined by non-linear least squares regression using the Levenberg-Marquardt method. For initial guesses, . the 20 parameters four per oscillator for the CHOA model were found for each composition by using the values from the next nearest composition and adjust- ing oscillator energies due to peak shifts. The rst regression analyses were performed on data for the . lattice matched sample no. 1295 with initial guesses for the oscillator parameters being chosen by assign- ing oscillators to the E , E q , and E transitions, 1 1 1 2 a broad linewidth oscillator to the region between
E q and E to ll-in the response between the
1 1 2 . critical points , and a broad linewidth oscillator above . E to adjust the baseline and the response . 2 The oscillators associated with the critical points can be initially estimated by tting to second derivative data; however, in this case the initial guesses were chosen by shifting the energies of the corresponding w x oscillators from our t to GaAs 13 . The pseudo- dielectric function for each InAlAs composition was then extracted on a point by point basis using the thicknesses found previously from the CHOA t. To reduce numerical artifacts, we then re-t the CHOA parameters to the resulting dielectric function and repeated the two-step process. The data were then
visually examined for interference oscillations thick-
. ness error-induced effects . Initial guesses for the regression were adjusted until an apparent best t was found. Compositional data were collected by DCXRD as- suming coherently strained lms. In principle, the Pendellosung fringes of the rocking curves will pro- vide an independent estimate for the epitaxial lm. . However, only the lattice-matched sample no. 1295 . and one strained sample no. 1296 showed clear Pendellosung oscillations. This could have been the result of interface quality problems, but may also be the result of thickness non-uniformity in the epitaxial layers. For the strained sample, the extracted thick- ness from DCXRD was ;29.0 nm based on a t- ( ) L.I. Kamlet, F.L. Terry Jr. rThin Solid Films 313314 1998 177182 179 by-eye alignment of the oscillation peaks and valleys. The regression procedure described above yielded 27.6 nm from the SE data. The thickness and composition data are summa- rized in Table 1. As can be seen, the measured thicknesses were greater than the growth targets in all cases, while the measured compositions are very close to the target values. The reasons for this discrepancy are not known. However, because of the thicker- than-desired InAlAs layers, the assumption of coher- ent strain may be erroneous, and, therefore, our de- rived model should only be regarded as a rst attempt at characterizing this material system. A study of both
thinner lms which are more likely to be coherently
. strained and of thick, fully-relaxed lms at various Al-fractions is needed. This was, unfortunately, bey- ond the nancial scope of our project. To examine further the possibility of relaxation of the strain, we examined three samples for variations . in cos with azimuthal angle at the HeNe laser . wavelength 632.8 nm on a Gaertner L116B rotating analyzer ellipsometer equipped with a rotational stage. Without surface roughness due to incoherent growth, w x no variation with azimuthal angle is expected 14 . The results are shown in Fig. 1. The lattice matched . sample no. 1295 was analyzed, and the results can be regarded as an error bound for this measurement, as there should be no variation for this sample. These
variations introduce signicant uncertainties as large
. as 53% for sample no. 1298 in the values of the imaginary part of the dielectric function, , but 2 . relatively small errors ;1% or less for the real part, . Furthermore, at this wavelength is small and is 1 2 therefore subject to large errors in extraction from rotating element ellipsometry. We have not yet syste- matically studied the azimuthal anisotropy of these samples at other wavelengths, but a few SE measure- ments taken at different substrate orientations did not indicate large differences at the shorter wave- lengths. Another possible cause of the observed azi- muthal variations is that in some cases the InAlAs
thicknesses are non-uniform across the samples as
. will be shown later , leading to possible sensitivity in the precise beam location during measurement. Other well-documented ellipsometry alignment issues, such as beam wobble, may also be a factor due to the layer thickness non-uniformity. The extracted pseudo-dielectric functions of the samples were tted using the modications to the Kim-Garland approach which we have presented pre- viously for the dielectric functions of lattice-matched w x In Ga As P on InP 10 . This parameterization 1yx x y 1yy scheme assumes that the dielectric function is isotropic and that the semiconductor is unstrained. The optical anisotropy expected in uniaxial III-V compounds is difcult to resolve with ellipsometry, as the angle of refraction varies only slightly with angle of incidence due to the high refractive index. Aspnes has examined this problem and notes that for rela- < < . tively high dielectric constant materials ;10 the surface normal component of the dielectric function will have little effect on the ellipsometrically mea- sured pseudo-dielectric function and that the mea- sured pseudo-dielectric function is the projection of the dielectric function tensor onto the line formed by the sample surface and the measurement plane of w x incidence 15 . De Smet has shown numerically that even for the case of uniaxial crystals, multiple angle of incidence ellipsometry measurements are insuf- cient to measure the normal and extraordinary com- w x ponents of the dielectric function separately 16 . Therefore, we ignore optical anisotropy in our model. In principle, the dielectric function model should be modied to account for changes in the band struc- ture due to strain. The most signicant effects are shifts in the energies, amplitudes, and linewidths of w x the critical points 13,17 ; however, these effects must be treated in any composition-dependent parameterization problem. The parameters needed to estimate the energy, amplitude, and linewidth changes accurately as a function of composition would pre- sumably be different for the strained vs. the relaxed case, but no other changes in the parameterization scheme should be needed due to these effects. We expected that the most signicant effects which would require changes in the model formulae to be the reduction in degeneracy and reduction of dimension- . . w x ality of the E X and E K transitions 18 . We 2 2 modied the Kim-Garland expression to include a two-dimensional critical point for parallel strain and a Table 1 A summary of the targetes thickness and compositions of the InAlAs samples studied, and the thicknesses and compositions obtained by SE and DCXRD Sample no. Target thickness Target Al SE thickness DCXRD Al Critical thickness . . . nm fraction nm fraction nm 1295 300.0 0.48 430.4 0.480 1296 15.0 0.33 27.6 0.305 23.25 1297 70.0 0.43 119.6 0.440 93.56 1298 70.0 0.53 102.7 0.550 76.81 1299 15.0 0.63 33.4 0.635 21.21 ( ) L.I. Kamlet, F.L. Terry Jr. rThin Solid Films 313314 1998 177182 180 . Fig. 1. Variation in cos at s623.8 nm with azimuthal angle . for the latticed matched lm no. 1295 and two strained layer lms. one-dimensional critical point for perpendicular . strain. For a uniaxial crystal grown in the 001 direc- tion, both types should occur. We t the dielectric functions using the methods w x detailed in earlier work 610,18 . The quality of the extracted pseudo-dielectric functions was not suf- cient to support second and third derivative tting methods. Therefore, only the pseudo-dielectric func- tions and their rst derivatives were used in the ts. Very good ts to the primary data were achieved, but the lack of higher derivative data increases the con- dence limits that can be placed on the critical point energies and lineshapes. In all cases the use of the reduced-degeneracy formulation resulted in worse ts than the isotropic formulation; therefore, only the isotropic results will be shown. The reasons for this are not clear, but possibilities include incoherent strain effects or insufcient resolution in our experi- mental data. Two separate ts using the modied Kim parame- terization scheme were performed. First, the model parameters were optimized for each composition in- . dependently individual ts . Second, a composition- dependent parametric model was generated by a global optimization over all the data sets. The individ- ual data and ts are shown in Fig. 2a and b. The E 1
and E q critical point energies along with
1 1 . parabolic ts of the energies vs. y from the individual ts are shown in Fig. 3. The variation of these ener- gies follows the trend expected from the critical point energies of the relaxed binary endpoints of InAs and AlAs. Also, the amplitude of the E critical point 1 increased relative to that of the E q critical point 1 1 as the In concentration was increased, and the oppo- site occurred as the Al concentration was increased. Unfortunately, no strong conclusions can be drawn regarding the strain state of the epitaxial lms from these observations, as both are consistent with what Fig. 2. The measured and best t pseudo-dielectric functions for . . the InAlAs lms studied here; a real part, b imaginary part. would be expected from compositional changes as w x well as strain-induced changes 1,3 . The strong strain-induced reduction of the E q energy shift 1 1 vs. In content for strained layers of InGaAs on GaAs w x observed by Pickering 19 is not evident in this data. This may also be an indication that our lms are not coherently strained. Only a small shift in the E 2 critical point was observed. The composition-dependent model was generated using the following procedure. The energies from the individual ts were tted to quadratic polynomial functions of the Al fraction y, and the linewidths to cubics using the DCXRD-determined y values. These factors were then held constant and the amplitude factors were tted as cubic functions of y by simulta- neously considering all data sets. Nearly equivalent quality ts to the measured data were obtained in all ( ) L.I. Kamlet, F.L. Terry Jr. rThin Solid Films 313314 1998 177182 181 . . Fig. 3. The E round points and E q square points critical 1 1 1 point energies from our parameter optimization for individual ts vs. the Al fraction, y. The dashed lines are quadratic functions tted to these critical point energies. . w x cases visually equivalent to the plots of Fig. 2 20 . . For example, the root mean square error MSE , 2 2 meas f i t meas f i t . . y q y rN where N ' / E 1 1 2 2 is the number of points, and the summation is over . the measured photon energies, E , for the individual . t to the lattice matched data no. 1295 was 0.217 while that of the composition-dependent t was 0.361. The error per point is typically well below 1% for
both and except for a narrow error spike
1 2 . ;"10% around E . As all available data were 2 needed for the compositional t, an objective test of the accuracy in extracting y for an intermediate com- position could not be run. We have made use of the InAlAs pseudo-dielectric function data in two process monitoring applications. The extracted pseudo-dielectric function data from sample no. 1297 were used in a Leitz MPV-SP micros- . cope-based spectral reectometer SR with an auto- mated stage to map the thickness uniformity of the growth run. The sample was not rotating during growth, and the growth conditions were not optimized for uniformity at this composition. The resulting non- uniformity is evident in the thickness map of Fig. 4. A sample from the thicker left edge of this wafer was measured by XTEM, and yielded about ;150 nm . lm thickness sample preparation was not optimal in at least qualitative agreement with the SR map. In a second application, we have used these data for pre- and post-ECR etch thickness measurements of In- AlAsrInGaAsrInP HBT device structures. The start- ing InAlAs and InGaAs could be determined and mapped for uniformity on the SR. These thicknesses were used in estimating required etch times. As an example, one pre-etch measurement yielded thick- nesses of 186"1 nm of InAlAs on top of 63.5"0.5 nm of InGaAs from the data shown in Fig. 5. Use of Fig. 4. A wafer map obtained by automated microscope-based spectral reectometry of the InAlAs thickness on sample no. 1297. ( ) L.I. Kamlet, F.L. Terry Jr. rThin Solid Films 313314 1998 177182 182 . Fig. 5. The reectance vs. wavelength data points and best t . solid curve for an InAlAsrInGaAsrInP HBT test structure yield- ing 186"1 nm of InAlAs over 63.5"0.5 nm of InGaAs. this procedure in verifying the accuracy of an optical emission endpoint detection scheme will be described
in a separate publication Hanish et al., unpublished
. results . 4. Summary We have made an initial investigation of the dielec- tric functions of In Al As epitaxial lms grown on 1yy y InP and have parameterized the ts to produce a numerically efcient compositional model for use in process monitoring applications. Our original intent was to examine coherently strained lms so that the paramaterization results could be applied to moni- toring applications in which variations around the nominally lattice-matched case are encountered. However, as discussed above, the lms in our study may not be coherently strained. Additional studies of coherently strained and fully relaxed lms in this materials system are needed. Acknowledgements This work was supported in part by DARPA under contract F33615-92-C-5972. The authors would also like to thank Drs. Xiangkun Zhang and Pallab K. Bhattacharya for growth of the MBE samples, and Arun Ramamoorthy for performing the TEM analy- sis. We would also like to thank the State of Michigan Center for Display Technology and Manufacturing for use of the GESP-5. References w x . 1 F.H. Pollak, M. Cardona, Phys. Rev. 172 1968 816. w x . 2 M. Chandrasekhar, F.H. Pollak, Phys. Rev. B 15 1977 2127. w x . 3 F.H. Pollack, Semiconductors Semimetals 32 1990 17. w x 4 H.W. Dinges, H. Burkhard, R. Losch, H. Nickel, W. Schlapp, . Thin Solid Films 233 1993 145. w x 5 J.H. Baek, B. Lee, S.W. Choi, J.H. Lee, E.H. Lee, W.S. Han, . Ungyong-Mulli. 10 1997 48. w x 6 C.C. Kim, J.W. Garland, H. Abad, P.M. Raccah, Phys. Rev. B . 45 1992 11749. w x 7 C.C. Kim, Ph.D. Thesis, Univ. of Illinois at Chicago, 1991. w x . 8 C.C. Kim, J.W. Garland, P.M. Raccah, Phys. Rev. B 47 1993 1876. w x . 9 C.C. Kim, S. Sivananthan, J. Appl. Phys. 78 1995 4003. w x . 10 L.I. Kamlet, F.L. Terry, Jr., J. Elect. Mater. 24 1995 2005. w x . 11 J.W. Matthews, A.E. Blakeslee, J. Crystal Growth 27 1974 118. w x 12 Y.-C. Chen, Ph.D. Thesis, Univ. of Michigan, 1992. w x . 13 F.L. Terry, Jr., J. Appl. Phys. 70 1991 409. w x 14 M.T. Sinn, J.A. del-Alamo, B.R. Bennett, K. Haberman, F.G. . Celii, J. Elect. Mater. 25 1996 313. w x . 15 D.E. Aspnes, J. Opt. Soc. Am. 70 1980 1275. w x . 16 D.J. De Smet, J. Appl. Phys. 76 1994 2571. w x . 17 H. Lee, D. Biwas, M.V. Klein, et al., J. Appl. Phys. 75 1994 5040. w x 18 L.I. Kamlet, Ph.D. Thesis, Univ. of Michigan, 1996. w x 19 C. Pickering, R.T. Carline, M.T. Emeny, N.S. Garawal, L.K. . Howard, Appl. Phys. Lett. 60 1992 2412. w x 20 Data and t parameters can be obtained by anonymous ftp from the site ftp.eecs.umich.edu under the directory peoplerfredty. The authors request an e-mail notication to fredty@umich.edu from those using data from this work.
Fitted electronic density functions from H to Rn for use in quantum similarity measures_ cis-diamminedichloroplatinum(II) complex as an application example (Journal of Computational Chemistry, vol. 20, issue 9) (1999)
Three and Four-Electrode Electrochemical Impedance Spectroscopy Studies Using Embedded Composite Thin Film Pseudo-Reference Electrodes in Proton Exchange Membrane Fuel Cells