Extended defects of ion-implanted

Transcription

1 Extended defects of ion-implanted GaAs K. S. Jones Department of MateriaIs Science & Engineering, University of Florida, Gainesville, Florida E. L. Allen, H. G. Robinson, and D. A. Stevenson Department of Materials Science & Engineering, Stanford University, Stanford, California 94305,... M. D. Deal and J. D. Plummer Integrated Circuits Laboratory Stanford University, Stanford, California (Received 6 June 1991; accepted for publication 26 August 1991) Ion-implantation-induced extended defect formation and annealing processes have been studied in GaAs. Mg, Be, Si, Ge, and Sn ions were implanted between 40 and 185 kev over the dose range of 1 X 10t3-1 X 1015/cm2. Furnace annealing after capping with S&N4 was performed between 700 and 900 C for times between 5 min and 10 h. Plan-view and crosssectional transmission electron microscopy results were correlated with secondary-ionmass spectroscopy profiles. The results indicate subthreshold (type-i) defect formation occurs at a dose of 1 X 10i4/cm2 for high-energy, light (Mg, Be) ions but not for heavier ions (Si, Ge, Sn) at shallower projected ranges ( < 500 A). Si and Ge implants at a dose of 1 x 10 s/cm2 both show extended defect formation upon annealing that is believed to be precipitation related (type-v defects). For Si implants, these dislocation loops are eliminated after 10 h at 900 C. Upon annealing 1 X 1015/cm2 Sn implants, unusual precipitate motion both toward the surface and into the crystal was observed. Type-II defects are observed but only in the as-implanted cross section. In addition, a layer of dislocatiqn loops formed at a depth much greater than the type-ii defect layer. These defects appear to be a new type of defect possibly related to either the different binary recoil distributions of Ga and As or differences in vacancy and interstitial diffusivities. It is shown that, with modifications to account for the binary nature of the target, the classification scheme developed for extended defects in silicon can be applied to implantation of gallium arsenide. I. lntroductlon Ion implantation of GaAs has been used for a number of years for both doping and isolation purposes in microelectronic applications, as well as for superlattice mixing in optoelectronic applications. The interaction between implantation-related point~defects, impurity diffusion, and extended defect formation is well documented in silicon.2*3 For dopant implantations in GaAs, it is generally recognized that point defects, rather than extended defects, are more important for electrical activation.4y5 However, it has been shown that dislocations can indirectly influence the electrical activation.6 7 In addition, recent results indicate the point defects introduced by the ion implantation process can affect the subsequent diffusion processesg- 2 as well as the process of extended defect formation. Finally, for superlattice intermixing studies, dislocation loop formation has been shown to retard the interdiffusion process.13 Thus, there is a need to understand the source of extended defects, in order to optimize the modelling of diffusion, electrical activation, and superlattice mixing in ion-implanted GaAs. The sources of ion-implantation-related defects in Si and GaAs have been investigated for a number of years. For silicon studies, this culminated in a five category classification scheme for defects based on their origin3 Type-I (subthreshold) defects form above a certain dose (1 x 1014/cm2) but prior to amorphization. They are generally observed for lighter ion implantation. Type-II (end of range), -111 (solid-phase regrowth related), and -IV (clamshell) defects are associated with amorphization and regrowth. Because of their dependence on the amorphization process, these defects in general are very sensitive to the point defects that arise from the elastic collision processes and, therefore, their formation kinetics.are more sensitive to ion mass than species. Type-V (precipitation related) defects arise upon annealing impurity profiles that exceed the solid solubility of the host matrix. The most common form are precipitates, although other defects such as dislocation loops can arise from the precipitation process. In general, most of the previous transmission electron microscopy (TEM) studies of implanted GaAs lack the combination of as-implanted and annealed cross-sectional samples necessary to determine the position of extended defects relative to the damage density distribution and the ion distribution. This correlation is essential to discern the source of the defects. In spite of the absence of such a correlation, the following observations have been made. Extrinsic dislocation loops have been reported upon annealing high-energy (>I90 kev), low-dose (<5X1014/cm2) Sef,14 P+,15 and Si+ (Ref. 169 implants. These dislocation loops are probably type-1 defects. Type-III defects always form upon solid-phase epitaxial regrowth of any amorphous layers in implanted GaAs, re gardless of the substrate orientation, if the amorphous layer is greater than 400 %, thick. 7-u These defects (mi J. Appl. Phys. 70 (1 I), 1 December 1991 American institute of Physics 6790

2 Be kev Mg 150 kev Si 40 kev Ge+ 110 kev Sn+ 185 kev FIG. 1. Plan-view TEM (bright field, guo) results of 1 X 10 4/cma implants after annealing at 900 C for 15 min for Be and Mg, 30 min for Si, Sn, and Ge. crotwins and stacking faults) dissolve upon annealing at 600 0C.1*23124 The only report of possible type-iv defects is for dislocation loops observed in the GaAs layers of implanted GaAs/AlAs superlattices. A number of type-v defects has also been reported. Metallic precipitate formation has been reported for Sn (Ref. 26) and Zn (Ref. 27) implants. Sadana4 noted that dislocation loops form in Se + -implanted GaAs and first suggested that the disldcations were related to exceeding the solid solubility. Opyd et ~1.~~ noted type-v dislocation loop formation for Si + implants, and Morita and co-workers27 also observed type-v dislocation loops in Zn -implanted GaAs. Dislocation loops were found in annealed Sn-implanted GaAs for doses between 5X 1014 and lx 1015/cm2 by Shim, Itok, and Yamamoto.28 These dislocations might be type V or a different type of defect, as will be discussed. They also observed precipitate formation for doses of > 1 X 1015/cm2. Finally, Ar bubbles have been observed upon annealing Ar + -implanted G~As.~ This paper represents the first attempt to extend the defect classification scheme for implanted Si to the implantation of GaAs. A number of species (Be, Mg, Si, Ge, Sn), at various doses and annealing conditions, have been studied in an effort to better understand and predict the conditions that result in extended defects formation and elimination. II. EXPERIMENTAL Undoped Furakawa (100) liquid-encapsulated Czochralski (LEC) semi-insulating GaAs wafers were implanted with Mg+, Be +, lzosn +, 74Ge +, or 2gSi + at energies between 40. and 185 kev and doses between 1 X 1013 and 1 X 10t5/cm2. The current density varied between 0.1 and 0.4,uA/cm2. The implants were done at room temperature. The samples were capped with =:900-A Si3N4 by plasma-enhanced chemical-vapor deposition (PECVD) prior to furnace annealing. The furnace annealing was done between 700 and 900 C for times from 5 min to 10 h. Both plan-view and cross-sectional TEM analysis was done on either a JEOL 200CX or JEOL 4000FX using two-beam bright field gea or g220 imaging conditions. Ill. RESULTS Figure 1 shows plan-view TEM results of 1 x 10t4/cm2 implants at various energies after 900 C! annealing. Category I defects are observed only in the case of Mg + or Be + implants. All dislocation loops were determined to be extrinsic. These defects persist for up to 4 h at 900 C. For Si, Sn, and Ge no dislocation loops were observed. When the dose for the Mg implantation was reduced to 1 X 10t3/cm2 no extended defects were observed upon annealing at 900 C for 15 min. In Fig. 2, a cross-sectional TEM analysis combined with secondary-ion-mass spectroscopy (SIMS) profiles (same scale) show that the subthreshold defects (for a 150-keV 1 i< 1014/cm2 Mg implant) are distributed in a wide band around the original projected range. Upon increasing the dose to 1 X 1015/cm2 the following observations were made. Implantation and annealing of Sn implants at a dose of 1 X 1015/cm2 is shown in cross section in Fig. 3. Again the cross-sectional TEM and the SIMS plots are on the same scale. The as-implanted morphology shows a surface amorphous layer 700 A thick and a layer of type-ii defects at 900 A below the surface. 4nnealing at 700 C! results in a well-defined band of precipitates as well as a layer of loops at a deeper depth. These loops are at a depth ( 1600 A) much greater than the location of the type-11 defects in the as-implanted microstructure (900 A). These loops are-therefore not type-ii defects. The Sn-related dislocation loops may be a form of type-1 defects, not observed in silicon, that are related to the recoil distribution, as will be discussed. Upon additional annealing between 700 and 900 C!, two different phenomena are observed. First, the band of precipitates was observed to move deeper into the crystal. This was confirmed by SIMS measurements and is plotted from TEM results in Fig. 4. This motion of the precipitate band is very unusual in that it involves uphill diffusion. Second, 6791 J. Appl. Phys., Vol. 70, No. 11, 1 December 1991 Jones et a/. 6791

3 Implanted Mg O15W.... Depth (microns) FIG. 2. Cross-sectional TEM (bright field, g2& micr&raph and SIMS plot of 150-keV 1 X 10 4/cmZ Mg + annealing. implant after 900% 15-min the dislocation loops formed at 1600 A are shown to dissolve completely after 30 min at 900 C. Upon implantation with llo-kev Ge + at a dose of 1 x 10 5/cm2, amorphization was again observed (Fig. 5) with solid-phase epitaxy occurring either during the implantation process by ion-beam-induced epitaxial crystallization (IBIEC), during the S&N4 capping process or during the sample preparation process. The type-ii defects are seen in a fine layer z A deep. The large dislocation loops that form upon annealing at 900 C! are believecj to be type-v (precipitatiop related) defects. The loops form at a depth ( A) which is close to the projected range as measured by SIMS (400 A) and shallower than the type-ii defects position. For 40-keV Si implants at a dose of 1 X 101 /cm2 again any amorphous layer has been regrown either by IBIEC or during sample preparation (Fig. 6). Between 5 and 30 min at 900 C, defects are observed to form between 400 and 900 A deep, which is slightly deeper than the SIMS projected range of 275 A. These defects appear to be type-v defects. Further annealing results in complete elimination of this dislocation layer after 10 h at 900 C. IV. DISCUSSION As.lmplanted 703-c Concentration (cm ) There are definite differences in type-1 defect formation in GaAs and Si. The threshold dose for type-1 defect formation in Si substrates does not show a strong energy or mass dependence.3 However, in GaAs there appears to be both a mass and energy dependence that may be the result of the different recoil distribution of Ga and As or differences in point-defect diffusivities. This is supported by the observation that the defects in the Mg + implant were distributed over a broad depth range whereas in implanted Si they are typically in a well-defined layer.3 A previous report by Stewart et al. of low-dose ( <10 4/cm2) 240-keV Si+ implants in GaAs also shows the type-1 defects distributed over a broader depth range ( A).29 8oo C- SQO C 0 Concentration (cm ) 4 s; $ GA Zi < c. ID D Es :!2 8 k w 1400rn FIG. 3. Cross-sectional TEM micrographs of 18%keV Sn, 1 X 10 5/cm2 implants and corresponding SIMS profiles. The anneals were done for 30 min. The as-implanted cross-sectional image is a weak beam dark field, g*20 micrograph. The amorphous layer is labeled (a). All annealed crosssectional images are bright field g,,, micrographs. The surface is indicated by an arrow Anneal Temperature ( C) (30 minutes) FIG. 4. Graph of the deeper precipitate layer motion upon 30-min annealing for the 185-keV Sn 1 X lo /cm* implant J. Appl. Phys., Vol. 70, No. 11, 1 December 1991 Jones et al. 6792

4 As-Implanted 900-C 30 minutes Concentration (cm? mg. 5. Cross-sectional~and plan-view (bright field gzzo) micrographs of a 1 lo-kev, 1 X 10 5/cm2 Ge + implant and the corresponding SIMS profile. The arrows in the as-implanted cross section denote the surface(s) and the type-11 defect layer (II). Note the well-defined network formation near RP upon annealing. The SIMS results show the projected range for the 1 x 10 4/cm2 Si, Ge, and Sn implants (no defects) is approximately 380 A whereas the projected range for the Mg and Be implants is 1900 and 5000 A, respectively. We have previously reported that even Soiver-energy 20-keV Si implants show no defects at doses of 1X 10 4/cm2 upon 900 C rapid thermal annealing (RTA)30 which would support a model that type-1 defect formation in GaAs is energy dependent. Other results for higher-energy, heavierion, 200-keV 31P + implants I5 (Rp zz 1950 A from TRIM 190) show loop formation at doses as low as 3 x 1013/cm2 and 300-keV Se+ (R z 1220 A from TRIM 190) and 240-keV Sif Rp z 2500 Af rom TRIM 90) shows a threshold dose as low as 1 X 10 3/cm2. 4~29 These results would imply that type-1 defect formation in GaAs is very sensitive to the implant energy, with the defects forming only for deeper implants. This would also support an argument that effects from the binary (Ga, As) nature of the target are more important for dislocation loop formation than the implanted ion concentration which is diluted upon increasing energy. Other Si+ and Se+ implant studies16 noted the dislocation loops formed at depths deeper than Rp which they have attributed to the migration of interstitials (implanted ions, Ga, and/or As recoils) to deeper depths prior to the nucleation of dislocation loops. There remains an unexplained previously reported result that 120-keV Se + implants yield dislocation loops at 6793 J. Appl. Phys., Vol. 70, No. 11, 1 December 1991 doses as low as 3 X 10 3/cm2, 4 despite the shallower projected range (RP A;, TRIM -90): This may indicate ihat defebt formation in Se + implants is somehow related to the chemistry of the ion. We are presently conducting additional studies to better determine the chemical and energy dependence of type-1 defect formation. The dislocation loops from the 185-keV 1 X 1015jcm2 Sn+ implants (Fig. 3) form at a depth (1600 A) much greater than the position of the type-ii defect layer (900 b;) in the as-implanted sample. Dislocation loop formation in Sn-implanted GaAs has also been reported by Shim and co-workers2s in this dose regime upon 900 C annealing, although the depth of the dislocation loops was not determined. As discussed previously,3 type-ii defects are,also called end-of-range dislocation loops and form in the damaged region just below the amorphous/crystalline interface. There is no mention in the literature of type-ii defect formation in GiAs. It is noted by studying the results of Opyd et al. 24 that for Si imp lants the type-ii defect layer observed to form just below the amorphous/crystalline interface apparently dissolves upon annealing at 600 C for 30 s. Our Si + and Ge + results support the observation that the type-ii defects in GaAs are very unstable. This is in stark contrast to type-ii defects in -iniplanted silicon, which are known to be. quite stable.3 30 The instability of type-ii defects would support the argument that the loops in the Sn-implanted sample are not type-ii loops. Simula- Jones et a/. 6793

5 As-Implanted 5 minutes 30 minutes Concentration (cm-3 s, g;, ;;, z:, ;;; 0 Cross-section Plan-view FIG. 6. Cross-sectional (weak beam dark field, gzzo) and plan-view (bright field, g,,,) micrograph of a 40-keV, 1 X 10 5/cm2 Si + implant and the correspoliding SIMS plot. The annealing temperature was 900 C. The arrows denote the surface. Note the development bf defects between 5- and 30-min annealing. tions using the Boltzmann transport equations developed in Ref. 32 show that after 185-keV Sn implantation, excess Ga interstitials exist from a depth of 500 to beyond 2000 A. The dislocation loops exist within this region, implying that these defects are of a new type, neither type I nor type II. If the excess Ga interstitials are responsible for extrinsic loops, then excess Ga vacvlcies should be available for Sn activation. Quantitatively, the concentration of interstitials in these loops is (from plan-view measurements) at most 2~ 1013/cm2 after 800 C 30-min annealing. Although this may be a significant fraction of the low-temperature.activated dose, observations that the activation continues to increase at higher temperatures after the loops have dissolved implies that if these extended defects affect the electrical activity it is only at lower annealing temperatures. These defects may be related to the variation in recoil distributions of the Ga and As, i.e., they may be a form of type-1 defects that form in a sample that has been amorphized. The loops appear to be relatively unstable with respect to normal type-1 or type-v defects having dissolved after 30 min at 900 C. It has been shown that the precipitates that form upon annealing the 1 X 1015/cm2 Sn+ implants are principally composed of the metallic P phase of tin.26 The formation of double concentration peaks upon Sn + post-implant anneals and their subsequent motion in opposite directions toward the surface and into the bulk constitutes separate uphill diffusion processes. Surface gettering could ac- count for the first peak, however, layer motion of the deeper peak and its associated precipitates is not well understood. In a separate experiment, the same results were obtained after rapid thermal annealing of uncapped samples, and therefore, the effect does not appear to arise from stress induced by the Si3N4 cap. It remains unclear what the driving force is for this phenomena. In order to attribute the dislocation loops observed upon annealing 1 X 10 5/cm2 Si + and Ge + implants to precipitation-related [type V) phenomena, it is necessary to determine if the peak concentration is above the solid solubility at the anneal temperature. Much of the Rutherford backscattering (RBS) solubility studies in implanted GaAs have focused on Sn, Sb, and Te.?3,34 They conclude that the solubility generally decreases between 700 and 900 C and varies with implant temperature. For the Si and Ge studies the 1 X 10 5/cm2 implant resulted in peak concentrations of 2 2 X 1020/cm3. The lack of solubility data for Ge makes it diecult to predict when precipitation occurs. However, given the previously discussed observations concerning the depth of the defects (near RJ and the instability upon annealing of other kinds (type II and III) of defects, it is reasonable to conclude that the defects observed upon annealing 1 X1015/em Ge implants are type-v defects. For the Si implants the defect formation process at 900 C, observed in Fig. 6, implies the loops are type-v defects. The electrical solubility of Si in GaAs at J. Appl. Phys., Vol. 70, No. 11, 1 December 1991 Jones et al. 6794

6 900 C is known to saturate at <5x 1018/cm3. Above this concentration Si becomes amphoteric. The solid solubility of Si in GaAs at 900 C as determined by RBS measurements has not been reported to the author s knowledge. From solid-source in-diffusion measurements the Si solubility has been estimated to be 10 at 900 0C,35 which is consistent with these being type-v defects. Again, the location of the defects, the instability of other types of defects, and the formation kinetics of the loops are consistent with type-v defects3 The formation of a high density of dislocations upon annealing higher-dose (> 1 x 10 5/cm2) silicon implants is consistent with previous studies. It has been previously noted that the loop morphology for higher-dose Si implants appear to vary with implant parameters such as dose rate. This is not generally observed for type-v defects formed in implanted silicon. V. CONCLUSIONS Type-I (subamorphization threshold) defects were shown to form upon annealing high-energy Mg and Be implants at a dose of =: 1 X 10t4/cm2. The defects- form over a broader depth range in GaAs than in silicon. GaAs appears to show a strong energ! dependence whereas Si does not. This would point to a mechanism of type-1 defect formation related to variations in the binary recoil distribution of differences in point-defect diffusivities. The type- II (end of range) defects never fully form well-defined dislocation loops and they dissolve at low temperatures (600 C, 30 s). This is very different from implanted silicon where the end-of-range defects consistently form and are one of the most stable forms of implantation-related extended defects. Type-III (regrowth related) defects are shown to dissolve by 700 C 30-min.anneaIs for Sn implants. Sn, Ge, and Si all produce type-v defects after a 1 x 10 /cm2 implant. For Sn implants a fine precipitate layer formed and split into two layers. The first layer of precipitates gettered near the surface and the second layer diffused deeper with increased annealing temperature or time. Both layers exhibit uphill diffusion. For Ge and Si implants the precipitation process results in dislocation loop formation. These type-v defects for the Si+ implants dissolve after 10 h at 900 C. Finally, a new category of defects was observed in the 1 i< lq15/cm2 Sri-implanted GaAs. Dislocation loops were observed at depths much greater than the amorphous/crystalline interface. It is proposed that these relatively unstable defects may be a new variation of type-1 defects that form in samples that have been amorphized. In general, the classification scheme developed for Si implantation applies to GaAs implantation with the addition of modifications to account for the dual nature of the interstitial recoil distributions in GaAs. ACKNOWLEDGMENTS. The authors would like to acknowledge the assistance of Alex Azan in the TEM sample preparation and the financial support of DARPA J. Appl. Phys., Vol. 70, No. 11, 1 December 1991 S. J. Pearton, Solid State Phenomena I/2, 247 (1988). *J. Narayan and 0. W. Holland, J. Electrochem. Sot. Solid-State Sci. Technol. 131, 2651 (1984). 3K. S.Jones, S. Prussin, and E. R. Weber, Appl. Phys. A 45, 1 (1988). 4D. K. Sadana, Nucl. Instrum. Methods Phys. Rea. B 7/S, 375 (1985). S. J. Pearton, J. M. Poate, F. Sette, J. M. Gibson, D. C. Jacobson, and J. S. Williams, Nucl. Instrum. Methods Phys. Res. 19/20, 369 (1987). F. Hyuga, J. Appl. Phys. 64, 3880 (1988). 7 S. Miyazawa and K. Wada, Appl. Phys. Lett. 48, 905 (1986). *H. G. Robinson, M. D. Deal, and D. A. Stevenson, Appl. Phys. Lett. 56, ). H. G. Robinson, M. D. Deal, and D. A. Stevenson, in Proceedings of Impurities, Defects and DSffusion in Semiconductors: Bulk and Layered Structures (Materials Research Society, Boston, MA 1990), p E. L. Allen, M. D. Deal, and J. D. Plummer, MRS Symp. Proc. 163, 685 (1990). E. L. Allen, M D. Deal, and J. D. Plummer, J. Appl. Phys. 67, 3311 (1990). E L. Allen, J. J. Murray, M. D. Deal, J. D. Plummer, K. S. Jones, and W. S. Rubart, J. Electrochem. Sot. 138, 3440 ( 1991). 13Y. Arakawa, J. S. Smith, A. Yariv, N. Otsuka, C. Choi, B. P. Gu, and T. Venkatesan, Appl. Phys. Lett. 50, 92 (1987). 14T. Bachmann and H. Bartsch, Nucl. Instrum. Methods Phys. Res. B 43, 529 (1989). R. Gwilliam, R. S. Deal, R. Blunt, and B. J. Sealy, MRS Symp. Proc. 92, 437 (1987)..- 16P. Bellon, J. P. Chevalier, and G. Martin, in Microscopy of Semiconducting Materials, 1987, edited by A. G. Cullis and P. D. Augustus (IGP, London, 1987), pp M. G. Grimaldi, B. M. Paine, M. M;ienp%i, M.-A. Nicolet, and D. K. Sadana, Appl. Phys. Lett. 39, 70 (1981). *D. K. Sadana, T. Sands, and J. Washburn, Appl. Phys. Lett. 44, 623 (1984). 19M. G. Grimaldi, B. M. Paine, M.-A. Nicolet, and D. K. Sadana, J. Appl. Phys. 52,4038 (1981). soy. I. Nassim, L. A. Christel, T. W. Sigmon, J. F. Gibbons, T. J. Magee, and R. Ormond, Appl. Phys. Lett. 39, 598 (1981). D. K. Sadana, H. Choski, J. Washburn, P. F. Byrne, and N. W. Cheung, Appl. Phys. Lett. 44, 301 (1984). *C! Licoppe, Y. I. Nissim, and C. d Anterroches, Phys. Rev. B 37, 1287 (1988). 23A. K. Rai, R. S. Bhattacharya, and P. P. Pronko, J. Appl. Phys. 54, 2307 (1983). 24W. G. Opyd, J. F. Gibbons, J. C. Bravman, and M. A. Parker, Appl. Phys. Lett. 49, 974 (1986). =J. S. Williams, R. G. Elliman, S. T. Johnson, D. K. Sengupta, and J. M. Zemanski, MRS Symp. Proc. 144, 355 (1989). 26M A Shahid, R. Bensalem, B. J. Scaly, P. N. Favennec, and M. Gauneau, Nucl. Instrum. Methods Phys. Res. B 30, 531 (1988). E. Morita, J. Kasahara, and S. Kawado, Jpn. J. Appl. Phys. 24, 1274 (1985). * T. E. Shim, T. Itok, and Y. Yamamoto, J. Appl. Phys. 61,4635 (1987). 29C. P. Stewart, R. T. Blant, G. R. Booker, and I. R. Sanders, Physica B 116, 635 (1983). 3o W. S. Rubart, K. S. Jones, L. Seiberling, and D. K. Sadana, MRS Symp. Proc. 157, 677 (1990). K. S. Jones, S. Prussin, and E. R. Weber, J. Appl. Phys. 62, 4114 (1987). 32L. A. Christel and J. F. Gibbons, J. Appl. Phys. 52, 5050 (1981). j3k. G. Orrman-Rossiter, S. T. Johnson, and J. S. Williams, Nucl. Instrum. Methods Phys. Res. B 7/8, 448 (1985). 34S. J. Pearton, J. S. Williams, K. T. Short, S: T. Johnson, J. M. Gibson, D. C. Jacobson, J. M. Poate, and D. 0. Boerma, J. Appl. Phys. 65, 1089 (1989). K. H. Lee, D. A. Stevenson, and M. D. Deal, J. Appl. Phys. 68, 4008 (1990). 3bI. S. Tashlykov, G. Carter, and M. J. Nobes, Sov. Phys. Semicond. 20, 497 (1986). T. E. Haynes and 0. W. Holland, Appl. Phys. Lett. 58, 62 (1991). Jones et a/. 6795