Optical Materials 30 (2007) 238–243 www.elsevier.com/locate/optmat
Optical properties of nanocrystalline silicon ﬁlms deposited by plasma-enhanced chemical vapor deposition Atif Mossad Ali
Advanced Bio-Nano Devices Laboratory, Department of Bioengineering and Robotics, Tohoku University, 6-6-01 Aza-Aoba, Aramaki, Aoba-ku, Sendai 980-8579, Japan Received 28 June 2006; accepted 8 November 2006 Available online 12 January 2007
Abstract Nanocrystalline silicon (nc-Si) ﬁlms were deposited by a plasma-enhanced chemical vapor deposition technique using SiF4/SiH4/H2 gas mixtures. The optical properties of the nc-Si ﬁlms were examined by varying the deposition temperature (Td) under two diﬀerent hydrogen ﬂow rate ([H2]) conditions. For these ﬁlms, we found two photoluminescence (PL) spectra at around 1.7–1.75 eV and 2.2– 2.3 eV. The peak energy, EPL, of the 1.7–1.75-eV PL band was found to shift as Td or [H2] changes. It was found that the decrease in Td acts to decrease the average grain size, hdi, and to increase both the optical band gap, Eopt g , and the EPL values. By contrast, the increase in [H2] decreased the hdi value, while increased the values of Eopt g , and EPL. Thus, as either Td decreases or [H2] increases, it is found that a decrease in hdi corresponds well with increases in Eopt g and EPL. As a consequence, it was suggested that an increase of EPL of the 1.7–1.75-eV PL band can be connected with an increase in Eopt g , through a decrease in hdi. However, the PL process can not be connected with the transition between both the bands, related to formation of nanocrystals. Based on these results, it was proposed that the use of both low Td and high [H2] conditions would allow to grow nc-Si ﬁlms with small grains. Ó 2006 Elsevier B.V. All rights reserved. PACS: 78; 78.20.e; 78.66.w Keywords: Nanocrystalline silicon; PECVD; Deposition temperature; Photoluminescence; Quantum-size eﬀect; Eﬀect of hydrogen addition
1. Introduction Recently, silicon is the main material of microelectronics, but it is not widely used in optoelectronics. The reason is to due to bulk silicon has an indirect optical gap and low photoluminescence (PL) eﬃciency. Nanocrystalline silicon (nc-Si) have been intensively studied in the last few years, because nc-Si is expected to exhibit a quantum size eﬀect and therefore has a potential for application to optoelectronics. Several methods of creating PL thin ﬁlms containing nc-Si have been reported [1–4]. Although there is a *
Tel.: +81 22 795 6909; fax: +81 22 795 6907. E-mail addresses: [email protected]
, [email protected]
1 Permanent address: Department of Physics, Faculty of Science, Assiut University, Assiut 71516, Egypt. 0925-3467/$ - see front matter Ó 2006 Elsevier B.V. All rights reserved. doi:10.1016/j.optmat.2006.11.042
large amount of research in the recent years, the mechanism underlying the visible luminescence is still unclear. There are two mechanisms, ﬁrst one is the quantum conﬁnement model and the second is the surface model, widely discussed to know the origin of PL. Fabrication of nc-Si ﬁlms, having a highly eﬃcient quantum-size eﬀect, requires both a reduction in the grain size, d, and an enhancement in the crystalline volume fraction, q. Then, an increase in the nucleation rate would become a key technique to be developed. It has been reported that the enhanced nucleation on the growing surface during deposition of nc-Si can be achieved by increasing the hydrogen ﬂow rate, [H2] [5,6], or the deposition temperature, Td . However, the increase in Td also caused an increase in d. Thus, the increase in [H2] leads to reduction of d [5,6], while the increase in Td should lead to increase in d . Furthermore,
A.M. Ali / Optical Materials 30 (2007) 238–243
it has been reported that formation of nc-Si is due also to higher etching activity of hydrogen radicals for the amorphous phase than the crystalline one [8,9]. In the present work, we deposited nc-Si ﬁlms using a plasma-enhanced chemical vapor deposition (PECVD) method. The purpose of this work is to make clear the causes aﬀecting the optical properties, such as the PL spectra and the optical band gap, Eopt g , in connection with a change in d.
nent) and crystalline phases (520 cm1 component). The photoluminescence spectra were also measured using the above Raman system. The vibrational spectra were measured by a FTIR spectrometer (JASCO FT/IR-610) at a normal light incident. The optical transmission spectra were measured using an UV/VIS/NIR spectrophotometer (JASCO V-570).
Nc-Si ﬁlms were deposited on Corning 7059 glass, (1 0 0) Si and fused quartz substrates by PECVD using a SiF4/ SiH4/H2 gas mixture. For SiF4, we used a SiF4 gas diluted with helium (95% He + 5% SiF4). In this paper, we represent the net ﬂow rates of the SiF4 gas as [SiF4]. The substrates were cleaned for 20 min using nitrogen plasma and subsequently for 20 min using hydrogen plasma, just before deposition of nc-Si ﬁlms. The samples were deposited on 0.3-mm-thick glass substrates for measurements of X-ray diﬀraction (XRD) and Raman scattering, on 0.3-mm-thick n-type (1 0 0) Si substrates with high resistivity of 1000–3000 X cm for measurements of Fourier transform infrared (FTIR) absorption, and on 0.3-mm-thick fused quartz substrates for measurements of PL and optical absorption. The ﬁlms with two diﬀerent series of thickness values were prepared by adjusting the deposition time: The thickness of one series was ﬁxed at 0.37 (±0.03) lm for measuring the XRD, the Raman scattering, the PL, and the FTIR. In another series, we adopt a ﬁxed value of 0.73 (±0.03) lm for measuring the optical absorption. The deposition temperature, Td, was varied from 95 °C to 250 °C. Other deposition conditions are summarized in Table 1. The average value, hdi, of the grain size d was estimated from the spectral width of the XRD using Scherrer’s formula , which will reﬂect those in the depth direction of the ﬁlms. The Raman spectra were measured by a Raman spectrometer having a double monochrometer (Jobin Yvon RAMANOR HG 2S) coupled with a cooled photo-multiplier tube (Hamamatsu R649S) and having an Ar–ion laser light at 488 nm. The observed Raman spectra were divided into amorphous phase (480 cm1 compo-
Fig. 1 shows the peak frequency, ER, of the Raman signal arising from the crystalline Si (c-Si) phase for the nc-Si ﬁlms, as a function of Td. As shown in Fig. 1, with an increase in Td, the ER increases and approaches a value around 522 cm1, which was observed for single c-Si by using the same Raman system with that in the present work. In addition, the increase in [H2] is found to increase the ER toward 522 cm1. Then, a decrease of ER in the range lower than 522 cm1 will reﬂect a change in d. Fig. 2 shows the deposition rates, Rd, for nc-Si ﬁlms, as a function of the reciprocal values of Td. The Rd value increases with an increase in Td but decreases with an increase in [H2]. As revealed in Fig. 2, the relationship between Rd and Td can be expressed as follows: DE Rd ¼ A exp ; ð1Þ RT d
Feed gas Substrates [SiH4] [SiF4] [H2] Deposition temperature, (Td) Total gas pressure RF power (PRF)
SiF4/SiH4/H2 Glass (Corning 7059), Si (1 0 0) and fused quartz 0.60 sccm 0.38 sccm 30 or 46 sccm 95–250 °C 0.3 Torr 20 W
where A is a constant, DE the activation energy for ﬁlm growth, and R the gas constant. Based on the results shown in Fig. 2, we ﬁnd DE = 1.2 eV for [H2] = 30 sccm and DE = 3.3 eV for [H2] = 46 sccm. The Rd value can be controlled by a change in the self-diﬀusion coeﬃcient of Si atoms. The high self-diﬀusion coeﬃcient, which will increase with increasing Td under a given condition of the host-network, is expected to increase Rd. For self-diﬀusion in amorphous Si, the DE value has been reported to be
Peak Frequency, ER, (cm-1)
Table 1 Deposition parameters for growth of nc-Si ﬁlms using PECVD method
3. Results and discussion
Single c-Si 521 520 519 518 [H2] = 30 sccm
[H2] = 46 sccm 516 80
Deposition Temperature (oC) Fig. 1. Peak frequency, ER, of the Raman signal arising from crystalline Si phases for nc-Si ﬁlms, as a function of Td. Dashed line shows the value of ER for single c-Si. Solid lines are drawn as a visual guide.
A.M. Ali / Optical Materials 30 (2007) 238–243
3.3 eV 1.5
[H2] = 30 sccm [H2] = 46 sccm
1000/Td (K ) Fig. 2. Logarithmic deposition rate, ln Rd, for nc-Si ﬁlms, as a function of the reciprocal value of Td. The solid lines were drawn, using a method of the least square. The values shown on each solid line show the activation energy, DE, in ﬁlm growth, obtained using Eq. (1).
2.8 eV , which is close to that found for the ﬁlms with [H2] = 46 sccm. However, the value of DE = 1.2 eV, observed for [H2] = 30 sccm, appears to be too small for interpreting in terms of only a change in the self-diﬀusion coeﬃcient. Fig. 3 illustrates the IR absorption spectra over the range 400–4000 cm1, which were measured under vacuum for nc-Si ﬁlms with diﬀerent Td and [H2] values. These spectra were measured within 2 days after deposition. As seen in Fig. 3, the absorption bands were observed at around 650, 800–900 and 2100 cm1, which are assigned to the wagging, bending and stretching motions of Si–H bonds, respectively. For the SiH stretching band, we can ﬁnd three satellite lines at around 2090, 2100, and 2140 cm1. The intensity ratio of these three satellite lines is found to be independent of the deposition conditions of both [H2] and Td (=95 °C and 180 °C). Origins of these three lines are unknown, though the absorption band around
[H2] = 46 sccm, Td = 250 C
[H2] = 30 sccm, Td = 250o C
2100 cm1 is known to arise from dihydrides (Si–H2) . The intensity of the spectra around 2100 cm1 is likely to decreases with increasing Td and/or [H2]. These Si–H2 bonds should exist in the grain boundary regions. We have proposed that Si–H2 bonds would be occupied in microvoid-like boundaries formed in nc-Si ﬁlms, in a previous article . The PL spectra observed for (a) nc-ﬁlms with [H2] = 30 sccm and for (b) those with [H2] = 46 sccm under diﬀerent Td conditions, as a function of photon energy, are plotted in Fig. 4. These spectra were observed at room temperature. In addition, the hd(1 1 1)i and hd(1 1 0)i values were shown for comparison. As seen in Fig. 4, the PL spectra have two separated lines at the peak energy, EPL, of 1.7–1.75 eV and 2.2–2.3 eV. The observed EPL values are similar to those found by Zhao et al.  and Edelberg et al. . As demonstrated in previous works [13,15], the spectrum with EPL = 1.7–1.75 eV may be ascribed to nanometer size grains in nc-Si ﬁlms , and that with EPL = 2.2– 2.3 eV may be due to defects related to oxygen , or to SiH-related bonds . Fig. 5 shows (a) the optical band gap, Eopt g , and (b) EPL of the 1.7–1.75-eV band observed for nc-ﬁlms, as a function of Td. The Eopt were deterg values pﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ mined by drawing the Tauc plots, ðahmÞ versus ðhm Eopt g Þ, using the optical absorption coeﬃcient, a,
(δ <110>) = 18 nm
[H2] = 46 sccm, Td = 180 C
[H2] = 30 sccm, Td = 180o C
[H2] = 46 sccm, Td = 95 C
[H2] = 30 sccm, Td = 95o C
Td = 95 oC
14.3 nm (δ <111>) =
[H2 ] = 46 sccm
(δ <110>) = 17 nm
Td = 95 oC
18.2 nm 1.6
(Si-H) wag 4000
(δ <111>) = [H2 ] = 30 sccm
PL Intensity (a.u.)
Wavenumber (cm-1) Fig. 3. IR absorption spectra over the range 400–4000 cm1, which were measured under vacuum for nc-Si ﬁlms with diﬀerent Td and [H2] values.
250 oC 2.4
Photon Energy (eV) Fig. 4. Photoluminescence (PL) spectra for nc-Si ﬁlms with (a) [H2] = 30 sccm and (b) [H2] = 46 sccm under diﬀerent Td conditions, as a function of photon energy.
A.M. Ali / Optical Materials 30 (2007) 238–243 2.3
[H2] = 30 sccm
[ H2] = 46 sccm
E opt (eV)
Fig. 5. (a) Optical band gap, Eopt g , and (b) the peak energy, EPL, of the 1.7– 1.75-eV PL band observed for nc-Si ﬁlms, as a function of Td. Solid lines are drawn as a visual guide.
observed at photon energy of hm. In these experiments, we used 0.73 (±0.03)-lm-thick ﬁlms to eliminate an aﬀection of thinner thickness. As revealed in Fig. 5, an increase in EPL corresponds well with an increase in Eopt g with varying Td or [H2], though the rates in the increase of EPL is considerably smaller than that of Eopt g . This result suggest that the radiative recombination between excited electron and hole pair, may be caused by states other than those at both the band edges. This will be discussed at a later stage again. In Fig. 4, we also notice that the absolute values of EPL = 1.7–1.75 eV and EPL = 2.2–2.3 eV are higher than the band gap for c-Si (1.12 eV at RT), though the values of both hd(1 1 1)i and hd(1 1 0)i for almost ﬁlms are larger than 10 nm and 7 nm, respectively. Such large grains will have a characteristic similar to that of c-Si. As revealed in Figs. 4 and 5b, EPL of the 1.7–1.75-eV band increases with a decrease in Td, and appears to increase with increasing [H2]. On the other hand, the intensities of both the PL bands increase with a decrease in Td or an increase in [H2]. Thus, the dependence of EPL on Td or [H2] for the 1.7–1.75eV band is close to that of the PL intensity. This is because the oscillator strength of the 1.7–1.75-eV band may increase with a decrease in d, as shown theoretically by Stevens Kalceﬀ and Phillips , that is, small grains will cause a relatively large PL spectrum, under a ﬁxed density of grains. This supports an assignment that the 1.7–1.75-eV band may be ascribed to nanometer size grains in nc-Si ﬁlms as stated above. Indeed, as seen in Figs. 4 and 5, the decrease in hdi is found to correspond well with an increase in EPL of the 1.7–1.75-eV band, along with an
increase in Eopt g . Thus, the characteristics of the 1.7–1.75eV PL band appear to be closely connected with a change in hdi, in agreement with a model proposed by Edelberg et al. . This result implies a contribution of a quantum size eﬀect to the PL spectra. Furthermore, note that the surface of small grains should be deﬁned well, in order to obtain a PL emission from the small grains related to a quantum size eﬀect. However, the values of EPL are larger than those inferred from the hdi values, as stated above. Next, we will discuss the origins causing such large EPL. The intensity of the emitted PL light should be obtained as the product of the number of grains and their respective oscillator strength values. If the PL process is controlled by small grains, the oscillator strength would increase with a decrease in d as stated above. On the other hand, grains with diﬀerent d values will be distributed in the ﬁlms. Then, the size of grains eﬀective in the PL process should be smaller than the values of the average in d. Furthermore, in a previous article , we assumed a result that grains with smaller d values preferentially may exist in shallower layer of the ﬁlms, then the above-mentioned relationship between hdi and EPL values is easy to be explained. The hdi values in Fig. 4 were estimated from the width values of the XRD spectra (Section 2). In the present work, since ˚ is used, the an X-ray light with the wavelength of 1.54 A penetration depth of the X-ray for Si is approximately 13 lm. On the other hand, the XRD signal from grains smaller than 3 nm can not be detected because of broadening of the spectrum. Therefore, the hdi values in Fig. 4 will give an average size of grains larger than 3 nm over the entire positions in 0.37-lm-thick ﬁlms used (Section 2). This may also be one of reasons causing the observed values of hdi larger than those inferred from the EPL values as stated above. On the other hand, the penetration depth of a 488-nm Ar ion laser light, used for both the PL measurements and the Raman scattering, was approximately 0.5 lm. 3.2. Band gap based on simple theory As discussed in Section 3.1, the values of d contributing to the EPL and Eopt values are larger than the hdi values g obtained through the XRD measurements, because the former will become eﬀective in smaller grains than those for the latter. In this section, we will discuss the band gap estimated using the shifts of the Raman spectra that will reﬂect the characteristics of the whole grains with diﬀerent size as well as the PL and the optical absorption measurements. As shown in Fig. 1, the Raman peak arising from crystalline phases shifts toward a low frequency side with decreasing Td. Supposing that the peak shift is due only to the conﬁnement of optical phonons in spherical nanocrystals, we can estimate the crystallite size in diameter, DR, as : rﬃﬃﬃﬃﬃﬃ B DR ¼ 2p ; ð2Þ Dm
A.M. Ali / Optical Materials 30 (2007) 238–243
where B is 2.24 cm1 nm2, and Dm the frequency shift in unit of cm1, which was deﬁned as the diﬀerence between the observed peak-frequency value and 522 cm1. The latter value was observed for single crystal Si as described in Section 3.1. Fig. 6 shows a relationship between hd(1 1 1)i and hd(1 1 0)i, and DR. When we compared the results obtained under a given crystal direction and a given [H2], we can ﬁnd a close correlation between the hdi and DR values. However, it is found that the absolute values of hdi observed are considerably larger than DR values and the rate in the increase of hdi are faster than that of DR, Furthermore, based on the results shown in Fig. 6, we ﬁnd a relationship of hdi = 3.69DR 7.28 (nm) for the ﬁlms with [H2] = 30 sccm and of hdi = 3.56DR 11.89 for the ﬁlms with [H2] = 46 sccm, in the measurements under a direction of the h1 1 0i axis that is the dominant texture in the ﬁlms. On the other hand, for the h1 1 1i texture, we ﬁnd a relationship of hdi = 2.61DR + 4.48 for [H2] = 30 sccm and hdi = 2.64DR + 0.05 for [H2] = 46 sccm. These formulas were obtained by ﬁtting the values of hdi vs. DR to a linear relationship, using a method of the least squares. As seen in these results, the linear relationships of hdi as a function of DR appear to be characterized by the crystal axis of grains, that is, the slope (3.63 ± 0.07) for the h1 1 0i texture is steeper than that (2.63 ± 0.02) for the h1 1 1i texture. In addition, as shown in Figs. 4 and 6, the sizes of the h1 1 0i grains are smaller than that of the h1 1 1i grains. As described in Section 3.1, the size of grains eﬀective the XRD spectra, whose width determines the hdi values, would be larger than those inferred from the shifts of the Raman spectra, which reﬂect the characteristics of the whole grains with diﬀerent d values. This is because grains smaller than 3 nm would be insensitive to the XRD spectra as stated above. This result may also explain the observation of a large intercept in the linear relationship of hdi vs. DR, as seen in Fig. 6.
Using the values of DR for the individual samples, we can evaluate the lowest excitation energy, E, under a simple conﬁnement theory for electron and hole [4,19,20], as follows: E ¼ Eg þ
2p2 h2 3:572 e2 þ 0:284ERy ; er D R mr D2R
where Eg is the energy gap of c-Si (1.12 eV at RT), mr is the reduced eﬀective mass of an electron-hole pair, er is the dielectric constant, and ERy is the Rydberg energy for the bulk semiconductor. The value of E correspond to the band gap of the ﬁlms. In the later two terms, 3.572e2/erDR corresponds to the coulomb term and 0.284ERy gives the spatial correlation energy. The later two terms are minor corrections, so we neglected them in the calculation used in this work, because the contribution of these two terms to the total energy will be less than 5% . Fig. 7 shows the E values (a solid curve) obtained based on Eq. (3), as a function of R. In Fig. 7, the experimental values of Eopt g (closed symbols) and EPL (open symbols) shown in Fig. 5a and b, respectively, are also shown for comparison, as a function of R (R = DR/2 is the radius of crystals), through the values of DR obtained using the experimental Dm values along with Eq. (2). As shown in Fig. 7, we can ﬁnd a qualitative agreement between the observed Eopt values (closed triangles and g closed circles) and a solid curve calculated using Eq. (3), though the former values are considerably larger than the latter. Furukawa and Miyasato  have found also similar
E, E g or EPL (eV)
<111>, [H2] = 30 sccm
<111>, [H2] = 46 sccm <110>, [H2] = 46 sccm
Average Grain Size (nm)
<110>, [H2] = 30 sccm 21
Eq. (3) opt
E g at [H2] = 30 sccm
E opt at [H2] = 46 sccm g
EPL at [H2] = 30 sccm
EPL at [H2] = 46 sccm
9 0.0 0
R (nm) 0
DR (nm) Fig. 6. Relationship between the average grain size, hd(1 1 1)i and hd(1 1 0)i, as a function of the diameter of grains, DR, calculated using Eq. (2). The solid lines were drawn, using a method of the least square.
Fig. 7. Lowest excitation energy, E, as a function of radius of crystals R (a solid curve), obtained based on Eq. (3). In this diagram, the experimental values of Eopt values (closed symbols) and EPL (open g symbols) values, which were shown in Figs. 5a and 5b, respectively, are also shown for comparison, as a function of R (=DR/2) through the DR values obtained using the experimental Dm values along with Eq. (2).
A.M. Ali / Optical Materials 30 (2007) 238–243
discrepancy between the theoretical and experimental results, and interpreted the discrepancy in terms of a diﬀerence in the surface shape of grains as boundary conditions in both the theoretical and experimental process. On the other hand, the change of EPL as a function of R is considerably smaller than those of E and Eopt g though the trend of the changes for EPL agreed with that for Eopt as shown in g Fig. 5. This result indicates that the PL process of the 1.7–1.75-eV band can not be connected with the transition between both the band edges, related to formation of nanocrystals. 4. Conclusions We prepared nc-Si ﬁlms by a PECVD method using SiF4/SiH4/H2 gas mixtures. The optical properties of the nc-Si ﬁlms were examined by decreasing Td from 250 to 95°C under [H2] = 30 and 46 sccm. We found two PL spectra at around 1.7–1.75 eV and 2.2–2.3 eV. The peak energy, EPL, of the 1.7–1.75-eV PL band was found to shift as Td or [H2] changes. It was found that the increase in Td causes the increases in hdi, and the decreases in both the Eopt g and EPL. By contrast, the increase in [H2] decreases the hdi, while increases the Eopt and EPL. Thus, as Td decreases or g [H2] increases, it was suggested that EPL increases with an increase in Eopt g , in good correspondence to a decrease in hdi. However, the observed hdi values were found to be considerably larger than the values of d inferred from the EPL and Eopt values, and some models for interpreting g the causes were proposed. Based on these results, it was suggested that the use of both low Td and high [H2] conditions would allow us the preparation of nc-Si ﬁlms with small grains.
Acknowledgement The author is pleased to thanks Professor S. Hasegawa for using his laboratory to do this work. References  J.H. Kim, K.A. Jeon, G.H. Kim, S.Y. Lee, Opt. Mater. 27 (2005) 991.  S. Veprek, T. Wirschem, M. Ruckschloß, H. Tamura, J. Oswald, Mater. Res. Soc. Symp. Proc. 358 (1995) 99.  G. Cicala, G. Bruno, P. Capezzuto, L. Schiavulli, V. Capozzi, G. Perna, Mater. Res. Soc. Symp. Proc. 452 (1997) 809.  E. Edelberg, S. Bergh, R. Naone, M. Hall, E.S. Aydil, J. Appl. Phys. 81 (1997) 2410.  D. Milovzorov, T. Inokuma, Y. Kurata, S. Hasegawa, J. Electrochem. Soc. 145 (1998) 3615.  S.K. Kim, K.C. Park, J. Jang, J. Appl. Phys. 77 (1995) 5115.  H.J. Lim, B.Y. Ryu, J.I. Ryu, J. Jang, Thin Solid Films 289 (1996) 227.  I. Solomon, B. Drevillon, H. Shirai, N. Layadi, J. Non-Cryst. Solids 164–166 (1993) 98.  S. Oda, M. Odobe, Mater. Res. Soc. Symp. Proc. 358 (1995) 721.  B.D. Cullity, Elements of X-Ray Diﬀfraction, Addison-Wesley, Reading, 1978, p. 102.  R.B. Iverson, R. Reif, J. Appl. Phys. 62 (1987) 1675.  D.V. Tsu, G. Lucovsky, B.N. Davidson, Phys. Rev. B 40 (1989) 1795.  A.M. Ali, T. Inokuma, Y. Kurata, S. Hasegawa, Jpn. J. Appl. Phys. 38 (1999) 6047.  X. Zhao, O. Schoenfeld, Y. Aoyagi, T. Sugano, Appl. Phys. Lett. 65 (1994) 1290.  A.M. Ali, T. Inokuma, Y. Kurata, S. Hasegawa, Mater. Sci. Eng. C 15 (2001) 125.  A.J. Kenyon, P.F. Trwoga, C.W. Pitt, J. Appl. Phys. 79 (1996) 9291.  X. Zhao, O. Schoenfeld, J. Kusano, Y. Aoyagi, T. Sugano, Jpn. J. Appl. Phys. 33 (1994) L649.  M.A. Stevens Kalceﬀ, M.R. Phillips, Phys. Rev. B 52 (1995) 3122.  Al.L. Efros, A.L. Efros, Sov. Phys. Semicond. 16 (1982) 772.  Y. Kayanuma, Phys. Rev. B 38 (1988) 9797.  S. Furukawa, T. Miyasato, Phys. Rev. B 38 (1988) 5726.