Low energy cathodoluminescence spectroscopy of SiO2 nanoparticles

Low-energy cathodoluminescence spectroscopy of erbium-doped gallium nitride surfaces T. M. Levina) and A. P. Young Department of Electrical Engineering, The Ohio State University, Columbus, Ohio 43210

J. Scha¨fer Center for Materials Research, The Ohio State University, Columbus, Ohio 43210

L. J. Brillson Department of Electrical Engineering and Center for Materials Research, The Ohio State University, Columbus, Ohio 43210

J. D. MacKenzie and C. R. Abernathy Department of Materials Science and Engineering, University of Florida, Gainesville, Florida 32611

共Received 16 March 1999; accepted 6 July 1999兲 We have used cathodoluminescence spectroscopy with variable incident beam energies to study the energy levels and activation of Er impurities in GaN as a function of depth below the free surface. The GaN films were doped in situ during either metalorganic molecular-beam epitaxy 共MOMBE兲 or molecular-beam epitaxy 共MBE兲. Besides the well-known Er3⫹ luminescence at 0.80 eV, we observe emissions at 1.2, 1.8, 2.2, and 2.3 eV, corresponding to higher energy Er 4f shell transitions. For unannealed MOMBE-grown GaN:Er, these higher energy emissions appear only for excitation depths of hundreds of nanometers. The MOMBE-grown GaN;Er annealed to 500 °C shows a dramatic increase in the 1.8, 2.2, and 2.3 eV peak intensities at shallow probe depths, with its yield increasing with increasing depth. These three features become pronounced at all depths after a 700 °C anneal. MBE-grown GaN:Er grown with lower C and O impurity levels than the MOMBE-grown sample exhibits strong emission at all these energies without annealing. The decreased emission at shallow 共tens of nanometer兲 probe depths suggests a depletion of activation Er in the near-surface region. Enhancement of near-surface Er3⫹ luminescence with annealing may be due to lattice reordering as well as impurity redistribution. © 1999 American Vacuum Society. 关S0734-2101共99兲00106-2兴

I. INTRODUCTION Rare earth-doped semiconductors have become a promising material system for integrated optoelectronic applications. Erbium 共Er兲 doping in a wide class of semiconductors hosts has been studied because of the intrashell erbium transition that occurs at 0.80 eV, coinciding with the loss minimum in silica-based optical fibers.1 Wide band gap hosts for Er impurities have the increased benefit of reducing the thermal quenching of the Er related luminescence as the temperature is increased.2 In most cases, the Er doping has been accomplished by ion implantation with a significant amount of work being However, done with oxygen co-implantation.3–8 implantation-induced damage of the GaN lattice could be a source of defects, degrading device performance. Previous work has focused almost exclusively on the 0.80 eV 共1540 nm兲 line associated with Er3⫹ due to its significance for optical fiber communications. At least four different chemical environments for implanted Er within GaN have been identified using photoluminescence 共PL兲 and photoluminescence excitation 共PLE兲 spectroscopies that focus on the 0.80 eV emission.7 It has been shown that co-doping of GaN, especially with fluorine or oxygen improves the luminescence a兲

Electronic mail: [email protected]

3437

J. Vac. Sci. Technol. A 17„6…, Nov/Dec 1999

efficiency of this transition markedly.3 These findings underscore the importance of the local chemical environment in the vicinity of the Er impurities for understanding and producing the 0.80 eV emission. Since the large mass of Er limits the depth and distribution of ion implanted Er doping profiles and contributes to damage in GaN films, it would be advantageous to incorporate Er into GaN films in a way that avoids these limitations. J. D. MacKenzie and C. R. Abernathy have demonstrated in situ doping of III–N semiconductors grown by molecularbeam epitaxy 共MBE兲 and metalorganic molecular-beam epitaxy 共MOMBE兲 with Er.9,10 To further study the luminescence of Er in GaN, we have performed low-energy cathodoluminescence spectroscopy 共CLS兲 in ultrahigh vacuum 共UHV兲 on samples of GaN:Er grown by MBE and MOMBE. Spectral emissions covering a broad spectra range were measured to search for other Er 4f emissions. In concert with the low energy CLS experiments, we have performed annealing studies to probe the processing conditions necessary to improve the sensitization of the Er3⫹ emissions. We report observations of increased sensitization of Er3⫹ related emissions with annealing temperatures between 500 and 700 °C in a sample-grown MOMBE. In contrast, a sample grown by MBE exhibited pronounced Er3⫹ intensity even without annealing. These results indicate that the local

0734-2101/99/17„6…/3437/6/$15.00

©1999 American Vacuum Society

3437

3438

Levin et al.: Low-energy cathodoluminescence spectroscopy

3438

chemical environment, determined by the growth technique and post-growth processing, plays a major role in Er3⫹ activation in GaN. II. EXPERIMENT Erbium-doped GaN layers 5000-Å-thick with Er concentration of 3⫻1018 cm⫺3 were grown at the University of Florida-Gainesville, by both MBE and MOMBE. The MOMBE-grown sample consisted of a 共0001兲 sapphire substrate, followed by an AlN buffer deposited at 435 °C with dimethyl ethyl alane amine as an aluminum precursor, followed by 2000 Å of unintentionally doped GaN with triethyl gallium as a Ga precursor. The 5000 Å Er doped film was deposited at 925 °C. The MBE GaN:Er layers were grown using an elemental Ga source and an electron cyclotron N2 source. The different Ga sources employed resulted in films having carbon and oxygen concentrations that differed by over two orders of magnitude, with C and O concentrations ⬍1⫻1019 cm⫺3 for the MBE-grown films and ⬇1 ⫻1020 cm⫺3 for the MOMBE-grown films. The C and O impurity concentrations were measured by secondary ion mass spectrometry. The samples were shipped to The Ohio State University under ambient conditions. Low-energy CLS was performed in UHV at Ohio State using electron beam energies between 0.6 and 4.5 keV. The beam was incident at 45° with respect to the surface normal of the sample. The beam spot size could be observed visually and varied between less than 1 mm to approximately 2 mm in diameter. The beam spot size varied with primary beam energy, with the largest spot sizes observed 0.5 keV. Previous electron beam induced current measurements indicate that the currents were on the order of 1 ␮A.11 The luminescence signal was collected by a CaF2 lens placed inside the UHV chamber, and directed through a sapphire window, and on to the entrance slit of a 0.25 m Leiss single pass flint glass prism monochromator. A LN2 cooled North Coast Ge diode detector was used for spectral measurements between 0.7 and 2.0 eV 共1700 nm⬍␭⬍620 nm兲, while a S-20 photomultiplier was used to cover the range from 1.8 to 3.75 eV 共680 nm⬍␭ ⬍330 nm兲. The detector output was measured with a lock-in amplifier. The response of the data collection system was measured using an Oriel calibrated lamp as a spectral source. The system response was deconvolved from the data taken with the S-20 photomultiplier, while data taken with the Ge photodiode were corrected only for the gain of the lock-in amplifier. The resolution of the monochromator is estimated to be ⫾15 meV for all photon energies detected. III. RESULTS Figure 1 illustrates low-energy CLS spectra for different incident beam voltages at the surface of the MOMBE-grown, air-exposed GaN:Er/Al2O3 sample. With increasing voltage, the electron beam excites free electron–hole pairs at increasing depths. For the range from 0.6 to 4.5 keV, the Everhart– Hoff relationship12 extended to low energies13 yields maximum penetration depths from 10 to 140 nm, respectively. J. Vac. Sci. Technol. A, Vol. 17, No. 6, Nov/Dec 1999

FIG. 1. CLS spectra taken at different incident electron beam energies of the MOMBE-grown GaN:Er sample after exposure to ambient conditions. The Bohr–Bethe range, R B as calculated from Ref. 13 is listed for each electron beam energy. The NBE transition dominates all spectra. The intrashell Er 4 f transitions are evident only in the spectra taken at 4.5 keV.

The corresponding depths of maximum electron–hole pair creation are a fraction 共roughly 1/3兲 of these Bohr–Bethe ranges. All of the spectra in Fig. 1 are dominated by the nearband-edge 共NBE兲 peak of GaN, which appears at 3.45 eV at room temperature. This peak has a full width at half maximum of 0.3. With an increasing electron beam energy and a depth below the free surface, the NBE emission intensity increases by approximately an order of magnitude. The major portion of this increase occurs for energies between 1.0 and 1.5 keV, corresponding to excitation depths of 20–30 nm. At the highest excitation energy of 4.5 keV, additional features with relatively low intensities are apparent at subband gap energies of 1.8, 2.2, and 2.3 eV. Figure 2 provides CLS spectra of air-exposed, MOMBEgrown GaN:Er/Al2O3 for similar electron beam energies and emission energies in the infrared region. Here the spectra are dominated by the well-known Er 4f emission line at 0.80 eV with a spectral line shape determined by the optical resolution of our monochromator. An additional feature at 1.21 eV appears for all but the lowest excitation depths. With an increasing beam energy and depth, the intensities of the overall spectra increase by a factor of 20 between 0.75 and 3 keV, comparable to the changes observed in the visible and nearultraviolet spectra. Figure 3 shows the set of CLS spectra for the same sample described previously after an annealing at 500 °C for 5 min in UHV. The temperature was measured by an optical

3439

Levin et al.: Low-energy cathodoluminescence spectroscopy

FIG. 2. CLS spectra in the infrared region of GaN:Er grown by MOMBE.

FIG. 3. CLS spectra of MOMBE-grown GaN:Er after a 500 °C anneal in UHV. The ER 4f transitions are slightly sensitized, but the NBE emission remains the dominant feature in the spectra. JVST A - Vacuum, Surfaces, and Films

3439

FIG. 4. CLS spectra of MOMBE-grown GaN:Er after a 700 °C anneal for 5 min in UHV. The Er 4f peaks now dominate the spectra in contrast to the spectra shown in Figs. 1 and 3.

pyrometer through a sapphire window. Similar to the unannealed case, the spectra tend to be dominated by the broad asymmetric NBE peak centered at 3.45 eV. However, the dominance of the NBE peak at the lowest 共0.6 keV兲 and highest 共4.5 keV兲 bean energies is noticeably absent. The low intensity peaks at 1.8, 2.2, and 2.3 eV are detected beginning with the spectrum measured at 1.0 keV, corresponding to a depth of roughly 20 nm, and are evident in all of the higher beam energy spectra. However, the overall intensity drops in the 3.0 and 4.5 keV spectra compared with the 2.0 keV spectrum. The infrared spectra taken after the 500 °C anneal 共not shown兲 are very similar to the spectra shown in Fig. 2 which were taken prior to the anneal. All infrared spectra are dominated by the Er 4f at 0.80 eV peak and show increasing intensity with an increasing beam energy. Figure 4 shows the CLS spectra taken after annealing at 700 °C for 5 min. In contrast to the spectra taken prior to and subsequent to the 500 °C anneal, the three subband gap peaks are prominent at all energies and penetration depths. Additionally, these spectra show commensurate intensity maxima for both the NBE peak and the set of three subband gap peaks in four of the five spectra. The NBE peak shape retains the line shape seen previously. The intensity of the three subband gap peaks vary independently of one another. The 0.1–0.2 eV linewidth of these subband gap peaks remain constant with beam energy and beam penetration depth. The integrated intensity of the spectra increases with increasing beam energy up to 2.0 keV. Then, the integrated intensity

3440

Levin et al.: Low-energy cathodoluminescence spectroscopy

FIG. 5. CLS spectra in the infrared region for the MOMBE-grown GaN:Er sample after a 700 °C anneal for 5 min. The 0.75 eV emission intensity increases with an increasing excitation energy and penetration depth.

decreases by a factor of 4 as the beam energy and penetration depth are increased through 4.5 keV. The infrared spectra taken after the 700 °C anneal are shown in Fig. 5. The 0.80 eV peak intensity increases with an increasing beam energy and penetration depth, with the peak in the 4.5 keV spectrum a factor of 30 times larger than the peak in the 0.6 keV spectrum. The peak at 1.21 eV is evident in all but the two spectra taken at the lowest beam voltages. Figure 6 illustrates the CLS spectra from the air exposed GaN:Er/Al2O3 sample grown by MBE, which has a lower C and O concentrations than the sample considered in Figs. 1 through 5. In contrast to the previous sample, these spectra have a qualitatively consistent shape, with the three peaks at 1.86, 2.23, and 2.32 eV present at all electron beam voltages, including those at 1.0 keV and below which constitute a more surface sensitive probe. The integrated intensity of the spectra increases by an order of magnitude with an increasing beam voltage and depth below the free surface. Additionally, the relative peak heights of the three peak heights remains rather consistent, with the 1.86 eV peak having 2/3 of the intensity of the 2.23 and 2.32 eV peaks. The 2.23 and 2.32 eV peaks heights are within 10% of one another. A striking feature of all of these spectra is the absence of the large NBE peak seen in the high C and O sample. The peak is noticeable in the three spectra taken at 3.0 keV. However, J. Vac. Sci. Technol. A, Vol. 17, No. 6, Nov/Dec 1999

3440

FIG. 6. CLS spectra of an air-exposed GaN:Er sample grown by MBE. Er 4f emissions at 1.8, 2.2, and 2.3 eV dominate the spectra, and are present without annealing.

it cannot be distinguished above the noise in all other spectra. The infrared portion CLS spectrum for the MBE-grown sample is shown in Fig. 7. The Er 4f peak appears at 0.80 eV and dominates the response. The emission at 1.23 eV is seen earlier and is present at all beam voltages and depths. The integrated intensity increases by a factor of 4–5 as the electron beam probes to greater depths within the sample. IV. DISCUSSION We observe multiplet Er-related peaks in the GaN:Er spectra by CLS. The peak energy and the comparative sharpness of these peaks with respect to a defect level emission indicates that these peaks can be attributed to intrashell Er 4f transitions. The 0.80 eV peak corresponds to the 4 I 13/2 → 4 I 15/2 transition. The peak at 1.2 eV is assigned to the 4 I 11/2→ 4 I 15/2 transition. Kim et al.8 have observed this transition in Er implanted GaN films as well as the 4 I 9/2 → 4 I 15/2 transition which occurs at 1.532 eV using PLE. In addition, we observe three other Er 4f peaks occurring at 1.8, 2.2, and 2.3 eV that are assigned, based on their energy, to the 4 F 9/2→ 4 I 15/2 , 4 S 3/2→ 4 I 15/2 , and 4 H 11/2→ 4 I 15/2 transitions, respectively. The transition at 1.5 eV observed by Kim et al. is not apparent in our spectra. The affect of annealing can be seen by comparing the spectra shown in Figs. 2, 3, and 4. These figures illustrate the spectra for the MOMBE sample without annealing, after the 500 °C anneal, and after the 700 °C anneal, respectively, for

3441

Levin et al.: Low-energy cathodoluminescence spectroscopy

photon energies between 1.5 and 3.75 eV. The intensity of the Er 4f peaks at 1.8, 2.2, and 2.3 eV are seen to increase dramatically relative to the NBE peak significantly after the 700 °C anneal. Without annealing, the Er 4f peaks at 1.8, 2.2, and 2.3 eV are evident only in the 4.0 keV spectrum. However after the 700 °C anneal, the intensity of the NBE peak and the Er 4f triplet peaks are comparable. Thus, the anneal serves to activate these transitions with an activation threshold of 500–700 °C. In contrast, we observe the 0.80 and 1.2 eV transitions without annealing. There are no significant differences between the integrated CL intensities in the spectra taken with the Ge detector at the same incident beam energy and voltage among the air-exposed, 500 and 700 °C spectra. This result is surprising, since the annealing temperatures used here are lower than growth temperature of 925 °C. Any lattice healing, Er-impurity or Er-related complexes that would form should have formed during growth. However, it is clear from the spectra of the higher energy transitions that annealing sensitizes these peaks. The effect of the surface can be studied by examining the intensity of the luminescence with an incident beam voltage and probe depth. Typically, we see that the integrated luminescence decreases by at least an order of magnitude as the beam energy is decreased and the probe depth approaches the surface. This reduction in intensity begins at depths of 200–500 Å, and further reductions in the integrated intensity continue as the probe depth is reduced. The surface can be viewed as an extended two-dimensional defect, which disrupts the periodicity of the crystal lattice. It is well known

FIG. 7. CLS spectra in the infrared region of air-exposed GaN:Er grown by MBE. JVST A - Vacuum, Surfaces, and Films

3441

that the large number of defects associated with the surface increase both the density and cross section of nonradiative recombination paths. In addition, the air-exposed samples were not given any special chemical treatment aside from the 500 and 700 °C anneals performed on the high C and O sample grown by MOMBE. Although no low-energy electron diffraction experiments were performed, it is reasonable to assume that the surface was not clean and ordered when spectra were taken. The presence of lattice disorder or impurity adatoms may lead to competing recombination paths near the surface, decreasing the luminescence. Typically, the relative size of the GaN NBE peak and the Er 4f peaks observed with the S-20 photomultiplier tube 共1.5–3.8 eV兲 stays fairly constant with incident beam energy. In the infrared region, the 0.80 eV Er peak dominates the spectrum, without any other significant peaks with which to compare. However, the change in the intensity of the 0.80 eV peak does not vary with incident beam energy in a very different way than the other Er 4f peaks or the NBE peak. Thus, the reduction in intensity with a decreasing beam energy can be understood as the effect of competing recombination paths due to the presence of the surface, affecting each of the significant features in a similar fashion. The effect of impurities and co-dopants has been shown to be crucial in improving the luminescence efficiency in GaN:Er. Many researchers have suggested that the excitation of the 0.80 eV line in GaN:Er is mediated by impurities or impurity-related complexes, the most popular of these ideas involving the presence of oxygen impurities. Kim et al.8 show that the photoluminescence spectrum near 0.80 eV measured using above gap 共325 nm兲 pumped GaN:Er can be reconstructed from the sum of two PL spectra obtained with 633 and 458 nm light. From this observation, they deduce that above gap excitation of the 0.80 eV Er 4f peak proceeds predominantly via a trap mediated process, rather than resonant pumping of one of the Er 4f excited states. Above gap excitation by PL or CL, for example, is expected to mimic free carrier recombination in a junction device and therefore provide insight into the mechanisms of recombination and their associated efficiencies. Our data shows that the presence of high C and O concentrations does not necessarily lead to higher sensitization of all Er 4f intrashell transitions within GaN. The MBEgrown sample, having more than three O impurity atoms for every Er impurity requires no annealing or extra processing to sensitize the Er 4f emissions with peaks observed at 1.8, 2.2, and 2.3 eV. In contrast, the MOMBE-grown sample, having more than 30 O impurities for every Er impurity required annealing to a temperature between 500 and 700 °C to sensitize the more energetic Er 4f transitions. These results indicate that the specific chemical environment surrounding the Er atoms is different in the two samples. Torvik et al.3 determined that the luminescence from GaN films grown by halogen chemical vapor deposition and co-implanted with Er and O, required an O:Er concentration ratio of 5:1 before the integrated PL intensity from the 0.80 eV peak saturated. Our results therefore indicate that

3442

Levin et al.: Low-energy cathodoluminescence spectroscopy

the presence of excess O above a certain level or the level of incorporation of O in the GaN matrix may reduce the Er luminescence sensitivity near the surface. Overall, the increase in emission with annealing of the high C and O GaN peaks at 1.8, 2.2, and 2.3 eV indicates a change in the local Er environment near the surface. We see further evidence for differences in the nearsurface local chemical environments by the relative behavior of the emission seen at 0.80 and 1.2 eV. The 0.80 eV peak is present at all beam energies and probe depths in the MOMBE-grown sample, whereas the 1.2 eV emission appears at 1.0 keV in the air-exposed spectra, at 2.0 keV in the 500 °C annealed spectra, and at 2.0 keV for the 700 °C spectra for the MOMBE-grown sample. In contrast, the emission at 1.2 eV peak observed in the MBE-grown sample appears at all incident beam energies, including the spectrum taken at 0.5 keV. Furthermore, the ratio of the intensity of the ⬇1.2 keV peak to the 0.80 eV peak is greater in the MBE sample than in the MOMBE sample. Annealing to 700 °C sensitized the emissions at 1.8, 2.2, and 2.3 eV even at the lowest depths in the MOMBE-grown sample. We do not observe similar behavior with the 1.2 eV peak. This difference in behavior with annealing indicates that the chemical environment in the sample is changing with depth in the MOMBEgrown material. Lattice healing or a redistribution of O in the lattice may cause these changes. However, at annealing temperature used in this study, significant redistribution of O is unlikely.14 Nevertheless, the differences in the precursors for MOMBE and MBE growth that give rise to the differences in C and O impurity concentrations also may lead to variations in the concentration of hydrogen in the lattice. The presence of significant concentrations of hydrogen may lead to passivation of ionized impurities or defects that are necessary for the trap-mediated excitation of the Er 4f levels. At temperatures of 500–700 °C, the hydrogen passivation may be eliminated or reduced, thus leading to the increased intensity of the higher energy Er 4f transitions observed in the MOMBE-grown sample after annealing. Pearton, Abernathy, and Ren15,16 have shown that annealing temperatures of 450 °C are sufficient to remove H passivation effects in GaN films. In contrast, we see a consistent line shape with beam energy and probe depth in the MBE-grown sample. Taken together, these results indicate differences in the local bonding environments between the two samples with depth below the free surface. V. CONCLUSION We have observed multiple Er 4f emissions in the nearsurface region in in situ doped GaN grown by both MOMBE

J. Vac. Sci. Technol. A, Vol. 17, No. 6, Nov/Dec 1999

3442

and MBE using CLS in UHV. The depth dependence of the spectra showing increased intensities with increasing probe depth is explained by the effect of surface disorder. The 4 I 13/2→ 4 I 15/2 transition occurring at 0.80 eV and 4 I 11/2 → 4 I 15/2 transition occurring at 1.2 eV are present in both samples without annealing. In contrast, the transitions observed at 1.8, 2.2, and 2.3 eV are observed in the MOMBE, grown sample only after annealing to 500–700 °C, while these transitions are prominent without annealing in the MBE-grown sample. These differences with annealing and with growth method are indicative of the differences in the local chemical environments surrounding the ER dopants and in particular the role that oxygen and hydrogen impurities play in the sensitization of the emissions. The depth dependence of the 1.2, 1.8, 2.2, and 2.3 eV emissions indicates that there is a change in the Er bonding environments in the MOMBE-grown sample, whereas the MBE-grown sample shows more uniformity with depth below the free surface. ACKNOWLEDGMENTS This work was supported in part by the Department of Energy for the low-energy CLS 共Grant No. DEFG0297ER45666兲 and by the National Science Foundation for the nanoscale depth analysis 共Grant No. DMR-9711851兲. 1

P. N. Favennec, H. L. Haridon, D. Moutonnet, and M. Salvi, Electron. Lett. 26, 1556 共1990兲. 2 P. N. Favennec, H. L. Haridon, M. Salvi, D. Moutonnet, and Y. L. Cuillo, Electron. Lett. 25, 718 共1990兲. 3 J. T. Torvik, C. H. Qiu, R. J. Feuerstein, J. I. Pankove, and F. Namavar, J. Appl. Phys. 81, 6343 共1997兲. 4 C. H. Qiu, M. W. Leksono, J. I. Pankove, J. T. Torvik, R. J. Feuerstein, and F. Namavar, Appl. Phys. Lett. 66, 562 共1995兲. 5 J. T. Torvik, R. J. Feuerstein, J. I. Pankove, C. H. Qiu, and F. Namavar, Appl. Phys. Lett. 69, 2098 共1996兲. 6 M. Thaik, U. Ho¨mmerich, R. N. Schwartz, R. G. Wilson, and J. M. Zavada, Appl. Phys. Lett. 71, 2641 共1997兲. 7 S. Kim, S. J. Rhee, X. Li, J. J. Coleman, S. G. Bishop, and P. B. Klein, J. Electron. Mater. 27, 246 共1998兲. 8 S. Sim, S. J. Rhee, D. A. Turnbull, X. Li, J. J. Coleman, S. G. Bishop, and P. B. Klein, Appl. Phys. Lett. 71, 2662 共1997兲. 9 C. R. Abernathy, J. D. MacKenzie, and S. M. Donovan, J. Cryst. Growth 178, 74 共1997兲. 10 J. D. MacKenzie, C. R. Abernathy, S. J. Pearton, U. Ho¨mmerich, X. Wu, R. N. Schwartz, R. G. Wilson, and J. M. Zavada, Appl. Phys. Lett. 69, 2083 共1996兲. 11 T. M. Levin and G. H. Jessen 共unpublished兲. 12 T. E. Everhart and P. H. Hoff, J. Appl. Phys. 42, 5837 共1971兲. 13 L. J. Brillson and R. E. Viturro, Scanning Microsc. 2, 789 共1988兲. 14 X. A. Cao et al., J. Electron. Mater. 28, 261 共1999兲. 15 S. J. Pearton, C. R. Abernathy, and F. Ren, Electron. Lett. 30, 527 共1994兲. 16 D. M. Hansen, R. Zhang, N. R. Perkins, S. Safvi, L. Zhang, K. L. Bray, and T. F. Kuech, Appl. Phys. Lett. 72, 1244 共1998兲.