JEM_1102-R7
7/14/04
5:38 PM
Page 881
Journal of ELECTRONIC MATERIALS, Vol. 33, No. 8, 2004
Regular Issue Paper
Correlation of CdZnTe(211)B Substrate Surface Morphology and HgCdTe(211)B Epilayer Defects J. ZHAO,1,4 Y. CHANG,1 G. BADANO,1 S. SIVANANTHAN,1 J. MARKUNAS,2 S. LEWIS,2 J.H. DINAN,2 P.S. WIJEWARNASURIYA,3 Y. CHEN,3 G. BRILL,3 and N. DHAR3 1.—Microphysics Laboratory, University of Illinois at Chicago, Chicago, IL 60607. 2.— United States Army RDECOM CERDEC NVESD, Fort Belvoir, VA, 22060. 3.—United States Army Research Laboratory, Adelphi, MD, 20783. 4.— E-mail:
[email protected]
We present results on the surface morphology and recombination lifetimes of molecular-beam epitaxy (MBE)-grown HgCdTe (211)B epilayers and correlate them with the roughness of the CdZnTe substrate surfaces. The substrate surface quality was monitored by in-situ spectroscopic ellipsometry (SE) and reflection high-energy electron diffraction (RHEED). The SE roughness of the substrate was measured after oxide desorption in the growth chamber. The RHEED patterns collected show a strong correlation with the SE roughness. This proves that SE is a valuable CdZnTe prescreening tool. We also found a correlation between the substrate roughness and the epilayer morphologies. They are characterized by a high density of thin elongated defects, “needle defects,” which appear on most samples regardless of growth conditions. The HgCdTe epilayers grown on these substrates were characterized by temperature-dependent, photoconductive decay-lifetime data. Fits to the data indicate the presence of mid-gap recombination centers, which were not removed by 250°C/24-h annealing under a Hg-rich atmosphere. These centers are believed to originate from bulk defects rather than Hg vacancies. We show that Te annealing and CdTe growth on the CdZnTe substrates smooth the surface and lower substantially the density of needle defects. Additionally, a variety of interfacial layers were also introduced to reduce the defect density and improve the overall quality of the epilayer, even in the presence of less than perfect substrates. Both the perfection of the substrate surface and that of its crystalline structure are essential for the growth of high-quality material. Thus, CdZnTe surface polishing procedures and growth techniques are crucial issues. Key words: Molecular beam epitaxy (MBE), HgCdTe, defects, surface morphology, ellipsometry, reflection high-energy electron diffraction (RHEED)
INTRODUCTION High-quality mercury cadmium telluride, suitable for device applications, is difficult to obtain by either bulk or epitaxial growth. Compared with other material growth techniques, molecular-beam epitaxy (MBE) growth is most suitable for the growth of high-quality complex heterojunctions and multilayer structures for future, infrared focal-plane arrays. (Received September 26, 2003; accepted April 9, 2004)
One of the challenges posed by the MBE growth of HgCdTe is the understanding and control of visible defects on the surface. The defects observed are of several varieties.1,2 Some, such as voids and microtwins, are introduced during growth; others are more related to imperfections or substrate preparation procedures. Even though substrate issues for the growth of HgCdTe have been reported previously,3,4 more studies are needed to fully understand and control substrate-related defects. Here, we present our results on the surface morphology of HgCdTe epilayers grown by MBE. These 881
JEM_1102-R7
7/14/04
5:38 PM
Page 882
Zhao, Chang, Badano, Sivananthan, Markunas, Lewis, Dinan, Wijewarnasuriya, Chen, Brill, and Dhar
882
layers are characterized by thin elongated defects (“needle defects”), which, now we believe, are substrate related. The CdZnTe substrates used for these growths were characterized using in-situ spectroscopic ellipsometry (SE) and reflection high-energy electron diffraction (RHEED). Then, the HgCdTe surface morphology was correlated with the CdZnTe substrate properties. EXPERIMENTS The HgCdTe epilayers were grown in a Riber 32P MBE system (Rueil-Malmaison, France) equipped with RHEED, an 88-wavelength spectroscopic ellipsometer, a pyrometer, and a Hg valved cell. Solid CdTe and Te were used as source materials. Elemental In was used as the n-type dopant. For this study, front-side-polished CdZnTe (211)B substrates with a Zn mole fraction of 3.5 1% were purchased from Nikko Materials USA, Inc. (Chandler, AZ).5 Their surface appeared smooth and mirrorlike. Before growth, the CdZnTe substrates were chemically cleaned using trichloroethylene, acetone, and methanol; etched in a 1% bromine/methanol solution; and rinsed in methanol and deionized water. Then, the substrates were dried and mounted between a spring plate and a graphite thermal diffuser plate on an In-free holder. Extra care was taken to ensure a good mounting repeatability for all substrates. Once introduced into the growth chamber, excess Te present on the sample surface was stripped at 250°C, and CdTe was deposited to reduce the surface roughness. In-situ SE and RHEED were used to monitor the oxide desorption, interface formation, and growth together with the substrate thermocouple.6 The pyrometer was used to help monitor the substrate temperature. Different kinds of interfacial layers were used between the substrate and the epilayer (Fig. 1). It was suggested by our previous qualitative findings that abrupt interfaces interposed between HgCdTe layers of different composition might help block the propagation of substrate-related defects. Most layers were grown with mid-wavelength infrared/very long
wavelength infrared interfacial layers. The growth rate of HgCdTe was 6–8 Å/sec; the Cd molar fraction was chosen to be about 0.23. Our SE is a rotating analyzer ellipsometer that can measure 88 wavelengths in the range from 1.6 eV to 4.3 eV (the Woollam M88, J.A. Woollam, Lincoln, NE). The evolution of the dielectric function during the substrate preparation procedure was recorded in real time. The ellipsometric roughness was measured using the standard two-layer model, comprised of a temperature-dependent library of CdZnTe pseudodielectric functions and a surface roughness layer of variable thickness modeled in the Bruggeman approximation (the EMA). This effective medium theory describes well the changes caused by microscopic roughness. Detailed description of the system and the data analysis can be found in our previous paper.7 The surface morphology of the HgCdTe epilayers was inspected using a Nomarski optical microscope and an atomic force microscopy (AFM). The minority-carrier recombination lifetimes of these HgCdTe layers were also measured using the photoconductive decay (PCD) method; temperature-dependent lifetime data were fit to extract the density and energy level of the dominant Shockley–Reed recombination centers. X-ray double-crystal rocking curve measurements gave information on the crystalline perfection of the layers grown. RESULTS AND DISCUSSION Figure 2 shows a comparison of the CdZnTe surface roughness obtained by SE during preparation and the corresponding RHEED patterns. Our CdZnTe substrates were (211)B oriented, and the RHEED patterns were taken from the [111] azimuth. To avoid fitting errors, the SE surface roughness was measured after removing the oxide created by the cleaning procedure. RHEED is very sensitive to surface imperfections. It is well known that roughness or imperfections in the crystalline structure will make the patterns spotty. However, the quantification of roughness from RHEED patterns is difficult, whereas the SE
a b Fig. 1. The typical structures used for the HgCdTe growth on CdZnTe with different interfacial layers.
JEM_1102-R7
7/14/04
5:38 PM
Page 883
Correlation of CdZnTe(211)B Substrate Surface Morphology and HgCdTe(211)B Epilayer Defects
a
b
883
d
c
Fig. 2. The RHEED patterns of CdZnTe substrates versus ellipsometric roughness. (a) 3 Å, (b) 10 Å, (c) 20 Å, (d) 31 Å.
measurement of surface roughness is simple and reproducible. As is shown in Fig. 2, only with SE roughness lower than 10 Å are the RHEED patterns nearly streaky; the higher the SE roughness is, the spottier the RHEED patterns are. Thus, the surface roughness of the substrates correlates well with the RHEED patterns, and SE can be used as a valuable tool for the screening of CdZnTe substrates in situ or ex situ. Another substrate preparation parameter that we tried to vary is the thickness of the CdTe layer grown on the CdZnTe. The HgCdTe samples grown can be divided into three groups according to the amount of CdTe deposited on the substrate: those without CdTe, with 8 Å of CdTe, or with 80 Å of CdTe. The effect of CdTe growth during the CdZnTe preparation will be discussed later. Here, we are concerned with its effect on the epilayer. Nomarski optical microscopy on epilayers grown on microscopically rough CdZnTe reveals elongated defects (needle defects) oriented in a preferential direction (Fig. 3). The detailed structure of one of these defects, as revealed by AFM, is shown in Fig. 4. These defects are very thin (about 0.2–0.3 µm), and their length is about 8 µm. Trenches were found by AFM at one side of some of the needle defects. A perpendicular section shows that the total height of these defects is about 10 nm (from the bottom of the
trench to the top of the ridge). The average length of the needle defects varies from layer to layer, from a few microns to greater than 10 µm. Some samples showed a low density of long defects; some had short but very numerous defects. No relationship between the length and the density was found. Experiments showed that the needle defects were occasionally accompanied by voids or twins in the epilayers. This indicates that the presence of needle defects is not well correlated with the HgCdTe growth conditions. To verify this conclusion, HgCdTe layers were grown on CdTe/Si substrates using the same growth conditions. As we expected, no needle defects appeared on these HgCdTe layers. The SE readings corroborate these findings. Even if the defects we are discussing here are optically macroscopic and, therefore, generate an incoherent, diffused component at near specular angles, the SE roughness parameter can still be used to characterize the coherent, polarized part of the signal reflected off the epilayers. Hence, the SE roughness reading provides at least quantitative information on the sample evolution. We find that the SE roughness increases during growth by about 5 Å on samples with higher densities of needle defects. All of these results indicated that the needle defects most probably originate from the CdZnTe substrate.
a
b
Fig. 3. Nomarski pictures (1,000, sized 100 µm 100 µm) of needle defects on the surface of a HgCdTe epilayer grown on CdZnTe substrates. (a) Needle defects with a large void; (b) Needle defects with high-density twins.
JEM_1102-R7
7/14/04
5:38 PM
Page 884
Zhao, Chang, Badano, Sivananthan, Markunas, Lewis, Dinan, Wijewarnasuriya, Chen, Brill, and Dhar
884
a
b
Fig. 4. Left: an AFM micrograph of a needle defect (5 µm 5 µm). Top right: section analysis perpendicular to the same needle defects. The total height of the needles is about 10–20 nm.
Fig. 6. X-ray double-crystal rocking curves of the CdZnTe substrate measured at different spots on the substrate. At most angles, double or multiple peaks were found.
Fig. 5. Plot of the HgCdTe needle-defect density versus the CdZnTe surface roughness as measured by in-situ ellipsometry. The squares represent samples that were grown on ~80 Å of CdTe interfacial layer. The lines on the plot are meant to show the trend. The diamonds and the triangles represent samples grown either without a CdTe layer or with ~8 Å of CdTe. They show much higher needle-defect densities.
All HgCdTe layers with high densities of needle defects had one thing in common: they were grown on substrates that had spotty RHEED patterns and a relatively thick SE roughness layer. Figure 5 shows the needle-defect density (the number of defects per unit area) versus the SE roughness for all samples grown. Data are grouped into three sets. The diamonds represent samples grown without CdTe at the interface, the triangles samples grown with 8 Å of CdTe, and the full squares samples grown with 80 Å of CdTe. In general, layers grown on substrates for
which the SE roughness was 10 Å had very low densities of needle defects or no such defects at all. Rougher substrates resulted in proportionally higher densities of needle defects. The interpolation lines on the plot are meant to show this trend. Compared with samples with only 8 Å of CdTe or without CdTe interfacial layers, the samples with 80 Å of CdTe layers showed much lower needle-defect densities. An 80-Å CdTe layer grown before the nucleation of HgCdTe helps to smooth the CdZnTe substrates and, therefore, lowers the defect densities. However, the presence of CdTe may reduce the benefit of lattice match. The growth of a lattice-matched CdZnTe layer could be a better solution. X-ray diffraction also was used to characterize the CdZnTe substrates. X-ray double-crystal rocking curve measurements revealed that some CdZnTe substrates had different diffraction patterns at
JEM_1102-R7
7/14/04
5:38 PM
Page 885
Correlation of CdZnTe(211)B Substrate Surface Morphology and HgCdTe(211)B Epilayer Defects
885
a
b Fig. 7. The recombination lifetime measured by the PCD method. The HgCdTe layer surface featured high-density needle defects. (a) Data from an as-grown HgCdTe layer with a dominant recombination-center level at 0.75Eg related to Hg vacancies. (b) Lifetime data after n-type annealing with a dominant recombination-center level at 0.50Eg related to needle defects.
different azimuths. Only along particular azimuths did these CdZnTe substrates show a single diffraction peak. At most angles, double or multiple peaks were found (Fig. 6). We are currently correlating the x-ray mappings of the substrates with the defect densities of the epilayers. We propose to analyze the x-ray results in detail in a future paper. Samples with high needle-defect densities also showed high etching pit densities and low carrier mobilities. Finally, the lifetime data in Fig. 7 show that, for as-grown HgCdTe, the dominant recombination center is around 0.75Eg above the valence band edge. We believe this center to be due to Hg vacancies. After N-type annealing in a Hg atmosphere, the dominant level moved to about mid-gap (0.50Eg). Chen and Schiebel8 reported that this mid-energy gap level is due to dislocations. This recombination center could be related to the presence of needle defects. Further studies are needed to understand fully the effect of needle defect on the electrical properties of HgCdTe materials. SUMMARY Needle-defect features appear on the surface of HgCdTe epilayers grown by MBE regardless of growth conditions. An excellent correlation was found between the degrees of roughness of the CdZnTe substrate surface, as monitored in situ by SE, and the density of needle defects. The growth of a CdTe buffer
layer on a CdZnTe substrate lowers the density of the defects. A dominant minority-carrier recombination center was present in these layers, and the relationship between the needle defects and this center is under investigation. ACKNOWLEDGEMENTS This work was supported by the Department of Defense Multidisciplinary University Research Initiative program administered by the Army Research Office under Grant No. DAAD19-01-1-406. We acknowledge use of facilities at the Center for High Resolution Electron Microscopy at Arizona State University. REFERENCES 1. L. He, Y. Wu, L. Chen, S.L. Wang, M.F. Yu, Y.M. Qiao, J.R. Yang, Y.J. Li, R.J. Ding, and Q.Y. Zhang, J. Cryst. Growth 227, 677 (2001). 2. M. Zandian and E. Goo, J. Electron. Mater. 30, 623 (2001). 3. R. Triboulet, A. Tromson-Carli, D. Lorans, and T. Nguyen Duy, J. Electron. Mater. 22, 827 (1993). 4. J.P. Tower, S.P. Tobin, M. Kestigian, P.W. Norton, A.B. Bollong, H.F. Schaake, and C.K. Ard, J. Electron. Mater. 24, 497 (1995). 5. Nikko Materials USA, Inc., 125 North Price Road, Chandler, AZ 85224. 6. J.D. Benson, A.B. Cornfeld, M. Martinka, J.H. Dinan, B. Johs, P. He, and J.A. Woollam, J. Cryst. Growth 175, 659 (1997). 7. G. Badano, J. Garland, and S. Sivananthan, J. Cryst. Growth 251, 571 (2003). 8. M.C. Chen and R.A. Schiebel, J. Appl. Phys. 71, 5269 (1992).
