Journal of ELECTRONIC MATERIALS, Vol. 42, No. 11, 2013

DOI: 10.1007/s11664-013-2691-z Ó 2013 TMS (outside the USA)

Analysis of Mesa Dislocation Gettering in HgCdTe/CdTe/ Si(211) by Scanning Transmission Electron Microscopy R.N. JACOBS,1,4 A.J. STOLTZ,1 J.D. BENSON,1 P. SMITH,1 C.M. LENNON,1 L.A. ALMEIDA,1 S. FARRELL,2 P.S. WIJEWARNASURIYA,2 G. BRILL,2 Y. CHEN,2 M. SALMON,3 and J. ZU3 1.—U.S. Army RDECOM, CERDEC Night Vision and Electronic Sensors Directorate, Fort Belvoir, VA 22060, USA. 2.—U.S. Army Research Laboratory, Adelphi, MD 20783, USA. 3.—Evans Analytical Group, Inc., Raleigh, NC 27606, USA. 4.—e-mail: [email protected]

Due to its strong infrared absorption and variable band-gap, HgCdTe is the ideal detector material for high-performance infrared focal-plane arrays (IRFPAs). Next-generation IRFPAs will utilize dual-color high-definition formats on large-area substrates such as Si or GaAs. However, heteroepitaxial growth on these substrates is plagued by high densities of lattice-mismatchinduced threading dislocations (TDs) that ultimately reduce IRFPA operability. Previously we demonstrated a postgrowth technique with the potential to eliminate or move TDs such that they have less impact on detector operability. In this technique, highly reticulated mesa structures are produced in as-grown HgCdTe epilayers, and then subjected to thermal cycle annealing. To fully exploit this technique, better understanding of the inherent mechanism is required. In this work, we employ scanning transmission electron microscopy (STEM) analysis of HgCdTe/CdTe/Si(211) samples prepared by focused ion beam milling. A key factor is the use of defect-decorated samples, which allows for a correlation of etch pits observed on the surface with underlying dislocation segments viewed in cross-section STEM images. We perform an analysis of these dislocations in terms of the general distribution, density, and mobility at various locations within the mesa structures. Based on our observations, we suggest factors that contribute to the underlying mechanism for dislocation gettering. Key words: HgCdTe thin films, dislocations, etch pit density, STEM, lattice-mismatched substrates, FIB

INTRODUCTION Realization of a viable low-cost, large-area alternative substrate to CdZnTe for HgCdTe-based infrared detector arrays has motivated over two decades of research.1–4 The research has primarily been focused on silicon, which continues to be the most technologically developed, inexpensive semiconductor substrate. In addition, Si is available in large sizes (over 12 inches) and is thermally

(Received April 3, 2013; accepted July 5, 2013; published online August 14, 2013)

3148

compatible with existing read-out integrated circuits to which detector arrays are mated during IRFPA fabrication. The most challenging obstacle to HgCdTe/CdTe/Si heteroepitaxial growth is the large lattice mismatch (19%), which inevitably causes unacceptable levels of threading dislocations (TDs), particularly for long-wave infrared (LWIR) detector applications.5 Thermal mismatch between Si and HgCdTe/CdTe can potentially generate TDs, but recent studies6 comparing CdTe growth on Si, GaAs, and Ge substrates have suggested that lattice mismatch is the primary factor. Lattice-mismatchinduced TDs reduce detector operability through generation of shunt currents at electrical junctions in the device layer.7 Ongoing studies indicate that

Analysis of Mesa Dislocation Gettering in HgCdTe/CdTe/Si(211) by Scanning Transmission Electron Microscopy

discrete impurities and defect clustering are important factors, but will likely work in concert with TDs to cause inoperable pixels in detector arrays.8–10 Therefore, improved detector performance is expected for material with lower TD density. Prior to growth of HgCdTe by molecular beam epitaxy (MBE), an 8-lm- to 15-lm-thick CdTe buffer layer is typically grown on Si(211) surfaces. Thermal cycle annealing (TCA) throughout the growth of this buffer layer has played a significant role in the reduction of surface etch pit density (EPD). The EPD is interpreted to approximate the TD density at the film surface. For the CdTe buffer layer, EPD levels in the 5 9 106/cm2 to 1 9 107/cm2 range are achieved.10 For HgCdTe, an EPD below 5 9 105/cm2 is desired for LWIR applications. It has not been feasible to implement TCA during MBE growth of the HgCdTe layer due to practical limitations in the amount of Hg overpressure needed to prevent desorption and to maintain stable surfaces. Nonetheless, postgrowth TCA has been performed and been shown to have a significant impact on TD density.11–13 In this approach, the as-grown HgCdTe/CdTe/Si heterostructure is placed in a quartz ampoule with a droplet of Hg. After evacuation, the ampoule is heated to 275°C and then cycled to temperatures of 385°C (or greater). Etch pit densities of approximately 1 9 106/cm2 have been routinely achieved, as measured by Benson defect decoration.9 A complementary dislocation reduction technique incorporates plasma-etched mesa structures prior to annealing.14,15 In much the same way that TDs move towards the film growth surface during epitaxy, the sidewalls of a mesa structure may produce a similar effect. The dislocations experience an image force due to the proximity of a sidewall. If given sufficient energy (through annealing), the dislocation will preferentially move toward the sidewalls, which allow for stress reduction. The technique (termed mesa dislocation gettering) has been applied to HgCdTe/CdTe/Si(211), where various mesa shapes and sizes were studied. Of the various geometries examined, the dislocation gettering effect was only observed in rectangular mesas where the short dimension was approximately 20 lm.16 A mechanism for these observations has not yet been proposed. In this work we used scanning transmission electron microscopy (STEM) to improve understanding of the mechanism associated with mesa dislocation gettering in HgCdTe/CdTe/Si(211). Focused ion beam (FIB) sample preparation allowed for examination of dislocation segments over a large electron-transparent area. We examine the surface and cross-sectional dislocation distribution within specific structures that exhibit the gettering phenomenon. Based on our observations, we suggest factors related to the geometry dependence of mesa dislocation gettering.

3149

EXPERIMENTAL PROCEDURES CdTe buffer layers were deposited onto Si(211) wafers in a VG-80 MBE reactor. These composite wafers were then transferred in vacuo to an adjacent MBE reactor (modified Fisons VG-80) for deposition of 10-lm-thick Hg(1x)CdxTe films. The epilayers were ‘‘single-color,’’ long-wave (x  0.2), and doped n-type with In. After epilayer growth, mesa delineation was performed using an inductively coupled plasma (ICP) reactor (Plasma-Therm). The nominal ICP etch depth was approximately 8 lm to 9 lm. A defect gettering mask was used which incorporated various geometrical shapes and sizes, in an effort to study their effectiveness in achieving the desired dislocation gettering effect. Details of this gettering mask and the ICP plasma processing conditions are discussed in the literature.16 The patterned and mesa-delineated samples were then subjected to four annealing cycles between 250°C and 494°C. The cyclic annealing apparatus and procedure have been described in greater detail in an earlier paper.12,13 Finally, samples were Benson defect decorated for examination of the surface etch pit density. Etch pit densities were initially examined using bright-field Nomarski microscopy and scanning electron microscopy. Samples for STEM analysis were prepared by FIB milling. The samples were initially coated with a layer of Pt prior to placing them in the FIB system (FEI single-beam system). Samples were milled with a 30-kV Ga+ beam with ion currents as low as 50 pA for final polishing down to electron transparency. The lift-out technique was applied to place the specimen onto STEM grids for subsequent analysis. STEM analysis was performed using a Hitachi dedicated STEM instrument (HD2300) operated at 200 kV. The dedicated STEM allows simultaneous viewing of the image and the diffraction pattern (convergent-beam electron diffraction). The STEM can also be operated in secondary-electron (backscattering) mode, which allowed for low-magnification observation of samples. All STEM images shown herein are bright-field images obtained with the beam direction along the h011i zone axis. It is worth mentioning the choice of STEM, rather than transmission electron microscopy (TEM). In comparison with STEM, TEM offers greater diffraction contrast, and dislocation lines will generally appear darker in the resulting image. On the other hand, bend contours, thickness fringes, and other features are also easily visible in TEM images. These artifacts (occasionally resulting from sample preparation) may obscure the viewing of dislocation lines. While the overall contrast in STEM is lower than that in TEM, the images are less prone to these artifacts, but the sharp contrast from dislocation line segments completely extending through the sample remains visible. Thus, STEM was initially chosen for this dislocation study in an effort to achieve simplified and uniform image contrast.

3150

Jacobs, Stoltz, Benson, Smith, Lennon, Almeida, Farrell, Wijewarnasuriya, Brill, Chen, Salmon, and Zu

RESULTS AND DISCUSSION Various mesa shapes, plasma-etched in long-wave HgCdTe/Si, were previously examined16 in an effort to gain better understanding of the dislocation gettering effect. These geometries included squares, circles, diamonds, rectangles, and others, oriented along various crystallographic directions. In that work, it was observed that only one geometry clearly and consistently showed some level of the dislocation gettering phenomenon: long rectangular bars oriented along h011i and h111i produced a distinct EPD distribution. Figure 1a shows a top-down SEM image of bars oriented parallel to h011i, where an EPD of >1 9 107/cm2 is typically observed at end regions. At a distance 20 lm away from the ends of these structures, there is a relative scarcity of etch pits and the EPD is typically well under 5 9 105/cm2. Figure 1b shows an SEM image of bars oriented parallel to h111i. In this case the EPD distribution appears to be concentrated along the center of the bar. For bars oriented at angles in between h011i and h111i, some degree of the two gettering phenomena was observed, with EPD in the center of the bars decreasing from >1 9 107/cm2 to <5 9 105/cm2 as the angle moves from h111i to h011i. It should also be noted that there is a dimensional limit for which the gettering effect is observed, as discussed below. These structures were chosen for our current study, as they represent the only geometry which clearly and reproducibly exhibits a gettering effect. We refer to h011i-oriented bars as structure 1 and h111i-oriented bars as structure 2. The distribution of defect-decorated surface etch pits was examined by Nomarski and SEM for both structures 1 and 2. STEM was then employed to examine the cross-sectional distribution of dislocations within important regions of these structures. Figure 2 shows a FIB image at the end (a) and middle (b) regions of a bar from structure 1. A relatively high EPD is observed at the end in comparison with the mid-region. A row of several etch pits toward the end region was targeted for subsequent cross-section STEM imaging. Likewise, in the midregion, with a relatively low EPD, a lone etch pit was targeted for cross-section STEM imaging. These regions (containing etch pits), were chosen to increase the probability of finding associated TD segments, and if possible, to identify the TD in terms of Burgers vector and mobility. Figure 3 shows STEM images of the end (a) and mid (b) regions of structure 1. A schematic is also shown in Fig. 3c to help visualize the cross-sectional regions being imaged. In both cases, two adjacent FIB sections were thinned to allow for observation of the dislocation distribution from center to sidewall of the mesa cross-sections. This was necessary due to practical limitations in the width of STEM samples prepared by FIB milling. It should also be noted that Fig. 3a, b are composite images where several STEM images have been overlaid, and in some cases

Fig. 1. Top-down SEM images of defect-decorated rectangular mesa structures in HgCdTe/CdTe/Si(211): (a) structure 1, bars oriented along h011i; (b) structure 2, bars oriented along h111i.

magnified, to enhance the viewing of TD segments throughout the mesa cross-section. For the endregion (a), we observe dislocation line segments throughout the sample, but somewhat concentrated toward the sidewall. For the mid region (b), the image is virtually free of dislocation segments at both the center and sidewall. The relatively large number of dislocation segments toward the CdTe interface is to be expected from lattice mismatch. The TD density has been determined from Fig. 3a, b. The calculation is based on measuring the overall line segment density within an estimated volume of material in the image. In particular, a grid is overlaid on the image and the total number of intersections, N, of dislocation segments with grid lines is determined. In addition, the total length, L, of horizontal and vertical grid lines is determined for the measurement area of interest. An estimated dislocation density is then determined as 2N/Lt, where t is the STEM sample thickness. This method has

Analysis of Mesa Dislocation Gettering in HgCdTe/CdTe/Si(211) by Scanning Transmission Electron Microscopy

3151

Fig. 2. Focused ion beam image of a defect-decorated mesa in structure 1. Etch pits from two regions: (a) end and (b) mid, were targeted for subsequent cross-section STEM.

been described in greater detail in the literature.17 Due to the different measurement technique, such values should not be compared with those determined by defect decoration (EPD). In Fig. 4 the dislocation density is plotted versus HgCdTe film thickness for end and mid-regions. In both cases it is observed that the TD density shows a 1/x fall-off (where x is the film thickness) from the CdTe interface. For the mid-region, the TD essentially falls to zero beyond a thickness of 6 lm, whereas for the end-region the density does not fall below 5 9 107/cm2. A similar procedure was carried out for examination of structure 2, where etch pits mostly align along the center of h111i-oriented mesa bars. The FIB image in Fig. 5 shows three regions, where pits were targeted for cross-sectioning and subsequent STEM analysis. These three regions consist of two sidewalls (1 and 2), and the mesa center where the surface EPD is relatively concentrated. Figure 6 shows corresponding STEM images for (a) sidewall 1, (b) mesa center, and (c) sidewall 2. While

Fig. 3. Composite STEM images of the (a) end and (b) mid regions of the structure 1 mesa from Fig. 2. (c) Three-dimensional depiction of the FIB foils taken from the mesa structure for STEM analysis. The electron beam is parallel to the h011i direction.

discrete dislocation segments are observed throughout the sidewall regions (a and c), a much higher density is observed at the mesa center (b). Furthermore, the entanglement of dislocation segments at the mesa center (b) is more pronounced throughout the cross-section and is not limited to the CdTe interface region, as is the case for the sidewalls (a and c). Figure 7 shows the dislocation density versus HgCdTe film thickness for sidewalls 1 and 2,

3152

Jacobs, Stoltz, Benson, Smith, Lennon, Almeida, Farrell, Wijewarnasuriya, Brill, Chen, Salmon, and Zu

Fig. 4. Dislocation density as a function of film thickness for the end and mid cross-sectional regions of structure 1.

and the mesa center. Once again, a 1/x fall-off occurs in all cases; however, the dislocation density for the mesa center remains relatively high throughout the cross-section. Note again that these calculations are based solely on the images. The FIB-thinned foil (150 nm to 250 nm thick) may not capture all dislocation segments associated with the targeted etch pits. Due to the larger area sampled, surface EPD is a better statistical estimator of dislocation density than direct measurement from cross-section STEM images. However, the observed trend is consistent. An investigation of dislocation mobility was carried out by a two-step procedure involving gÆb and u 9 b analysis. Here, g is the vector corresponding to a specific set of Bragg diffracted lattice planes in the crystal, b is the dislocation Burgers vector, and u is a unit vector parallel to the dislocation line. With the STEM instrument in diffraction mode, the sample was tilted so that only particular sets of g reflections were excited by the incident electron beam. Under a unique tilt condition, the diffracting planes are parallel to a dislocation’s displacement field so that gÆb = 0. For this unique condition, a dislocation with its characteristic b should not be visible in the corresponding image. Thus, by examining a dislocation under different sets of g reflections, its characteristic Burgers vector b can be predicted from existing tables in the literature. The crystallographic orientations of dislocation lines, u, were determined from diffraction patterns of the corresponding images, in addition to the [011] stereographic projection. The slip planes associated with specific dislocations were determined from u 9 b. Since the mobile slip planes in the zincblende crystal structure are known, it is then possible to predict the mobility (or immobility) of dislocations. This dislocation analysis procedure is well established,17

Fig. 5. Focused ion beam images of a defect-decorated mesa in structure 2. Etch pits from three regions: (a) sidewall 1, (b) mesa center, and (c) sidewall 2, were targeted for subsequent crosssection STEM.

and its use in STEM is described in greater detail in a previous paper where dislocations in nonpatterned TCA samples were investigated.18 Table I presents a summary of these results in terms of the percentage of sessile dislocations determined for each region in the two mesa structures. Dislocation segments at the mesa sidewalls of structure 2 were mostly mobile (i.e., on {111} slip planes). Similarly, segments analyzed in structure 1 (Fig. 3a, b) were also mostly mobile. In comparison with the other regions examined, dislocation segments found along the center of structure 2 (Fig. 6b) had a relatively greater likelihood of being sessile (i.e., lying on {100} or {110} planes). It must be acknowledged, however, that the relative scarcity of dislocation segments in other images may prevent a sufficient comparison of all regions in structures 1 and 2. In addition, it should be noted that Burgers vector and mobility could not be unambiguously determined for all dislocations. Nonetheless, a higher density of dislocation segments indicates a greater likelihood for entanglement and subsequent immobility. This observation supports a previous assertion16 that dislocations are likely pinned (i.e., immobile) along the center of mesa structure 2. It was experimentally determined that there exists an approximate dimension at which the onset of dislocation gettering phenomena occurs in the mesa structures (Fig. 1). For structure 1 (bars oriented along h011i), a minimum width of roughly 20 lm is required to discern gettering at the end regions. For widths larger than this, the distribution of surface etch pits becomes increasingly uniform. At widths less than 20 lm, very few etch pits, <1 9 104/cm2, are observed in all regions of the structure. For a fixed width of 20 lm, the bar length

Analysis of Mesa Dislocation Gettering in HgCdTe/CdTe/Si(211) by Scanning Transmission Electron Microscopy

3153

Fig. 6. Stitched STEM images of (a) sidewall 1, (b) mesa center, and (c) sidewall 2 of the structure 2 mesa from Fig. 5. The schematic (d) provides a three-dimensional depiction of the FIB foils taken from the mesa structure for STEM analysis. The electron beam is parallel to the h011i direction.

must be at least 50 lm (i.e., roughly more than twice the bar width) to see the gettering phenomena.16 For structure 2, (bars oriented along h111i), 20 lm appears to be the approximate width for which etch pits are mostly distributed along the center of the mesa bars. Above this bar width, the etch pits became distributed more and more uniformly. Below this bar width, the etch pits are distributed more uniformly but eventually decrease to<1 9 104/cm2. For a fixed width of 20 lm, a bar length of roughly 50 lm is the minimum required for observation of the structure 2 etch pit distribution. In a paper published by Xhang et al., an EPD of less than 2.0 9 104/cm2 (i.e., virtually TD free) was

demonstrated in 70-lm square mesas patterned in ZnSe/GaAs.19 Image-force-induced TD movement via patterning and annealing were cited for the low EPD. An upper limit of mesa size was not determined, and thus it is likely that the minimum dimension is greater than 70 lm. We suggest that growth orientation and mesa geometry are significant factors that could explain the differences in gettering phenomena observed in our work. The ZnSe films were (100) oriented and patterned with square mesas whose sides were parallel to h011i directions. Owing to the zincblende crystal structure (for both ZnSe and HgCdTe), the glide system is h011i{111}. Thus, the (100)-oriented ZnSe mesa

3154

Jacobs, Stoltz, Benson, Smith, Lennon, Almeida, Farrell, Wijewarnasuriya, Brill, Chen, Salmon, and Zu

individual TD segments analyzed. Nonetheless, there are insufficient data (due to the relatively limited area available from STEM imaging) to make any definite conclusions. It remains unclear what gives rise to the observed distribution of dislocations in these two mesa structures. It is suggested, however, that the orientation of the {111} planes (with respect to mesa sidewalls) is key to uncovering the underlying mechanism. CONCLUSIONS

Fig. 7. Dislocation density as a function of film thickness for sidewalls 1 and 2, and the mesa center of structure 2.

Table I. Percent sessile dislocation segments (of the total analyzed in each region) for different regions in mesa structures 1 and 2 Mesa Structure and Region Structure 1, h011i End region Mid region Structure 2, h111i Sidewall 1 Sidewall 2 Mesa center

Percent Sessile Dislocation Segments 13 17 14 25 46

structures should have fourfold symmetry in terms of TD gettering toward any of the four sidewalls. This is decidedly not the case for the mesa structures currently under investigation. The HgCdTe films are (211) oriented and patterned with rectangular mesas with sides parallel to h011i and h111i directions. At best, there appears to be a twofold symmetry in terms of dislocation gettering phenomena. In addition, we have previously shown that dislocation gettering does not occur for square mesas patterned in HgCdTe/CdTe/Si(211) heterostructures.16 It is worth noting that defect-decorated sidewalls have revealed etch pits that, on occasion, appear to line up along directions parallel to two {111}-type slip planes. Assuming a (211) growth orientation,  and (111)  these are the (111) slip planes that, respectively, intersect the growth surface along     213 and [231 cross-hatch lines. In a previous paper16 we proposed this to be evidence of enhanced migration along these particular slip planes. In the present study we did not observe a greater occurrence of any particular {111} slip plane(s) for the

We performed an investigation of mesa dislocation gettering structures in HgCdTe/CdTe/Si(211) heterostructures. Two rectangular mesa structures (one along h011i and the second along h111i) exhibiting different dislocation gettering phenomena were examined via STEM. We demonstrated a direct correlation between the distribution of surface etch pits and that of dislocation segments in the cross-sectional structures. Dislocation density calculated directly from the STEM images provided a quantitative comparison of different regions within the rectangular mesa structures. Analysis of TD segments suggests a greater likelihood of sessile dislocations along the center of structures aligned to the h111i direction. By comparing our observations with those in a study reported by Xhang, we surmise that growth orientation, and subsequently the orientation of the {111} slip planes with respect to mesa sidewalls, are key factors to explain the differences in our observations. A potentially useful study would be to repeat the mesa dislocation gettering experiments for MBE HgCdTe/CdTe/Si with a (100) rather than (211) growth direction. The patterned structures would incorporate square mesas with four h011i-oriented sidewalls, in addition to the rectangular structures discussed in this paper. Such study may provide more clues to achieve deeper understanding of the mechanism responsible for geometry-dependent mesa dislocation gettering. REFERENCES 1. J.M. Arias, R.E. DeWames, S.H. Shin, J.G. Pasko, J.S. Chen, and E.R. Gertner, Appl. Phys. Lett. 54, 1025 (1989). 2. R. Sporken, S. Sivananthan, K.K. Mahavadi, G. Monfroy, M. Boukerche, and J.P. Faurie, Appl. Phys. Lett. 55, 1879 (1989). 3. L.A. Almeida, L. Hirsch, M. Martinka, P.R. Boyd, and J.H. Dinan, J. Electron. Mater. 30, 608 (2001). 4. M. Carmody, J.G. Pasko, D. Edwall, R. Bailey, J. Arias, M. Groenert, L.A. Almeida, J.H. Dinan, Y. Chen, G. Brill, and N.K. Dhar, J. Electron. Mater. 35, 1417 (2006). 5. S.M. Johnson, D.R. Rhiger, J.P. Rosenbeck, J.M. Peterson, S.M. Taylor, and M.E. Boyd, J. Vac. Sci. Technol. B10, 1499 (1992). 6. R.N. Jacobs, J. Markunas, J. Pellegrino, L.A. Almeida, M. Groenert, M. Jaime-Vasquez, N. Mahadik, C. Andrews, and S.B. Qadri, J. Cryst. Growth 310, 2960 (2008). 7. L.O. Bubulac, J.D. Benson, R.N. Jacobs, A. Stoltz, M. Jaime-Vasquez, L.A. Almeida, A. Wang, L. Wang, R. Helmer, T.D. Golding, J.H. Dinan, M. Carmody, P. Wijewarnasuriya,

Analysis of Mesa Dislocation Gettering in HgCdTe/CdTe/Si(211) by Scanning Transmission Electron Microscopy

8. 9.

10. 11. 12. 13.

M.F. Lee, M.F. Vilela, J. Peterson, S.M. Johnson, D.F. Lofgreen, and D. Rhiger, J. Electron. Mater. 40, 280 (2011). M. Jaime-Vasquez, R.N. Jacobs, J.D. Benson, A.J. Stoltz, L.A. Almeida, L.O. Bubulac, Y. Chen, and G. Brill, J. Electron. Mater. 39, 951 (2010). J.D. Benson, L.O. Bubulac, P.J. Smith, R.N. Jacobs, J.K. Markunas, M. Jaime-Vasquez, L.A. Almeida, A. Stoltz, P.S. Wijewarnasuriya, G. Brill, Y. Chen, U. Lee, M.F. Villa, J. Peterson, S.M. Johnson, D.D. Lofgren, D. Rhiger, E.A. Patten, and P.M. Goetz, J. Electron. Mater. 39, 1080 (2010). J.D. Benson, R.N. Jacobs, J.K. Markunas, M. Jaime-Vasquez, P.J. Smith, L.A. Almeida, M. Martinka, M.F. Vilela, and U. Lee, J. Electron. Mater. 37, 1231 (2008). G. Brill, S. Farrell, Y.P. Chen, P.S. Wijewarnasuriya, M.V. Rao, J.D. Benson, and N. Dhar, J. Electron. Mater. 39, 967 (2010). S. Farrell, G. Brill, Y. Chen, P. Wijewarnasuriya, M.V. Rao, N. Dhar, and K. Harris, J. Electron. Mater. 39, 43 (2010). J.D. Benson, S. Farrel, G. Brill, Y. Chen, P.S. Wijewarnasuriya, L.O. Bubulac, P.J. Smith, R.N. Jacobs,

14. 15. 16. 17. 18. 19.

3155

J.K. Markunas, M. Jaime-Vasquez, L.A. Almeida, A. Stoltz, U. Lee, M.F. Vilela, J. Peterson, S.M. Johnson, D.D. Lofgreen, D. Rhiger, E.A. Patten, and P.M. Goetz, J. Electron. Mater. 40, 1847 (2011). P. Mu¨ller, H. Zogg, A. Fach, J. John, C. Paglino, A.N. Tiwari, M. Krejci, and G. Kostorz, Phys. Rev. Lett. 78, 3007 (1997). M. Yamaguchi, J. Mater. Res. 6, 376 (1991). A. Stoltz, J.D. Benson, M. Carmody, S. Farrell, P.S. Wijewarnasuriya, G. Brill, R.N. Jacobs, and Y. Chen, J. Electron. Mater. 40, 1785 (2011). D.B. Williams and C.B. Carter, Transmission Electron Microscopy. A Textbook for Materials Science (Berlin: Springer, 2009). R. Jacobs, J.D. Benson, A. Stoltz, L.A. Almeida, S. Farrell, G. Brill, M. Salmon, and A. Newell, J. Cryst. Growth 366, 88 (2013). X.G. Zhang, A. Rodriguez, X. Wang, P. Li, F.C. Jain, and J.E. Ayers, Appl. Phys. Lett. 77, 2524 (2000).