We report on the growth of near-minimum bandgap (0.1 eV) bulk InAs0.54Sb0.46 on GaAs with pronounced photoluminescence. Combining strain-mediating techniques effectively manages the ∼10% lattice mismatch. An interfacial misfit (IMF) dislocation array allows a GaAs substrate and a GaSb buffer layer to act as a direct substitute for a conventional GaSb substrate. We further increase the lattice constant with a linearly graded metamorphic buffer layer of AlGaInSb, upon which we grow an AlInSb virtual substrate with the targeted lattice constant of the active InAsSb region. We observe that the graded buffer not only manages the mismatch between GaSb and InAsSb but also reduces the density of residual threading dislocations from the GaSb/GaAs IMF by a factor of ∼10×. Material characterization demonstrates the viability of using the coupled approach of metamorphic, graded buffers and IMF arrays to manage the large lattice mismatch between the substrate and the mixed group V ternary InAsSb.
I. INTRODUCTION
InAsSb is a ternary, mixed group V alloy, and it has the smallest bandgap among the conventionally alloyed III-V compounds.1 GaSb is the only available III-V substrate with a lattice constant within the broad range between binary InAs and InSb constituents. Although InAs0.91Sb0.09 can be grown lattice matched to GaSb, its bandgap corresponds to absorption only at a specific wavelength within the mid-wave infrared (MWIR) range (3–5 μm). Reaching longer wavelengths requires higher Sb compositions and a substrate with a correspondingly larger lattice constant.2,3 Recent findings that the InAsSb bandgap bowing parameter, which determines the minimum bandgap of the alloy, is large enough to reach the long wavelength (LWIR) band (8–12 μm) make this a goal worth pursuing.4–7 As a direct-bandgap material, InAsSb has a spectral absorption that for practical purposes is indistinguishable from that of HgCdTe,8,9 making it a promising, lower-cost, III-V alternative for LWIR detector applications. With compositional grading and a virtual substrate (VS), we obtain InAsSb films on GaSb, which are unrelaxed and unstrained; in other words, this material exhibits an inherent lattice constant and bandgap. This is in contrast to earlier work10 that resulted in a material with systematically larger bandgaps, described with a smaller bowing parameter. It is therefore critical to maintain these conditions when attempting to grow InAsSb on a substrate with an even larger lattice mismatch.
It has been a long standing practice to bridge large lattice constant mismatches with thick monolithic buffer layers. Typically, this involves growing the buffer layer well beyond its critical thickness, allowing relaxation by threading and misfit dislocations, upon which the active film is grown. This has been used extensively over the years, for example, in structures consisting of wide bandgap nitrides, particularly before the relatively recent introduction and availability of affordable, high quality freestanding GaN. For instance, GaN templates grown by hydride vapor phase epitaxy (HVPE) onto Al2O3 or Si are commercially available. As another example, for many years, the cost, availability, and size limitations of CdZnTe substrates led to the extensive study of CdTe and ZnTe buffer layers to bridge the 19% lattice mismatch between Si and HgCdTe.11
Grading schemes,12 blocking layers,13 or other multilayer approaches14 used in many semiconductor materials work to confine the mismatch-related defects and minimize their vertical propagation into the active region. The optoelectronic properties of different semiconductor materials, however, are not equally defect-tolerant.15 III-V semiconductors, in general, are typically not defect-tolerant although the properties of the III-nitrides are widely thought to be more tolerant to dislocations than the III-As or III-Sb alloys.16 Device structures, even made of the same materials, are not equally defect-tolerant, for example, majority carrier devices are usually more tolerant to defects than minority carrier versions. Similarly, the optical properties of MWIR detector structures tolerate a higher dislocation density than LWIR structures, and in both cases, dislocation tolerance decreases with operating temperature.11,17 In any case, the less defect tolerant the material, device structure, or operating condition, the more critical it is to minimize the dislocations threading through the active structure. Although grading, blocking, and multilayer schemes can greatly reduce defect densities, they all leave some degree of residual strain, which could lead to deleterious defects, particularly when very thick layers for device structures need to be grown. These residual strains can affect the bandgap or other optical properties, as previously reported for InAsSb.7
For these reasons, over the last several years, we have developed a VS technology based on the theoretical considerations of Tersoff.18 As described in the previously mentioned references, we linearly grade AlGaInSb or GaInSb from the lattice constant of GaSb to a higher target value, typically ∼2.5% larger. The graded layer nearly completely relaxes throughout most of its thickness. The dislocations propagate through only the bottom ∼3/4 of the grade, leaving the top portion unrelaxed and tetragonally distorted. This results in a small residual strain. Onto this, we grow an AlInSb or GaInSb VS with a bulk lattice constant equal to the in-plane lattice constant at the top of the grade, which negates the residual strain. InAsSb is grown lattice matched to the VS. Because it is unstrained, unrelaxed, and not ordered, InAsSb exhibits inherent properties and, specifically, inherent bandgap.7
There are practical limits to this technique. A ∼2.5% increase in the lattice constant from GaSb, typically needed to reach the lattice constants of InAsSb corresponding to the LWIR material, requires a grade thickness close to 3 μm. While this is feasible, if we were to grow this structure on GaAs, with the same grade rate, we would need the VS to have a lattice constant that is ∼10%–11% larger than that of GaAs. This would correspond to a ∼10 μm thick grade, which represents a substantial MBE machine time cost. Furthermore, it would not be possible to grade continuously while changing only the group III elements. One possible approach would involve grading the group V elements from GaAs to GaSb, followed by a group III grading of Ga(In)Sb. This is unlikely to work in one growth run since one cannot quickly remove As from the chamber, and therefore, Ga(In)Sb would likely incorporate As, which would affect the lattice constant in an uncontrolled manner. Another approach could involve group III grading GaAs to InAs and then group V grading InAsSb. This would also be difficult to do in a controlled way, particularly after the As valve and shutter are open for a few hours, at flux levels that would be different from what is needed for InAsSb growth. Although these methods may not be impossible, the difficulties related to controlling the mixed group V elements and the sheer length of time needed for the growth make them impractical.
A different approach for managing large lattice mismatches is the interfacial misfit (IMF) dislocation array approach. Interfaces between certain binary semiconductors, such as GaAs and GaSb,19–21 induce laterally propagating 90° misfit dislocations confined to the interface between the two layers. The 90° misfit dislocation has a higher energy of formation than and relieves twice as much strain as the more common 60° misfit dislocation, which typically forms at interfaces having a lattice mismatch below 2%. The 90° misfit dislocations generally occur at highly mismatched (greater than 6%) interfaces and in combination with 60° misfit dislocations for mismatches of ∼3%–4%.22 These ranges are possibilities,23 not strict boundaries, since IMF arrays have been reported for mismatches as small as 0.4% for GaP/Si24 and as large as 13% for AlSb/Si.25 The preference for 60° or 90° dislocation formation also depends on the growth temperature.26,27 It is worth pointing out that “perfect” IMF arrays consisting only of 90° misfit dislocations do not exist. Some 60° dislocations always occur. The glide of 90° misfit dislocations is insufficient to completely relax the film.
The mismatch between the InAsSb alloys having compositions of interest for LWIR and GaAs is 10%–11%, which falls within the mismatch range where IMF arrays can be grown. However, for the IMF array to relieve strain effectively, without producing coincidental threading defects, the interface must be very abrupt with nearly all strain mitigated within the first few monolayers of growth. To our knowledge, there are no reports of IMF arrays formed between binary and ternary alloys.
MWIR InAsSb on GaSb/GaAs IMF arrays was studied by Craig et al., who grew two different InAsSb structures, including mismatched InAs0.79Sb0.21 absorbers within nBn structures, and more closely lattice matched InAs0.87Sb0.13 p-i-n structures.28 Our approach combines the IMF and the metamorphic graded buffer/VS approach to bridge the much larger lattice mismatches required for growing InAsSb with compositions required for LWIR absorption.
Our intention was not to produce an optimized IMF between GaSb and GaAs, which requires considerable care and practice. Part of this experiment was to determine if the second phase of strain remediation, the lattice shifting buffer layer, could also function as a dislocation-blocking layer and remediate a moderate density of threading dislocations arising from the IMF array. In order to study the defect morphology in transmission electron microscopy (TEM) and have meaningful statistics despite the field of view limitations, there must be a sufficient density (exceeding 106 cm−2). Optimization of the IMF layer was therefore not needed and actually undesirable for this particular experiment.
II. EXPERIMENTAL METHODS
The material was grown in a modular Gen II system using solid sources. As and Sb were delivered using valved cracker sources. The substrate temperature was measured using a K-space BandiT system operating in both transmission band-edge (GaAs) and pyrometry modes (GaSb). Conventional GaAs MBE practices were used to grow a ∼0.5 μm thick GaAs buffer layer at 600 °C, which was then left in the growth chamber overnight to reduce the background pressure of As, which is known to disturb the formation of IMFs due to the formation of an unintended, rough interface layer.19,20 After an initial preheat at ∼550 °C, we then lowered the substrate temperature to 490 °C for the deposition of GaSb.
Specific problems arise while measuring the substrate temperature on structures with combinations of materials with large differences in the bandgap. For GaAs, the BandiT system can use the light from the substrate heater and monitor the spectral position of the transmission edge, which shifts due to the temperature dependence of the bandgap—ideally this relationship is calibrated for each type of substrate (characterized by the thickness, number of polished sides, doping, etc.). For a material with a narrower bandgap, such as GaSb, the spectrometer in our instrument does not cover the transmission edge and we reprogram the BandiT system and use it as a wide-spectral range pyrometer. To use it in this mode, we have calibrated the instrument through the viewport using a blackbody source. In principle, it should therefore be possible to use the transmission mode for GaAs and the pyrometry mode for thick overlayers of GaSb. Unfortunately, there is a wide intermediate GaSb thickness range in which the overlayer disturbs the GaAs transmission reading and the transmission interferes with the pyrometer reading.
It is important to notice that the thickness of the GaSb buffer layer in this experiment was not chosen to further reduce the number of threading dislocations but was solely based on the desire to enable substrate control in the pyrometry mode. To minimize the overall growth time, we arbitrarily decided to terminate the GaSb deposition at ∼2 μm when it was possible to operate the BandiT system in the pyrometry mode, which is our standard procedure for the growth of InAsSb onto VS/graded layers. At this point, we initiated a previously developed GaSb-based VS and InAsSb growth recipe using a 415 °C temperature for the narrow-bandgap material. Figure 1 shows a schematic of the structure, with thicknesses measured by transmission electron microscopy (TEM). It is likely that the thickness of the GaSb buffer layer was inadequate and that some substrate heater radiation reached the pyrometer and skewed the temperature reading which will be further discussed.
The compositions and strain were analyzed by high-resolution x-ray diffraction (XRD) using a Panalytical MRD-X'pert Pro. Four triple axis scans were collected for symmetric (004) and asymmetric (115) reflections with a 180 degree rotation in phi. Asymmetric reciprocal space maps were also collected with a linear detector array. The structure and defect morphology were examined using cross-sectional TEM. The samples were prepared by tripod polishing and Argon ion milling to electron transparency and then examined using a JEOL 2010F TEM operating at 200 keV. Photoluminescence (PL) was excited by a 0.98 μm diode laser. The excitation area was 1 mm2. The PL spectra were measured using a Nicolet Magna-860 Fourier transform infrared (FTIR) spectrometer, operating in the continuous-scan mode with a 14-μm cut-off wavelength external HgCdTe photodetector.
III. RESULTS AND DISCUSSION
Figure 1 shows the compositions of the layers as measured by XRD. The symmetric (004) triple axis XRD spectra are shown in Fig. 2. The position of InAsSb is slightly to the right of, rather than coincident with, the position of the VS peak. Through analysis of the symmetric and asymmetric scans, we find that the InAsSb composition is 46% Sb, which is just short (by less than 1%-point) of the target composition, and therefore there is a tensile mismatch of ∼0.24% with the underlying VS. It is worth pointing out that previous studies have shown that the composition of InAsSb is very sensitive to the substrate temperature, which may have been slightly different on the GaSb/GaAs structure from for our typical growth processes on GaSb substrates.29 The Matthews–Blakeslee prediction of the critical thickness of InAs0.54Sb0.46 on the In0.35Al0.65Sb virtual substrate is roughly 650 Å. The XRD indicates that InAsSb is completely relaxed relative to the VS. The (115) reciprocal space map that includes the reflections from all the layers between the GaSb buffer and the InAsSb is shown in Fig. 3. The dashed line corresponds to 100% relaxation, and the vertical line is the pseudomorphic growth line for the InAsSb relative to the VS.
Figure 4 shows a high-resolution TEM image of the GaSb/GaAs interface along the zone axis, with a Burger's circuit drawn around one of the dislocation cores. The negative for this image has a magnification of 500 k. The diffraction pattern for this structure (not shown here) indicates no tilting of the planes across the interface, which is in agreement with the XRD measurements. We measure an average dislocation spacing of 57 Å, which agrees well with the value of 56 Å measured by Huang et al.19 This corresponds to 96.2% relaxation of GaSb. A theoretically perfect IMF array between these two alloys would have 13 GaSb sites corresponding to 14 GaAs sites. As also observed by Huang, this agrees with the average values collected from several TEM images collected along the interface for this sample although there are locally small variations. Even with a perfect IMF array, a mismatch of 0.13% at the interface remains, corresponding to a maximum GaSb relaxation of 98.3%. Therefore, some residual strain remains in our GaSb buffer, as it does even in the most perfectly grown IMF arrays. Considering the thickness of the device structures grown on top of the GaSb, to minimize threading dislocations, we must take in to account the effect of this residual strain and remediate it.
It is worth pointing out that a literature survey shows that most of the TEM images included in discussions of IMF arrays are typically at a fairly high resolution, either to show the Burger's circuits or to show the “dotted line” appearance of the dislocation cores so that the average spacing can be measured at the GaSb/GaAs interface. These resolutions, however, are not reasonable for demonstrating that there are no threading defects throughout the structure. The images shown in much of the literature could still have dislocation densities in ∼upper 108 cm−2 and have a “clean” area of the size shown in many of the papers. For this reason, some of the studies included estimates from etch pit densities observed using SEM, which samples a much bigger area and is useful for qualitatively comparing samples. However, since certain dislocations do not readily form etch pits at the surface, dislocation densities obtained from etch pit densities are notoriously lower than those found in TEM for the same sample.30 Hong et al. determined that this discrepancy could be as much as 2–4 orders of magnitude for some materials.31 One notable study by Tan et al.32 does include low-resolution TEM images (20 k magnification) of IMF arrays grown at different substrate temperatures. They report a defect density of 3 × 108–2 × 109 cm−2 for GaSb/GaAs arrays grown at non-ideal temperatures and a defect density of ∼1 × 108 cm−2 for an optimized temperature of 450 °C.
The negative for the lower resolution image in Fig. 5 has a magnification of 25 k. This area includes the same region seen in Fig. 4 and clearly shows threading defects propagating into the GaSb buffer layer. A rough estimate of the dislocation density corresponds to a low value of 109 cm−2, which is very similar to the densities observed in comparable low-resolution TEM images for non-optimized IMF arrays in the work by Tan et al.32
Figure 6(a) shows a cross-section of the region from the GaAs buffer layer to the bottom of the lattice-shifting grade. Figure 6(b) shows the sample from the top part of the GaSb buffer to the InAsSb surface. The wavy features at the top of the InAsSb surface [Fig. 6(b)] are thickness fringes. We cannot show the entire structure from the GaAs substrate to the InAsSb surface in one image due to the inherent limitations of TEM related to the sizes of the condenser and objective lens apertures. It is also very difficult to get TEM samples to be electron transparent over such a big area without thickness variations inducing distracting fringes or without having the sample starting to curl or bend in the presence of the electron beam. The negative's magnification was 10 000×, which is an extremely low magnification for a TEM.
XRD and TEM indicate that the lattice-shifting grade is behaving in the same manner as seen in our previously published structures grown on GaSb substrates.3–6 The top, mostly unrelaxed portion of the grade, is visible just below the grade/VS interface, and the thickness of this region is in line with the predictions by Tersoff.18
Defect densities were averaged over many images. The density of defects in the VS is approximately 108 cm−2, which is high, but it is a full order of magnitude below the defect density in the GaAs buffer layer arising from the GaAs/GaSb IMF array. The grade effectively increases the lattice constant from GaSb to that of the target InAsSb without generating additional dislocations in the active region, and it reduces the dislocations propagating from the substrate. Huang et al.19 report defect densities in the high order of 105 cm−2 for GaSb/GaAs grown under optimized conditions determined after extensive experiments, including specific substrate temperatures, and Sb soaking before GaSb growth initiation. With the caveats mentioned before related to reported defect densities, it should be possible to produce GaSb buffer layers with significantly lower threading dislocation densities than (deliberately) used in this study, which can then be further reduced by the VS structure. This enables the growth of more complicated non-binary, high quality, materials having very high lattice mismatches with the starting substrate.
A high-resolution image of the InAsSb absorption layer along the zone axis is shown in Fig. 7. There are no defects or strange contrast features seen at these magnifications.
The diffraction pattern along the zone axis is shown in Fig. 8(a), with an intensity profile (8b) parallel to a set of spots along the direction through the spot. We did not evaluate the intensity profile through the transmitted beam since it would overwhelm the contrast. The profile shows the presence of CuPt ordering (denoted by the small intensity spikes labeled with arrows in the profile).
It is not surprising that there is ordering in the InAsSb layer. The defect density of the InAsSb layer was slightly higher than that of the VS but was within the error bars of the density measurement. The small 0.24% mismatch between the VS and the InAsSb drives the ordering, which is consistent with our previous observations that slightly mismatched samples present ordering.7,33
The PL results are shown in Fig. 9. The spectra were measured over a range of excitation powers with a density as low as 5 W/cm2 [Fig. 9(a)]. A blue shift of the PL maximum with excitation can be attributed to band filling [Fig. 9(b)]. A superlinear dependence of the radiative recombination rate with excitation was observed, as shown in Fig. 9(c). Accounting for a thermal broadening of 2 kT ∼ 13 meV, we can estimate the bandgap to be 0.1 eV. The first comment that can be made is that the luminescence yield is very strong. In fact, this sample showed higher PL intensity than any other structures grown on GaSb substrates. The maximum of the PL peak shifts with increasing, but relatively low, pump intensities, which indicates that band filling is enabled by an unusually long minority carrier life time. Quantification of this and further details will be the subject of future studies.
The higher optical quality is a very intriguing observation for several reasons. The sample did not contain any additional layers compared to structures grown directly on GaSb substrates. The only significant difference in this experiment compared to other growth processes is the substrate temperature. It is clear that it resulted in a change of the As/Sb incorporation and therefore a slight lattice mismatch to the VS. This in turn induced CuPt ordering. However, we have not seen evidence for enhanced PL due specifically to ordering before. The most likely explanation, pending further investigations, is therefore that the substrate temperature was closer to the optimum than what we have been using before. We have previously pointed out that no quality optimization has been done to the LWIR InAsSb materials.34 These observations imply that optical parameters such as minority carrier lifetimes can significantly improve over the 185 ns result that we have obtained so far.8 Lifetimes in the micro-second range have been reported for InAsSb lattice matched to GaSb35 and might therefore also be possible for the LWIR material, which would allow it to truly rival HgCdTe as a LWIR detector material.
The growth of materials with a different lattice constant from the substrate spawned much development work over decades, as mentioned in the introduction. Typically, this was due to a desire to replace a costly substrate that was only available at smaller diameters with a larger and more inexpensive alternative (for example, InGaAs on GaAs instead of InP and GaAs on Si). In such a scenario, the substrate cost-savings must be larger than the extra epi-reactor time inevitably incurred by any mismatch accommodation scheme. The material quality deterioration, which usually follows, must also be considered. However, a larger wafer yields more devices, which drives down the unit cost. To make the economic considerations more complicated, the target usually moves since the size and the cost of the more exotic substrate typically improve over time. A similar development is currently taking place with GaSb substrates.
In our case, however, the situation is somewhat different in that there is no binary substrate available for LWIR InAsSb. Lattice mismatch accommodation must therefore inevitably be part of the production cost. IR detector arrays are some of the largest-area devices in production, meaning that the wafer diameter substantially affects fabrication costs. In addition, the advantage in mechanical strength exhibited by GaAs over GaSb would influence the fabrication yield in a potential future InAsSb fabrication process. Typically, LWIR absorber structures are also thick to begin with, making significant demands on epi-reactor time. In this scenario, it is quite reasonable to consider growth on GaAs since the extra growth time required will be a smaller fraction of the entire process. The extra growth time added in the process used here is represented by the GaSb thickness (and overhead time associated with the GaAs preparation). We used an excessive thickness of GaSb to allow pyrometer-control. In a standardized process, the layer could be thinner and the temperature control during the transition could be done via optimized thermo-couple control.
IV. SUMMARY
We have demonstrated narrow bandgap (0.1 eV) bulk InAs0.54Sb0.46 on GaAs with very pronounced photoluminescence. A combination of an IMF-array and the VS approach allows the growth of InAsSb onto GaAs with its inherent bandgap. We routinely grow InAsSb films with Sb compositions in the range of 40%–60% onto VS/grades on GaSb substrates that have dislocation densities below ∼1 × 106 cm−2, which is below the level that can be detected by TEM. This study shows that the grade/VS further reduces threading dislocations remaining from a non-ideal IMF array by an order of magnitude. Therefore, it is reasonable to expect that InAsSb films grown on GaAs substrates with an optimized IMF/VS approach will have the same defect densities as typical for InAsSb grown on GaSb. The very high PL yield implies that growth condition optimization of LWIR InAsSb can significantly improve the optical quality of the material.
ACKNOWLEDGMENTS
The work performed at Stony Brook University was supported by the U.S. Army Research Office Grant No. W911NF-16-2-0053.