Dependence of growth temperature on the electrical properties and microstructure of MBE-grown AlN/GaN resonant tunneling diodes on sapphire

Interest in resonant tunnel diodes (RTDs) has recently been rekindled following reports of >1 THz oscillations in these devices,^1–3 after nearly two decades of relatively little activity after reports of >700 GHz oscillations in InAs-based RTDs.⁴ Recent demonstrations of significant power output (>1 mW) from a double-barrier RTD will likely sharpen this interest.⁵ While these notable results have come from III–V materials, the larger band offsets and higher thermal conductivity to be found in III-nitrides suggest the potential for even higher output power and power density. Repeatable, room temperature, stable, hysteresis-free negative differential resistance (NDR) was observed in III-nitride RTDs for the first time in 2016, from AlN/GaN double-barrier RTDs grown on low dislocation-density, freestanding GaN substrates.^6,7 While the peak current density, J_P, and peak-to-valley current ratios (PVCRs) of these early GaN-based RTDs were typically low, improved device designs quickly yielded increases in J_P of more than two orders of magnitude with little to no degradation of PVCR.^8,9

While it appeared that low dislocation-density GaN substrates were the key enablers of these advances, Wang et al. demonstrated in 2018 that GaN-based devices grown on GaN-templated sapphire also exhibited repeatable room temperature NDR.¹⁰ Growden et al. have since reported a record-high peak current density, approaching 1 MA/cm², from an AlN/GaN double-barrier RTD grown on a GaN template on sapphire.¹¹ 

Little is so far known about the effect of growth conditions on the electrical properties of GaN-based RTDs. In this work, we report on the effect of growth temperature on the electrical and structural properties of AlN/GaN double-barrier RTDs grown by rf-plasma assisted molecular beam epitaxy (rf-MBE) on GaN templates grown by metalorganic chemical vapor deposition (MOCVD) on sapphire substrates.

Three samples were grown by rf-MBE in a Scienta Omicron PRO-75 MBE deposition chamber equipped with a Veeco Uni-Bulb plasma source for supplying active nitrogen (N*), dual-filament effusion cells for evaporation of elemental Ga and Al, a single-filament medium high temperature cell for Si, and reflection-high-energy electron diffraction.

The layer structure of each sample was identical to that used previously^8,11 and is shown in Fig. 1. The structures were grown on 18 × 18 mm² squares diced from a single MOCVD-grown GaN template on (0001)-oriented (i.e., c-plane) sapphire. The GaN template consisted of a 2 μm-thick Fe-doped layer followed by a 2 μm-thick unintentionally doped layer. The threading dislocation density (TDD) of the MOCVD GaN layer was estimated to be ∼3 × 10⁸ cm⁻² by atomic force microscopy (AFM). The template surface was covered with a protective layer of photoresist to protect it during dicing. Immediately prior to loading each square into the MBE chamber, the protective coating of the photoresist was stripped using organic solvents (acetone, methanol, and isopropanol), and the GaN template was subjected to a final ex situ wet chemical clean including a solvent degrease and an “SC1” (1:1:5 NH₄OH:H₂O₂:H₂O) clean. After cleaning, the wafer was placed onto an indium-free molybdenum wafer mount and loaded into the ultrahigh-vacuum (UHV) system. The same wafer mount was used for all samples to minimize unintentional thermal differences between growths. Each wafer was degassed for 30 min at 700 °C in the analysis chamber prior to transfer into the deposition chamber for growth.

Each sample was grown continuously, without interrupts. The growth temperature for each sample was kept constant during growth but varied between samples. Samples A, B, and C were grown at 760, 810, and 860 °C, respectively, as measured by a thermocouple mounted behind the substrate. We have separately observed rapid desorption of metallic Ga from a GaN surface to occur at ∼900 °C. The growth was initiated by first exposing the substrate wafer to the N* plasma for 2 min, after which Ga and Si shutters were opened simultaneously. The gallium shutter was kept open for the entire growth, including during the growth of the AlN barriers. The nitrogen rf plasma power and flow were maintained at 275 W, 0.80 SCCM, and the nominal flux ratio of the gallium to active nitrogen was estimated to be ∼1.3, while the ratio of the Al to active nitrogen flux during growth of the AlN barriers was nominally 1:1. Due to the well-known preferential incorporation of Al over Ga, the composition of the barriers was expected to be AlN despite the presence of excess Ga.¹² 

The RHEED patterns of samples A and B were characterized by streaks throughout the growth, indicative of epitaxial growth and a smooth surface, except for the initial ∼1 min after opening the Ga and Si shutters, during which the RHEED streaks were modulated by spots, suggesting a slight roughening of the surface. The spot modulation quickly faded, and the RHEED pattern exhibited only streaks for the rest of the growth. No spot modulation was observed from sample C until after the growth of the AlN barriers. The spot modulation gradually increased in intensity until the end of the growth, but at no point did the streaks disappear.

The as-grown samples were characterized by optical microscopy, AFM, x-ray diffraction (XRD), and cross-sectional transmission electron microscopy (TEM). Approximately 400 resonant tunnel diodes were fabricated on each sample as described in detail elsewhere.¹¹ The nominal areas of the device mesas were 3 × 4 , 4 × 5 , 7 × 10 , or 8 × 12 μm². Current–voltage (I–V) curves were measured at room temperature from nearly every device.

Gallium droplets, indicative of excess gallium during growth, were observed by optical microscopy at the corners of sample A, grown at 760 °C, but not at the center. No droplets were observed on samples B and C, which were grown at higher temperatures. We assume that any excess gallium re-evaporated from samples B and C, as well as from the droplet-free region of sample A during growth. The droplets on sample A were removed by dipping the as-grown sample in HCl at 60 °C for 5 min prior to further characterization and device fabrication.

AFM was performed at three locations on each sample, and little variation in the surface morphology between locations was observed. The results revealed a clear trend to rougher surface morphology as the growth temperature was increased, as shown in Fig. 2. Each AFM image represents a 20 × 20 μm² field of view near the sample center. Sample A [Fig. 2(a)], grown at 760 °C, exhibited a z-range and a root-mean-square (RMS) roughness of 12.2 and 1.77 nm, respectively. Samples B and C, grown at 810 and 860 °C, respectively, exhibited z-range (RMS roughness) values of 41.7 nm (4.14 nm) and 75.6 nm (7.17 nm), respectively. Atomic steps were observed in the 5 × 5 μm² images from all three locations on sample A, but no atomic steps were seen on either sample B or sample C.

Symmetric ω–2θ XRD patterns taken about the GaN (0002) reflection revealed broad envelopes around the central GaN diffraction peak of each sample; a representative pattern is shown in Fig. 3(a). Dynamical simulations¹³ for the intended structure predicted the existence of Pendellösung fringes, but these were not observed in the experimental data. However, a similar series of devices grown on freestanding GaN substrates exhibited clear evidence of Pendellösung fringes [Fig. 3(b)], from which the thicknesses of the upper GaN layers, and hence the GaN growth rate, could be precisely determined.¹⁴ It was possible from that series to determine the GaN growth rate, which was very nearly 3.0 nm/min. A trend toward a reduced growth rate with increasing temperature was detected. However, the reduction in growth rate was less than ∼3% over the full range of growth temperature, which was from 760 to 900 °C (thermocouple) for that series.

Cross-sectional TEM images revealed higher densities of threading dislocations in the MBE-grown layers of each sample than in the MOCVD-grown GaN template (Fig. 4). Differences in the top GaN layer thicknesses from their nominal value are ascribed to local differences in etching during device fabrication. The regrowth interface between the MOCVD-grown template and the MBE-grown GaN is clearly visible in each image, and threading dislocations are easily seen to be generated at that interface. In addition, sample C exhibited a large number of “V-defects” in the active region, near the AlN barrier layers (Fig. 5). These can be seen in Fig. 4(c) and, at higher magnification, in Fig. 5(a). However, the AlN barrier layers and GaN quantum well were well-defined in regions well removed from threading dislocations [Fig. 5(b)]. The TDD decreased as the growth temperature increased. While the TDD of the samples was not measured directly, it was estimated from the transmission electron micrographs to be ∼10¹⁰ cm⁻² for sample A, ∼10⁹ for sample C, and intermediate between these values for sample B. Furthermore, high-resolution images of the regrowth interfaces between the rf-MBE-grown GaN:Si buffers and the MOCVD-grown substrates revealed disordered interfacial layers ∼5–10 nm thick in sample A, ∼2 nm thick in sample B, and ∼1 nm thick in sample C. This behavior is strikingly different from homoepitaxial GaN growth by MBE on freestanding hydride vapor-phase epitaxy (HVPE)-grown GaN substrates,^15,16 for which it is relatively straightforward to suppress both the formation of a visible regrowth interface and the generation of new threading dislocations. The ex situ “SC1” wet chemical clean by itself is evidently insufficient to properly prepare the surface of the GaN template for growth. In contrast, we have observed that the “SC1” clean of a separate ammonothermally grown GaN wafer, in conjunction with in situ Ga deposition and thermal desorption as described in Ref. 15, resulted in homoepitaxial GaN layer growth with no new TDs and no visible regrowth interface (Fig. 6). Thus, the use of an in situ cleaning procedure in addition to an ex situ wet chemical clean is highly recommended for future device growth and fabrication purposes.

The exact cause of the defective MBE-grown layers is not known. However, Xie et al. reported the growth of GaN layers by MBE on an MOCVD-grown template in which no new threading dislocations were generated at the regrowth interface between the MBE- and MOCVD-grown GaN layers, and TEM micrographs of the interfacial region revealed no visible regrowth interface. Prior to growth, the template had been cleaned with boiling aqua regia followed by boiling KOH and then annealed in UHV at 800 °C. In contrast, when they examined a similar layer grown on a GaN template in which boiling KOH was not used, and the template was annealed at 650 instead of 800 °C, the regrowth interface was clearly visible, though no new threading dislocations were generated.¹⁷ In our experience, oxygen is the most significant impurity at the regrowth interface between rf-MBE-grown homoepitaxial GaN and an HVPE-grown GaN substrate, as determined by secondary ion mass spectroscopy. We have found that we can reliably grow GaN by rf-MBE on HVPE-grown GaN substrates without generating new threading dislocations or producing a visible regrowth interface if there are fewer than 0.25–0.50 monolayers of oxygen on the surface.¹⁶ There are numerous means to remove this oxygen, including ex situ wet chemical clean and high temperature anneal prior to growth, consistent with Xie’s result. We speculate that the surface of our template contained residual impurities such as oxygen, which resulted in visible layers at the regrowth interface, seen as spot-modulated RHEED patterns at the beginning of growth, and subsequent degradation of the crystal quality, but that higher growth temperature led to partial desorption of these impurities, resulting in thinner interfacial layers, streakier RHEED patterns at the start of growth, and lower dislocation density.

The surface morphologies observed on samples A–C by AFM are similar to those seen previously from nearly identical samples grown on freestanding, low dislocation-density GaN substrates.¹⁸ Surface pits and atomic steps appeared simultaneously in the images presented in the earlier report, with no apparent correlation between the step edges and the size or location of the pits, suggesting that the pits formed after the conclusion of growth.¹⁸ We speculate that the surface pits of sample B also form due to thermal decomposition of the surface after the completion of growth, occurring immediately after cessation of the elemental fluxes but before the sample temperature had significantly decreased. Such a process is not inconsistent with the observations of Koblmüller et al., who reported a correlation between higher densities of surface pits on GaN and reduced equilibrium adlayer coverages of gallium during growth.¹⁹ A similar process may also be at work in sample C, although a significant number of surface pits are correlated to V-defects, which appear to form at the AlN barriers and propagate to the surface, indicating they form during growth rather than after it.

The surface roughness and dislocation density of samples A–C are anticorrelated, suggesting a common, underlying cause. As noted above, higher growth temperature is correlated with thinner interfacial layers and lower dislocation density, and it plausibly leads to reduced Ga adlayer coverage. Thus, increased growth temperature can be seen to reduce the density of extended defects and increase surface pitting. Fortunately, the increased density of surface pits does not appear to have a deleterious effect on the electrical properties of the devices.¹⁸ 

Representative room temperature J–V curves from each of the samples are shown in Fig. 7. Negative differential resistance was observed on each of the samples. However, the PVCRs were typically low, between 1.01 and 1.11, as was the fraction of devices of devices exhibiting NDR. In general, as the growth temperature increased, the peak current density, J_P, also increased, as shown in Fig. 7. The peak current density is highly sensitive to barrier thickness.¹¹ 

Hence, this increase is likely due to the unintentional reduction in the effective AlN barrier thickness as the growth temperature increases.

Approximately 10% of the devices on sample A exhibited room temperature NDR. Of these, ∼90% exhibited PVCRs between 1.01 and 1.03, while the PVCRs of the remainder (1% of the total) were ∼1.1, with peak current densities J_P∼ 15–21 kA/cm². On sample B, only three devices (about 1% of the total) exhibited room temperature NDR. The peak current density of the three devices on sample B ranged from 70 to 165 kA/cm², with PVCRs between 1.003 and 1.04. Repeatable room temperature NDR was observed in 12 devices (∼3%) on sample C, with PVCRs from 1.02 to 1.1, and J_P from 630 to 930 kA/cm². However, NDR was only observed in the smallest devices (nominally 3 × 4 μm²) of this sample, likely due to strong self-heating effects in these devices.

While the device yields and PVCRs were low, these results indicate that the presence of threading dislocations, even at high density, does not preclude NDR. Similar behavior was observed in InAs/AlSb RTDs, which exhibited current densities of ∼370 kA/cm² and PVCRs of 3.2 despite a TDD of ∼10⁹ cm⁻² in the devices.^20,21 Other factors such as barrier uniformity, barrier layer thickness, or the abruptness of the interfaces between the AlN barriers and the GaN quantum wells could play a larger role in determining device performance. While the shortest and simplest path to realizing III-nitride RTDs appears to involve epitaxial growth on low dislocation-density, freestanding GaN substrates, improvements in growth on GaN templates may well enable a lower-cost alternative.

We have investigated the effect of growth temperature on the electrical and structural properties of rf-plasma MBE-grown AlN/GaN double-barrier RTDs on MOCVD-grown GaN templates on sapphire. Three growth temperatures in the range from 760 to 860 °C were investigated. AFM revealed increased surface roughness as the growth temperature increased. XRD patterns were consistent with the intended structure but lacked sufficient detail for precise determination of any layer thicknesses. X-ray diffractometry on a similar series grown on freestanding GaN did exhibit Pendellösung fringes, enabling the growth rate to be determined. The GaN growth rate in this series was observed to decrease by ∼3% as the growth temperature increased from 760 to 900 °C. Negative differential resistance was observed on each sample at room temperature. However, the fraction of devices exhibiting room temperature NDR ranged from 1% to ∼10%, while the peak-to-valley current ratio ranged from 1.003 to about 1.1. Peak current density, J_P, increased with increasing growth temperature, from ∼20 to ∼930 kA/cm² as the growth temperature increased from 760 to 860 °C, likely due to the unintentional reduction in AlN barrier thickness as the growth temperature increased. TEM revealed higher densities of threading dislocations in all MBE layers compared to the MOCVD GaN template and a clear trend toward lower dislocation density as growth temperature increased. We speculate that this result is due to a reduction in the residual impurity concentration on the MOCVD GaN surface at the beginning of growth as the growth temperature is increased. This result also indicates that the presence of threading dislocations, even at high densities, does not preclude NDR in AlN/GaN RTDs.

This work was supported by the Office of Naval Research and the Under Secretary of Defense for Research and Engineering. DISTRIBUTION STATEMENT A. Approved for public release. Distribution is unlimited. This material is based on work supported by the Under Secretary of Defense for Research and Engineering under Air Force Contract No. FA8702-15-D-0001. Any opinions, findings, conclusions, or recommendations expressed in this material are those of the author(s) and do not necessarily reflect the views of the Under Secretary of Defense for Research and Engineering. The use of facilities in the John M. Cowley Center for High Resolution Electron Microscopy at Arizona State University is also acknowledged.