Karl D. Hobart

The roles of dislocation motion and interdiffusion on the interfacial morphology and thermal properties of the Ge–Si interface were elucidated using transmission electron microscopy, energy dispersive x-ray spectroscopy, and thermoreflectance measurements. Epitaxial Ge thin films deposited on Si (001) substrates by molecular beam epitaxy were annealed from 400 to 800 °C both in situ and ex situ in nitrogen to modify the interface morphology. Multiple transmission electron microscopy techniques monitored the evolution of defects and diffusion before and after different annealing treatments, revealing a strong correlation between dislocation motion and interdiffusion at the interface, producing a corrugated interface morphology. The characterization supports the fact that dislocations affect the interface structure by serving as vacancy sinks during their reconfiguration at elevated temperatures. Directed vacancy fluxes then enhance interdiffusion near dislocation cores, resulting in the periodic corrugation of the interface. Thermal measurements were performed to extract the thermal boundary resistance from the resulting interface morphologies produced by each annealing condition. Annealing resulted in increased thermal boundary resistance at the Ge–Si interface, largely attributed to alloy scattering becoming dominant as interdiffusion proceeds.

Rapid wafer-scale defect identification and quantification are essential to achieve high yielding and reliable materials for device fabrication. Since defects are microscopic and the measurement scales are macroscopic, a quick, non-destructive automated process is needed to assess wafer quality. Although methods for wafer-scale defect visualization are well established, the appearance of defects can vary in size, shape, intensity, and signal to noise across a single sample. Using a traditional pattern matching technique would require extensive optimization, and if blind spots in the method were observed, corrections may be labor intensive to implement. This paper discusses using machine learning models to detect and classify defects in SiC and demonstrates that an implementation of the U-Net architecture is effective at identifying four types of critical defects in SiC with ultraviolet photoluminescence: stacking faults, polytype inclusions, threading screw/mixed dislocations, and basal plane dislocations. The resulting model can evaluate both 6-inch and 8-inch SiC wafers in minutes.

Aluminum nitride (AlN) is well suited to address evolving needs of future power systems for increased density, efficiency, and voltage handling. Eventual commercialization of power devices requires maturation of substrate technology to ensure fabrication of consistent devices impacted by the spatial distribution of defects and substrate inhomogeneities. Commercial 50 mm AlN substrates grown by physical vapor transport were characterized as received and after 1 μm of homoepitaxial growth by metalorganic chemical vapor deposition. Optical profilometry determined macro-roughness to be sub-nm while atomic force microscopy reveals consistent step edges and 0.1 nm micro-roughness, despite observation of sporadic 100–400 nm wide, ∼1 nm deep depressions. Sub-bandgap photoluminescence reveals broad spectroscopic signals, which slowly vary spatially across the wafer, corresponding with faint visible coloration. Raman spectroscopy indicates uniform, high crystal quality but shows increased background scattering associated with micrometer-scale features visible in optical microscopy. High-resolution x-ray diffraction validates the crystal quality with median (0002) full-width at half maximum of 17.6 arc sec and x-ray topography identified varying regions with dislocation density from ∼102 cm−2 to ∼mid-105 cm−2. Electron microscopy revealed nano-pipes that may be the origin of the nanoscale depressions and that at least some bright optical features are aluminum platelets with coherent atomic boundaries. After epitaxial growth, surface morphology retained smooth macro- and micro-roughness with clear step edges, demonstrating viability of device-relevant epitaxial layer thicknesses for power devices. Three wafers from one boule were characterized, indicating consistent wafer-to-wafer characteristics, critical for future device commercialization.

In this work, we demonstrate device fabrication and characterization of ultrawide bandgap (UWBG) β-(AlxGa1−x)2O3 channel metal–semiconductor field-effect transistors (MESFETs) with Si-implanted source/drain contacts. Films of Si-doped β-(AlxGa1−x)2O3 and unintentionally doped (UID) β-Ga2O3 were grown on an Fe-doped (010) β-Ga2O3 substrate using close-injection showerhead metal–organic chemical vapor deposition (CIS-MOCVD). The Al concentration (x) of the β-(AlxGa1−x)2O3 film was estimated to be ≈21% using x-ray diffraction (XRD), and the Si doping concentration of the n-type β-(Al0.21Ga0.79)2O3 film was measured to be 2 × 1018 cm−3 using secondary ion mass spectroscopy (SIMS). Both gate-recessed and non-recessed MESFET structures were fabricated and had on/off ratios of ≈105. The gate-recessed MESFETs had a lower saturation drain current (15 mA/mm) than the non-recessed structures (26 mA/mm), but they also had a slightly lower off-state leakage current. The gate-recessed structures also enabled better modulation of the channel conductivity, which led to a positive threshold voltage shift of +8V compared with the non-recessed structures. The maximum breakdown voltages of 730 and 858 V were measured for the gate-recessed and non-recessed MESFETs, respectively. Because the thermal conductivity of the disordered alloy β-(Al0.21Ga0.79)2O3 channel is expected to be much lower than β-Ga2O3, thermoreflectance imaging was used to assess the device-level thermal performance. At a DC power density of 0.58 W/mm (VGS = 0 V), an area-averaged gate temperature rise of 54 K was measured. This work demonstrates the potential of leveraging the large-area, low-cost, high-quality β-Ga2O3 substrate platform to develop next-generation power electronics using UWBG β-(AlxGa1−x)2O3 as the active semiconductor.

We report the heteroepitaxial growth of spinel Cr2MnO4 thin films on (2¯01) β-Ga2O3 substrates via off-axis RF magnetron sputtering. Structural analysis confirms (111)-oriented epitaxial films with optimized growth at 950 °C. Fabricated Cr2MnO4/β-Ga2O3 vertical diodes exhibit rectifying behavior with high on/off ratios (∼108), turn-on voltages near 0.85 V, and low reverse leakage current. Transmission electron microscopy and atomic force microscopy analyses show the films exhibit a Stranski–Krastanov growth mode, with three-dimensional texturing of potential interest for device performance. Capacitance–voltage measurements reveal intrinsic p-type conductivity with carrier concentration ∼7 × 1017 cm−3. The results suggest that Cr2MnO4 may be a promising candidate for Ga2O3-based bipolar devices.

We report back-end-of-line growth of nanocrystalline diamond (NCD) on ultrawide bandgap (UWBG) high Al content aluminum gallium nitride (AlGaN) channel high electron mobility transistors for thermal management. A thin (∼15 nm) silicon nitride (SiNx) interlayer was deposited to protect the device surface before performing a low temperature (500 °C) NCD growth process in an attempt to protect the gates on these fully fabricated devices. Notably, atomic force microscopy showed that the maximum lateral grain size exceeded 300 nm even though the film thickness was ∼250 nm. Comparing electrical (DC) performance before and after NCD growth, the gate leakage increased by ∼102 after NCD growth. Despite the lower NCD growth temperature, intermixing of the Ni and Au was observed in the Schottky gate metal stack; however, we believe there is another mechanism, possibly hydrogen-related, that is responsible for the measured increase in gate leakage. Regarding thermal management, the device-level thermal resistance (quantified using the average gate temperature rise measured by thermoreflectance imaging) was reduced by 29% through the incorporation of the top-side diamond film. Using time-domain thermoreflectance, the thermal conductivity of the ≈250 nm thick NCD film was measured to be 45 ± 25 W m−1 K−1. This is expected to be at least 5× greater than the thermal conductivity of the thin disordered AlGaN alloy. There could also be a coupled electrothermal component contributing to the reduced temperature rise from electric field spreading and consequent heat spreading. This study demonstrates a promising first step toward device-level thermal management of high power UWBG Al-rich AlGaN devices.

The cracking and local strain relaxation in (010) (AlxGa1−x)2O3 films grown on Ga2O3 substrates are assessed in terms of film composition and thickness. We utilize x-ray diffraction and electron microscopy techniques combined with simulation and modeling to investigate that film cracking on curvature (flatness) has a directly proportional relationship with film thickness and/or aluminum content. Cross section transmission electron microscopy reveals that cracks along both the (001) and (100) cleavage planes penetrate into the substrate. The diffuse scattered intensity observed in reciprocal space maps (RSMs) is directly correlated with the tilt that is introduced due to the change in the deformation conditions near the cracks. While asymmetric RSMs show that the layers are fully strained, the diffuse scattering distribution in reciprocal space can be interpreted to show that the cracking relaxes and locally tilts the lattice ∼350 nm from the crack edges, which is consistent with the larger radius of curvature associated with the films with higher crack densities. For example, a 200 nm (Al0.13Ga0.87)2O3 thick film has an average inter-crack spacing of 3.3 µm, so most of the epitaxial layer is fully strained except near the cracks where it deforms elastically and is consistent with the gallium oxide Poisson ratio. A reciprocal space model was developed, which imports the strain and tilt distributions (based on finite element modeling) to match the features observed in the experimental maps. We also note that previous studies involving (AlxGa1−x)2O3 films may show evidence of cracking as observed in their symmetric and asymmetric RSMs.

Homoepitaxial Si-doped (010) β-Ga2O3 epitaxial layers with thicknesses ranging from 1.2 to 12 μm were grown simultaneously via close-injection showerhead metal-organic chemical vapor deposition on both Czochralski and edge-defined, film-fed grown (010) β-Ga2O3:Fe substrates. Structural characterization was performed via x-ray diffraction [full-width at half maxima of 21–72 arcsec depending on epilayer thickness and substrate, measured from the symmetric (020) reflection], x-ray topography, Raman spectroscopy, cathodoluminescence spectroscopy, and transmission electron microscopy. Surface characterization was performed via atomic force microscope (AFM) and Nomarski imaging, and electrical characterization was performed via electron paramagnetic resonance spectroscopy and Hall effect measurements. In addition, thermal conductivity was measured via steady-state thermoreflectance. AFM imaging revealed a number of extended defects forming in the epilayer, and transmission electron microscopy imaging detailed a V-shaped pit defect formed by misoriented twin defects likely originating from a defect at or near the substrate surface. The Hall effect data (NS = 2.81–3.1 × 1017 cm−3 and μH = 81.5–103.5 cm2/V s), surface roughness (1.78–4.25 nm rms), and thermal conductivity measurements (19.3–19.8 W/m K) yielded comparable results among all samples, independent of epilayer thickness. These results suggested that even thicker, high quality epitaxial (010) β-Ga2O3 could be grown via this technique, provided that the development of the observed extended defects can be mitigated.

Wide bandgap semiconductors, such as SiC, are expected to surpass traditional silicon-based technologies in high-power electronics applications, owing to their potential for reduced size, weight, and enhanced power efficiency. SiC, in particular, has evolved significantly, reaching a maturity that enables its adoption in high-voltage applications including power modules. However, the current state-of-the-art SiC epitaxial wafers face significant challenges, primarily due to the presence of threading dislocations, basal plane dislocations, stacking faults, polytype inclusions, and screw dislocations. These structural imperfections lead to electron and hole carrier scattering, which impairs device performance. Addressing this, the present research introduces a novel approach employing a convolutional neural network (CNN) trained to identify and quantify dislocation defects with high precision. Analysis of the models shows it both highlights the features of interest and delineates the shape, size, and number of the dislocations with higher accuracy than the manual labeling used to train the model, suggesting that CNN has successfully captured the subtle details needed to identify dislocations defects.