Crack-free AlN films with threading dislocation density (TDD) below 109 cm−2 are needed for deep-UV optoelectronics. This is typically achieved using pulsed lateral overgrowth or very thick buffer layers (>10 μm), a costly and time-consuming approach. A method for conventional metalorganic chemical vapor deposition growth of AlN/SiC films below 3 μm with greatly improved quality is presented. Focusing on substrate pretreatment before growth, we reduce average film stress from 0.9 GPa (tension) to −1.1 GPa (compression) and eliminate cracking. Next, with optimized growth conditions during initial deposition, AlN films with x-ray rocking curve widths of 123 arc-sec ( ) and 304 arc-sec ( ) are developed, and TDD is confirmed via plan view transmission electron microscopy (TEM) to be 2 108 cm−2. Film stress measurements including x-ray 2θ-ω, reciprocal space mapping, and curvature depict compressively stressed growth of AlN on 4H-SiC due to lattice mismatch. The thermal expansion coefficient mismatch between AlN and SiC is measured to be and is found to be constant between room temperature and 1400 °C. TEM confirms the existence of dense misfit dislocation (MD) networks consistent with MD formation near SiC step edges and low MD density regions attributed to nearly coherent AlN growth on SiC terraces. These low-TDD, crack-free AlN/SiC buffers provide a platform for deep-UV optoelectronics and ultrawide bandgap electronics.
31 January 2019 in Semiconductor Science and Technology
The impact of AlGaN growth conditions on AlGaN:Si resistivity and surface morphology has been investigated using metalorganic chemical vapor deposition. Growth parameters including growth temperature, growth rate, and trimethylindium (TMI) flow have been systematically studied to minimize the resistivity of AlGaN:Si. We observed a strong anticorrelation between AlGaN:Si conductivity and growth temperature, suggesting increased silicon donor compensation at elevated temperatures. Secondary ion mass spectrometry and positron annihilation spectroscopy ruled out compensation by common impurities or group-III monovacancies as a reason for the observed phenomenon, in contrast to theoretical predictions. The underlying reason for AlGaN:Si resistivity dependence on growth temperature is discussed based on the possibility of silicon acting as a DX center in Al0.65Ga0.35N at high growth temperatures. We also show remarkable enhancement of AlGaN:Si conductivity by introducing TMI flow during growth. A minimum resistivity of 7.5 mΩ cm was obtained for n-type Al0.65Ga0.35N, which is among the lowest reported resistivity for this composition.
Discovering ways to increase the LED light extraction efficiency (LEE) should help create the largest performance improvement in the power of UV AlGaN LEDs. Employing surface roughening to increase the LEE of typical AlGaN UV LEDs is challenging and not well understood, yet it can be achieved easily in AlGaN LEDs grown on SiC. We fabricate thin-film UV LEDs (similar to 294-310 nm) grown on SiC-with reflective contacts and roughened emission surface-to study and optimize KOH roughening of N-face AlN on the LEE as a function of roughened AlN pyramid size and KOH solution temperature. The LEE increased the most (2X) when the average AlN pyramid base diagonals (d) were comparable to the electroluminescence (EL) wavelength in the AlN layer (d similar to lambda(EL): 42-52 pyramids/mu m(2)), but the LEE enhancement diminished when d was much larger than lambda(EL) (d -5.5 lambda(EL); 2-3 pyramids/mu m). The UV LEDs had a 10 mu p-GaN contact layer, and the forward voltage was similar to 6 V at similar to 8 A/cm(2), with a voltage efficiency (VE) of similar to 70%. The VE of the LEDs did not change after KOH roughening. This work suggests important implications to increase the LEE of AlGaN LEDs. (C) 2019 Optical Society of America under the terms of the OSA Open Access Publishing Agreement
Graph neural networks (GNNs) for crystal property prediction are typically evaluated by a single aggregate error, which can mask where, and for which classes of materials, these models fail. In this study, we present a space-group-, centering-type-, and band gap-resolved error analysis of GNN band gap prediction on the Materials Project dataset. As a channel for this analysis we use Crystal-X, a deliberately simple model: a standard graph convolutional backbone with two minor architectural modifications, an asymmetric edge convolution and a neighbor-feature transformation, that supplement bond information often treated as secondary in node-centric models. Crystal-X is not a state-of-the-art model: it reaches a band gap MAE of 0.256 eV on the MP 2018.6 dataset, behind ALIGNN (0.22 eV) and PotNet (0.20 eV), though ahead of older baselines such as CGCNN (0.39 eV), SchNet (0.415 eV), and MEGNet (0.33 eV) while using only the nine-property CGCNN atomic feature set. Its value here is as a controlled, low-complexity testbed for the error analysis. That analysis reveals systematic patterns that aggregate MAE conceals: errors concentrate in underrepresented band gap ranges and in low-symmetry and non-centrosymmetric space groups; per-group errors for sparsely populated space groups are dominated by sampling noise; and modest, as-yet-unverified gains from edge-aware convolutions appear in monoclinic and non-primitive-centered systems. We argue that this kind of granular, symmetry-resolved evaluation should accompany aggregate benchmarks when assessing crystal GNNs.