Search for: All records

High-level synthesis (HLS) is an automated design process that transforms high-level code into optimized hardware designs, enabling rapid development of efficient hardware accelerators for various applications such as image processing, machine learning, and signal processing. To achieve optimal performance, HLS tools rely on pragmas, which are directives inserted into the source code to guide the synthesis process, and these pragmas can have various settings and values that significantly impact the resulting hardware design. State-of the-art ML-based HLS methods, such as harp, first train a deep learning model, typically based on graph neural networks (GNNs) applied to graph-based representations of the source code and its pragmas. They then perform design space exploration (DSE) to explore the pragma design space, rank candidate designs using the trained model, and return the top designs as the final designs. However, traditional DSE methods face challenges due to the highly nonlinear relationship between pragma settings and performance metrics, along with complex interactions between pragmas that affect performance in non-obvious ways. To address these challenges, we propose compareXplore, a novel approach that learns to compare hardware designs for effective HLS optimization. compareXplore introduces a hybrid loss function that combines pairwise preference learning with pointwise performance prediction, enabling the model to capture both relative preferences and absolute performance values. Moreover, we introduce a novel Node Difference Attention module that focuses on the most informative differences between designs, enhancing the model’s ability to identify critical pragmas impacting performance. compareXplore adopts a two-stage DSE approach, where a pointwise prediction model is used for the initial design pruning, followed by a pairwise comparison stage for precise performance verification. Experimental results demonstrate that compareXplore achieves significant improvements in ranking metrics and generates high quality HLS results for the selected designs, outperforming the existing state-of-the-art method. 

The development of high performance wide-bandgap AlGaN channel transistors with high current densities and reduced Ohmic losses necessitates extremely highly doped, high Al content AlGaN epilayers for regrown source/drain contact regions. In this work, we demonstrate the achievement of semi-metallic conductivity in silicon (Si) doped N-polar Al0.6Ga0.4N grown on C-face 4H-SiC substrates by molecular beam epitaxy. Under optimized conditions, the AlGaN epilayer shows smooth surface morphology and a narrow photoluminescence spectral linewidth, without the presence of any secondary peaks. A favorable growth window is identified wherein the free electron concentration reaches as high as ∼1.8 × 1020 cm−3 as obtained from Hall measurements, with a high mobility of 34 cm2/V·s, leading to a room temperature resistivity of only 1 mΩ·cm. Temperature-dependent Hall measurements show that the electron concentration, mobility, and sheet resistance do not depend on temperature, clearly indicating dopant Mott transition to a semi-metallic state, wherein the activation energy (Ea) falls to 0 meV at this high value of Si doping for the AlGaN films. This achievement of semi-metallic conductivity in Si doped N-polar high Al content AlGaN is instrumental for advancing ultrawide bandgap electronic and optoelectronic devices. 

Interface engineering in heterostructures at the atomic scale has been a central research focus of nanoscale and quantum material science. Despite its paramount importance, the achievement of atomically ordered heterointerfaces has been severely limited by the strong diffusive feature of interfacial atoms in heterostructures. In this work, we first report a strong dependence of interfacial diffusion on the surface polarity. Near-perfect quantum interfaces can be readily synthesized on the semipolar plane instead of the conventionalc-plane of GaN/AlN heterostructures. The chemical bonding configurations on the semipolar plane can significantly suppress the cation substitution process as evidenced by first-principles calculations, which leads to an atomically sharp interface. Moreover, the surface polarity of GaN/AlN can be readily controlled by varying the strain relaxation process in core–shell nanostructures. The obtained extremely confined, interdiffusion-free ultrathin GaN quantum wells exhibit a high internal quantum efficiency of ~75%. Deep ultraviolet light-emitting diodes are fabricated utilizing a scalable and robust method and the electroluminescence emission is nearly free of the quantum-confined Stark effect, which is significant for ultrastable device operation. The presented work shows a vital path for achieving atomically ordered quantum heterostructures for III-nitrides as well as other polar materials such as III-arsenides, perovskites, etc. 

N-polar AlGaN is an emerging wide-bandgap semiconductor for next-generation high electron mobility transistors and ultraviolet light emitting diodes and lasers. Here, we demonstrate the growth and characterization of high-quality N-polar AlGaN films on C-face 4H-silicon carbide (SiC) substrates by molecular beam epitaxy. On optimization of the growth conditions, N-polar AlGaN films exhibit a crack free, atomically smooth surface (rms roughness ∼ 0.9 nm), and high crystal quality with low density of defects and dislocations. The N-polar crystallographic orientation of the epitaxially grown AlGaN film is unambiguously confirmed by wet chemical etching. We demonstrate precise compositional tunability of the N-polar AlGaN films over a wide range of Al content and a high internal quantum efficiency ∼74% for the 65% Al content AlGaN film at room temperature. Furthermore, controllable silicon (Si) doping in high Al content (65%) N-polar AlGaN films has been demonstrated with the highest mobility value ∼65 cm2/V-s observed corresponding to an electron concentration of 1.1 × 1017 cm−3, whereas a relatively high mobility value of 18 cm2/V-s is sustained for an electron concentration of 3.2 × 1019 cm−3, with an exceptionally low resistivity value of 0.009 Ω·cm. The polarity-controlled epitaxy of AlGaN on SiC presents a viable approach for achieving high-quality N-polar III-nitride semiconductors that can be harnessed for a wide range of emerging electronic and optoelectronic device applications. 

To date, it has remained challenging to achieve N-polar AlN, which is of great importance for high power, high frequency, and high temperature electronics, acoustic resonators and filters, ultraviolet (UV) optoelectronics, and integrated photonics. Here, we performed a detailed study of the molecular beam epitaxy and characterization of N-polar AlN on C-face 4H-SiC substrates. The N-polar AlN films grown under optimized conditions exhibit an atomically smooth surface and strong excitonic emission in the deep UV with luminescence efficiency exceeding 50% at room temperature. Detailed scanning transmission electron microscopy (STEM) studies suggest that most dislocations are terminated/annihilated within ∼200 nm AlN grown directly on the SiC substrate due to the relatively small (1%) lattice mismatch between AlN and SiC. The strain distribution of AlN is further analyzed by STEM and micro-Raman spectroscopy, and its impact on the temperature-dependent deep UV emission is elucidated. 

Hurricanes are one of the most catastrophic natural forces with potential to inflict severe damages to properties and loss of human lives from high winds and inland flooding. Accurate long-term forecasting of the trajectory and intensity of advancing hurricanes is therefore crucial to provide timely warnings for civilians and emergency responders to mitigate costly damages and their life-threatening impact. In this paper, we present a novel online learning framework called JOHAN that simultaneously predicts the trajectory and intensity of a hurricane based on outputs produced by an ensemble of dynamic (physical) hurricane models. In addition, JOHAN is designed to generate accurate forecasts of the ordinal-valued hurricane intensity categories to ensure that their severity level can be reliably communicated to the public. The framework also employs exponentially-weighted quantile loss functions to bias the algorithm towards improving its prediction accuracy for high category hurricanes approaching landfall. Experimental results using real-world hurricane data demonstrated the superiority of JOHAN compared to several state-of-the-art learning approaches.