Publications

In article number 1906146, Robert O. Ritchie, Ashwin J. Shahani, and co-workers investigate the origins of certain eutectics that grow into spiral geometry. The spiral formation follows a two-step process, wherein the first step is mediated by the low solid–liquid interfacial energy of a precursor phase, and the second step by crystallographic defects on the precursor. The cover features a computed X-ray nano-tomography image of spiral eutectics in Mg-Zn alloy showing multiple spiral eutectic colonies with the eutectic MgZn2 and Zn phases shown in dark and light false colors (image courtesy of H. Bale).

Foreign-object damage associated with the ingestion of debris into aircraft turbine engines can lead to a marked degradation in the high-cycle fatigue (HCF) life of turbine components. This degradation is generally considered to be associated with the premature initiation of fatigue cracks at or near the damage sites; this is suspected to be due to, at least in part, the impact-induced residual stress state, which can be strongly tensile in these locations. However, recent experimental studies have shown the unexpected propensity for impact-induced fatigue crack formation at locations of compressive residual stress in the vicinity of the impact site. To address this issue, in situ and ex situ spatially-resolved X-ray diffraction and numerical modeling are utilized to show that the initial residual stress state can be strongly relaxed during the fatigue loading process. The magnitude and rate of relaxation is strongly dependent on the applied loads. For a Ti–6Al–4V turbine blade alloy, little relaxation was observed for an applied maximum stress of 325 MPa (0.35σ y , where σ y is the yield stress), and cracks tended to form in subsurface zones of tensile residual stress away from the damage sites. In contrast, at an applied maximum stress of 500 MPa (0.54σ y ), equal to the smooth-bar 107-cycle endurance strength, cracks tended to form at the damage sites in zones of high stress concentration that had initially been in strong compression, but had relaxed during the fatigue loading.

Current trends in Machine Learning~(ML) inference on hardware accelerated devices (e.g., GPUs, TPUs) point to alarmingly low utilization. As ML inference is increasingly time-bounded by tight latency SLOs, increasing data parallelism is not an option. The need for better efficiency motivates GPU multiplexing. Furthermore, existing GPU programming abstractions force programmers to micro-manage GPU resources in an early-binding, context-free fashion. We propose a VLIW-inspired Out-of-Order (OoO) Just-in-Time (JIT) compiler that coalesces and reorders execution kernels at runtime for throughput-optimal device utilization while satisfying latency SLOs. We quantify the inefficiencies of space-only and time-only multiplexing alternatives and demonstrate an achievable 7.7x opportunity gap through spatial coalescing.

A recently developed As2S3 chemical treatment was used to passivate the perimeters of self-aligned heterojunction bipolar transistors (HBTs). The As2S3 chemical treatment significantly lowered the base current resulting in a two order of magnitude reduction in the collector current density at which dc current gain was observed (β=1). No degradation with time has been observed in the electrical characteristics of the chemically treated HBTs. This absence of degradation is attributed to the impermeability to oxygen of the As2S3 glass which coats the perimeter of the HBT after chemical treatment.

Controlling the spatial arrangement of molecular catalysts on electrodes is critical to developing an optimal electrocatalyst. Mo-S clusters have shown great promise in catalyzing hydrogen evolution for the generation of carbon-free fuel from water. Here we report a synthetic approach to organize these molecular clusters into ordered dimers, cages, and chains through the use of organic linkers, as solved by single-crystal X-ray diffraction. We find that the linkage through the coordination bond between thiolate and Mo₃S₇ leads to (1) a 40-fold enhancement in turnover frequency compared with the unlinked cluster and (2) the periodic arrangement of clusters on the electrode with control over their distance, orientation, and density, thus enabling hydrogen evolution at high catalyst loading. The materials developed here require an overpotential of only 89 mV to achieve a current density of 10 mA cm^-2, outperforming other Mo-S catalysts (both molecular and solid-state).

We present an approach for building an active agent that learns to segment its visual observations into individual objects by interacting with its environment in a completely self-supervised manner. The agent uses its current segmentation model to infer pixels that constitute objects and refines the segmentation model by interacting with these pixels. The model learned from over 50K interactions generalizes to novel objects and backgrounds. To deal with noisy training signal for segmenting objects obtained by self-supervised interactions, we propose robust set loss. A dataset of robot's interactions along-with a few human labeled examples is provided as a benchmark for future research. We test the utility of the learned segmentation model by providing results on a downstream vision-based control task of rearranging multiple objects into target configurations from visual inputs alone. Videos, code, and robotic interaction dataset are available at https://pathak22.github.io/seg-by-interaction/

Disordered rocksalt Li-excess (DRX) compounds have emerged as promising new cathode materials for lithium-ion batteries, as they can consist solely of resource-abundant metals and eliminate the need for cobalt or nickel. A deeper understanding of the lithium-ion transport kinetics in DRX compounds is essential for enhancing their rate performance. This study employs first-principles calculations, cluster expansion techniques, and kinetic Monte Carlo simulations to investigate the Li+ transport properties in DRX Li2–xVO3, where 0 ≤ x ≤ 1. Our findings underscore (i) the necessity of accounting for both tetrahedral and octahedral Li occupancy when predicting the transport properties in DRX materials, (ii) the factors influencing the variation in the diffusion coefficients with Li content in Li2–xVO3, and (iii) the impact of Li+ correlated motion on the kinetics of Li+ transport. We reveal that the relative stability of tetrahedral and octahedral Li determines the number of active sites within the percolation network, subsequently affecting the Li+ transport properties. Furthermore, we demonstrate that the wide site energy distribution causes correlated motion in Li2–xVO3, which hinders Li+ transport.

Alloying III–V compounds with small amounts of nitrogen leads to a dramatic reduction of the fundamental band-gap energy in the resulting dilute nitride alloys. The effect originates from an anticrossing interaction between the extended conduction-band states and localized N states. The interaction splits the conduction band into two non-parabolic subbands. The downward shift of the lower conduction subband edge is responsible for the N-induced reduction of the fundamental band-gap energy. The changes in the conduction band structure result in a significant increase in electron effective mass and a decrease in the electron mobility, and lead to a large enhancement of the maximum doping level in GaInNAs doped with group VI donors. In addition, a striking asymmetry in the electrical activation of group IV and group VI donors can be attributed to mutual passivation process through formation of the nearest neighbour group-IV donor nitrogen pairs.