Tag: Machine Learning in Materials Science
Model identifies capping layers that may limit niobium oxide formation
What the study found A predictive framework was developed to select metal capping layers that inhibit niobium oxide formation, which is associated with two-level systems that degrade niobium-based superconducting quantum computing devices. The authors report that their model identified Zr, Hf, and Ta as effective diffusion barriers, and that Zr, Ta, and Sc were especially…
Machine-learning models captured CDW behavior in NbSe2
What the study found Machine-learning interatomic potentials (models that predict how atoms interact) can be trained to capture charge density wave (CDW, a periodic distortion in a material’s structure) behavior in mono- and bilayer NbSe2. The study found that CDW lattice distortions were relatively easy to learn, but vibrational properties were more difficult. Why the…
Autonomous electron-beam fabrication controlled defect structures in 2D materials
What the study found The study found that a fully autonomous approach can fabricate atomic-level defects in two-dimensional materials using machine learning and automated electron-beam control. As a proof of concept, the authors achieved controlled fabrication of MoS-nanowire edge structures in a MoS2 monolayer. Why the authors say this matters The authors say the approach…
EAC-Net predicts charge density with high accuracy
What the study found EAC-Net is a model for predicting real-space charge density that combines accuracy with a physically grounded atomic decomposition. The authors report that it can achieve errors typically below 1% across the periodic table and generalize well to diverse chemical environments. Why the authors say this matters The study suggests that EAC-Net…
Post-training pruning makes moment tensor potentials faster
What the study found The study found that a post-training, cost-aware pruning strategy can remove expensive basis functions from Moment Tensor Potentials (MTPs) with minimal loss of accuracy. Applied to nickel and silicon-oxygen systems, the resulting models were up to seven times faster than standard MTPs. Why the authors say this matters The authors say…
ACE model describes diverse Si-H structures
—
in EngineeringWhat the study found The authors present a machine-learned interatomic potential for the silicon-hydrogen system that can describe crystalline and amorphous bulk structures, surfaces, and molecules. They say it covers a wide range of Si-H phases. Why the authors say this matters The study suggests this may help explore large structural models of amorphous silicon-hydrogen…
Physics-guided machine learning improved waveform prediction under sparse data
—
Physics-guided machine learning framework predicts quasi-isentropic waveforms from sparse data, achieving 96% accuracy and reducing computational resource requirements for material design.
