A generalizable, uncertainty-aware neural network potential for GeSbTe with Monte Carlo dropout

Sung-Ho Lee (1) , Valerio Olevano (1, 2) , Benoit Sklénard (1)

1 Université Grenoble Alpes, CEA, Leti, F-38000, Grenoble, France
2 CNRS, Institut Néel, F-38042, Grenoble, France

A Bayesian neural network potential (NNP) achieved with the Monte Carlo dropout approximation method is developed for GeSbTe alloys. The Bayesian NNP is shown to be more generalizable than its classical counterpart, yielding reasonable predictions on structures that are not directly in the training configurations, and is able to output uncertainty estimates for the predictions. Its application to a molecular dynamics (MD) simulation is also presented, and the validity of the obtained trajectory is evaluated by comparing it to Density Functional Theory (DFT).

Domains: Materials Science [cond-mat.mtrl-sci]; Artificial Intelligence [cs.AI]

Sung-Ho Lee, Valerio Olevano, Benoit Sklénard. A generalizable, uncertainty-aware neural network potential for GeSbTe with Monte Carlo dropout. Solid-State Electronics, 2023, 199, pp.108508. ⟨10.1016/j.sse.2022.108508⟩. ⟨cea-03930383⟩