Machine-learning Potentials Enabling Large-scale Simulation and Accelerated Material Discovery

Wonseok Jeong, Changho Hong, Jeong Min Choi, Jisu Jung, Suyeon Ju, Kyeongpung Lee, Dongheon Lee, and Seungwu Han*

Department of Materials Science and Engineering, Seoul National University,Seoul 08826, South Korea

ABSTRACT: Recently, machine-learning (ML) approaches to developing interatomic potentials are attracting considerable attention because it is poised to overcome the major shortcoming inherent to the classical potential and density functional theory (DFT), i.e., difficulty in potential development and huge computational cost, respectively. In particular, the high-dimensional neural network potential (NNP) suggested by Behler and Parrinello is attracting wide interests with applications demonstrated over various materials. In this presentation, we first introduce our in-house code for training and executing NNP called SIMPLE-NN (SNU Interatomic Machine-learning PotentiaL packagE-version Neural Network). The package features GDF weighting that compensates for sampling biases in the training set and ‘replica’ ensemble to detect atomic configurations with high uncertainties in large-scale MD simulations. We further discuss on the fundamental aspect of ML potentials that enables the transferability of the potential. We show that the ML potential is essentially a manifestation of O(N) method of DFT, which is realized in terms of atomic energies.

As application examples, we present our recent results on phase change behavior of chalcogendies and silicidation process in semiconductor fabrication. Furthermore, we will show that ML potentials can be used as highly accurate surrogate models in exploring large space of crystal structures. This enables finding the stable crystal structure for complicated multicomponent systems.

Keywords: machine-learning potential; large scale simulation; density-functional theory.