Deep learning inter-atomic potential for irradiation damage in 3C-SiC

TL;DR

This paper introduces a deep learning-based inter-atomic potential for 3C-SiC that improves the accuracy of molecular dynamics simulations of irradiation damage, aligning closer with ab-initio results than traditional potentials.

Contribution

The authors developed a novel deep learning inter-atomic potential that outperforms existing analytical potentials in simulating radiation damage in 3C-SiC.

Findings

Deep-learning potential closely matches DFT predictions for defect energies.

It provides more consistent results than traditional empirical potentials.

Enhanced accuracy in modeling primary irradiation damage processes.

Abstract

We developed and validated an accurate inter-atomic potential for molecular dynamics simulation in cubic silicon carbide (3C-SiC) using a deep learning framework combined with smooth Ziegler-Biersack-Littmark (ZBL) screened nuclear repulsion potential interpolation. Comparisons of multiple important properties were made between the deep-learning potential and existing analytical potentials which are most commonly used in molecular dynamics simulations of 3C-SiC. Not only for equilibrium properties but also for significant properties of radiation damage such as defect formation energies and threshold displacement energies, our deep-learning potential gave closer predictions to DFT criterion than analytical potentials. The deep-learning potential framework solved the long-standing dilemma that traditional empirical potentials currently applied in 3C-SiC radiation damage simulations gave large disparities with each other and were inconsistent with ab-initio calculations. A more realistic depiction of the primary irradiation damage process in 3C-SiC can be given and the accuracy of classical molecular dynamics simulation for cubic silicon carbide can be expected to the level of quantum mechanics.