Neural-Network Potential for Defect Formation Induced by Knock-On Irradiation Damage in 4H-SiC

Understanding the microscopic mechanism of the irradiation damage in silicon carbide (SiC) is of great importance for improving the irradiation resistance and the ion implantation processes of SiC-based devices. Currently, the atomic-scale simulations of the cascade collisions caused by irradiation in SiC are bottlenecked by the low accuracy of molecular dynamics (MD) with classical interatomic potentials and the low efficiency of ab initio MD (AIMD). In this study, a neural network potential (NNP) is constructed for the simulations of irradiation damage in 4H-SiC using the stochastic surface walking (SSW) for the potential energy surface (PES) exploration. This potential is not only able to provide accurate structural and elastic properties, but also capable of predicting the defect properties and threshold displacement energies (TDEs) that well agree with the first-principles results. More importantly, using this NNP, the directional dependence of the TDEs can be determined based on a set of high throughput calculations, and the minimal TDEs and the corresponding collision directions for Si and C can be predicted, which are in good agreement with the experimental results. This potential provides an efficient and accurate tool to accurately simulate the cascade collisions and gain fundamental understanding of the irradiation damage mechanisms of 4H-SiC.