A machine-learning interatomic potential to understand primary radiation damage of silicon

Abstract

Harsh radiation environments cause displacement damages in semiconductor components, resulting in performance degradation. Molecular simulations provide a unique approach to study the dynamic processes of radiation-induced defect production, clustering, and evolution to design and reinforce novel semiconductor components. In this paper, we developed a more efficient machine learning (ML) potential with DFT accuracy to investigate the radiation damage in silicon material. The accuracy of the potential was verified by comparing the static properties, defect formation energy, and threshold displacement energy with experiments and DFT data. By simulating the excitation of a single PKA we found that with the ML potential PKA has an impact over a larger spatial area compared to the empirical potentials. Finally, by simulating multiple PKA excitations in sequence, we found that the first several excited PKAs produce an amorphous region. For the later excited PKAs, the energies are dissipated when they cross through the amorphous region, resulting in a 36% decrease in newly generated defects.