The Application of Deep Learning on Room Temperature Conductivity of Li10GeP2S12 Type Solid State Electrolyte

Abstract

Room temperature ionic conductivity of the solid-state electrolyte is the essencial element of the commercialization of solid-state lithium-ion batteries. First-principles method is a good tool as it simulates the atomic level dynamics, and provides accurate insides of the ion moving velocity, path, and the microscopic environment. However, the first-principles method is time and computational resources-consuming. It is usually applied to systems with hundreds of atoms and the simulation time is less than 100 picoseconds. The classical molecular dynamics method applies to systems containing thousands of atoms and the simulation time can reach the scale of nanoseconds. However, it is based on the experimental force-field parameters and can only be applied to limited systems with experimentally-proofed good parameters. The method used in this paper is a combination of the first-principles simulation and classical dynamics. Deep-learning tool Deepmd-kit is used to train the first-principles simulation data of the targeted system into a tailor-made force-field model. The verified model for the specific system will then be used in the classical molecular dynamics simulation. With the first-principles precise and classical efficiency, the molecular dynamics simulation of ion movement at room temperature is achieved. The results of Li10GeP2S12 and Li10SiP2S12 showed good agreement with the experiments.