Revisiting the in-plane and in-channel diffusion of lithium ions in a solid-state electrolyte at room temperature through neural network-assisted molecular dynamics simulations†

Developing superionic conductor (SIC) materials offers a promising pathway to achieving high ionic conductivity in solid-state electrolytes (SSEs). The Li₁₀GeP₂S₁₂ (LGPS) family has received significant attention due to its remarkable ionic conductivity among various SIC materials. Ab initio molecular dynamics (AIMD) simulations have been extensively used to explore the diffusion behavior of Li⁺ ions in Li₁₀GeP₂S₁₂. These simulations indicate that Li⁺ ions diffuse rapidly along a one-dimensional (1D) chain direction, specifically along the c-axis, a process known as in-channel diffusion. In addition, these computational studies have identified additional diffusion pathways within the ab planes, referred to as in-plane diffusion. However, there are still notable limitations in the time scale associated with AIMD simulation techniques for studying the dynamic behavior of Li₁₀GeP₂S₁₂ at room temperature. In this study, we trained a deep potential (DP) model for the LGPS system and performed a 300-nanosecond DeePMD simulation to investigate the diffusion behavior of Li₁₀GeP₂S₁₂ at room temperature. The neural network (NN) assisted simulation showed that the framework structure of LGPS remained remarkably stable over the entire 300-nanosecond period. Following this, our investigation focused on the two-dimensional (2D) diffusion pathways within the ab plane (in-plane diffusion mechanism) and the 1D diffusion channel along the c-axis (in-channel diffusion mechanism). Upon analyzing the DeePMD simulation results, we identified two distinct pathways for in-plane Li⁺ diffusion within the ab plane: the Li-2 and Li-4 pathways. We determined the energy barriers for the two diffusion pathways to be 0.23 eV and 0.34 eV, respectively, in qualitative agreement with recent experimental and theoretical results. For the in-channel diffusion along the c-axis, our calculated energy barrier was approximately 0.083 eV, closely matching the previous one-particle potential (OPP) analysis. Our results confirm experimental and theoretical studies, indicating that lithium ions experience significantly less resistance when diffusing through the in-channel pathway compared to the in-plane migration mechanism. However, our findings suggest that despite having a higher energy barrier than the Li-4 pathway, the Li-2 pathway remains a viable option for in-plane lithium-ion migration.