Investigation of Phase Stability and Ionic Conductivity of Solid Electrolytes Li₁₀MP₂S_12-xO_x (M = Ge, Si, or Sn) with Universal Neural Network Potential

Li₁₀GeP₂S₁₂ (LGPS)-type structures are promising materials for solid electrolytes for all-solid-state lithium-ion batteries with high lithium-ion conductivity. One of the main challenges of these structures is stability. It has been reported that stability can be improved by substituting oxygen for sulfur, but there is a trade-off, as it reduces lithium-ion conductivity. Therefore, it is worth revealing the effects of oxygen substitution on stability and lithium-ion conductivity, and finding the optimal substitution ratio of the oxygen species that balances both properties.<br/><br/>Density functional theory (DFT) calculations have been widely applied to battery materials. However, because of the computational cost of DFT calculations, it is hard to apply for partially substituted structures that require considering various configurations of substituted elements. In contrast, empirical interatomic potentials are much faster than DFT calculations, but it is difficult to construct an empirical interatomic potential which can handle complex structures composed of as much as five elements with enough accuracy.<br/><br/>In this work, we investigated phase stabilities and lithium-ion diffusion barriers of Li₁₀MP₂S_12-xO_x (M=Ge, Si, or Sn) with PFP, a neural network interatomic potential which supports any combination of 72 elements. Phase stabilities are calculated considering various configurations of substituted structures, and lithium-ion diffusion barriers are calculated with molecular dynamics (MD) simulations with finite temperatures. The accuracy of PFP was validated by comparing forces with DFT calculations for structures randomly selected from the MD trajectories, and it exhibited good agreement with a mean absolute error of 0.027 eV/Å without fine-tuning for the target system. Our results reproduced that oxygen substitution increases phase stability and increases the lithium-ion diffusion barrier, i.e., decreases lithium-ion conductivity. Analysis of lithium-ion diffusion paths reveals that the disappearance of diffusion channels that are not adjacent to the oxygen atoms increases the lithium-ion diffusion barrier. It suggests that there is a critical mass for oxygen which will close all diffusion paths, and Li₁₀SnP₂S_11.5O_0.5 would be a promising solid electrolyte material, as it balances both stability and lithium-ion conductivity.