High-Throughput Data-Driven Prediction of Stable High-Performance Na-Ion Sulfide Solid Electrolytes

Abstract

Designing solid electrolytes (SEs) for solid-state batteries is a challenging issue. Herein, by using data-driven techniques and a multi-stage density functional theory molecular dynamics (DFT-MD) sampling workflow that is developed, 523 443 978 cell structure samples of in-silico Na-based sulfides with isolated tetrahedral framework units are generated and a largely unexplored chemical space of the filtered 170 samples is examined to find novel SE candidates. Given the DFT-accurate MD ion trajectory configurations for the largest cell structure dataset so far, the three (meta)stable solid-solution series with high Na-ion conductivity σNa,300K = 10−3–10−2 S cm−1: Na5−2xAl1 −xVxS4, Na5−2xAl1−xTaxS4, and Na5−2xIn1−xSbxS4 (0.375 ≤ x ≤ 0.625) are suggested. Moreover, robust descriptors for Na-ion self-diffusion coefficient DNa,300K accumulated from the sampling workflow by exhaustive multiple regression modeling is revealed, indicating that crystal structures with wide NaS3 solid angles and high-valence dopants with small ionic radii result in high Na-ion diffusivity.