Computational screening of sodium solid electrolytes through unsupervised learning

Abstract

All-solid-state Na-ion batteries have emerged as alternatives of all-solid-state Li-ion batteries owing to the global abundance of Na element. However, the attempts to seek a commercially viable Na-ion solid-state electrolyte (SSE) are challenging due to the relatively poor understanding of the structures which are effective for conducting Na-ion compared to Li-ion SSE. In this study, unsupervised machine learning is performed to reveal the major characteristics of possible Na-ion SSEs. The descriptor vector that consists of 180 quantitative structure-properties for 12,670 Na-ion contained SSE structures was utilized as training data for the unsupervised clustering via the hierarchical density-based spatial clustering of applications with noise. The resulted clusters identified 12 structure groups including experimentally proven Na-ion superconductor ones such as NASICONs and thiophosphates. The post hoc analysis of the clusters reveals that the structure groups with high conductivity shares the similar characteristics implying the existence of mobility ion channels and the weak interactions between Na-ions and the proximate atoms. The ab initio molecular dynamics simulation results are presented to confirm the promising Na-ion SSE group shows characteristic tendencies in Na ion diffusivity distinguishable from other groups. The Na-ion SSEs map provided in this study will serve as fundamental guidelines to develop novel Na-ion SSEs for all-solid-state Na-ion batteries