Unsupervised machine learning accelerates solid electrolyte discovery

摘要

Traditional organic liquid electrolytes used in commercial Li-ion batteries would incur serious safety issues due to their flammability and volatility[1].The exploration and design of solid electrolytes with high room-temperature Li-ion conductivities(sRT)are important to improve the safety and cycle life of Li-ion batteries[2].Although previous investigations have proven that various physical factors correlate with Li-ion diffusion in solids,there is no unified theory to explain the similarity among distinctive crystal structures of solid-state Liion conductors(SSLCs).In addition,the exploration of a vast composition-structure space of thousands of materials is extremely difficult.The current investigations mainly rely on‘trial and error’and are limited to a few kinds of candidates,such as lithium thiophosphates,garnets,sodium super ionic conductors(NASICONs),perovskites and argyrodites[3].Therefore,the discovery of novel SSLCs from wider materials on basis of available knowledge is significant.