Rapid discovery of inorganic-organic solid composite electrolytes by unsupervised learning

Inorganic-organic solid composite electrolytes (SCEs) have been widely concerned owing to their excellent film forming performance, good wettability and low flammability. However, their high polymer crystallinity leads to the low ionic conductivity (sigma), seriously impeding practical applications. Discovering SCEs with high sigma through trial-and-error experiments and high-throughput calculations from massive material search space is an impractical task. The severe scarcity of experimental data on known SCEs even limits the utilization of supervised learning. Here, we adopted an unsupervised learning (UL) model to discover new SCEs with high sigma based on <50 known experimental data. Our model revealed the key physical factors that affected the u and clustered most of the known SCEs with high sigma into four groups. From that we rapidly identified 49 promising SCEs with high u, compared them with previous experimental results, and found two structures with the lowest Li+ migration activation energy (only 0.212 eV). This work fully exploited the potential of UL to overcome the limitations of data scarcity in material discovery. Importantly, we shortened the screening period of SCEs by similar to 23 years, providing a new idea for the rapid discovery and targeted design of materials for solid-state batteries.