Harnessing artificial intelligence to holistic design and identification for solid electrolytes

Despite extensive studies, the development of solid-state batteries (SSBs) has not yet met expectations, owing mainly to the lack of suitable solid electrolytes (SEs) that exhibit low electronic conductivity (σe), high ionic conductivity (σi), and good stability. Here, we propose an effective target-driven framework for holistic identifying promising garnet-type SEs. Using artificial intelligence (AI) technologies, we accurately predict the σe with a mean absolute error of 0.25 eV, achieving a computed speed that is ~109 faster than ab initio calculations. Successfully, from 29,008 garnets, we discovered 12 promising super Li-ion conductors for SEs with σe < 3.6 × 10−30 S cm−1, σi > 10−4 S cm−1 (up to 3.24 S cm−1), and good thermal stability at room temperature and high temperature based on rigorous ab initio validation. These emerging SEs are expected to be used in Li-ion SSBs, thus improving the safety, performance, and lifetime of state-of-the-art energy storage technology. This approach directly cuts across at least 95 years of computational cycles to screen SEs, resulting in significant cost savings and helping us enter an electrified future that relies less on fossil fuels. The data that support the machine learning model of this study are available at: https://www.materialsproject.org.