Discovering Physical Limits of Battery Materials with Physics-based Machine Learning

We compile data and physics-based machine learned models for solid Li-ion electrolyte performance to assess the state of materials discovery efforts in solid-state batteries. Candidate electrolyte materials must satisfy several requirements, chief among them fast ionic conductivity and robust electrochemical stability. In order to probe the interplay of these properties, we first build and validate a machine learning-based model for predicting ionic conductivity. We find this model offers a 3x improvement over trial-and-error searches, and successfully identifies several new materials that demonstrate exceptional ionic conductivity. Then, drawing on DFT-based electrochemical stability models, we examine the predicted performance of thousands of candidate materials and quantify the likelihood of breakthrough solid electrolyte discoveries. Among other insights, this analysis suggests that two electrolytes are likely to be necessary in solid-state Li-ion batteries with Li metal anodes. This work is an effort to extract as much information as possible from today’s limited existing data in order to provide a clear path forward for accelerating tomorrow’s efforts.