Machine Learning Accelerated Electrolyte Discovery for Batteries

Battery chemistries with high energy densities are required for long duration energy storage. Moving away from intercalation to electrodeposition endows lithium metal batteries (LMBs) with energy densities that surpass conventional lithium-ion. However, there are no electrolytes to date that can enable efficient electrodeposition and dissolution of lithium metal without significant degradation and lithium dendritic growth. Therefore, electrolyte discovery holds great promise for lithium metal batteries. Unfortunately, current approaches to electrolyte discovery have focused primarily on trial-and-error. In our work, we pursue a data-driven approach. We curate the largest dataset of important properties for small molecule electrolytes – ionic conductivity, oxidative stability, and Coulombic efficiency. We build machine learning models capable of predicting these properties and show that the models are consistent with known chemical principles and intuition. Deploying these models on large unlabeled datasets allowed us to identify new and promising electrolytes that have never been explored for LMBs. Our data science driven approach for electrolyte discovery is a paradigm shift that addresses challenges facing electrolytes in a wide range of electrochemical devices.