HN Theater @HNTheaterMonth

Discovering promising new materials is central to our ability to design better batteries, but research progress over the last several decades has been limited by an incomplete understanding of the materials physics and inefficient guess-and-check searches. Our research seeks to overcome these limitations by leveraging new approaches inspired by machine learning to make predictions of material performance from existing experimental data. Focusing on solid-state electrolyte materials, we build a data-driven model for predicting ionic conductivity from experimental data on crystal structure and ionic conductivity from the literature. We use the resulting model to guide an experimental search for high ionic conductivity electrolyte materials, and find that incorporating machine learning yields a 3-5x increase in the discovery of high ionic conductivity materials over a comparable guess-and-check effort. This approach also enables us to quantify the likelihood of additional future breakthrough materials discoveries, and we find, among other insights, that one is nearly 100x more likely to realize a stable solid-state battery with two electrolytes rather than one.