Predicting Reactivity\nand Passivation of Solid-State\nBattery Interfaces

In this work, we build a computationally inexpensive,\ndata-driven\nmodel that utilizes atomistic structure information to predict the\nreactivity of interfaces between any candidate solid-state electrolyte\nmaterial and a Li metal anode. This model is trained on data from <i>ab initio</i> molecular dynamics (AIMD) simulations of the time\nevolution of the solid electrolyte–Li metal interfaces for\n67 different materials. Predicting the reactivity of solid-state interfaces\nwith <i>ab initio</i> techniques remains an elusive challenge\nin materials discovery and informatics, and previous work on predicting\ninterfacial compatibility of solid-state Li-ion electrolytes and Li\nmetal anodes has focused mainly on thermodynamic convex hull calculations.\nOur framework involves training machine learning models on AIMD data,\nthereby capturing information on both kinetics and thermodynamics,\nand then leveraging these models to predict the reactivity of thousands\nof new candidates in the span of seconds, avoiding the need for additional\nweeks-long AIMD simulations. We identify over 300 new chemically stable\nand over 780 passivating solid electrolytes that are predicted to\nbe thermodynamically unfavored. Our results indicate many potential\nsolid-state electrolyte candidates have been incorrectly labeled unstable\nvia purely thermodynamic approaches using density functional theory\n(DFT) energetics, and that the pool of promising, Li-stable solid-state\nelectrolyte materials may be much larger than previously thought from\nscreening efforts. To showcase the value of our approach, we highlight\ntwo borate materials that were identified by our model and confirmed\nby further AIMD calculations to likely be highly conductive and chemically\nstable with Li: LiB<sub>13</sub>C<sub>2</sub> and LiB<sub>12</sub>PC.