Machine-Learning Framework for Designing Stable Interfaces in All-Solid-State Lithium-Ion Batteries

All-solid-state lithium-ion batteries are promising next-generation energy-storage systems, but interfacial instability between cathodes and solid electrolytes remains a major barrier to long-term durability. Interfacial coatings can mitigate these reactions, yet coating selection is limited by the vast chemical design space and incomplete database coverage. In this study, coating discovery is formulated as a prediction-and-design problem in which interfacial reaction energies define a composition-to-reactivity map that generalizes to unseen compounds. Reaction energies are calculated for 809 phase-stable Li-containing compounds from the Materials Project against 10 oxide cathodes and 7 sulfide solid electrolytes. Unsupervised clustering identifies distinct reactivity groups, and composition-based analysis reveals signatures, including polyanion tolerance and fixed-valence cations, that define a low-reactivity design envelope. Using composition-derived physicochemical descriptors, an ensemble regressor predicts pair-averaged reaction energies for newly enumerated compositions. Within this envelope, charge-balanced Li─M─O and Li─M─A─O compositions (A = B, P, Si) are enumerated, pre-screened, and validated by phase-diagram analysis. This workflow enables interpretable machine-learning-guided expansion beyond existing databases for scalable coating discovery.