Data-Driven Design Strategies for Chemically Stabilizing Cathode-Coating Interfaces Employing Interpretable Machine Learning

Abstract

The chemical stability of cathode–coating interfaces plays a critical role in enabling high-energy, high-voltage battery systems; however, the key descriptors governing interfacial compatibility remain poorly understood. In this study, we construct a large-scale thermodynamic reactivity landscape comprising 41,372 cathode–coating pairs and systematically map their solid–solid interfacial reaction energies. Statistical analysis reveals distinct chemistry-dependent trends, highlighting the dominant role of anion compatibility in determining interfacial stability. In particular, cross-anion mismatches, including oxide–fluoride pairings, significantly increase the thermodynamic driving force for interfacial phase reconstruction, whereas fluorine-based coating chemistries, especially the MF2-type alkaline earth metal fluorides, generally suppress reactivity relative to oxide counterparts. Oxide coatings further exhibit nonlinear, descriptor-coupled behavior, in which high-valence cations tend to enhance interfacial stability, albeit with increased sensitivity to the cathode state of charge. To interrogate the high-dimensional descriptor space, we develop a hierarchical feature framework and train interpretable machine-learning models that achieve strong predictive and generalizable performance. These results suggest that interfacial reactivity is governed by general physicochemical descriptors rather than specific phases, as quantitatively elucidated using SHAP analysis. Collectively, this work establishes a descriptor-driven compatibility framework for the rational design of cathode–coating interfaces in next-generation high-voltage energy storage systems.