ABSTRACT Intergranular fracture due to the penetration of metal dendrites is an important failure mode in inorganic solid electrolytes (ISEs). Adequate fracture resistance at ISE grain boundaries (GBs) is thus critical for realizing practical solid‐state batteries (SSBs). Using a machine learning (ML) technique, this work connects attributes of ISE GBs to their fracture resistance and demonstrates that GBs in distinct materials, by virtue of their local structure and composition, can behave in similar ways. Altering interfacial characteristics is thus suggested as a potential method for tuning material properties without changing crystal chemistry. We train a mixture‐of‐experts (MoE) based ML model for estimating work of adhesion ( W ad ), the thermodynamic threshold energy for fracture, at ISE GBs and construct structure maps using latent features extracted from the model. A global picture of GB W ad across ISE chemistries is thus presented and complex relationships between crystallographic attributes of GBs, their local non‐stoichiometry, and W ad are revealed. While this ML‐assisted technique may provide useful insights for processing ISEs with higher fracture‐resistance by tuning GB chemistry, it also highlights the value in analyzing latent variables from ML models for unraveling complex structure‐property relationships.
Cheenady et al. (Tue,) studied this question.