Machine-Learning-Accelerated Discovery of High-Performance Na 5 X(Si/Ge) 4 O 12 (X = Y, Ln, Bi) Solid-State Electrolytes

Developing high-performance solid-state electrolytes is essential for realizing safe, high-energy-density all-solid-state batteries. In this work, we systematically screen the Na 5 XSi 4 O 12 and Na 5 XGe 4 O 12 (X = Y, Ln, Bi) series using universal interatomic potentials, first-principles calculations, and specialized machine learning potentials. Most candidates exhibit favorable thermodynamic and electrochemical stability and robust electronic insulation, underscoring their promise as solid-state electrolytes. In the silicate series, ionic conductivity varies nonmonotonically with ionic radius, where Na 5 BiSi 4 O 12 achieving an outstanding room-temperature conductivity of 13.74 mS/cm. This behavior is attributed to enhanced Bi–O bond covalency, which exerts an inductive effect that smoothens the migration landscape. By contrast, the germanate analogues display inferior properties due to stronger electrostatic stabilization and rougher energy landscapes. Geometric analysis further reveals that increasing the substituent ionic radius induces XO 6 octahedral distortion and constricts migration bottlenecks in both series. These findings provide a comprehensive roadmap for the rational design of this solid-state electrolyte family.