Machine-learning-accelerated discovery of amorphous lithium-ion electrolytes

All-solid-state lithium-ion batteries require high-performance electrolytes to replace flammable liquids. Amorphous solid-state electrolytes (SSEs) are particularly promising due to their isotropic transport pathways, compositional tunability, and superior dendrite resistance. We present an integrated, multistage computational workflow to accelerate the discovery of these materials. First, a machine-learning model trained on elemental descriptors enables rapid screening across a vast chemical space. Promising candidates are then evaluated via melt-quench simulations using universal interatomic potentials to ensure structural stability in the amorphous phase. Finally, high-accuracy, system-specific moment tensor potentials are developed through active learning for large-scale molecular-dynamics simulations to robustly quantify lithium mobility. This workflow identified five promising amorphous electrolytes: LiErSe 2 , LiDySe 2 , LiInSe 2 , LiTbSe 2 , and LiAlTe 2 . Notably, LiAlTe 2 exhibits a high predicted room-temperature ionic conductivity of 7.7 mS/cm. In all cases, amorphous phases outperformed their crystalline counterparts, showing lower activation energies and higher conductivities. Our results provide concrete candidate materials and establish structural disorder as a powerful design principle for next-generation SSEs.