Cation-Disordered High-Entropy Garnet Structures as Solid-State Electrolytes for All-Solid-State Batteries: Machine Learning-Driven Discovery

Exploring the vast chemical space of high-entropy (HE) solid-state electrolytes (SSEs) has become a highly active area of battery studies owing to the exceptional performance of all-solid-state batteries (ASSBs) with higher energy density and improved safety. The compositional complexity and extensive chemical space inherent to HE SSEs pose significant challenges for their investigation via conventional methodologies such as experimental approaches and density functional theory (DFT) calculations. In this study, we propose a novel material screening methodology aimed at accelerating the exploration of promising HE SSEs while maintaining reasonable computational costs and efficiency. Specifically, we introduce a machine learning (ML)-based screening framework to evaluate 4348 cation-disordered high-entropy (CDHE) garnet-type SSE candidates derived from the well-known Li7La3Zr2O12 structure. First, the ML-based surrogate model was incorporated to screen electron-conducting (bandgap (Eg) Ehull) > 0.040 eV/atom). Second, the crystal Hamiltonian graph neural network (CHGNet), a recently developed machine learning interatomic potential, was adapted to find the most stable atomic configuration of each composition with the following further DFT relaxations. Third, the CHGNet potential was fine-tuned on the CDHE garnet-type materials. Then, the elastic properties were calculated with retrained CHGNet to confirm whether the candidate materials could suppress dendrite formation and exhibit interfacial stability. Finally, molecular dynamics (MD) simulations equipped with fine-tuned CHGNet potential were conducted to investigate the lithium diffusion characteristics of CDHE garnet-type materials, resulting in the confirmation of three promising CDHE garnet-type candidates with ionic conductivities exceeding 10–4 S/cm at room temperature.