Accelerating discovery and design of high-performance solid-state electrolytes: a machine learning approach

Abstract

Solid-state batteries (SSBs) have the potential to fulfil the increasing global energy requirement, outperforming their liquid electrolyte counterparts. However, the progress in SSB development is hindered by the conventional approach of screening solid-state electrolytes (SSEs), which relies on human knowledge, introducing biases and requiring a time-consuming, resource-intensive trial-and-error process. As a result, a wide range of promising Li-containing structures remain unexplored. To accelerate the search for optimal SSE materials, it is crucial to understand the chemical and structural factors that govern ion transport within a crystalline lattice. We utilize logistic regression-based machine learning (ML) to identify and quantify key physio-chemical features influencing ion mobility in NASICON compounds. The dopant-related features that influence the ionic conductivity are further used to design doped SSEs for Li-ion batteries. Our innovative design approach results in NASICON electrolytes with significantly improved migration barriers and ionic conductivity, validated through density functional theory-based calculations. Specifically, this approach successfully identifies two doped SSEs with high ionic conductivity: Li₂Mg_0.5Ge_1.5(PO₄)₃ and Li_1.667Y_0.667Ge_1.333(PO₄)₃. Li₂Mg_0.5Ge_1.5(PO₄)₃ has the lowest barrier energy of 0.261 eV, surpassing the previously best-known doped material, Li_1.5Al_0.5Ge_1.5(PO₄)₃ (LAGP), which has a migration barrier of 0.37 eV. Additionally, Li_1.667Y_0.667Ge_1.333(PO₄)₃ is identified to have the second-lowest migration barrier height of 0.365 eV. By focusing the training of the machine learning model on a specific class of materials, our approach significantly reduces the time, resources, and size of the dataset required to discover novel materials with targeted properties. This methodology is readily adaptable to the design of materials in various other fields, including catalysis and structural materials.