Unveiling potential lithium ionic conductors through machine learning and atomic simulation approaches

Abstract

This study delves into the critical realm of solid-state electrolytes (SSE) to address the safety concerns associated with conventional liquid electrolytes in lithium-ion batteries. Specifically, machine learning (ML) method was used to expedite the discovery of novel SSE materials. A comparative analysis involving random forest, support vector regression, XGBoost, and compositionally-restricted attention-based network models showcases the efficacy of the XGBoost model. This study extends its impact by integrating Ceder's statistical model of ionic substitution, resulting in the creation of 18,155 compounds. A meticulous screening process, guided by criteria such as high ionic conductivity, cost-effectiveness, and low toxicity, culminated in the identification of 287 potential lithium-ion conductors. In addition, we also employed density functional theory calculation for the selected candidate. This comprehensive approach exemplifies the synergy of ML and computational methodologies in accelerating the discovery and screening of materials for SSE applications, thereby contributing valuable insights to the ongoing advancements in energy storage technologies.