From Composition to Ionic Conductivity: Machine Learning-Guided Discovery and Experimental Validation of...

The discovery of solid-state electrolytes (SSEs) with high lithium-ion conductivities is critical for advancing all-solid-state batteries. However, prior efforts have largely focused on structure-driven design. This study presents a composition-based machine learning framework, Elements-To-Ionics (E2I), for the accurate prediction and optimization of the ionic conductivities of argyrodite-type SSEs using only their elemental compositions. Guided by these predictions, a series of Si-Sn, Ge-Si, and Ge-Sn co-substituted argyrodites are synthesized. Li_6.7Ge_0.595Si_0.105P_0.3S₅I achieves the highest ionic conductivity (7.2 × 10^-3 S cm^-1) with a low activation energy (0.20 eV). Using hot-pressing to optimize the conductivity, values comparable to those of Li₁₀GeP₂S₁₂-type superionic conductors are achieved (>10^-2 S cm^-1). The developed model reliably identifies both high- and low-conductivity regions and significantly reduces the experimental workload. These results highlight the potential of composition-based informatics for accelerating the discovery of high-performance SSEs within complex chemical spaces, and provide a valuable methodology for the development of next-generation solid-state battery technologies.