Accelerating the Search for New Solid Electrolytes—Exploring Vast Chemical Space with Machine Learning-Enabled Computational Calculations

Advancements in all-solid-state lithium batteries, crucial for enhancing safety and energy density, depend heavily on the discovery of new solid electrolytes (SEs). The practical application of SEs is often limited by their low ionic conductivities and chemical instabilities. In response, we report a machine learning (ML) assisted high-throughput virtual screening (HTVS) methodology to identify new SE materials. This method broadens the chemical space for potential SE materials by substituting elements in known structures, generating a vast number of candidate materials. These candidates were initially screened using ML models trained on existing materials databases to predict properties such as formation energy, electrochemical stability window, and Li-ion conductivity. By choosing ML models that utilize composition or modified X-ray diffraction patterns, which do not require structure optimization, we efficiently narrowed down the list of potential SE materials from the vast chemical space. A select number of candidate materials derived from this screening process were subsequently validated through density functional theory (DFT) calculations and Ab-initio molecular dynamics (AIMD) simulations to confirm their stability and conductive properties. The results of the validation identified several promising oxysulfide SE materials that exhibit both high Li-ion conductivity and enhanced stability, which are required for effective SEs. In summary, we leveraged an ML-assisted HTVS strategy to narrow down the vast chemical space and highlighted the capability of this strategy to identify novel and promising SE materials with reduced computational costs.