Discovery of Effective Halide Solid Electrolytes for Solid-State Rechargeable Batteries via Machine Learning and DFT Calculations

Halide solid electrolytes (SEs) are a strong candidate for next-generation lithium-based solid-state batteries for their potential to possess a balance of key properties including ionic conductivity, mechanical properties, and electrochemical stability window (ESW) and can be synthesized using environmentally friendly processes. However, there is a lack of halides simultaneously fulfilling all the mentioned key properties, and searching for the right candidates via experiments is proven challenging. In this work, we develop a computational approach combining machine learning (ML) and DFT calculations, to discover promising halide SEs that satisfy several bulk properties via multiproperty predictions. Various ML and deep learning (DL) models are compared to predict ionic conductivity, bulk and shear moduli, and ESW. The CatBoost, Light Gradient Boosting (LGBM), and Skorch Neural Network (NN) models are found to yield high prediction accuracies for the mentioned properties, with minimum average classification accuracies and average R 2 scores exceeding 80% and 0.70, respectively. DFT verifications are performed on Rb 2 LiBiCl 6, LiHF 2, and Rb 2 LiAlF 6, with the results suggesting Rb 2 LiAlF 6 as a promising candidate for high voltage battery applications. Overall, we demonstrate that the current ML + DFT approach is useful in screening potential halide solid-state electrolytes that can satisfy several key SE properties.