Machine-learning-aided screening of inorganic lithium solid-state electrolytes with a wide electrochemical window

Abstract

We construct an electrochemical window (ECW) dataset of over 16 000 Li-containing compounds using a thermodynamic approach for solid-state electrolytes (SSEs). A data-driven ECW prediction framework is developed, with the classification model achieving >0.98 accuracy and the regression model yielding mean absolute errors of 0.19/0.21 V for the left/right ECW limits. Screening 69 243 compounds identifies promising SSE candidates, enabling accelerated discovery of electrochemically stable materials.