Simultaneous Exploration of Candidates for Electrode and Electrolyte Materials for All-Solid-State Batteries Using Predicted Rating from a Recommender System

Discovering unknown materials demands significant time and resources due to the reliance on repeated trial-and-error experimentation. Consequently, numerous machine-learning approaches have been employed to accelerate the practical search for materials. In this study, we focused on the top 20 candidate materials in the Li-M-M′-O system with the highest predicted ratings identified by the recommender system. Through ambient and high-pressure solid-state synthesis, seven compositions were discovered from the nominated candidates. The discovery rate of materials improves with increasing the predicted rating, and it has been demonstrated that this predicted rating correlates with the feasibility of experimental material synthesis. Among the discovered compositions, Li_5+xGe_2–xAl_1+xO₈ with x = 0.5 exhibited the highest ionic conductivity of approximately 1 × 10^–5 S cm^–1 at 473 K. In addition, LiMo_4/3V_2/3O₆, whose composition was optimized based on the synthesis results, exhibited mixed ion and electron conductivity and showed lithium (de)intercalation activity. Searching using the predicted rating of the recommender system as a guideline improved the efficiency of material exploration. Finally, solid electrolytes and electrode materials were discovered simultaneously for all-solid-state batteries.