Publication date: 10th April 2024
High-throughput computational screening and machine learning offer significant potential for exploring diverse chemical compositions and identifying novel inorganic solids [1,2]. Despite these advancements, identifying high-performance materials with specific properties remains challenging. In response, we propose "Materials Discovery by Interpretation (MDI)," a hybrid approach designed to efficiently develop high-performance materials by leveraging both machine learning and interpretation. We applied MDI to explore fast proton-conducting oxides at an intermediate temperature of 300C. Our machine learning model identifies new host materials for proton conductors, nominating rhombohedral SrSnO3 [1]. In the first attempt, we observed proton conductivity in Sc-doped SrSnO3, which exhibited a relatively high bulk proton conductivity of 0.001 S/cm at 380C. Our interpretation of fast proton conductors focuses on symmetric cubic oxides with high Sc solubilities [3]. Realizing our interpretation in a stannate-host perovskite has led to heavily Sc-doped BaSnO3. Using this methodology, we discovered proton-conducting oxides exhibiting 0.01 S/cm conductivity at 300C and high chemical stability [4].
This work is supported by JST CREST (JPMJCR18J3) and GteX (JPMJGX23H7 and JPMJGX23H0).