| 초록 |
Solid electrolytes are considered as alternatives to organic liquid electrolytes in lithium-ion battery applications because they are safer, longer-lasting and more energy-dense. However, solid electrolytes discovered to date have demonstrated high conductivity but poor stability or vice versa. Therefore, it is essential to develop new materials that satisfy both conductivity and stability conditions. In this work, we performed high-throughput screening on chemical space, larger than any existing database, to discover new solid electrolytes. To achieve this goal, we used supervised and unsupervised machine learning approaches to predict various chemical properties of unoptimized crystal structures. We initially generated broad chemical space by substituting atoms. Second, the properties of candidate materials are predicted using supervised machine learning. To predict Li-ion conductivity, unsupervised machine learning was used to classify highly Li-conductive materials. Finally, all the candidate materials were explicitly evaluated by performing density functional theory calculations. We expect this work provides promising candidate materials for developing new solid electrolytes. |
