Deep-neural-network-guided screening for material properties of secondary batteries and permanent magnets
|
이주형 GIST The demand for high performance energy materials is ever increasing for various applications including portable electronics, electric zero-emission vehicles and stationary energy storage systems. Among others, tremendous research efforts have been devoted to discovering next-generation battery cathode materials for secondary batteries and environmentally benign rare-earth-free permanent magnets. However, finding a magic chemical compound that satisfies all design criteria such as high energy density, flat voltage curve, minimal volume change, low ion migration barrier and high Curie temperature makes a highly complex problem due to the vast materials space to explore. In this talk, I will describe our recent development that implements efficient and accurate deep neural network (DNN) models to address this important issue, and present results of a massive, accelerated screening of more than 32,000 candidate compounds. As for battery cathode materials, a total of three alternative cathode active materials for Mg-ion batteries are identified through DNN and further investigated with ab initio density functional theory calculations. Our calculations demonstrate that the proposed candidates not only attain higher energy density than known cathode materials but also maintain low ion-migration energy barrier as well as minimal volume change. In addition, our second DNN model predicts the Curie temperature of various magnetic materials more accurately than previously reported models. It is expected that these results will play an important role to guide an expedited design of novel materials with desired properties. |