Massive Screening of High Capacity Battery Cathode Active Materials Using Deep Neural Networks. 심층신경망을 이용한 고용량 배터리 양극 활물질 대량탐색.

Abstract

Ever since the first commercialization of a lithium-ion battery by Sony in 1991, the energy storage system has shaped the lifestyles of millions of people. For example, the energy storage system significantly contributed to the popularization of portable electronics such as smartphones and laptop computers. Besides its successful application to small-size portable devices, high capacity batteries have also received great attention as a promising solution to transitioning renewable energy society. Innumerable research efforts have been devoted to developing high energy density battery materials for transportation and housing applications, but research progress has not been made fast enough to meet the high capacity demands from markets. Designing battery materials require optimization on various engineering parameters such as high voltage, high energy density, and low volume change during charge/discharge cycles. Satisfying all those design criteria is indeed a very complex problem to solve at once, and this nature of complexity has been an enormous barrier to overcome for battery researchers in the past decades. In this study, an accelerated materials discovery using deep neural networks and materials database is introduced. A total of four battery properties were trained with the proposed deep neural network and showed the significant prediction improvement over the reference network. The proposed network even approaches to near-zero loss whereas the reference network overfits during the training phase. The fast and accurate prediction performance of the proposed network is well illustrated in this study, and the pace and scale of novel battery materials discovery would elevate to the next unprecedented level with the proposed design algorithm.