A novel cathode material design for Mg-ion batteries using deep learning and first-principles calculations

Abstract

The demand for high energy density energy storage systems is ever increasing due to the growing market demands from the transportation and housing sectors. Although notable energy density growth is achieved in the current market-dominant cathodes for Li-ion batteries, the long-term supply of strategic materials such as cobalt and natural graphite is under debate due to the geopolitical uncertainty in countries of origin. The intrinsic solution to the problem mentioned above is to find alternate cathode chemistries comparable to or surpassing the battery performance of conventional cathode chemistries. However, finding novel cathode chemistry is an exceptionally challenging and time-consuming process since it is practically impossible to explore an infinite number of chemical spaces through experimental procedures. This dissertation proposes the end-to-end method of massive cathode candidate search using deep learning and theoretical validation using the physics-based density functional theory. The self-attention-based deep learning model is proposed and analyzed in the series of case studies of representative cathode chemistries, LiCoO2 and LiFePO4. The proposed method identifies a novel class of garnet cathode, Mg3V2(SiO4)3, for Mg-ion batteries, which is backed by a thorough investigation of theories of physics and electrochemistry. The method proposed in this dissertation is scalable and also expandible to other materials domain, which has a high potential to open up new horizons for novel materials development.