Data-driven design of NASICON-type electrodes using graph-based neural networks

Abstract

Sodium superionic conductor (NASICON)-type cathode materials are considered promising candidates for high-performance sodium-ion batteries (SIBs) because of the abundance and low cost of raw materials. However, NASICON-type cathodes suffer from low capacities. This limitation can be addressed through the activation of sodium-excess phases, which can enhance capacities up to theoretical values. Thus, this paper proposes the use of transition metal (TM)-substituted Na3V2(PO4)2F3 (NVPF) to induce sodium-excess phases. To identify suitable doping elements, an inverse design approach is developed, combining machine learning prediction and density functional theory (DFT) calculations. Graph-based neural networks are used to predict two crucial properties, i. e., the structural stability and voltage level. Results indicate that the use of TM-substituted NVPF materials leads to about 150 % capacity enhancement with reduced time and resource requirements compared with the direct design approach. Furthermore, DFT calculations confirm improvements in cyclability, electronic conductivity, and chemical stability. The proposed approach is expected to accelerate the discovery of superior materials for battery electrodes. This work presents innovative NVPF cathodes customized for excess sodium storage. By integrating synthesizability, electrode structural stability, and operating voltage window as critical screening features, we devise a unique inverse design strategy combining machine learning and electronic structure calculations. The proposed materials theoretically demonstrate a substantial 150 % increase in capacity and enhancements in various performance aspects, indicating the potential to greatly accelerate the exploration of advanced battery electrode materials. image