Machine-learning-assisted discovery of boron-doped graphene with high work function as an anode material for Li/Na/K-ion batteries

Abstract

Work function (WF) modulation is a crucial descriptor for carbon-based electrodes in optoelectronic, catalytic, and energy storage applications. Boron-doped graphene is envisioned as a highly promising anode material for alkali metal-ion batteries (MIBs). However, due to the large structural space concerning various doping concentrations, the lack of both datasets and effective methods hinders the discovery of boron-doped graphene with a high WF that generally leads to strong adsorption. Herein, we propose a machine-learning-assisted approach to discover the target, where a Crystal Graph Convolutional Neural Network was developed to efficiently predict the WF for all possible configurations. As a result, the B₅C₂₇ structure is found to have the highest WF in the entire space containing 566 211 structures. In addition, it is revealed that the adsorption energy of alkali metals is linearly related to the WF of the substrate. Therefore, the screened B₅C₂₇ is investigated as an anode for Li/Na/K-ion batteries, and it possesses a higher theoretical specific capacity of 2262/2546/1131 mA h g⁻¹ for Li/Na/K-ion batteries compared with that of pristine graphene and other boron-doped graphene. Our work provides an effective way to locate possible high-WF structures in heteroatom-doped systems, which may accelerate future screening of promising adsorbents for alkali metals.