Machine learning driven insights into lithiation mechanisms at the silicon-graphite interface within composite electrode

Abstract

Silicon-graphite composite electrodes are considered as ideal candidates for high-performance lithium-ion batteries due to their outstanding theoretical specific capacity. A comprehensive understanding of the lithiation mechanism at the silicon-graphite interface is crucial for improving battery performance, particularly in terms of enhancing cycling stability and reversible capacity. In this work, we have trained a machine learning potential and performed large-scale molecular simulations to investigate the lithiation mechanisms at individual a-Si, a-Si/graphite-I, and a-Si/graphite-II interfaces. The results demonstrate that the trained potential achieves accuracy comparable to density functional theory, making it well-suited for high-precision simulations of lithiation dynamics in large-scale systems. Our findings reveal that, compared to the reference a-Si system, graphite effectively mitigates direct lithium-silicon contact, reducing interfacial reactions and promoting a diffusion-dominated lithiation process. Notably, the a-Si/graphite-I interface exhibits the fastest lithiation rate while efficiently suppressing the diffusion of silicon atoms toward the lithium source. This confinement facilitates the formation of a dense lithiated phase, significantly minimizing silicon loss and enhancing both the reversible capacity and cycling stability of the electrode. Our study provides valuable theoretical insights for the performance enhancement of silicon-graphite composite electrode materials.