Amorphous silicon and silicon oxides (SiOₓ, 0 ≤ x ≤ 1) are promising anode materials for lithium-ion batteries due to their high theoretical energy capacity. However, their practical implementation is hindered by substantial volume changes during cycling. A detailed atomic-level understanding is essential to improve their stability and performance. This project focuses on developing accurate and transferable machine learning force fields (MLFFs) for amorphous SiOₓ. Initial amorphous structures were generated using ab initio molecular dynamics (AIMD) simulations with the Vienna Ab initio Simulation Package (VASP) via a melt-and-quench approach. Different quench rates were investigated to minimize training errors and improve MLFF reliability. The resulting MLFFs significantly reduce the computational cost compared to AIMD simulations, enabling simulations at larger length scales and longer timescales. This approach allows efficient investigation of structural evolution and lithiation mechanisms in Si-based anodes, supporting the design of more durable, high-capacity lithium-ion anode materials.
