Advanced Machine Learning Interatomic Potential for Accelerated Atomistic Simulations of Lithiation Dynamics in...

Area of Science:

• Materials Science
• Electrochemistry
• Computational Chemistry

Background:

• Silicon anodes offer high capacity for lithium-ion batteries but suffer from significant volume expansion during lithiation, limiting cycle life.
• Understanding the atomic-scale dynamics of lithiation in silicon is crucial for designing stable anodes.

Purpose of the Study:

• To develop an efficient computational model for simulating lithiation dynamics in silicon-carbon anodes.
• To investigate the role of carbon shell thickness on volume expansion and lithium distribution.

Main Methods:

• Development of a neuroevolution potential (NEP) model for atomistic simulations.
• Utilizing a direct sampling strategy to optimize training data size.
• Performing large-scale simulations of silicon-carbon core-shell structures during lithiation.

Main Results:

• The NEP model achieved ~70,000x speedup over ab initio molecular dynamics with near-density functional theory accuracy.
• A ~4 nm carbon layer was found to minimize volume expansion to <1% and enable an inner-expansion/outer-locking mechanism.
• The model accurately predicted atomic forces, radial distribution functions, and lithium diffusivities across various structural configurations.

Conclusions:

• Machine learning interatomic potentials are powerful tools for bridging computational feasibility and atomistic accuracy in materials design.
• A ~4 nm carbon shell is optimal for suppressing volume expansion in silicon-carbon anodes, enhancing cycling stability.
• This work provides fundamental insights for designing next-generation high-performance silicon-carbon anodes.