Machine learning force field for thermal oxidation of silicon

Area of Science:

• Materials Science
• Computational Chemistry
• Semiconductor Physics

Background:

• Silicon (Si) and silicon dioxide (SiO2) are foundational to the semiconductor industry.
• Down-scaling of nanoelectronic devices necessitates ultra-thin oxide layers and a deeper understanding of the Si/SiO2 interface.
• Classical force fields lack accuracy, while ab initio methods are computationally expensive.

Purpose of the Study:

• To train a machine learning force field (MLFF) for simulating the dry thermal oxidation of silicon substrates.
• To achieve a balance between accuracy and computational cost in modeling the Si/SiO2 interface.
• To provide a publicly available MLFF for the research community.

Main Methods:

• Density functional theory (DFT) calculations were used to generate training data for the MLFF.
• A novel MLFF was developed and trained using DFT-generated data.
• Simulations of the dry thermal oxidation process were performed using the trained MLFF.

Main Results:

• The MLFF accurately reproduces atomic configurations at the Si/SiO2 interface, consistent with ab initio simulations and experimental data.
• MLFF simulations show vastly improved results compared to classical and reactive force fields.
• The developed MLFF offers a computationally efficient yet accurate approach to modeling Si oxidation.

Conclusions:

• Machine learning force fields represent a significant advancement for simulating semiconductor fabrication processes.
• The trained MLFF provides a valuable tool for understanding and optimizing the Si/SiO2 interface in next-generation electronics.
• Public availability of the MLFF promotes further research and development in nanoelectronics.