Atomistic Simulation of HF Etching Process of Amorphous Si₃N₄ Using Machine Learning Potential.

Area of Science:

• Materials Science
• Computational Chemistry
• Semiconductor Physics

Background:

• Atomistic understanding of dry-etching is vital for advanced semiconductor devices.
• Molecular dynamics (MD) simulations are powerful but limited by unreliable force fields for etching.
• Accurate simulation of amorphous Si3N4 etching with HF is needed.

Purpose of the Study:

• To develop an accurate neural network potential (NNP) for atomistic simulations of amorphous Si3N4 etching with HF molecules.
• To establish a computational framework for scale-bridging from atomistic simulations to continuum models.
• To enable more precise and efficient design of semiconductor etching processes.

Main Methods:

• Development of a neural network potential (NNP) trained on diverse datasets including baseline, reaction-specific, and general structures.
• Iterative refinement of the NNP through comparison with density functional theory (DFT) results.
• Execution of MD simulations using the trained NNP to analyze etching phenomena.
• Development of a continuum model based on MD simulation outputs.

Main Results:

• The trained NNP accurately simulates amorphous Si3N4 etching with HF at an atomistic level.
• Simulations reveal key etching processes: preferential sputtering, surface modification, etching yield, and threshold energy.
• MD simulations provide detailed analysis of etching product distribution.
• A derived continuum model effectively reproduces surface composition from MD simulations.

Conclusions:

• The developed NNP and computational framework facilitate accurate atomistic etching simulations.
• This approach enables scale bridging, connecting atomistic details to continuum behavior.
• The work paves the way for enhanced semiconductor device design and manufacturing precision.