FWave: a fast 3D resist model with dual-branch architecture for EUV lithography simulation

Area of Science:

• Semiconductor Manufacturing
• Computational Lithography
• Materials Science

Background:

• Rigorous physical resist models are accurate but computationally intensive, limiting simulation speed.
• Extreme ultraviolet (EUV) lithography introduces stochastic effects, causing irregular pattern edges not well-captured by traditional models.
• Predicting inhibitor concentration during post-exposure bake (PEB) is crucial for accurate lithography simulation.

Purpose of the Study:

• To develop a fast and accurate 3D resist model for predicting inhibitor concentration in EUV lithography.
• To address the computational expense and stochastic effects inherent in EUV lithography simulations.
• To improve critical dimension (CD) control in contact hole patterning.

Main Methods:

• Proposed FWave, a dual-branch neural network architecture.
• Utilized a Fourier Neural Operator (FNO) branch for global feature extraction.
• Employed a Discrete Wavelet Transform (DWT) branch to capture edge variations from stochastic effects.
• Implemented a spatially weighted mean squared error (WMSE) loss function for precise contact hole region fitting.

Main Results:

• FWave achieved sub-nanometer critical dimension (CD) errors in EUV contact hole simulations.
• Demonstrated a speedup of approximately 2000x compared to rigorous physical models.
• Successfully integrated global feature capture and edge variation analysis.

Conclusions:

• FWave offers a computationally efficient and accurate solution for EUV lithography simulations.
• The model effectively handles stochastic effects and improves CD accuracy.
• FWave represents a significant advancement in accelerating semiconductor process modeling.