Efficient source and mask optimization based on interpretable hypergraph auto-encoding network.

Area of Science:

• Semiconductor Manufacturing
• Computational Lithography
• Deep Learning Applications

Background:

• Source and Mask Co-optimization (SMO) is crucial for advanced lithography.
• Current pixelated SMO methods are computationally intensive and time-consuming.
• Efficient SMO techniques are needed to improve lithography process windows and image fidelity.

Purpose of the Study:

• To develop an efficient source and mask co-optimization (SMO) method.
• To leverage hypergraph deep learning for accelerated mask optimization.
• To enhance lithography image fidelity and process robustness.

Main Methods:

• Developed a novel mask clip selection method using sparse signal reconstruction.
• Employed a fast gradient-based algorithm for source pattern optimization.
• Introduced a hypergraph auto-encoding network for accelerated mask optimization, utilizing layout features and a lithography physical imaging model for self-supervised training.

Main Results:

• The proposed method significantly improves lithography image fidelity.
• Enhanced process robustness was demonstrated through simulations.
• Achieved superior computational efficiency compared to existing SMO techniques.

Conclusions:

• The hypergraph deep learning framework offers an efficient solution for SMO.
• The developed methods accelerate the optimization process while maintaining high fidelity and robustness.
• This work presents a significant advancement in computational lithography.