Model-informed deep learning for computational lithography with partially coherent illumination.

Area of Science:

• Semiconductor Manufacturing
• Computational Lithography
• Deep Learning Applications

Background:

• Computational lithography is crucial for optimizing optical lithography systems.
• High computational complexity is a major challenge in current computational lithography methods.
• Enhancing imaging fidelity and computational efficiency remains a key research objective.

Purpose of the Study:

• To propose a model-informed deep learning (MIDL) approach for computational lithography.
• To improve the computational efficiency and image fidelity of lithography systems with partially coherent illumination (PCI).
• To address the computational cost and complexity associated with traditional methods.

Main Methods:

• Developed a dual-channel network structure for MIDL, derived from an approximate compact imaging model of PCI lithography.
• Implemented an unsupervised training method using an accurate lithography imaging model to eliminate labeling costs.
• Leveraged deep learning principles integrated with physical models for lithography optimization.

Main Results:

• The MIDL approach significantly enhances computational efficiency compared to conventional methods.
• Demonstrated substantial improvements in the image fidelity of lithography systems under PCI.
• The dual-channel network structure effectively overcomes the vanishing gradient problem, boosting prediction capacity.

Conclusions:

• MIDL offers a promising solution for accelerating computational lithography.
• The proposed method achieves superior performance in both speed and accuracy for PCI lithography.
• This research paves the way for more efficient and effective semiconductor manufacturing processes.