Fast inverse lithography based on a model-driven block stacking convolutional neural network.

Area of Science:

• Semiconductor Manufacturing
• Computational Lithography
• Deep Learning Applications

Background:

• Optical proximity correction (OPC) is vital for advanced lithography to mitigate optical proximity effects (OPE).
• Conventional pixel-based OPC methods often result in complex mask patterns that are difficult and costly to manufacture.
• Existing methods may require extensive labeled datasets, increasing training time and complexity.

Purpose of the Study:

• To develop a novel, manufacturing-conducive inverse lithographic approach for OPC.
• To expedite mask pattern generation using a deep learning framework.
• To enhance the diversity and feasibility of mask patterns in advanced lithography.

Main Methods:

• A model-driven, block stacking deep learning framework is employed for inverse lithographic OPC.
• The approach is founded on vector lithography modeling, reducing the need for large labeled datasets.
• A wave function collapse algorithm is utilized to generate diverse target mask patterns.

Main Results:

• The proposed end-to-end approach effectively generates manufacturing-friendly mask patterns.
• The deep learning framework significantly speeds up the mask generation process.
• Numerical experiments confirm the approach's capability to manage mask complexity in advanced OPC.

Conclusions:

• The novel inverse lithographic OPC method enhances the feasibility and economic viability of advanced lithography.
• This approach addresses the manufacturing challenges associated with conventional OPC techniques.
• The integration of deep learning and wave function collapse offers a powerful paradigm for future mask design.