Inverse lithography technology based on a physics-guided reinforcement learning framework.

Area of Science:

• Semiconductor Manufacturing
• Computational Lithography
• Artificial Intelligence in Engineering

Background:

• Inverse lithography technology (ILT) is crucial for enhancing lithography system resolution and chip manufacturing yields.
• Practical ILT applications face challenges due to computational complexity and mask manufacturability.
• Existing ILT methods require significant computational resources and struggle with mask fabrication constraints.

Purpose of the Study:

• To propose a novel ILT framework utilizing reinforcement learning (RL) to overcome computational and manufacturability limitations.
• To improve the efficiency and practical applicability of ILT in semiconductor manufacturing.
• To integrate physical prior knowledge into the ILT process for enhanced mask design.

Main Methods:

• Developed an ILT framework employing reinforcement learning (RL) agents.
• Integrated physical information through interactions between RL agents and forward lithography simulations.
• Utilized a keypoint sequence derived from the inverse lithography gradient map for RL agent actions.
• Employed a Manhattan correction unit to ensure mask manufacturability.

Main Results:

• The RL-based ILT framework demonstrated improved computational efficiency compared to traditional methods.
• Enhanced mask manufacturability was achieved by incorporating physical constraints.
• Simulation results confirmed the superiority of the proposed RL-based approach in ILT.
• The keypoint sequence-based action strategy ensured efficient computation.

Conclusions:

• The proposed RL-based ILT framework offers a significant advancement in semiconductor manufacturing.
• This approach effectively addresses the computational complexity and mask manufacturability issues in ILT.
• Reinforcement learning provides a powerful tool for optimizing lithography processes and improving chip yields.