LithoGRPO: Fast Inverse Lithography via GRPO Reinforced Flow Matching
Researchers have developed LithoGRPO, a novel framework for Inverse Lithography Technology (ILT) that combines flow matching with GRPO-based reinforcement learning. This approach allows for the efficient generation of optimized masks used in semiconductor manufacturing to improve pattern transfer fidelity. LithoGRPO uniquely leverages a physics-based reward function within the reinforcement learning process, enabling optimization under complex, process-aware constraints. The system also incorporates a fast shot-counting algorithm for manufacturability evaluation, achieving significant speedups while maintaining mask ranking. AI
IMPACT This AI framework could accelerate semiconductor manufacturing by improving the precision of circuit pattern transfer.