LithoGRPO: Fast Inverse Lithography via GRPO Reinforced Flow Matching

2026-05-29

AI Summary

This work addresses pattern distortion in optical lithography caused by diffraction effects by proposing a novel inverse lithography mask optimization method that integrates flow-matching generative models with GRPO reinforcement learning. It is the first to unify flow matching and reinforcement learning for mask synthesis, enabling efficient generation of high-fidelity and diverse mask patterns under complex process constraints. The approach introduces a physics-informed reward function and a fast hotspot-counting algorithm, which together significantly accelerate manufacturability evaluation while preserving ranking consistency. Experimental results demonstrate that the proposed method outperforms existing optimization and learning-based approaches in both mask quality and computational efficiency, achieving state-of-the-art performance.

Abstract

In semiconductor manufacturing, lithography projects circuit layouts onto silicon wafers through an optical mask. As circuit features shrink below the wavelength of light, optical diffraction causes the printed patterns to deviate from their intended layouts. Inverse Lithography Technology (ILT) addresses this challenge by generating optimized masks that enhance the fidelity of pattern transfer onto wafers. While ILT resembles an image synthesis task, its reliance on explicit physical metrics for mask evaluation limits the applicability of existing generative models. We introduce LithoGRPO, an ILT framework that integrates the flow-matching paradigm with GRPO-based reinforcement learning (RL) fine-tuning, enabling efficient exploration of diverse masks for a given target layout. Unlike purely generative or optimization-based approaches, RL in LithoGRPO exploits the explicitly defined, physics-based reward function of ILT, enabling optimization under complex, process-aware constraints. To the best of our knowledge, this is the first framework that unifies flow matching and RL for mask optimization. To improve RL sampling efficiency, we propose a fast shot-counting algorithm for manufacturability evaluation, achieving over 130x speedup while preserving the mask ranking of the traditional shot-count metric. Extensive experiments demonstrate that LithoGRPO achieves state-of-the-art performance over both optimization-based and learning-based methods, while maintaining efficient mask generation.

Problem

• Inverse Lithography
• Optical Diffraction
• Mask Optimization
• Semiconductor Manufacturing
• Pattern Fidelity

Innovation

• Inverse Lithography Technology
• Flow Matching
• Reinforcement Learning
• GRPO
• Shot Counting