![[2602.19027] Pushing the Limits of Inverse Lithography with Generative Reinforcement Learning](https://arxiv.org/static/browse/0.3.4/images/arxiv-logo-fb.png){kind=logo}
[2602.19027] Pushing the Limits of Inverse Lithography with Generative Reinforcement Learning
Summary
This article presents a novel approach to inverse lithography using generative reinforcement learning, significantly improving mask quality and throughput in semiconductor manufacturing.
Why It Matters
The advancements in inverse lithography are crucial for semiconductor manufacturing, where precision and efficiency directly impact production costs and technology capabilities. This research addresses optimization challenges, potentially leading to faster and more effective manufacturing processes.
Key Takeaways
- Introduces a hybrid framework combining generative AI and reinforcement learning for mask synthesis in lithography.
- Achieves significant reductions in edge placement error (EPE) and increases throughput compared to traditional methods.
- Demonstrates the ability to escape non-convex optimization traps, enhancing mask quality.
- Utilizes a pretraining and fine-tuning approach to improve the generative model's performance.
- Validates the method on established datasets, showcasing its effectiveness in real-world applications.
Computer Science > Machine Learning arXiv:2602.19027 (cs) [Submitted on 22 Feb 2026] Title:Pushing the Limits of Inverse Lithography with Generative Reinforcement Learning Authors:Haoyu Yang, Haoxing Ren View a PDF of the paper titled Pushing the Limits of Inverse Lithography with Generative Reinforcement Learning, by Haoyu Yang and Haoxing Ren View PDF HTML (experimental) Abstract:Inverse lithography (ILT) is critical for modern semiconductor manufacturing but suffers from highly non-convex objectives that often trap optimization in poor local minima. Generative AI has been explored to warm-start ILT, yet most approaches train deterministic image-to-image translators to mimic sub-optimal datasets, providing limited guidance for escaping non-convex traps during refinement. We reformulate mask synthesis as conditional sampling: a generator learns a distribution over masks conditioned on the design and proposes multiple candidates. The generator is first pretrained with WGAN plus a reconstruction loss, then fine-tuned using Group Relative Policy Optimization (GRPO) with an ILT-guided imitation loss. At inference, we sample a small batch of masks, run fast batched ILT refinement, evaluate lithography metrics (e.g., EPE, process window), and select the best candidate. On \texttt{LithoBench} dataset, the proposed hybrid framework reduces EPE violations under a 3\,nm tolerance and roughly doubles throughput versus a strong numerical ILT baseline, while improving final mask quali...