# Pushing the Limits of Inverse Lithography with Generative Reinforcement Learning

arXiv: https://arxiv.org/abs/2602.19027 · HTML: https://mlsanity.com/paper/2602.19027/

## Key facts

- Cited by 1 papers in the map (data through 14 Oct 2026).
- Builds on 6 of its 20 references that are in the map.
- First posted on arXiv 22 Feb 2026.

## Builds on (most cited first, 6 of 6)

- [A Style-Based Generator Architecture for Generative Adversarial Networks](https://mlsanity.com/paper/1812.04948/) (2018)
- [Image-to-Image Translation with Conditional Adversarial Networks](https://mlsanity.com/paper/1611.07004/) (2016)
- [DAMO: Deep Agile Mask Optimization for Full Chip Scale](https://mlsanity.com/paper/2008.00806/) (2020)
- [Generic Lithography Modeling with Dual-band Optics-Inspired Neural Networks](https://mlsanity.com/paper/2203.08616/) (2022)
- [ILILT: Implicit Learning of Inverse Lithography Technologies](https://mlsanity.com/paper/2405.03574/) (2024)
- [GPU-Accelerated Inverse Lithography Towards High Quality Curvy Mask Generation](https://mlsanity.com/paper/2411.07311/) (2024)

## Built on by (most cited first, 1 of 1)

- [LithoGRPO: Fast Inverse Lithography via GRPO Reinforced Flow Matching](https://mlsanity.com/paper/2606.00228/) (2026)

## Abstract (arXiv)

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 textttLithoBench 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 quality. We also present over 20% EPE improvement on textttICCAD13 contest cases with 3× speedup over the SOTA numerical ILT solver. By learning to propose ILT-friendly initializations, our approach mitigates non-convexity and advances beyond what traditional solvers or GenAI can achieve.