Efficient SRAF Generation via Diffusion Models

In semiconductor manufacturing, optical proximity correction and sub-resolution assist features (SRAF) are critical techniques for achieving high-fidelity wafer images, especially as semiconductor device critical dimensions shrink. However, traditional SRAF generation methods often face challenges in scalability, adaptability, and efficiency. This paper introduces a novel method that uses a conditional generative diffusion model for SRAF generation to improve efficiency and flexibility. We treat the SRAF generation task as an image-to-image translation problem, converting the input layout to include optimal assist features. Experimental results show that our proposed approach achieves a 5.57× speed-up over commercial tools while maintaining comparable accuracy in terms of edge placement error and process variation band.