An efficient inverse lithography technology method based on detail-enhanced Pix2Pix network

Mask optimization has arisen as a vital challenge in the flow of VLSI manufacturing, primarily because the critical dimension of integrated circuits stands much smaller than the light source's wavelength. Inverse lithography technology (ILT), a notable resolution enhancement technology, is known for its efficacy in enhancing mask printability. However, its extensive computational complexity has obstructed its broad-spectrum adoption. In our paper, we introduce a detail-enhanced Pix2Pix network, founded on GAN principles, to speed up the ILT process. This network has the capacity to generate quasi-optimal masks from given target layouts, thereby reducing the amount of traditional ILT steps required to produce high-quality masks. Our experiments on the ICCAD 2013 benchmarks demonstrate that, in comparison to the latest cGAN-based method, the L2 error, PVB, and runtime in our work have seen reductions of 7.2%, 5.9%, and 18.4% respectively. Thus, our approach not only expedites the ILT process but also guarantees enhanced printability.