Deep learning-driven digital inverse lithography technology for DMD-based maskless projection lithography

Digital micromirror device (DMD)-based maskless projection lithography has gained significant attention for its maskless, flexible, and cost-effective characteristics. However, when dealing with target layouts smaller than the wavelength scale, optical proximity effects (OPE) pose a significant challenge in achieving precise resist patterns that closely match the intended layout. To address this issue, we introduce an innovative technique called deep learning-driven digital inverse lithography technology (DDILT). This method optimizes the DMD modulation coefficient to implement optical proximity correction (OPC), significantly improving the printability of target layouts. DDILT allows for reverse engineering target layouts that do not align with the DMD pixel grid, effectively overcoming the limitations imposed by DMD pixel size. Notably, DDILT operates at a speed advantage of at least three orders of magnitude compared to alternative methods. In the case of binary amplitude modulation, it reduces edge distance error (EDE) to just 35% of the original value, providing a rapid and efficient solution for addressing large-scale layout challenges.