Machine learning assisted SRAF placement for full chip

Sub-Resolution Assist Features (SRAF) are widely used for Process Window (PW) enhancement in computational lithography. Rule-Based SRAF (RB-SRAF) methods work well with simple designs and regular repeated patterns, but require a long development cycle involving Litho, OPC, and design-technology co-optimization (DTCO) engineers. Furthermore, RB-SRAF is heuristics-based and there is no guarantee that SRAF placement is optimal for complex patterns. In contrast, the Model-Based SRAF (MB-SRAF) technique to construct SRAFs using the guidance map is sufficient to provide the required process window for the 32nm node and below. It provides an improved lithography margin for full chip and removes the challenge of developing manually complex rules to assist 2D structures. The machine learning assisted SRAF placement technique developed on the ASML Brion Tachyon platform allows us to push the limits of MB-SRAF even further. A Deep Convolutional Neural Network (DCNN) is trained using a Continuous Transmission Mask (CTM) that is fully optimized by the Tachyon inverse lithography engine. The neural network generated SRAF guidance map is then used to assist full-chip SRAF placement. This is different from the current full-chip MB-SRAF approach which utilizes a guidance map of mask sensitivity to improve the contrast of optical image at the edge of lithography target patterns. We expect that machine learning assisted SRAF placement can achieve a superior process window compared to the MB-SRAF method, with a full-chip affordable runtime significantly faster than inverse lithography. We will describe the current status of machine learning assisted SRAF technique and demonstrate its application on the full chip mask synthesis and how it can extend the computational lithography roadmap.