Lithography hotspot detection with ResNet network

The ever-shrinking feature size and complicated mask design is posing great challenges to lithography wafer image printability, where certain patterns are prone to circuit failures with process variations, thus hotspot detection to identify problematic layout regions for further refinement is vital in the physical verification flow. In this paper, we propose a lithography hotspot detection method based on ResNet neural network with enhanced data augmentation. Experimental results show the improved recall rate of over 98% for 28nm technology node outperforming the benchmark of ICCAD2012 contest.