Machine learning-assisted pattern optimizations for fixing design for manufacturability (DFM) rule check violations

An in-design, Machine Learning (ML)-assisted Pattern OPtimizer (POP) is developed for improving Manufacturability Analysis and Scoring (MAS), i.e., ML MAS-POP, fixing DFM rule check violations automatically by leveraging the Placeand-Route (PnR) hotspot repair tool. First, DFM rules are converted to patterns using pattern classification automation. Corresponding fixing hints are prescribed for each pattern. In addition, a topology-aware ML model is developed to dynamically predict a sequence of fixing hints given the input pattern context. Once the patterns, fixing hints, and the ML model have been developed, they are inserted in the PnR hotspot repair tool, enabling the automatic detection and fixing of the DFM violations by applying a dynamic sequence of predicted fixing hints. The ML MAS-POP methodology was demonstrated on block-level designs. Results show that 10k+ DFM via-metal line-end enclosure violations were fixed automatically for a 200 x 200 um² design, an 81% fixing rate.