Modeling Sampling Strategy Optimization by Machine Learning Based Analysis

Optical proximity correction (OPC) model has become a necessity in advanced lithography in order to improve design to wafer fidelity. Typically, a limited set of test patterns are measured for OPC model calibration. More test patterns are used when node goes to smaller. However, more modeling data usually mean heavier metrology workload, longer model optimization time, and more computational resource demands. To balance the resource & time consumption and model accuracy, here we proposed a novel way to optimize modeling sampling strategy using machine learning analysis. The proposed approach uses machine learning platform (MLP) to generate feature vector and does proper hyper-space coverage analysis for sampling reduction. This method can significantly reduce metrology burden and improves model tuning cycle without sacrificing model accuracy and robustness.