ETCH Model Based on Machine Learning

This paper introduces a machine learning based etch bias prediction model and demonstrates its capability of extracting the features of the one-dimension and two-dimension layout post etch and predicting of etch bias. For 1D pattern, this model achieves absolute error less than ±2nm and average relative error below 10%. For 2D layout we achieved average absolute error less than 4nm with a standard deviation of 3nm, and average relative error of 15%. These results are promising for the developing the model-based etch proximity correction model towards the high volume manufacturing.