Predicting Material Removal Rate in Chemical Mechanical Polishing (CMP) Using Explainable Machine Learning Methods

Chemical Mechanical Polishing (CMP) is a crucial process in integrated circuit manufacturing. To determine the material removal rate in this process, three different models were employed for learning and prediction. Among them, the XGBoost (eXtreme Gradient Boosting) algorithm yielded the best performance, achieving an R-squared value of 0.87. Using SHAP (SHapley Additive exPlanations) in conjunction with the XGBoost model, an assessment of feature importance was conducted. Utilizing these identified features as inputs for training, the R-squared value improved to 0.89 under the XGBoost algorithm. These findings were summarized, with a detail discussion feature importance based on the physical and chemical processes associated with CMP.