Predictive modeling of material removal rate in chemical mechanical planarization with physics-informed machine learning

Abstract

Chemical mechanical planarization (CMP) is a high-precision and complex manufacturing process that removes material with chemical and mechanical forces in order to achieve highly planar surfaces. Significant research efforts have been devoted to developing physics-based and data-driven approaches to predicting the material removal rate (MRR) in the CMP process. Both physics-based and data-driven methods have advantages and disadvantages. A novel physics-informed machine learning approach is introduced to combine a physics-based model of MRR in CMP with a data-driven model of MRR in CMP. The physics-based model takes into account the contact between a polishing pad and abrasives and the contact between abrasives and a wafer. The data-driven model trained by a machine learning algorithm predicts the asperity radius and asperity density of the polishing pad using the polishing pad wear and conditioner wear estimated by the physics-based model. The predicted asperity radius and asperity density of the polishing pad are then used to estimate MRR in CMP. Experimental data collected from a CMP tool under varying operating conditions are used to train and validate the predictive model. Experimental results have shown that the physics-informed machine learning method is capable of predicting MRR in CMP with high accuracy.