Assessment of Physics-Based and Data-Driven Models for Material Removal Rate Prediction in Chemical Mechanical Polishing

Abstract

Material removal rate (MRR) during the chemical mechanical polishing (CMP) process affects the control of product quality. Complexity of various parameters makes it challenging to predict MRR accurately. We addressed this challenge by integrating physics-based modeling with data-driven statistics. First, we analyzed the raw data using data profiling techniques. Then, we extracted features from a physical point of view. Finally, we constructed two Random Forest models respectively based on the feature selection results via the Generic Algorithm. Experiments show that the features we extracted embody key information of each process. The final score predicted by this approach ranked in the second place in a Data Challenge Competition.