Enhancing Material Removal Rate Prediction on Wafer Surface: A PSO-DBN-OSELM Approach

Aiming at the problem of insufficient pre-precision of wafer surface Material Removal Rate (MRR) in chemical mechanical polishing, this paper proposes a combined model based on Deep Belief Network (DBN), Online Sequential Extreme Learning Machine (OS-ELM) and improved Particle Swarm Optimization (PSO) algorithm. First, the source data is preprocessed by techniques such as feature selection and normalization, and then the abstract features in the data are extracted by DBN and used as the input of OSELM for regression training. Finally, through the improved PSO algorithm, according to the greedy training method of the DBN-OSELM model, a specific method of optimizing the network structure and parameters is proposed to achieve high-precision prediction. The experimental results show that the RMSE of the PSO-DBN-OSELM model is 6.2% lower than that of the existing MRR prediction model, which indicates that it has higher prediction accuracy.