Predicting the Material Removal Rate in Chemical Mechanical Planarization Based on Improved Neural Network

Chemical Mechanical Planarization (CMP) technology in integrated circuit manufacturing plays a critical role in realizing local and global flatness of silicon wafer surface. There are still some problems to be solved in industrial production, such as long period of CMP process debugging and various polishing materials, which lead to a more complicated Material Removal Rate (MRR) calculation during the CMP process. Based on the 2016 PHM Challenge dataset, this paper proposes a prediction method based on improved genetic algorithm to optimize BP neural network, which can be used to predict MRR more accurately. It is more difficult to predict MRR because of the many evaluation metrics involved in the CMP process. In this paper, grey relation analysis and principal component analysis methods are introduced to extract and reduce the dimension of 19 evaluation indicators. After reconstruction, six principal components are obtained for input to the BP neural network. The improved adaptive genetic algorithm is used to optimize the weight threshold of BP neural network, and the prediction results are compared with the same type of algorithms. The experimental results show that the BP neural network improved by feature selection and adaptive genetic algorithm has the smallest error with the real results and the highest prediction accuracy for MRR.