A Neural Network-Based Approach to Material Removal Rate Prediction for Copper Chemical Mechanical Planarization

Abstract

In modern integrated circuits, chemical mechanical planarization (CMP) has emerged as one of the most important solutions for surface global planarization. The surface uniformity and quality control are greatly dependent on the material removal rate (MRR) of the polished wafers. The construction of accurate physical CMP models to predict MRR is a great challenge due to the complexity of the coupling interplay of mechanical, chemical and design pattern effects in the CMP process. In this work, CMP experiments are designed and performed under different process conditions to obtain the removal rates and a data-driven neural network-based approach is developed to predict the MRR and reveal the relationship between the removal rate and polishing parameters for copper CMP. It is shown that the predicted results of the removal rates are consistent with the experimental data with an optimized network structure. The investigation of the neural networks (NNs) to model the MRR indicates that the NN-based method can provide a general way of capturing the removal rate profiles regardless of the complexity of the polishing mechanism. Therefore, the present CMP model has good potentials of assisting in achieving the surface uniformity control of the copper wafer in semiconductor manufacturing.