Modeling and optimization of Chemical Mechanical Planarization (CMP) using neural networks, ANFIS and evolutionary algorithms

Abstract

Scope and Method of Study: The purpose of this study is to provide new approaches to improve the current modeling and optimization of the chemical mechanical planarization or polishing (CMP) process. Neural Networks (NN), ANFIS (Adaptive-based-Network Fuzzy Inference System), and Evolutionary Algorithms (EA) were applied to construct process models of the material removal rate (MRR) and the within wafer non-uniformity (WIWNU). Furthermore, MultiObjective Evolutionary Algorithms (MOEA) are firstly applied to search the optimal input settings for CMP process to trade-off the higher MRR and lower Non-Uniformity by using the previously constructed models.

Findings and Conclusions: The modeling approaches using NN, ANFIS, and EA can fully capture the highly non-linear dynamics of complex CMP process, and successfully provide accurate models of MRR and WIWNU under sufficient training data. The uniform distribution of training data from the factorial designing experiments, providing entire information of input and output spaces, can effectively reduce the training errors for constructing more accurate process models. In addition, the fine-tuning techniques for re-modifying ANFIS models constructed by sparse data can effectively improve modeling accuracy. The size of training data can only slightly influence the generalization capabilities of EA models, if the pre-determined formulae chosen are correct enough. The results also show the simulation of MOEA optimization can certainly provide the accurate guidance to search the optimal input settings for CMP process to produce lower non-uniform water surfaces under higher MRR.