Application of an adaptive neuro-fuzzy inference system for the optimal analysis of chemical-mechanical polishing process parameters

This paper examines the machining parameters during the wafer flattening process by, chemical-mechanical polishing (CMP). There are very few data available from CMP experiments for wafer flattening. This study adopted an adaptive neuro-fuzzy inference system (ANFIS) to predict the surface roughness in the absence of CMP experiments. An integrated concept like ANFIS combines the advantages of the two systems of fuzzy control and neuro networks. Next, the feasible directions algorithm and sequential approximation algorithm from the local search method are combined with ANFIS. During the process of combination, the value from the optimisation theory is replaced by that from the ANFIS, so that, the roughness value of the wafer surface can be predicted. Alternatively, the optimal values of various process parameters can also be predicted. To sum up, verification through experiments indicates that the optimal experimental values of process parameters are identical with those predicted by the optimisation theory, and ANFIS. Thus, the optimal precise value can be simulated and predicted within the parameters of the experimental design. The predicted optimal result is compared with the optimal experimental result of Kung and Dai to show that the predicted optimal result is acceptable. As a result, the CMP process parameters investigated in this study. can be controlled.