Virtual Metrology Equipped with a Variability Analyzer in Chemical Mechanical Polishing

Abstract

Virtual metrology equipped with a variability analyzer is reported. Twenty-six process variables are modeled to predict run-to-run material removal rates in chemical mechanical polishing. The mean absolute percentage error of about 0.5 % is accomplished in prediction. Factors leading to the run-to-run variations are analyzed using exploratory factor analysis. Three major factors are found and analyzed using two different approaches. The relevant process features corresponding to the factors are extracted and possible physical explanations provided. The results suggest the proposed virtual metrology is potential to analyze variability in chemical mechanical polishing and extendable to the other processes.