Fuzzy predictive R2R control to CMP process

For chemical mechanical polishing (CMP) process characteristics of nonlinear, time-varying and not easily being in-situ measured, this paper proposes a CMP process fuzzy predictive run-to-run (R2R) controller named FPR2R. CMP T-S fuzzy predictive model is off-line and on-line identified by algorithms of fuzzy clustering and recursive least squares with forgetting factor, thus problem of constructing accurate mathematical model of complicated CMP is solved and error of modeling is reduced. Recipe is calculated by multivariable generalized predictive control (GPC) method, therefore it improves control precision. Simulation results illustrate that proposed CMP FPR2R controller is better than EWMA control scheme about performance, variation in various runs of products is reduced substantially, process drifts and shifts is suppressed significantly. Compared to EWMA, root mean squared error for material removal rate (MRR) is decreased.