Artificial neural network model-based run-to-run process controller

In this paper, we present an artificial neural network (ANN) model-based controller for a batch semiconductor manufacturing process. The proposed controller is an integration of ANN, statistical process control (SPC), and automatic process control (APC) techniques. An ANN model trained with design of experiments (DOE) data is used to map the input-output relation of the process. The controller model is then extracted from the ANN process model by Taylor expansion and inversion. For application to a noisy process, the exponential weighted moving average (EWMA) technique is first used to filter out the output noise and detect the process shift/drift. Based on feedback, the controller tunes the settings to compensate for the process shift/drift. Experimental data on a laboratory chemical vapor deposition (CVD) reactor is used to demonstrate the effectiveness of the proposed run-to-run controller. A comparison shows that the proposed controller performs better than other similar controllers. Finally, a total cost criterion is proposed to provide optimum parameters for a run-to-run controller