Neural network based linearization and control of sputter processes

A model-based design method is developed to control nonlinear multiple-input multiple-output sputter processes. The method consists of three control structures. Artificial neural networks are identified to decouple the subsystems and linearize the plant. Local linear controllers are designed to ensure set-point following of the closed-loop system with the desired performance specifications. A gain-scheduled controller is proposed to compensate the nonlinear influence of an additional actuator system. The robustness of the control loops is discussed. Experiments demonstrate the effectiveness of the controller.