Neural network based adaptive tracking controller for a reactive ion etching system

The theoretical development of a nonlinear adaptive tracking control architecture for reactive ion etching (RIE) systems is presented. This design is based on a dual time scale assumption of RIE dynamics. Two multi-layer feedforward neural networks are employed in this architecture. One is trained off-line to approximately invert the nonlinear dynamics of the etching plant, and the other, which is capable of online learning, is used to dynamically cancel the inversion error. Stability analysis is provided using the Lyapunov method, and a weight adjustment rule for the online learning neural network is derived that, can guarantee closed-loop stability