Abstract

Reactive Ion Etching (RIE) is a critical technology for modern VLSI circuit fabrication and is used at many stages of the manufacturing process. Several real-time control strategies such as Proportional-Integral ( PI ) self-tuning, Linear Quadratic Gaussian ( LQG ), stochastic adaptive control, Deurocontrol, robust and hierarchical control based on both linear and nonlinear models of the Plasma Generation Subsystem (PGS) are developed to improve plasma characteristics in the Reactive Ion Etching process. The proposed approaches result in superior accuracy and performance when compared to results that are available in the literature. The identification process (prediction error approach) to determine linear Auto Regressive Moving Average ( ARMA ) models of the PGS is based on the computationally efficient recursive least squared ( RLS ) procedure. This is an alternative to the use of Kalman filter that is based on state estimation. The massively parallel processing, nonlinear mapping, and self-learning abilities of neural networks are exploited in the development of intelligent control systems. Neurocontrollers enhance RIE manufacturability and may be used for process optimization, control, and diagnosis. A hierarchical real-time control strategy is developed that automatically selects during each specific operating interval the best real-time control strategy for tracking the dc self bias voltage and fluorine concentration set points. It is shown that the proposed methodology results in higher performance and is computationally more efficient than that using a single control strategy that is dependent on a range of operating conditions.