Neural network control of a plasma gate etch: Early steps in wafer-to-wafer process control

Abstract

A gate oxide thickness controller for a plasma etch reactor has been developed. This controller is for 0.9-/spl mu/m technology. By monitoring certain processes, signatures are fed forward into a neural network trained by the backpropagation method. It is possible to predict in real time the correct over-etch time on a wafer-by-wafer basis. Computer simulations indicate that the neural network is equivalent to humans for this task. The uniqueness of this controller is compared with a previous controller for a 1.25-/spl mu/m technology gate etch process.<>