Process Control to Improve Yield in the Plasma Etching Process Using an Adaptively Trained Neural Network

Abstract

In this paper, we present a process analysis system that can analyze causes with expert proficiency for a given result after undergoing various processes. Also, the plasma etching process that affects yield is controlled, using an adaptively trained neural network, to predict an output before the real process. In modeling, a method that utilizes the trend history of input data shows considerable advantage in both learning and prediction. The research regards CD (Critical Dimension), which is crucial in high integrated circuits, as the output variable of the model. Based on the model using this method, we propose an algorithm to analyze and control the effect of input variables for predicted defects. Both the weight of input variables and their trend history are considered in for this algorithm.