A study in dynamic neural control of semiconductor fabrication processes

This paper describes a generic dynamic control system designed for use in s emiconductor fabrication process control. The controller is designed for an y batch silicon wafer process that is run on equipment having a high number of variables that are under operator control. These controlled variables i nclude both equipment state variables such as power, temperature, etc. and the repair, replacement, or maintenance of equipment parts, which cause par ameter drift of the machine over time. The controller consists of three pri ncipal components: 1) an automatically updating database, 2) a neural-netwo rk prediction model for the prediction of process quality based on both equ ipment state variables and parts usage, and 3) an optimization algorithm de signed to determine the optimal change of controllable inputs that yield a reduced operation cost, in-control solution. The optimizer suggests a set o f least cost and least effort alternatives for the equipment engineer or op erator. The controller is a PC-driven software solution that resides outsid e the equipment and does not mandate implementation of recommendations in o rder to function correctly. The neural model base continues to learn and im prove over time. An example of the dynamic: process control tool performanc e is presented retrospectively for a plasma etch system. In this study, the neural networks exhibited overall accuracy to within 20% of the observed v alues of .986, .938, and .87 for the output quality variables of etch rate, standard deviation, and selectivity, respectively, based on a total sample size of 148 records, The control unit was able to accurately detect the ne ed for parts replacements and wet clean operations in 34 of 40 operations. The controller suggested chamber state variable changes which either improv ed performance of the output quality variables or adjusted the input variab le to a lower cost level without impairment of output quality.