Application of artificial neural networks to real-time control of plasma processes

摘要

Abstract: Plasma processes typically involve a number of coupledcontrol parameters which exhibit complex interactions.These parameters jointly effect the plasma processparameters which may or may not be measurable.Ultimately, the process parameters affect the waferparameters which we would ultimately like to control.The inherent multivariate nature of the problem makesconventional control methodologies difficult to apply.Artificial neural networks (ANNs) offer a promisingalternative because of their ability to learn thedesired control behavior by direct observation of theprocess. Once enabled as the controller, the ANNcontinues to improve its model of the process behaviorand thus compensates for slow drifts in the process.!9