Control relevant RIE modeling by neural networks from real time production state sensor measurements

In the present paper we address the problem of control relevant process modeling from production data for the n-well reactive ion etching processed by LAM Rainbow Etchers. Due to physical constraints we consider building an empirical neural network model using one lot of data which usually contains 24 wafers. Using the existence result of feedforward networks as universal approximators, we experimentally developed different network structures as models of the etching process under investigation. Our results are built upon extensive simulations on different lots of the process