-

It is of prime importance to obtain information about irregular patterns on the output control chart of a semiconductor manufacturing process as well as to be able
to predict the occurrance of such patterns. Here, processing of the gate in a 27MHz submicron GaAs MESFET of a monoJithic microwave integrated circuit is chosen as an example, the method 1for extracting control charts, use of process data is described and the time series and neural network methods are used to predict the patterns. Unlike equipmental based modeling this method can predict slow changes in the output. Time series modeling of the output: is done by classical Box-Jenkins. The ability to predict the irregular patterns as well as the effect of the forcasted length is investigated, using an optimum neural network model and a series of simulation.