Single wafer rapid thermal processing (RTP) becomes one of the key technologies in semiconductor manufacturing due to its faster wafer processing with reduced thermal budget and better control of processing conditions. As the standard size of silicon wafer grows and integration of integrated circuits increases, better control to improve the processing time, uniformity and repeatability of processing is still highly required. In RTP, identification and control are complicated because of highly nonlinearity, drift and time varying nature of the wafer dynamics. Physical models for the wafer dynamics have been available, but they are not utilized fully for identification and control. Here, a method to obtain nonlinear models which use a physical model of the single wafer rapid thermal processing and a set of local linear models is proposed. In obtaining local linear models, the closed-loop identification method with loosely tuned PI controllers is used to maintain operating points and suppress drifts. From the set of local linear models, three nonlinear models by imposing different assumptions on the physical model are obtained. Model parameters are well coincide with those in literatures and estimates with physical property data. The second and third models use analytic nonlinear terms whose coefficients are not identified but estimated analytically through physical property data. They can provide a simpler identification method where local linear models at different temperature levels are not needed. The identified models have been tested by simulations using the first nonlinear model for the process. A simple linearizing controller is used for closed loop simulations and excellent control performances have been obtained.

Extended Abstract Status: Not Uploaded