Application of neural fuzzy network to pyrometer correction and temperature control in rapid thermal processing

Temperature measurement and control are two difficult problems in the rapid thermal processing (RTP) system. For many applications such as rapid thermal processing chemical vapor deposition (RTCVD) [1] and rapid thermal oxidation (RTO) [2], large changes in wafer emissivity can occur during film growing, leading to erroneous temperature measurements with a single wavelength pyrometer. The error in the inferred temperature will affect the temperature control of the RTP system. In order to correct the temperature reading of the pyrometer, a neural fuzzy network is used to predict the emissivity changes for the compensation of measured temperature. As for the temperature control, to overcome ill performance of the temperature tracking system due to the inaccuracy of the identified model, another neural fuzzy network is used in the RTP system for learning inverse control simultaneously. The key advantage of neural fuzzy approach over traditional ones lies on that the approach does not require a mathematical description of the system while performing pyrometer correction and temperature control. Simulation results show that the adopted neural fuzzy networks can not only correct the pyrometer reading accurately, but also be able to track a temperature trajectory very well.