Adaptive fuzzy control of temperatures in a semiconductor processing furnace

The ability to control temperatures in semiconductor processing furnaces is critical to the manufacturing of semiconductor devices. Temperature control is made difficult by certain characteristics of furnaces: nonlinearity, long time constants, actuator saturation and asymmetry, and a large operating temperature range. In addition, the controller must perform well in response to large disturbances that occur when a cold wafer load is first inserted into the furnace as well as handling small disturbances that may occur in steady state. In this work, a general approach to fuzzy control is developed that takes these characteristics into account and provides significant performance advantages when compared to a linear controller.; Because of the long time constants and high operating costs of semiconductor processing furnaces, simulation is essential. A computationally efficient lumped heat capacitance model was developed to simulate the dominant center zone of a vertical furnace. Although it is not exact, the model is shown to be accurate enough to use for controller development and was used throughout this work to design and evaluate prototype control strategies.; A commonly used approach to fuzzy control was found to be unsuitable for use with the furnace. Large steady-state errors were produced by the combination of nonlinear gains with the proportional control action internal to the fuzzy controller, and were eliminated by using a modified fuzzy PI controller. Integrator windup was eliminated by incorporating its prevention into the membership functions and rule base. When used with nonlinear or asymmetric response characteristics, the resulting fuzzy PI controller is shown to have a unique capability to obtain best-case performance in response to both large and small disturbances using a single set of controller parameters.; Fuzzy gain scheduling was applied to both a linear PI controller and an asymmetric fuzzy PI controller. A single gain scheduler improved the performance of the linear controller at high temperatures and stabilized the system at low temperatures. Gain-scheduled fuzzy controllers were able to produce optimum performance at a desired stabilization temperature plus substantially improved performance for both large and small steps when operated hands-off over a 400-1000{dollar}spcirc{dollar}C temperature range.