Many modern thermal processing systems involve temperature control of heated plates. In most Rapid Thermal Processing (RTP) systems, the ‘plate’ is a single wafer that is heated from one or both sides by an array of tungsten halogen lamps. In many other systems such as Metal-Organic Chemical Vapor Deposition (MOCVD) and epitaxial deposition (epi) systems, the plates are often thicker carriers or susceptors on which one or more wafers are placed. Typically these plates are heated by radiation from hot filaments (including tungsten halogen lamps) and the temperature is measured using pyrometers. In many cases the temperature of the system is controlled using a proportional-integral-derivative (PID) controller. However, in cases where the temperature must be changed rapidly while maintaining good temperature uniformity and with tight performance limits, it can be difficult or even impossible to achieve the desired performance using PID controllers. In addition, the dynamic response of these systems typically changes considerably with operating conditions such as temperature, process gas composition, wafer emissivity, or wall emissivity (e.g., in systems where walls coat during process). This can make it difficult to tune a PID controller to give good performance over a broad range of operating conditions.
To overcome some of these limitations, PID controllers are often gain-scheduled by using different PID gains for the different operating conditions, which typically improves performance. However, the limitations of PID control still apply, and gain-scheduling increases the amount of effort needed to tune the gains. Alternatively, even more complex controllers can be designed, such as Linear Quadratic Gaussian (LQG) controllers, but this requires the knowledge and hand of an expert in the field of systems and control. An alternative model-based control approach has been adopted by SC that can achieve good performance for a wide range of operating conditions. In this approach, a mathematical model of the physics of the system is directly incorporated into the feedback controller, thus allowing the controller to have knowledge of how the system dynamics change with the operating temperature. The design of this controller is relatively simple compared to an LQG controller; apart from the plant model – which can be derived from the chamber geometry, material properties, etc. – the design of the controller comes down to a few simple choices, such as the selection of the bandwidth of the desired closed-loop system. In this paper, the robustness and performance of this new model-based controller is compared to that of a gain-scheduled PID controller, as well as an LQG controller for a range of plate properties and operating conditions. All three controller methods are compared in terms of handling system modeling uncertainty, tracking performance (overshoot, settling time, etc.), and noise and disturbance accommodation properties.
Simulation studies show that the new model-based approach as well as the LQG controller yield significantly better performance and exhibit a much smaller sensitivity to system properties variation than a gain-scheduled PID controller at the expense of a slight increase in sensitivity to sensor noise. In addition, the model-based controller has the advantage of a simplified and more intuitive design than the mathematically more complex LQG controller.