For device optimization purposes, especially for closed-loop optimizations, it is necessary to provide a complete definition of the optimization task. This includes the description of the device with its geometry and doping profile, the definition of the optimization parameters which will be varied during the optimization process, and the simulation procedure with a clear definition of the optimization goal.
The resulting optimization setup can be compared with a black box evaluator which offers a set of optimization parameters at its input and a performance metric, also called ``optimization target'', and constraints at the output (Fig. 3.1). The goal is to find a set of optimization parameters which deliver the best performance metric while keeping the constraints within a specified range. The metric can be, for example, the drive current, the gate delay time, and so on. Constraints can, for instance, be leakage currents. Possible optimization parameters are doping parameters, geometry parameters, or parameters defining the operating condition of the device, for example, the supply voltage.
This setup process has to be performed carefully since a lot of a-priori knowledge about a ``good'' device can, on the one hand, reduce the number of optimization parameters and, therefore, the optimization effort. But, on the other hand, it can restrict the optimization possibilities too much so that no perceptions or ideas about new device design methods can be found.
In this work, a two-stage optimization approach is chosen combining the advantage of a very general approach in the first stage and a reduced parameter set in the second stage. This makes it possible to benefit from the knowledge about the performance improvement of the first stage by analyzing its result and decreasing the complexity of the optimization setup in the second stage. The output of the second stage will also confirm that the assumed reasons for the performance gain are correct.