The optimization field for TCAD applications [79] can be categorized
into three major tasks. These are *inverse modeling* [80], calibration of simulator
models [81,82] and the tuning of certain process parameters
[83].

A TCAD optimization task is an iterative minimization or maximization process, where the optimizer controls a set of free parameters of a model within certain upper and lower bounds to minimize or maximize a given target. A simulator that takes the model as input is used to compute an error vector. The target is then computed as the quadratic mean of the dimensional error vector

(5.1) |

The error vector is a modified relative error that is scaled to values between to avoid numerical overflows. Depending on the optimization algorithm either the above given target or the error vector is passed to the optimizer which changes the free parameters according to its optimization algorithm in order to take a step towards the desired minimum or maximum.

Several iterations are performed until a truncation criterion is
reached. For the case of a local optimizer the criterion is usually a minimum
change of the target value. For a global optimizer additional criteria like,
e.g. a maximum iteration number might be defined. An optimization
*framework* integrates optimizer and simulator and spreads the simulation jobs
over a cluster of workstations. Fig. 5.1 depicts this
scenario. Several concrete sub-problems of this abstract optimization task
exists which are briefly sketched in the following sections.

2003-03-27