9.1 A General Purpose Inverse Modeling Setup

As the rigorous calibration of simulation models can be a cumbersome procedure, models are often poorly calibrated based on a very limited number of measurements. Thus, the models deliver satisfying results only as long as they are used in the vicinity of the conditions of the calibration. Due to such a restricted calibration within small subspaces of the model's parameter ranges, it can occur that models are miscalibrated which means that differences between measurements and simulation are fitted by changing a parameter which had better been kept constant. Although the model accuracy has slightly been improved for the measurements taken into account for that kind of calibration, the accuracy might even have worsened for operating regions which remained unconsidered by the calibration procedure.

To minimize the risk of miscalibration, it is of utmost importance to include as many measurements in the calibration procedure as available. This increases the probability that the overall accuracy really improves due to calibration and a physically sound set of parameters can be found. However, time restrictions of TCAD engineers will inhibit rigorous calibrations unless the calibration procedure is automated as far as possible. Moreover, these calibrations can require significant amounts of simulation time which raises demand for an efficient processing of the evaluations of the simulation model.

SIESTA offers remedy for both of these obstacles. A Levenberg-Marquardt optimization experiment (see Section 4.2.2 on page ) can be utilized to perform an automatic calibration, and the models presented in offer an efficient way to include multiple sets of measurements with a minimal maintenance effort. Based on these features a ground-breaking inverse modeling system can be built up which leads to better TCAD models and, therefore, adds value to any TCAD system.