15.3 Summary

Two optimizers of the SIESTA framework were used for the inverse modeling tasks. First the global optimizer EGO (cf. Section 9.3.1) was used for finding interesting starting points and identifying global minima. In the second step the local optimizer DONOPT (cf. Section 9.3.5) was used for refining these points. The combination of both approaches allows to take advantage of the global screening of genetic algorithms and the fast progress of good local optimizers.

The calibrated model was later used by process engineers to simulate arsenic pre-deposition processes. The large number of measurements, the distinctness of different process conditions in Table 15.1, and the good final agreement provide the confidence that the simulations can be trusted for a wide range of process conditions. The robustness of SIESTA, its ability to deal with complicated setups, and the choice of optimizers were crucial for arriving at this result.