Derivative-Free Design Optimization of Sonnet-Simulated Structures Using Shape-Preserving Response Prediction and Space Mapping

A two-level derivative-free algorithm for design optimization of structures simulated using Sonnet em is introduced. The presented technique exploits a coarse-discretization model of the structure of interest that is optimized on a coarse grid using pattern search. Space mapping optimization is subsequently performed with the underlying surrogate model created using limited amount of coarse-discretization Sonnet model data and a recently introduced shape-preserving response prediction methodology. Our technique is demonstrated through the design of two microstrip filters.