Descripción
Efficient neural space mapping methods for highly accurate electromagnetics-based design optimization, statistical analysis and yield estimation are described and contrasted in this chapter. Statistical analysis of RF and microwave circuits is crucial for manufacturability, typically requiring a massive amount of high-fidelity simulations. This chapter describes several methodologies for accurate and inexpensive yield estimations after an efficient design optimization process based on space-mapping. Comparisons are presented between different strategies for developing neural models that implement suitable mappings. It is first described how to develop an input space mapping approach to enhance available coarse models for efficient statistical analysis. This is realized in a linear fashion, using a Broyden-based approach, as well as in a nonlinear manner using neural networks. The corresponding neural space mapping is trained using reduced sets of full-wave EM data. It is illustrated how these input space mapping techniques can be employed to perform efficient yield estimations. Finally, the combination of linear input mappings with neural output mappings to develop more accurate and equally efficient surrogate models is described. This last strategy allows the transformation of conventional equivalent circuit models into accurate vehicles for EM-based statistical analysis and design. The design optimization and yield prediction of synthetic problems as well as practical microstrip circuits, using commercially available CAD tools, illustrate the methodologies.
