Recent advances of neural network-based EM-CAD

Abstract

In this article, we provide an overview of recent advances in computer-aided design techniques using neural networks for electromagnetic (EM) modeling and design applications. Summary of various recent neural network modeling techniques including passive component modeling, design and optimization using the models are discussed. Training data for the models are generated from EM simulations. The trained neural networks become fast and accurate models of EM structures. The models are then incorporated into various optimization methods and commercially available circuit simulators for fast design and optimization. We also provide an overview of recently developed neural network inverse modeling technique. Training a neural network inverse model directly may become difficult due to the nonuniqueness of the input-output relationship in the inverse model. Training data containing multivalued solutions are divided into groups according to derivative information. Multiple inverse submodels are built based on divided data groups and are then combined to form a complete model. Comparison between the conventional EMbased design approach and the inverse design approach has also been discussed. These computer-aided design techniques using neural models provide circuit level simulation speed with EM level accuracy avoiding the high computational cost of EM simulation.

Obstructive uropathy caused by an inguinal hernia in a kidney transplant recipient: Report of hernia cure by the shouldice technique

“…It is important that the surrogate is physically based [10], so that it can give a reliable prediction of the original structure's behavior under the modification of its designable parameters. One of the most successful techniques in microwave engineering exploiting physically based surrogate models is space mapping (SM) [11–20]. SM replaces direct optimization of a CPU‐intensive structure (“fine” model) by iterative optimization and updating of so‐called coarse models that are less accurate but cheaper to evaluate.…”

Rapid design optimization of antennas using space mapping and response surface approximation models

“…A variety of function approximation methods are available including polynomial approximation [16], neural networks [36][37][38][39][40], kriging [16,41,42], multidimensional Cauchy approximation [43], or support vector regression [44]. Here, the coarse model is constructed using kriging interpolation.…”

Response correction techniques for surrogate-based design optimization of microwave structures

“…Several methods exploiting this approach have been proposed recently including manifold mapping (MM) [30], multipoint response correction [31], adaptive response correction (ARC) [32], and shape-preserving response prediction (SPRP) [33]. Formally, output SM [16] and its variations [34] also belong to this group, as well as the neuro-SM techniques [35][36][37].…”