Abstract
The analysis and optimization of design parameters for fabricated and simulated electromagnetic (EM) structures may be expensive in terms of costs and computations. In response, the analysis of an EM structure can be conducted by dividing the analysis area into two parts: (1) an internal zone with a detailed model of the device, and (2) an external zone representing the input-output relationship of device terminals without considering interior components or structural parameters. The external zone can be represented as a frequency-dependent network equivalent (FDNE). The macromodeling of intricate EM structures as FDNEs, such as equivalent circuits and systems that approximate same input-output relationships at the terminals, offers an alternative analysis method with a low computational cost with acceptable accuracy. Additionally, acquiring the required materials or crafting the geometry for the desired EM structure may pose challenges or incur significant expenses, while macromodeling techniques construct an equivalent model using more affordable and readily available components such as R/L/C elements. Therefore, macromodeling of EM structures proves beneficial in analyzing and modeling EM devices, offering cheaper and computationally efficient models with sufficient accuracy. There exist numerous algorithms through out literature, each designed for a specific task in macromodeling process, however, only a handful of automated tools are available to perform macromodeling of EM structures, beginning from network parameters to generate a reduced-order passivity-enforced equivalent circuit. On the other hand, as internal components/parameters remain inaccessible through macromodeling, optimizing the internal parameters becomes unfeasible. To address this issue, the internal zone of EM structures can be analyzed using advanced numerical methods alongside with Artificial Intelligence (AI) methodologies such as machine learning (ML) algorithms and neural network modeling techniques, which are able to offer fast, reliable, and low-cost solutions.
This work introduces modeling methodologies for both the internal and external zones. Initially, it focuses on external zone modeling by proposing canonical equivalent circuits for improper rational transfer functions derived from fitting network parameters of EM structures. It proposes equivalent network representation techniques and presents methods for reducing the order of equivalent circuits. Subsequently, it introduces SROPEE, a novel algorithm and automated Python-based open-source software (OSS) designed to synthesize the reduced-order passivity-enforced equivalent circuits of networks represented by scattering parameters (S-parameters). The developed Python-based OSS may be deployed as a stand-alone tool, or integrated into existing process technology flows and CAD tools to expedite design automation. Finally, it delves into internal zone modeling by exploring numerical techniques for EM analysis, and introduces AI-based modeling methodologies derived from integrating numerical techniques with machine learning models and neural networks, aimed at analyzing various EM problems.