Event Details

The majority of today’s technology depends on the design and fabrication of high-performance electromagnetic components, whether in the radio-frequency (RF) or photonics domain. In the traditional workflow of electromagnetic design, human designers iteratively refine devices to yield desired performance specifications, a process that is both labor-intensive and heavily reliant on intuition and experience. As such, inverse design techniques have gained ever-increasing prominence, owing to their capability to automate the design of novel, nonintuitive devices achieving performance superior to that of conventional designs and exceeding the bounds of human creativity. However, recent inverse design approaches either require resource-intensive full-wave simulation at each optimization iteration or involve using approximations, such as replacing the EM solver with deep learning-based surrogate models. Unfortunately, although surrogate models can significantly accelerate the time it takes to predict the objective function for a candidate design, they require extensive time for training data generation and do not guarantee correct results over the full design space. In this talk, I will introduce the Precomputed Numerical Green Function (PNGF) method, which addresses these issues by matching the accuracy of full-wave 3D electromagnetics simulation while achieving the speed of deep-learning models. The PNGF method reduces the total runtime of inverse design from multiple days or weeks when using commercial EM solvers down to several minutes without requiring any training or preexisting datasets. Utilizing the current equivalence principle, an efficient single-step fully parallelized precomputation process is used to solve for the Green’s function of the simulation domain numerically, encapsulating the interactions between the optimization region and the remaining static portions of the device and design environment that remain unchanged throughout the optimization. The subsequent optimization requires no full-wave simulations and has linear time complexity with respect to the size of the optimization region. I present several example design studies, verified with simulation and experimental results, which demonstrate speedups beyond 16,000x for the full design process compared to commercial solvers.

October 3, 2025

Join Zoom Webinar

Host: Steve Crago
POC: Amy Kasmir

Speaker Bio

Constantine Sideris is an Associate Professor of Electrical and Computer Engineering and the Andrew and Erna Viterbi Early Career Chair at the University of Southern California. He received the B.S., M.S., and Ph.D. degrees with honors from the California Institute of Technology in 2010, 2011, and 2017 respectively. He was a visiting scholar at UC Berkeley’s Wireless Research Center in 2013. He was a postdoctoral scholar in the Department of Computing and Mathematical Sciences at Caltech in 2017-2018 working on integral equation methods for electromagnetics. He was the recipient of an ONR YIP award in 2023, an NSF CAREER award in 2021, an AFOSR YIP award in 2020, and an NSF graduate research fellowship in 2010. Constantine's research lies at the intersection of analog/RF and photonic integrated circuits and computational electromagnetics, with applications in biomedical technology and wireless communications. His interdisciplinary approach bridges applied mathematics, computation, physics, electrical engineering, and medicine to develop next-generation technologies. His current projects explore portable magnetic resonance spectrometers, wearable devices for real-time monitoring and analysis of biological signals, ingestible "smart" pills, and neural interfaces. On the computational side, he is pioneering highly efficient algorithms for solving Maxwell’s equations and integrating them with optimization techniques to develop design automation platforms for complex radiofrequency and nanophotonic systems.