Inverse electromagnetics design with physics-driven neural networks

Jonathan Fan
Assistant Professor of Electrical Engineering
Stanford University

In this talk, Fan will present new algorithmic approaches to the inverse design of freeform electromagnetic devices. His focus will be on an optimization strategy based on physics-driven neural networks, termed GLOnets, in which the global optimization process is reframed as the training of a generative neural network. He will discuss how this method incorporates physics and physical constraints through the interfacing of Maxwell’s equations with machine learning, and he will frame the discussion around examples of metasurfaces and thin film stacks operating near physical design limits. These ideas will help set the stage for hybrid physics- and data-driven approaches to be used in defining the next frontier of electromagnetics engineering.

Read Jonathan Fan's bio.

The MIT.nano Seminar Series offers monthly talks from researchers across the spectrum of nanoscience and nanoengineering. This series is free and open to the public.