Dataless training of generative models for the inverse design of metasurfaces.
Year: 2019
Type: preprint
Abstract: Metasurfaces are subwavelength-structured artificial media that can shape andlocalize electromagnetic waves in unique ways. The inverse design of
metasurfaces is a non-convex optimization problem in a high dimensional space,
making global optimization a huge challenge. We present a new type of global
optimization algorithm, based on the training of a generative neural network
without a training set, which can produce high-performance metasurfaces.
Instead of directly optimizing devices one at a time, we reframe the
optimization as the training of a generator that iteratively enhances the
probability of generating high-performance devices. The loss function used for
backpropagation is defined as a function of generated patterns and their
efficiency gradients, which are calculated by the adjoint variable method using
the forward and adjoint electromagnetic simulations. We observe that
distributions of devices generated by the network continuously shift towards
high-performance design space regions over the course of optimization. Upon
training completion, the best-generated devices have efficiencies comparable to
or exceeding the best devices designed using standard topology optimization. We
envision that our proposed global optimization algorithm generally applies to
other gradient-based optimization problems in optics, mechanics and
electronics.
Source: arXiv (Cornell University)
Authors: Jiaqi Jiang, Jonathan A. Fan
Language: English
Field: Materials Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W2951650104