Abstract

Instead of using FDTD simulations for all the inverse design steps, we proposed to use neural network-based fitting to estimate the output of the FDTD simulations, and improve the design. We observed clear acceleration in the improvement of metric.

© 2017 OSA

PDF Article