On-chip Optical Attenuators Designed by Artifical Neural Networks

We have proposed an efficient optimization strategy for inverse design of on-chip digital meta devices. A prediction mechanism is applied for estimation of the figure-of-merits during the direct binary search process. As such, the optimization is significantly accelerated. Based on 700 training samples obtained via the finite difference time domain method, we are able to accelerate the optimization procedure of integrated and ultra-compact optical attenuators with arbitrary attenuations.