Machine-Engineered Active Disorder for Digital Photonics

Abstract

Resolving spatial and temporal complexities in wave-matter interactions is essential for controlling the light behavior inside disordered and nonstationary systems and therefore achieving high capacity devices. Although these complexities have usually been studied separately, a few examples exploiting both degrees of freedom have derived intriguing phenomena such as hyper-transport in evolving disorder and topological phenomena in synthetic dimensions. Here, engineering active disorder-disordered structures with external modulation-is proposed by employing deep neural networks. A functional regressor and a material evaluator are developed to enable inverse design of active disorder with target wave responses and evaluation of disordered structures according to the wave response controllability, respectively. By machine engineering deep-subwavelength disorder including a phase change material, functional disorder for light is revealed, which leads to angle-selective or broadband digital switching. A generative configuration of the neural network utilizing a single wave metric is also developed to realize a family of disordered structures with independent engineering of multiple wave properties, in contrast to the traditional engineering of disorder with a specific order metric. This approach establishes realization of reconfigurable devices by exploiting the spatiotemporal complexity in wave mechanics.