Inverse Design of Reconfigurable Plasma Metamaterials for Optical Computing

Inverse design (or equivalently, machine learning) methods are commonly used to create high-efficiency optical devices that perform exotic functions that otherwise would not be possible to create using conventional ‘human design’ methods. In this study, we apply inverse design methods to produce fully reconfigurable, multi-function, two-dimensional plasma metamaterial (PMM) devices composed of low-temperature plasma discharge tubes. Autograd-compliant finite difference frequency domain simulations are used to design waveguides, demultiplexers, and all-optical logic gates for use in optical computing. Demultiplexing and waveguiding are demonstrated for PMM devices composed of realistic plasma elements with non-uniform plasma density profiles, collisional damping, and resistance to experimental error, allowing for future in-situtraining and experimental realization of these designs.