Toward Programmable Moiré Computation

Abstract

Recent advances in optical quantum computation set up a broad discussion on quantum supremacy and its practicability. Lack of programmability and extreme working conditions remain the challenges, calling for a programmable computation scheme. The quasi-2D layered materials introduce new architectures for the optical neural networks (ONNs), which support various programmable computations following the on-demand layer design. Compared with the traditional ONNs, Moiré ONNs architectures are more flexible to manufacture via layer number or twist angle control. A general Penn's model to demonstrate the mechanism inside is developed: the dielectric constant control through the layer and twisted bilayer angle dependence, respectively. Theoretically, this device can conduct demo computations ranging from boson sampling to image classification, where quantum computing shows its significant advantages. Instead of redundant 3D-printing and lithography in traditional ONNs, the Moiré computation framework can train different tasks through programmable twists on single layers without replacing materials.