Multichannel Optical Computing With Multi-plane Light Converters

Abstract

Optical computing revolutionizes information processing by harnessing light's unique advantages, such as high bandwidth, parallelism, and energy efficiency. Unlike electronic systems, optical systems process data at light speed with minimal energy loss, making them ideal for computationally intensive tasks like AI, neural networks, and image processing [1]. Recently, linear and nonlinear optical relations have been utilized to realize photonic neural networks with complex media, fibers, and beam shaping techniques [2,3,4]. However, these studies always convert multichannel information to grayscale to encode it to laser light, and such a transformation causes information losses, which decreases the performance of the photonic neural networks. © 2025 Elsevier B.V., All rights reserved.