Various forms of linear transforms, such as the Fourier transform, are widely used for information processing in different applications. These transformations are typically implemented in the digital domain using an electronic processor, and the computational speed is limited by the capacity of the electronic chip used, which becomes a bottleneck for large data and image sizes. The solution to this problem is to replace the digital processor with a compatible optical processor and use light to process the information.
In a new treatise published in Light: Science and applicationA team of optical engineers and their colleagues, led by Professor Aydogan Ozcan of the Department of Electrical and Computer Engineering at the University of California, Los Angeles (UCLA), have developed a deep learning base. Design method For all optical calculations of any linear transformation.This all optical processor It uses spatially designed diffractive surfaces when manipulating light waves to calculate the linear transformation required as light passes through a series of diffractive surfaces. In this way, the transmission of input light through these diffractive surfaces completes the calculation of the desired linear transformation at the rate of light propagation. In addition to computational speed, these all-optical processors consume no computational power, except for lighting. Light, Becomes a passive, high-throughput computing system.
The analysis performed by the UCLA team Deep learningThe base design of these all-optical diffraction processors allows you to accurately synthesize any alignment. Transform The accuracy and diffraction efficiency of the optical conversion between the input and output surfaces and the resulting light conversion will improve significantly as the number of diffraction surfaces increases, demonstrating that the computing capabilities of deeper diffraction processors are more powerful. I am.
The success of this method has been demonstrated by performing a wide range of linear transformations, including, for example, randomly generated phase and amplitude transforms, Fourier transforms, image substitutions, and filtering operations. This computing framework is widely applied to any part of the electromagnetic spectrum to design an all-optical processor using spatially designed diffractive surfaces and universally perform linear transformations of any complex value. I can do it. It can also be used to form an all-optical information processing network to perform desired computational tasks between the input and output planes and provide a passive, power-free alternative to digital processors.
Onur Kulce et al, total optical compositing of any linear transformation using a diffractive surface, Light: Science and application (2021). DOI: 10.1038 / s41377-021-00623-5
Quote: Light uses a digital processor (September 24, 2021) obtained from https: //phys.org/news/2021-09-desired-linear-digital-processor.html on September 24, 2021. Calculate the required linear transformation without.
This document is subject to copyright. No part may be reproduced without written permission, except for fair transactions for personal investigation or research purposes. The content is provided for informational purposes only.
Light computes any linear transformation without a digital processor
Source link Light computes any linear transformation without a digital processor