4.8 Article

Chromatic Plasmonic Polarizer-Based Synapse for All-Optical Convolutional Neural Network

Journal

NANO LETTERS
Volume -, Issue -, Pages -

Publisher

AMER CHEMICAL SOC
DOI: 10.1021/acs.nanolett.3c02194

Keywords

Surface plasmon polariton; Polarizer; Opticalsynapse; Optical convolutional neural network

Ask authors/readers for more resources

This study presents a novel optical synapse based on surface plasmon resonance polarizer for implementing convolutional filters and optical convolutional neural networks. The synapse consists of nanoscale crossed gold arrays that strongly respond to the polarization angle of incident light. The experimental results show that the presented synapse achieved excellent performance in classifying the MNIST handwritten digit data set, with a classification accuracy of over 98% after training on 1,000 images and testing on a separate set of 10,000 images. This research provides a promising approach for designing artificial neural networks with efficient hardware and energy consumption, low cost, and scalable fabrication.
Emergingmemory devices have been demonstrated as artificial synapsesfor neural networks. However, the process of rewriting these synapsesis often inefficient, in terms of hardware and energy usage. Herein,we present a novel surface plasmon resonance polarizer-based all-opticalsynapse for realizing convolutional filters and optical convolutionalneural networks. The synaptic device comprises nanoscale crossed goldarrays with varying vertical and horizontal arms that respond stronglyto the incident light's polarization angle. The presented synapsein an optical convolutional neural network achieved excellent performancein four different convolutional results for classifying the ModifiedNational Institute of Standards and Technology (MNIST) handwrittendigit data set. After training on 1,000 images, the network achieveda classification accuracy of over 98% when tested on a separate setof 10,000 images. This presents a promising approach for designingartificial neural networks with efficient hardware and energy consumption,low cost, and scalable fabrication.

Authors

I am an author on this paper
Click your name to claim this paper and add it to your profile.

Reviews

Primary Rating

4.8
Not enough ratings

Secondary Ratings

Novelty
-
Significance
-
Scientific rigor
-
Rate this paper

Recommended

No Data Available
No Data Available