Expand this Topic clickable element to expand a topic
Skip to content
Optica Publishing Group

Four-dimensional neural network architecture

Open Access Open Access

Abstract

This paper describes a four-dimensional (4-D) optical-interconnect architecture that has the potential to provide much larger numbers of interconnects than can be achieved with electronics. Many applications require more interconnections than can be realized by using electronics (approximately equal to those of a stupid honeybee). For example, vision applications, including infrared search and track, may require more than 1010 interconnects and 1013 interconnects per second. This interconnect architecture makes use of holographic interconnects stored in spectral-hole-burning (SHB) materials. The use of holography and spectral selectivity provides access to the fundamental capacity of the material to achieve a large storage capacity of interconnect values. In addition to the three spatial dimensions available with volume holographic techniques, a fourth independent dimension, laser frequency, is available. Information can be stored and read independently at many laser wavelengths in SHB materials. Our 4-D interconnect concept provides a way to fold the access to the four spectral and spatial dimensions into a compact package. The use of four independent dimensions allows implementation of a completely general 2-D-to-2-D interconnect, which makes optimum use of the space-bandwidth product of the spatial light modulators that supply the input and training patterns.

© 1990 Optical Society of America

PDF Article
More Like This
Photonic neural-network architecture based on incoherent-coherent holographic interconnections

B. K. Jenkins, A. R. Tanguay, S. Piazzolla, G. C. Petrisor, and P. Asthana
MVV2 OSA Annual Meeting (FIO) 1990

Wavelength-multiplexed volume holograms for fully interconnected neural networks

Robert T. Weverka, Kelvin Wagner, and Mark Saffman
MVV6 OSA Annual Meeting (FIO) 1990

Optical disk based architectures for pattern recognition and neural network implementation

M. A. Neifeld, A. Yamamura, Seiji Kobayashi, and D. Psaltis
MX2 OSA Annual Meeting (FIO) 1989

Select as filters


Select Topics Cancel
© Copyright 2024 | Optica Publishing Group. All rights reserved, including rights for text and data mining and training of artificial technologies or similar technologies.