An analog neural network processor with programmable topology

Bernhard E. Boser, Eduard Sackinger, Jane Bromley, Yann LeCun, Lawrence D. Jackel

Research output: Contribution to journalArticle

Abstract

The architecture, implementation, and applications of a special-purpose neural network processor are described. The chip performs over 2000 multiplications and additions simultaneously. Its data path is particularly suitable for the convolutional topologies that are typical in classification networks, but can also be configured for fully connected or feedback topologies. Resources can be multiplexed to permit implementation of networks with several hundreds of thousands of connections on a single chip. Computations are performed with 6-b accuracy for the weights and 3 b for the neuron states. Analog processing is used internally for reduced power dissipation and higher density, but all input/output is digital to simplify system integration. The practicality of the chip is demonstrated with an implementation of a neural network for optical character recognition. This network contains over 130,000 connections and was evaluated in 1 ms.

Original languageEnglish (US)
Pages (from-to)2017-2025
Number of pages9
JournalIEEE Journal of Solid-State Circuits
Volume26
Issue number12
DOIs
StatePublished - Dec 1991

Fingerprint

Topology
Neural networks
Optical character recognition
Neurons
Energy dissipation
Feedback
Processing

ASJC Scopus subject areas

  • Electrical and Electronic Engineering

Cite this

An analog neural network processor with programmable topology. / Boser, Bernhard E.; Sackinger, Eduard; Bromley, Jane; LeCun, Yann; Jackel, Lawrence D.

In: IEEE Journal of Solid-State Circuits, Vol. 26, No. 12, 12.1991, p. 2017-2025.

Research output: Contribution to journalArticle

Boser, BE, Sackinger, E, Bromley, J, LeCun, Y & Jackel, LD 1991, 'An analog neural network processor with programmable topology', IEEE Journal of Solid-State Circuits, vol. 26, no. 12, pp. 2017-2025. https://doi.org/10.1109/4.104196
Boser, Bernhard E. ; Sackinger, Eduard ; Bromley, Jane ; LeCun, Yann ; Jackel, Lawrence D. / An analog neural network processor with programmable topology. In: IEEE Journal of Solid-State Circuits. 1991 ; Vol. 26, No. 12. pp. 2017-2025.
@article{f773c4cc13064abe9264f06f8c0da632,
title = "An analog neural network processor with programmable topology",
abstract = "The architecture, implementation, and applications of a special-purpose neural network processor are described. The chip performs over 2000 multiplications and additions simultaneously. Its data path is particularly suitable for the convolutional topologies that are typical in classification networks, but can also be configured for fully connected or feedback topologies. Resources can be multiplexed to permit implementation of networks with several hundreds of thousands of connections on a single chip. Computations are performed with 6-b accuracy for the weights and 3 b for the neuron states. Analog processing is used internally for reduced power dissipation and higher density, but all input/output is digital to simplify system integration. The practicality of the chip is demonstrated with an implementation of a neural network for optical character recognition. This network contains over 130,000 connections and was evaluated in 1 ms.",
author = "Boser, {Bernhard E.} and Eduard Sackinger and Jane Bromley and Yann LeCun and Jackel, {Lawrence D.}",
year = "1991",
month = "12",
doi = "10.1109/4.104196",
language = "English (US)",
volume = "26",
pages = "2017--2025",
journal = "IEEE Journal of Solid-State Circuits",
issn = "0018-9200",
publisher = "Institute of Electrical and Electronics Engineers Inc.",
number = "12",

}

TY - JOUR

T1 - An analog neural network processor with programmable topology

AU - Boser, Bernhard E.

AU - Sackinger, Eduard

AU - Bromley, Jane

AU - LeCun, Yann

AU - Jackel, Lawrence D.

PY - 1991/12

Y1 - 1991/12

N2 - The architecture, implementation, and applications of a special-purpose neural network processor are described. The chip performs over 2000 multiplications and additions simultaneously. Its data path is particularly suitable for the convolutional topologies that are typical in classification networks, but can also be configured for fully connected or feedback topologies. Resources can be multiplexed to permit implementation of networks with several hundreds of thousands of connections on a single chip. Computations are performed with 6-b accuracy for the weights and 3 b for the neuron states. Analog processing is used internally for reduced power dissipation and higher density, but all input/output is digital to simplify system integration. The practicality of the chip is demonstrated with an implementation of a neural network for optical character recognition. This network contains over 130,000 connections and was evaluated in 1 ms.

AB - The architecture, implementation, and applications of a special-purpose neural network processor are described. The chip performs over 2000 multiplications and additions simultaneously. Its data path is particularly suitable for the convolutional topologies that are typical in classification networks, but can also be configured for fully connected or feedback topologies. Resources can be multiplexed to permit implementation of networks with several hundreds of thousands of connections on a single chip. Computations are performed with 6-b accuracy for the weights and 3 b for the neuron states. Analog processing is used internally for reduced power dissipation and higher density, but all input/output is digital to simplify system integration. The practicality of the chip is demonstrated with an implementation of a neural network for optical character recognition. This network contains over 130,000 connections and was evaluated in 1 ms.

UR - http://www.scopus.com/inward/record.url?scp=0026384824&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=0026384824&partnerID=8YFLogxK

U2 - 10.1109/4.104196

DO - 10.1109/4.104196

M3 - Article

VL - 26

SP - 2017

EP - 2025

JO - IEEE Journal of Solid-State Circuits

JF - IEEE Journal of Solid-State Circuits

SN - 0018-9200

IS - 12

ER -