Techniques for power reduction in an SIMD implementation of the VQ/SOM algorithms

David C Hendry, Roberta Cambio

Research output: Contribution to journalArticle

1 Citation (Scopus)

Abstract

Hardware implementations of the VQ (vector quantization) and SOM (self organizing map) permit the deployment of these computationally intensive algorithms as single chips or IP cores. This paper discusses the design of an IP core based on an SIMD (single instruction multiple data) processor array
for such an implementation with emphasis on those aspects of the design which lead to a low power implementation. Power reduction techniques described are: local memory sharing between processors; processor instruction set and data path organization; implementation of the winner take all calculation;
and use of a thresholding algorithm to permit power down of processors during the distance calculation. It is shown that with a typical 0:13 um low power semiconductor process and with a clock speed of 100MHz the power dissipation per processor is approximately 1mW without use of thresholding. Including thresholding reduces this power to less than 0.5mW per processor. Area for a
256 processor array with 256 8-bit vector elements per processor is 3.5mm 2.5 mm.
Original languageEnglish
Pages (from-to)291-300
Number of pages10
JournalNeurocomputing
Volume74
Issue number1-3
Early online date27 Mar 2010
DOIs
Publication statusPublished - Dec 2010

Fingerprint

Vector quantization
Self organizing maps
Semiconductors
Parallel processing systems
Clocks
Energy dissipation
Semiconductor materials
Hardware
Data storage equipment
Intellectual property core
Datasets

Keywords

  • vector quantization
  • self organising map
  • VLSi
  • SIMD
  • low power
  • system on chip

Cite this

Techniques for power reduction in an SIMD implementation of the VQ/SOM algorithms. / Hendry, David C; Cambio, Roberta.

In: Neurocomputing, Vol. 74, No. 1-3, 12.2010, p. 291-300.

Research output: Contribution to journalArticle

Hendry, David C ; Cambio, Roberta. / Techniques for power reduction in an SIMD implementation of the VQ/SOM algorithms. In: Neurocomputing. 2010 ; Vol. 74, No. 1-3. pp. 291-300.
@article{63071997db974f939f33ae29dbd21bf5,
title = "Techniques for power reduction in an SIMD implementation of the VQ/SOM algorithms",
abstract = "Hardware implementations of the VQ (vector quantization) and SOM (self organizing map) permit the deployment of these computationally intensive algorithms as single chips or IP cores. This paper discusses the design of an IP core based on an SIMD (single instruction multiple data) processor arrayfor such an implementation with emphasis on those aspects of the design which lead to a low power implementation. Power reduction techniques described are: local memory sharing between processors; processor instruction set and data path organization; implementation of the winner take all calculation;and use of a thresholding algorithm to permit power down of processors during the distance calculation. It is shown that with a typical 0:13 um low power semiconductor process and with a clock speed of 100MHz the power dissipation per processor is approximately 1mW without use of thresholding. Including thresholding reduces this power to less than 0.5mW per processor. Area for a256 processor array with 256 8-bit vector elements per processor is 3.5mm 2.5 mm.",
keywords = "vector quantization, self organising map, VLSi, SIMD, low power, system on chip",
author = "Hendry, {David C} and Roberta Cambio",
year = "2010",
month = "12",
doi = "10.1016/j.neucom.2010.03.005",
language = "English",
volume = "74",
pages = "291--300",
journal = "Neurocomputing",
issn = "0925-2312",
publisher = "ELSEVIER SCIENCE BV",
number = "1-3",

}

TY - JOUR

T1 - Techniques for power reduction in an SIMD implementation of the VQ/SOM algorithms

AU - Hendry, David C

AU - Cambio, Roberta

PY - 2010/12

Y1 - 2010/12

N2 - Hardware implementations of the VQ (vector quantization) and SOM (self organizing map) permit the deployment of these computationally intensive algorithms as single chips or IP cores. This paper discusses the design of an IP core based on an SIMD (single instruction multiple data) processor arrayfor such an implementation with emphasis on those aspects of the design which lead to a low power implementation. Power reduction techniques described are: local memory sharing between processors; processor instruction set and data path organization; implementation of the winner take all calculation;and use of a thresholding algorithm to permit power down of processors during the distance calculation. It is shown that with a typical 0:13 um low power semiconductor process and with a clock speed of 100MHz the power dissipation per processor is approximately 1mW without use of thresholding. Including thresholding reduces this power to less than 0.5mW per processor. Area for a256 processor array with 256 8-bit vector elements per processor is 3.5mm 2.5 mm.

AB - Hardware implementations of the VQ (vector quantization) and SOM (self organizing map) permit the deployment of these computationally intensive algorithms as single chips or IP cores. This paper discusses the design of an IP core based on an SIMD (single instruction multiple data) processor arrayfor such an implementation with emphasis on those aspects of the design which lead to a low power implementation. Power reduction techniques described are: local memory sharing between processors; processor instruction set and data path organization; implementation of the winner take all calculation;and use of a thresholding algorithm to permit power down of processors during the distance calculation. It is shown that with a typical 0:13 um low power semiconductor process and with a clock speed of 100MHz the power dissipation per processor is approximately 1mW without use of thresholding. Including thresholding reduces this power to less than 0.5mW per processor. Area for a256 processor array with 256 8-bit vector elements per processor is 3.5mm 2.5 mm.

KW - vector quantization

KW - self organising map

KW - VLSi

KW - SIMD

KW - low power

KW - system on chip

U2 - 10.1016/j.neucom.2010.03.005

DO - 10.1016/j.neucom.2010.03.005

M3 - Article

VL - 74

SP - 291

EP - 300

JO - Neurocomputing

JF - Neurocomputing

SN - 0925-2312

IS - 1-3

ER -