Always-On Speech Recognition Using TrueNorth, a Reconfigurable, Neurosynaptic Processor

Wei Yu Tsai, Davis R. Barch, Andrew S. Cassidy, Michael V. Debole, Alexander Andreopoulos, Bryan L. Jackson, Myron D. Flickner, John V. Arthur, Dharmendra S. Modha, John Morgan Sampson, Vijaykrishnan Narayanan

Research output: Contribution to journalArticle

8 Citations (Scopus)

Abstract

Deep neural networks (DNN) have been shown to be very effective at solving challenging problems in several areas of computing, including vision, speech, and natural language processing. However, traditional platforms for implementing these DNNs are often very power hungry, which has lead to significant efforts in the development of configurable platforms capable of implementing these DNNs efficiently. One of these platforms, the IBM TrueNorth processor, has demonstrated very low operating power in performing visual computing and neural network classification tasks in real-Time. The neuron computation, synaptic memory, and communication fabrics are all configurable, so that a wide range of network types and topologies can be mapped to TrueNorth. This reconfigurability translates into the capability to support a wide range of low-power functions in addition to feed-forward DNN classifiers, including for example, the audio processing functions presented here.In this work, we propose an end-To-end audio processing pipeline that is implemented entirely on a TrueNorth processor and designed to specifically leverage the highly-parallel, low-precision computing primitives TrueNorth offers. As part of this pipeline, we develop an audio feature extractor (LATTE) designed for implementation on TrueNorth, and explore the tradeoffs among several design variants in terms of accuracy, power, and performance. We customize the energy-efficient deep neuromorphic networks structures that our design utilizes as the classifier and show how classifier parameters can trade between power and accuracy. In addition to enabling a wide range of diverse functions, the reconfigurability of TrueNorth enables re-Training and re-programming the system to satisfy varying energy, speed, area, and accuracy requirements. The resulting system's end-To-end power consumption can be as low as 14.43 mW , which would give up to 100 hours of continuous usage with button cell batteries (CR3023 1.5 Whr ) or 450 hours with cellphone batteries (iPhone 6s $6.55 Whr ).

Original languageEnglish (US)
Article number7750640
Pages (from-to)996-1007
Number of pages12
JournalIEEE Transactions on Computers
Volume66
Issue number6
DOIs
StatePublished - Jun 1 2017

Fingerprint

Speech Recognition
Speech recognition
Reconfigurability
Classifiers
Classifier
Battery
Computing
Processing
Pipelines
Neural Networks
Range of data
Extractor
Feedforward Neural Networks
Power Function
Energy Efficient
Network Structure
Leverage
Natural Language
Neurons
Power Consumption

All Science Journal Classification (ASJC) codes

  • Software
  • Theoretical Computer Science
  • Hardware and Architecture
  • Computational Theory and Mathematics

Cite this

Tsai, W. Y., Barch, D. R., Cassidy, A. S., Debole, M. V., Andreopoulos, A., Jackson, B. L., ... Narayanan, V. (2017). Always-On Speech Recognition Using TrueNorth, a Reconfigurable, Neurosynaptic Processor. IEEE Transactions on Computers, 66(6), 996-1007. [7750640]. https://doi.org/10.1109/TC.2016.2630683
Tsai, Wei Yu ; Barch, Davis R. ; Cassidy, Andrew S. ; Debole, Michael V. ; Andreopoulos, Alexander ; Jackson, Bryan L. ; Flickner, Myron D. ; Arthur, John V. ; Modha, Dharmendra S. ; Sampson, John Morgan ; Narayanan, Vijaykrishnan. / Always-On Speech Recognition Using TrueNorth, a Reconfigurable, Neurosynaptic Processor. In: IEEE Transactions on Computers. 2017 ; Vol. 66, No. 6. pp. 996-1007.
@article{48ab2a4db4834174ae54b5dfe57b26fb,
title = "Always-On Speech Recognition Using TrueNorth, a Reconfigurable, Neurosynaptic Processor",
abstract = "Deep neural networks (DNN) have been shown to be very effective at solving challenging problems in several areas of computing, including vision, speech, and natural language processing. However, traditional platforms for implementing these DNNs are often very power hungry, which has lead to significant efforts in the development of configurable platforms capable of implementing these DNNs efficiently. One of these platforms, the IBM TrueNorth processor, has demonstrated very low operating power in performing visual computing and neural network classification tasks in real-Time. The neuron computation, synaptic memory, and communication fabrics are all configurable, so that a wide range of network types and topologies can be mapped to TrueNorth. This reconfigurability translates into the capability to support a wide range of low-power functions in addition to feed-forward DNN classifiers, including for example, the audio processing functions presented here.In this work, we propose an end-To-end audio processing pipeline that is implemented entirely on a TrueNorth processor and designed to specifically leverage the highly-parallel, low-precision computing primitives TrueNorth offers. As part of this pipeline, we develop an audio feature extractor (LATTE) designed for implementation on TrueNorth, and explore the tradeoffs among several design variants in terms of accuracy, power, and performance. We customize the energy-efficient deep neuromorphic networks structures that our design utilizes as the classifier and show how classifier parameters can trade between power and accuracy. In addition to enabling a wide range of diverse functions, the reconfigurability of TrueNorth enables re-Training and re-programming the system to satisfy varying energy, speed, area, and accuracy requirements. The resulting system's end-To-end power consumption can be as low as 14.43 mW , which would give up to 100 hours of continuous usage with button cell batteries (CR3023 1.5 Whr ) or 450 hours with cellphone batteries (iPhone 6s $6.55 Whr ).",
author = "Tsai, {Wei Yu} and Barch, {Davis R.} and Cassidy, {Andrew S.} and Debole, {Michael V.} and Alexander Andreopoulos and Jackson, {Bryan L.} and Flickner, {Myron D.} and Arthur, {John V.} and Modha, {Dharmendra S.} and Sampson, {John Morgan} and Vijaykrishnan Narayanan",
year = "2017",
month = "6",
day = "1",
doi = "10.1109/TC.2016.2630683",
language = "English (US)",
volume = "66",
pages = "996--1007",
journal = "IEEE Transactions on Computers",
issn = "0018-9340",
publisher = "IEEE Computer Society",
number = "6",

}

Tsai, WY, Barch, DR, Cassidy, AS, Debole, MV, Andreopoulos, A, Jackson, BL, Flickner, MD, Arthur, JV, Modha, DS, Sampson, JM & Narayanan, V 2017, 'Always-On Speech Recognition Using TrueNorth, a Reconfigurable, Neurosynaptic Processor', IEEE Transactions on Computers, vol. 66, no. 6, 7750640, pp. 996-1007. https://doi.org/10.1109/TC.2016.2630683

Always-On Speech Recognition Using TrueNorth, a Reconfigurable, Neurosynaptic Processor. / Tsai, Wei Yu; Barch, Davis R.; Cassidy, Andrew S.; Debole, Michael V.; Andreopoulos, Alexander; Jackson, Bryan L.; Flickner, Myron D.; Arthur, John V.; Modha, Dharmendra S.; Sampson, John Morgan; Narayanan, Vijaykrishnan.

In: IEEE Transactions on Computers, Vol. 66, No. 6, 7750640, 01.06.2017, p. 996-1007.

Research output: Contribution to journalArticle

TY - JOUR

T1 - Always-On Speech Recognition Using TrueNorth, a Reconfigurable, Neurosynaptic Processor

AU - Tsai, Wei Yu

AU - Barch, Davis R.

AU - Cassidy, Andrew S.

AU - Debole, Michael V.

AU - Andreopoulos, Alexander

AU - Jackson, Bryan L.

AU - Flickner, Myron D.

AU - Arthur, John V.

AU - Modha, Dharmendra S.

AU - Sampson, John Morgan

AU - Narayanan, Vijaykrishnan

PY - 2017/6/1

Y1 - 2017/6/1

N2 - Deep neural networks (DNN) have been shown to be very effective at solving challenging problems in several areas of computing, including vision, speech, and natural language processing. However, traditional platforms for implementing these DNNs are often very power hungry, which has lead to significant efforts in the development of configurable platforms capable of implementing these DNNs efficiently. One of these platforms, the IBM TrueNorth processor, has demonstrated very low operating power in performing visual computing and neural network classification tasks in real-Time. The neuron computation, synaptic memory, and communication fabrics are all configurable, so that a wide range of network types and topologies can be mapped to TrueNorth. This reconfigurability translates into the capability to support a wide range of low-power functions in addition to feed-forward DNN classifiers, including for example, the audio processing functions presented here.In this work, we propose an end-To-end audio processing pipeline that is implemented entirely on a TrueNorth processor and designed to specifically leverage the highly-parallel, low-precision computing primitives TrueNorth offers. As part of this pipeline, we develop an audio feature extractor (LATTE) designed for implementation on TrueNorth, and explore the tradeoffs among several design variants in terms of accuracy, power, and performance. We customize the energy-efficient deep neuromorphic networks structures that our design utilizes as the classifier and show how classifier parameters can trade between power and accuracy. In addition to enabling a wide range of diverse functions, the reconfigurability of TrueNorth enables re-Training and re-programming the system to satisfy varying energy, speed, area, and accuracy requirements. The resulting system's end-To-end power consumption can be as low as 14.43 mW , which would give up to 100 hours of continuous usage with button cell batteries (CR3023 1.5 Whr ) or 450 hours with cellphone batteries (iPhone 6s $6.55 Whr ).

AB - Deep neural networks (DNN) have been shown to be very effective at solving challenging problems in several areas of computing, including vision, speech, and natural language processing. However, traditional platforms for implementing these DNNs are often very power hungry, which has lead to significant efforts in the development of configurable platforms capable of implementing these DNNs efficiently. One of these platforms, the IBM TrueNorth processor, has demonstrated very low operating power in performing visual computing and neural network classification tasks in real-Time. The neuron computation, synaptic memory, and communication fabrics are all configurable, so that a wide range of network types and topologies can be mapped to TrueNorth. This reconfigurability translates into the capability to support a wide range of low-power functions in addition to feed-forward DNN classifiers, including for example, the audio processing functions presented here.In this work, we propose an end-To-end audio processing pipeline that is implemented entirely on a TrueNorth processor and designed to specifically leverage the highly-parallel, low-precision computing primitives TrueNorth offers. As part of this pipeline, we develop an audio feature extractor (LATTE) designed for implementation on TrueNorth, and explore the tradeoffs among several design variants in terms of accuracy, power, and performance. We customize the energy-efficient deep neuromorphic networks structures that our design utilizes as the classifier and show how classifier parameters can trade between power and accuracy. In addition to enabling a wide range of diverse functions, the reconfigurability of TrueNorth enables re-Training and re-programming the system to satisfy varying energy, speed, area, and accuracy requirements. The resulting system's end-To-end power consumption can be as low as 14.43 mW , which would give up to 100 hours of continuous usage with button cell batteries (CR3023 1.5 Whr ) or 450 hours with cellphone batteries (iPhone 6s $6.55 Whr ).

UR - http://www.scopus.com/inward/record.url?scp=85019174924&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85019174924&partnerID=8YFLogxK

U2 - 10.1109/TC.2016.2630683

DO - 10.1109/TC.2016.2630683

M3 - Article

AN - SCOPUS:85019174924

VL - 66

SP - 996

EP - 1007

JO - IEEE Transactions on Computers

JF - IEEE Transactions on Computers

SN - 0018-9340

IS - 6

M1 - 7750640

ER -

Tsai WY, Barch DR, Cassidy AS, Debole MV, Andreopoulos A, Jackson BL et al. Always-On Speech Recognition Using TrueNorth, a Reconfigurable, Neurosynaptic Processor. IEEE Transactions on Computers. 2017 Jun 1;66(6):996-1007. 7750640. https://doi.org/10.1109/TC.2016.2630683