Analog inference circuits for deep learning

Jeremy Holleman; Itamar Arel; Steven Young; Junjie Lu

doi:10.1109/BioCAS.2015.7348310

Analog inference circuits for deep learning

Jeremy Holleman, Itamar Arel, Steven Young, Junjie Lu

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution › peer-review

5 Scopus citations

Abstract

Deep Machine Learning (DML) algorithms have proven to be highly successful at challenging, high-dimensional learning problems, but their widespread deployment is limited by their heavy computational requirements and the associated power consumption. Analog computational circuits offer the potential for large improvements in power efficiency, but noise, mismatch, and other effects cause deviations from ideal computations. In this paper we describe circuits useful for DML algorithms, including a tunable-width bump circuit and a configurable distance calculator. We also discuss the impacts of computational errors on learning performance. Finally we will describe a complete deep learning engine implemented using current-mode analog circuits and compare its performance to digital equivalents.

Original language	English
Title of host publication	IEEE Biomedical Circuits and Systems Conference
Subtitle of host publication	Engineering for Healthy Minds and Able Bodies, BioCAS 2015 - Proceedings
Publisher	Institute of Electrical and Electronics Engineers Inc.
ISBN (Electronic)	9781479972333
DOIs	https://doi.org/10.1109/BioCAS.2015.7348310
State	Published - Dec 4 2015
Externally published	Yes
Event	11th IEEE Biomedical Circuits and Systems Conference, BioCAS 2015 - Atlanta, United States Duration: Oct 22 2015 → Oct 24 2015

Publication series

Name	IEEE Biomedical Circuits and Systems Conference: Engineering for Healthy Minds and Able Bodies, BioCAS 2015 - Proceedings

Conference

Conference	11th IEEE Biomedical Circuits and Systems Conference, BioCAS 2015
Country/Territory	United States
City	Atlanta
Period	10/22/15 → 10/24/15

Funding

This work was partially supported by the NSF SHF program through grant #CCF-1218492 and by the DARPA UPSIDE project through agreement #HR0011-13-2-0016.

Keywords

analog CMOS
deep learning
error modeling
neuromorphic computing

Access to Document

10.1109/BioCAS.2015.7348310

Cite this

Holleman, J., Arel, I., Young, S., & Lu, J. (2015). Analog inference circuits for deep learning. In IEEE Biomedical Circuits and Systems Conference: Engineering for Healthy Minds and Able Bodies, BioCAS 2015 - Proceedings Article 7348310 (IEEE Biomedical Circuits and Systems Conference: Engineering for Healthy Minds and Able Bodies, BioCAS 2015 - Proceedings). Institute of Electrical and Electronics Engineers Inc.. https://doi.org/10.1109/BioCAS.2015.7348310

Holleman, Jeremy ; Arel, Itamar ; Young, Steven et al. / Analog inference circuits for deep learning. IEEE Biomedical Circuits and Systems Conference: Engineering for Healthy Minds and Able Bodies, BioCAS 2015 - Proceedings. Institute of Electrical and Electronics Engineers Inc., 2015. (IEEE Biomedical Circuits and Systems Conference: Engineering for Healthy Minds and Able Bodies, BioCAS 2015 - Proceedings).

@inproceedings{be72be49920341ab84f3aba268ff99fa,

title = "Analog inference circuits for deep learning",

abstract = "Deep Machine Learning (DML) algorithms have proven to be highly successful at challenging, high-dimensional learning problems, but their widespread deployment is limited by their heavy computational requirements and the associated power consumption. Analog computational circuits offer the potential for large improvements in power efficiency, but noise, mismatch, and other effects cause deviations from ideal computations. In this paper we describe circuits useful for DML algorithms, including a tunable-width bump circuit and a configurable distance calculator. We also discuss the impacts of computational errors on learning performance. Finally we will describe a complete deep learning engine implemented using current-mode analog circuits and compare its performance to digital equivalents.",

keywords = "analog CMOS, deep learning, error modeling, neuromorphic computing",

author = "Jeremy Holleman and Itamar Arel and Steven Young and Junjie Lu",

note = "Publisher Copyright: {\textcopyright} 2015 IEEE.; 11th IEEE Biomedical Circuits and Systems Conference, BioCAS 2015 ; Conference date: 22-10-2015 Through 24-10-2015",

year = "2015",

month = dec,

day = "4",

doi = "10.1109/BioCAS.2015.7348310",

language = "English",

series = "IEEE Biomedical Circuits and Systems Conference: Engineering for Healthy Minds and Able Bodies, BioCAS 2015 - Proceedings",

publisher = "Institute of Electrical and Electronics Engineers Inc.",

booktitle = "IEEE Biomedical Circuits and Systems Conference",

}

Holleman, J, Arel, I, Young, S & Lu, J 2015, Analog inference circuits for deep learning. in IEEE Biomedical Circuits and Systems Conference: Engineering for Healthy Minds and Able Bodies, BioCAS 2015 - Proceedings., 7348310, IEEE Biomedical Circuits and Systems Conference: Engineering for Healthy Minds and Able Bodies, BioCAS 2015 - Proceedings, Institute of Electrical and Electronics Engineers Inc., 11th IEEE Biomedical Circuits and Systems Conference, BioCAS 2015, Atlanta, United States, 10/22/15. https://doi.org/10.1109/BioCAS.2015.7348310

Analog inference circuits for deep learning. / Holleman, Jeremy; Arel, Itamar; Young, Steven et al.
IEEE Biomedical Circuits and Systems Conference: Engineering for Healthy Minds and Able Bodies, BioCAS 2015 - Proceedings. Institute of Electrical and Electronics Engineers Inc., 2015. 7348310 (IEEE Biomedical Circuits and Systems Conference: Engineering for Healthy Minds and Able Bodies, BioCAS 2015 - Proceedings).

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution › peer-review

TY - GEN

T1 - Analog inference circuits for deep learning

AU - Holleman, Jeremy

AU - Arel, Itamar

AU - Young, Steven

AU - Lu, Junjie

PY - 2015/12/4

Y1 - 2015/12/4

N2 - Deep Machine Learning (DML) algorithms have proven to be highly successful at challenging, high-dimensional learning problems, but their widespread deployment is limited by their heavy computational requirements and the associated power consumption. Analog computational circuits offer the potential for large improvements in power efficiency, but noise, mismatch, and other effects cause deviations from ideal computations. In this paper we describe circuits useful for DML algorithms, including a tunable-width bump circuit and a configurable distance calculator. We also discuss the impacts of computational errors on learning performance. Finally we will describe a complete deep learning engine implemented using current-mode analog circuits and compare its performance to digital equivalents.

AB - Deep Machine Learning (DML) algorithms have proven to be highly successful at challenging, high-dimensional learning problems, but their widespread deployment is limited by their heavy computational requirements and the associated power consumption. Analog computational circuits offer the potential for large improvements in power efficiency, but noise, mismatch, and other effects cause deviations from ideal computations. In this paper we describe circuits useful for DML algorithms, including a tunable-width bump circuit and a configurable distance calculator. We also discuss the impacts of computational errors on learning performance. Finally we will describe a complete deep learning engine implemented using current-mode analog circuits and compare its performance to digital equivalents.

KW - analog CMOS

KW - deep learning

KW - error modeling

KW - neuromorphic computing

UR - http://www.scopus.com/inward/record.url?scp=84962745226&partnerID=8YFLogxK

U2 - 10.1109/BioCAS.2015.7348310

DO - 10.1109/BioCAS.2015.7348310

M3 - Conference contribution

AN - SCOPUS:84962745226

T3 - IEEE Biomedical Circuits and Systems Conference: Engineering for Healthy Minds and Able Bodies, BioCAS 2015 - Proceedings

BT - IEEE Biomedical Circuits and Systems Conference

PB - Institute of Electrical and Electronics Engineers Inc.

T2 - 11th IEEE Biomedical Circuits and Systems Conference, BioCAS 2015

Y2 - 22 October 2015 through 24 October 2015

ER -

Holleman J, Arel I, Young S, Lu J. Analog inference circuits for deep learning. In IEEE Biomedical Circuits and Systems Conference: Engineering for Healthy Minds and Able Bodies, BioCAS 2015 - Proceedings. Institute of Electrical and Electronics Engineers Inc. 2015. 7348310. (IEEE Biomedical Circuits and Systems Conference: Engineering for Healthy Minds and Able Bodies, BioCAS 2015 - Proceedings). doi: 10.1109/BioCAS.2015.7348310

Analog inference circuits for deep learning

Abstract

Publication series

Conference

Funding

Keywords

Access to Document

Other files and links

Fingerprint

Cite this