Skip to content

Energy proportional neural network inference with adaptive voltage and frequency scaling

Research output: Contribution to journalArticle

Standard

Energy proportional neural network inference with adaptive voltage and frequency scaling. / Nunez-Yanez, Jose.

In: IEEE Transactions on Computers, Vol. 68, No. 5, 8531784, 01.05.2019, p. 676-687.

Research output: Contribution to journalArticle

Harvard

APA

Vancouver

Author

Nunez-Yanez, Jose. / Energy proportional neural network inference with adaptive voltage and frequency scaling. In: IEEE Transactions on Computers. 2019 ; Vol. 68, No. 5. pp. 676-687.

Bibtex

@article{ef8296535439481aa729e8d35dbaeb8e,
title = "Energy proportional neural network inference with adaptive voltage and frequency scaling",
abstract = "This research presents the extension and application of a voltage and frequency scaling framework called Elongate to a high-performance and reconfigurable binarized neural network. The neural network is created in the FPGA reconfigurable fabric and coupled to a multiprocessor host that controls the operational point to obtain energy proportionality. Elongate instruments a design netlist by inserting timing detectors to enable the exploitation of the operating margins of a device reliably. The elongated neural network is re-targeted to devices with different nominal operating voltages and fabricated with 28 nm (i.e., Zynq) and 16nm (i.e., Zynq Ultrascale) feature sizes showing the portability of the framework to advanced process nodes. New hardware and software components are created to support the 16nm fabric microarchitecture and a comparison in terms of power, energy and performance with the older 28 nm process is performed. The results show that Elongate can obtain new performance and energy points that are up to 86 percent better than nominal at the same level of classification accuracy. Trade-offs between energy and performance are also possible with a large dynamic range of valid working points available. The results also indicate that the built-in neural network robustness allows operation beyond the first point of error while maintaining the classification accuracy largely unaffected.",
keywords = "convolutional neural network, DVFS, energy efficiency, FPGA",
author = "Jose Nunez-Yanez",
year = "2019",
month = "5",
day = "1",
doi = "10.1109/TC.2018.2879333",
language = "English",
volume = "68",
pages = "676--687",
journal = "IEEE Transactions on Computers",
issn = "0018-9340",
publisher = "IEEE Computer Society",
number = "5",

}

RIS - suitable for import to EndNote

TY - JOUR

T1 - Energy proportional neural network inference with adaptive voltage and frequency scaling

AU - Nunez-Yanez, Jose

PY - 2019/5/1

Y1 - 2019/5/1

N2 - This research presents the extension and application of a voltage and frequency scaling framework called Elongate to a high-performance and reconfigurable binarized neural network. The neural network is created in the FPGA reconfigurable fabric and coupled to a multiprocessor host that controls the operational point to obtain energy proportionality. Elongate instruments a design netlist by inserting timing detectors to enable the exploitation of the operating margins of a device reliably. The elongated neural network is re-targeted to devices with different nominal operating voltages and fabricated with 28 nm (i.e., Zynq) and 16nm (i.e., Zynq Ultrascale) feature sizes showing the portability of the framework to advanced process nodes. New hardware and software components are created to support the 16nm fabric microarchitecture and a comparison in terms of power, energy and performance with the older 28 nm process is performed. The results show that Elongate can obtain new performance and energy points that are up to 86 percent better than nominal at the same level of classification accuracy. Trade-offs between energy and performance are also possible with a large dynamic range of valid working points available. The results also indicate that the built-in neural network robustness allows operation beyond the first point of error while maintaining the classification accuracy largely unaffected.

AB - This research presents the extension and application of a voltage and frequency scaling framework called Elongate to a high-performance and reconfigurable binarized neural network. The neural network is created in the FPGA reconfigurable fabric and coupled to a multiprocessor host that controls the operational point to obtain energy proportionality. Elongate instruments a design netlist by inserting timing detectors to enable the exploitation of the operating margins of a device reliably. The elongated neural network is re-targeted to devices with different nominal operating voltages and fabricated with 28 nm (i.e., Zynq) and 16nm (i.e., Zynq Ultrascale) feature sizes showing the portability of the framework to advanced process nodes. New hardware and software components are created to support the 16nm fabric microarchitecture and a comparison in terms of power, energy and performance with the older 28 nm process is performed. The results show that Elongate can obtain new performance and energy points that are up to 86 percent better than nominal at the same level of classification accuracy. Trade-offs between energy and performance are also possible with a large dynamic range of valid working points available. The results also indicate that the built-in neural network robustness allows operation beyond the first point of error while maintaining the classification accuracy largely unaffected.

KW - convolutional neural network

KW - DVFS

KW - energy efficiency

KW - FPGA

UR - http://www.scopus.com/inward/record.url?scp=85056358969&partnerID=8YFLogxK

U2 - 10.1109/TC.2018.2879333

DO - 10.1109/TC.2018.2879333

M3 - Article

VL - 68

SP - 676

EP - 687

JO - IEEE Transactions on Computers

JF - IEEE Transactions on Computers

SN - 0018-9340

IS - 5

M1 - 8531784

ER -