Energy-efficient accelerator architecture for stereo image matching using approximate computing and statistical error compensation

Eric P. Kim, Naresh R Shanbhag

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Abstract

Modern nanoscale processes exhibit stochastic behavior that can no longer be ignored. Statistical error compensation (SEC) has shown significant benefits in achieving energy efficiency and error resiliency by embracing the stochastic nature of the underlying process. Approximate computing (AC), on the other hand, employs deterministic designs that produce imprecise results to achieve energy efficiency. In this paper, we bridge the two design paradigms by utilizing SEC and AC in the design of a machine learning accelerator core. ANT, a form of SEC, was applied to an AC based stereo image matching implementation in a 45 nm process. Simulation results show that ANT combined with AC achieves energy savings of 44% compared to a conventional system, and 32.7% compared to an AC only system, while its performance degradation is less than 4%. This result shows that embracing the stochasticity of the architecture is crucial in achieving high energy efficiency, and that AC and ANT are synergistic.

Original languageEnglish (US)
Title of host publication2014 IEEE Global Conference on Signal and Information Processing, GlobalSIP 2014
PublisherInstitute of Electrical and Electronics Engineers Inc.
Pages55-59
Number of pages5
ISBN (Electronic)9781479970889
DOIs
StatePublished - Feb 5 2014
Event2014 IEEE Global Conference on Signal and Information Processing, GlobalSIP 2014 - Atlanta, United States
Duration: Dec 3 2014Dec 5 2014

Publication series

Name2014 IEEE Global Conference on Signal and Information Processing, GlobalSIP 2014

Other

Other2014 IEEE Global Conference on Signal and Information Processing, GlobalSIP 2014
CountryUnited States
CityAtlanta
Period12/3/1412/5/14

Keywords

  • Approximate computing
  • Low power
  • Statistical error compensation
  • Voltage overscaling

ASJC Scopus subject areas

  • Signal Processing
  • Information Systems

Fingerprint Dive into the research topics of 'Energy-efficient accelerator architecture for stereo image matching using approximate computing and statistical error compensation'. Together they form a unique fingerprint.

  • Cite this

    Kim, E. P., & Shanbhag, N. R. (2014). Energy-efficient accelerator architecture for stereo image matching using approximate computing and statistical error compensation. In 2014 IEEE Global Conference on Signal and Information Processing, GlobalSIP 2014 (pp. 55-59). [7032077] (2014 IEEE Global Conference on Signal and Information Processing, GlobalSIP 2014). Institute of Electrical and Electronics Engineers Inc.. https://doi.org/10.1109/GlobalSIP.2014.7032077