Publication detail

Adaptive verifiability-driven strategy for evolutionary approximation of arithmetic circuits

ČEŠKA, M. MATYÁŠ, J. MRÁZEK, V. SEKANINA, L. VAŠÍČEK, Z. VOJNAR, T.

Original Title

Adaptive verifiability-driven strategy for evolutionary approximation of arithmetic circuits

Type

journal article in Web of Science

Language

English

Original Abstract

We present a novel approach for designing complex approximate arithmetic circuits that trade correctness for power consumption and play important role in many energy-aware applications. Our approach integrates in a unique way formal methods providing formal guarantees on the approximation error into an evolutionary circuit optimisation algorithm. The key idea is to employ a novel adaptive search strategy that drives the evolution towards promptly verifiable approximate circuits. As demonstrated in an extensive evaluation including several structurally different arithmetic circuits and target precisions, the search strategy provides superior scalability and versatility with respect to various approximation scenarios. Our approach significantly improves capabilities of the existing methods and paves a way towards an automated design process of provably-correct circuit approximations.

Keywords

Approximate computing,  Energy efficiency,  Circuit optimisation,  Genetic programming

Authors

ČEŠKA, M.; MATYÁŠ, J.; MRÁZEK, V.; SEKANINA, L.; VAŠÍČEK, Z.; VOJNAR, T.

Released

22. 6. 2020

ISBN

1568-4946

Periodical

APPLIED SOFT COMPUTING

Year of study

95

Number

106466

State

Kingdom of the Netherlands

Pages from

1

Pages to

17

Pages count

17

URL

https://www.fit.vut.cz/research/publication/12312/

BibTex

@article{BUT168169,
  author="Milan {Češka} and Jiří {Matyáš} and Vojtěch {Mrázek} and Lukáš {Sekanina} and Zdeněk {Vašíček} and Tomáš {Vojnar}",
  title="Adaptive verifiability-driven strategy for evolutionary approximation of arithmetic circuits",
  journal="APPLIED SOFT COMPUTING",
  year="2020",
  volume="95",
  number="106466",
  pages="1--17",
  doi="10.1016/j.asoc.2020.106466",
  issn="1568-4946",
  url="https://www.fit.vut.cz/research/publication/12312/"
}

Documents

preprint.pdf
main.pdf