Skip to main content

Lund University Publications

LUND UNIVERSITY LIBRARIES

Scalable and power efficient approximate multipliers for error-resilient applications in 22 nm

Åberg, Victor LU orcid ; Castillo Mohedano, Sergio LU ; Westring, Kristoffer LU ; Allfjord, Alex LU ; Andersson, Per LU ; Rodrigues, Joachim LU and Nouripayam, Masoud LU (2026)
Abstract
Approximate multipliers (AMs) show great potential to achieve substantial area and power savings when replacing their exact counterparts in error-resilient applications. In this paper, we propose an AM architecture, scalable across operand widths. Our 8-bit AM achieves a 1.91% Mean Relative Error Distance (MRED) and a power-area-delay product (PADP) gain of >78.2% over a Wallace multiplier. We also introduce an evaluation methodology focused on post-implementation hardware savings. Beyond standalone benchmarking, measurements of a 22nm chip confirm substantial power reductions of 32.9% when replacing exact multipliers with our AM.
Please use this url to cite or link to this publication:
author
; ; ; ; ; and
organization
publishing date
type
Chapter in Book/Report/Conference proceeding
publication status
in press
subject
host publication
2026 IEEE European Solid-State Electronics Research Conference (ESSERC)
pages
6 pages
language
English
LU publication?
yes
id
7fddeb7e-e4e1-42c8-9556-778bdeecda76
date added to LUP
2026-08-25 11:45:49
date last changed
2026-09-09 15:10:45
@inproceedings{7fddeb7e-e4e1-42c8-9556-778bdeecda76,
  abstract     = {{Approximate multipliers (AMs) show great potential to achieve substantial area and power savings when replacing their exact counterparts in error-resilient applications. In this paper, we propose an AM architecture, scalable across operand widths. Our 8-bit AM achieves a 1.91% Mean Relative Error Distance (MRED) and a power-area-delay product (PADP) gain of >78.2% over a Wallace multiplier. We also introduce an evaluation methodology focused on post-implementation hardware savings. Beyond standalone benchmarking, measurements of a 22nm chip confirm substantial power reductions of 32.9% when replacing exact multipliers with our AM.}},
  author       = {{Åberg, Victor and Castillo Mohedano, Sergio and Westring, Kristoffer and Allfjord, Alex and Andersson, Per and Rodrigues, Joachim and Nouripayam, Masoud}},
  booktitle    = {{2026 IEEE European Solid-State Electronics Research Conference (ESSERC)}},
  language     = {{eng}},
  title        = {{Scalable and power efficient approximate multipliers for error-resilient applications in 22 nm}},
  year         = {{2026}},
}