Design of Majority Logic-Based Approximate Booth Multipliers for Error-Tolerant Applications

Approximate computing at the nanoscale provides sufficiently accurate and often adaptive results to improve hardware efficiency for error-tolerant applications. Differently from conventional Boolean logic-based designs, many emerging nanotechnologies extensively assemble circuits using the voter-bas...

Full description

Saved in:
Bibliographic Details
Published in:IEEE transactions on nanotechnology 2022, Vol.21, p.81-89
Main Authors: Zhang, Tingting, Jiang, Honglan, Mo, Hai, Liu, Weiqiang, Lombardi, Fabrizio, Liu, Leibo, Han, Jie
Format: Article
Language:English
Subjects:
approximate multiplier
Approximation
Approximation algorithms
Artificial neural networks
Boolean algebra
Booth encoding
Circuit design
Design
Error analysis
Generators
Hardware
Image coding
Image processing
Logic
Logic gates
majority logic
Multipliers
Nanoscale devices
nanotechnologies
partial product generator
Shape
Citations: Items that this one cites
Items that cite this one
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Approximate computing at the nanoscale provides sufficiently accurate and often adaptive results to improve hardware efficiency for error-tolerant applications. Differently from conventional Boolean logic-based designs, many emerging nanotechnologies extensively assemble circuits using the voter-based majority logic (ML). In this letter, we investigate designs of approximate radix-4 Booth multipliers based on ML. Initially, we propose two new radix-4 Booth partial product (PP) generation methods by exploiting the characteristics of ML. Based on these methods, approximate PP generators are designed to produce single-sided or double-sided errors. The PPs are then reduced by using the features of errors to construct approximate multipliers. Specifically, complementary strategies guided by an analysis of error effects are developed to compensate for the accuracy loss and to reduce the hardware overhead during the PP reduction. The reduced PPs are then compressed by using full adders. Four approximate multipliers are proposed to offer various accuracy requirements for different applications. These designs show superior performance in power and area for emerging nanotechnologies. As case studies, image processing, a multiple-layer perceptron and a multi-task convolutional neural network are presented to show the validity and advantages of the proposed designs.
ISSN:1536-125X
1941-0085
DOI:10.1109/TNANO.2022.3145362