SiMul: An Algorithm-Driven Approximate Multiplier Design for Machine Learning
- Authors
- Liu, Zhenhong; Yazdanbakhsh, Amir; Park, Taejoon; Esmaeilzadeh, Hadi; Kim, Nam Sung
- Issue Date
- Jul-2018
- Publisher
- IEEE COMPUTER SOC
- Keywords
- approximate computing; hardware; machine learning; multiplier; neural network
- Citation
- IEEE MICRO, v.38, no.4, pp.50 - 59
- Indexed
- SCIE
SCOPUS
- Journal Title
- IEEE MICRO
- Volume
- 38
- Number
- 4
- Start Page
- 50
- End Page
- 59
- URI
- https://scholarworks.bwise.kr/erica/handle/2021.sw.erica/5810
- DOI
- 10.1109/MM.2018.043191125
- ISSN
- 0272-1732
- Abstract
- The need to support various machine learning (ML) algorithms on energy-constrained computing devices has steadily grown. In this article, we propose an approximate multiplier, which is a key hardware component in various ML accelerators. Dubbed SiMul, our approximate multiplier features user-controlled precision that exploits the common characteristics of ML algorithms. SiMul supports a tradeoff between compute precision and energy consumption at runtime, reducing the energy consumption of the accelerator while satisfying a desired inference accuracy requirement. Compared improves the energy efficiency of multiplication by 11.6x to 3.2x while achieving 81.7-percent to 98.5-percent precision for individual multiplication operations (96.0-, 97.8-, and 97.7-percent inference accuracy for three distinct applications, respectively, compared to the baseline inference accuracy of 98.3, 99.0, and 97.7 percent using precise multipliers). A neural accelerator implemented with our multiplier can provide 1.7x (up to 2.1x) higher energy efficiency over one implemented with the precise multiplier with a negligible impact on the accuracy of the output for various applications.
- Files in This Item
-
Go to Link
- Appears in
Collections - COLLEGE OF ENGINEERING SCIENCES > DEPARTMENT OF ROBOT ENGINEERING > 1. Journal Articles
![qrcode](https://api.qrserver.com/v1/create-qr-code/?size=55x55&data=https://scholarworks.bwise.kr/erica/handle/2021.sw.erica/5810)
Items in ScholarWorks are protected by copyright, with all rights reserved, unless otherwise indicated.