Please use this identifier to cite or link to this item: https://hdl.handle.net/20.500.11851/1977
Title: Error Recovery Through Partial Value Similarity
Authors: Eker, Abdulaziz
Ergin, Oğuz
143001
Keywords: Computer architecture
Forecasting
Branch prediction
Issue Date: 2016
Publisher: IEEE
Source: Eker, A., & Ergin, O. (2016, September). Error recovery through partial value similarity. In 2016 IEEE International Symposium on Defect and Fault Tolerance in VLSI and Nanotechnology Systems (DFT) (pp. 103-106). IEEE.
Abstract: Soft errors arose as a critical problem for microprocessor designers due to shrinking feature sizes and increasing clock rates. Many attempts appeared to protect the register file which is the main storage component in modern microprocessors. Exploiting existing replica values in the register file is a recently proposed method to correct errors that occur on the register values. In this paper, with the observation that partial matches are clustered in the least significant bytes of the similar values in the register file, we propose to extend the coverage of existing replica based error correction schemes. Our schemes almost double the coverage of previously proposed mechanisms. The proposed method can recover 39.8% of the stored values in the register file by using this similarity and single-bit parity protection. The power overhead of this method constitutes only 3.1% of the register file's power budget.
Description: 29th IEEE International Symposium on Defect and Fault Tolerance in VLSI and Nanotechnology Systems (DFT) (2016 : Univ Connecticut, Storrs, CT)
URI: https://ieeexplore.ieee.org/document/7684078
https://hdl.handle.net/20.500.11851/1977
ISBN: 978-1-5090-3623-3
Appears in Collections:Bilgisayar Mühendisliği Bölümü / Department of Computer Engineering
Scopus İndeksli Yayınlar Koleksiyonu / Scopus Indexed Publications Collection
WoS İndeksli Yayınlar Koleksiyonu / WoS Indexed Publications Collection

Show full item record

CORE Recommender

Page view(s)

18
checked on Dec 26, 2022

Google ScholarTM

Check

Altmetric


Items in GCRIS Repository are protected by copyright, with all rights reserved, unless otherwise indicated.