LI Yan, HU Yueming, ZENG Xiaoyang. Cost-Effective TMR Soft Error Tolerance Technique for Commercial Aerospace: Utilization of Approximate Computing[J]. Journal of Electronics & Information Technology, 2024, 46(5): 1604-1612. doi: 10.11999/JEIT231288
Citation:
LI Yan, HU Yueming, ZENG Xiaoyang. Cost-Effective TMR Soft Error Tolerance Technique for Commercial Aerospace: Utilization of Approximate Computing[J].
Journal of Electronics & Information Technology
, 2024, 46(5): 1604-1612.
doi:
10.11999/JEIT231288
LI Yan, HU Yueming, ZENG Xiaoyang. Cost-Effective TMR Soft Error Tolerance Technique for Commercial Aerospace: Utilization of Approximate Computing[J]. Journal of Electronics & Information Technology, 2024, 46(5): 1604-1612. doi: 10.11999/JEIT231288
Citation:
LI Yan, HU Yueming, ZENG Xiaoyang. Cost-Effective TMR Soft Error Tolerance Technique for Commercial Aerospace: Utilization of Approximate Computing[J].
Journal of Electronics & Information Technology
, 2024, 46(5): 1604-1612.
doi:
10.11999/JEIT231288
Cost-Effective TMR Soft Error Tolerance Technique for Commercial Aerospace: Utilization of Approximate Computing
doi:
10.11999/JEIT231288
LI Yan
,
HU Yueming
,
ZENG Xiaoyang
Fudan University State Key Laboratory of Integrated Chips and Systems, Shanghai 201203, China
Funds:
The National Natural Science Foundation of China (62204045), Shanghai Pujiang Program (22PJD005)
Abstract
Triple Modular Redundancy (TMR), as the most prevalent and effective technique for soft error mitigation technique, inevitably incurs substantial hardware overhead while meeting high fault-tolerance requirements. To achieve the trade-off between area, power and fault coverage and meet the requirement of low-cost and high-reliability circuit design, Approximate Triple Modular Redundancy (ATMR) is investigated and a Dynamic Adjustment Multi-Objective Optimization Framework based on Approximate Gate Library (ApxLib+DAMOO) is investigated. The basic optimization framework employs Non-dominated Sorting Genetic Algorithm II (NSGA-II), achieving rapidly approximation through parity analysis and the pre-established ApxLib. Subsequently, the framework introduces two novel mechanisms: dynamic probability adjustment and parity expansion. The first mechanism dynamically updates the mutation probability of gates in the genetic algorithm based on testability analysis, while the second mechanism performs recognition and reconstruction for binate gates to achieve dual optimization of efficiency and effectiveness in optimization. Experimental results indicate that the proposed optimization framework achieves an additional Soft Error Rate (SER) reduction of up to 10%～20% compared to traditional NSGA-II with the same hardware overhead, while reducing 18.7% of execution time reduction averagely.
Keywords:
Soft error tolerance
,
Triple modular redundancy
,
Approximate computing
,
Multi-objective optimization
© 2018 JOURNAL OF ELECTRONICS & INFORMATION TECHNOLOGY
京ICP备20021838号-8
Institute of Electronics, Chinese Academy of Sciences, P.O.Box 2702, Beijing
100190
Tel：010-58887066
Fax：021-64253812
Email：
[email protected]
Supported by:
Beijing Renhe Information Technology Co. Ltd
