一种数据流驱动的计算通道容错方案

doi:10.16180/j.cnki.issn1007-7820.2020.07.002

摘要/Abstract

摘要：

单粒子翻转发生在数据流驱动的计算通道中时,会产生错误结果甚至计算通道无法正常工作。针对这一问题,文中提出了一种基于地址回卷技术的双模冗余容错方案。该方案采用双计算通道比较,检测结果是否发生错误;利用地址回卷技术还原地址信息,驱动原数据重新计算;对相关性数据流计算进行特定优化回卷。文中基于数据流驱动运算器对方案进行具体实现,实验结果显示,该方案实现了99%以上的可靠率,面积开销仅为三模冗余的76%,时间开销比双模冗余节约最多50%,实现了面积和时间消耗的平衡。

关键词: 单粒子翻转, 数据流驱动, 容错, 双模冗余, 计算通道, 地址回卷

Abstract:

When a single event upset occurred in the data flow-driven computation channel, it produces erroneous results or even the computational channel will not function properly. For this reason, this study proposed a dual-mode redundant fault-tolerant scheme based on address rewinding technology, which used double computing channel comparison to detect whether the error occurs. The address rewinding technique was used to restore the address information and drive the original data to be recalculated. A specific optimized rewind on the correlation data stream calculations was also performed in the study. Based on the data stream driven operator, the scheme was implemented. The experimental results showed that the scheme achieved a reliability rate of over 99%, the area cost was only 76% of the three-mode redundancy, and the time overhead was up to 50% compared with the dual-mode redundancy, indicating the scheme successfully achieved a balance between area and time consumption.

Key words: SEU, data stream drive, fault tolerance, DMR, computer channel, address rewind

中图分类号:

TN47

项阳,孙晓霞,宋宇鲲. 一种数据流驱动的计算通道容错方案[J]. 电子科技, 2020, 33(7): 6-11.

XIANG Yang,SUN Xiaoxia,SONG Yukun. A Fault-tolerant Scheme for Data Flow Driven Computing Channels[J]. Electronic Science and Technology, 2020, 33(7): 6-11.

图/表 13

图1

图2

图3

表1

图4

图5

图6

图7

图8

图9

图10

表2

表3

参考文献 16

[1]	Morgan K. SEU-induced persistent error propagation in FPGAs[J]. IEEE Transactions on Nuclear Science, 2005,52(6):2438-2445.
[2]	山蕊, 李涛, 蒋林. 动态可重构阵列处理器数据流处理单元的设计和实现[J]. 微电子学与计算机, 2017,34(1):106-109.
	Shan Rui, Li Tao, Jiang Lin. Design and implementation of data stream processing unit for dynamic reconfigurable array processor[J]. Microelectronics & Computer, 2017,34(1):106-109.
[3]	李浩洋, 宋宇鲲. 一种支持多种工作模式的可重构计算单元的设计[J]. 微电子学与计算机, 2017,34(10):53-58.
	Li Haoyang, Song Yukun. Design of a reconfigurable computing unit supporting multiple working modes[J]. Microelectronics & Computer, 2017,34(10):53-58.
[4]	谭龙生, 梁华国, 王存. 一种用于嵌入式软核的动态可重构容错方案[J]. 合肥工业大学学报, 2015,38(11):1052-1057.
	Tan Longsheng, Liang Huaguo, Wang Cun. A dynamic reconfigurable fault tolerant scheme for embedded soft cores[J]. Journal of Hefei University of Technology, 2015,38(11):1052-1057.
[5]	邱根, 毕东杰, 彭礼彪. 一种应用于SRAM型FPGA寄存器的容错设计[J]. 中国测试, 2018,44(9):75-79.
	Qiu Gen, Bi Dongjie, Peng Libiao. A fault tolerant design for SRAM type FPGA registers[J]. China Testing, 2018,44(9):75-79.
[6]	孙野, 陈晖照. TMR + Scrubbing SRAM加固型FPGA辐照实验研究[J]. 电子科技, 2013,26(10):139-141.
	Sun Ye, Chen Huizhao. Study on SEE experiment of TMR + Scrubbing SRAM FPGA[J]. Electronic Science and Tecgnology, 2013,26(10):139-141.
[7]	刘光辉. 使用时间冗余保证处理器的可靠性[J]. 计算机工程与应用, 2011,47(21):17-22.
	Liu Guanghui. Using time redundancy to ensure processor reliability[J]. Computer Engineering and Applications, 2011,47(21):17-22.
[8]	Abate F. New technique for improving the performance of lockstep architecture for SEEs mitigation in FPGA embedded processors[J]. IEEE Transactions on Nuclear Science, 2009,56(4):1992-2000.
[9]	Wang Jing, Yang Xing, Zhao Yuanfu, et al. Multi-bits error detection and fast recovery in RICS core[J].Journal of Semiconductors, 2016, 36(11): 115009(1-8).
[10]	Ichinomiya Y. Improving the robustness of a softcore processor against SEUs by using TMR and partial reconfiguration[C].Charlotte:IEEE Annual International Symposium on Field-Programmable Custom Computing Machines, 2010.
[11]	王晶, 申娇, 丁利华, 等. 基于周期粒度的级间寄存器备份机制[J]. 电子学报, 2018,46(10):2486-2494.
	Wang Jing, Shen Jiao, Ding Lihua, et al. Inter-stage register backup mechanism based on periodic granularity[J]. Chinese Journal of Electronics, 2018,46(10):2486-2494.
[12]	郝亚男, 高欣, 许仕龙. SRAM型FPGA的SEU容错技术研究[J]. 中国集成电路, 2015,10(5):31-36.
	Hao Yanan, Gao Xin, Xu Shilong. Study on the SEU tolerant techniques for SRAM-based FPGAs[J]. China Integrated Circuit, 2015,10(5):31-36.
[13]	薛芸男, 李振, 蒋承翔, 等. 面向航空环境的多时钟单粒子翻转故障注入方法[J]. 电子与信息学报, 2014,86(6):1504-1508.
	Xue Yunan, Li Zhen, Jiang Chengxiang, et al. Multi-clock single-event fault injection method for aviation environment[J]. Journal of Electronics & Information Technology, 2014,86(6):1504-1508.
[14]	Wang Yongqing, Ma Yuanxing, Liu Donglei, et al. An SEU tolerant approach for space-borne Viterbi decoders[J]. Chinese Journal of Electronics, 2014,24(3):857-861.
[15]	Esteve Hoyos S, Evans H D, Daly E. From satellite lon flux data to SEU rata estimation[J]. IEEE Transaction on Nuclear Science, 2004,5(51):2927-2935.
[16]	于航, 王晶, 周继芹, 等. 面向单粒子翻转效应的模拟故障注入技术[J]. 计算机工程与设计, 2016,37(1):107-111.
	Yu Hang, Wang Jing, Zhou Jiqin, et al. Simulation fault injection technique for single event flipping effect[J]. Computer Engineering and Design, 2016,37(1):107-111.

实部运算结果	虚部运算结果	编码	状态
一致	一致	00	正常
不一致	一致	01	回卷
一致	不一致	10	回卷
不一致	一致	01	回卷

容错方案	原运算时间 /周期数	冗余运算时间 /周期数	时间开销比/%
DMR	2 048	3 076~4 096	150.2~200.0
Roll-Back	2 048	2 052~2 062	100.2~100.7
TMR	2 048	2 048	100.0

方案	FF	LUT
原型	4 276	6 688
DMR	6 885	9 234
TMR	9 013	12 940
Roll-Back(本文方案)	7 156	9 677