FPGA软错误防护方法研究

发布时间：2018-06-23 02:14

本文选题：单粒子翻转 + 软错误　；参考：《西安电子科技大学》2015年硕士论文

【摘要】：基于SRAM结构的FPGA(SRAM-based FPGA),自从问世以来一直不断向大规模,高密度,低功耗方面发展,已达到Virtex-7系列的千万门规模,凭借这些优势,SRAM型FPGA在星载设备方面备受人们关注;但其FPGA的逻辑门电路都是动态加载到SRAM中,其中配置位占FPGA存储单元的比重在99%以上;在辐射环境下,相比于ASIC、基于反熔丝结构的FPGA等,SRAM型FPGA更容易受到单粒子效应,尤其是单粒子翻转效应(single-event upset,SEU)的影响。如何提高设备的抗单粒子翻转可靠性已经成为SRAM-based FPGA系统设计时必须重点考虑的问题。本文通过对常见防护方法的防护效果、代价进行建模、理论分析、数据分析,进而期望为工程实践中防护方法的选择提供参考。主要工作分为如下三个方面:1、将空间高能粒子对器件辐射抽象为泊松过程,在此基础上对常用的抗单粒子翻转防护方法:三模冗余防护方法、周期刷新防护方法以及汉明码防护方法,在理论上推导分析,并对防护前、后抗单粒子翻转可靠性进行定性对比分析。2、针对整个FPGA工程采用某种单一的防护方法进行防护,代价过大,于是对其按照一定规则划分为可以独立综合映射模块,分析各个模块的抗单粒子翻转可靠性,对易出错模块进行防护,以达到用最小的代价换取性能最大限度提升。所以,对FPGA工程按照一定规则划分为模块,针对于模块实现以下5种防护方法:基于模块的三模冗余防护方法;以牺牲速度为代价提高可靠性的基于模块的时间冗余防护方法;等时间间隔对FPGA重新擦写的周期刷新防护方法;将周期刷新和三模冗余结合在一起提升可靠性的基于模块刷新三模冗余防护方法;针对于BRAM进行防护的汉明码防护方法。3、对这些防护方法进行性能方面定量分析,结合FPGA的特性,从面积、速度、功耗三方面分析防护带来的代价。分析XC4VSX55型FPGA实验数据,与理论模型预估结果基本一致。可以为工程应用中防护方法的选择提供参考。
[Abstract]:SRAM-based FPGA (SRAM-based FPGA) has been developing to large scale, high density and low power consumption since its inception. It has reached the scale of tens of millions of Virtex-7 series. With these advantages, SRAM FPGA has attracted much attention in space-borne devices. But the logic gates of FPGA are loaded into SRAM dynamically, in which the proportion of configuration bit is more than 99% of the memory cell of FPGA. In radiation environment, compared with ASIC-based FPGA, FPGA based on anti-fuse structure is more vulnerable to single particle effect. In particular, the effect of single particle reversal (single-event upset SEU). How to improve the reliability of SRAM-based FPGA has become an important problem in the design of SRAM-based FPGA system. Through modeling, theoretical analysis and data analysis of the protection effect and cost of common protection methods, this paper hopes to provide a reference for the selection of protection methods in engineering practice. The main work is divided into the following three aspects: 1, abstracting the radiation of space high-energy particles to the Poisson process. Based on this, the common anti-single-particle inversion protection method: the three-mode redundancy protection method, is proposed. The periodic refresh protection method and the hamming code protection method are deduced and analyzed in theory, and the reliability of anti-single particle inversion before and after protection is qualitatively compared and analyzed .2. a single protection method is adopted to protect the whole FPGA project. The cost is too high, so it can be divided into independent integrated mapping modules according to certain rules. The reliability of each module can be analyzed, and the error-prone module can be protected in order to achieve the minimum cost in exchange for maximum performance improvement. Therefore, the FPGA project is divided into modules according to certain rules, and the following five protection methods are implemented for the modules: three mode redundancy protection methods based on modules, time redundancy protection methods based on modules at the expense of sacrificing speed, and time redundancy protection methods based on modules at the expense of sacrificing speed. A periodic refresh protection method for rewriting FPGA at equal intervals; a modular refreshed three-mode redundancy protection method which combines periodic refresh and trimode redundancy to enhance reliability; Aiming at the hamming code protection method. 3, the performance of these protection methods is analyzed quantitatively, and the cost of protection is analyzed from three aspects of area, speed and power consumption combined with the characteristics of FPGA. The experimental data of XC4VSX55 FPGA are in good agreement with the theoretical model. It can provide reference for selection of protection methods in engineering application.
【学位授予单位】：西安电子科技大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：TN791

【相似文献】