基于数据差异性的软件抗辐射技术研究

Single Event Effects (SEEs) are a kind of transient fault phenomenon in semiconductor circuit, which is caused by the external radiation from the high energy neutrons from cosmic rays, and etc. SEEs are always one of vital reasons for affecting the reliability of space computers. With the continuously progress of integrated circuits, the performance of modern processors is improved significantly, but their dependability is increasingly affected by SEEs. Comparing with the hardware-implemented fault tolerance for SEEs, the software-implemented methods are attractive because of their advantage on costs and flexibility. However, traditional techniques of software fault tolerance have the weakness of error detecting capacity and error recovery. Firstly, this project plans to analyze error model for SEEs. Secondly, based on data diversity and instructions duplication, the error detecting method will be investigated for improving the capacity of detecting errors. Thirdly, this project will focus on the techniques about light weight error recovery for SEU, the optimization of check point and fine-grained recovering errors. At last, GCC compiler is alternated for implementing the proposed techniques. Based on software implemented radiation-hardening techniques, high reliability and high performance onboard computers can be constructed via COTS components. The research of this project will be in favor of extending the service life, reducing the cost, and improving the performance for future satellites.

空间环境下高能粒子辐射所产生的单粒子效应是半导体电路中的一种瞬态故障现象，是影响航天计算机可靠性的重要原因。随着集成电路制造工艺的持续进步，处理器性能在大幅度提高的同时，其可信性也正日益面临着单粒子效应的威胁。与硬件加固技术相比，针对单粒子效应的软件容错技术由于在实现成本和灵活性等方面的优势而备受关注。但传统软件容错技术在错误检测能力和错误恢复方面存在不足。本项目将首先研究面向单粒子效应的错误分析模型；在此基础上，研究基于数据差异性和指令复算的错误检测方法，以提高软件容错的检错能力；然后，研究轻量级错误恢复方法和检查点优化策略，以及细粒度错误恢复技术；最后，通过改造GCC编译器实现所提出的软件容错技术。基于软件实现的抗辐射技术，可以构建基于商用器件的高可靠、高性能的空间信息处理平台。本项目的研究可为延长卫星在轨飞行寿命、降低卫星研制成本、提高性能指标奠定技术基础。

随着集成电路制造工艺的持续进步，继性能和功耗问题之后，单粒子效应所导致的计算可信性已成为一个日益严峻的课题。已有软件抗辐射加固技术在错误检测和错误恢复方面存在时空开销大等不足，应用范围受限。本项目首先面向寄存器单粒子效应问题研究了程序可靠性分析模型，并基于该模型提出了一种程序可靠性优化方法；提出了一种基于指令复算和数据差异性的故障检测技术，以及对关键数据进行优先加固的可配置故障检测方法；分别针对数据流错误和控制流错误研究了细粒度的错误恢复技术，可显著降低错误恢复的时空开销；在此基础上，基于LLVM编译器具体实现了上述软件容错方法，并通过改造二进制插桩框架Pin实现了一个故障注入工具，可以对容错方法的有效性进行实际验证。这些研究成果有效提高了软件容错技术的应用能力，从而为构建高可靠、高性能的空间信息处理平台提供技术支持。