Investigation of soft error mitigation techniques on resource-constrained IoT Edge systems.

软误差由许多影响引起，但这里主要关注的是由局部源、宇宙射线或被α发射器污染的包装材料的辐射打击引起的误差。不管原因是什么，软错误的最终结果都是一个（或多个）比特从预期状态被翻转。检测和纠正这些错误的传统方法包括三模冗余（或仅用于检测的双模冗余），或者在软件中，在同一处理器上多次执行相同的代码并比较结果，或者在硬件中，多个处理器执行一次代码并比较结果。TMR的成本很高，无论是硬件的面积、内存和功耗，还是软件的时间、内存和功耗。其他技术的成本也同样高。嵌入式系统通常在其资源方面受到高度限制，并且仍然受到限制性设计需求的约束。特别是实时系统，可能无法承受执行TMR所需的执行时间增加3倍，也无法承受RAM的类似增加。对于远程系统，或那些在难以接近的位置，类似的电力消耗增加可能需要更频繁的维修或更换，如果电池供电。

Soft errors result from many effects, but the primary concern here is for those caused by radiation strikes from local sources, cosmic rays, or packaging materials contaminated with alpha emitters. The ultimate result of a soft error, regardless of its cause, is one (or more) bits flipped from their expected state. Traditional methods of detecting and correcting these occurrences include Triple Modular Redundancy (or Dual for detection only) either in software, executing the same code multiple times on the same processor and comparing the results, or in hardware, where multiple processors executed the code once and compare the result. TMR is costly, either in area, memory and power for hardware, or time, memory and power for software. Other techniques have similarly high costs. Embedded systems are often highly constrained in their resources and still subject to restrictive design requirements. Real-time systems, in particular, may not be able to afford a 3x increase in execution time to perform TMR, nor a similar increase in RAM. For remote systems, or those in inaccessible locations, a similar increase in power consumption may necessitate more frequent servicing or replacement if battery operated.