Collaborative Research: CSR-EHS: Towards an Integrated Framework for Low Power Reliable Real-Time Embedded Systems

随着电池供电的便携式/嵌入式设备的普及，有效的能源管理变得极其重要，研究团体最近开发了各种能源管理技术。然而，考虑到随着技术规模的扩大和设计余量的减少，瞬态故障率不断增加，迫切需要将可靠性要求纳入这些框架。这对于安全关键型实时嵌入式系统尤其重要。最近的报告表明，广泛流行的能量管理技术——动态电压和频率缩放（DVFS）加剧了这一问题，由于在低电源电压下运行会增加暂态故障率，因此对系统可靠性产生了惊人的负面影响。本项目全面研究实时嵌入式系统中能量管理与暂态故障恢复之间的相互作用。首先，考虑到辐射、温度和电磁干扰的影响，正在开发dvfs系统的广义瞬态故障模型。其次，专注于操作系统级算法和工具，该项目正在为不同类型的嵌入式和实时系统开发一个集成的、可靠性感知的电源管理框架。该框架能够对节能和总体可靠性进行权衡分析。本文还研究了通过监测系统运行时的执行情况来动态提高系统性能的自适应方案。该项目预计将产生重大的技术影响，因为能效和可靠性是下一代实时和嵌入式系统的基本要求。

With the proliferation of the battery-powered portable/embedded devices, effective energy management became extremely important and the research community has recently developed various energy management techniques. However, there is an urgent need to incorporate reliability requirements to these frameworks, in view of the ever-increasing transient fault rates linked to scaled technology sizes and reduced design margins. This is particularly important for safety-critical real-time embedded systems. Recent reports indicating that the problem is exacerbated by the widely-popular energy management technique, dynamic voltage and frequency scaling (DVFS), which has an alarmingly negative effect on system reliability because of increased transient fault rates associated with operation at low supply voltages.This project undertakes a comprehensive study of the interplay between energy management and transient fault recovery in real-time embedded systems. First, a generalized transient fault model for DVFS-enabled systems is being developed that considers the effects of radiation, temperature and electromagnetic interference. Second, focusing on operating-system-level algorithms and tools, the project is developing an integrated, reliability-aware power management framework for different types of embedded and real-time systems. The framework enables a trade-off analysis for energy savings and aggregate reliability to be carried out. Adaptive schemes that dynamically improve the system performance by monitoring system's execution at run-time are also investigated. The project is expected to have a substantial technological impact as energy-efficiency and reliability are known to be fundamental requirements for next generation real-time and embedded systems.