CAREER: Design Optimization for Robustness to Single-Event Effects

CCF-0746850 PI名称：Kartik Mohanram标题：Career：针对单一事件效果的稳健性设计优化机构：William Marsh莱斯大学摘要：技术趋势和经济因素是完全基于商用硬件和操作系统的主流计算和通信系统广泛发展的幕后推动力。然而，作为一个社会，我们对这类系统的可靠性的重视程度已经随着我们对它们的依赖而增加，这与我们生活的顺利运行相称。单事件效应(SEE)引起的软错误是影响这类主流商品系统可靠性的一种重要且可能占主导地位的失效模式，本研究将开发基于优化的低成本SEE可靠性感知和SEE可靠性驱动设计解决方案，以最大化对SEE的稳健性，通常被称为SEE硬化。SEE硬化是一种有吸引力的低成本解决方案，可以提高可靠性，因为它不需要来自硬件或操作系统的任何运行时支持。SEE硬化还可用于补充和降低传统故障检测和容错技术的开销成本。这项工作产生的SEE硬化优化算法提供了SEE硬度和面积延迟功率之间的无缝折衷，从而实现了与目标应用程序生命周期内的关键程度和可靠性要求相称的经济高效的解决方案。这项研究的一个主要影响是通过将其覆盖范围扩大到缺乏财政资源来获取定制解决方案的域，从而实现无处不在的低成本、高度可靠的计算。通过学术和行业合作，该项目将开发一个综合试验台和基于网络的资源，以促进对可靠系统设计的广泛研究，这是一个正在迅速获得重要性和兴趣的领域。

CCF - 0746850 PI name: KArtik MohanramTitle: CAREER: Design Optimization for Robustness to Single-Event EffectsInstitution: William Marsh Rice UniversityABSTRACT:Technology trends and economic factors are driving forces behind the widespread move to mainstream computing and communication systems based entirely on commodity hardware and operating systems. Yet, the premium that we as a society place on the reliability of such systems has increased commensurate with our reliance on them for the smooth operation of our lives. Soft errors resulting from single-event effects (SEEs) are an important-and possibly dominant-failure mode that impact the reliability of such mainstream commodity systems.This research will develop low-cost SEE-reliability-aware and SEE-reliability-driven design solutions based on optimization to maximize robustness to SEEs, commonly termed SEE-hardening. SEE-hardening is an attractive low-cost solution to increase reliability since it does not require any runtime support from either the hardware or the operating system. SEE-hardening can also be used to complement and reduce the overhead cost of traditional fault detection and tolerance techniques. The optimization algorithms for SEE-hardening resulting from this work provide seamless tradeoffs between SEE-hardness and area-delay-power, enabling cost-effective solutions commensurate with the criticality and reliability requirements over the lifetime of the target application. A major impact of this research is to enable ubiquitous low-cost highly reliable computing, by expanding its reach to domains that lack the financial resources to acquire custom solutions. Through academic and industry collaborations, this project will develop an integrated testbed and web-based resources to facilitate broad research in reliable system design, an area that is rapidly gaining in importance and interest.