Resilience and Real-Time Constrained Energy Optimization in Embedded Processor Systems

Abstract—Low-power embedded processing relies on dy-namic voltage-frequency scaling (DVFS) in order to optimize energy usage and therefore battery life. DVFS allows the processor to continuously adapt voltage and frequency to the minimum that still meets a program’s current performance requirements. However, low-voltage operation exacerbates the incidence of soft errors. Similarly, high-voltage operation (to meet real-time deadlines) is constrained by power dissipation (and associated thermal) maxima- as dictated by aging limits. In this paper, we introduce a novel modeling framework called PEARL to examine a class of embedded system applications relevant to mobile, airborne vehicles. Using PEARL, we in-vestigate the problem of assigning optimal voltage-frequency settings to individual segments within example workflows. The goal of this study is to understand the limits of achievable energy efficiency (performance per watt) under varying levels of system-resilience targets. The analysis results show that: (a) the resilience constraints limit achievable efficiency; and (b) higher variability in power dissipation across workflow seg-ments provides more opportunities to boost efficiency, despite stringent resilience constraints. Keywords-low power, embedded systems, soft error resilience, real-time constraints I