Tutorial T3: Error Resilient Real-Time Embedded Systems: Computing, Communications and Control

摘要

Real-time embedded systems encompassing computing, communications and control are pushed towards always operating close to their limits of performance under virtually all operating conditions, for reasons of maximizing throughput per watt or just throughput, making the hardware susceptible to transient errors and noise due to reduced operating margins. In addition, the effects of process variations, electrical degradation and thermal hot-spots introduce additional concerns compounding the overall error resilience problem. In this tutorial, we present key resilience issues with the digital and mixed-signal/RF processing, control and sensor subsystems of real-time embedded systems. First, failure mechanisms and associated fault models are discussed for both permanent failures as well as transient errors particularly in the light of the adaptive nature of such systems (vis.a. vis. low power operation). Next, off-line testing and diagnosis of digital, analog and mixed-signal/RF circuits and subsystems is discussed, covering both catastrophic defects as well as parametric failure conditions. Advanced methods for postmanufacture tuning of digital and analog/RF devices to maximize manufacturing yield and field reliability are presented. Issues related to testing and tuning of adaptive systems with analysis of test cases are also analyzed. Finally, we present on-line testing methods for robustness of the complete embedded system to transient errors and induced noise. Applications to linear as well as nonlinear control of such real-time systems are discussed. We use a robotic system with a video sensor to illustrate the core concepts.