ARMOR: Adaptive Reliability Management by On-the-Fly Redundancy in Multicore Embedded Processors

Multicore processors are expected to play a key role in the future of critical embedded systems such as automotive and avionics. This is primarily due to their advantages offered to the embedded systems such as increase in processing capability and reduction in power, size, and cost. However, reliability is one of the most compelling factors for critical applications. Improvement of the reliability may adversely affect the parameters of an embedded system such as power and resource utilization. This paper proposes an adaptive yet proactive method by which, in response to the requests from part of the application software, the reliability of the system is enhanced using physical redundancy. Having considered the resource and power limits in an embedded system, the method assigns some cores to perform the targeted critical tasks in a resilient form. It takes advantage of generated slacks by on-the-fly managing frequency of the cores without violating hard deadlines of the real-time system. Analytical results show that this method is up to 12 times more efficient than standby sparing in a multicore processor considering combination of energy consumption and resource utilization factors.