Run-time adaptation method for mitigation of hardware faults and power budget variations in space-borne FPGA-based systems
Dimple Sharma, Lev Kirischian, Valeri Kirischian · 2017
This paper proposes a method for FPGA-based space-borne systems supporting multi-task applications to enable run-time combined adaptation to variations in system power budget and variations in available hardware resources due to occurrence of transient or permanent faults. The adaptation method assumes that each task in the workload has multiple implementations, which can operate at different clock frequencies, have different resource utilization and hence different power consumption. Also, while the critical tasks have strict performance specifications, the non-critical tasks can execute in a range of performance values. Depending on the existing power consumption and resource constraints, the method explores the design space of the task variants at run-time to find a suitable combination of implementation variants of all the tasks such that performance of the critical tasks is sustained, and performance of non-critical tasks is adjusted in order to satisfy a reduced power budget and/or create spare resources to enable relocation of affected task components in the case of hardware faults. The paper presents a detailed analysis of how the decision-making adaptation method selects a system configuration in the different scenarios of power budget depletion and occurrence of faults. It also analyzes the the worst-case overhead of the mechanism when implemented on the Arm Cortex A9 processor in the Zynq XC7Z020 SoC.