Identification of critical variables using an FPGA-based fault injection framework

Abstract

The shrinking nanometer technologies of modern microprocessors and the aggressive supply voltage down-scaling drastically increase the risk of soft errors. In order to cope with this risk efficiently, selective hardware and software protection schemes are applied. In this paper, we propose an FPGA-based fault injection framework which is able to identify the most critical registers of an entire microprocessor. Further-more, our framework identifies critical variables in the source code of an arbitrary application running in its native environment. We verify the feasibility and relevance of our approach by implementing a lightweight and efficient error correction mechanism protecting only the most critical parts of the system. Experimental results with state estimation applications demonstrate a significantly reduced number of critical calculation errors caused by faults injected into the processor.