Reducing the Cost of Protection against Soft Errors using Profile-Based Analysis

The negative impact of the aggressive scaling of technology nodes on the sensitivity of CMOS devices to soft errors has been well studied in the past. Technology scaling makes processors more susceptible to transient faults. Errors caused by high-energy particle strikes in proces-sors can result in unexpected behavior and incorrect results. With the smaller and cheaper CMOS devices pervasive in mainstream comput-ing, it is necessary to protect these devices against soft errors; an in-creasing rate of faults makes the need to protect the critical application running on commodity processors against soft errors more important than ever. The existing methods of protecting against such faults gen-erally have high area or performance overheads and so are used only in mission-critical domains. In order to protect against the soft errors caused by high energy particle strikes, the detection of these errors is necessary so that recovery can be triggered. In this work, a profile-based software-only application analysis and transformation solution to detect soft-errors is presented. The goal is to develop a low cost solution which can be deployed for off-the-shelf commodity processors. The solution works by intelligently duplicating instructions which are likely to affect the program output, and com-paring results between original and duplicated instructions. The intel-ligence of our solution is provided by the use of control flow, memory dependence, and value profiling to understand and exploit the common-case behavior of applications. Our solution is able to achieve 92 % fault coverage with a 20 % instruction overhead. This represents a 41 % lower performance overhead than the best prior approaches with approxi-mately the same fault coverage