The Reliability of Software Algorithms and Software-Based Mitigation Techniques in Digital Signal Processors

Digital signal processors (DSP) are microprocessor- like processing elements that are specifically designed for signal and image processing algorithms. DSPs share many of the same failure mechanisms as microprocessors, including component crashes, program crashes and silent data corruption (SDC). The root of these problems is often single- event upsets (SEUs) in caches or control circuitry. In testing we have found DSPs are very sensitive to SDC, where computationally incorrect output data is produced but program execution is not affected. SDC can be problematic as it is hard to detect and could lead to the corruption of on-orbit data. We found that SDC could be reduced through the use of triple modular redundancy. In this paper we present data on how software on DSPs is susceptible to radiation-induced upsets and software techniques for reducing SDC.