Some guidelines to enhance application-level robustness in linear computation

Application robustness i.e., the ability to provide a graceful degradation in performance when the algorithm solving the application is perturbed in its structural parameters, has an immediate effect on the design of a reliable circuit In fact, if the application is robust, we will experience a reduced impact of errors induced by faults affecting the application parameters. This situation is interesting for all those applications (e.g., embedded systems, image/signal processing on remote stations) for which an error-affected behaviour is acceptable as long as the error is kept within a priori defined bound. In this paper, we focus the attention on a linear computation by providing sensitivity tools to test the robustness degree of an application and off-fine techniques to improve it.