Self-Adaptation for Availability in CPU-FPGA Systems Under Soft Errors

We introduce a model-based reliability estimation to preserve application availability in CPU-FPGA systems exposed to soft errors under varying environment conditions. The estimation is used as an in-system method to select a suitable configuration for changing radiation conditions. This allows systems to autonomously adapt their configuration in order to balance between reliability and performance. Such a self-adaptation goes beyond the state-of-the-art, where adaptation relies on preplanned reactive mode changes. By autonomously evaluating new configurations, our self-adaptation process is capable of increasing the availability by selecting the configuration with the desired application reliabilities for the current environment conditions.