RUSH

Space presents a challenging environment for computing. Extended development times and radiation tolerance requirements leave hardware performance a decade or more behind the terrestrial state-of-the-art at the time of deployment. Additionally, once deployed, hardware changes are impractical, encouraging a trend towards increased software programmability. At the same time, topside pressure from application advancements is forcing space-based platforms to improve throughput and latency while reducing power consumption. A popular approach to addressing the tension between these requirements is the heterogeneous processing architecture. By providing multiple hardware tools that optimally support a subset of the anticipated workload, a heterogeneous architecture can offer performance and power solutions to the application developer. However, programming these systems is extremely challenging due to variations in toolsets and data sharing interfaces. As a result, data sharing and dynamic workload scheduling across heterogeneous architectures are often suboptimal and hindered by poor scalability. In this research and development effort, we study the feasibility of unifying a heterogeneous processing platform a unique programming model This platform is called the Assimilation Dynamic Network (ADN). The ADN employs a mesh network and virtual tiles on FPGAs and scalable multicore processors to create a cleaner and innovative programming model.