Modern Heterogeneous Systems-on-Chip (SoCs) that couple general-purpose and application-specific processors, a GPU, and reconfigurable fabric supporting dynamic partial reconfiguration, represent a promising class of platforms to implement modern cyber-physical systems.
When subject to real-time and dependability requirements, the process of dynamic partial reconfiguration must be predictable and must assure time guarantees. In other words, the maximum time needed for each dynamic partial reconfiguration performed on the platform shall be bounded to a known maximum value.
To reach such a goal, in this work, we review the challenges of dynamic partial reconfiguration predictability in real-time systems, and present concepts that allow to assure reconfiguration time predictability and bounding. A preliminary case study, involving the use of dynamic partial reconfiguration in an aerospace application, is considered to support the theoretical work.