Virtualization of accelerators in embedded systems for mixed-criticality: RPU exploitation for fusion diagnostics and control

Abstract

The need for computing systems featuring high performance, safety, security, and predictability requirements, has become essential for fusion diagnostics and control. In this scenario, the next-generation Multiprocessors System-On-Chip (MPSoCs) characterized by modern processors and hardware accelerators can expand the potential of control infrastructure. To enhance the usability, performance, and dependability of these new technologies and to reduce the deployment cost, the concept of virtualization sees its first application in embedded systems. This work aims to propose three high-level architectural solutions that leverage virtualization to exploit accelerators and Real-Time Processing Units (RPUs) present on MPSoCs to realize Mixed-Criticality Systems (MCSs). That is, systems capable to manage on a single chip real-time tasks for critical automatic control together with other value-added services on the same hardware. The problems to be solved to implement these proposals will be explained in detail to guide future works. To highlight the potential of virtualization for real-time embedded systems in fusion science scenarios we explore some use cases regarding Vertical Stabilization (VS) algorithms for plasma magnetic control that can benefit from the proposed solutions.