A RISC-V Multicore and GPU SoC Platform with a Qualifiable Software Stack for Safety Critical Systems

Abstract

This paper introduces the METASAT platform, a novel RISC-V multicore and GPU System-on-Chip designed as a prototype for future safety-critical space missions. The architecture integrates a space-grade NOEL-V multiprocessor with the SPARROW AI accelerator and the Vortex GPU to handle performance-intensive AI workloads. Crucially, the platform includes a qualifiable software stack utilizing baremetal, RTEMS, and the XtratuM hypervisor, enabling the adoption of high-performance hardware in institutional space environments.

Report

Key Highlights

• Presents the METASAT hardware platform, a prototype developed under a Horizon Europe project specifically targeting future safety-critical space systems.
• Introduces a novel architecture for the space domain by integrating high-performance massive parallel processing units (GPUs and AI accelerators).
• The system is built entirely upon the RISC-V specification, encompassing both the multicore CPU and the parallel accelerators.
• A primary innovation is the deployment of a comprehensive, qualifiable software stack necessary for deployment in institutional space missions, moving beyond the current limitations of 'New Space' applications.

Technical Details

• Core Components: The SoC integrates the NOEL-V multiprocessor (an established space-grade processor), the SPARROW AI accelerator, and the Vortex GPU.
• Instruction Set: All primary processing systems within the platform adhere to the RISC-V specification.
• Workloads: The platform is designed to handle increased performance requirements for future space-related workloads, specifically citing AI.
• Software Abstraction: The qualifiable software stack includes three layers/options for varying criticality:
  • Baremetal
  • RTEMS (Real-Time Operating System)
  • XtratuM hypervisor

Implications

• Enabling High Performance in Space: This work addresses the need for increased computational performance, particularly for AI workloads, which are crucial for future space missions, by safely incorporating parallel architectures like GPUs.
• RISC-V Ecosystem Maturity: The project validates and strengthens the position of RISC-V as a robust, capable instruction set architecture for highly regulated, safety-critical systems, spanning from traditional CPUs (NOEL-V) to novel accelerators (SPARROW and Vortex).
• Qualification Barrier Reduction: By providing a pre-defined, qualifiable software stack, the platform significantly lowers the barrier for adopting advanced, high-performance hardware in stringent institutional missions where software certification is mandatory.