METASAT Demonstrator: Mixed Criticality, Accelerated AI Computing for Future Space Systems
summary
A prototype SoC runs on an VCU118 FPGA and QEMU emulator combining a NOEL-V multicore with a custom SPARROW SIMD unit and a VORTEX GPU; three AI apps run in parallel and timeshare the accelerator and GPU. Cloud screening segmentation and ship detection produce pixel masks and bounding boxes, with cloud inference ~3× faster than scalar. FDIR uses mission data from the German Space Agency, predicts temperature curves and flags anomalies when the score exceeds 2.
AI-written summary · about
related talks
19mMission-Critical AI in Space and Sky: SWaP-Constrained Intelligence Wi...
56mInterplanetary compute: The Moon, Mars, and beyond
07mDemo: Neural Network Acceleration on Metis, Powered by RISC-V
30mAI On-board of Satellites, Towards Autonomous Scientific Instruments
49mMonitoring global events with AI-driven satellite swarms
10mDemo: Secure and Open Source RISC-V Based Flight Computer
21mMTIA: Meta's First Generation of AI Accelerators
57mBuilding GPU-accelerated multi-agent apps