Mission-Critical AI in Space and Sky: SWaP-Constrained Intelligence Wi...
summary
Unified RISC-V–controlled Agilex FPGA SoCs run deployed CV and LM models (YOLO, MobileNet, UNET) on an AI-native CubeSat FLAP-0 and a UAV ISR node SKY-0, with BitDLA doing FPGA-accelerated inference and NIOS V softcore for orchestration. A shared layer FSHMEM-MOA² enables hybrid CPU–FPGA memory sharing; measured results claim up to 10× node-to-node comms acceleration and ~2× LLM compute speedup, a sustained-throughput metric (SIT) peaking near 339 SIT ops, and planned dev kits in 2028.
AI-written summary · about
chapters (11)
- 0:00Introduction to mission-critical AI
- 1:49Platforms for space and UAVs
- 3:53Flap Zero and Sky Zero overview
- 4:15NEOS 5 soft core processor
- 6:02Visual Designer Studio demo
- 8:09Hardware and software workflow
- 11:03Hybrid architecture orchestration
- 11:47Rick’s Bench for AI efficiency
- 15:23Open-source benchmark goals
- 16:03Future of AI compute orchestration
- 17:05Q&A session
related talks
06mMETASAT Demonstrator: Mixed Criticality, Accelerated AI Computing for Future Space Systems
33mAI and SDR: Software Meets Hardware Again...
13mKeynote: Taking RISC-V Intelligence Everywhere
30mAI On-board of Satellites, Towards Autonomous Scientific Instruments
53mBuilding an AI Accelerator with High-Level Synthesis
19mAccelerating Edge AI with the Help of RISC-V
30mRISC-V: Enabling Open Physical AI
49mMonitoring global events with AI-driven satellite swarms