#edge ai
Edge Ai: 49 AI articles covering edge ai news, analysis, and research
Articles
How 60 GHz Radar Improves Low-Power Presence Sensing in IoT Devicesβ10
60 GHz radar enables accurate, low-power presence sensing in IoT devices, detecting stationary people while preserving privacy and cutting energy use.
Hardware-Software Co-Design in the AI Era: A 2026 Perspectiveβ9
Hardware-software co-design is reshaping AI computing by 2026, enabling domain-specific architectures, chiplets, and AI-driven design automation for faster, eff...
Building Trust Into Physical AI Systemsβ9
Discover how physical AI systemsβrobots, autonomous vehicles, and industrial gearβbuild trust through safety, transparency, and robust verification for real-wor...
Humanoid Compute and Security: A More Complex Challenge Thanβ9
Humanoid robots demand far more complex compute and security than autonomous vehicles, requiring distributed processing, real-time safety, and robust defenses.
When Edge AI Lies: Fault Injection and False State in Liveβ10
Fault injection reveals false states in edge AI perception pipelines, exposing hidden failures that undermine real-time reliability and safety.
EBook β Accelerate Silicon Design for Physical AI (Part 1)β9
Discover how silicon design is evolving for Physical AI in 2026βreal-time processing, energy efficiency, and safety innovations.
Redefining Roles for Edge and Cloud AI in 2026β10
Explore the evolving roles of edge and cloud AI in 2026, from reduced latency and privacy to scalable cloud powerβdiscover their collaborative future.
Meet FreeToken: An Edge-Native MoE Serving Engine that Runs 753Bβ7
FreeToken enables running massive 753B-parameter MoE models on a single workstation GPU, democratizing high-scale AI inference for individual developers.
The Future of AI Compute: Why a Single Chip Architecture Won't Sufficeβ10
The future of AI compute demands heterogeneous architectures, not single-chip solutions, to meet growing power, cost, and performance needs.
Meet Needle 2: An Open 45M-Parameter Tool-Calling Model Thatβ7
Meet Needle 2: a 45M-parameter tool-calling model in a 14MB binary, running sessions in 28MB of RAM for edge devices.
