![[personal profile]](https://www.dreamwidth.org/img/silk/identity/user.png){kind=small}
Израильский видео блогер подробно расспрашивает нашего президента про ИИ на 10-ваттном чипе.
Summary
Background
- Krishna Rangasayee has 30+ years in semiconductors, software, and AI.
- Formerly at Xilinx (18 years); founded SiMa six years ago to focus on physical AI.
SiMa’s Mission
- Cloud AI dominated 2015–2025; 2025–2035 will be the decade of physical AI.
- Physical AI will embed in robotics, automotive, medical, aerospace, and defense.
- SiMa builds full-stack solutions: custom chips + AI software.
Key Innovations
- Multimodality: supports audio, video, text inputs/outputs.
- Broad model support: CNNs, transformers, LLMs, all on one chip.
- Ultra-low power: <10 watts, 10x more efficient than typical alternatives.
- Ease of use: models from Hugging Face can run with a button click.
Why Edge AI Matters
- Security & privacy: critical for defense, medical, aerospace.
- Latency: real-time decision-making (e.g., robotics, vehicles) needs on-device AI.
- Cost: cloud compute is expensive; local AI reduces TCO.
Defense Use Cases
- Drones, robotics, mobility platforms (land, sea, space), satellites.
- Satellites can process images on-board and transmit only “delta” data.
- Drones benefit from longer flight times via power-efficient AI.
- Defense requires ruggedization, reliability, and long lifecycle support.
Future of Edge AI in Defense
- AI-enabling existing equipment (not replacing, but upgrading).
- Machine-to-machine communication: seamless data sharing across forces.
- Faster innovation cycles: years-long defense adoption shrinking to 1–2 years.
- Personalized compute for soldiers — context-aware, memory-enabled AI assistants.
Challenges
- Bridging the fast pace of AI with slow defense upgrade cycles.
- Ensuring 12–16 year reliability in harsh conditions.
- Managing cybersecurity and ethical risks as devices proliferate.
Takeaway
SiMa is positioning itself at the core of the coming shift to physical AI: multimodal, power-efficient, secure, and future-proofed AI at the edge, with strong relevance for aerospace and defense.
