Google and Synaptics preview edge AI microcontroller running inference under 6 milliwatts
lmoroney · x · 2026-10-19
Google and Synaptics previewed a microcontroller for multimodal on-device AI, unveiled Oct 14 at Google Research's annual event ahead of an Oct 20 launch. It pairs Google's Coral NPU with Synaptics' Torq NPU and Arm CPUs, and will ship in Google devices.
Key points:
- Coral's M2 milestone adds the first Coral vector processor, claimed to use 40% less energy than competing processors and run on-device inference at an average draw under 6 milliwatts
- Google co-designs hardware and software, owning both TPUs and models like Gemini; a 2017 Google study showed 30–80x better performance per watt than CPU+GPU baselines
- Target: "ambient AI" that keeps personal data on-device, responds fast, and preserves battery — from wearables to industrial sensors
Related event: Google and Synaptics Preview Ultra-Low-Power On-Device AI Microcontroller(2 posts)→
More from Infra
- LTX-2.5 on one GPU: 130s → 43s per clip by fixing 384 silent fast-attention fallbacks — _BatmanRises_ · 2026-10-19
- NVIDIA GPUs Hold Value Far Above Depreciation: H100 at 61%, B200 at 164% — sudoraohacker · 2026-10-19
- Huawei's Hi-One NPO uses ILS with a redundant laser; scarce 300mW+ laser supply drives Chinese vendors' choice — zephyr_z9 · 2026-10-19
- Compute wins the AI race: Meta caught up and xAI scaled fast by building compute — haider1 · 2026-10-19
- Polymarket puts 15% odds on CoreWeave bankruptcy by end of 2027 — AccBalanced · 2026-10-19
- CoreWeave plunges 25% in a week as credit default swaps hit distress levels — MartinGTobias · 2026-10-19