Your next smartwatch could run AI without ever talking to the cloud — and sip less power than a dim LED while doing it. That is the promise behind the edge AI chip Synaptics formally unveiled at its Tech Day in San Jose on October 20, a microcontroller co-developed with Google Research over nearly two years that can process voice, vision, motion and contextual sensing at the same time.
The chip sits in Synaptics’ Astra family of AI-native processors — the SL2610/SL2600 series — and combines Google’s open-source Coral neural processing unit “M2” with Synaptics’ own Torq NPU and Arm-based CPU cores. According to trade publication R&D World, Synaptics CEO Rahul Patel said the silicon will end up inside Google devices, though he declined to name which ones or when they will ship.
The headline numbers are striking. Google Research vice president Avinatan Hassidim said the Coral M2 design uses 40 percent less energy than rival processors, and that on-device inference runs at an average draw of under six milliwatts, Crypto Briefing reports. The chip also carries high-density DRAM support, hardware-rooted security and an open-source compiler stack the two companies built on MLIR over 18 months — “free from the walled-garden software plays you are used to with microcontrollers,” as Patel put it.
What the Headline Numbers Don’t Tell You
Here is the part the launch coverage mostly skipped: Hassidim never named the workloads or the comparison chips behind the 40-percent claim. Forty percent less energy than what, measured how? That missing baseline is the difference between a breakthrough and a marketing slide, and no independent test data has been published yet.
The competitive picture needs context too. Qualcomm’s Snapdragon platforms ship with Hexagon NPUs rated up to 45 TOPS and went into more than 800 million AI-capable chips in 2025, targeting phones and premium wearables that can absorb far more power. Synaptics isn’t trying to beat Qualcomm at its own game — it’s carving out the tier below, where devices run on tiny batteries and cloud round-trips are too slow, too power-hungry or too creepy for always-on sensing. Astra’s pitch: the first chip to handle genuinely multimodal AI workloads — voice, vision and motion simultaneously — down at that power level.
Google also confirmed the roadmap: the next Coral design, M3, is complete and adds native 32-bit floating-point support to run Gemma 3 270M — a 270-million-parameter language model — directly on-device. No shipping date yet.
Meanwhile in Gadgets:
If the wearable angle has you curious about what’s actually on shelves, read our rundown of Google’s $179.99 Fitbit Edge — shipping October 27, the most likely early candidate for exactly this kind of silicon. And for the privacy-first flip side of smart eyewear, see Meta’s camera-free Ray-Ban Audio glasses, which ditched the camera entirely to dodge the creep factor.
Why This Matters
Google co-designing silicon for devices it doesn’t even build itself tells you where the company thinks the next platform is: not another phone, but ambient AI that lives on your body, processing your context locally and privately. On-device inference means voice and vision data never has to leave your wrist — a meaningful selling point in a year when camera-equipped wearables are drawing regulatory heat.
But keep the excitement honest. Until independent testers verify the 40-percent claim against named workloads and named rivals, treat it as a direction, not a measurement. The real story to watch isn’t the Tech Day demo — it’s which Google device ships this first, and whether the battery-life and privacy promises survive contact with the real world.
