Google and Synaptics have previewed a microcontroller designed to run multimodal AI in devices constrained by power, memory and cost. The chip combines Google’s Coral M2 processor with Synaptics’ Torq NPU and Arm CPUs, targeting products from wearables to industrial sensors.
Watch Desk analysis
What happened
Synaptics CEO Rahul Patel outlined the chip at Google Research’s Research@ event on 14 October. Synaptics plans to unveil it formally on 20 October, according to reporting by RD World Online.
The companies say the chip can process voice, vision, motion and contextual sensing at the same time. It supports high-density DRAM, GPU and multimedia processing, hardware-rooted security, and an open-source compiler built with Google Research on MLIR. Developers can use standard machine-learning frameworks and model formats, and can already prototype on Coralboard, a development platform announced earlier this year.
Google Research’s Avinatan Hassidim says the M2 processor uses 40% less energy than other processors on the market and can run on-device inference at an average draw below six milliwatts. The comparison workloads and processors were not specified. Patel said the chip will appear in Google devices, but neither which devices nor when they will ship has been disclosed.
Why it matters
Running AI on a device can mean quicker responses and less need to send personal information to the cloud. The pitch here is not just another compact processor: Google and Synaptics are pairing hardware with an open compiler and support for familiar tools, aiming to make development less dependent on a single proprietary software stack.
That could matter to developers building battery-powered products, where an AI feature is only useful if it fits within real limits on power and memory. The energy figures are company claims without enough detail to compare them independently, and a preview is not a shipping product. Still, this is a concrete attempt to make local multimodal AI practical beyond phones and laptops.
Our read
The most interesting part is the combination of low-power hardware and an open development path. That is a stronger proposition than a chip name and a promise of future cleverness. Developers should watch for the formal launch details, especially supported models, measured performance and actual device availability. For now, the sub-six-milliwatt figure is intriguing, but the missing comparison details deserve a raised eyebrow.
What to watch
- What Synaptics discloses at its 20 October launch about specifications, availability and performance.
- Which Google devices use the chip, and when they are expected to ship.
- Whether developers can run useful models within the stated power and memory limits.
- How Google’s upcoming Coral M3 design, which Hassidim says is complete, supports Gemma 3 270M on-device. Neither company has given a shipping date for M3-based silicon.
Discussion spark: For small devices running AI locally, what should matter most: battery life, privacy, or an open development ecosystem?
Sources and evidence
- Source update (16 October 2026, 22:18 UTC)
Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.