A Google Co-Designed Edge AI Chip Wants Your Wrist

Your next smartwatch might run AI without ever phoning the cloud. Synaptics is formally unveiling a new Edge AI Chip at its Tech Day in San Jose on October 20, 2026 — a microcontroller co-developed with Google Research over nearly two years that can process voice, vision, motion and contextual sensing at the same time, according to trade publication R&D World.

The part sits in the Astra family of AI-native processors from Synaptics and combines Google’s Coral neural accelerator with the company’s own Torq NPU and Arm-based CPUs. Synaptics chief executive Rahul Patel outlined the design at Google’s annual Research@ gathering in Mountain View earlier this month, saying the silicon will end up inside Google devices, though he declined to name which ones or when they will ship.

The efficiency numbers are the headline. The Coral M2 design at the chip’s core uses 40 percent less energy than rival processors on the market, Avinatan Hassidim, a vice president at the search giant’s research arm, said at the Mountain View event. He added that on-device inference runs at an average draw of under six milliwatts — roughly the power of a dim indicator LED — while conceding that he did not name the workloads or the comparison chips behind the figure.

For Hassidim, the point is what he calls ambient AI: software that acts on a user’s context while keeping personal information on the device, responding quickly and preserving battery life. He argued that current edge chips burn too much energy and answer too slowly for those experiences, and that a fragmented market of proprietary designs makes building for them harder than it should be. “With Coral, we want to address this challenge,” he said.

What It Means for Your Wrist

The first products built around this Edge AI Chip are most likely to be the ones you wear. Target hardware ranges from battery-powered wearables to industrial vision systems and cameras, with smartwatches and fitness trackers the most visible to everyday buyers. On-device sensing means voice commands and health tracking could keep working in airplane mode, and your biometric data would not need to leave the watch to be useful. The part also carries hardware-rooted security plus graphics and multimedia processing, and Patel flagged support for high-density DRAM as the lever that lets the design scale across device classes.

That combination matters because the privacy story flips. Today, asking your watch a question usually means sending your voice to a server farm and waiting for an answer. A microcontroller that hears, sees and reasons locally could shrink that round trip to zero — faster responses, longer battery life, and nothing uploaded. It is the same pitch behind Google’s push to put more intelligence directly into its products, and it lands as AI features keep demanding more of our workdays, as the AI productivity paradox shows.

An Open Stack, Not a Walled Garden

Developers get unusual freedom with this one. Synaptics and the search giant’s research lab spent 18 months building an open-source compiler on MLIR, the Multi-Level Intermediate Representation framework, and the chip works with standard machine learning frameworks and model formats. “You are free from the walled-garden software plays you are used to with microcontrollers,” Patel said. Builders can already prototype on Coralboard, a development platform released earlier this year that pairs the accelerator with Astra processors.

The roadmap is already public. After the San Jose unveiling — a noon press and analyst briefing plus a demonstration showcase — the next milestone is the M3 design, which Hassidim said is complete. It adds native 32-bit floating-point support so compact models such as Gemma 3 270M can run entirely on device. Synaptics has committed to building chips on future Coral designs, but neither company has named a shipping date for that silicon.

Google is no stranger to silicon partnerships. In 2023 it teamed with Qualcomm on a RISC-V platform for Wear OS, including the supporting software ecosystem. The Synaptics collaboration goes further by pairing Google’s accelerator design with an open compiler stack aimed squarely at microcontrollers. A much older Google study once claimed 30 to 80 times higher performance per watt for its first tensor processors versus the Intel and Nvidia hardware of the day — a reminder that the company has been chasing efficient AI hardware for the better part of a decade.

Every Edge AI Chip announcement comes with caveats, and this one is no exception. Patel would not say which of the company’s devices get the part or when, and Hassidim left his efficiency comparisons unanchored to specific workloads. Still, if the silicon delivers, the next generation of wearables could listen, watch and sense without a cloud round trip — a quiet shift in where AI actually lives, and one more reason the gadget race of the next few years will be fought at the edge. For the broader device landscape, see Oppo’s latest global phone launch.

Source: R&D World, “Google teases Synaptics edge AI chip ahead of Oct. 20 launch”