Google and Synaptics have introduced an open-source, low-power edge AI microcontroller built for devices that never get to plug into a wall for long.
The chip belongs to Synaptics’ Astra SL2610/SL2600 series. It targets wearables, smart glasses and industrial sensors.
What’s inside the chip
The microcontroller combines several pieces of hardware into one platform. Google contributes its open-source Coral NPU “M2,” which includes a vector processor. Synaptics adds its own Torq NPU, Arm CPU cores and multimedia processing capabilities.
The design is built for simultaneous multimodal processing, handling voice, vision, motion and contextual sensing tasks at the same time.
The headline power figure is the most striking one. Average on-device inference draws less than 6 milliwatts, according to the research findings on the chip.
Performance claims also position the Coral NPU as a leader on efficiency. It reportedly delivers approximately 40% lower energy use compared to prevailing edge AI processors.
Timeline and the open-source angle
The partnership itself was announced in January 2025, combining Google’s Coral NPU with the Astra SL2610 series.
Developer platforms, including the Coralboard powered by Astra SL2610, were showcased at Google I/O 2026. The chip was then previewed at Google Research’s Research@ event on October 14, 2026. A formal demonstration is scheduled for Synaptics’ Tech Day on October 20, 2026.
The companies emphasize an open-source software stack developed over an 18-month period. That stack uses MLIR/IREE-based tools, which aim to standardize machine learning frameworks and reduce fragmentation in edge computing for IoT applications.
The Coral NPU technology is also being targeted for Google products.
Why Google keeps coming back to the edge
This isn’t Google’s first move into specialized AI hardware. The project builds on the company’s earlier work with Coral and with TPUs, its tensor processing units. Folding Coral NPU technology into Synaptics’ platforms extends that strategy: Google supplies the open architecture, and a chipmaker with an established device business turns it into shipping silicon.
What this means for device makers and the edge AI market
The most immediate stakeholders are hardware makers building wearables, smart glasses and industrial IoT systems. A sub-6-milliwatt inference draw could loosen battery life constraints considerably for always-on features.
The October 20 demonstration at Synaptics’ Tech Day will offer the next real look at how the platform performs.
Disclosure: This article was edited by Diego Almada Lopez. For more information on how we create and review content, see our Editorial Policy.