Synaptics has announced the Astra SL2600 Series, a new family of open-source Edge AI processors designed to power what it calls the “cognitive Internet of Things”. The launch introduces the Astra SL2610 product line, comprising five pin-to-pin compatible processor families – the SL2611, SL2613, SL2615, SL2617, and SL2619 – that address applications ranging from battery-powered consumer products to industrial vision systems.
Each processor is built on Synaptics’ new Torq Edge AI platform, combining its own neural acceleration technology with Google’s open-source Coral NPU based on the RISC-V architecture. This integration makes Synaptics the first silicon vendor to bring Google’s open NPU to market in production silicon.
Vikram Gupta, Senior Vice President and General Manager, Edge AI IoT Processors at Synaptics, said: “With the Astra SL2610 product line, Synaptics is redefining what’s possible for Edge AI. Through industry-leading power efficiency and breakthrough multimodal AI acceleration, these processors deliver the architectural foundation for customers to design scalable, next-generation IoT solutions.”
Open-source momentum reaches the Edge
The launch reflects a wider shift in the semiconductor industry as AI workloads migrate from the Cloud towards the Edge. While training large models remains the domain of hyperscale data centres, inference tasks – which account for around 80 to 90% of AI workloads – are increasingly being executed closer to devices. This move requires processors that balance performance, efficiency, and accessibility across highly fragmented IoT ecosystems.
Speaking during a technical briefing hosted by Publitek Connect, Nebu Philips, Senior Director, Technical Product Marketing at Synaptics, explained that this fragmentation is not only a hardware issue but also a software one. Proprietary compilers, tools, and frameworks can limit developers’ freedom to move between vendors or iterate rapidly.
“Today … [if] you buy Qualcomm silicon, you get Edge Impulse tooling. You buy from Renesas or STMicro, you get Imagimob tooling. It’s all nicely packaged with the silicon … but then you’re limited on choice,” said Phillips.
“What Google has done is completely opened it up, all the way from the compiler to the runtime … the whole build flow is the same across industry, and it’s all open … You don’t have to licence it anymore.”
A flexible, developer-first platform
At the hardware level, the Torq platform combines two AI engines: Synaptics’ transformer- and CNN-optimised accelerator, and Google’s Coral NPU, which Phillips described as a RISC-V-based “localised scalar computer”. The architecture allows new machine learning operators to be implemented easily without re-taping silicon, extending the lifetime of OEM investments.
The SL2610 processors integrate Arm Cortex-A55 and Cortex-M52 with Helium, along with a Mali GPU for graphics and multimedia. They also feature layered security built into the silicon, including an immutable root of trust, threat detection, and an application crypto coprocessor for secure AI inferencing.
Synaptics’ focus on open-source extends beyond hardware. Its adoption of Google’s IREE/MLIR compiler and runtime – part of the LLVM ecosystem – provides a unified development flow for deploying AI models across heterogeneous architectures. Phillips described the move as “a game changer for the industry, because it’s sort of the last sort of basket of vendor lock-in in the whole AI tech stack.”
Long-term collaboration with Google
Synaptics’ partnership with Google Research is a multi-generational engagement. The company will continue integrating future versions of the Coral NPU into its upcoming products, maintaining alignment with Google’s roadmap for open, flexible AI at the Edge.
Phillips said: “We are the lead partner. Over time, there’s going to be more, but we believe that we get a bit of time-to-market and venture advantage.”
Google Research Vice President and Head of Google Research, Yossi Matias, added:
“Google Research is excited to collaborate with Synaptics on Edge AI. By integrating our Coral NPU ML accelerator, we are helping create an ecosystem that will simplify development and unlock powerful new experiences. This partnership is a catalyst to move the industry forward and help make devices even more useful.”
Availability
The Astra SL2610 product line is sampling to customers now, with general availability planned for Q2 2026. Developers can register for early access through the Astra Machina development kit, which provides a hands-on environment for testing and deployment.
