Synaptics Astra SR200 Google Coral NPU: Coralboard SL Ships, SR200 Chips Still Sampling

Synaptics introduced its Astra SR200 series on Oct. 20, a three-chip family of AI microcontrollers built around Google Research’s open-source Coral NPU technology. The chips span 200 to 400 GOPS of AI processing power, but they remain in the sampling stage, with no public pricing or firm production schedule.

What’s actually orderable today is a separate, smaller board. The companies’ existing Coralboard SL developer kit reached general availability through DigiKey the same day, running a 1-TOPS implementation of the Coral NPU. That gap between a shipping developer board and an announced-but-pre-production chip family is the detail embedded teams need to track before committing design time.

Google doesn’t manufacture the Synaptics silicon itself. It open-sourced the Coral NPU architecture under an Apache 2.0 license, allowing chip vendors to incorporate the design into their own processors without royalties. Synaptics’ Coralboard SL carries what the company describes as the first implementation of the Coral NPU in commercial hardware.

Synaptics Astra SR200 specifications, availability, and Coralboard SR status

Synaptics calls the SR200 series an “xMCU,” a category it says sits between conventional microcontrollers and full application processors, pairing AI-native compute with high-bandwidth memory built into the chip. The family is aimed at wearables, battery-powered sensors, connected cameras, and industrial vision and voice systems.

Three pin-compatible chips make up the lineup. The SR210 targets 200 GOPS for always-on voice and sensing tasks. The SR230 steps up to 300 GOPS and adds vision and display support. The SR250 tops out at 400 GOPS with dual vision and display paths for more demanding edge applications. All three pair Arm Cortex-M85 and Cortex-M52 cores with the Torq AI engine integrating Coral NPU technology.

Coralboard SL, the board developers can buy now, runs on Synaptics’ existing Astra SL2619 processor, with dual 2GHz Arm Cortex-A55 cores, a 1-TOPS Torq/Coral NPU, and 2GB of DDR4 memory. It moved from a limited-edition preview to full global availability through DigiKey on Oct. 20.

A second board, Coralboard SR, will carry the SR250 chip, but Synaptics hasn’t announced an order date or price. Teams budgeting around it should confirm timing directly with Synaptics before planning a design cycle.

Memory is where SR200 breaks from typical microcontroller design. The chips pack up to 8MB of integrated SRAM alongside custom DRAM that scales to hundreds of megabytes. Synaptics says the architecture uses substantially less power than typical system DRAM, although real-world battery performance has not yet been independently established.

On security, Synaptics lists hardware isolation, cryptographic acceleration, and secure boot among the SR200’s capabilities. The company describes PSA Level 3 and CC EAL4+ as certification targets, so buyers should verify certification status rather than assume those certifications are complete.

Why the Coral NPU open-toolchain strategy matters for product teams

Google’s argument for open-sourcing Coral NPU is portability, part of a broader industry push to move AI inference onto local hardware rather than the cloud, an approach also visible in Nvidia’s RTX Spark bringing local AI processing to Windows laptops.

The Coral architecture uses a standards-based RISC-V design paired with an open-source MLIR/IREE compiler. Google says this approach could allow developers to optimize a model once and deploy it across different chip vendors, devices, and power envelopes. That remains a goal rather than something demonstrated across a broad range of competing commercial silicon.

The Apache 2.0 license removes royalty obligations for semiconductor companies that want to build Coral into their processors. It doesn’t make Synaptics’ finished SR200 chips free or the entire commercial product open source: Synaptics layers its own Torq processing technology around the Coral core.

For industrial buyers, SR200 variants add 1Gb Ethernet with IEEE 1588 support and two CAN-FD interfaces intended for physical AI and robotics applications. Those capabilities give industrial teams a reason to track the family even while the chips remain in sampling.

On the model side, Synaptics says its Torq compiler supports Google’s Gemma 3 270M, LiquidAI’s LFM2.5, and Ultralytics’ YOLO26. Synaptics cites roughly 12 to 16 tokens per second for Gemma 3 270M depending on quantization, but independent testing on shipping SR200 hardware isn’t yet available.

The new processors also arrive as companies race to put increasingly capable AI into compact wearable hardware. TechRepublic has tracked that shift in products such as Meta’s camera-free smart glasses, where local processing could influence power use, responsiveness, and privacy.

The practical split is straightforward. Coralboard SL is shipping and usable now for engineers who want to prototype with the Torq/Coral toolchain. The SR200 chips and Coralboard SR are announced but not yet generally available, with no disclosed chip pricing or confirmed production date.

That distinction matters for evaluation planning. Coralboard SL runs on Linux-class hardware with 2GB of DDR4, while the SR200 family targets a different MCU-class memory and power envelope. A model that performs well on Coralboard SL isn’t necessarily a reliable proxy for how it will behave on SR200 silicon once that hardware ships.

Until Synaptics discloses chip pricing, a production date, and an order window for Coralboard SR, the SR200 remains primarily a technology and specifications announcement rather than a buying decision. Teams that want to get ahead of the toolchain can start with Coralboard SL, but performance and deployment assumptions will need to be reverified on SR200 hardware.

Also read: Apple’s planned smart glasses offer another look at how AI processing is moving into smaller wearable devices.