Synaptics has introduced the Astra SR200 Series, which it calls an xMCU: a microcontroller with enough compute and memory to run multimodal edge AI on-device. Target applications include AI wearables, battery-powered sensors, connected cameras, and industrial vision and voice systems.
The pitch is that these workloads outgrow MCU memory but can’t afford the power and cost of an applications processor. Synaptics says the SR200 handles jobs that traditionally required Linux and discrete DRAM.
Memory is the main differentiator. Up to 8MB of SRAM pairs with custom high-bandwidth DRAM that scales to hundreds of megabytes. Synaptics claims it runs at one-third the power of typical system DRAM and removes the need for external memory in most designs.
Compute pairs an Arm Cortex-M85, running up to 800 MHz, with a Cortex-M52 for always-on sensor fusion, connectivity, and low-power inference. Arm says this’s the first time it has combined the two. An optional Cadence Tensilica HiFi 5 DSP handles audio and acoustic ML. The Torq AI engine integrates Google Research’s Coral NPU and uses an open-source MLIR/IREE compiler instead of a proprietary toolchain.
Three pin-compatible parts share one SDK: the SR210 at 200 GOPS, the SR230 at 300 GOPS with vision and display, and the SR250 at 400 GOPS with dual vision and display paths.
Security targets PSA Level 3 and CC EAL4+. Industrial variants add 1Gb Ethernet with IEEE 1588 TSN and two CAN-FD interfaces. The parts support Zephyr RTOS. Sonos, Verisure, and Jorjin Technologies are among the companies quoted as evaluating or using the Astra platform.