Can microcontrollers handle AI workloads while depending on application processors and external memory? A processor family offers AI acceleration and bespoke DRAM for on-chip processing.
Synaptics has launched its Astra SR200 processor family, part of a new xMCU range. This includes processors to support multi-modal artificial intelligence (AI) processing directly at the device. These are targeted at wearables, battery powered sensors, smart cameras, set top boxes, plus industrial vision and audio.
The SR200 series processors overcome the limitations of traditional MCUs by providing a powerful combination of multiple processing cores, AI engine and bespoke DRAM. This enables more complex AI models to be processed on device.
The family uses the company’s Torq AI engine with Google Research’s Coral NPU. It also includes an open source MLIR/IREE compiler to allow machine learning frameworks and models. Zephyr RTOS is also supported.
There are three models within the pin-compatible range offering differing processing capabilities. The SR210 provides up to 200 billion operations per second (GOPS) for voice processing and sensor applications. The SR230 provides 300 GOPS with vision and display plus optional audio. The SR250 provides 400 GOPS, vision and display plus dual vision/display pathways.
The processors provide up to 8Mb of SRAM, plus bespoke DRAM up to hundreds of megabytes. Synaptics claims the memory can provide up to a third less power for processing than typical system DRAM. The processors also include a high performance Arm Cortex-M85 core with speeds up to 800MHz. There is also a Cortex-M52 core for energy efficient processing. An optional Tensilica HiFi 5 DSP provides for audio processing.
Other features include hardware security, crypto acceleration, tamper detection and secure boot. Industrial versions also include 1Gb Ethernet, IEEE 1588 time sensitive networking and two CAN-FD interfaces.
The SR200 series is designed to provide developers with scalable processing capability whilst keeping a uniform hardware architecture and software development kit.
Click here for the official announcement.
