Synaptics has released a new microcontroller series integrating AI acceleration and implementing multiple domains to mitigate power. The Synaptics SR100’s main processing elements are an Arm Cortex-M55 CPU and an Ethos-U55 NPU. For low-power operation, the MCU has a Cortex-M4 core with a proprietary micro NPU. Targeting computer vision, the chip also has camera interfaces, an image preprocessor, and a JPEG encoder in addition to conventional MCU peripherals.
As IoT designers turn to neural networks for computer vision, they require new MCUs that can handle AI functions. Microprocessors with built-in AI acceleration are too big, costly, and power hungry. Older MCUs didn’t have the computational chops to handle a neural network. Companies such as Synaptics, therefore, are developing MCUs with built-in AI accelerators (NPUs). The new Synaptics SR-Series MCUs complement the company’s SL microprocessors, which employ more powerful Arm Cortex-A cores, NPUs, GPUs, and video hardware.
Arm has enabled these designs with its CPUs and NPUs. Released in 2020, the M55 and U55 deliver a notch less performance than the Cortex-M85 and Ethos-U85. Synaptics configures the M55 with Arm’s Helium extensions, SIMD instructions supporting INT8. The CPU, therefore, can handle some AI operations, such as smaller or less time-critical neural networks or activation functions unsupported by the U55.
Continually running the main CPU and NPU, however, would unnecessarily use energy in designs where these functions only operate occasionally to handle discrete events. The SR100, therefore, has the M4 and a 10 GOPS NPU to triage stimulus, deciding whether the bigger cores must awake. To further conserve energy, the low-power M4 domain can also sleep. An ultra-low-power sensing domain is always on to detect visual or audible changes in the environment or monitor other time series. This multidomain gearing approach has proven effective in other vendors’ designs to simultaneously minimize long-term energy use and offer high peak performance. Bringing this approach and AI acceleration to MCUs, Synaptics helps enable a new generation of small, low-cost, battery-operated industrial and consumer devices.