Boost to AI at the edge with launch of NPU-accelerated 32-bit microcontrollers
The STM32N6 microcontroller series from STMicroelectronics with new machine-learning capabilities makes it possible to perform computer vision, audio processing, sound analysis and more in consumer and industrial applications at the edge.
STMicroelectronics has launched a series of 32-bit STM32 microcontrollers which integrate accelerated machine-learning (ML) capabilities alongside a high-performance 800 MHz Arm® Cortex®-M55 CPU core.
The new STM32N6 MCUs are the first to embed the Neural-ART Accelerator, a neural processing unit (NPU). With the on-board NPU, the STM32N6 provides 600 times more machine-learning performance than a current high-end STM32 MCU. This enables cost-sensitive and power-conscious consumer and industrial products to implement AI video and audio functions in small embedded systems.
The high-speed, low-power ML inferencing performed by the STM32N6 is due to the Neural-ART Accelerator NPU, which features nearly 300 configurable multiply-accumulate (MAC) units which can implement up to 600 giga-operations per second (GOPS). The NPU includes dedicated streaming engines which optimize data flow and minimize internal buffer usage and power. The accelerator supports on-the-fly weight decompression and real-time data encryption and decryption.
The high STM32N6 ML performance is complemented by the system control side of the MCU. The Cortex-M55 core achieves a high CoreMark® score of 3,360. The MCU also includes 4.2 Mbytes of RAM, providing sufficient memory to support data-intensive AI and multimedia workloads. Two 64-bit AXI interfaces provide the high intra-chip bandwidth required to unlock the full power of the Neural-ART Accelerator.
To support AI vision applications, the STM32N6 incorporates an image signal processor (ISP) which provides direct signal processing, enabling the use of simple and affordable image sensors. This ISP can be configured using the free ST ISP IQTune software, a cutting-edge tool which enables the developer to customize image signal processing parameters such as exposure time, contrast and color balance.
ST also provides development software to support the implementation of AI functions. The Edge AI Suite is a comprehensive set of software tools for the development of edge machine-learning applications. The software tools enable developers to implement AI models in various formats such as TensorFlow Lite, Keras, and ONNX.