STMicroelectronics has released the powerful STM32N6 series of MCU with an integrated Neural Processing Unit (NPU) for energy-efficient edge AI applications. An NPU is a specialized hardware component designed to accelerate tasks related to machine learning and artificial intelligence. This ARM-powered MCU has Arm Helium vector processing, which will offload the CPU while performing Digital Signal Processing (DSP) tasks.
The STM32N6’s Neural-ART Accelerator features 300 configurable MAC units and dual 64-bit AXI memory buses, achieving 600 GOPS(Giga Operations Per Second). Optimized for power (3 TOPS/W), it supports TensorFlow Lite, Keras, and ONNX, offering a wide variety of compatibility with these tools and frameworks commonly used in ML and AI.
With 4.2 MB RAM, it supports advanced GUIs and can store double frame buffers for 1280×800 displays. It features Octo/Hexa SPI flash interfaces for efficient asset management, a NeoChrom GPU, and H.264/JPEG encoders. These enable rich UIs, video streaming, and combined GUI and neural network processing on one MCU, offloading tasks via its NPU and multimedia hardware.
These MCUs are available in six different packages ranging from 169 to 264 pins and from 0.4 mm to 0.8 mm pitch.
The STM32N6x7 line of MCUs supports AI acceleration while the STM32N6x5 line does not.
On the communication side, the STM32N6 features a Gigabit Ethernet module, six SPI and two I3C interfaces, two 12-bit ADCs, four 32-bit advanced timers, and more.
The new series of MCUs includes an Image Signal Processor (ISP) for enhanced machine vision applications, leveraging its NPU for high performance. The ISP processes raw image data from cameras, handling tasks like noise reduction and color correction. It also supports MIPI CSI-2, the widely used camera interface in mobile devices, eliminating the need for an external ISP for compatible cameras.
The Edge AI Developer Cloud which is a part of ST Edge AI Suite provides pre-trained models, a board farm, and STM32-specific neural networks. ST Edge AI Core optimizes models from popular frameworks for the accelerator. STMicroelectronics is also optimistic about its collaborations with partners like NVIDIA TAO Toolkit, AWS STM32 ML at the Edge Accelerator, and Hugging Face to ensure seamless AI development.
Developers can explore the STM32N6570-DK Discovery kit featuring the STM32N657X0H3Q microcontroller with ST Neural-ART Accelerator.
Leave a Reply