ST rolls out AI-microcontroller family

New Products | December 10, 2024

By Peter Clarke

ST has announced details of its STM32N6 family of 32bit ARM-based microcontrollers with a neural acceleration unit called Neural-Art on-chip.

The addition of machine learning capabilities to a microcontroller makes it possible to run computer vision, audio processing, sound analysis and more consumer and industrial applications at the edge rather than sending data to cloud or enterprise data center, the company said.

ST first disclosed the STM32N6 in May 2022, some 30 months ago (see ST to launch its first neural microcontroller with NPU) it is now rolling the product out in high volume. The family includes chis with and without hardware cryptographic support and in six different packages. It has been supplying specially-selected customers with the STM32N6 parts since October 2023.

ST has subsequently disclosed plans to further develop its NeuralArt NPUs.

ST adds to roadmap for AI-capable microcontrollers

The Neural-Art Accelerator, delivers 600x more machine-learning performance than a conventional high-end STM32MCU, ST claims.

The Neural-Art accelerator includes nearly 300 configurable multiply-accumulate (MAC) units that can provide up to 600 giga operations per second (GOPS).
The STM32N6 is has an 800MHz Cortex-M55 core, which provides a CoreMark score of 3360 and 4.2Mbytes of SRAM; a large memory to support AI and multimedia algorithms.
The STM32N6 incorporates an image signal processor (ISP) that provides direct signal processing, enabling the use of simple and affordable image sensors. This ISP can be configured using ST’s free ISP IQTune software, a tool that permits customizing image signal processing parameters such as exposure time, contrast or color balance.
The MCU is supported by ST’s Edge AI Suite of software tools for the development of edge machine-learning applications, including the possibility of hosting proprietary AI models in various formats such as TensorFlow Lite, Keras and ONNX.
ST hosts a “zoo” of AI models available to users.

“We are on the verge of a significant transformation at the tiny edge. This transformation involves the increasing augmentation or replacement of our customers’ workloads by AI models. Currently, these models are used for tasks such as segmentation, classification, and recognition. In the future, they will be applied to new applications yet to be developed,” said Remi El-Ouazzane, president of the microcontrollers group at ST, in a statement.

The packaging available varies from the VFBGA169 6 by 6 array up to the VFBGA264 14 by 14 array.