The new STM32Cube.AI extension pack for ST’s STM32CubeMX software tool generates optimized code to run neural networks on the STM32 MCUs

By Gina Roos, editor-in-chief

STMicroelectronics
expands the STM32CubeMX ecosystem
for the STM32 microcontrollers with the addition of advanced artificial intelligence
(AI) features. With the new STM32Cube.AI extension pack, product developers can now convert
pre-trained neural networks into C-code that calls functions in optimized libraries that can run on STM32 MCUs, said ST. The toolbox
includes the STM32Cube.AI mapping tool along with software examples running on
ST’s SensorTile reference board.

 

The
new STM32 neural-network developer toolbox brings AI to microcontroller-powered
edge, nodes, and embedded devices for a variety of applications including IoT,
smart building, industrial, and medical. The STM32Cube.AI delivers a new FP-AI-SENSING1 software function pack that provides examples of code for
human activity recognition and audio scene classification based on neural
networks.

These
code examples can be used with the ST SensorTile reference board to capture and label the sensor data
before the training process, enabling the board to run inferences of the
optimized neural network, and the ST BLE Sensor mobile app, which acts as the remote control and display.

The
STM32Cube.AI extension pack (part number: X-Cube-AI) can
be downloaded via ST’s STM32CubeMX MCU configuration and software
code-generation ecosystem. The tool currently supports Caffe, Keras (with
TensorFlow backend), Lasagne, ConvnetJS frameworks, and several IDEs, including
from Keil, IAR, and System Workbench.

Engineering services are
available for developers through qualified partners as part of the ST Partner Program and
the dedicated AI & Machine Learning (ML) STM32 online community. ST will demonstrate
applications developed using the STM32Cube.AI extension pack running on STM32
microcontrollers in a private suite at CES 2019.