AI tool suite enables neural network optimization and deployment on STM32 devices

The ST Edge AI Suite software from STMicroelectronics facilitates the deployment of AI models on STM32 microcontrollers and microprocessors, Stellar automotive microcontrollers, and MEMS smart sensors. 

The suite is compatible with common AI frameworks, including Amazon Web Services, Keras, LARQ, MATLAB, Microsoft Azure, Nvidia TAO Toolkit, ONNX, Pytorch, and TensorFlow Lite. The suite, which is compatible with all STM32 microcontrollers, provides the software tools required to take advantage of the AI acceleration capabilities in the new STM32N6 series of MCUs, the first STM32 MCUs to include the ST Neural-ART Accelerator, a neural processing unit (NPU). 

Functions supported by the ST Edge AI Suite include: 

• Data logging: capturing the sensor data necessary for AI model training
• Auto ML: automatically generating optimized machine learning algorithms
• Model optimization: optimizing AI models and generating associated code for target devices
• Validation and testing: ensuring model performance meets deployment criteria
• Online benchmarking: testing model performance on ST hardware via the cloud

Embedded developers can also benefit from the ST model zoo, a set of pre-trained models which accelerate the deployment of common types of AI functions on supported devices.

Comprehensive documentation guides the developer through the deployment process.

Tools provided in the ST Edge AI Suite are optimized for use in specific types of device. The ST Edge AI Core CLI version and the ST Edge AI Developer Cloud allow users to optimize and evaluate AI model performance on any ST hardware. 

For STM32 MCUs, the STM32Cube.AI (X-CUBE-AI) tool performs neural network optimization. NanoEdge AI Studio is an AutoML tool for algorithm generation.

Different tools apply to STM32 MPUs: developers can use AI for OpenSTLinux (X-LINUX-AI) and the STM32MP2 offline compiler for Linux^® operating system-based AI frameworks.

For MEMS sensors with a machine learning core, the online ST AIoT Craft tool and the MEMS Studio desktop tool can be used for data analysis, algorithm design, and model optimization. The MLC model zoo provides pre-optimized models. Other MEMS sensors feature an intelligent sensor processing unit (ISPU): the MEMS Studio enables data analysis and model optimization for these devices, and the ISPU model zoo provides pre-optimized models.

For Stellar MCUs, StellarStudioAI is a software package for neural network optimization and deployment.

The tools featured in the ST Edge AI Suite are free of charge, including for commercial use.