STMicroelectronics extends embedded AI solutions

These tools facilitate moving AI and ML to the edge of an application. At the edge, AI/ML delivers substantial advantages, which include privacy by design, deterministic and real-time response, greater reliability, and lower power consumption.

NanoEdge AI Studio is an automated ML tool for applications that do not require the development of neural networks. It is used with STM32 microcontrollers (MCUs) and MEMS sensors that include ST’s embedded intelligent sensor processing unit (ISPU). For developers needing to use neural networks, STM32Cube.AI is an AI model optimiser and compiler for STM32. The two new releases deliver features that help design and implement high-performance AI/ML solutions quickly and with minimum investment.

NanoEdge AI Studio version 3.2 now contains an automatic datalogger generator that increases development productivity. Its inputs include the ST development board and developer-defined sensor parameters, such as data rate, range, sample size, and number of axes. With these, NanoEdge AI Studio produces the binary for the development board without the developer having to write any code.

Because dataset quality directly impacts machine learning performance, the new data-manipulation features in NanoEdge AI Studio allow the user to clean and optimise the captured data in the NanoEdge AI Studio in a few clicks. A new validation stage has also been added, which helps users assess their algorithms by showing inference time, memory usage, and common performance metrics such as the accuracy, and F1-Score. It also highlights more information about the pre-processing and ML model involved in the selected library. The newest enhancement to the NanoEdge AI Studio adds more pre-processing techniques and ML models for anomaly detection and regression algorithms that boost performance. In addition, the tool supports creation of smart libraries that can predict future system states using multi-order regression models.

STM32Cube.AI version 7.3 is a tool for developing cutting-edge AI/ML solutions. Fully integrated into the STM32 ecosystem, it enables conversion of pretrained neural networks into optimised C code for the industry’s most popular family of 32-bit Arm Cortex-core MCUs. The enhanced STM32Cube.AI adds greater flexibility for neural-network (NN) optimisations. The tool can adapt existing neural networks to achieve performance demands, fit within memory limitations, or, in a balanced optimisation, get the best of both. The update also brings support for TensorFlow 2.10 models and new kernel performance improvements.