ST consolidates STM32 edge AI tooling into four-part workflow

STMicroelectronics has redesigned the workflow inside ST Edge AI Suite, replacing a looser set of resources with four integrated tools aimed at embedded machine-learning development on STM32 microcontrollers.

Model Explorer provides a web interface for browsing pre-trained models that were previously reached through the Model Zoo GitHub repository. Dataset Factory handles capture and transformation of raw data into usable training sets. Model Factory covers application-targeted training, customisation and quantisation, with optimisation steps aimed at STM32 memory and compute constraints. Model Bench gives a web interface for benchmarking trained models remotely on ST's own board farm, and replaces the previous ST Edge AI Developer Cloud service.

Stephane Henry, Edge AI Solution Group VP at STMicroelectronics, said the company now offers the largest collection of AI models of any microcontroller vendor, citing over 400 pre-trained models including a large selection aimed at sensing applications, which he referred to as physical AI. This figure and the ranking claim are STMicroelectronics' own characterisation and have not been independently verified.

ST said Edge AI Suite currently supports over 260,000 projects, a number it describes as continuously growing, drawn from users ranging from large technology companies to startups and independent developers across consumer, industrial, medical and automotive applications.

The suite connects to the wider STM32Cube ecosystem, which ST says spans over 4,000 commercial part numbers, along with software libraries, IDE integration and evaluation hardware ranging from Nucleo boards to Discovery kits and expansion cards for wireless connectivity, vision, time-of-flight and motor-drive applications.

STMicroelectronics said the new tools will be available in late November, with registration open for notification ahead of that release. The current version of ST Edge AI Suite remains accessible in the meantime.