STMicroelectronics are offering a set of AI solutions for its family of 32-microcontrollers. This solution makes it possible to use your existing microcontroller to run on DNNs. This development makes it possible for microprocessors to create and run DNNs themselves.
Artificial Neural Network (ANN), is a neural network, consisting of more than two layers (input, output layers and as a minimum one more) and are characterised by a certain amount of complexity. These networks are designed to deal with mathematical modeling, and to process data in more sophisticated way. ANN is widely used in AI applications.
STMicroelectronics’ AI solution, called STM32Cube.AI is an extension pack, that allows users to run ANNs, enabling AI on the existing STM32 microcontrollers.
The AI enabling process is very straightforward, and consists of few steps.
First capturing physical data occurs by using a variety of sensors, like light, temperature, humidity and others. Data capturing and control can be performed using the ST BLE Sensor app for smartphones. Physical data capture can be performed with the SensorTile module, offering huge functionality.
After physical labelling, data sets are characterised and assigned to specific outputs. The developer can decide what ANN topology they will use for a desired application. Assigned sets of data are then used for ANN training.
STMicroelectronics and partners like Bluewind, Neovision, Cartesiam, NALBI, SIANA Systems, Leonard+Bauer, provide engineering services to support ANN architects and data scientists.
Second, data should go through neural networks in order optimise proper error criteria and create pre-trained ANN. Mentioned above STM32Cube.AI software tool allows this step.
Next, pre-trained ANNs should be converted into optimised code suitable for running on an MCU. ST also helps designers to prototype desired applications using Function Packs. These Function Packs are a set of examples, combining embedded low-level drivers, middleware libraries and application samples, that can be used by developers as the template of their design.
Summing up, STM32Cube.AI offers developer popular deep learning training tools, sensing tools, and allow ANNs to run on existing STM32 MCUs, plus many more functional possibilities.