This repository provides examples, tutorials, firmware, and other development resources for the ISPU, a dedicated ultralow-power, high-performance, programmable core, able to execute real-time processing directly inside the sensor.

The ISPU allows implementing algorithms written in C code running directly inside the sensor instead of on an application processor, enabling consistent reduction in power consumption, latency, and cost. Any type of algorithm can be implemented, from artificial intelligence to signal processing.

Thanks to the sensor hub functionality, which allows connecting up to four external sensors, the ISPU can process their data as well, thus enhancing the intelligence of sensors that do not have any onboard processing.

The output of the algorithms can be read from the application processor at any time. Furthermore, there is the possibility to generate an interrupt when there is new relevant information in the output, so that the application processor can otherwise sleep and save power.

The sensors embedding the ISPU are supported by the ISPU-Toolchain, the X-CUBE-MEMS1 software package, MEMS Studio, a graphical application to develop and test solutions, and the software included in this repository.

Pretrained artificial intelligence models can be easily converted to code optimized for the ISPU using ST Edge AI Core. Other tools from the ST Edge AI Suite that are based on ST Edge AI Core can also be used: MEMS Studio enables a graphical approach with its ISPU Model Converter feature, while ST Edge AI Developer Cloud allows converting and benchmarking models with only a web browser.

For more information, please explore the page on ST's website dedicated to the MEMS Sensors Ecosystem for Machine Learning.

This repository is structured as follows:

• A docs folder, containing the documentation for the ISPU DSP library in HTML format. In order to visualize the documentation, clone or download the repository and open the index.html file with a web browser.
• An examples folder, containing templates and example projects as well as libraries to start programming with the ISPU, together with prebuilt files ready to be used with the sensors. Additionally, it contains instructions on how to set up the development environment.
• A host_firmware folder, containing various firmware for boards hosting sensors embedding the ISPU.
• A model_zoo folder, containing a collection of artificial intelligence models optimized for the ISPU, that can be used as is or can be modified and retrained to fit the needs of the user.
• A tutorials folder, containing tutorials describing how to create example solutions using different ST hardware kits and software tools.

Here is where to find the resources helpful when using ST Edge AI Core, MEMS Studio's ISPU Model Converter, and the ST Edge AI Developer Cloud:

• Templates for integration into the final application (template_stedgeai), available in the examples folder and organized by device. A guide on how to use them to deploy an artificial intelligence model on the ISPU is included.
• Templates for validation on target hardware (template_stedgeai_validate), available in the examples folder and organized by device. A guide on how to perform validation on target is included.
• Host board firmware for validation on target hardware, available in the host_firmware folder. A guide on how to prepare the board for validation is included.
• Tutorials for developing an application using ST Edge AI Core, available in the tutorials folder. They cover all steps starting from the data collection and finishing with the test of the final application.
• Model zoo for retraining and deploying ready-to-use models with simple automated procedures, available in the model_zoo folder. A guide is included for each model of the zoo.

For more help with ST Edge AI Core, please refer to the HTML documentation distributed with the tool, available in its installation folder.

More information: http://www.st.com

Copyright © 2023 STMicroelectronics