Issue link: https://resources.mouser.com/i/1481902
Intelligence at the Edge 24 M icrocontrollers (MCUs) are the ubiquitous computing chips that manage most battery- powered IoT devices and embedded electronic systems found across all industries. STMicroelectronics (ST) will ship more than two billion 32-bit STM32 MCUs this year and offers special STM32 AI tools and function packs that allow smart devices to infer independently from the cloud. Figure 1: The key steps behind neural networks. (Source: STMicroelectronics) Today, with the emergence of ever more advanced hardware, efficient software tools, and highly optimized AI models, it is feasible to run machine learning (ML) algorithms, even complex neural networks (NN), on an MCU. But MCU devices have less computational power, memory capacity, and supplied energy than traditional AI computers with fast CPU, GPU, and NPU cores. This makes the implementation of ML workloads on MCUs considerably more challenging. For this reason, ST's primary objective has been achieving favorable AI/ ML performance with limited device resources and developer AI know-how. Figure 1 shows the typical development process for embedding ML on an MCU, from data collection and labeling, NN model training and conversion, to the final deployment. Data collection is beyond the scope of this article, but keep in mind that data is the most important part for ML. It is impossible to build a NN model without data, just as a talented chef cannot cook a tasty meal without ingredients. Training a NN model begins with determining what framework to use. This is a case-by-case decision, Embedded Neural Networks Approach Allen Ren, STMicroelectronics