STMicroelectronics and Schneider Electric Demonstrate how AI is Moving to MCUs
09-12-2020 | By Robin Mitchell
The world of MCUs is always moving fast, and the recent demonstration by STMicroelectronics and Schneider Electric show how AI is now entering the MCU world. How has MCU technology changed over the years, what was demonstrated, and how will it help future designs?
How has MCU technology changed?
Microcontrollers have been at the centre of most electronics produced for the past few decades, and this is due to their strong I/O capabilities while being programmable. A standard processor allows for complex computer systems to be constructed with large memories and advanced user interface peripherals. Still, such a system is almost always bulky, power-consuming, and complex to program.
A microcontroller integrates all the parts needed for a small CPU to operate as well as additional I/O features such as bus controllers, transceivers, and peripherals. While microcontrollers are nowhere near as powerful as a dedicate processor, their ability to be easily integrated into circuits makes them incredibly versatile and useful in most applications.
However, while microcontrollers have been around for several decades, their power and capabilities have dramatically improved over the years. The first microcontrollers would have several bytes of variable storage and a few thousand instructions for programs. Still, now microcontrollers can have thousands of kilobytes of RAM and megabytes of ROM. These same microcontrollers can integrate their own power conversion circuitry which eliminates the need for an external regulator. They can also have extremely advanced peripherals such as TCP/IP stacks for internet communication.
But, the introduction of AI into MCUs is what will revolutionise the industry. AI technology allows for systems to not only intelligently react to incoming data, but can also learn from this data to improve its own performance over time. Currently, most AI systems are limited to processors on main computers or those found in mobile devices, but MCUs rarely integrate such capabilities.
What did STMicroelectronics and Schneider Electric demonstrate?
STMicroelectronics, who are famous for their STM range of microcontrollers, are jointly demonstrating a new system developed with Schneider Electric that puts AI on an STM32. The new device is a prototype IoT sensor that can help with building management by intelligently determining building occupancy and usage.
The use of STM32Cube has allowed for Schneider Electric engineers to effectively use hardware and peripherals present on the STM32 range of microcontrollers while also streamlining the project. The STM32Cube platform also has an integrated AI toolchain to integrate AI capabilities into its microcontrollers, and this demonstration shows how AI is now entering the MCU world.
The prototype sensor utilises the LYNRED ThermEye imager with a Yolo-based neural network that runs on an STM32H723. The MCU integrates a 550MHz ARM Cortex-M7 core with up to 1Mb of FLASH and 564KB of SRAM. Multiple security peripherals include hashing hardware acceleration, secure firmware install, and secure boot, while also integrating up to 35 different communication interfaces such as FD-CAN, USB 2.0, and camera interface.
Why is AI on MCUs significant?
Bringing AI to MCUs provides engineers with a whole new dimension of design and capability. The past decade has demonstrated how important AI is in everyday applications; it can be used to improve efficiency, predict issues, and improve overall functionality over time. But AI can also be useful for creating systems that are tailored to specific individuals without the need for complex programming or careful selection of variables; just let the system observe the system it is intended to work with and eventually it will be able to develop a neural network to run off.
Since its introduction, AI has only been possible on powerful computation systems such as datacentres and desktop computers, and now that mobile technology has caught up, can also be integrated into mobile devices. However, real-world applications of AI are generally found where there are simple MCUs with sensors such as IoT and IIoT. Due to the simplicity of these devices, they are not able to run their own AI neural nets and often require the use of a remote data centre to process requests.
However, if AI capabilities can shift to MCUs, the ability to process data internally (also known as on the edge) on an MCU would bring about a new era of data processing. For example, one of the greatest concerns with IoT devices is the sending of private information. Instead, if data can be processed on the IoT device without the need to transmit private data, the security of such devices is significantly increased. The ability to run AI on-device also reduces latency times as the need to establish a server connection, upload the data, and wait for a response is no longer needed.
There are MCUs on the market that has been able to run AI for many years. Still, this demonstration of using an STM32 shows how everyday microcontrollers are now becoming ever more powerful, and the use of STM32Cube.AI tool to easily integrate AI systems into an MCU demonstrates how MCUs will now finally start to have AI capabilities.
Read More
