ST recently announced the mass-market availability of the STM32MP23x (STM32MP235, STM32MP233, and STM32MP231) product line. After releasing the STM32MP25 a year ago, the new devices target industrial applications that have higher cost constraints but can still benefit from a neural processing unit (NPU), a heterogeneous architecture (Cortex-M33 and dual-core Cortex-A35), a real-time operating system beside Linux, powerful networking interfaces with Time-Sensitive Networking (TSN), and more. Put simply, it makes machine learning at the edge more accessible for applications that run perfectly fine with a 16-bit DDR4/LPDDR4/DDR3L memory controller and a video accelerator that only offers H.264 hardware decoding because the system simply won’t need to be encoding video streams.
Table of Contents
STM32MP23x: Focusing power where it matters
Thinking about upgrades
As a picture is worth a thousand words, let’s paint the story of an automated bottling factory, where processes are complex, machines must perform precise operations guided by numerous sensors under the orchestration of controllers, and engineers are finding new uses for machine learning, from predictive maintenance to defect detection and process optimization. As many systems on the factory floor already run on Linux and most of the technical staff is familiar with the UNIX operating system, integrators are looking for a microprocessor to help them upgrade some of the existing machines.
Mitigating circumstances
In such a scenario, including two Cortex-A35 at 1.5 GHz in the STM32MP23x means that developers can run their Linux OS while creating very low-power runtimes that could run on an included Cortex-M33. In a nutshell, the new devices offer the same flexibility as the STM32MP25x since it has the same processors and runs at the same frequencies but relies on a more cost-effective design. Too often, teams push back system upgrades due to costs. A device like the STM32MP23x can help mitigate those issues, ensuring a smaller gap between each system iteration, which also makes development more straightforward.
STM32MP23x: Optimizing intelligence and making machine learning at the edge ubiquitous
More machine learning at the edge
Simplifying development is particularly important in the age of machine learning at the edge because companies know they can benefit from it but often balk at the cost. Let’s take the example of our automated bottling manufacturing line. Numerous case studies already show how sensors on motors can help with predictive maintenance, significantly reducing overall operational costs. There are also an increasing number of AI systems that can help catch defects. For such a factory, being able to discard bottles with defects can provide considerable gains in customer satisfaction and vastly lower issues with defective bottles. That’s where AI comes in, but too many don’t fully understand what that means.
More than just an NPU
Indeed, while an NPU is essential, it’s not the only thing needed. That’s why the STM32MP23x has a neural processing engine capable of 0.6 TOPS but also has a camera interface with Lite-ISP capable of rapidly processing images from the sensor. There’s also a hardware accelerator for decoding images at up to 500 Mpixels per second, and MIPI DSI and LVDS interfaces for a human-machine interface. Too often, competing solutions will only focus on the NPU. By offering an MPU like the STM32MP23x, engineers can provide people working at the automated bottling factory with a way to capture images, process them right away, and put them through a neural network without using multiple MPUs or co-MCUs.
STM32MP23x: Answering the unique challenges of industrial applications
Doing more with less
Integration remains the driving force behind microprocessors. They must do more. For a smart factory, like the automated bottling line in our example, this is critical because it helps reduce the points of failure and helps design a more comprehensive system. For instance, if the MPU responsible for running the machine learning algorithm can also power the UI projected on a display used by the employees, it can improve productivity and optimize workflows. However, that demands that the MPU take a comprehensive approach to industrial applications. The automated bottling manufacturing line has a myriad of other systems, and all must coordinate with one another.
Less uncertainty and more solutions
Consequently, the STM32MP23x keeps a lot of the features of the STM32MP25x that shine in industrial applications, like CAN-FD buses, an SDIO 3.0 controller, support for USB Type-C Power Delivery, a USB 2.0 Host/Device controller, and two Gigabit Ethernet module with support for time-sensitive networking. This last feature makes creating a deterministic network possible. In a factory where motor control applications demand exact data synchronization and low latency, TSN is imperative to ensure predictable information delivery. It also enables engineers to prioritize packets to plan for emergencies and even build redundancies to improve their infrastructure’s robustness. Put simply, the heart of a factory can beat on one well-made microprocessor.
- Learn more about the STM32MP235, STM32MP233, and STM32MP231
