After introducing the STM32MP2 series in 2023, ST is now launching the STM32MP25 and announcing the release of the STM32MP23 and STM32MP21. The STM32MP25 will be in mass production by the first half of this year, while the STM32MP23 will arrive by the end of 2024. As for the STM32MP21, we will also be sampling it by the end of 2024, with production expected in the first half of 2025.
By opening our roadmap, we want to give greater visibility into the new STM32MP2 series and help our community understand the spirit driving our innovations. Hence, while this blog post will primarily focus on the STM32MP25, it also serves as a peek behind the curtains of our operations as we continue to answer the new challenges in security and advanced edge AI computing in Industry 4.0 applications. That’s why, among other things, the STM32MP25 is our first MPU with a neural processing unit (NPU) capable of 1.35 TOPS.
STM32MP25: Advanced computing capabilities
NPU
Additionally, we are shipping a comprehensive suite of tools to help developers take advantage of the new NPU. Compared to 2015, developers no longer need to be in a doctoral program to run an optimized neural network algorithm. In fact, the NPU in the STM32MP25 is already available for experimentation, even if developers don’t have access to a device. Indeed, it is already possible to run neural network applications on it through the Board Farm of the STM32Cube.AI Developer Cloud, meaning that developers can concretely see how their neural network would run on the new device before they even get it in their hands. The STM32MP25 will be available on the Board Farm during Embedded World (April 9 to 11).
Graphics and video
The STM32MP25 features two Cortex-A35 running at up to 1.5 GHz, one Cortex-M33 at 400 MHz, and a 32-bit DDR4/LPDDR4/DDR3L memory controller. There’s also a new GPU and a new VPU. The new GPU is capable of rendering 3D UIs in 1080p. The new video processing unit features a traditional H.264 decoder and comes with a hardware encoder to optimize video capture. To help developers create an HMI, the new device also houses three display outputs supporting LVDS for protocols like FPD-Link, and DSI for MIPI-DSI.
Supporting the growth of connected applications
Connectivity
The STM32MP25 will be the only STM32MP2 to provide a PCI Express Gen 2 controller, a USB 3 controller, and three Ethernet ports. One is directly connected to the Gigabit Media Access Control, or GMAC. The other two are connected to a switch cascaded behind a second GMAC. Consequently, it becomes possible to design very efficient equipment for industrial applications that can manage network packages without waking the processors. Additionally, the Ethernet controllers support Time-Sensitive Networking (TSN) endpoints for deterministic applications, such as control applications for industrial systems, or audio and video flows. They also support the precision time protocol (PTP) for cellular tower equipment or satellite navigation that uses a synchronized clock to coordinate packet transmissions.
Security
Connected applications have a lot more security needs. Hence, all STM32MP2s target a SESIP Level 3 and PSA certifications. Concretely, developers can use TrustZone to create segregated environments on the Cortex-A and Cortex-M to protect from intrusions. The STM32MP2 even goes a step beyond since it is also possible to isolate resources, such as memories or peripherals, to make them only accessible from given traffic initiators. The new devices also offer the same security features that helped make the STM32MP1 popular in sensitive applications. It, therefore, targets PCI pre-certification for Point of Sales, among other use cases. From secure provisioning to over-the-air updates, developers can create mission-critical applications while satisfying the more stringent demands of governments and customers.
Ecosystem
STM32MP23 and STM32MP21
We will release more information about the STM32MP23 and STM32MP21 as we near their mass-market release in 2024 and 2025, respectively. Still, we are thrilled to share their configuration to help our community plan for them and adopt the STM32MP2 series more rapidly and efficiently.
STM32MP23
The STM32MP23 will feature two Cortex-A35 at up to 1.5 GHz and one Cortex-M33 at 400 MHz. Given its position as a more cost-effective solution that would process significantly less data than the STM32MP25, it has a 16-bit DDR4/LPDDR4/DDR3L memory controller and two ethernet controllers supporting GMAC. Similarly, the video accelerator only features an H.264 decoder since it will not need to encode data, and the future MPU also features a smaller GPU while still supporting DSI, LVDS, and RGB displays.
STM32MP21
The STM32MP21 has one Cortex-A35 at 1.5 GHz, one Cortex-M33 at 300 MHz, a slower 16-bit memory controller than the STM32MP23, and the same Ethernet controllers. Similarly, as the device targets very different graphical needs, it offers the display parallel interface found on the other two STM32MP2s, omits the GPU, but still supports parallel and CSI interfaces for cameras to ensure developers can easily interface with various sensors.
The STM32MP2s working together
The whole series can work together to create an ecosystem of products. Let’s take a medical imaging analysis application. The camera interface and processing capabilities of the STM32MP21 mean that it could apply various filters to clean the data received before sending it to an STM32MP25, which would run the neural network. Developers can thus increase the overall accuracy of their system without adding another workload on the STM32MP25, which would be able to run inferences, show results on its display, and securely send relevant data to the cloud. And since several packages in the STM32MP2 series are pin-to-pin compatible, developers could reuse many of the same design cues and code to reduce their time to market.
