Last week, I visited the Embedded World (2025). Only the exhibition, not the conference part. It was great to be back again, meeting old friends and making new acquaintances in the embedded business. I already told you the dramatic story of how I got there. This blog is about what I saw at the show – trends and technologies.
The Show
The last time I went to the EW was back in 2019, when I was at Intel and really only peripherally involved with the embedded business world. This time it was for real, as I am now at ASTC working with VLAB and we have real customers and partners to meet with and visit at the conference.
Unlike many other events, the Embedded World is still going strong. According to Elektroniktidningen, the number of visitors and exhibitors in 2025 is about like it was in 2018.
Anyway, let’s go through some observations from the show.
Artificial Intelligence – at the Edge
“AI” was everywhere. Sometimes it meant “machine learning”, sometimes large language models. Quite often it was about vision applications for industrial production or vehicle applications. It is clear that AI acceleration hardware is appearing in even very small microcontrollers. Not to run LLMs, but smaller models that would have been called machine learnings a few years ago.
There were examples of full-blown LLM usage. A rather extreme example was found in the NXP booth where they showed off the hardware they got with their recent Kinara acquisition. They were running 8B-size LLMs locally on the demo machine. The demo used NXP i.MX9 chips as the hosts for the Kinara accelerators. The accelerators were attached over PCIe (gen 3), with 16 GiB of local RAM – not totally unlike a regular PC. Still, I think they said it was using just 15W, which is pretty good.
At another extreme, ETAS showed their Embedded AI Coder. This is a software tool that takes trained AI models in ONNX or TensorFlow Light formats and compiles them into C code. This means that tiny models can be run on tiny hardware. The demo used a dozen kilobytes on an ARM Cortex-M0!
Automotive
Automotive is always big at the Embedded World. Basically, if you wanted to demonstrate your software technologies or software tools at the embedded world, you did it using either AI or Automotive or both. There were also many vendors who specialize in Automotive and only serve general embedded in a peripheral way.
Here are some examples.
This demo in the ARM booth showed a system from Cipia tracking the attention and activities of a driver. It could detect a user picking up a phone or drinking from a cup. It all ran on various ARM processors.
QA Systems brought a real Porsche onto the show floor and put it next to another driving simulator.
The Wind River booth featured a demo platform from their parent company Aptiv. The skateboard is a fully drivable electric car chassis with real ECUs running it. Note the model of the James Webb telescope hanging in the air above it.
Open Source
Open source has a big presence in the embedded systems field. That is both old news and new news. Open-source real-time operating systems have been around for a long time. But in automotive, the acceptance for open-source software seems quite new. The logic that is beginning to emerge is that open-source software makes sense for non-differentiating aspects of a product. Share the boring fundamental functionality and innovate at higher levels of the stack.
The Eclipse foundation had a booth just to demonstrate their embedded open-source projects including SDV (Software-Defined Vehicle), open hardware based on RISC-V, and the ThreadX open-source operating system. I did not know this before, but apparently ThreadX was acquired by Microsoft and turned into “Azure RTOS” in 2019, and then it got turned into an open-source project in 2023. I still recall the wonderful booths and give-aways ThreadX used to have at the Embedded Systems Conference in the early 2000s…
The biggest open-source hit must have been Zephyr. This RTOS was everywhere. They also had their own booth, but demos based on Zephyr were running in many other booths. Debugger and compiler companies targeted Zephyr, and hardware companies were running Zephyr on their chips. Zephyr has been on the radar for a long time, I talked to Intel folks about it back in 2018 or so, and I have a t-shirt at home from its original launch as Wind River Rocket back in 2015.
RISC-V
RISC-V was everywhere at the show. Or at least it felt that way.
The RISC-V community booth gathered RISC-V cores from Synopsys, SiFive, Andes, and Semidynamics. There was a display of physical chips featuring RISC-V processors including devices from Renesas (R9A) and Microchip (Polarfire SoC FPGAs) and smaller players like Cortus (Ulyss).
RISC-V was also prominently featured as a target architecture for software tools, in particular debuggers. IAR, Lauterbach, Tasking, and others showed off tools specifically targeting RISC-V.
The biggest news around RISC-V must have been the unveiling of future RISC-V-based AURIX chips from Infineon. It seems that Infineon will eventually replace the venerable TriCore with RISC-V in their main line of automotive controllers.
Rust
The Rust programming language is not quite as hot as RISC-V, but it did show up here at there. The promise of code with fewer safety problems is very attractive for safety-critical systems. There was a particularly interesting setup where Vector ran Rust on their Autosar operating system using a compiler from HighTec. All demoed in the Infineon booth since the setup ran on an Infineon chip. Partnerships are de-rigeur in Automotive.
Panther Lake
Since I did not leave Intel that long ago I was pleasantly surprised to find a Panther Lake chip exhibited in the Intel booth. However, it was also quite funny in that the chip was just sitting in a box with no explanation of what it was and why it was there. It was just Panther Lake.
Great Demos
Finally, there were a couple of really well-done demos on the floor that I felt deserve a shout-out.
Lauterbach demonstrated code coverage and how hard it can be to achieve 100% code coverage using a simple snake game! This was a really clever setup where the game was running on a microcontroller that was being traced in real-time using a Lauterbach debug box. The user controlled the snake using a joystick and the goal was to play the game to achieve 100% code coverage. You needed to play a few games in a row within the allotted time, since some of the code coverage came from dying/ending the game in various ways. Like colliding with each wall separately.
Renesas 365 (powered by Altium) had a demo where a robot arm drew a picture of you, using some embedded edge AI. The flow started with a camera capturing a photo of you. Then, the photo was processed on an embedded MCU (the specific type of which I forgot) to produce a line drawing. This line drawing was then executed by the robot arm, under the control of a second MCU.
The point of this complex demo was to show how the Renesas 365 tools could combine chip selection, board design, and management of the software layer in a single tool.