Hello, and welcome to the STM32 Summit. I am here in the heart of Silicon Valley to share with you the latest in STM32 innovation. Our mission is to empower a global community of close to 2 million developers to deliver products faster and with greater ease. Coming up, we will introduce the new STM32A, bringing STM32 into automotive and high-safety systems. We will launch the STM32G5 and see how it supports application from robotic hands to data center power supply management. The STM32H7 also gets an exciting upgrade. We will share the latest AI ecosystem update, including our brand-new STM32Cube for Agentic AI. First, our main theme today is technology convergence. For STM32, this marks an important evolution of the platform, an evolution that I genuinely believe is one of the most exciting things we have done since we launched the family nearly 20 years ago.
Embedded systems are entering a new phase. Across automotive, industrial automation, and robotics, customers are moving beyond isolated hardware-defined products towards systems that evolve over time, connected, secure, software-defined, and increasingly able to act autonomously in the real world. The opportunity for humanoid robotics alone is rapidly approaching $100 billion per year, with the overall industrial robotics market representing many multiple of this. That does not mean these markets are becoming the same. Each will keep its own standards, certification, and lifecycle expectation. But underneath those differences, something important is changing. The technology foundations are converging, and we are making sure STM32 evolve with them.
Embedded systems are entering a new phase. Different markets have different requirements and face different realities, but the technology foundations are converging. AI, sensing, connectivity, control, safety, and security. STM32 brings these technologies together across markets, enabling physical AI systems to perceive, decide, and act in the real world. This is technology convergence. Success is no longer about a single MCU. It is about choosing a platform, the unified STM32 MCU platform, powered by portfolio consistency and continuity, a common ecosystem foundation, and shared process technologies. Build faster, build on success, evolve with confidence. Welcome to the convergence era, powered by STM32.
That video captured this transformation. Across automotive, industrial, and robotics, the same forces are reshaping systems: AI, connectivity, sensing, control, safety, security, and software lifecycle management. Equipment manufacturers are pushing legacy solutions to their limit. We see it in next-generation vehicles, autonomous robots, and heavy industrial drives. How do they manage system complexity as software-defined features multiply? How do they increase software reuse and reduce redundant effort? How do they scale their architecture? How do they sustain certified safety and security across long life cycles? This convergence is making a new generation of physical AI systems possible, systems that can perceive, decide, and act in the real world. That can mean a smart meter catching abnormal behavior from the grid before it becomes a fault.
It can mean an autonomous mobile robot finding its own path through a crowded warehouse floor, or a software-defined vehicle making a real-time decision, a decision that in the past could have required a full ECU redesign to change. The applications differ, but the core capability increasingly do not. Embedded intelligence, secure connectivity, deterministic control, and safety mechanism. That is why the platform matters beyond any single device. You need a complete foundation that lets you reuse tools, skills, software assets, and IP across product families and application domain. You need confidence that what you build today will still create value as requirements shift tomorrow. That is the role of the unified MCU platform around STM32 that we are introducing today. It is designed to help you develop faster, increase reuse, lower risk, and optimize total cost of ownership while meeting domain-specific needs. It rests on three foundations.
First, portfolio consistency and continuity, Arm Cortex-M cores, and shared architectural principles across families. Second, a common ecosystem foundation. STM32Cube, the Arm ecosystem, development tools, software components, and support models all help cut complexity and accelerate development. This is backed up by our ST Community with over 300,000 visitors every month, and is being augmented by our increasingly popular AI-powered assistant, the STM32 Sidekick. Third, shared process technology. This include 40 nanometer flash and 18 nanometer FDSOI with embedded phase-change memory. They deliver the performance, density, and reliability needed for faster innovation across domain, and this rely on a highly resilient supply chain we manage ourselves. These foundation matters because they allow us to bring critical capabilities to different markets quickly and reliably. Let me now double-click on our first foundation, Arm Cortex-M cores, a longstanding technology foundation familiar to all of us.
ST and Arm have worked together closely for more than two decades, and that cooperation has helped shape the STM32 portfolio we know today, and help us ship more than 20 billion devices. Let's take a short moment to see why Cortex-M has been so important, and why it continues to be the architecture of choice for STM32 and the close to 2 million developers working with our ecosystem. Today, together with Arm, we are sharing an important next step in the future of Cortex-M real-time compute roadmap. ST and Arm are continuing to work together to deliver the unified STM32 platform and support the technology convergence needed for physical AI. We believe that the industry needs less focus on creating another computer architecture and more focus on helping MCU technologies converge around a common foundation that supports software reuse.
To explain what we're doing together, it's my great pleasure to bring in Drew Henry, Executive Vice President of Arm Physical AI Business Unit. Drew, it's great to have you with us. Can you explain why the ST-Arm partnership and Arm Cortex-M are so critical to technology convergence and physical AI?
Sure, Remi, and thank you very much for giving the opportunity to join you. We've been working together for 20-plus years, our two companies, and the result of that has been such a remarkable investment in the Cortex-M architecture. We've shipped over 100 billion Cortex-M devices in our history now. It's an important part of our platform. As we're walking now into this new world, this new world of physical AI, I think it's going to be even more important. Physical AI is this thing where we're going to embed AI technologies into things that mine, grow, the agriculture space, manufacture, manage transportation systems, and the like. The ability to bring AI into those devices is going to be really quite transformative to the productivity in all those areas and grow to substantial GDP growth across the world.
We believe that this is a $200 billion compute market in the middle of next decade or so, probably one of the biggest compute markets, if not the biggest compute market ever. This is an area that we're investing in very heavily, and it's fantastic to be working really closely with you all as we do this, because these systems we're designing are very complicated. You can't build the computer brain without also building all of the devices that have to actuate around it, and that's where Cortex-M becomes so important to us, and that's why software compatibility across this area is so important for us. That's why today we are very pleased to announce that ST will be a lead partner in defining the future of our Cortex-M portfolio. This is an ecosystem-wide effort.
We can't do something like this entirely by ourselves, Cadence is going to be a key design partner for us in this effort, and together we will be defining the future of the Cortex-M roadmap to meet the opportunities that the entirety of the Arm partnership needs.
Thank you, Drew. We are very happy to continue this partnership as we work together to secure the future of Cortex-M. You mentioned the important role it will play in physical AI. ST unified STM32 platform is an industry first and a game-changer for developers. Can you also share how our work together fits into Arm's broader ecosystem vision for physical AI?
Yeah. As we both know, Remi, one company can't do this alone. Our ecosystem needs to work together to be able to accelerate the adoption of these technologies and really get this benefit that I was talking about before, GDP growth. We believe deeply in the power of ecosystems, and we announced just in September the Arm Total Design Program for physical AI. We have over 100 companies that are now part of this, including ST, and we're so grateful for that. Together with all those companies, we want to do things like define really kind of what capability that all robots have. How do we create a framework for language for that?
How do we accelerate development of our industry for the things that we want to be able to do so that physical AI technologies and the benefits of it can be really adopted more quickly into all the different application areas that we know are useful for them? Most importantly, of course, is to make sure that we've got a solid Cortex-M roadmap. So our partnership with you is key for this and a critical part of really this entire thing we're trying to build in the space of physical AI while we define the future for high-performance MCUs as being a really important platform for all these devices that we want to build.
Drew, thanks for being with us today. It was great to host you. Thank you, and I hope to see you soon. We are really excited about this key collaboration as we're ensuring the evolution of Cortex-M, the architecture of choice for over 100 billion devices so far and for the future of physical AI. With this convergence, customer and software developers will be able to reuse more of their code base across application, from cars to robots and industrial system. For ST, it will enable new products such as our STM32A and STM32V8 family to take the next steps in performance and versatility. This helps developers address the most demanding applications where performance is key, but system power consumption and cost are highly constrained. We are excited about the future of Cortex-M we are creating together. Stay tuned for more.
Our unified platform foundations are solid, built over decades, and designed to support innovation for decades to come. They matter because they let us bring advanced capabilities to customers faster. On top of them, we bring the technology building blocks customers need to create secure, software-defined systems and physical AI. For edge AI, we combine Arm CPUs, ST neural accelerators, and more than 10 years of MCU AI software experience. For connectivity, we support the shift toward deterministic Ethernet-based architectures. For security, we already offer STM32s with integrated SHA algorithm accelerators and PQC-optimized cryptographic libraries, the first MCU maker to offer a fully certified solution. For safety, we support increasingly rigorous requirements through qualified hardware and software bricks, diagnostics, safety analysis, manuals, and assessment support. We are now extending that platform approach to cover a wider range of application.
With that in mind, let's hear from my colleagues, Maria and Patrick, about a key addition to our lineup.
Thanks, Remi. Today, we are introducing the new STM32A series. The STM32A combines two decades of experience in designing general-purpose microcontrollers and automotive MCUs to create a new series addressing both automotive and high-safety embedded applications. ST has a strong and established track record in automotive microcontrollers. For more than 20 years, we have supported car makers and tier ones with automotive-grade MCUs, shipping billions of devices. They are designed for reliability, functional safety, security, quality, and long product life cycles. We understand what this market requires: robust qualification, automotive-grade development practices, safety evidence, long-term availability, and support from design-in to production. Why now? Vehicle architectures are changing. They are moving from many isolated electronic boxes toward more centralized, software-defined, and zonal approaches. As you heard just now, similar trends are appearing in industrial automation, robotics, and emerging physical AI systems.
Embedded designs are becoming increasingly connected, intelligent, software-driven, and safety-critical. Across these domains, you, our customers, face a common challenge. You need the right balance of cost, compute, memory, connectivity, integration, safety, and security. Yes, I know it's a lot. You need it without fragmenting your software, your tools, and development flows. That is why we are introducing the STM32A and STM32 series, built for vehicle and high-safety applications. It brings STM32 scalability, compatible IP, software concepts, tools, and ecosystem continuity in automotive-grade and high-safety environments. Let's hear from some of our customers and partners about why they are excited about the STM32A and our unified STM32 platform.
As the automotive industry continues its transformation, we at the Volkswagen Group are shaping the mobility of tomorrow as a global automotive tech driver. We highly value our strategic partnership with STMicroelectronics. The STM32A family is an important step towards future automotive electronics, as offering us more choice with a unified platform and software ecosystem.
Software-defined systems are reshaping every market we serve, from data centers and buildings to industry and infrastructure. This transformation is calling for connectivity, scalability, security. In such a context, STMicroelectronics is a key partner. The breadth of STM32 portfolio is an enabler of this software-defined transformation, and it's allowing us to maintain a solid software foundation.
Humanoid robots combine high-level intelligence with many distributed control functions for motion, sensing, power, and safety. Extending the familiar STM32 ecosystem toward higher-safety applications while preserving the scalability, software continuity, and development efficiency is what we need to build more capable humanoid robots.
STM32A is important because it extends familiar STM32 foundations into higher-safety environments with support for modern automotive software and seamless continuity with STM32Cube ecosystem. This continuity helps developers reuse software, simplify validation, and scale solutions across different STM32 devices with greater efficiency and confidence.
At Lutron, we know that innovation is about more than choosing the right device. It's about choosing the right platform and support. The STM32 family scales across our product lines, so our engineers carry their code and experience from one product to the next. With excellent documentation, intuitive devices, and responsive engineering support from ST, we bring the next generation of intelligent lighting to life faster.
Those perspectives show why our unified platform and STM32A matter across automotive, industrial, and intelligent embedded systems. Patrick, can you take us through the first of our products in our roadmap?
Sure. Today, we are unveiling the first part of this roadmap with three product lines. STM32A4 targets entry-level automotive nodes. These applications need low cost, compact memory, crypto support, and LIN implementation. Think distributed sensing, actuation, door, window, lighting, and local control functions. The same profile is also relevant for high-safety industrial and infrastructure edge nodes, including local sensors, actuators, access control points, safety monitors, and compact control units. STM32A5 moves up to cost-efficient body and comfort control. It offers up to 4 MB of memory, enhanced security, and support for safety applications up to ASIL B. Beyond the vehicle, this level of capability also fits mid-range safety-critical applications in industrial automation, robotics, and energy systems. These applications need more memory, stronger security, and functional safety support. STM32A9 addresses higher function zonal endpoints in vehicles.
It is designed for deterministic real-time control, low latency, Ethernet-oriented connectivity, stronger integration, and safety support up to ASIL D. Its split-lock core architecture gives customers flexibility. They can prioritize ASIL D when required or unlock twice the ASIL B performance when ASIL D is not needed. These capabilities also matter beyond automotive. In industrial robots and physical AI applications, real-time response, networking, safety, and security all have to work together.
Thanks, Patrick. Together, STM32A4, A5, and A9 are the first steps in the STM32A roadmap. They extend STM32 into automotive and other high-safety systems, creating a scalable portfolio with broader customer choice. With that, back to you, Remi.
Now, it's my great pleasure to welcome my friend Deepu Talla, Vice President of Robotics and Edge AI at NVIDIA. Deepu leads NVIDIA efforts to bring AI into the physical world. He's the architect behind platforms like NVIDIA Isaac, Metropolis, and Jetson, which have become foundational building blocks for the physical AI ecosystem. Deepu, welcome.
Remi, thank you for having me.
My pleasure. Let's start with the first question. Robotics becoming so big, why do you think it represents actually the ultimate version of AI convergence?
I mean, for 50 years, we always thought that robotics is the ultimate application of AI, bringing AI into the physical world. But the problem is that the accuracy requirements, the reliability requirements, the safety requirements for robotics is extremely high compared to the digital AI, and technology has not been good enough to bring it until very recently. Recently, necessary technologies have matured sufficiently. Things like end-to-end models with reasoning, multimodal, simulation technologies. These are what's enabling us to solve some of these very difficult problems that we had no shot before. Now, of course, a robot is extremely complicated because you have many different sensors. Of course, you need to have a good brain to process all that.
Actuation, because most of the robots that we want to use is going to move the physical world, move atoms, and actuation is going to become extremely important as well. As you know, it's not just a brain problem, it's not just a mechatronics problem, it's a combination of all of these things. One of the most important things that we need to solve is, how do we make sure that all these sensing to processing to actuation is going to be seamlessly solved? This is why one of the technologies that we have invested openly is what's called Holoscan Sensor Bridge, and this is something that enables us to seamlessly get data from sensors, all sorts of sensors over Ethernet, and then also hook up with all sorts of actuators over Ethernet.
Yeah, HSB has been a big game changer for all of us in the industry. Deepu, how do you see the role of ST in this actually physical AI world on a moving forward basis?
Yeah. ST is a phenomenal company, right? You have such a broad portfolio. You have lot of portfolio in sensing. Of course, you have portfolio in power management that's going to be extremely important in robotics. Now, increasingly, safety and security is going to be super important as well, especially when you're starting to think about general purpose humanoids. These will likely involve 50 to 100 actuators, especially even if you think of just the hand itself, there's probably going to be 20 to 30 degrees of freedom. So the number of actuators are going to be many. It's quite likely that this robotics problem is going to be solved in multiple phases, right?
There's going to be likely a system zero that's going to be operating very close to the sensing and actuation, and ST can play a phenomenal role there from a safety perspective, from processing some edge AI right there, obviously from a sensor and actuation. Then when you take that high-speed processing and then bring it to the general purpose brain, or most likely a humanoid brain might be here because the volume's bigger in a humanoid form factor for power dissipation, you can now provide regional safety and then bring that safety back to the main processor that's going to sit in the main robot. Yeah.
The work that we've done together, we started actually to go and implement a Holoscan Sensor Bridge in STM32H7 as a POC. We are firm believer of what you've done here. We actually are going to go in the coming months to release the HSB software as a native component in our STM32Cube ecosystem. It will allow HSB now to be valid for the vast majority of the STM32 products that we are delivering out there. We believe in your vision in term of sensing and actuation. We have this large portfolio from RGB sensors to MEMS. You talked about gate drivers or FET products, and obviously microcontrollers with STM32 and now STM32A.
It's also our intention to allow to pack things together, because I also foresee that the packaging piece, the real estate is going to become a bigger and bigger differentiator or important element moving forward. Deepu, just moving forward on a different theme, which was simulation, and I know that you've been working on this from day one.
That's right.
Please tell me why simulation is becoming the foundational element to really make physical AI a reality.
Yeah. In fact, simulation is one of the major technology unlocks that happened that's enabling this revolution to happen, because testing a robot in the real world is extremely difficult. It's expensive, it's not safe, and of course, you're going to be operating on wall clock time if you're operating in the real world. Simulation can speed up these simulations, these evaluations in a safe manner, in a cheap manner, orders of magnitude, thousands if not millions of times faster. Anything that we can bring into simulation, the time horizon for us to actually train and test AI is going to be sped up significantly. It's been such a hard problem to solve because we are in the chip industry for 30 plus years, Remi.
Yeah.
Everything that we tape out before we tape out, we validate it, we emulate it, we simulate it before we send it to the fab. For robotics, the sim-to-real gap was sufficiently large. It's not completely closed yet, but it's gotten reasonably good enough now that we've been working on this for 15 years using a combination of AI and graphics and building the physics engines, building all the simulation environments, and now making all of this available to AI agents that can use all these different libraries. We've always been thinking about simulation, and this is a major technology unlock. This is why in one of the partnerships, the thing we always talk about is, yes, Holoscan Sensor Bridge is kind of like the plumbing that enables the sensing, the brain, and the actuation to go together.
Anytime we bring all of those capability into simulation, we can build, develop, test orders of magnitude faster before we put it in the real world.
Yeah, it makes perfect sense. You've been so consistent on this, and it's our commitment to you to develop actually libraries of all our product portfolio, be it sensing, MEMS, microcontrollers, or actuation devices, and actually make them available in this environment to go and enable actually the vision.
Yeah, I think that's phenomenal for all the developer ecosystem.
That's great. That's great, Deepu. Deepu, one last word. You've been working on physical AI from the beginning, and I remember that was 10, 15 years ago. We're going to see phenomenal growth in the years to come. Last word of wisdom, or last words of wisdom for the audience.
I think the golden age of robotics is just arriving in time for us, right? If you look back in the last six months or so, AI has really become useful in the digital world, finally. If you look at what the frontier models of general-purpose models from last year, these used to be the trillion parameter models, that same level of intelligence is now available in models that are 30 times smaller. These are starting to run real-time at the edge. In fact, they run on NVIDIA Jetson. If you wait one more click, six months, a year from now, it is highly likely that level of intelligence is going to come down to single-digit billion parameter models, and ultimately even smaller than that, which means they would be able to run on microcontrollers.
Some parts of the models would be able to run on microcontrollers, which means this frontier intelligence that was never available before is suddenly available to us. Coupling that with simulation technology, I think the next 10, 20 years, the golden age of robotics, physical AI, is upon us, and I think it's going to be a fun time for all the developers.
Fantastic, Deepu. Two thank you. First, for joining me today in the STM32 Summit, and second, and more important, for driving us forward in this industry.
My pleasure, Remi.
Take care.
Thank you.
What you have just seen with the STM32A and our work with NVIDIA on robotics are not separate stories. They are chapters in the same book. They point to the same strategy, one unified MCU platform, a platform designed to help you develop faster across evolving markets. What does this mean for you in practice? A unified MCU platform helps you scale over time. It opens the door to cross-domain reuse, where reuse actually makes sense. It shortens time to market by reducing fragmentation. It lowers your investment risk through proven technologies, ecosystem maturity, and long-term continuity, and it reduces total cost of ownership. Your teams can reuse knowledge, tools, and software across product variants and generation instead of rebuilding them each time around. Our answer to all of this is a unified MCU platform built on STM32 strength and extended with STM32A and adaptation for physical AI.
With that foundation in place, let's move to the next wave of STM32 innovation and delivering the next 20 billion STM32 devices together. Now, I will turn it over to my colleagues. Albert and Alexa will give you more details on the new products I mentioned earlier. Guys, take it away.
Thank you, Remi. Let's kick off part two with the launch of an exciting new product, the STM32G5, a new series of mission-driven, general-purpose microcontrollers. Alexa has all the details.
Albert, way back in 2019, we launched the STM32G4, a mixed-signal MCU that targeted both digital power and motor control applications. Over the past seven years, those two worlds have become increasingly complex and demanding. Power-hungry data centers are multiplying on a seemingly daily basis. Humanoid robots are becoming more sophisticated, and motor control systems more complex. The solution is the STM32G5. In a nutshell, this is our first mixed-signal Cortex-M33 and our most versatile one yet.
There are different versions for motor control and digital power, correct?
Absolutely. For a digital power application, we will have versions of the STM32G5 that offer more high-resolution timers. For motor control systems, we will build in more analog peripherals. But crucially, the G5 remains a general-purpose microcontroller, and thanks in part to its Cortex-M33 running at 280 megahertz, it represents a significant leap in performance. In fact, we are talking about a CoreMark score more than double that of the STM32G4. The new device targets the same security certifications our customers have come to expect, while also offering post-quantum cryptographic support. This is also part of our accessibility mission. That is why it comes with a timer featuring a resolution of 111 picoseconds, a new position encoder, an integrated programmable logic array with 16 logic elements, and an analog-to-digital converter supporting 5 mega samples per second compared to the previous four.
As well as all of that, we are offering a pseudo-differential mode to improve the signal-to-noise ratio.
Well, Alexa, I know one exciting potential market is for better motor control and robotics. Let's see the STM32G5 in action with our customer, Ruiyan Robots.
In humanoid robotics, one of the biggest challenges is dexterity. At Ruiyan Intelligent Control, we see the hand as one of the clearest examples of that challenge. It has to be strong enough for demanding tasks, but also agile and precise enough to handle something delicate, like glass. Finding that balance is not easy. Inside a compact palm, we need to control multiple motors, manage sensing and protection, and keep the system reliable and cool. When you are designing for something like 16 degrees of freedom, every bit of performance and integration counts. That is why we rely today on the STM32G4 microcontroller. It gives us the mixed signal capabilities we need to bring that precise control into the hand with a single microcontroller. Now, we are excited about the upcoming STM32G5. We have tested it, and it delivers exactly what humanoid robotics needs.
Up to twice the performance, support for more peripherals, and stronger built-in security. With it, we can control more motors, respond fast when speed and precision matter, and support a growing number of sensors and other functions inside the hand. That matters because the hand is doing much more than moving. It needs to sense, react in real time, and adapt. For us, the STM32G5 feels like a natural next step. Together with ST, we are helping bring dexterous robot hands closer to real-world use. From easing repetitive tasks for workers in factories to enabling more precise automation for medical laboratories, with the ultimate goal of making everyday life easier. Looking forward to our future cooperation together. Come on.
Many thanks to everyone at Ruiyan Robots for helping us make that video. Alexa, let's focus a minute on security.
Well, that remains a key concern regardless of the application. The STM32G5 offers robust built-in security to support current and upcoming compliance efforts. It targets the same SESIP Level 3 and PSA Certified Level 3 security certifications that our customers have come to expect. As well as all of those features, we are also offering post-quantum cryptographic support. That means you get familiar security features like TrustZone and Root of Trust, knowing you can also run the PQC libraries that are becoming increasingly necessary to comply with emerging regulations. Of course, we've not forgotten about our ecosystem support. The first G5 devices are available alongside Nucleo boards and discovery kits. The STM32G5 also natively supports the STM32CubeMX2 ecosystem. We're also really excited to be developing a dual-core version that will be released next year.
That is great. Where can the audience go to find out more?
The key destination is the Summit webpage, which you can find here. You can also find that link inside the YouTube description and in the comments on LinkedIn. In addition, depending on your region and location, there is an online webinar on November 17 and local STM32 Summit on Tour events around the world later this year.
Alexa, many thanks for your time today. Before you go, we have another great video showcasing the STM32G5. This time, it is from Delta Electronics.
At Delta Electronics, our mission is to deliver innovative, clean, and energy-efficient solutions for a better tomorrow. Digital power is at the heart of that mission, and it requires control platforms that can keep pace with increasingly complex systems. To keep pushing efficiency and power density, we partner with companies offering leading embedded processing solutions. A key challenge in our latest generation of high-density converters was controlling faster switching at higher frequencies without losing efficiency or reliability. We needed real-time control that could close tighter, faster loops and integrated more functions on a single chip to shrink the board. We found the answer in the STM32G5. We have tested it and found that it brings a level of performance and flexibility needed for the next generation digital power applications. Its combination of processing power, integrated peripherals, and built-in security make it a strong option for our designs.
With the STM32G5, our new converters are more efficient and significantly more compact than the previous generation. For our customers, that means lower energy losses, a smaller footprint, and higher reliability in the demanding AI and EV applications. Working with ST gave us access to a platform that supports both our current needs and our future ambitions. Together, we are helping shape the next generation of digital power solutions, smarter, more efficient, and ready for real-world scale.
Our next story is about another great product, or to be exact, the latest version of an already successful one, the STM32H7. Let me hand over the stage to Ram Kumar to provide the facts.
Thank you, Albert. Let me introduce the STM32H7P. This is our first STM32H7, combining a Cortex-M7, a DSP accelerator named Hardware Signal Processor, and an 82 GOPS NPU, also known as ST's Neural-ART Accelerator. It's our most powerful STM32H7 with flash memory, capable of reaching up to 520 megahertz and a CoreMark score of nearly 2,700, which is 150% higher than most competing solutions. We know that high-performance projects can have very different needs. This is even more true today, as digital power, digital signal processing, or edge AI may require a Cortex-M7 with innovative peripherals and accelerators. In fact, we learned a lot from the development of the STM32G5 to make this a truly high-end general-purpose device with unique capabilities. We have added high-resolution timers and a plethora of comparators in a configuration never seen before on an STM32H7.
The feedback from early customers have been overwhelmingly positive, especially the fact that the STM32H7P is roughly the same price as the 1 MB STM32H7, despite having twice the amount of flash. To achieve such positioning, we took a careful look at what our customers are creating. Many want STM32H7 with more processing capabilities to optimize their designs. We designed a mission-driven product that removes unused communication protocols, TFT controllers, or the Chrom-ART Accelerator to prioritize the digital power applications and edge AI. That's why the STM32H7P focuses on real-time performance, more memory, and innovative accelerators. Let me bring all of that potential to life. We already have alpha customers using one STM32H7P to replace multiple microcontrollers.
One of our customers had an MCU in a solar inverter that required so much edge AI and raw performance that it could manage an arc fault circuit interrupter on no more than four panels. By switching to the STM32H7P, they could run 20 panels on a single device. In short, by creating a mission-driven, general-purpose, high-performance MCU, we can now offer faster edge AI inference, deterministic real-time control, and optimized digital power. The STM32H7P also comes with exposed pads, enabling industrial mission profiles aiming up to 105 degrees Celsius. That is it from me. Back to Albert and the next chapter in this year's STM32 Summit.
At ST, we want everyone to benefit from AI. Even embedded systems developers with no experience in data science or machine learning should be able to build projects that take full advantage of our edge AI capabilities while using the latest agentic innovations. Miguel can explain more.
Absolutely, Albert. ST was one of the pioneer MCU makers in edge AI democratization. In fact, I have lived and breathed this world since 2015. We were at CES, and I remember we hosted some private demos for clients to show them our first edge AI chips. In 2019, we launched the STM32Cube.AI tools, and earlier this year, we unveiled its successor, STM32Cube AI Studio. Today, we have the largest collection of AI models of any microcontroller vendors. As Albert said, we want everyone to be part of this AI revolution, which is why we are announcing today a redesign of our current offering. Instead of Model Zoo, users will now have access to the Model Explorer, a web interface that help them browse and find the right model using filters.
Once that done, users will enter the Model Factory, a set of services and tools to help train customized models to ensure they are optimized for an application. Finally, developers can use Model Bench to benchmark models on ST boards, just like they used to do on Developer Cloud.
What about the various features from the previous versions that people are familiar with and they find useful?
No worries, they are still there. In fact, we have even expanded the user experience with more intuitive access to our models, providing new quantization tools for both ST models and third-party ones. We are also announcing Dataset Factory, a utility to help organize, manage, and clean your data. It sits next to Model Factory to make data science more intuitive and efficient. That is not all. Today, AI is also a coding companion. To address the increasing usefulness of large language models in development workflows, we are launching STM32Cube for Agentic AI. It is an agnostic solution that adds context for coding assistance. You simply install documentation, guidelines, code examples, and other resources, and the coding agent of your choice will use them to produce code that is not only more accurate, but also uses significantly fewer tokens.
You can even have multiple contexts to tailor your agent to various projects, and a code reviewer will check API calls and potential hardware errata conditions.
What do the results look like so far?
Well, in our first experiments, asking an agent to create a blinking light program required 100,000 tokens, and the application didn't even work. After using STM32Cube for Agentic AI, the coding assistant was able to create a functional application and reduce the number of tokens used by a staggering 93%. Generally speaking, we expect a token saving of around 50%-60% in typical use cases, based on STM32 configurations and hardware access. We are even helping agents select the most relevant STM32 ecosystem, such as tools and middleware, to help development operations. Similarly, ST recently released STAGI documentation, which was written with agentic development in mind. This is how ST makes AI more meaningful, so we can all move forward. We make it more accessible, and we build solutions to make your work vastly more efficient.
If you go to the summit webpage, you can register online to make sure you are kept up to date and notified about availability.
Miguel, many thanks for the update. Well, that's it from us. To close this year's STM32 Summit, let me hand it back to Remi.
Well, that wraps up this year's STM32 Summit. We have taken a deep dive into the era of convergence and seen exciting new ST innovation and products in action. You can find more on the event website, including upcoming webinars and the STM32 Summit on Tour events. For now, thanks for watching, and see you next time.