Elektor at embedded world North America 2026: Day 2 Update
September 24, 2026
on
on
Elektor continued exploring embedded world North America 2026 in Anaheim on Wednesday, September 23. After a first day focused on technologies ranging from FPGAs and RF test to quantum-secure communications, Day 2 brought us deeper into edge AI, robotics, motor control, embedded security, smart-city applications, and the tools engineers use to build and maintain increasingly complex systems.
If there was one theme connecting many of our conversations today, it was the continued movement of intelligence toward the edge. AI inference, speech recognition, computer vision, and other capabilities that once required powerful computers or cloud services are increasingly finding their way directly into embedded devices.
At the same time, the development environment surrounding those devices is evolving. Cross-platform development, remote updates, security, and long-term device management are becoming as important as the hardware itself.
Here are some of the technologies and demonstrations that caught our attention on Day 2.
The platform also provides integration with Visual Studio Code and its extension ecosystem, giving development teams more flexibility in how they build their workflows. This is particularly significant for professional embedded teams working on compliance- and safety-critical applications, where developers increasingly expect modern development workflows without giving up specialized embedded tooling.
Built on the Arm v8-M architecture, the GD32F503 series operates at frequencies up to 252 MHz and combines DSP hardware acceleration with a single-precision floating-point unit. The additional processing capability targets compute-intensive, real-time motor-control applications where precise control and fast response are critical.
We also looked at the GD32H7, which GigaDevice is positioning for embodied AI and robotics. The platform features a 600 MHz Cortex-M7 with DSP, FPU, TMU, and FAC capabilities for real-time workloads including vision, SLAM, and AI/ML inference. GigaDevice also highlights interfaces and features intended for robotics, including multiple camera support, ToF LiDAR connectivity, motor-control interfaces, and multimodal human-robot interaction.
It was an interesting example of how high-performance MCUs are moving beyond traditional control applications and into systems where sensing, motion, and local intelligence need to operate together in real time.
Importantly, the system also incorporates liveness detection, helping distinguish a real person from attempts to fool a recognition system with an image or other representation. The demo provided another practical example of the growing ability to perform computer-vision workloads directly on embedded hardware while keeping sensitive data at the edge.
The eBOX630B is a fanless embedded edge-AI computer supporting Intel Core Ultra 7/5 processors with integrated CPU, GPU, and NPU resources. The system can connect to multiple IP cameras while performing AI inference locally, with edge-to-cloud integration available when needed.
The demonstration illustrated an important direction for practical AI deployment: instead of streaming every video feed to a data center, significant analysis can take place close to where the cameras and sensors are operating.
Torizon brings together hardware, an operating system, development tools, remote updates, and device monitoring into an integrated environment. The objective is to make embedded development more iterative while providing infrastructure for managing deployed products over their operating lifetime.
That lifecycle perspective is increasingly important. For many connected products, shipping the hardware is no longer the end of development. Devices may require security patches, software updates, monitoring, and new functionality for years after deployment. Torizon is designed to provide a scalable approach to those requirements while emphasizing reliability and security.
According to the demonstration material, DEEPCRAFT STT supports models ranging from approximately 6 million to 32 million parameters, with configurable latency and memory requirements depending on the implementation. Infineon specifies memory requirements starting at approximately 4.4 MB ROM and under 3 MB RAM, with target active power ranging from roughly 20 mW to 65 mW.
Those numbers are particularly interesting because they demonstrate how capabilities normally associated with considerably larger computing platforms are moving into the MCU domain. Local speech processing can also reduce dependence on cloud connectivity and enable voice interfaces where latency, privacy, or power consumption matters.
A robotics MCU can combine motor control, cameras, LiDAR, and AI inference. A security system can recognize a face and determine whether it is looking at a real person without sending the image to the cloud. An embedded computer can analyze multiple camera streams for a smart city. And even speech-to-text is moving onto MCU-class hardware.
But more capable devices also create more complex software and lifecycle requirements. That is why the development platforms we saw from companies such as IAR Systems a key part of the story.
The embedded AI era isn't just about putting AI into a product. It's about giving engineers the hardware, software, security, development tools, and infrastructure to build intelligent products that can operate reliably in the real world.
Stay tuned for more updates!
If there was one theme connecting many of our conversations today, it was the continued movement of intelligence toward the edge. AI inference, speech recognition, computer vision, and other capabilities that once required powerful computers or cloud services are increasingly finding their way directly into embedded devices.
At the same time, the development environment surrounding those devices is evolving. Cross-platform development, remote updates, security, and long-term device management are becoming as important as the hardware itself.
Here are some of the technologies and demonstrations that caught our attention on Day 2.
Embedded Development Goes Cross-Platform
Shawn Prestridge of IAR Systems gave us a technical demonstration of the latest IAR Platform, including the company's cross-platform IDE. IAR now offers native Linux support alongside Windows for its embedded development environment. Built using the Qt Framework, the IDE is intended to provide a consistent development experience across operating systems while retaining the compiler, debugging, analysis, and safety-oriented capabilities associated with IAR's toolchain.The platform also provides integration with Visual Studio Code and its extension ecosystem, giving development teams more flexibility in how they build their workflows. This is particularly significant for professional embedded teams working on compliance- and safety-critical applications, where developers increasingly expect modern development workflows without giving up specialized embedded tooling.
Motor Control and MCUs for Embodied AI
Lon Howze, Sr. FAE Manager, Americas at GigaDevice, walked us through several of the company's latest MCU technologies, including the GD32F503 for advanced motor control.Built on the Arm v8-M architecture, the GD32F503 series operates at frequencies up to 252 MHz and combines DSP hardware acceleration with a single-precision floating-point unit. The additional processing capability targets compute-intensive, real-time motor-control applications where precise control and fast response are critical.
We also looked at the GD32H7, which GigaDevice is positioning for embodied AI and robotics. The platform features a 600 MHz Cortex-M7 with DSP, FPU, TMU, and FAC capabilities for real-time workloads including vision, SLAM, and AI/ML inference. GigaDevice also highlights interfaces and features intended for robotics, including multiple camera support, ToF LiDAR connectivity, motor-control interfaces, and multimodal human-robot interaction.
It was an interesting example of how high-performance MCUs are moving beyond traditional control applications and into systems where sensing, motion, and local intelligence need to operate together in real time.
Face Recognition with Liveness Detection at the Edge
At Microchip Technology, Kedar Remane demonstrated a face-recognition system with liveness detection. Facial recognition is useful for applications such as access control and automation, but sending images to the cloud for processing can introduce latency, connectivity requirements, privacy concerns, and ongoing service costs. The demonstration instead showed an edge-based approach that performs recognition locally.Importantly, the system also incorporates liveness detection, helping distinguish a real person from attempts to fool a recognition system with an image or other representation. The demo provided another practical example of the growing ability to perform computer-vision workloads directly on embedded hardware while keeping sensitive data at the edge.
Multi-Camera Edge AI for Smart Cities
Jody Cheng of Axiomtek showed us how edge AI can scale beyond a single camera. The company's demonstration featured MicroIP AI Analytics running on Axiomtek's eBOX630B, providing real-time, multi-camera video analytics for smart-city applications. Potential uses include traffic monitoring and enforcement, critical-zone intrusion monitoring, pedestrian safety, and real-time hazard detection.The eBOX630B is a fanless embedded edge-AI computer supporting Intel Core Ultra 7/5 processors with integrated CPU, GPU, and NPU resources. The system can connect to multiple IP cameras while performing AI inference locally, with edge-to-cloud integration available when needed.
The demonstration illustrated an important direction for practical AI deployment: instead of streaming every video feed to a data center, significant analysis can take place close to where the cameras and sensors are operating.
Torizon and the Software Lifecycle of Embedded Products
Not everything we saw was about adding more processing power. Dan Egner of Toradex gave us a detailed overview of Torizon, the company's Linux-based platform for developing and maintaining embedded products.Torizon brings together hardware, an operating system, development tools, remote updates, and device monitoring into an integrated environment. The objective is to make embedded development more iterative while providing infrastructure for managing deployed products over their operating lifetime.
That lifecycle perspective is increasingly important. For many connected products, shipping the hardware is no longer the end of development. Devices may require security patches, software updates, monitoring, and new functionality for years after deployment. Torizon is designed to provide a scalable approach to those requirements while emphasizing reliability and security.
Speech-to-Text Moves onto the MCU
Niall Lyons of Infineon introduced us to several technologies, including DEEPCRAFT Speech-to-Text (STT) running on PSOC Edge. The machine-learning-based software brings speech transcription directly to supported PSOC Edge E83 and E84 MCUs, targeting applications including voice interfaces, captioning, and speech understanding.According to the demonstration material, DEEPCRAFT STT supports models ranging from approximately 6 million to 32 million parameters, with configurable latency and memory requirements depending on the implementation. Infineon specifies memory requirements starting at approximately 4.4 MB ROM and under 3 MB RAM, with target active power ranging from roughly 20 mW to 65 mW.
Those numbers are particularly interesting because they demonstrate how capabilities normally associated with considerably larger computing platforms are moving into the MCU domain. Local speech processing can also reduce dependence on cloud connectivity and enable voice interfaces where latency, privacy, or power consumption matters.
Intelligence Is Moving Closer to the Hardware
After two days at embedded world North America, a broader picture is starting to emerge. On the first say, we saw technologies spanning programmable logic, RF characterization, embedded data collection, post-quantum security, debugging, and sensing. Today, many of the demonstrations showed what happens when those underlying capabilities are combined with significantly more processing at the edge.A robotics MCU can combine motor control, cameras, LiDAR, and AI inference. A security system can recognize a face and determine whether it is looking at a real person without sending the image to the cloud. An embedded computer can analyze multiple camera streams for a smart city. And even speech-to-text is moving onto MCU-class hardware.
But more capable devices also create more complex software and lifecycle requirements. That is why the development platforms we saw from companies such as IAR Systems a key part of the story.
The embedded AI era isn't just about putting AI into a product. It's about giving engineers the hardware, software, security, development tools, and infrastructure to build intelligent products that can operate reliably in the real world.
Stay tuned for more updates!
Read full article
Hide full article
Discussion (0 comments)