Companies Showcase Edge AI Tech at Embedded World

- FPGAs offer significant advantages over GPUs for edge AI in aerospace due to their lower and more predictable latency, as they process data as it arrives rather than in batches. This deterministic performance is critical for real-time applications like sensor fusion and autonomous navigation. While requiring specialized programming knowledge, FPGAs provide greater power efficiency and the flexibility to be reconfigured for evolving AI workloads. - The certification of AI and machine learning software for avionics under the DO-178C standard presents a major hurdle, primarily due to the challenge of proving determinism in AI/ML systems. Current regulations for the highest safety-critical levels (DAL-A) do not permit AI/ML where logic decisions change in real-time. Companies are actively working with regulatory bodies like the FAA and EASA to establish acceptable means of compliance for ML-based systems. - Semtech's integration of LoRa connectivity with edge AI enables long-range, low-power communication for sensor data, which is crucial for applications like predictive maintenance in industrial IoT. By processing data locally on the device, this combination reduces network bandwidth requirements, lowers cloud processing costs, and extends the battery life of remote sensors. - Infineon's microcontroller portfolio, including the PSOC™ and AURIX™ families, provides the real-time processing capabilities, safety features, and security needed for edge AI in robotics and automotive systems. Their solutions are designed to support the shift towards software-defined vehicles, enabling features like over-the-air (OTA) updates and zonal E/E architectures. - Model-Based Systems Engineering (MBSE) is increasingly being integrated with AI to manage the complexity of designing AI-based systems. AI-driven MBSE tools can automate the generation of system models from natural language requirements, improve traceability, and automatically verify model consistency, which helps to accelerate the development and certification process. - Alp Lab's System-on-Modules (SoMs) are designed to be hardware-agnostic, allowing developers to use a unified SDK across different silicon vendors. This approach aims to reduce vendor lock-in and streamline the development and long-term maintenance of edge AI applications in industrial and robotic fields. - The trend in resource-constrained edge AI is moving towards "TinyML," which focuses on running machine learning models on microcontrollers with extremely limited memory (kilobytes) and power (milliwatts). This requires significant model optimization techniques, such as quantization (using lower-precision integers instead of floating-point numbers) and pruning to reduce model size and computational demand. - For aerospace applications, edge AI hardware must be ruggedized to withstand harsh environments. NVIDIA's Jetson AGX Orin and Xavier NX are examples of platforms that combine GPU acceleration with deterministic networking for sensor fusion and autonomous navigation in defense and aerospace.