Embedded system technologies are the centerpiece for intelligent decision-making as data gets closer to the source. Edge AI uses efficient microcontrollers, AI-optimized processors, and other compute platforms with accelerators for low-power inference. Chiplets and modular SoCs are changing what performance per watt means, delivering flexible computing that is optimized for edge applications. In the November/December issue , we look at the latest technology advances and trends that have a significant impact on edge computing, with a focus on edge AI, physical AI, and agentic AI. Contributors tackle issues such as physical edge computing, embedded edge AI system design, energy consumption, agentic AI in software development, development platforms, microcontrollers, and chiplets. Contributing writer Filippo Di Giovanni makes a case for physical edge computing as the next frontier for AI, bringing intelligence to the physical world. The benefits: real-time local response, lower energy use, improved privacy, and keeping systems running when cloud links fail. Di Giovanni also addresses engineering hurdles such as data center AI hardware that cannot just be ported to a sensor node powered by a tiny battery and why it requires specialized AI hardware. One of the biggest bottlenecks in edge computing is energy consumption , contributing writer Stefano Lovati said. He reports that energy consumption challenges can be solved with model optimization, specialized silicon architectures, and smarter thermal management. Lovati also explains why more tera operations per second is not the answer to the problem. While it is the most used metric to evaluate edge AI system performance, it does not fully adapt to battery-powered devices operating within strict thermal design power constraints, he said. Instead, inference per joule is more appropriate, measuring how many AI predictions or tasks a system can complete for each unit of electrical energy consumed. Higher values of inference per joule translates into better efficiency, thus lowering operating costs and cooling requirements in edge devices. Giordana Francesca Brescia covers the fundamentals of designing edge AI for real-time systems . It isn't just a faster chip, she said; it's the system architecture, including sensors, buffers, memory, compute, accelerators, and task management, all tuned for latency. She also discusses a common cause of latency, and it isn't found in the AI model but in the operations that precede it. Low latency in edge AI is not achieved simply by choosing a faster processor or smaller model; it is the result of broader architectural decisions. Brescia also looks at chiplets and modular SoCs and how they are redefining performance per watt by balancing CPU, accelerators, memory, and I/O while managing data movement, latency, and heat. Modularity offers greater design flexibility, but it also requires careful analysis of data transfers, latency, memory, and thermal dissipation, she said. Brescia also explores how chiplets can improve performance per watt through specialization. Contributing writer Saumitra Jagdale reports that streamlined development platforms are critical to edge AI deployment. To accelerate model optimization, heterogeneous hardware, and security hardening with real-time performance constraints, silicon vendors must optimize their development workflows, he said. Jagdale reviews several development platforms from leading silicon vendors. Software development agents connect those fragmented workflows, contributing writer Abhishek Jadhav reports. He explores how agentic AI is accelerating edge AI software development . Rather than only generating the code, a development agent works with compilers, test frameworks, profilers, debuggers, documentation, and other engineering tools to shorten the development cycle that previously required manual labor. Download (Please click PDF icon to download.) Cover image: Adobe AI Generated The post ESD eBook Novembe