IPC International Community magazine an association member publication
Issue link: https://iconnect007.uberflip.com/i/1543307
98 I-CONNECT007 MAGAZINE I FEBRUARY 2026 searching through a specific set of documents. These models are constantly being updated, and they must be selected based on the target task. Model management can be painful. Other models must be trained on actual design data, which requires a deep bench of pre-existing designs that must be categorized appropriately. An old through-hole design isn't much good for train- ing to do ultra HDI, power supply designs won't help for high-speed digital, and ITAR data can't be used for consumer designs. IP Security First, decide what constitutes IP because it's not just finished design data. What about the constraints used, or the process in which it was developed? Industry efforts to build large design libraries for model training have failed due to IP concerns. Most AI engines are running in the cloud to achieve economies of scale and performance. Many engineering teams have a problem transmit- ting design data to the cloud, not knowing how the data will be used. For example, will my designs be used to train a model my competitors can use? As EDA vendors, we have strict constraints around use of customer data for that very reason. The same is true with service bureaus. Verifiable and Accurate Results Accuracy is critical to engineering, so engineers must understand how AI results will be verified. Getting different results every time an AI engine is utilized reduces trust. Taking longer to review and verify results than to design manually is unac- ceptable, as are results that don't meet critical performance and manufacturability requirements. Engineers have learned to trust algorithmic auto- mation for tasks like batch DRCs or manufacturing outputs. The same learning curve will be neces- sary for AI-based automation. Adoption Resistance Organizational and individual inertia make us all resistant to change. We have to clearly understand the value and the risks. In this regard, AI is no different, but the hype around the potential of AI to replace large swaths of the engineering process or even engineering teams doesn't encourage its adoption. The industry will also need to define appropriate compensation for an engineer who trains AI for tasks that may replace them. Siemens' Approach to AI in Electronic Systems Design Siemens has experience deploying AI applications at scale, with a bench of over 1,000 AI develop- ers and the ability to leverage applications across disparate domains, supported by an underlying architecture that enables multiple applications to share common development tools and design data. We also bring a strong foundation for AI, with data management to categorize designs, and veri- fication engines that can be deployed to automati- cally review AI-generated results. Siemens design tools have deployed AI to: • Enable design space exploration to optimize designs for performance • Assist engineers new to a product with natural language-based documentation assistance • Predict next tasks in a design process based on an engineer's prior activity • Find information in component datasheets quickly • Predict component supply chain issues with availability and cost • Optimize selection of decoupling capacitors Conclusion While AI promises a transformative future for PCB design, its current application is most effec- tive in optimizing smaller, specific tasks. Over- coming challenges like model training, IP security, ensuring verifiable results, and addressing adop- tion resistance are crucial for AI's broader integra- tion. Siemens is actively deploying AI to enhance various aspects of electronic systems design, demonstrating a commitment to leveraging AI as a powerful tool to augment human expertise and drive innova- tion. I-CONNECT007 David Wiens is a product marketing manager at Siemens.