IPC International Community magazine an association member publication
Issue link: https://iconnect007.uberflip.com/i/1543307
112 I-CONNECT007 MAGAZINE I FEBRUARY 2026 A Machine-learning Approach Grounded in Design Intent This tool is built on machine learning, not large language models or generic generative AI. Its focus is on learning design intent and routing behavior from real PCB data, then applying that knowledge to new designs in a controlled, explainable way. One of the key ideas behind AIPR is hierarchical planning. Human designers do not route boards net by net in isolation. They think in terms of func- tional blocks, interconnect density, signal classes, and technology choices. AIPR mirrors this process by analyzing a design at multiple functional levels, from board-level structure down to detailed rout- ing patterns. This approach allows the AI to make decisions that reflect higher-level intent. For example, recog- nizing memory interfaces, high-speed buses, or repeated circuit blocks informs how routing space is reserved and how signals are grouped. The result is routing that looks and behaves more like human-driven layout, rather than traditional auto- router output. Beyond Traditional Autorouting Autorouting has existed for decades, but its adop- tion has been uneven. Many designers avoid it because results often require extensive cleanup or obscure the underlying design intent. AIPR addresses this gap by focusing on partial auton- omy and collaboration rather than full automation. The Smart Autorouter in CR-8000 AIPR gener- ates routing plans quickly, often in seconds, with little to no setup. More importantly, it stops when continuing would compromise design consistency. This is a deliberate design choice. The objective is not to force completion, but to give designers a high-quality starting point that reduces manual iter- ation and preserves clarity. Additional AI-assisted capabilities reinforce this collaborative model: • Pattern-aware routing reuse, where the system identifies similar circuit regions and adapts routing intelligently rather than copy- ing carelessly • Breakout and escape routing assistance for dense pin fields, generating multiple viable solutions that designers can quickly evaluate • Design knowledge capture, enabling learned routing styles and strategies to be reused across projects and teams These capabilities reflect a consistent theme. AI handles repetitive, pattern-heavy tasks while engi- neers retain control over architectural and techni- cal decisions. Learning Without Compromising IP or Trust A recurring concern with AI adoption is data ownership, so we keep learning local. Customer designs are not used to train shared or external models. Training occurs within the customer envi- ronment, and learned "brains" can be scoped by project, technology, or team. This ensures that proprietary design knowledge remains protected while still enabling internal knowledge transfer. This architecture also supports gradual adoption. Teams can start with Zuken-trained baseline intelli- gence and progress toward customer-trained and adaptive models as confidence grows. AI as an Engineering Multiplier In a pragmatic approach, AI is most effective when it augments human expertise, rather than attempt- ing to replace it. In PCB design, this means accel- erating planning, reducing repetitive effort, and helping less-experienced designers benefit from established best practices, without removing accountability or transparency from the process. Our work reflects that philosophy, treating AI as a design partner that applies learned experience consistently and quickly, while leaving judgment and responsibility with the engineer. As AI adoption in CAD tools expands, the distinc- tion between novelty and value will matter. In PCB design, value comes from respecting the disci- pline's constraints, complexity, and need for preci- sion. AIPR demonstrates that when AI is applied thoughtfully, it can enhance productivity and design quality without compromising engi- neering rigor. That's practical value. I-CONNECT007 Steve Watt is a PCB engi- neering manager at Zuken.