IPC International Community magazine an association member publication
Issue link: https://iconnect007.uberflip.com/i/1543955
16 I-CONNECT007 MAGAZINE I MARCH 2026 I struggle with organization, so I use AI tools to accelerate my general workflows, especially when I am hyper-focused on my work. Agentic AI can handle those mundane tasks for me, and I see the same principle applies to PCB layout. Ben, you've stated previously that using neural nets and machine learning to assist PCB layout is not new. Why is today different? Neural network concepts for routing algorithms and pathfinding date back to at least the 1970s, when computers in engineering were still new and theoretical, and not many people had the comput- ing power required to really try them. For example, even when NeuroRoute launched in 1995, it required a Pentium-class computer. Boards were mostly still through-hole. Even though surface mount technology emerged in the late 1980s, only the biggest manufacturers were using it. Training was still primitive and limited by the amount of available computing power. More of this technology was used in the semiconductor space, probably because that part of the electronics in- dustry was still using mainframes. So, what has changed? Today, two major changes have made this pos- sible. First is the availability of cheap but power- ful cloud-compute resources. Algorithms can run across multi-core CPUs and thousands of GPU cores. Second, the broad adoption of ChatGPT has contributed to a cultural mindset shift. The result is that, now, engineers in electronics, hardware, and PCB design are beginning to search for how AI will help them. Tell me specifically about the Quilter tool. Is it a standalone product or an add-on tool? Quilter is an automated place-and-route tool for printed circuit boards and nothing else. It's a standalone service that does not replace your current CAD tools. In fact, we work with the major CAD tools: Cadence Allegro, Altium Designer and Siemens Xpedition. That was another reason I was impressed with Quilter: their laser focus on solving this single challenge. Quilter will leave your pre-routed areas alone. If you have an RF or microwave section and you've already carefully routed that critical area within the board boundary, but the other components are left outside the board, Quilter will just work on those. The intelligent placement algorithms continue to improve over time. Quilter is based on reinforcement learning. Tradi- tional supervised learning would require millions of paired board designs for training, and that dataset doesn't exist, certainly not as open source. Instead, Quilter iterates on a user's design using circuit comprehension, with additional user input that is automatically detected. For example, we can detect bypass capacitors in the design using heuristics, as well as the ground and power nets, allowing the user to override or modify them if we've missed something. If you upload the entire project with all the schematics, you will see that, for example, this C-10 capacitor is directly wired to pin 12 of this microcontroller, which is on the VCC-3V3 net. Therefore, Quilter assigns that capacitor to that pin as a dedicated bypass cap, and the placer will place it as close to that pin as possible, in accordance with good design practices. The model is then punished or rewarded based on physical rule checks and physics. What is Quilter's output like? What does the de- signer get back? A designer can get multiple outputs and then choose their best option. They can be ranked " In fact, we work with the major CAD tools: Cadence Allegro, Altium Designer and Siemens Xpedition. That was another reason I was impressed with Quilter: their laser focus on solving this single challenge."