Simplifying PCB Decision-Making with AI

AI integration in design and simulation software packages is becoming the norm. In fact, it’s getting harder to find a CAD or FEA (Finite Element Analysis) vendor who is not adding AI-powered functions into its flagship product. A similar movement is also taking place in printed circuit board (PCB) design tools, but because of its under-the-surface, incremental nature, it may be less obvious. In this article, we set out to identify where and how AI is changing PCB design.

Parallel Developments

Dave Wiens, product marketing manager, Siemens, outlines how AI-powered tools work in Siemens’ NX and Solid Edge (for mechanical CAD, CAM, CAE) and Xpedition (electronic design). “For instance, with our Xpedition 2409 release in September 2024, we rolled out a predictive command tool, which we leveraged directly from the NX and Solid Edge teams,” Wiens reveals.

In both CAD and electronic design products, natural language processing has also been added to improve user experience. “Here, the domain [mechanical or electronic] is not critical. It’s just the ability to process natural language. The standard technology is available in third-party tools like AWS and others,” says Wiens. In Siemens’ PartQuest and Connect, both parts of the company’s electronic design automation (EDA) offerings, large language model (LLM) integration resulted in chatbots that understand natural language.

In Siemens’ Xpedition and HyperLynx (solver for simulating signal and power integrity), AI is behind the software’s predictive commands. “The predictive command uses machine learning to follow user actions and generate a domain- and tool-specific process for future actions,” explains Wiens.

“Compared to mechanical software, PCB software may be lagging a bit behind in natural language support,” Seth Hindman, director of product management and strategy, Autodesk, notes. “That’s probably because, as an industry, the PCB sector is still trying to figure out where to place its bet with AI. At Autodesk, we’re trying to have a unified UI (user interface) so you can query both mechanical and electrical design with natural language.”

Autodesk’s integration of natural language manifests in the evolution of Autodesk Assistant, available throughout its software portfolio, including Autodesk Fusion with PCB design tools. “Our approach to AI is to behave like a companion,” explains Hindman. “It would ask you simple questions, like, ‘Do you want to go left or right?’ But if we ask enough of these small questions along the way, we can automate 80% of the work for you.”

“One challenge is the pace of development of AI,” Dr. Kyle Miller, research and product manager, Zuken, says. “The pace of development in the PCB space has typically been much slower. Replacements for the transformers—what the ‘T’ in ChatGPT stands for—are already being proposed and evaluated. Newer and more efficient training methods have led to the release of open-source LLMs like DeepSeek-R1, which most people have probably heard of. This model shows immense promise as an assistant for many different applications, using much less compute resources than ChatGPT.”

Mechanical and Electrical Connections

In 2023, Autodesk struck a partnership with Cadence, a leading PCB software developer. The agreement made it possible to easily exchange design data between Autodesk Fusion and Cadence Allegro X or OrCAD X. In 2016, Autodesk acquired CadSoft, which developed EAGLE, an EDA package with PCB design functions.

“We maintain how PCB designs are described in EAGLE, but we’ve also embedded it in Fusion so it can automatically generate a 3D model of the board. That makes it easy for mechanical and electrical engineers to collaborate to make sure the PCB design fits into the product’s enclosure,” explains Hindman. “Collaboration between mechanical and electrical designers is squarely in our line of sight. That’s part of our strategy,” Hindman says.

Since Siemens offers software packages for both mechanical and electronic design, Wiens says, “It’s a natural function that they work well together, because they’re under the same roof, and they don’t need to go through other protocols to exchange data.” The streamlined connection can be seen in the integration of NX and Xpedition, Wiens points out.

Automated Routing

Automated routing in PCB boards began showing up in PCB design packages even before AI became a buzzword. There’s still an ongoing debate as to whether such tools are truly AI-driven or simply rule-based automation. Either way, they have eliminated tedious manual work and sped up the PCB design process significantly.

In 2023, at PCB West conference in Santa Clara, CA, Zuken announced its AI strategy. (For more, read “Zuken Launches AI-Powered PCB Routing Tool, as First Step in Long-Term AI Strategy,” September 2023.) The first incarnation of the strategy was Zuken’s Autonomous Intelligent Place and Route (AIPR), powered by AI.

“Things like design setup, reuse and adaptation can be applied in a way that allows for personal or company style, rules and preferred trade-offs,” Miller says. “AI can be used to fill in missing information that is crucial to perform PCB layout correctly. LLMs used in an agentic flow can supply a deeper understanding of the circuit, for example, by predicting signaling technologies or IC [integrated circuit] functionality, enabling better layout.”

However, the miniaturization of products and the need to integrate electronics into organic-shape Internet of Things devices are starting to test the automation algorithms. “People are having to put PCBs into products that have never had electronics before. They’re finding that the rules of thumb they’d relied on historically are starting to fail,” Hindman notes.

Classic PCB boards are rectangular and rigid in design; however, many newer boards are designed to be flexible, to accommodate the space constraints in the newer devices that house them. The latest PCB designs are multilayered, better visualized in 3D models than 2D schematics. Training data for optimal layout in rigid, flat boards are easier to obtain, but data for flexible boards are much harder to find, Wiens notes. Competition among rivals makes these board designs closely guarded secrets, locking them out of the common AI or machine learning training pool.

“If you train your AI using the layouts in rigid, flat boards, you can’t apply that AI algorithm to optimize the flexible board,” Wiens points out. “That means larger companies have an advantage over the startups, because they have a greater volume of data for training.”

Startups might turn to open-source data for AI development, but that may not help much in dealing with newer PCB designs. “Quite often, buyers beware. The data might work on a limited class of simple design, but not on complex board setups,” says Wiens.

“The challenges in applying AI to the PCB domain are unique due to the mixture of different data types used and the requirement for both accuracy and precision,” Miller notes. “Zuken has multiple dedicated AI teams working across the world, integrating AI into many aspects of the design process, from schematics, PCB, simulation to support.”

Simplifying Decision-Making

In mechanical design simulation, AI opens the possibility to use reduced order models (ROMs) or surrogate models to bypass the time-consuming physics-based simulation. A similar development can be seen in electrical design tools.

“The Siemens HyperLynx solver [for simulating electromagnetics, PCBs and IC packages] uses the HEEDS engine—the one powering Siemens Simcenter to explore design spaces,” Wiens says. “Let’s say you have eight variables, such as materials, spaces and properties, and you want to run a sweep of 10 permutations for each. You’d end up with 24 million permutations. Even if the simulation is not very complex, it still takes you about 10 minutes each, so you’re looking at years. So it’s not realistic. But with the HEEDS engine, we can reduce that workload to about overnight.”

“You can expect to see the industry investing in solutions that can make recommendations based on your requirements,” predicts Hindman. “Because the [AI-powered] Assistant has a broader view of the system and can tap into more experiences, it might suggest, for example, don’t use this resistor; use this one instead. It could not only flag components that are no longer available but also suggest replacements. Then there are lots of mundane, repetitive tasks PCB designers have to do, like having to manually create component assets to populate the libraries. You could expect to see AI creating these automatically off a data sheet.”

Many people might view them as simple automation, but Hindman thinks it’s laying the groundwork for “objective-based design.” Human PCB designers might be able to perform trade-offs based on one or two objectives at a time (cost and material choices, for example), but AI, with its superior memory and computing capacity, can consider far more parameters and objectives in tandem and offer a design solution that can satisfy all objectives.

Wiens sometimes sees AI creeping into conversations where it doesn’t belong. “Often, when a customer shows me a problem they want to apply AI to, I tell them, ‘You know, you don’t need AI to solve that.’ Sometimes the best solution to your problem is not AI.”

More Autodesk Coverage

More Siemens Digital Industries Software Coverage

More Zuken Coverage