Earlier this year, Cadence Design Systems unveiled its AI-assisted PCB design ambitions. The first phase of the transition to machine learning in this space will be a rollout to a small group of early access customers.
They use Allegro X AI software to place components on the layout, route critical nets, and perform signal and power integrity checks on partial layouts before a human engineer completes the job. I expect it to run.
This announcement marks another step in advancing machine learning in electronic design automation (EDA) in both chip and system design.
In the chip design arena, vendors are finding machine learning to be a useful tool for speeding up certain types of analysis, but it can take over tasks that previously required human intervention. There are only a few entries into the tool. The most common use cases include techniques used to speed up simulations or to determine which formal verification algorithms are best to introduce for a given circuit. However, PCB design is seeing a small influx of start-ups with ambitions to automate more processes.
In developing its AI-based products, DeepPCB has taken slogans from chip design open source projects initiated by DARPA at the end of the last decade. Andrew Kahn, professor of electrical engineering at the University of California, San Diego (UCSD), used his OpenROAD tools developed as part of a development plan by his group and others to explain the ultimate purpose of his suite. used the slogan “No Humans Involved”. DARPA Efforts. The rationale driving OpenROAD is to be able to take a design in the form of RTL and generate a production-ready GDS II without direct intervention, allowing it to run within a computation time of the order of a day. That was it. Similarly, DeepPCB claims that its cloud-hosted system, currently in beta, can deliver complex boards within his 24 hours.
The free beta release is for up to 150 components and two layers, but DeepPCB plans a commercial version of the tool that can handle larger designs once the beta phase is over, but enthusiasts We will also maintain simpler free options for supporting the maker community.
Cadence, on the other hand, is more focused on more complex designs, which is one reason why the first release only runs partial routes. The export from AI mode is a regular Allegro file, so the designer can use standard delete and replace procedures if the automated selection later proves to be problematic. Thorgat Sen, vice president of research and development at Cadence, said the models used in X AI take into account expected congestion to avoid stymiing human designers. I’m here.
creative partner
Cadence is focused on more complex designs, and that’s where AI-assisted demand comes in. “We believe that PCB design is becoming more and more complex as IC pinouts are getting much larger. Doing this in a few hours is a big change,” says Sen.
AI-based tools can come up with more efficient and unexpected placement options, as we saw in another project conducted by Cadence and Google in machine learning-powered placement and routing of ICs. “PCB designers can gain insight from tools they might have never thought of before. They can be creative partners in helping design,” says Sen.
Constraints are important despite the ability to come up with new layouts, as demonstrated by some users of online AI tools. For example, without a constraint that favors path length, the tool might choose to avoid vias by creating routes around the obstructing paths. This changes the way the tool presents options to the user, as routing is no longer interactive.
“This is similar to digital IC design, where constraints need to be defined. But anyone who has used Allegro X is already familiar with our approach to constraints,” Sen said. increase. “The long-term intention is to enable end-to-end locations and routes.”
A white paper published by Siemens mentions the possibility of an end-to-end design, but the emphasis is on the company working on its own machine learning algorithms to provide interactive assistance.
“Our approach is to leverage AI and machine learning algorithms to empower electronic system engineers,” said David Wiens, product manager for the Xpedition tool line at Siemens EDA. “Therefore, we are essentially focused on optimizing existing design processes rather than auto-generating ‘moonshots’ for entire electronic products.”
Siemens has explored several use cases for machine learning in PCB design that can speed up design without removing human engineers from the primary task of layout and routing. One example is the creation of library models. This involves taking details from datasheets and creating valid symbols and attributes that layout tools can use to ensure proper clock, power, and ground routing and compatibility in the final design. often Proposed layout.
Another possibility that highlights a similar issue with data collection lies in the automation of constraints used to control at least partially automated layout and routing. Instead of requiring engineers to manually re-enter design rules and recommendations for different materials and manufacturing environments, AI tools could potentially parse the information from the guidelines themselves and create constraint directives.
Zuken plans to present some of its own research results on machine learning for automatic routing support at a series of seminars in late spring. The company also said it is working on a format for engineering autocomplete. With this feature, the software learns what steps operators often follow in sequence and displays the most likely commands in menus or hotkeys, reducing the amount of input required. Similarly, future tools will analyze previous designs to create design rules for future layouts or to determine optimal distances and routing methods between components for frequently reused types of circuits. There is a possibility.
The use of AI may also enable more proactive work to determine how changes in materials, manufacturing techniques and components will impact the cost and reliability of the final design. . Sen said one of his potentials for the Allegro X tool is to take advantage of the speedup the tool offers to support more what-if analysis before teams choose a particular approach. It is said that
Celus, one of the AI startups entering the space, is trying to avoid being labeled an EDA tool for its software. Instead, it was designed to be inserted in phases before his EDA tools, such as the Allegro, CR-8000 and Xpedition, are normally invoked. The idea behind Celus’ AI-based tools is to take functional requirements and create schematics, bills of materials, and basic floorplans that lead to designs that work faster.
Machine learning is still in its infancy in PCB design, but it looks like it will become part of the design process. The big question is how high will it ultimately be pushed? Will you end up taking over most of the work, or will you be stranded on a problem that offers insight but can’t keep humans out of the loop entirely?
