SANTA CLARA, CA – What role will artificial intelligence truly play in electronics design, and what will the impact be on hardware engineers?
Zuken took a step toward answering that question today with its announcement at PCB West of a new AI-based tool for printed circuit place-and-route. Yet the first public mention of AIPR for CR-8000 – the actual rollout will come in the first quarter next year – poses not only a dramatic vision for a highly automated future of design but a host of new questions as well.
The new tool itself is an extension of Design Force, Zuken’s layout, routing and verification tool within the CR-8000 platform. It’s AI, explained Kyle Miller, Ph.D., who architected the engine, involves all three basic types of machine learning: supervised, unsupervised and reinforcement. AIPR stands for Autonomous Intelligent Place and Route, and like previously announced AI-based CAD tools, it starts with routing. The “Basic Brain” performs so-called smart routing by means of exposure to Zuken’s database of PCB designs built in CR-8000. Over time, it mimics human routing, with channels organized in logical ways. Smart placement is next, at an undisclosed time.
According to Bob Potock, vice president of sales and marketing, Zuken will add IPC-2581 capability as part of the next-generation Dynamic Brain, allowing designs from other ECAD systems to be incorporated and learned.
The first two stages are working up to Autonomous Brain, a goal-based utility that the product designers, including Miller, say will use text-based inferencing whereby it detects descriptions of different parts of boards. According to Miller, four functionality levels will be used to inform local and planning decisions.
The system, notes Steve Watt, manager of PCB engineer, can learn from both good and bad designs. “The brain can be untaught if it is sent a dirty design,” he said. Zuken has tested it on about 100 designs, most of the high-speed, digital variety.
Adds Miller: “PCBs are complex. They involve numerical data, geometrical data, the layers in the board, text, constraints … Autonomous Brain is multi-modal; it combines all of these data and extracts the designer’s intent.”
With designers in high demand due to the aging out of many veterans and the length of time and amount of knowledge it takes to develop expertise, some of the concerns about AI replacing humans are eased. But can AI-based tools be realistically used in anything but local environments? Zuken is still working through the issues of cloud-based system, as users point to security concerns. (Miller will be among a group of experts tackling this issue on a free panel session titled AI in Electronics: What Can We Expect? on Sept. 20 at PCB West.)
And how are time-based licenses affected? Miller indicated it takes about five days to learn to use AIPR. But once mastered, Zuken tests showed it eliminated autorouter setup time, and cut autorouting time to 30 sec. from 15 min. Potock noted the conundrum of issuing licenses for products that, on paper, reduce the time of use from hours or days to mere seconds. At this time, it appears Zuken will make AIPR available as a perpetual license.
Given the broad industry resistance to using autorouters, it remains to be seen how tools like AIPR (and others, which are coming right behind) will be integrated into general industry use. That said, the trend in board design is away from the traditional dedicated specialist, toward layout and placement being a small function of engineers’ overall responsibilities. That shift may finally tilt the field toward automation, and if Zuken’s vision is correct, even near-complete abdication to the machine.
Zuken will exhibit AIPR at the PCB West exhibition on Sept. 20 (show hours are 10 a.m. to 6 p.m.) at the Santa Clara (CA) Convention Center. A webinar is planned for November.