Were you unable to attend Transform 2022? Check out all of the summit sessions in our on-demand library now! Watch here.
Celus, a provider of AI-powered automation software for electronics engineering, is looking to improve tools that help design printed circuit boards. The company has also pioneered an electronic part search engine that transforms how engineers select electronic components in the multi-trillion-dollar electronics industry.
Electronic designers typically compare target specifications via a manual process of searching across digital product catalogs filled with tables of part specifications. Celus’ approach uses a digital twin representing design choices to automatically identify appropriate options and then design the circuitry around the component. A more automated approach could have as transformative an impact on the way engineers find products as the introduction of new search ranking techniques by Google had on the evolution of how consumers find knowledge.
Another significant innovation the company is working on includes tools to help automate how the individual components are laid out on a printed circuit board. Celus plans to combine machine learning and classic optimizers to improve the layout, reduce the size and cost, or achieve other design goals.
Engineers can essentially recompile the design to take advantage of newer alternatives or if a component goes out of stock due to a chip shortage.
“It is like a library update in software design,” Celus CEO Markus Pohl told VentureBeat.
Augmenting a complex process
Engineers like Markus Klausner, CTO at Viessmann Climate Solutions, a Celsus customer, used to spend countless hours matching design specs with parts to design the control systems for heat pumps, refrigerators, and air conditioners. He said that even though the electronics catalogs are getting better at organizing information, he still spends a lot of time matching new design requirements with part specifications.
“Celus automates this process by linking up with the distributor, so you don’t have to do the job manually,” Klausner told VentureBeat.
Since Celus also automates the design process, Klausner finds that its tools can currently create a design at the same level as an inexperienced engineer, but even still, it helps save time. Celus’ tools, he noted, are also getting better all the time, and he expects them to catch up with senior designers in the future.
Erick Brethenoux, Gartner distinguished VP analyst, told VentureBeat that new techniques for automating PCB design like this are an example of the recent progress in how generative AI is helping to design chips, buildings, security systems and other types of products and systems. This involves a combination of AI and traditional engineering techniques to optimize different aspects of the design process.
“Generative AI is a collaborative process between humans and machines to come up with better designs,” Brethenoux said. These tools help drive efficiency because they can come up with new ideas and also help optimize for different requirements such as cost, size, or performance.
Connecting between chips
Pohl said he hopes Celus’ tools will bring efficiency to the $273 billion enterprises spent on electronic component sales in 2021 according to Grand View Research estimates. It could also streamline some of the $1.3 trillion Grand View estimates enterprises spent on engineering outsourcing in 2021 that included some electronics engineering. A more efficient electronics design process will also allow physical industries like car companies to design new product variations to take advantage of new chips or avoid supply chain shortages.
Electronic design is an old idea, but has traditionally focused on improving how engineers lay out circuits in silicon chips. Electronic design automation leaders include companies like Cadence Design, Synopsis, and Siemens EDA (Formerly Mentor Graphics). These tools focused on digital logic early on and gradually added support for analog circuits. Although these tools have improved at automating various aspects of the process, they have tended to focus on the circuits within each chip rather than the circuits connecting them.
Pohl said the idea came to him while working on an electronic vehicle prototype, after spending countless hours reading through data sheets to find an appropriate component. It felt like the opposite of the kind of automation he was used to in selecting and customizing software components for the same design.
“I was shocked at how electronics engineers worked and was pretty sure I was doing it wrong,” he said.
He surveyed electronics engineers from across the industry and found they were all sourcing components the same way. In 2016, he collaborated with a small team to see if they could automate some of this process. Then, in 2018, they got the prototype working and launched the company.
Pohl argues that designing the printed circuit boards consisting of multiple chips is a much more complex engineering problem than designing the chips themselves. Designing the logic on silicon chips is relatively homogenous and is a more repeatable process. Chips designers also had a greater need for design automation as they scaled individual chips to support billions of transistors.
“They reached a level of complexity where it was unfeasible to do it any other way,” Pohl said. “Board-level design will soon hit that level as the complexity is rising.
Building out an ecosystem
Celus tools complement existing electronic design automation (EDA) tools. The company — fresh off a $25.6 million series A round — is also working on partnerships with several leading electronic computer-aided design (ECAD) design tools for managing data about circuit designs. It hopes to do so with vendors like Altium, Autodesk, and Siemens EDA to help automate the design these tools produce and get them ready for manufacturing.
Other startups have also emerged to help automate various aspects of the PCB design process, including companies like CircuitMind, JITX, and InstaDeep. Altium recently acquired Gumstix, another PCB-design startup. In addition, Zuken, an established Japanese ECAD vendor, has also been working on a machine learning PCB design tool.
Earlybird Venture Capital led the funding, with participation from DI Capital and existing investors Speedinvest and Plug and Play.