The Hidden Feature of AI Startups: A Much-Needed Human Touch
CAD as we know it for printed circuit design came into existence in the mid-1960s. And while some industry designers still remember the days (more fondly than I would!) of hand-taping components and traces, then using a camera to produce films for fabrication, it didn’t take too long before computers started taking over.
In 1970, Racal-Redac, which later was acquired by Zuken, released its original PCB, schematic and silicon layout tool. A few years later, Scientific Calculations introduced SCICARDS for generating photoplots from Gerber files. Of course, the dimensions back then were epic in size – pads were 70 mils or more, and lines and spaces were 25 mils.
By 1976, the EDA market was starting to pop with companies, and not just for design. Makoto Kaneko founded Zuken. A trio of professors at the University of Texas – James Truchard, Bill Nowlin and Jeff Kodosky – launched National Instruments. A former Tektronix engineering manager, Doug Campbell, started Polar Instruments.
A few years later, another group of Tektronix engineers – Tom Bruggere, Gerry Langeler and Dave Moffenbeier – founded Mentor Graphics; Aryeh Finegold and David Stamm formed Daisy Systems; and ECAD Inc., the predecessor to Cadence, was launched. By the mid-1980s, P-CAD (eventually bought by Altium and then Accel), Pacific Numerix (later part of Ansys), OrCAD (now owned by Cadence), and Protel (which became Altium) were up and running. The circle was becoming complete.
From there, mergers and acquisitions (and a few bankruptcies) became a standard part of the PCB design landscape. Eventually, the shakeout was so extreme that only the four major CAD vendors we now know, plus CadSoft (suppliers of Eagle), were left standing, along with a relatively small handful of side players offering specialty analysis or CAM. Almost all the rest of the place and route developers had been acquired.
Things stayed quiet for more than a decade. During that time, a few startups entered the space, often relying on open-source development to make a go. These tools were often used by hobbyists who couldn’t afford to spend thousands of dollars on mainstream CAD programs. Of them, KiCAD, OsmondPCB (for Apple) and Pulsonix (which may have been the most sophisticated among them at the time) were probably the most widely known.
Today, however, we just might be entering another golden age of software development. Spurred by strides in artificial intelligence – not to mention millions of dollars in private investment – one can hardly say large language model without catching the ear of another founder of an AI startup.
Not that PCD&F/CIRCUITS ASSEMBLY isn’t doing its part. We have interviewed many of these companies in these pages or on our PCB Chat podcast: Arch Systems, Breadboard, Cybord, Celus, Darwin AI, Flux, JITX, Luminovo and Quilter are some of the firms we have highlighted over the past few years. They all have great stories to tell, many of which have similar origins: highly educated, highly motivated engineer takes job in hardware design, finds existing processes slow or antiquated, and decides to make them better. It’s the classic entrepreneur story, one that now household names like Bill Hewlett, David Packard and Steve Jobs would all relate to.
And the tech is often really cool, if not always ready for prime time. AI as it stands today does not replicate the core knowledge, let alone an understanding of the tradeoffs, an experienced PCB designer brings to every job. Nor does it solve the inescapable anxiety that comes with turning over key elements of a design, namely routing, to a computer program. With reason, most designers eschew much of the automation already available in their CAD tools. To truly make a mark, AI will have to get much, much better at 100% completion while also demonstrating it can appropriately balance design intent and manufacturing (and cost) realities. Short of that, we will end up right back where we are: sophisticated solutions to addressing the entire design flow, but a gap in the core place-and-route functions.
Even as their products evolve, however, many of the startups have captured lots of eyeballs and more than a few repeat users. Another not-so-surprising development I’ve noticed is the tendency for a localized bent to the actual sales: while some of the AI startups have captured the attention of some of the truly elite customers, they mostly have very small numbers of paying clients.
But as someone who survived the dot-com bubble, I recall all too vividly the hangover that came after the celebration ended. Some measure of sobriety is still in order.
There are some key differences between then and now. Today, capital comes primarily from private equity, not IPOs. Founders (and their benefactors) haven’t been cashing in their chips (yet). Stability, at least in the PCB sector, is the norm.
But reportedly more than $1 trillion has been invested in AI companies across all industries. There is no way that won’t come with some disastrous endings. Even in our space, based on what I’ve seen, the hype among big-monied investors has outpaced the reality, both from a technical standpoint and from what customers are willing to pay for.
Like the ECAD space in the 1980s and ’90s, there will be a rapid proliferation period followed by a weeding out. Most of the companies we see today will not exist in their current form in five years. Some will be bought, some will fold.
No matter what happens to their algorithms, however, I do hope the founders all stick around. They have infused the industry with the remarkable passion only youth can afford, and we all benefit – even crusty veterans like me – from their positive energy and sense of purpose.
P.S. Want an insider’s view on how AI could overcome the current gap for solving complex designs? Attend Charles Pfeil’s keynote on Oct. 9 at PCB West. And while you are there, check out the AI for Electronics panel and the design community meetup (sponsored by PCEA and JITX) going on that same day.
Mike Buetow is president of PCEA (pcea.net); mike@pcea.net.