Fullscreen HTML5Med Res (??MB)HTML4

Tracks and tribulations Techno Talk A ramble through the evolution of PCB layout, from tape and scalpels to overcaffeinated autorouters and today’s AI-powered tools that are starting to earn their keep. There was once a time (“gather around, children”) when designing a printed circuit board (PCB) didn’t involve clicking a mouse, dragging traces across a screen, or pressing a button marked “Autoroute and Pray.” In those far-off days of yore, PCB layout was a hands-on, eyes-straining, finger-sticking affair performed on large sheets of translucent drafting film. Instead of pixels, we used precisely manufactured rolls of adhesive tape that came in various widths, each corresponding to a different track size. Want a 10-mil trace? Reach for the 10-mil tape. (In this case, “mil” means one thousandth of an inch. The term comes from “milli-inch,” not millimetre, which has caused no end of confusion over the years.) Pads and component footprints were not dragged from a library—they were rub-down transfers. You’d carefully position them over the layout and burnish them into place using a stylus, all the while muttering arcane incantations to the gods in the hope that they landed exactly where intended, because lifting and repositioning them was… let’s say “suboptimal”. Working at 2× or 4× scale was common practice, which meant your “small” board might occupy something the size of a dining table. Once complete, the artwork would be photographically reduced to its final dimensions, at which point any tiny misalignment you hadn’t previously noticed would suddenly become glaringly obvious. The most common tape colour was black, but for those working with double-sided boards (gasps of astonishment), an alternative was to use red tape for one side (usually the top) and blue tape for the other (often the bottom). This made it much easier to visually separate the layers and avoid going cross-eyed when your sheets of drafting film were stacked on a light table. On the one hand, this was a painful way of doing things. People joked that you 42 never saw a layout designer wearing a woolly pullover, even in the coldest weather. This was because you didn’t want any errant fibres to become attached to your film, only to be immortalised as almost invisibly thin “wires” destined to cause hard-to-diagnose faults by shorting out real tracks. On the other hand, there was something rather wonderful about it all. You could see the board taking shape beneath your hands. Every track was placed with intent. Every component had a physical presence. And it was as much an artisanal craft as it was an engineering discipline. System overheating! I remember circa 1982 when the company I was working for invested in a proper computer-aided design (CAD) system for PCB layout. This involved two large cathode-ray tube (CRT) screens mounted on top of (and integrated into) a meaty metal table, which also included a chunky workstation. The screen on the left was blackand-white; the one on the right was in colour. We used a keyboard and command-line interface (CLI) associated with the left-hand screen to issue cryptic commands like V (for “via”) 1250,875 or T (for “track”) 1000,500 1500,500 to instruct the system to place a via at a specific X-Y location Max the Magnificent or to draw a track segment between two points. As each command was executed, the corresponding object appeared on the right-hand screen. This system was not as user-friendly as one might have hoped, especially given how much of our hard-earned lucre we’d invested in it. One day, the right-hand screen started flashing an enigmatic message: SE392. The text was in red on a white background, so we knew it had to be something important. The system had been delivered with a veritable cornucopia of documentation. We’re talking about an entire shelf groaning under the weight of thick, ring-bound manuals. We finally located the “Messages” section in the first manual, which helpfully outlined the different categories. This was where we discovered that “SE” indicated a “System Error”. It also directed us to an appendix in the last manual that contained the messages themselves. There were hundreds of them. When we finally tracked down SE392, we discovered that its corresponding description tersely read: “System Overheating”. It was right. When we turned back to look at the workstation, we discovered it was on fire! Push and shove As the years went by, the computer-aided layout tools grew increasingly advanced. One great advancement was the use of interactive graphics. Suddenly, with the aid of a mouse (or, in some cases, a light pen or digitising tablet), we could place components and route tracks directly on the screen, which felt nothing short of magical. Even so, PCB layout was still somewhat constrained. Routing was largely “Manhattan” in style, with tracks running horizontally or vertically and meeting at crisp 90° corners. The idea of casually introducing 45° segments came later Practical Electronics | July | 2026 as the tools matured and became more accommodating. Autorouters began to make an appearance in the late 1970s and early 1980s. By the mid-1980s, they were starting to show up as standard features in PCB layout tools. In theory, these magical constructs would take your netlist and—at the click of a button—produce a fully routed board. In practice… well… The problem was that early autorouters didn’t understand intent, priority, or subtlety; they simply connected point A to point B in whatever way seemed expedient at the time, often leaving behind a trail of unnecessary vias and meandering tracks that looked as if they’d been laid out by an overcaffeinated spider. As a result, no self-respecting layout designer would admit to using an autorouter, not least because it was almost invariably faster (and certainly more satisfying) to do the job properly by hand than to clean up the enthusiastic chaos that the autorouter left behind. It was in the latter half of the 1990s that I first saw something that took my breath away. I was chatting with a layout designer who was working on a complex multi-layer board. He demonstrated how he could use his mouse to “push and shove” things like tracks, vias, and components, and how the system automatically re-routed everything around to accommodate these changes on the fly. I was flabbergasted (my flabber has rarely been so gasted to this day). Rip-up and retry until it is done As the 1990s gave way to the early 2000s, autorouting algorithms became increasingly capable. This coincided with signal integrity concerns rearing their ugly heads. Suddenly, it wasn’t just about connecting point A to point B; we had to worry about things like controlled impedance, differential pairs and length matching for highspeed signals. The autorouters rose to the challenge (at least in part) by allowing designers to define constraints and priorities. In practice, however, most experienced layout designers adopted a hybrid approach. The critical nets—clocks, impedance-controlled differential pairs, high-speed buses—were routed by hand, with great care and no small amount of muttering. Only once the important signals were safely tucked away would we unleash the autorouter on the remaining “mundane” connections, treating it less as a master craftsman and more like an enthusiastic assistant who needed close Practical Electronics | July | 2026 supervision. The results were certainly better than the overcaffeinated spider efforts of earlier years, but we still kept one hand hovering nervously over the “rip-up and retry” button. Colour me impressed! I think we’re all aware of the impact of generative AI, starting with the public release of ChatGPT in November 2022. In the past couple of years, several companies have begun offering AI-powered PCB layout tools, with various degrees of success. Suffice it to say that if layout designers were less than enthusiastic about using autorouters, you can only imagine their sentiments with respect to AI-powered layout. I have a friend (no, seriously; I do) called Ben Jordan. Ben has decades of experience in PCB layout. He currently works for a company called Quilter (www.quilter.ai), which offers a physics-driven, AI-powered PCB layout tool of the same name. The folks at Quilter are well aware of the scepticism PCB designers have toward AI-powered layout tools. To counter this, they recently did something rather clever. They used Quilter to create a complete, working singleboard computer (SBC). They started by taking an existing open-source reference design from NXP. This design involved a baseboard with a system-on-module (SOM) mounted on top. They defined the outlines and connector placements for both boards, then passed these—along with the schematics and bills of materials (BOMs)—to professional real-world PCB designers, who quoted 200 hours per board (400 hours total). Next, they fed the same information into Quilter, which completed the SOM in a matter of hours and the baseboard in about a day. Now, this is the clever part, because “the proof of the pudding is in the eating”, as they say. They sent the Quilter-generated layouts out for fabrication and assembly. With the assembled boards back in hand, they first ran tests involving real-world applications with the NXP SOM mounted on the Quilter baseboard. Next, they ran through the same tests using the Quilter system-on-module mounted on the NXP baseboard. Finally, they replaced some NXP baseboard-SOM combos running in real-­world systems with their Quilter counterparts, which passed with flying colours, prompting me to say, “Colour me impressed”! But wait, there’s more! It’s one thing to hear about this sort of thing… it’s another thing to see it. I was in the middle of a video conference call with Ben (he was using his main computer, of course) when I mentioned that I’d love to see the Quilter-generated baseboard-SOM combo in action. Ben responded that he had one on his desk. It already had a camera plugged in because he was testing it with some AI models, so he joined our call a second time via the development board, which left me staring at a brace of Bens. To be honest, I couldn’t tell which was which. Just to add a dollop of metaphorical cream on top of the allegorical cake, while we were still talking, Ben fired up the AI model he was playing with on his Quilter-generated baseboardSOM combo. Now, tools like Quilter don’t (yet) produce the same level of finely honed, production-optimised layouts that you’d expect from a seasoned human expert with decades of experience. Where Quilter really shines is in the earlier phases of the design lifecycle. If your goal is to get a working prototype into your hands as quickly as possible—to validate an architecture, test an interface, or simply see if the blasted thing boots—then shaving weeks or months off the schedule is nothing short of transformative. In this context, “good enough and available tomorrow” can be far more valuable than “perfect and available next quarter”. If history has taught us anything, it’s that each new wave of automation has been met with equal parts excitement and scepticism. AI-powered layout tools like Quilter feel like the next step along that same path. They’re not here to replace skilled designers (at least, not any time soon), but they are changing the way we work, especially in the early stages of design. If they can get us from idea to working hardware faster, that’s a future I’m more PE than happy to route toward! 43