Estimated reading time: 6 minutes

All Systems Go: Time Traveling to 2030 for ML-Augmented PCB Design

In our previous column, “Accelerate Your PCB Designs with Machine Learning,” we explained how artificial intelligence (AI) is an umbrella term embracing technologies that empower machines to simulate human behavior, while machine learning (ML) is a subset of AI that allows machines to automatically learn from past data and events without explicitly being programmed to do so. As ML systems become increasingly complex and capable, the distinction between AI and ML is becoming increasingly blurred.

We also discussed how we are currently in the early years of the second era of AI, and how ML has started to appear in PCB layout applications. Remembering that we are still in the early days of ML deployment in the PCB space, we talked about the types of tasks ML can help with today, such as detecting when we start to perform repetitive low-level activities and assuming the responsibility of implementing these tedious, time-consuming, and error-prone tasks, thereby allowing us to stop doing the dull and boring things and freeing us up to start doing only the cool and interesting things.

Since we’ve already looked at some of the more practical ideas we can implement in the short term, we thought we might use this column to zoom a little further out into the future, to brain-stem-storm on where things might go in the next five, 10, 15, or even 20 years.

Of course, as the baseball-playing natural philosopher, Yogi Berra, famously noted, “It’s tough to make predictions, especially about the future.” When we think back to the beginning of the current millennium in the form of the year 2000, which is only 22 years ago at the time of this writing, few people would have predicted the technologies we have today, like smartphones with multiple cameras, GPS and maps, and apps that allow us to do things like video conference with family and friends around the world, identify tunes that are currently playing on the radio, guide us to our cars in parking lots and help us tune our ukuleles (seriously, it’s a thing). How about the ubiquity of today’s wireless networks and cellular communications, with low Earth orbit (LEO) satellite constellations like SpaceX’s Starlink starting to come online? And then there’s virtual reality (VR) and augmented reality (AR), and of course, AI and ML.

Remember that the first iPhone didn’t appear until 2007 (15 years ago), the modern era of AI and ML only kicked off circa 2012 (10 years ago), and that consumer VR in the form of the Oculus Rift made its first appearance in 2016 (only six years ago). Who among our number would have predicted any of these applications and technologies 20 years ago? So, how accurate will any predictions we make be here? Well, let’s take some guesses, and then in 2030 and 2040, we’ll look back to see how well we did.

In the case of AI and ML, today’s models are scaling to incredible heights in terms of size and sophistication. Amazing applications are being developed in other domains, so what might AI and ML models trained on humongously large data sets bring to the PCB design and layout space?

A good starting point might be to look at GitHub Copilot, which has been trained on billions of lines of code and which uses the OpenAI Codex to suggest code statements and entire functions in real-time right in the software developers’ editors. One of our embedded software developer friends was recently telling us how Copilot can start making suggestions as soon as she types in a function name. As a result, she says her personal productivity has improved dramatically.

Suppose we were to incorporate something like this sort of AI/ML capability into our PCB tools. For example, let’s assume a designer selects a certain complex component, and the AI/ML immediately “looks inside the part” (i.e., accesses the datasheet) to discover the various voltages required (including the core voltage), the electrical interfaces supported by the GPIOs, any special memory interfaces like GDDR and PCIe (and which generations of these interfaces), and so forth. Right from the get-go, the AI/ML could start thinking about—and suggesting options pertaining to—power supplies, breakout patterns, thermal concerns and other considerations.

Today, we are largely constrained to viewing our 3D designs using 2D displays in the form of computer monitors—we are also limited to manipulating our tools using interfaces like keyboards and mice—but new technologies are on their way. For example, one company has just announced a technology that will provide true 3D holographic walls and tables that provide “stunning high resolution, perfect depth of focus and 180-degree to 360-degree viewing angles” without the user having to use any form of glasses or contact lenses (apart from anything they wear normally).

Figure 1: Even today’s PB AR systems offer tremendous visualization capabilities.

Imagine having the ability to view a PCB design in 3D. To be able to use hand gestures to rotate the design, zoom in, grab things, move them, pull the design apart and rearrange things—think Tom Cruise in “Minority Report” but without any obligation to wear his retro-futuristic steampunk gloves.

Another possibility will be to include multiphysics visualizations, including heat transfer, electrostatics and magnetostatics, stress and strain, and computational fluid dynamics (CFD). Imagine being able to see a glorious 3D view of a design in its entirety, with animations of things like signals propagating through traces (think “Tron”), heat being transferred, electromagnetic fields interfacing and interfering with each other, and so on. Critical signals approaching the limits of their propagation delays could be indicated using color; similarly for signal integrity (SI) and power integrity (PI) concerns. During all this, the AI/ML could be offering optimizations and suggestions.

Figure 2: Multiphysics visualizations provide a whole new way of seeing designs and systems.

Do you think this is “far future?” We haven’t even started yet. When most of us hear the word PCB we typically think of a traditional 2D board. Yes, of course, it has depth in terms of layers, but we predominantly visualize it in the context of a 2D X-Y plane. Our belief is that, in the not-so-distant future, we’re going to start thinking of “boards” in the context of 3D X-Y-Z volumes.

3D printing technology has come a long way. In addition to plastics, it’s now possible to print metals, including silver, gold, copper, and even stainless steel. It’s also possible to print glass. New printhead technologies allow conductive inks to be printed with resolutions of 0.5 µm.

What we are visualizing is an AI/ML-based true 3D PCB 3.0 generative design and implementation process that starts with someone saying, “We need to create a system with these capabilities that fits in this volume.” Right from the get-go, engineers and layout designers will be working together with the AI/ML, with each new suggestion generating a cascade of options and possibilities.

For example, it’s currently possible to 3D print a standalone coaxial cable. If a “board” is to be implemented as a 3D-printed entity, then such cables could be created inside the board as an integral part of the board. Silicon chips and chiplets, along with other components, could also be fabricated into the board, with both metallic and optical waveguide interconnects being implemented as part of the 3D print. Similarly, thermal conductors and cooling pipes could be fabricated as an integral part of the 3D structure.

It's not beyond the bounds of possibility that by 2030 or 2040 a group of designers and engineers scattered around the world—along with one or more humanoid avatars representing AI/ML systems—could be working together using 3D holographic visualizations to create something we wouldn’t even recognize as being a PCB.

These are just a few of the ideas that we’ve been bouncing around between us. What do you think about all this? Where do you think things are going? And do you think we are overstating the possibilities or understanding the potential of next-generation technologies?

Jorge Gonzalez is a lead software engineer at Cadence. Luke Roberto is a principal software engineer at Cadence.

Download The System Designer’s Guide to… System Analysis by Brad Griffin along with its companion book The Cadence System Design Solutions Guide. You can also view other titles in our full I-007eBook library here.