Join the Age of AI Mailing List. You'll be in good company with CEOs, investors and innovators around the planet.
“Electronics circuit design.”
Most engineers, even some hardcore techies, don’t get a feeling of bubbly enthusiasm when they hear the aforementioned words. (It’s also an awkward conversation-pauser at parties if asked what you do — just say “I make electronic gadgets,” jeezus!)
Because it’s not fun for most people, including engineers (and because to say that the electronics business is “huge” is laughably modest), therein lies an opportunity.
In this episode, I speak with Tobias Pohl, the CEO and Co-founder of Celus, which builds AI-assisted tools for circuit design.
The goal of their tools isn’t to replace the engineer, but to reduce the time they spend on certain design problems from weeks to hours.
We dive into the structure of the industry, how the electronics engineering process works (in simple English), and what it really takes to build USEFUL AI software for the purpose:
1:49 — Why engineers hate electronics!
03:33 — The current process of Electronics Engineering, and why it’s so damn hard
13:00 — Nerd ego, and why “experience” in electronics design is a double-edged sword
19:59 — What can you automate? The value of machine learning in the engineer’s process
28:47 — How machine learning streamlines and upgrades circuit designs
41:16 — Finding the right source of data for training Celus’ ML models
45:37 — The challenge of synthetic data for ML
49:00 — Celus’ journey as an AI software startup
56:06 — Celus’ target market and sales process: do big companies like Apple and Sony have their own electronics design tools?
01:01:40 — Why the “build vs buy” dilemma is not a problem for software startups anymore
Aman’s 2-Minute Summary and Key Takeaways
The overall process of circuit design:
At the highest level, it starts with an idea for a “black box” circuit which interacts with the environment/external systems in some way. Then, you have a long list of requirements that constrain what kind of inputs, outputs and physical size the black box will have.
This is given to the engineer, who now has to figure out how to design a circuit on paper that solves that problem, using standalone components that already exist — sensors, microcontrollers, etc. This is called the “schematic.” This is the most valuable part of the process, where the real IP lies. The market leader in this space is — wait for it — Microsoft Powerpoint. Literally.
But in any case, this is the first hurdle, because, in Tobias’ words, there are millions of components. Many engineers still look through long PDF lists of parts. Choosing the right components, for a circuit that you haven’t even designed yet, is tough!
The next step is layout design. The abstract drawing of the design is converted into the actual physical architecture, which would fit on a printed circuit board (PCB). You have to choose what goes where, physically. Here, you have to take into account what you are optimizing for, because many companies follow certain conventions and design patterns for their layout!
After the layout, it finally goes to printing/manufacturing.
How ML can help, and the challenges:
The first challenge is the “cold start” — choosing the right modules and components, which you will use to design the circuit — it’s somewhat of a chicken-egg problem.
The Celus software first uses ML to understand the requirements for the design, and then suggest reusable modules that could solve the problem at hand.
Once you have a few suggested modules, it further generates a long list of components that you could use to make those designs.
It’s like going to a shopping site and saying “I want to design a house,” and the site actively suggests to you all the things you’d need, different suggested designs, and guides you on how to go about it.
Then, going from schematic to complex PCB layout — AI is a great tool to create best options for circuit layouts, because it’s a hard skill to learn but also subjective. The human can make tweaks on top of suggested layouts.
An interesting challenge in using AI in electronics design is about intellectual property. You don’t want AI that’s implicitly learning companies’ trade secrets and passing them around! So Celus only trains their ML models open-source project data, which comes with its own set of challenges we discuss extensively in the episode!
But that being said, Celus is already gaining market share and working with enterprise customers, because of the speed gains and user friendliness upgrades by using such design software.
We got into chatting about random topics like fundraising.
One thing Tobias said that stuck with me: if thinking about a potential investor and their face doesn’t make you feel happy and excited to have them on your team, don’t take their money!
(If you’re in the position to say no, ofcourse.)
(Ethics Policy: These opinions are 100% my own as an independent observer and educator. I don’t own stock in guests’ companies or their competitors, nor do I get paid by them in any form for any reason at the time of publishing, unless specifically stated. Episodes are also not intended to be an automatic endorsement of any company or its products and services.)