JITX startup uses AI to automate the development of complex printed circuit boards

AI has the potential to eliminate the boring complexity of the PCB design process.

Anyone can learn how to design printed circuit boards, but only an experienced engineer will be able to create a layout that is both well optimized and not prone to melting, explosions, or bringing the device controlled by the board into a state in which it explodes or melts. Experienced engineers are usually busy, their work is expensive, and their character is very nasty - especially if you force them to do work that does not fully reveal their genius.

JITX is a startup founded by a team of electronics and mechanical engineering engineers from the University of California at Berkeley that creates an AI that can help develop optimized printed circuit boards in hours rather than weeks. As a result, the work of the engineer will move from manual labor to supervision. You tell the system at a fairly high level that you care, and it develops a working printed circuit board, on which everything that you do not care about has already been decided for you. Your expert engineering experience focuses only on the right things, and the system produces printed circuit boards - only better, faster, and cheaper.

The head of JITX is Duncan Haldane, the author of the “hyperactive jumping spring robot” named Salto-1P. A rather serious part of the work on Salto-1P (as well as on other robots) was spent on the development of special hardware, including printed circuit boards. Haldane and his colleagues discovered that every time they started doing something new, they had to start from scratch. “We realized how little time we spend on our core business — research,” says Haldane. - All our time was spent on low-level development of iron. The effort to develop a special iron is a huge barrier that stands in the way of new creative systems. ”
')
The goal of JITX is to make the development of hardware more similar to the development of software. This is not the first company trying to do something in this direction, and there are already a lot of auxiliary tools for the development of printed circuit boards. However, JITX believes that its more comprehensive and holistic approach is unique. The company writes in a press release:

The inspiration for our key technology was the technology that was used to develop computer chips. The appearance of hardware description languages (HDL) in the 80s revolutionized the development of chips. HDL has changed the basic principles of schema development. Instead of drawing the schemes manually, the engineers expressed the desired behavior of the scheme with the help of code, and then the algorithms automatically translated this code into the necessary copper forms. This workflow allows you to create today's chips with billions of transistors. We are organizing the same workflow in the design of printed circuit boards.

Development of printed circuit boards requires the use of knowledge from many areas - electrical engineering (circuit design, RF design, signal and energy compliance), engineering (temperature and vibrations) and production (cost optimization, design taking into account production requirements, suitability for mass assembly and testing ). It is not surprising that almost all these subtasks are difficult to computerize, so we have to use ingenious representations and heuristics to get suitable solutions. For each of these areas, you need to keep track of millions of details, and it is high time to get computers to do all this accounting.

Here is an early demonstration of JITX work recorded by Haldane last November so you can see how it all works:


“What they are trying to do is needed by most people working in the field of iron,” says Ted Larson. Larson manages OLogic, a robot design and development company focused on embedded systems. OLogic has worked on many robots with which you probably know, but have no right to talk about their work. “One of the problems that they identified and with which I agree is that Silicon Valley began to create terribly bad equipment,” says Larson. - The assumption that the development of iron is difficult, and that there are not enough people who could not do so expensive? It's straight to the point. ”

However, there are reasons why the approach to the development of iron, as to the development of software, has not yet been implemented, says Larson. "Many people want to make the development of electronics similar to software development, but there are problems with using this approach with schemes that go beyond certain prototypes." To develop circuits that can be produced and subjected to tests for compliance with standards, experience is needed, as well as to select all the necessary components that will work together exactly as it should. Larson thinks that JITX is ideal for projects that are somewhere between the prototype stage and the production stage, especially considering the time savings and cost of traditional approaches. And he says that the system has potential. "They are at the very beginning of the journey, and they have tremendous opportunities to improve the whole process."


Three boards developed by JITX AI

So far, JITX has been using tools designed for internal use. You tell them what the board should do, and they attract engineers who are being helped by the AI ​​to work on this task and issue the most efficient pay. On average, JITX produces printed circuit boards 3 times faster and is 25% cheaper than experienced engineers working without help. The ultimate goal is to further increase automation, expand it beyond simple printed circuit boards. But for now, JITX will join the Y Combinator summer workshop to develop ideas with little external support.

We talked with the head of the company, Duncan Haldane , by email.

IEEE Spectrum : Can you describe a typical PCB design process, and what’s wrong with it?

Duncan Haldane : First you need to figure out what the board should do. What are the actuators, what sensors, how they communicate with each other, what kind of processing is needed. And then there is hard work.

We googling, we seemingly finding the optimal sets of component parts, digging in the specifications of a hundred pages thick to figure out how to put all this together. After that, we model all the components in our favorite CAD (manually rewriting information from PDF), and draw a diagram. On the scheme we add symbols denoting all parts of the board, we try to line them up nicely (on one page, or fifty, depending on the complexity of the circuit), manually connect all the contacts, drawing the lines denoting wires. We hope that there were no mistakes when drawing lines. At this stage, we invite other engineers who will stare at this scheme for several hours, trying to find errors, since the only error will mean that your board will not work (and may explode).

When we have a scheme, you can develop the board itself. The goal is to understand where to place all the components so that you can draw all these copper tracks connecting the contacts. We need to think about a bunch of different physics, and for this we run all sorts of simulations, and manually mark the results on the board (by drawing a copper path of the correct form). Also at this moment you need to think about how this board will be produced, assembled and tested. We again call other engineers to stare at the board design for several hours trying to find errors. We hope that they will find all the mistakes, because now we are ready to pay for the production of the fee.

What's wrong with it? Yes all. This is the worst of what man does. Imagine that you have done all this, and then again for the next project you start from scratch.

Why no one has done what JITX is doing now?

This is happening now because the industry is in despair. The development of printed circuit boards has become such a delicate thing that teams of people draw all these boards around the clock and in shifts in order to be on time. Moreover, new factories are highly automated, which is why a bottleneck between companies and the market is under development. And this urgency appeared when the AI ​​simply smears the best of the previous methods on the wall. We combine new AI results with this deepening market challenge.

What part of the development can be automated, and what will it look like for the end user?

You can automate everything, you just need to find the right approach. For the end user, it will look almost the way our tools work today. The user says that it is important to him, and not how to do it. For example, we order a board with the BLE system and a microphone (this is “what”), and our software selects the corresponding key components from the library, solves issues with power and component values, assigns contacts, plans location, marks out tracks, and issues a board and circuit. (What is it like"). If the form of the board is important for you, add this restriction, if the component location is important for you, add this restriction, if you know which BLE chip you need, add it. Development tools must be smart enough to deal with all the millions of parts you don’t care about and optimize your circuit for what’s important to you.

What are the limitations of the system you are working on?

The biggest limitation is that developers sacrifice control over trifles for the sake of automation, accuracy and speed. This is a very big difference compared to today's tools.

What are you working on with DARPA?

We are working as part of the Electronics Resurgence Initiative (ERI), a massive, collaborative effort to reinvent the process of creating electronics. Specifically, we are participating in the IDEA program, according to which the development of electronics will be conducted without human intervention. To do this, we are working on companies producing semiconductors and components to create an extensive library of electronic components, create software that turns high-level project goals into real circuit boards, and optimization tools that find the best versions for the boards you need.

How do you see the distant future, and how does your current work fit into it?

In the future, we want to scale the technology of automatic development of printed circuit boards to give more creative freedom to the developers of hardware. All receive custom-made printed circuit boards, and we guarantee that the problem remains solved. Then we want to return to electromechanical development. When designing a printed circuit board, many problems arise during disputes with a mechanical engineer about how to get another half millimeter of extra space to cram another component. Robots are electromechanical devices, and we are working to ensure that someday our software can develop improved robots that can assemble the next generation of even better quality robots, execute code on improved printed circuit boards, and so on.

Today we ourselves use our development tools to quickly create cool printed circuit boards for other companies. This is the best way to make sure that the tools work, that they can be used, and that every minute of development is not wasted while we strive for our goal.