Happy Holden and Charles Pfeil Discuss the Past and Future of PCB Design, Part 2
In Part 2 of this feature interview, I continue my conversation with industry icons Happy Holden and Charles Pfeil at the recent AltiumLive 2017 event in San Diego. In this half of the interview, we discuss the potential for artificial intelligence in EDA software tools, the graying of the designer pool, and what can be done to draw more young people to PCB design.
To read Part 1 of this interview, click here.
MATTIES: You mentioned you sold your service bureau to a fab shop, and you worked there for a while after. That was a strategic move on their part it seems.
PFEIL: It was. The designs were better because we were there all the time. We were in the same building as the fab shop, so it was good. With a service bureau, the boards aren't always necessarily made at the same place, but generally, yes. If you do design and fab, you can actually take short cuts.
MATTIES: This is really what you guys are advocating.
HOLDEN: That's how I got into design. I was the manufacturing engineering manager complaining to the VP about the quality of design. He said, "Well you obviously seem to understand the problems. They're going to make you the new design service manager so you can fix them." I should've learned from the Army: You don't bitch and complain or somebody's likely to make a dramatic decision. I didn't choose to get into design. It chose me. When we took the set of designers who used to report to the product division and now they were reporting to the manufacturing division, they went from the least important person in the product division of PhD gurus, to the most important people in manufacturing, because they dictated our quality, throughput, and profitability. They became the most important asset we had, even though they were doing the same job, just difference in emphasis.
MATTIES: After you sold your business you worked for the fabricator for a while. What fabricator was that?
PFEIL: It was Automata in Virginia.
HOLDEN: A fabulous shop, a real innovator.
PFEIL: I sold it to them in '85. They actually had an LDI machine, and were testing it in '87.
HOLDEN: That's why I visited them.
PFEIL: It was nice, but it was too slow. So it took some time for that process to get improved. But they were always innovating. One of the jobs they gave me was to set up an impedance testing process so they could understand what kind of impedance they would have on their boards. That was really interesting to do that.
MATTIES: So now you're with Altium. How long have you been there, and what's your primary function here?
PFEIL: It’s been two and a half years, and I’m a product manager working in development, with a focus on routing technology. Specifically right now we have a tool called ActiveRoute, which is adding automation to interactive routing. So you can select routes or select connections, give it a guide where to go, route on multiple layers, automatic tuning. Things like that.
MATTIES: Do you see a day where design is fully automated?
PFEIL: What a question. No.
MATTIES: Logic would say otherwise, right?
PFEIL: The reason why is the following: First of all, what CAD company out there is going to want to spend five to eight years’ worth of resources to develop that?
MATTIES: If they started 10 years ago it'd be the leader today.
PFEIL: There's always that. But second, when you think about PCB design, first of all you look at all the different technologies that you have to support. From an RF design to an emulation card, they're totally different. It requires a different way of thinking about it, different rules, different high-speed requirements. Different ways of creating the traces based on the behavior that's desired.
So to start with you'd have to start, "OK, I'm going to support this technology." You could say, "I'm going to support a board with a processor and standard interfaces." There are a lot of boards like that right now. What do you have to do to fully automate it? You need to have a full understanding of the circuit and the behavior or the physical aspects that are needed to fulfill the behavior requirement.
You could say, "Well, there are reference designs." For example, if I get a reference design for a DDR4, okay it's on four layers. Well, my design isn't four layers. So this reference design isn't going to work for me. I need to place some memory components on both side of it. So just with that one circuit you've got n number of different possibilities. How does a computer know which one to choose? You’ve got to be involved.
MATTIES: I’m thinking something along the lines of IBM Watson, right?
PFEIL: Yeah. Give me IBM Watson. We'll have it in a year.
MATTIES: But seriously, it's there.
PFEIL: I agree with that.
MATTIES: Even though it's a challenge collecting the data and building the algorithms.
PFEIL: Yes. It's a knowledge. You need the knowledge. You need to know about circuit. You need to know about the component technology. Frankly, not all that information's out there, especially on a new microchip. How are you going to get that information? If you're a partner with Intel you can get it, but you don't get all of it. These are the kind of things that will have to be overcome.
MATTIES: Because as we're talking, we're at the AltiumLive 2017 event here in San Diego, and you guys are just introducing your new version of Altium Designer 18. It's all about speed. It's faster than the previous, a lot faster, with new features and new capabilities. Lots of new menus. A lot of intuitiveness has been built into the menus it looks like. The point being is it's faster. The boards that we're building and designing with this tool are getting faster and faster and faster.
PFEIL: ActiveRoute is, I'd say, on average is 25 times faster than manual routing. We keep adding automation, but it's still under user control. How do you communicate the user control that makes a functioning circuit to IBM Watson?
MATTIES: This is a layman's point of view. I don't develop software. I just see it almost as a wizard. If we have a crisis of bringing young engineers or designers in, then we only need tools to offset that deficiency. It's a wizard. Here's my physical space. Here's what I need this board to do. I want it to do this, this, this, this, this, and it's a program. Here are the components you need. Here's what in stock.
PFEIL: Happy, you were working on a kind of wizard for microvia designs, right?
HOLDEN: Yeah. It's one of my most famous stories. First, HP labs developed a DFM design advisor because we were working very heavily on design for manufacturing. We developed new metrics that helped us get better cost and electrical performance. When HP completed that design advisor, DCA, they tested it with different young engineers. My group used it and just loved it. It essentially automated and made us more productive and everything else. But when we went out to new engineers and new designers, after two or three questions they would turn around and say, "What do I ask it now?" They kept saying, "What do I ask it next?"
The PhDs went back and said, "Here's the problem. Happy has had 25 years and his people have an average of 15 years. They know all the questions. This guy is just out of college. He hasn't learned the questions yet. What we need to do is not build a better machine to answer, but build an artificial intelligence that can figure out the questions." So the young engineer just has to steer this thing, but it's going to create the questions.
That was explored. That's when I learned that self-learning artificial intelligence is better than a human. In fact, this is the thing that Elon Musk is warning us about. That is self-learning artificial intelligence that overtakes us. I've seen it. It's been demonstrated and I've seen it demonstrated. With the 55 complex boards from HP we benchmarked it with, it took the two to twelve weeks and made it two hours. It took 15% to 60% off the manufacturing cost and improved it 100% of the time.
PFEIL: So, are they still using that?
HOLDEN: No, they never used it. It was only a research development tool, so it's sitting on the shelf.
MATTIES: But you validated the concept.
HOLDEN: We validated the concept. Now with more and more people focusing in AI, it's surely going to be rediscovered or reinvented. We're reaching a cusp point. If the current designers retire, and the new guys don't master the challenges…like Dan Beeker discussed this morning, the OEMs have nothing else they can do other than look for the AI answer, with the design no longer being done by humans. It could be done by humans, if we master the challenges, or it's going to be done by AI. Because I've seen it done in AI, so I know it works and it's there.
PFEIL: The question is, who is going to take on that project?
MATTIES: The person with the vision of the future, because we know that it's only going to get faster. We know the IBM types of systems are only going to gain more traction.
PFEIL: Would it be a GoFundMe project? I think compute time is pretty expensive.
MATTIES: It could be. It could also just be an organization that recognizes this as the future when they invest in it. I don't know. It just seems to me that it's a path. It may be 20 years down the road. It may be five years down the road. Who knows? It depends on the motivation.
HOLDEN: I know there is a challenge coming that Dan outlined. Because I don't want the third alternative, which is that we outsource all design to India or China. If that happens, then we've lost the game.
MATTIES: Back to your original point of dealing with fabricators that employ designers, it's better to have that communication.
PFEIL: You mentioned what a designer goes through in Japan. This must have been six years ago, but I went to Viasystems and gave a presentation to them. There were 40 PCB designers in the room and 100% of them had master's degrees in engineering. Now they're going through the PCB design education and application. These designers, and they're getting the training. That's not happening in the U.S., not at that scale.
MATTIES: I was at PCB West, and Judy Warner (Altium’s director of community engagement) introduced me to a young lady who works for a pharmaceutical company. She is a mechanical engineer and she's doing their enclosures and cases. They don't normally do electronics, but they had a need, so they invited her to see if she would be interested in circuit design. And she has embraced it. She just sees it has a natural progression. She said, "I can make beautiful products this way."
PFEIL: Maybe the innovation has to come from the designers coming up who understand a different approach to design rather than someone like me who has 50 years of pretty much going in a certain direction. Maybe I'm blind to what is possible in the world of AI.
MATTIES: That's exactly what I thought. It's going to be the young retraining, because they're not going to stand for the inefficiencies and nuances of what we've been doing. They're going to own it. There seems to be a gap between designers in age groups. There's one large group of designers that is over 50 and then a group that’s 30 and younger, it seems. There’s a 20-year gap where we didn't have these designers coming in and learning.
Any advice that you would give a young designer, now that we've just admitted that after 50 years, your advice might not be valued?
PFEIL: The most important thing is being creative and being able to apply that creativity to your task at hand. That will become a lifelong enjoyment. But if you're not able to do that, take up something else like music.
HOLDEN: I would add that if you don't want to spend your time reading as much as you can and learning from others, then you chose the wrong profession and area. This is too dynamic and too complex for anybody that doesn't want to put in the time and work. It's not a lazy man's area.
PFEIL: That's true.
MATTIES: Thank you both.
PFEIL: You're welcome.