Innovations in AI: Transforming the Electronics Industry

On this episode of the OnTrack Podcast, we welcome Phil Marcoux, Advisor to the Printed Circuit Engineering Association (PCEA). Phil is gearing up for a fascinating webinar on AI in the electronics industry, so we thought we'd chat with him about what sees as the most potentially transformative aspects of this revolution in the electronics space.

Listen/watch to discover how Innovations in AI are transforming the electronics industry, reshaping everything from design to production. Phil and host, Tech Consultant Zach Peterson, explore real-world applications, the importance of data for AI's success, and how the industry is moving towards a more interconnected and intelligent (pun intended) future.

Listen to the Episode:

Watch the Episode:

Key Highlights

• AI & Electronics Design: Explores the symbiotic relationship between artificial intelligence and electronics design, highlighting how AI is revolutionizing the field.
• Are We in an AI Bubble?: Raises critical inquiry into whether the current surge in AI development constitutes a bubble, prompting reflection on the sustainability and potential pitfalls of AI advancement.
• AI Can't Read Your Mind: Discusses the inherent limitations of AI in deciphering human thoughts and intentions, emphasizing the distinction between AI capabilities and human cognition.
• Trends from the Early Days of AI: Examines trends emerging from the early stages of AI development, providing valuable insights into its evolution and predicting potential future trajectories.
• A Potential Role for AI: Speculates on the future landscape of electronics design with AI integration, envisioning potential applications and transformative impacts on the industry.

Further Resources

• Sign up for Phil's FREE Webinar
• Connect with Phil Marcoux
• Check out more episodes of the OnTrack Podcast
• Take advantage of exclusive 15 Days Free Altium Designer Access

Transcript:

Zach Peterson: Right, it's not just the collaboration between them, right? I mean, we have the tools and even with like All Team 365, you can now get people into the same workspace. It's now how does the AI even get that data and then create that feedback? So you're almost adding a third system into it in order to enable that part of it.

Phil Marcoux: That's correct, there has been, oh, a languishing effort for a communication standard within the IPC for a long time.

Zach Peterson: Yeah, that would be IPCCFX?

Phil Marcoux: Yeah, it just continues to languish. So, maybe it takes, maybe it's gonna take a new effort to, or a new alternate effort to provide that communication path.

Zach Peterson: Hello everyone, and welcome to the Altium OnTrack podcast. I'm your host, Zach Peterson. Today we're talking with Phil Marcoux, advisor for the Printed Circuit Engineering Association. We've had Phil on the podcast in the past. And coming up soon, he will be part of a very interesting webinar on AI in electronics. For those of you who have been paying attention to anything in software, you know that AI is the big craze right now, and inevitably it's making its way into the electronics industry. So we're gonna learn about some of those developments today. Phil, thank you so much for joining us.

Phil Marcoux: Oh, my pleasure.

Zach Peterson: How have you been since we last had you on?

Phil Marcoux: Oh, well, let's see. I have a new grandchild that we're helping to spoil. I've, let's see, had a chance to go out and do some more kayaking since we last talked. A little bit of traveling. Other than that, we're doing great.

Zach Peterson:Awesome, that's great. And of course, since we last talked, there have been, it feels like an infinite number of industry developments from M&A, nearshoring, friendshoring, and then of course, the two letters that seem to be making their way into just about everything, from software to manufacturing, which is AI. And you're going to be part of an SMTAPCEA panel on AI and electronics, is that correct?

Phil Marcoux: Yeah, Mike Butto graciously invited me to moderate it.

Zach Peterson: So this is a really interesting area, and there are established EDA players who are making inroads with AI in design. There are a lot of startups. And then I think there are also some manufacturing applications for AI in electronics.

Phil Marcoux: Oh, the funding being provided to some of these companies is just mind-boggling. Even one of my neighbors recently was funded to do a startup, and has already launched his product. I'm located here in Silicon Valley. One of my other volunteer activities is volunteering with an organization called Score, which is funded by the U.S. Small Business Administration. And because I'm part of the Silicon Valley chapter, we get clients from all over the, well, actually all over, well, all over the country asking us about helping them with their startups. AI has come in as one of the predominant activities. Matter of fact, we joke with people that if you wanna help your funding effort with your startup, in your elevator speech, make sure every fifth word is AI. But to me, AI is very misunderstood. And, oh. It's really a misnomer for what really is happening, and what its real potential is. And I fear that as a result of the mischaracterization and the misunderstanding, it's only going to delay it. And possibly even cause it to have its acceptance hampered. I liken it to exactly what happened with us, oh, going back 40 years in the early days of SMT. As many people know, I was part of that early SMT revolution. And there were periods of time in the early '80s where SMT almost never had a chance to exist. I see many of the same parallels. When I started my first company back in 1981 in SMT, we sought to purchase as much automation, and, without realizing it, as much early AI as possible. So that we didn't have to hire a whole bunch of people doing activities that were very tedious, and hampered the quality. To me, that was the earliest incorporation of AI. I also didn't realize when I was talking to Mike about the AI webinar that, oh, five years ago now, myself and a fellow named Dr. Chris Leichey, who was the chief statistician for two of the largest medical companies, Abbott, and Johnson & Johnson. And we wrote an article that Mike published. And it discussed, well, what they called at the time, SMT 2.0 or Factory 2.0. And what Chris kept emphasizing was the need for CFT, Critical Function Testing. Now, to me, going back to this notion that AI is a misnomer, people have to realize that the fundamentals of what is necessary to make AI a usable tool is all human-generated. AI will not exist without data. Where does that data come from? You know, the critical data in the electronics industry, whether it be design or manufacturing, comes from measuring results. Usually downstream results. You know, what's the electrical test data on a product? What's the physical appearance of the product? What's the overall quality of the product? Those are all measurements that are done downstream. And then if there's no feedback to the upstream processes, you know what happens? Well, if you remember anything about control theory, you know that if you're running open loop, you're running outta control. So you gotta have feedback. Well, in order to get data, which is critical to the existence of AI, a human-generated sensor, a human-generated process, a human-generated activity needs to be monitored, measured, and the data fed back. The promise to me of AI is to allow the data that has been gained in recent real time and from relevant sources, which is a critical activity here, allowing that data to be gathered using human intelligence, usually sensors created by humans, and being fed back to the upstream tools, and incorporated so that the process can be improved and enhanced in the factory floor. For example, in the simplest sense. Post solder paste deposition, one should be measuring where all the solder deposits are, and if the solder deposits are missing in certain areas. That machine, that optical machine, should be intelligent enough in these days to feedback the data to an intelligent screen printer and say, you're screwing up. You need to straighten out. Now, in artificial intelligence, we're assuming that's going to be the machine itself making those determinations, and correcting itself. That's to replace the human who may otherwise take too long to get back and correct that process. In another, in the critical to function format, if, for example, a circuit board has several BGAs on it, you know we need to have the AOI machine be intelligent enough to realize that the BGA areas need to have greater scrutiny and more testing. And more alarms set at a lower AQL level than might otherwise be. And the article that Chris and I did back in 2018, we cited, as an example, a board that had BGAs on it, 10,000 solder deposits. If we take, as an example, 12 missing solder deposits, you know if those were chip resistors or chip capacitors, we wouldn't necessarily break a sweat. If, however, those are in a critical area under a BGA, we'd need to break a sweat, and stop the process, and get that corrected. Now, how does that go from the manufacturing floor back to the design function? And that's a bridge that we're still trying to explore. And that's one of the activities that I hope to be of some value in facilitating with the PCEA. On the panel that we're doing March 6th, we have, oh, four companies that are participating. Two of which provide help and utilize data from their data banks to help designers select appropriate components. One of them, Lenovo, for example, helps a designer decide if the components they wanna put in their bill of materials really is readily available. The other one, Circuit Mind tries to help designers decide if a certain package selection is appropriate. Or maybe even breaking down the selections. You know, one example I've seen is where a circuit may attempt to use one large BGA that has, let's say, a couple thousand balls on it, and maybe it would be appropriate because of the availability of parts to break that down to multiple products, and have multiple BGAs. Ideally, if the data making that decision can be coupled with the knowledge of which factory that circuit board wants to be manufactured in, now we've really created a valuable bridge and valuable tool that enables the designer to not only get his design out more readily, but with greater assurance that it's going to work, and work well over time. And be manufactured well over time. So the critical thing that has to happen in the AI world for electronics to me is the generation of the data. Where is the source of that? And then to provide the linkages and communication from the manufacturing to the design source. I'm hoping that that's not too elusive. Again, going back to the early days of SMT, people don't realize that in the early days of SMT, those of us that were early practitioners tried to hold cards close to our chest. We had these treasured things called land patterns and design guidelines that we professed were unique to our particular process. And God forbid, you should take your product to somebody else 'cause it won't work. And lo and behold who comes up but a fellow named Dieter Bergman, the late Dieter Bergman on the IPC. And he gathered us all together, and basically told us we were full of it. And to get our act together if we wanted to make this thing happen. So that resulted in the IPC 782 document, the design guidelines. Another activity that occurred that, again, I'm hoping to avoid in the AI era is what happened around 1984 when the main users of SMT, namely the military defense contractors, realized that they were having field failures. And we were, a number of us were called together to address that situation. And we decided that the best way to work together was through a council. And that emerged what was called the Surface Mount Council, which functioned for actually 12 years. From that came a number of industry guidances, standards, status and action plan. And really was key to helping facilitate the orderly acceptance of SMT. A couple of us are talking about doing something similar for AI in order to enable it to be facilitated in an orderly fashion. We have had discussions with just plain too many, too many different companies that are either intrigued or frightened of AI as potential users. We have a number of companies who are exceedingly well-funded, who want to offer products based on leveraging what they've developed in AI tools. There's just, the two sides have gotta be brought together in order to decide how to best utilize the capabilities.

Zach Peterson: I think the big theme here is that, and it's something that I think a lot of designers have to realize is that AI and electronics is not just design. But I think for a lot of us in the design-side, the design part of it gets so much attention. And you brought up something interesting, which is that feedback between manufacturing, going back to design and ultimately improving, let's say product quality and yield. I think that's a big area where you have to bring all of these players together in these types of councils and panels like you're talking about. And the one thing I didn't think about, but you've brought up that's really interesting is some sort of standardization around AI. I hadn't thought about that. And I think for things like land patterns and what we do in design, of course it makes a lot of sense. But I'm wondering how does that look like for AI, and is it different for design versus manufacturing?

Phil Marcoux: Well, we're gonna try and touch on that on the March 6th panel. And then a couple of us have already started to talk about where do we go after that, and to try to incorporate some of the other players that we're aware of who have been giving this more thought, and then to solicit, get their feedback and help to decide. We also wanna bring in some of the major EMS players who were already making attempts to incorporate, and say, okay, what have you folks found to be your needs? And then from that, my desire would be to help coordinate a status and action plan. And to make that available to the industry. And get a group of competent people to help make that happen, and revisit it on a periodic basis and make whatever improvements and changes to it so the facilitation happens.

Zach Peterson: Another thing you've mentioned a few times is that some of these companies are exceedingly well-funded. And I think this brings up the question, and it certainly comes up from folks in financial media, is are we in any kind of AI bubble? Are we in a situation where AI is still a solution looking for a problem, and people are just willing to throw money at it until that solution gets married to a problem?

Phil Marcoux: From my experience, clearly that's happening. I have talked with companies who have been funded with no defined product.

Zach Peterson: That's interesting, this feels like the dot-com era a little bit.

Phil Marcoux: Actually, it's quite frustrating. Having had the experience of having to raise funds for my prior companies and making the pilgrimage to what we called VC gulch up in Palo Alto for funding, and then being jilted by younger folks with far less polished, we thought we were going there with nice, polished business plans, anyway they would come out with funding with far less polished business plans than we had. So it was frustrating then. It's frustrating now to have somebody come and talk to us. You know, again, going back to my activities in the Score organization, and say, okay, I received this funding, I need a product, what do you guys suggest? I would say, wait a minute, this is not how it should work.

Zach Peterson: Yeah, shouldn't it go in the other direction?

Phil Marcoux: It should go the other way, yeah. You need a product. Is it appropriate to incorporate AI? I mean, there's activities that are people that are pursuing that really don't need AI. But again, it has the dot-com bubble potential and funding pizazz to it at the moment, yes.

Zach Peterson: Well, I mean, ever since Chat GPT became household name, I've been doing a little bit of VC advisory. And about once a month I get a request to do some advisory for a VC fund. Usually on the East Coast, not in Silicon Valley. But they have been looking at companies that are essentially in the space or in the state that you described where they've basically just included AI in their marketing copy or in their domain name. And now they think that this is a candidate to invest in. And in some sense, I understand, because, I mean, they have to appear to their partners to be doing something as far as allocating capital. But my criticism of that is pretty stark. You're gonna screw up your cap table, and you're probably gonna lose money on this.

Phil Marcoux: On the one hand, it's good because it shows interest and multiple sources for the same activity, which certainly gives people choices. I mean, for example, I actually produced, I used a AI app to produce a little promo video for the panel discussion. I actually tried three different apps, and found one that I was more comfortable with. And was amazed at how quick and easy it was. Another video that I actually produced is for a nonprofit that I am on the board of. And we did, actually we just simply took the PowerPoint presentation and put it into app, and out pops the full-blown video and audio. Now people will tend to marvel at that and say, wow, that's absolutely great. And it certainly takes away the need for a human. Well, yes and no. And let's take Chat GPT as an example. To me, Chat GPT is wonderful in so far as, yeah, you can plug in a phrase that you want to describe or explore, or have blown out. For example, I asked Chat GPT to gimme the list of questions to pose to the panel. And within a minute I got back 25 questions. Now I could take those at face value, or I could read through 'em and say, based on what I think should be asked, and what the direction of the panel should be given the time constraints, I'm gonna edit this thing down. Now, what did Chat GPT do for me? Well, it got me over the potential hump of having writer's block, and sitting for two hours, trying to come up with 25 questions. So it gave me something to edit. In the standards, that work that I used to do with IPC, we always found that getting someone to volunteer to write the original content was really, really tough. But if somebody did that, there were always plenty of people more than happy to sit around and edit it, kibbitz over it. And hopefully as long as you had a good chairman to keep things in order, move things along, and improve it. So I always found that if we had a template to attack and work on, it was always easier to get productive work done. So to me, Chat GPT, these other tools, the myriad of little startups, all they do is help fortify the need for people to learn and understand what the potential value of these tools are. But they also need to be educated in knowing we're not stripping them of their jobs or intelligence. Well, there are some jobs that are gonna disappear, but all we're doing is giving them the opportunity to do many of those jobs better. And for those people that do lose jobs, to move on to learn newer jobs that result from this AI revolution.

Zach Peterson: Yeah, I would totally agree with everything you just said. And especially about things that are generated, especially related to technical knowledge. I always find that you have to read through it, proofread it, correct it. There's something that gets over-generalized or taken out of context, and then you have to go back and fix it. As the person who is frankly more knowledgeable than the AI. Maybe that will change, but I haven't seen any evidence of that, even with all of the hype that's been going on over the last, over a year at this point. One thing I'm noticing here, and it it sounds like you probably agree with me, is that AI in this domain of electronics, is not something where it reads your mind and gives you a design. It's really going to, it sounds to me like it's really going to be very targeted tasks that one AI tool can do really well. And it can give you some time back or allow you to focus on things like quality, allow you to focus more on the front-end engineering, whatever it may be. But I think that's the direction it's going on the EDA side. Would you agree?

Phil Marcoux: Oh, yeah, yeah, yeah. I mean, in the case of, well, two of the representative companies on the panel, I mean the tool that they're targeting is the bill of materials. How do you select suppliers and the packages that best assure you that you're going to have a good source, predictable source of supply, and a reliable set of packages? I mean, that's a critical function that has not easily been addressed in the past because it's relied on usually a different department. And many cases you had to go plead with those people to tell you what suppliers were good and what packages were good. In the case of auto-routing, the enhancements there may not be quite as dramatic. They may be more subtle. But to me, if the designer has the ability to get knowledge from the factory that the product is gonna be built in, and if that allows the auto-router to make subtle changes in the auto-routing that better enhance the yield, then it's well worth the investment and the cost of incorporating that.

Zach Peterson: Yeah, I think that's a big one that folks who focus more on the design side totally miss. Because I think most of the time they look at place and route, and they totally miss the potential application of DFM back into the place and route process, because that's where you really start to get a lot of cost that gets added, whether it's during prototyping or whether you're moving into production and have to now do quality control. That's where some of the costs can start to pile up. But everyone's so focused on automated place and route, and they totally miss some of these points around DFM, DFA, whatever the case may be.

- Well, sadly, their metrics are, they've gotten a whole list of projects. Time is of the essence. Time of the market is critical, speed of design is valued higher than necessarily the quality of the design.

Zach Peterson: So one thing that you brought up earlier as we've been talking was early days of AI. And I'm wondering do any of the points that we've just brought up, things like targeted application of AI, feedback from production back to design, were you seeing those same trends early on during the SMT buildup?

Phil Marcoux: Well, some of them, I mean, thinking earlier, and thinking of one of these situations, 'cause my first factory, we did both design and the assembly. So I mean, at that time there were no other knowledgeable outside sources to go to. So we had to do both. There were a couple of cases. And going back to the idea of running open loop. There were many times that we were doing so many designs that our designers strictly had as their only metric, get the product design, get it to the board fabricator as quickly as possible. If we had to do workarounds because of bad design or design flaws afterwards, so be it. It was cheaper to spend that labor after the board failed, and had to find the workaround than it was to necessarily do the design-checking before sending it to the fabricator. You know, time was too critical on the design side. A little less critical on the assembly side. There were a couple occasions in my early days that brought back big memories, because I physically had to pay for them. Where big projects, the circuit boards were shipped out, came back and there were just so many misses and failures. Missing vias, traces, even traces missing, that type of thing, that was just prohibitive to try and do the workarounds after the board was assembled. And I was thinking what artificial intelligence did I incorporate into our facility? Well, it reminded me of what I call the beer-check. We actually invoked a requirement that before any design could go out, the design project engineer, the fellow that did, the person that did the actual design, had to call in his fellow designers and offer them a beer for every mistake they found in the design. And lo and behold, it was amazing how the quality improved. Now, it didn't necessarily help our medical bills. But on the other hand, it certainly helped quality. Now, it came to an abrupt end when we had a couple of huge projects, and there were just so many mistakes that had been found in a couple of the boards that we realized we were gonna go broke buying all the beers. And so, doing a cost analysis, it turned out it was cheaper to buy some of the newer-fangled design rule-checking software than it was to keep buying beers for everybody. So, there ya go. One of the earliest applications of feedback. And I don't know if you wanna call it artificial intelligence, but it certainly was some form of intelligence.

Zach Peterson: That's interesting, beer-check, I'm gonna remember that one. So, I mean, with this feedback now being implemented, I still feel like in the design side of things, we miss that. We end up waiting until something goes out for prototyping and it comes back for testing or inspection. And then suddenly you realize something you missed. Or you start scaling into your first production run, all the quality checks are done, and then you get that inevitable email from the fabricator or assembler. Hey, we have these lists of problems, and you need to figure it out. And I think it's the "Here's the list, figure it out." How do you transition from that? Do you see that as an opportunity for AI to come in, and kind of connect the dots between here's your list of quality issues, here's what you can do in the design to potentially make it better. And here's a list of let's say five things for each point that can be done.

Phil Marcoux: Yeah, I mean, that's a tremendous opportunity. But it's also a tremendous challenge. How do you get the factory to provide suitable data to build the necessary database that enables that AI-empowered tool to provide the information in a timely fashion? And that's the challenge that I'm hoping that the ongoing effort through the PCEA is gonna try and overcome.

Zach Peterson: Right, it's not just the collaboration between them, right? I mean, we have the tools. And even with like All Team 365, you can now get people into the same workspace. It's now how does the AI even get that data and then create that feedback? So you're almost adding a third system into it in order to enable that part of it.

Phil Marcoux: That's correct, there has been, oh, a languishing effort for a communication standard within the IPC for a long time.

Zach Peterson: Yeah, that would be IPCCFX?

Phil Marcoux: Yeah, it just continues to languish. So, maybe it's gonna take a new effort to, or a new alternate effort, to provide that communication path.

Zach Peterson: That's interesting because I remember reading about, and I think even writing about, IPCCFX, like, I don't know, four or five years ago at this point. And I hadn't really heard anything about it since. And I guess that means that yeah, it's been languishing. Do you think it takes something like AI where now we have a solution that requires a standardized communication methodology in order to push some of those standards forward?

Phil Marcoux: Oh, absolutely, absolutely. And one of the benefits is because we have so many AI companies that have the wherewithal to help push, cajole, yank, whatever, just for their own survival and existence, they're gonna be able to force. And I see them in a great position to help us enable making this happen.

Zach Peterson: Do you think there's a danger of anything becoming like a defacto standard? And I bring this up because I think sometimes one company with a great idea will implement a solution, and it just takes off, takes over. And whatever method they've implemented for getting the data, processing it, outputting it for EDA, that becomes a defacto standard. Even if it's not an IPC standard, it's still something that everybody uses. Do you see that as a danger or a problem?

Phil Marcoux: Oh, that's a potential problem, yeah. But I mean, that occurs when there's an absence of participation, adequate participation by the industry.

Zach Peterson: Okay, so even though this may be a solution provider that's plugged in with all the EMS companies and the EDA folks, that's not enough. You need like external, you need like more oversight from other folks in the industry who didn't necessarily develop the standard?

Phil Marcoux: Yeah, I mean, one of the beauties of having so many companies that are funded to implement their AI tools is the fact that many of them are going to say, wait a minute, my solution's a little bit better than the other guy you're giving the gauntlet to. And if we're doing, if we're serving the industry properly, we will be open to allowing them to offer their solution in addition to whatever we consider to be the main rider. As an example, going back in history again, there was a period of time when one particular SMT package was considered the most prominent, the PLCC. And I was part of the semiconductor company that had created the SMT, the SOIC package. And there was a period of time when the SOIC package was attempted to be pushed out of the way in favor of the PLCC. Fortunately, that didn't happen. So again, because of the multiplicity of different players and different needs, the idea of having multiple surface mount IC type packages was given the chance to be heard, and survive.

Zach Peterson: Yeah, now we have all of these standardized packages that people can build to, and it makes things like part-swapping much easier, especially when-

Phil Marcoux: Yeah, although I will say one wrinkle, there was also the wrinkle that there were competing SOICs that one was metric and one was English, or imperial standards. And that gave us heartburn for a while.

Zach Peterson: So we're getting towards the end of our time, but one thing I wanted to ask you before we wrap up is just, what do you see as the future, looking maybe five or 10 years into the future? Whether it's from the EDA side or the manufacturing side, or even like the analysis side. I bring that up because some simulation tools are implementing AI as kind of like a situation-specific design tool to help you narrow into something that's gonna give you the best simulation results.

Phil Marcoux: Yeah, what I see, let's say five years down the line will be every machine that's is used in the factory floor will have some type of data gathering and data communication device built into it. And have some ability to communicate using a standardized machine language. Some cases the machines could have little tiny, tiny ML devices that are quite cheap. But anyway, it will enable those machines, even some of them doing mundane processes to communicate back what it is they're doing and how well they're doing it. And on the opposite end of it will be a repository, a database that will gather that, assemble it into a fashion that the AI tools can then utilize to feed upstream as far as design. And help create products faster with greater reliability and greater quality when they're being manufactured. That to me is what could happen as early as five years.

Zach Peterson: Okay, well, as all of this stuff develops, of course we're gonna come back to you to talk about it, and help educate the audience.

Phil Marcoux: Yeah, appreciate, it. Yeah, I welcome the opportunity.

Zach Peterson: Great, thank you so much for joining us today. For everyone that's out there listening, we've been talking with Phil Marcoux, advisor for the Printed Circuit Engineering Association. To sign up and register for that webinar on March 6th, make sure to check out the link in the description. Also make sure to check out the show notes. If you're watching on YouTube, make sure to hit the subscribe button. You'll be able to keep up with all of our podcast episodes and tutorials as they come out. And last but not least, don't stop learning, stay on track, and we'll see ya next time. Thanks, everybody.