Transcript
Sam: Welcome to my very first podcast, where we are exploring how AI is evolving in semiconductor manufacturing. Hopefully this ends up at least semi-insightful, and if not, well, you’re here for episode one and you’re part of an experiment. I’m your host, Sam Duchscherer and today I’m joined by our Product Marketing Manager for Productivity, Rich Burda.
Rich, great to have you here.
Rich: So happy to be here, Sam. I didn’t realize this was the first podcast. Happy to be on the first episode.
Sam: I couldn’t ask for a better first person to interview. So, to kick things off, could you share a bit about your background and what really inspired you to step into this current role at Applied Materials?
Rich: Sure, Sam. You know, rather than go through my educational background and all the details, my degree was in mechanical engineering. And after a few early career job switches, I found my way one way or another to a 300-millimeter fab (startup fab). It’s an IBM fab in East Fishkill, NY, just 80 miles from where I grew up—a fab I did not know existed before then—and I was an industrial engineer, hired in as an industrial engineer.
And I guess I was the one either smart enough to raise their hand when anyone said “Had anyone heard of RTD?” and I had, due to some simulation work I had done before. Based on that, I was the lead to figure out how to write a dispatch rule (what a dispatch rule was) for the start of IBM’s 300-millimeter fab. With that, I grew with the fab and really was a part of a lot of really enjoyable projects beyond just getting the fab up and running with dispatching rules, to then getting automated radical transport and dispatching up and under way, which at the time was first of the kind. Then, we got into local area scheduling and optimization, another first for that fab that I was involved with. So, that day I raised my hand saying, “You know, I’ll take a look at RTD,” really led to so many more things.
And just as far as, you know, why I’m here is I personally have just had so much enjoyment and so much pride when I think back on all the work I did in RTD and the system supporting RTD and semiconductor and supply chain. To be later in my career and get the opportunity to be a part of these tools and their roadmap and the people who are using them, it’s just been a really exciting change for me and I’m very happy to be here.
Sam: That’s great. I actually really love hearing people’s background and experience on how they got into this industry, so being an RTD user for that many years is really compelling.
Building on your background though, why do you think this is such an exciting moment for the productivity space? Productivity referring to dispatching, scheduling, planning, and reporting?
Rich: Yeah, Sam, I think that we’re really at an inflection point from a couple different ways. I was thinking first of the operational challenges. Now, it’s the nature of semiconductor manufacturing the technologies are always advancing, and the processes are advancing—continuing to change and evolve. And that continues to go on whether you’re talking about just more and tighter queue time restrictions and all sorts of scheduling challenges. But then, we’re also seeing at this point the need for a higher level of control and automation for all post-fab operations. So it used to be all these challenges were about the fully automated fab. Well, now just the nature of the business that control and automation is needed in post-fab—where things are a little different—so I think there’s a challenge there that’s part of this inflection point.
And one of the other things I think we’re seeing is less availability of domain knowledge and technical skills in manufacturing, and semiconductor manufacturing specifically. So, we’re beginning to see where those skills maybe aren’t as available; maybe it’s a bigger demand or less supply, but we definitely see that as another one of these challenges that’s an inflection point. And now at this time, we have dramatic new IT tools becoming available with AI. So, when I say an inflection point and what’s exciting to me, that’s what’s exciting. We have these new challenges. It’s not just continuing down the path. And we have some new tools that you know, we can maybe use to address those challenges.
Sam: Are you referring to just like large language models as dramatic new IT tools or something else?
Rich: I know, Sam, you live in this area. I’m stepping back; what AI tools make sense, whether it’s agentic chatting with data or reinforcement learning trying to enhance some of the existing scheduling tools—all these. I wasn’t speaking of anything specifically, and I guess that’s why I see it as exciting.
You know, I’m not sure anyone can tell you exactly where we’re going, but we have some tools coming on board which we’re assessing. And when I say we, I mean Applied speaking for the industry. We’re assessing to see, “What can we use, where? What makes sense?” But I think it’s an opportunity to do some things that we haven’t been able to do before.
Sam: Yeah, yeah. I completely agree and definitely the inflection points you mentioned, I would love to maybe talk to you for another hour on all of them. But since I don’t have that time, I’d like to kind of dive into AI.
So, in my experience when people hear AI, they often think of quality use cases around automated optical inspection or anomaly detection. Why—since you elaborated on your RTD experience, even scheduling and optimization—why do you see AI for productivity as something that shouldn’t be overlooked?
Rich: Yes, Sam, here’s where I get to play the role of the cranky old-school operations person. AI is too broad of a term. You know, when you get in these discussions, it can mean so many things to so many different people.
We need to be more specific in the discussions about the AI methods under discussion, whether it’s large language models, reinforcement, learning, agentics. So that’s just a comment when we start this discussion. And I guess given my background, coming from where I have, I see AI broadly are systems that are automated and self-adjusting with the ability to intelligently take actions for optimal results without direct human interaction.
The way I see it, our customers have been building these systems to run their factories for decades now. So there is an opportunity to use reinforced learning and other AI to enhance those areas, but in some ways I think – I don’t want to get in trouble calling our technologies now AI–but I think we use them like a lot of people would like to see AI used. In some ways I think we are probably (we, the industry—semiconductor manufacturing) is a great, great proving ground for these technologies to see how they can be put to use.
Sam: Yeah, so you mentioned it’s too broad of a term. Have you noticed in your experience what customers might appeal to what different type of AI term?
Rich: Yeah, I think a lot of people are confused. I think there’s confusion in a lot of places. I know there are places where you can talk very specifically about very specific projects, but really the area that excites me is the agentic side and being able to talk to the data in a way that may allow semiconductor—let’s say management –to break some of the silos that right now restrict domain knowledge between the various technical disciplines of semi manufacturing. I’m talking about equipment engineering and industrial engineering, the process team, integration team, materials planning, maintenance team.
Right now, each of those have their own human experts, and there’s very little opportunity for those silos to mix. And there are places where if you could mix them, it could help your fab to become very much more adaptable and adjustable to change. So if there’s an area that excites me, it’s kind of that agentic area and the opportunity not only for this specific productivity benefit, but where you can cross these boundaries right now that are not crossed. And ask an agentic AI agent a question about your productivity yesterday, and it’ll incorporate answers from maintenance, and availability, and the WIP flow, and process concerns, kind of crossing those silos. And that’s the kind of vision and where I see there really could be a game changer, really make things a lot more interesting.
Sam: AI breaking down the silos. Agreed, that would have the biggest impact. I guess why do you think it’s been taking a little bit longer than what the industry has hoped for? I mean, you know we’ve talked about silos for it seems quite some time now.
Rich: Yes, some of these Sam, aren’t–some of these roadblocks—are not technical at all. These silos are in place because different organizations can get protective of their territory. And you know, if your organization isn’t set up for that level of cooperation, it might be hard to have that happen.
Sam: Yeah, change management; that’s the keyword that we always talk about. So, other than change management, is there any other key milestones that customers should focus on to accelerate adoption on AI and breaking down these silos, or how do we get there, Rich?
Rich: So I mean specifically on that question, I think a key piece of AI on the operation side is prediction. And I would say a lot of our customers in the industry have some level of prediction but it’s really rudimentary, and I think maybe to set some specific goals for prediction of operations. And some are different than others; some very short term (shorter term), you’re talking about what’s happening by the end of the shift and you might have other tools trying to predict your WIP profile a week from now, two weeks from now, three weeks from now. So I think it’s different for different tools, but I think AI is bringing some of that capability. (Some might have given up on it a little bit a while ago because it was too hard.) I think those are some of the things that AI tools might already be here that can enable that sort of prediction already.
Sam: Why Applied though, right? I mean all these startups, all these other companies are doing AI. It seems like everyone’s doing AI, right? Why is Applied Materials the best equipped to lead our customers on this AI journey?
Rich: Yeah, I think it is. When I say a specialized industry, in a lot of ways, I think we as an industry are leading and some of that is just the nature of semiconductor manufacturing.
I read about other projects outside of our domain and you hear about the challenges of data, and how the data needs to be cleansed, and the quality of the data it just isn’t there to support this. Based on the operations and all the automation in the fab already, I think we take for granted our data is generally—I don’t want to say ready to use, there’s always work to do—but I think there’s some hurdles on data that we’ve already passed. And I think that’s different for a start-up who’s looking at all the different types of industries that they are trying to automate and use their tools on. Here, I think it’s a little bit different because it is more isolated from the data side and the challenges are still there, but it’s different because the data is already coming in. It’s being used in analytical tools to build schedules and optimization.
You know, in some ways we’re ahead in a in a lot of areas. And I don’t know that your average startup that isn’t addressing specifically semiconductor will recognize that and will be bringing the expertise and experience to get from where we are to where we need to be.
Sam: Yeah, I feel like when I was asking that question, I was thinking how I would answer. Luckily, we came to the same answer, so that’s good. Well, if you haven’t noticed working with me, I like to keep things fun. I like to laugh a lot, right? So in in my first podcast, I wanted to do a lightning round of just random questions. Let’s see how you do on these. No wrong answer, obviously.
Rich: Sure, I’m getting nervous here.
Sam: So the first one is, what is one myth about AI in factories you’d love to bust?
Rich: Yeah, I might have kind of touched on this before, but if I would say semiconductor, this is the myth: Semiconductor factories have been slow to adopt AI.
I think they’re the proving grounds for many of the new AI tools. You know, the benefits here will be from putting the power of AI methods to good use. You read about some projects, other AI projects, and it’s sometimes hard to really understand the costs that went in. But you’ll hear these stories and, in order to do the AI projects, there’s all sorts of business process re-engineering and data projects that need to happen for those to come in place. And then it’s hard to separate the benefits from all of that foundational work, then from the AI tools. So, in semiconductor, the infrastructure generally is already in place and we’re ready to focus on how to use these new tools.
I know that was a terrible lightning round answer. You’re looking for quick hitters, but that’s what I got. I’ll try to do better.
Sam: You busted the myth. That was a good answer. All right, next question. This one’s fun and I really am curious what you’re gonna land on here. If scheduling were a sport, which one and why?
Rich: Oh boy, Sam. All right, I’m going to give this one a try. We go to where we’re familiar, right? And I’m going to go to rowing, which is a sport that I enjoyed for many years. And you can picture the image of an eight rowing gracefully, moving across the water. You have eight rowers and you know they all have to be in perfect time. They have to be powerful; they have to be fast, and it’s just very much coordination. And I think that’s scheduling; that’s semiconductor manufacturing. You have these flows that are very complicated when they work. It looks beautiful. It is beautiful. But what people who maybe haven’t rowed don’t understand is that things are going great and then the wind picks up from either the headwind or a tailwind or a sidewind, and everyone in that boat is going to have to make an adjustment. And they might not make the same adjustment, which is going to throw off the timing. And there are all sorts of variables like that. You know, you think about the chop. Is the water flat? Is it choppy? How is the current flowing? So again, maybe more for me than most people, but I just think of trying to keep you know that flow synchronized and everyone working together in perfect timing when you have all of these factors working against you. So again, probably not the answer you’re going to get from a lot of people, but I guess I still translate a lot of my thoughts about things to the sport that I know best.
Sam: Wow, I don’t know who my audience is gonna be on this podcast, but that was very relatable that I think everyone could take something away from. So that was great. OK, the last (I only came up with three. I’m not that creative) but the last question on this lightning round is what’s the most underrated productivity metric that deserves more love?
Rich: Oh, Sam, we’re venturing into legendary IE lunchtime argument territory here. So okay, you know it’s different for different operations, but where I’ve been, the metric that I always went to was furnace batch size distributions. Furnaces seem simple and boring compared to some other tools, but they have a big effect on productivity with their batch sizes. With batch tools, they’re not the greatest for flow in general; you’d rather have single lots or single wafers flowing through, so that has an effect. But if you can get good data on batch size distribution, there’s all sorts of stories at the edges. You know, you find some of these small batches –much smaller than planned—and then there’s a story usually. Why? Why did that happen? Was there excessive time in queue? Was there a hot lot that, you know, somehow had to catch up? So. again, if we’re the productivity metric or report that deserves more love, I’ll always go with the furnace batch size distribution.
Sam: So have you been, uh, in a lunch lunchtime argument then?
Rich: Oh, I was not kidding. You know, you do get a group of six to eight IEs going to lunch maybe once a month. It’s usually not what doesn’t get the most love. It’s pick your metric. You have only one metric to manage your fab. What is it? And you know people argue cycle time and, you know, mask layer moves or who knows. But yeah—I’m not going to say embarrassed to say, more proud to say—yeah, we’ve had some of those lunchtime arguments.
Sam: That’s too funny. Well, that was great. You know, with time, I’d like to wrap things up, but definitely this last question may take a little longer to answer. But if there’s one key take away you want customers to remember about AI in the productivity space, what would it be?
Rich: I think it would be this, Sam, not even so much technical. I think being in the industry for a while and you know, in automation for a while, I think there’s a natural mindset to be a little bit skeptical and push back, and be ready to say that, you know, AI isn’t ready yet. And all of that may be true, but I would ask folks to fight that and maybe turn that energy around to say there is a lot of work in this area, and this is happening and it’s a lot of excitement. And you know, with some creativity, you could lead this discussion and there may be ways to use AI that others haven’t thought of, that might come natural to you where you can really make use of this faster and do things that you’ve always wanted to and haven’t been able to do. And you can do them now with AI.
Sam: So don’t be a skeptic, you know, get out of your comfort zone. Is that a good summary?
Rich: Well, it’s, you know, a lot of us love being skeptics. I would say don’t forget to put on your creative hat and think about how you could use it. Because, you know, give yourself some credit, you may be in a better position to know how to use these tools than the people you see on your social media feed all the time.
Sam: Yeah. That’s a great answer. I’m excited about what’s ahead of us. So, with that, Rich, thank you so much for being the first test subject. I really appreciate your insight.
Rich: Sam, it was fun. Thanks for inviting me and you’re doing a great job.
