Transcript
Sam: Hello, welcome to our podcast. I’m your host, Sam Duchscherer, and today I’m joined by Dan Meier, who is the Director of MES Strategy at Applied Materials. Dan, welcome.
Dan: Thanks, Sam. Great to be here.
Sam: Well, Dan, you are my third guest on this podcast and I’m going to be honest, I’ve done some homework on you. I’m mainly referring to your blogs on our Applied Smart Factory website, right? There was one blog where you talked about the history and evolution of MES from this digitized paper, to where it’s now a backbone that connects different systems in the factory. So, my first question is: What would you say was the biggest inflection point in that evolution?
Dan: Interesting question. So the history of MES, you know really is very, very rich and it goes back a long way. But the first thing that I think was an inflection—and I’m going to mention two—is repeatability. So being able to define a process flow that you can do basically the same thing again and again, even if that’s in written form, paper form, or just verbally communicated in a consistent manner. That’s how things started out, as the ability to define how you do things and do it the same way again and again.
The second thing that I think was a big inflection point and maybe even bigger than repeatability was the ability to automate that repeatability—so it accelerates repeatability. So now if I can repeat something, do it the same way again and again, the ability to do that faster, to be able to do that with fewer resources, to be able to do that on an accelerated timeline—all of that is wrapped up in this evolution of the MES.
Sam: So, I also read from you that integration is key and actually the hard part. So how does that fit into repeatability and automation?
Dan: I think integration is really, really important. I spent a lot of years as a factory operations manager, so when I think about how from a day-to-day perspective the MES and that integration across the MES impacts the manufacturing team, you know, I think of it from a perspective of all the moving parts that the MES can have. If you think back—I don’t know, this is probably 20 years ago—Microsoft Office wasn’t always Microsoft Office. They were only separate applications. You had a word processing application and a spreadsheet application and you could work with them together, but they all looked a little different, and they all felt a little bit different. And at some point, Microsoft integrated those into the Office suite, and evolved those so that, for example, when you change something in a toolbar like a font size, no matter what system you’re in, what application you’re in in Microsoft Office, it looks the same. It feels the same. It works in exactly the same way. So to me, when I think of manufacturing software, the Holy Grail is that all of our systems would be able to work together in the same way to reduce the cognitive load, to reduce the training time, to make sure that the manufacturing folks on the factory floor can do something in one system and expect it to work the same way in another system as well.
The second part of this is really about data—data flowing logically between systems. So back to the Microsoft analogy: Word and Excel way, way back when, were two separate applications and you couldn’t necessarily exchange information between them. But today you can, for example, embed an Excel worksheet within a PowerPoint and you can actually embed the whole PowerPoint within Word. So, to be able to—in the MES in our manufacturing software—to be able to have data that flows between our systems (MES data that flows into the scheduling and dispatching systems, for example, MES information that flows into the SPC system and the product quality systems, for example), that logical flow between the systems, I think, is really, really key. So to me, that’s good integration. That’s what the manufacturing team is looking for on a kind of day-to-day factory basis. It really provides benefits across the entire manufacturing space.
Sam: So, I loved how you brought in productivity and quality into this discussion and tied that to MES. You gave some examples, but are there any others where you see MES directly moving the needle in those two areas?
Dan: Oh, interesting that you mentioned that. I was just thinking about this earlier today. The benefits of the MES that I see, you know, the things that really kind of get me out of bed every day really benefit things at the opposite ends of the scale of manufacturing.
The space that we play in most of the time really is regarding full automation in high volume manufacturing facilities. So, this means fewer people on the factory floor and more consistency across the factory from a behavioral standpoint, from a data standpoint, from an operational standpoint. So to be able to have all of these systems, it’s really interesting because if you think of how you would automate something, once you start running the automation you start running into these corner cases—these exceptions, these boundary conditions—that you hadn’t thought about before. And so what we have to do from an automation perspective is to look at all those boundary conditions and be able to resolve that in order to be able to make full automation something that’s real and not just, you know, a fancy name that looks good on a PowerPoint.
To me, pushing that direction of full automation is really an exciting thing. But the flip side of this as well is how we can bring these big factory capabilities to smaller factories. One of the things that we’re working on right now is being able to improve a lot of our user interfaces. For example, I want to be able to define process flows in a more (I guess you would say) visual sort of manner, to be able to do that in a way that people are used to now rather than just in tables and tables of data or rather than in a static list of process steps and things like that—do things in a more graphical manner in how people are used to working in other areas today. Great; that works well for large factories, but to bring those kinds of capabilities down to small factories as well, I think, is really very, very important.
Sam: Yeah, seeing is believing. I love that example of how the process flows in a visual manner. But where do you see AI in that picture—in this big factory capability to smaller factory?
Dan: Wow, AI is interesting. So, there are just so many layers of this. It’s interesting because I’m going to digress for just a second to talk about how we define MES and how companies in general define MES. We tend to think within our product family as MES in a pretty narrow scope. I can define process flows, I can define lots, and I can track lots through those process flows. And that’s pretty much the limit of what the MES is all about. But there’s a lot of interfaces to other systems.
We know that within the SmartFactory products, we also have productivity applications. This would be maybe your scheduling and dispatching systems, for example. We also have product quality applications, so think of SPC and fault detection and things like that. So, these are systems that are working closer to the actual data, but they don’t have necessarily as much to do with how that data flows through the entire factory. Think of it as vertical processing rather than the MES being horizontal processing of lots as they progress through the factory. But that means that the MES needs to have interfaces to all those other systems so that, for example, if you set up a sample plan in the wafer fab, I’m not going to run all lots through every metrology step. That doesn’t make a lot of sense when I can create a statistical sample plan that gives me a statistically significant confidence that I can have good product quality even though I don’t measure every lot, right?
So where does that come in? Well, typically MES has a knob that says, okay, for this particular metrology step, run 20% of lots and 80% of lots don’t get that particular metrology step. Well, there’s a lot of analysis that goes into that to determine what the right percentage is based on your run rate, based on the number of lots that you have in the factory, based on the amount of time that you’re going to be running the process, etc. But it’s static. So what happens if the manufacturing conditions change? You have to go back and manually update things. To me, it would make a lot more sense to be able to have AI in those product quality systems and in the productivity systems that are looking at those conditions in the fab, looking at the data that’s coming from the MES, and looking at the data that’s coming from SPC and from dispatching, to be able to automatically make those decisions about how to adjust the sample rate and feedback that to the MES—and have the MES update in real time how many lots need to be sampled.
And that’s just one of many kinds of examples where the MES really provides the interface for action, but the AI that’s behind that really kind of needs to be done elsewhere based on a whole different set of domain and data.
Sam: Do you see that example kind of being this human in the loop starting out … where that sample plan feeds into MES where they might accept or reject it at first?
Dan: Absolutely. I love that term human in the loop. We really have to build confidence in any of our systems. And I know that going from just, you know, back in that history of the MES having a set process flow, as you handed that off to somebody else, they had to figure it out, and they had to have built some confidence that yes, this was the right thing. It was going to give me what I expected in the end and even though they didn’t have as much flexibility in the process steps, I was going to be able to improve the rate at which I do things.
So, it’s no different than that. This is a new technology, a new expectation. Over time, we have to build confidence in this new system before we’re able to really turn it loose in an automated standpoint. We already do that today—again, in terms of product quality, advanced process control, what we call run-to-run capabilities. It’s automatically using metrology information, feeding back to adjust process recipes in pretty much real time for things, so we already have the tip of the iceberg. We’ve already gone below just the tip of the iceberg with this, and I think there’s just a whole world of possibilities that lay below the surface with AI and what we can do to optimize fab processing.
Sam: So, what naturally ties with AI and MES are digital twins, right? Where do you see digital twins becoming truly useful to MES from your point of view?
Dan: It’s really interesting to see digital twins kind of as just another report. We use tons and tons of reports in the factory today, and reports tend to be somewhat siloed. I’ve got my diffusion reports, I’ve got my production reports, I’ve got my equipment engineering reports, etc. And typically, the people who are looking at the reports usually look at their reports in their domain.
But digital twin is not just reports. If you’re thinking of digital twins from the perspective of a virtual representation of the factory floor just as if I was standing at the ceiling of the factory and looking out across the factory floor, I can see the entire factory. And a digital twin should be virtually able to give us that kind of information, whether that’s from a physical representation or simply from a data representation. You know that’s not an integration, but an implementation issue. I can imagine a number of different scenarios, but the importance of all of this is that we can take all the underlying data that used to be in those siloed dashboards and we can begin to integrate that with the ability to explore beyond the silos.
How often do we find something within our domain that we’re required to go out of the domain? In the fab environment, you know there’s always that natural tension between photolithography and etch. The photo people say, “Oh, it’s an etch problem.” The etch people say, “Oh, it’s a photo problem.” But what happens if you could view each other’s data to be able to find out, “All right, what’s really happening here?” Maybe it’s neither of those problems. Maybe it’s a CMP problem, maybe it’s a diffusion problem. So, to be able to explore beyond your silo, to not have to make those initial opinions based on that siloed limited set of data, I think the digital twin can be really, really useful for it.
The other thing I think is going to be really great about digital twin is to be able to ask questions of your data. If I can ask questions in just a natural language, “What’s my cycle time trend?” “What’s my yield trending?” I think that’s really, really a very, very helpful thing.
Traditionally, we ask our IT folks to be able to come up with a new report or a new dashboard (something like that). And you know, there’s a fairly long cycle time to be able to get a new report completed, and even then, you know, it’s not everything that the process engineer who requested it wanted. So there’s that iterative cycle and it’s okay, but it takes too long. Well, what happens if you can just ask the question? So instead of weeks to be able to come up with a new dashboard, we can take minutes of typing in a question, seeing an answer, and typing in follow-up questions or asking follow-up questions to be able to get to what we want.
One more piece about this that I think is something in the future and what’s really exciting to me is when we integrate agentic AI into the conversation. I love the idea of being able to say, “Well, what’s my cycle time trend?” But I love even more the idea of being able to say, “Why is the cycle time trending up?” “Why is my yield trending down?” And again, because we’ve got this integration of data across the factory, to be able to have AI that’s looking at all of this data not just to notice the trends and to surface the trends, but to begin to say, “Okay, this is happening because that thing is happening in the fab.” I think that’s really important.
I was visiting a customer site a couple of months ago and they had asked before we went into the customer site, “Hey, can you tell us what you’re doing with AI these days?” The production manager in the meeting—as we were starting to talk about AI—he stopped, interrupted in the middle. He said, “You know, I don’t need AI. I need ‘A-Why’: Why are these things happening? Why is cycle time getting worse? Why is yield getting worse?”
And to me, that’s really the promise of what we want to be able to do with AI in our environment, is to be able to integrate all the information and have AI that not only looks at the trends but also suggests the ‘whys’ behind them and from a predictive standpoint, what to do about it.
Sam: So in your example there, I think it went back to one of your main points about how integration is key and actually the hard part. And when you’re listing the AI examples, I’m just curious. Do you see AI helping with the physical integration aspect of it?
Dan: Yeah, this is really interesting because I think people expect all of this stuff to be an event. That is to say, I’ve got a product, I buy the product, I install it and boom, I’ve got, you know, a magic wand. But where we are today with the technology (even though the technology is evolving very, very quickly with AI, with digital twins and things like that), it’s a real process to be able to integrate that in an optimal way within a factory.
So right now, within our SmartFactory product family, we have quite a few AI solutions that I would call point solutions. That is to say, we’re optimizing run-to-run control to optimize process flows and get SPC charts centered up again. We’re optimizing cycle time. We’re able to predict based on actual fab conditions when a lot will finish and things like that. So, these are point solutions that are optimizing these local areas. But optimizing point solutions doesn’t necessarily optimize entire fab output.
And so, I think that’s the thing that, as we evolve these point solutions, to have something in the future that is an AI orchestration engine—that’s orchestrating across many different AI point solutions—to be able to come up with that fab-wide orchestration, that fab-wide optimizer that may say, “Okay, well, we need to focus on optimizing this area, but do not focus on optimizing this other area because it doesn’t improve the factory output; it doesn’t improve the factory quality.” And those are the things that, you know, really it’s time and experience and a lot of data that will lead us to creating that kind of solution. It’s not going to be something that we do overnight.
Sam: Yeah, I’m definitely excited for that holistic vision, though, that you painted. And I know we’re doing some impressive stuff. I’m sure if I had another podcast we could talk for another hour, but let’s …
Dan: We should, have a podcast interviewing you about all the AI activities that would be interesting.
Sam: Yeah, but let’s switch gears and get to my favorite part of the podcast, which is actually the lightning round where it’s just quick questions, quick answers. Say the first thing that comes to mind. Does this sound good? This is how I like to end the podcast.
Dan: I’m terrified.
Sam: All right, so the first one is, if you’re a fab leader listening to this podcast, what are some practical steps you can take to modernize your MES without boiling the ocean?
Dan: Yeah, that’s a great question. So, changing the MES is, if you’ve got an MES already, changing to a new one can be a years-long process. It entails a lot of time and a lot of risks for factories. So being able to formalize the processes that you have within your existing MES is, I think, one of the big things that can be done—again, without boiling the ocean. Just making sure that the steps are correct, making sure that the escapes from those steps are defined. For example, I’ve got a metrology step. What happens if the metrology is good? We typically handle that kind of case very well. But what happens if the metrology is bad? Do I rework? Do I scrap? Being able to formalize that kind of thing, the conditions, the rules, so that you don’t have subjective opinions happening in the middle of the night or on the weekend within the factory. You have everybody doing the same thing the same way. To me, that’s I think the better thing. It’s really just getting back to basics and making sure that everything is formalized within the MES and how to process.
Sam: Good answer. Second one, what’s the MES capability you’d never give up once you’ve had it?
Dan: I can think of a few capabilities that I’d never give up if I ever get them. But that’s what product roadmaps are for, right? To get to that point. For me, reporting I think is one of the things that I would never give up of what I have today, being able to see what’s happening in your factory and whether that is simply to see the state of the equipment that are there, to see the trend in yield, to see the trend in cycle time, just that high-level reporting to be able to see the factory condition I think is the thing that was, for me, the thing that I would take to the island.
Sam: Alright, last one. And this one, like I said, I’ve done my homework on you. You used to play the French horn, correct?
Dan: You know, in a past life and a past career, I was a professional musician in a symphony orchestra.
Sam: Yeah. So, this last question is, if you had to explain the value of MES using a music analogy, how would you describe it?
Dan: It’s a symphony. That sounds like an odd answer, but when I think about it, the MES is really the orchestration engine for the factory. It really guides and determines what happens with everything in the factory and how you build your product. And when I think of a symphony orchestra (I’ve played in the symphony orchestra for many years professionally), it’s the same sort of thing. You’ve got a lot of moving parts. You’ve got a lot of individuals who are doing their own thing. But, the conductor up at the front is bringing it all together to make sure that you have one harmonious result that your customers (the audience) will like. So, to me that’s it with the MES. The MES has to really penetrate the entire factory, orchestrate the entire flow in order to be able to get a quality product out in a timely manner in a way that that pleases our customers.
Sam: Well, Dan, with that, I just want to say I appreciate your time and like most guests, you were truly insightful.
Thanks, Dan.
Dan: Well, thanks for lots of thoughtful questions, Sam.
