How Texas Instruments Builds Foundational Semiconductors and What They Enable
Channel: Alex Kantrowitz
Published at: 2025-12-23
YouTube video id: qPXZ23XddZ4
Source: https://www.youtube.com/watch?v=qPXZ23XddZ4
Let's take an inside look at the semiconductor manufacturing process here in Sherman, Texas, as Texas Instruments opens up a new fabrication plant. And did you know that semiconductors are actually TI's core business, we're joined today by Jeff Moroni, CTO for power at Texas Instruments in a conversation brought to you by Texas Instruments. Jeff, welcome. >> Thank you. Thanks for having me. So, we're here at the opening of a big Texas Instruments semiconductor plant. >> Let's start broad. Can you describe the state of the semiconductor supply chain? Where are they being built and do we have enough? >> Yeah. So, let me first start actually with a um a little bit of a broader view of what's happened in the last 10 years in semiconductors. I think it'll it'll answer your question. So, I would actually venture to say that 10 years ago, the average person probably didn't even know what a semiconductor was. And I think a lot has changed over the last 10 years. And people realize for a few reasons. One, because it's in the news. But two, because semiconductors are in our lives everywhere. And you start to appreciate it more and more when you can touch and feel it in your cars or in your cell phones in your um you see the robotics and the AI um infrastructure and I think it's important to understand why is that and really there's been since semiconductor inception two things that have I would say been constant ambitions miniaturaturization and making devices genon more affordable and that's by the way actually TI's passion statement is We want to make semiconductors more and more affordable. But the byproduct of that is that they show up everywhere. Every year semiconductors get more and more affordable and they show up in new applications and they solve new problems. >> Right. Like the smart refrigerator. >> The smart refrigerator. It's everywhere. Yep. Wireless uh washing machines. That's right. And uh >> I actually have thought a bit about this. I'm not sure if I'm uh if I'm missing any industries, but I can't think of any industries over a long time scale that have been anti-inflationary other than semiconductors. Semiconductors now are cheaper than they were 20 years ago and that is because of the scale that we can drive in semiconductors. It's because of the miniaturaturization. It's because of facilities like this. So if I kind of come to the supply chain piece then there's really two categories of semiconductors. There's the foundational chips which is what's built here and then there's the sort of leading edge nodes that you see in GPUs and CPUs and um and you know those kinds of products. And I think semiconductors probably also became very familiar for people um during the semiconductor supply um crunch during the COVID cycle. Everybody started to realize, oh, I don't I can't buy a car or I can't get this because there's no semiconductors. And people started to learn what semiconductors were. A lot of that supply chain challenge actually came from foundational chips. Chips like the chips that are built in this factory that TI builds. And the reason is that you you hear about all the investments that go on around the world in the US including on the leading edge nodes for processors and for AI applications. Uh but a lot of the um the um supply needs goes to the foundational chips that support all of these functions in the world and not very many people actually invest in that. Uh so TI is one of the few and we're definitely the biggest in terms of investment. We've announced $60 billion I think over the last few years in investments in foundational chips. So for me the supply chain and the sort of um the the situation that we have we see more semiconductors in the world. We see more need for foundational chips and there's fewer and fewer people building kind of purpose-built capacity to support these chips. And so it's I think a very unique opportunity. It also serves to show I think the criticality of some of these technologies that we build. >> Okay. That's very interesting. So when we heard about this supply chain crunch for when it came to semiconductors, I think everybody like had in their head, okay, they're being manufactured outside of the US and they are the same. The semiconductor in a computer or in a data center is the same as the semiconductor in your car. They're very different though. >> Yeah, I mean I think there's a lot of different functions. I mean certainly there's processors in all of those applications, but none of them work without the foundational chips. And these are really analog mixed signal chips. They're um embedded processing chips. Chips that serve the function of powering a processor or serve the function of um providing a signal chain so that a real world sensor can kind of um get translated into something that can show up in a you know in a haptic interface for your phone or things like that. So there's tons of analog around us and they all require these foundational chips and it's very different. The technologies that are required are completely different when you compare those two categories. >> So we're here in Sherman, Texas. Yeah, >> this is day one of the production um of your new facility to manufacture >> this this chip. >> Yep. This is uh there's many many chips in this. So, this is a 300 millm wafer and inside of this uh depending on of course the size of each chip, there's can be thousands, tens of thousands, hundreds of thousands of even chips on each one of these individual 300 mm wastes. >> So, you're telling me that something this size can hold hundreds of thousands of chips? >> Hundreds of thousands of chips. exciting. It's a funny story actually, but um I forget exactly when this was, maybe five or six years ago. We ended up building a chip small enough that it broke um one of our systems. It was a Y2K moment. You know, you go from 1999 to 2000. And the system wasn't designed to handle as many chips as we could fit in this. And so, we had to actually redo the software so that we could handle hundreds of thousands of chips in our system and the way that we track it for a wafer like this. >> This is a dumb question. Why are they built in a circle? Because I I think of chips as a square, but this is a circle. >> Yeah. There's a long uh history of that. So they're actually singulated. So they're cut up into squares and they're packaged that way. A lot of the semiconductor tools just from history have always been um built to accommodate circular or circle wafers. So these start with ingots. The ingots are built into circular silicon wafers. And then we have processes for example where we spin on different materials. And in order to spin those materials on it's much more natural to do it in a circular pattern than a square. So there's a lot of reasons why it's been that way from the beginning, but yeah, ultimately the product becomes a square when we cut it up. >> Okay. So I I just talk a little bit more about the new plant. >> How long did it take to build? Um your CEO uh was just talking a little bit about the amount of materials. Uh apparently there's been enough concrete poured here to build a highway from New York to DC. Yes. >> So talk a little bit about the undertaking and then what the chips built here are going to enable >> building the building itself. Of course, there's all of the work that goes into actually designing it and um making sure that it passes all of the standards that we require from a clean room perspective because these devices require ultra clean environment. >> Why? >> Because any contamination causes us to have to scrap units and so I mean we call it yield but eventually it becomes a yield loss. So we want these chips as pristine as possible. And there's been decades of work going into making sure that clean rooms are as clean as possible so that the devices we build ultimately work and yield. So all of that infrastructure, all of the design, all of the planning that goes into it. It takes time even just to build the shell. And so three, four years is not uncommon when it comes to this. So one of the reasons why I think it's so important in the supply chain commentary is that we don't know what the market is going to do. we um we all we know is what we can do in the market to perform and if the market wants to buy a lot more chips because there's um there's you know a new application a new demand we need to have the buildings ready we need to have the tools ready so that we can support that and so that's I think to me just the why it's so hard to achieve this scale because you have to build this in anticipation of demand by the time the demand is here if you don't have a facility like this you're too late and so I think that to me is um is very exciting It's also a very unique opportunity and what it enables. I mean, just take a car for example. I think in today's cars depends, of course, on your model, but uh it's not uncommon to see over 3,000 chips in a car. Most of them are foundational technology chips. And >> what are all those chips doing? >> Yeah, exactly. So, I mean, you move your seat, there's a motor drive. You move your window, there's a motor drive. There's radar that um can detect um objects outside for safety. Those are foundational chips. There's LED headlights that they can individually dim. So you can beam steer, you can illuminate, you know, maybe a dog running across the street for safety purposes, that function is all enabled through um foundational chips. So basically any electronic LED lights that illuminate the inside of your cabin for aesthetics. U audio amplifiers in your car so you can hear, those are all foundational chips. So basically everything that happens inside of that car is supported or enabled in some way by a foundational chip. Those are all technologies that are built in fabrication plants like this. There's new applications like robotics. People see, you know, robotics, you know, these humanoid robots walking around, motor drives, sensors, interfaces so that they can feel, so they know how much pressure they're applying. Uh vision systems, all foundational chips. >> And so this new plant will produce tens of millions a day. >> A day. Yes, that's right. And there's actually I mean we're in one of them, but there's going to be four here eventually. So we're we see sort of a scale of um of um a quarter of what this eventually will be. And the scale of that I think is just extremely important for the reasons I mentioned earlier. More and more semiconductors in our lives, more and more foundational chips, fewer and fewer people investing in facilities that can support foundational chips. So we think the scale of this is actually a huge competitive advantage for us. And it also helps serve our customers. We're able to give them the devices they want when they need them. And you know, it's interesting. We're sitting here at Texas Instruments. Um, one thing I learned recently is that semiconductor manufacturer is actually the biggest business that TI has, bigger than the hardware that you guys make. >> Uhhuh. Yeah. We uh I I kind of joke actually I even joke with my sister-in-law, but you hear Texas Instruments and a lot of people think calculators, but we make a lot more than uh calculators. And uh the the the business that we have here is actually a long kind of evolution of investments really over the last 30 years. We um we started building these um semiconductor chips and I would say that it started actually with the advent of the integrated circuit which was invented by Jack Kilby who worked at TI and the story is that he had a summer internship where he was sort of left to um left to do what he wanted to do and during that summer internship he invented the integrated circuit and that integrated circuit became the backbone for everything I just described when it comes to the foundational chips. He won the Nobel Prize for it, I think, a Nobel um science prize for it um a few decades later. But that is where this all started and you can kind of see now what what the world looks like as a result of some of those inventions. >> Okay. Um you know, we kind of made a distinction or you made a distinction earlier that there's two types of chips uh and there's like the foundational chips and the other type that you might see in an AI data center. >> Yeah. But maybe that world merges, right? Because um the way that the AI industry is moving, it seems like the models are becoming more efficient. Um there's a trend in AI, the small language model, right? Maybe you don't have to fire off all the world's knowledge to be able to, you know, get an LLM to do something for you. Maybe you can have something that's purposely built, purpose built. >> Yes. >> Maybe in a car, for instance, you have an LLM or a small language model. uh that can deal with car functions >> and that's going to run many of those will run on device. >> Talk a little bit about how you know some of those tens of millions of chips a day might end up powering some of these uh AI ondevice activities and whether TI and whether you anticipate that to be a growing business line here. >> Absolutely. So actually there's two categories but they're actually similar. they're just um different in the implication it has for the foundational chips. So let's talk about data center first. So um I would say that the single biggest challenge that we face when it comes to the trend in um AI is power. There's a macro scale of actually grid level power in the infrastructure required to support the grid level power. And then there's even the micro level. If you actually look at these processors, they're about this big. >> And um if if I >> AIP, >> an AI processor. Yeah, it's about that big. I mean it's you see some of the presentations and it's like they're holding what looks like a Grammy award. >> You know what the rest of that stuff is around it? It's power >> really. >> So there's you know on some of these um particular products there's maybe one two or three of these AI processors and then it's on a big board and that big board has all kinds of stuff around it. A lot of that is power management actually. And um the way I try to describe it actually just to to sort of make it hit home. Okay, so one of these chips, these processors, it can take somewhere on the average of two kilowatts, which is not an uncommon amount of average power for a household, a full house in the US to take. It's in this amount of power or in this size >> like per day or like per per minute per minute >> instantaneous. So I mean it can run that much power instantaneously. >> Okay. >> And it actually is a very dynamic system. It goes from it can go from no power to that two kilowatts and it can do it hundreds of thousands of times faster than you can blink. And while doing that the power that is given to that chip needs to stay tightly regulated so that the billions of transistors in there which by the way are also tens of thousands of times smaller than a human hair. They all work properly. And in order to do that power management becomes the biggest challenge just at a micro level. The chips that do that are built as foundational chips in factories like this Sherman Sherman factory. And I think that that to me is maybe the most exciting area to be when it comes to foundational chips and power right now. >> Okay. So, so in AI data centers, the chips built here are being used to regulate power going to those bigger. So the the chips here are part are part of of the data center picture. >> That's right. They don't the data center picture doesn't exist without chips like this. And what about on device then? >> So on the device on the processor itself. Yeah. So on the processor itself there is um there is uh a lot of um of computation a lot of um of additional circuitry that's in there. But for the that all is built on what we call the leading edge nodes. >> Okay. >> And so all of the stuff that powers it comes from the foundational chips and then uh and then the processor takes it from there. >> Can I ask you so as the AI buildout has exploded? I think that word doesn't even do it justice. >> No. Um has has the demand I mean obviously if your components are being used to regulate these data centers like have you felt that? >> Oh yeah. >> How have you felt it? >> I felt it in I would say one distinct way um the pace of innovation. So you you know we go to these customers I personally go to our customers we'll stand at a whiteboard and we'll think of ideas um focusing on what their problems are because they don't necessarily know what the solution is but they all know what their problems are and a good idea comes out. It could be from us. It could be from them. It could be joint. The next day they're ready for it. When can we have samples? When can we have a proof of concept? And I think it just goes to show how big the challenge is when it comes to power specifically. They're hungry for solutions. They really need the solutions. And if they don't get them, it stalls their ability to build out the infrastructure. And so I feel that very real in terms of just the pace that we have to move with them in order to solve the problems that that um that are coming. There's also, you know, the kind of um the manufacturing side of it. They have very high demand when it comes to a couple of things. One is just how much you can supply, how many units you can ship to them. And the second is the quality. Uh if you just imagine that one unit on this board goes down, that rack could go down and it's a very missionritical application. And so reliability, quality, those are very very important to them. And that's where owning your own manufacturing actually becomes a huge advantage as well because we control everything from the technology to the number and the types of products that we build to the quality that comes out of the fabrication facility. >> Okay. So you're you're involved in power in so many different ways. Yes. Um and and here you really have this opportunity to figure out how to power a data center, right? The one that's just come online. >> Like you you've already given a little bit of an indication of the undertaking that it takes to do that. >> Um I was listening to a podcast recently where you know one of the leading tech CEOs was saying that >> he has the GPUs he needs but he can't get them into warm shelves uh because power is such a challenge. So can you just describe the scope of the of the issue here with power? How big is it? um does do you anticipate that the resources are going to be there to scale up and and then drilling down a little bit >> how are you able to do it here? >> Yeah. Uh absolutely I anticipate that and um I think it's already happening. So start at the macro level actually um if you look at the global energy consumption it's actually um increasing at a rate higher than it ever has historically in let's just even the last year >> right the the power I've been speaking with some experts about this actually demand for power was relatively stable >> for many years coming into I think 2023 or 2024 >> and now you can see the increase in the numbers which is like it's it's a big increase it's not like a expon exponential or hockey stick, but seeing it has like all of a sudden the antennas have gone up among people being like, "Oh, >> here we go. >> Here we go." And you know, um another supporting fact there is that um the population rate in the world has not been increasing at the same rate as it used to. >> So that means that each person is consuming on average, if you normalize it that way, more power than they were 10 years ago. And where is it going? So it's not going to things that scale with population size. It's going to electronics. It's going to these processors. It's going to electric vehicles that need to be charged. It's going to the entire infrastructure around it. So, I think the fact that power consumption is increasing at a rate faster than the um the population suggests that more and more electronics are out there. They're consuming more and more power and it will become a limiting factor. So, I I think there's probably two roles that we play in that story. And I kind of think of it as like a numerator and a denominator. On the numerator you have how much energy can we produce? How much energy is generated at a grid level and there's lots of opportunity there when it comes to especially diversity of energy sources. There's of course solar, there's wind, there's all kinds of additional more decentralized energy sources. There's also batteries which are a massive part of even the installed electricity base in Texas. And the beauty of these diverse energy sources is that they have they they can be used at different times. You can charge during the day, you can charge a battery during the day, and then at night when everybody's got their ovens on, you can discharge that battery. And so you get, I think, a dynamic in the system that helps this problem. All of those energy sources, all of these things I talked about require power management, power processing from foundational chips. I keep using that term. And so that's the numerator. There's also a denominator, which is that once we have the power, how are we responsibly processing it? So that power that comes from the grid can safely power a processor with as little power loss as possible. And that's where high efficiency, high power density, these are kind of terms we use in the industry comes into pay into play. And that's really built through the foundation of these technologies. We have better, faster, smaller devices that go on chips like this or wafers like this that consume less energy and can um generate less heat. And so we kind of play a part I think both in the generation and then also how we process and manage the energy that we get. >> It's very interesting thinking about this this power as a limiting factor. >> Yep. >> I don't I don't know if there's a real solution yet for how enough of it is generated for the demand and how that's done sustainably. But >> um it is interesting here you suggest that >> there is there's a a generation so like generating a power solution >> but also a technology solution here. Absolutely. Yeah, I think that I mean ultimately there's always going to be a limit where the amount of power you generate will become your limiting factor. But I think like I said the exciting thing there is that there's such a diversity of energy sources available that um can be used in different ways and are almost I would say required in different ways depending on the application and depending on what you need the power for. And they all require power processing. They all require sensors. they all require, you know, um things to keep the the system regulated properly and safe. And so there's another reason why we see more and more semiconductors in our lives now than we did 10 years ago. And I don't think it's hard to guess that 10 years from now we're going to see even more. >> Oh yeah. Um early stage research. >> Yes, >> it is important here. Uh, another thing that I've learned recently, um, talk a little bit about how, um, TI, talk a little bit more about TI's process for encouraging early stage research and then how you take that and turn it into applications. >> Yeah, it's it's so important. Um, you know, ultimately we're a technology company and so we live and um, survive based off of our technologies and I think to me where I always start and where I think you have to start is the customer. You have to start with the customer in mind. What problem do you need to solve with them? And you know, I've said it before, but you go and talk to these customers, not all of them know what they want, but they all know what problem they have. And if you focus on the problem, that's where the magic happens. You can start illuminating a problem and know that it's a real problem. And you get a diverse set of people, diverse set of skill sets and different capabilities into a room to think about that problem. And a lot of magic starts to happen and ideas are born. And I'd say that's the genesis of a lot of early stage research even at at TI. I think from there where where it gets more challenging and I think even more interesting is usually where we think the idea is going to end is not where it ends. It's research. It's got um unknowns and we have to learn and you know I say that 10 you know you want to see 10 feet in front of you if you have a flashlight that can illuminate 10 feet. If you want to see 11 feet the only way to do it is to take a step forward by one foot. And I think that that's how we think about research. You have to move. You have to go. You have to try. And you're going to learn. You're going to pivot. When you pivot, you're going to maybe go in a different direction. But the worst thing you can do is is is stay still. So once we have the problem, we encourage and we I think facilitate try it, see what happens, and learn. And as long as we're persistent, good things come out of it. So you you have a lab here, Kilby Labs, that's meant to do exactly that. So what is the process that Kilby Labs uses? >> So um Kilby Labs, as I mentioned, Jack Kilby was our uh kind of the inventor of the integrated circuit. So we're named after Kilby Labs. Uh we're named after Jack Kilby and um we do the advanced research for TI. So I'm actually a part of Kilby Labs and we're really chartered with developing the high-risk, highly disruptive technologies across all of the areas that TI supports. So that's our charter in kind of um in simple terms and it spans all of these different foundational technologies, foundational chips that I've mentioned. I think that the interesting thing, you know, I've been in Kilby Labs for a long time. We've learned a lot of uh a lot of things over the years, even outside of the technology. And um I think that, you know, 10 or 12 years ago, maybe even 15 years ago, we would have a great idea. We would develop that idea. We would do it in insulation inside of Kili Labs. we'd be really proud of what we accomplished. Um, we would demonstrate a proof of concept and then we would throw it over the fence to a product team that was supposed to to take it to product and they wouldn't do anything with it and it would frustrate us. And I think what we learned over the years is that just as important as the idea is also the model for how you develop these ideas. So the model that we use right now and I think it's actually quite successful is we co-develop these ideas with the businesses with the technology teams that are going to be required or that are are responsible for making the product. So from day one we start with the customer problem like I mentioned we kick off a joint project they um they being our our partners across TI invest resources we invest resources we co-develop this idea from the beginning we go down that journey where we pivot and we learn and we have to turn and along the way everybody knows why we made the decisions we made why did we go this direction why did we not go that direction and there's full and complete ownership uh we do that until we demonstrate the high-risk concepts so the goal is to fail fast if we um if we're successful, we have something where the risk level of it looks no different than any other product that TI builds, any other technology that TI builds. At that point, that same core team moves full-time out of Kilby and into the um to the productization teams that are responsible for building the products and driving it to high volume. And so now there's complete ownership. There's no baton passing. It's one team that's from there from the or that's there from the beginning all the way to the end. We installed this model I would say about 10 years ago and I think it's really demonstrated a lot of benefits when it comes to actually solving customer problems moving as quickly as we need to to develop the technologies and ultimately doing um um doing what's necessary to get these technologies off of paper and into like high volume into facilities like this. So for Kilby I think that model has been a journey but it's very I think important because otherwise we just do research for research sake but our goal is to do research to solve customer problems. >> Can you give me like one concrete example of something that's come out of the lab? >> Yeah I can give you a lot actually but uh where do I start? The first one okay so uh not the first one but the first one that comes to my mind actually we announced this um a year ago um November of last year. Um uh we call it um magpack. So this is basically a packaging technology that allows us to integrate passives. So these passives are are um things that surround our semiconductor chips. >> So it's like a packaging that's going to appear in here. >> It would appear inside of here that would allow us to integrate that functionality in our into our product in a way that nobody else has ever been able to do. >> Okay. >> And um the journey started about 10 years ago. Um, I hired somebody who is an expert in magnetic material into the team and I told him to tell us what to do because we're not experts in magnetic material. He is. And um, we spent I I was naive I would say at the beginning because I thought it was going to take two years but it took a lot longer and we we pivoted, we learned, we solved problems and about a year ago we were able to release that. I think there's six or 10 products on the web that started in Kilby Labs about 10 years ago. Another example that started actually even before that is auto radar. So we have these um chips that actually can sense um like it's a radar on your car and it can sense um objects car um you know um cars in other lanes and it's basically one of the more key functions in enabling self-driving that started in Kilby Labs a number of years ago as well. So we've had many success stories. We actually uh we pride ourselves on being a a sort of a um a um applied R&D and so we want to see these things go to product. We want to see them solve real customer problems. I think we've got a pretty good uh pretty good track record there. >> Couple more for you before we go. >> Yeah. Yeah. >> So TI has the entire supply chain. >> Why is that important? >> Yeah. There's a lot of reasons why. Um you know I think that controlling your own destiny is probably the simplest answer. We control the technologies we build this factory >> mean design production >> everything design production even the um you know ti.com everything from the initial technology to the manufacturing of it to the products that we build to how we engage with our customers we own the entire chain and so it gives us the opportunity and the ability to control our own destiny and that I think spans all of those areas we can control our factories We can control what technologies we install in those factories. You know, I mean this particular factory, it was purpose-built ground up from the layout of it, from the tools that we installed, from the way that we run the factory to support these analog foundational chips. And really exciting, I think, in the future is that we now also have the opportunity to define next generation technologies to take advantage of that. And so to me, controlling the entire supply chain is got technology elements to it. It's got just quality elements to it. It's got customer supply. It's got all of the pieces that I think is very unique for us. There's very few people who are able to do this at scale the way that we are. And I think it just gives us the ability to control our destiny in a way that um in a way that differentiates us. >> Last one. Um the technology world is changing fast. Yes. >> I think that's an understatement. I can't even imagine like what's going to come next year because it's just been bananas. Um what are you most excited about? Yeah, I am excited. You know, maybe it's uh maybe it's just me, but I think that I kind of mentioned the semiconductor thing at the beginning that 10 or 15 years ago, the average person probably hadn't even heard the word semiconductor. What's the most exciting thing to me is that it's still the whiteboard story. You go to customers and they really want to collaborate. They understand how vital of a role our technology and our products play in their future. And to me, the pace of innovation increases. The ability to solve problems jointly is much better than it is if we just show up with something that we thought was a good idea. But if we can jointly develop these ideas together, jointly um um envision the technologies. And I see an appetite for that more so than I've ever seen. I mean, sorry, excuse me. This year, I probably visited more customers than I have in my entire career. And it's because they want to talk about this stuff. understand the importance of it and the supply chain, the technology, the criticality of semiconductors. So that to me is what gets me really excited because now technology solutions are not a nice to have or something you just sort of um you know just sort of claim on paper. They have to have it. They need it in order to solve their problems. And it's just very real when you go talk to them and you realize how badly they want to work with us and how badly they need the new technology. >> Jeff, thanks so much. Great to speak with you. Yeah, it was great. Thank you. Thanks for having me. >> Very interesting. Thanks again. >> Yep. >> All right, everybody. Thank you so much for watching. We will see you next time.