Companies mentioned: GFS 0.00%↑
The history of GlobalFoundries starts with AMD, then moves to private ownership, along with technology sharing arrangements, followed by a famous readjustment of the company to not pursue leading edge EUV technology in favour of building a customer base around legacy process nodes – or to quote the new term, ‘essential nodes’ for an semiconductor industry that produces far less silicon on the leading edge than most realise. For every smartphone SoC or machine learning processor comes an army of power delivery, control silicon, and all sorts of chips to support it. A modern smartphone, according to one report, contains only 23% ‘leading edge’ (7nm or better) silicon. As a result, foundry offerings like GlobalFoundries are in high demand.
At the recent GlobalFoundries Tech Summit, GTS, I had the opportunity to talk to the CEO for an update on the health of the company, as well as a few questions that I had been asked over and over again in the community. Dr. Thomas Caulfield has been President and CEO of GF since March 2018, previously serving as SVP and GM of Fab 8 in Malta, NY. Dr. Caulfield has led the company through the pivot away from EUV to the IPO in October 2021. In previous roles, Dr. Caulfield spent 17 years at IBM, ultimately serving as VP of IBM’s 300nm microelectronics division. Dr. Caulfield holds a doctorate in Material Science and Engineering from Columbia.
This interview is available in video below, followed by a transcript. The transcript has been edited for readability.
Ian Cutress: We last spoke properly in this format years ago, you had just become CEO, and we’ve come a few years down from then - we’ve gone through a pandemic for a start, and a chip shortage. The first question is simply ‘How’s business?’
Tom Caulfield: Business for GlobalFoundries, over the period of time you're talking about, has been really great for our company and our customers and our employees. If you recall, in 2018, we decided to stop chasing Moore’s Law to be a distant 3rd player in that race. We focused our energies and our R&D where we can create really differentiated products for end markets which help our customers differentiate their products. They [end up capturing] value for themselves, and we get to capture some of that value for ourselves. So when I think about the longer range view of this company from 2018, to where we are today, it’s been a great journey for GlobalFoundries.
IC: Customers up? Revenue is up?
TC: Everything is up and to the right! Profitability is up and to the right.
IC: One thing that I've noticed at this event - you’re no longer referring to your production as ‘Legacy Nodes’. You're talking about this concept of ‘essential chips’. What’s that all about?
TC: I think our industry is not very good at marketing itself! We went from ‘there are two types of chips: advanced chips, and legacy (or mature)’. It really doesn't do justice to the level of innovation that goes into these chips. We can't get stuck thinking that the only innovation in this industry is driving the scale of transistors. There is tons more innovation that is actually much more relevant, especially to the world we live in. So we started to call them ‘feature-rich’ chips, because that's what we’re adding - we’re adding new features. We take existing technology platforms and are making them more vibrant and germane to the industry. So we went to ‘essential’ because at the end of the day, without these chips, many of the applications we’ve talked about would never occur, so they become essential to the end markets.
IC: I know you guys play a lot in smartphone and automotive, and I saw a slide here today where it is not just leading-edge silicon that goes into those devices.
TC: Right, Exactly. I’ll get a couple of statistics: about 70% of the market for semiconductors is actually 12nm and above. Only 30% are in single-digit nanometers, and why is that? Moore’s Law was always an economic law first and foremost, and it was to scale transistors because you got a lower cost per transistor but you also got better power per transistor and better performance. If you fast forward to 28nm, and certainly by 16nm, 14nm, the economics changed. The cost per transistor went up, so the only reason you would use those technologies is for applications that absolutely demanded either the performance per transistor, or the power per transistor. You would never pay more for an application than you needed to, and that really created the pervasive deployment of semiconductors and why these essential chips became more essential. They [have] the right functionality, the right feature set and at the best economics.
IC: When we look at the foundry space, there's obviously a number of players, some leading edge, some more essential. Most foundries in your position are expanding - what is GlobalFoundries doing to expand?
TC: Our business is a manufacturing service company we can only grow revenue by growing more capacity. You're right - there are a number of foundries, but it's a finite number I would argue. There are only five foundries on the planet of any scale - let's choose scale at two billion dollars revenue, and that's a pretty low bar I think given that in 2022, GlobalFoundries was 8 billion dollars. So who are those five foundries? In Taiwan it's TSMC and UMC, then there’s GlobalFoundries with our global footprint, then Samsung Foundry and SMIC in China. So we have a high concentration of foundry in one region of world. Almost 70 to 80% of the capacity is in Taiwan, and on top of that we have high concentration in one company. TSMC is twice the size of the revenue of the other four I just mentioned, times two! So it's a big concentration, and it's a big industrial issue in concentration to have one company concentrated one part of the planet.
So back to your question - we need to add capacity to create scale so our customers continue to rely on us - we've been building out our global footprint. Let's start where I was just two weeks ago. In 2021 we put our first pylon in the ground to add capacity at our Singapore facility. In June of 2022, we moved the first tool in, and then July of 2023 we celebrated with a ribbon cutting. With this factory up and running, it added a significant amount of capacity - over 400,000 wafers a year of manufacturing to our footprint. In upstate New York in our Fab 8, we're continued to build out the floor space there with capacity we've invested in - over 1-1.2 billion dollars over the last couple of years. In Dresden, over the last two and a half years, we invested 2.4 billion dollars to bring that capacity up. We also announced last June a partnership with STMicro where we'll expand capacity in a joint partnership with them, and so that's capacity we have in flight.
As more and more opportunity for demand materializes will continue to grow our footprint and how we do that is we grow globally. The last thing I'll leave you with we just last week we announced the submission to the US CHIPS office our expansion plans, and co-investments, and funding from the US Government to continue to grow our footprint the United States.
IC: How much does this CHIPS act in the US and the EU CHIPS act affect how you're expanding?
TC: It doesn't affect the timing of when we expand. We will only create capacity when there's certainly and demand. For us it's about certainty, durability, and profitability. We are not a product company - we only build capacity to serve our customers, and our customers need to commit to that capacity for us to make those investments. Many times we make those investments in concert or in partnership with them. We make investments in capacity in the markets that we view are doable, on technology platforms that we have that serve markets that we think are doable. For example, in automotive. Then of course, like any good company or anybody that makes an investment, we want to get a return on those investments. So profitability is an important part of that investment thesis. So for us with the European government, or U.S governments (or other governments), what they bring to the table is their desire to create capacity in their regions. They provide some of the funding that helps close that economic model on profitability.
IC: A lot of people don't realize just how much research goes on in nodes that aren't leading-edge. I've just come out of a presentation where showcasing all the different variants of 22nm and 45nm class nodes that GlobalFoundries has. Can you go into the R&D that GlobalFoundries does on these essential nodes?
TC: Let’s first start with how dedicated you are to R&D. You vote with your dollars - so somewhere around 7-8% of our revenue we put towards R&D, and that represents a few thousand people doing R&D. Where do you do the R&D, and how do you choose to do the R&D? For us, we focus on end-markets where we think real value can be created that's differentiated. This is to help our customers serve those markets, and by extension if we're helping them serve the markets we also capture value for ourselves.
So we do a very detailed in-market assessment. We call them ‘battlegrounds’ of where GlobalFoundries can create differentiation. Differentiation comes in many forms.
What do we add to the platforms of our technology?
Is that high voltage? Is it embedded memory for security for microcontrollers?
Is it low power in the off state for untethered battery-powered devices?
We choose the types of features we need to add. Sometimes those features are actually in the software. Or it's IP for designing – the complex or foundational IP that enables the platform to deliver to those marketplaces. So worldwide, we have over 2,000 people doing R&D. In Europe alone we have over 500 people doing R&D. It's to continue to add features that serve these end markets, and build out the platform of our technologies.
IC: Any piece of recent development that makes you feel warm and fuzzy inside?
TC: Warm and fuzzy? *laughs* I think I'm really excited by our embedded memory technology. Especially one announcement we made recently on our 28nm. Not only does it feature performance, endurance, and reliability, it is super competitive. We actually create the memory cell with half the mask steps of our competition. So we have a performance benefit, and a cost advantage in creating that technology. Those are the kind of things that make warm and fuzzy!
IC: I think for me it's the photonics - the 45nm photonics process. I’m seeing a lot of people wanting to use that and having very few options elsewhere.
TC: I think that's another good one! I don't try to choose my favourite children. But yes, we're the only company that integrates RF-SOI, CMOS and photonic processing on a single die and that creates huge opportunities in and markets we serve.
IC: The chip shortage caused a lot of havoc in the industry - this whole thing about ensuring a supply chain and then on top of that a secure supply chain. We are here in Germany today, and obviously Germany's a big point for the automotive manufacturers. Using those two bits of information, are we out of the woods in terms of not meeting supply?
TC: Let me reflect on that question a little bit. For me it was just amazing that you had an entire world economy depending on an industry called ‘semiconductors’, and it really it was obscure. No-one really knew about it until you couldn't buy a car, or a washing machine, or some type of appliance! Everybody said ‘what's a semiconductor?’ and they learned a lot. They learned the criticality of this technology. So it did create a lot of havoc, and it created a lot of action to create more capacity. It created action by governments around the world, [with them] saying we need to make sure for, you know, sovereign security we need supply chain security, and economic security. [The governments said] that [we need to] create some of this capacity in our own country. It created a froth of activity, but I think that was a reflection of just how important semiconductors are to the world and to the world economy.
So where are we from a supply demand mismatch? It's a very tough question, in that all these end-markets have different unique attributes and different types of supply. I would say the auto industry is much better in balance than where it was in 2021. But I think we're still at a point, for the types of chips that are needed and given the transformation from internal combustion engine cars to autonomous connected and electrified cars, the types of chips needed to do that means we will still going to be chasing that capability for some time to come. But we much better alignment compared to 2021.
IC: I speak to other companies in the industry and they say they’re trying to meet as much demand as possible, but it's still going to be a few years out to meet this demand. These are your customers.
TC: So you confirm my answer!
IC: Yes yes, I can confirm! So the current boom in chips right now, especially in the media, is this whole thing about artificial intelligence and machine learning. We've just spoken about essential nodes, but typically when we talk about machine learning we talk about training and the need for this high-density, high-performance chips. So where does GlobalFoundries meet the market in machine learning?
TC: I think the big opportunity for AI is what we call ‘Intelligence at the Edge’. The simplest place to put AI is in the cloud – the cloud will do the training of these large models, but for these large models that inference will take place at the edge. It's too costly and the economics don't work if they try to do everything in the cloud. Remember, at the edge is where something is sensed. The analog world comes in: is it an image, or is it a sound, or is it heat? Something gets sensed and it needs to be acted upon. That acting is the inference from some model that was securely put on that device. That device senses this analog world, and does something with that information and takes an action. Perhaps it then takes the data, parses it, compresses it, and sends it to the cloud, after a certain amount of data that's been accumulated. Then those models get retrained. So when you think about the future of AI, [in the] cloud will be where models are trained and generated, but inference will take place at the edge. The two of those are very powerful because you'll have the ability to have all your appliances, all your devices, with your smartphone and your display, your television, your computer – they’ll be AI-enabled and that's because the inference will take place on that device for the quick response to whatever signal or sensing it has.
IC: Is GlobalFoundries going to mostly play at the edge?
TC: We are. We’d like to be the leader for inference at the edge.
IC: What about infrastructure?
TC: Infrastructure is super important, especially when you think about infrastructure for energy and infrastructure for communications. We have leading technology for RF communication, whether it's a short range Bluetooth, NFC, mid-range WI-Fi or for cell tower-based in millimeter wave, or 5G. Our technology is the heart of that, both from an infrastructure side and a smart mobile side.
IC: If there is one thing you want the people watching to know about GlobalFoundries in 20 seconds, what would you want it to be?
TC: It would be that innovation can't narrowly be defined by transistor scaling. Yes, it's a necessary part of the of the technology requirements of our industry. But it's nowhere sufficient. These essential chips - the types of features we add - bring life to all the appliances we use. You use secure pay on your mobile device - well guess who makes those those devices? GlobalFoundries does, because we put embedded memory in a microcontroller that has NFC connectivity. This means it has a secure element to make sure that your transactions are done in a secure way. When you're cell phone connects to a 5G tower, it's all the silicon GF makes – that’s not made on a single digit nanometer, but made on differentiated essential chips that build out these features for those mission critical applications.
IC: So into the after show – here are some pointy questions from the audience for you!
TC: This is like Bill Maher overtime!
IC: What happened to the EUV machines that you had in Malta (Fab 8)? I remember seeing them because I visited when they were being installed!
TC: When we pivoted the company in 2018, we had no need for those machines. I believe we sold one to a research organization, and the other one ASML took back, refurbished it, and sold it to someone else. By ‘took back’, they gave us you know ‘some’ number of cents on the dollar.
IC: So there was a restocking fee?
TC: A very steep one.
IC: What has happened to 12FDX?
TC: 12FDX is still in our roadmap. Creating a technology well ahead of the market need doesn't serve anybody, and for us it's making sure that we develop these technologies for the future in partnership with a customer with a specific end use in mind. So we continue to keep 12FDX on our roadmap, and we continue to gauge customers to make sure the definition of that technology and where we're going to spend R&D is critical to an end market
IC: You recently made an announcement about an office and development in Penang, Malaysia, despite having a fab in Singapore. Why not Singapore?
TC: What we're trying to create is a global footprint that is flexible. The Penang facility services all the fabs around the world, whereas the Singapore team is for the Singapore team. So we've tried to take a global network of fabs and have them run in a way that feels like they’re a virtual single fab. There's certain things we do locally. But the European Union is an example of things we do globally. The Malaysia site is part of how we want to manage some of the engineering globally.
IC: Are there any technologies on the horizon that would imply analog signal density increases?
TC: It's really difficult to compress or shrink analog. That's why some of these essential nodes stay around for a long time - because in shrinking them if you don't have enough digital content in a design, you don't get very much shrink. So mostly it's about how to integrate analog with digital in the most holistic way.
IC: Do you have plans to bring photonics down to 22nm and 12nm?
TC: Right now we don't see the need. For the feature size that's driving the roadmap, it's more about adding more of the photonics compute capability than the size of the transistors doing that.
IC: When will GlobalFoundries be the number one Foundry?
TC: In my lifetime (chuckles)! So it depends on what you want to define as number one. If it's in revenue, in dollars, well we have a long way to go - our competition is eight times bigger than us. But if it's number one in innovating around essential chips, I think hard to argue we're not already number one on that.
IC: And a final pointy one. Recently Intel failed in acquiring Tower due to regulatory issues. Is there anything between Intel and GlobalFoundries to know about?
TC: We don't comment on any kind of rumours or anything like that. But nice try!
IC: Awesome, thank you so much Tom, and thanks for being a great sport.
Great video