Episode 59 • 2 March 2023

Chris Miller on the History of Semiconductors, TSMC, and the CHIPS Act

Spotify Apple Google Pocket Casts Other apps

Contents

Chris Miller is an Associate Professor of International History at Tufts University and author of the book “Chip War: The Fight for the World’s Most Critical Technology” (the Financial Times Business Book of the Year). He is also a Visiting Fellow at the American Enterprise Institute, and Eurasia Director at the Foreign Policy Research Institute.

Over the next few episodes we will be exploring the potential for catastrophe cause by advanced artificial intelligence. But before we look ahead, we wanted to give a primer on where we are today: on the history and trends behind the development of AI so far. In this episode, we discuss:

How semiconductors have historically been related to US military strategy
How the Taiwanese company TSMC became such an important player in this space — while other countries’ attempts have failed
What the CHIPS Act signals about attitudes to compute governance in the decade ahead

Transcript

Introduction

Luca 0:06

Hey, you’re listening to hear this idea. In our next few episodes, we will be exploring advanced artificial intelligence. We’ll be speaking to people about governing this transformative technology and how to prevent potential catastrophes caused by advanced AI. But before we look ahead, we figured that we should begin by getting some grounding on where we are today day in the history and trends behind the development of AI so far. To really understand where we are now, we should probably understand some things about the supply chain behind advanced computing. For as much as AI seems to be made up of disembodied lines of code that we can access from anywhere via the Internet, it actually requires a lot of physical inputs, from incredibly complicated hardware to huge data centers, to talented labs of people and much more.

Luca 00:50

Thinking about these real world inputs means thinking about economic and geopolitical issues surrounding who controls production, where this production takes place, and what this production ultimately gets used for. And answering these questions highlights some risky possibilities, such as US. Chinese tensions and businesses under pressure to outcompete each other. But they also point to some potential solutions, such as where in this complicated system governance can intervene and what actors underlying motivations are so that we can avoid a race at the bottom that neglects safety. To help give context to these questions, we spoke to Chris Miller, who is an associate professor of International history at Tufts University and author of the book Chip War the Fight for the World’s Most Critical Technology, which has won much praise, including Theft Business Book of the Year.

Luca 01:35

Chris’s book tells the decade long history of how we got from the invention of the transistor in 1947 to where we are today. So again, this episode is mostly about historical context. We do not explicitly talk about existential risks or specific policy interventions, but we do think that many listeners, especially those new to this topic, can get a lot of useful background information from it. We look at how chips have historically been related to US. Military strategy, how the Taiwanese company TSMC became such an important player in this space, but other countries attempts have failed, and what the recent chips act signals about coming attitudes for the decade ahead. So, without further ado, here’s Chris Miller.

Fin 02:14

All right, Chris Miller, thanks for being on the show.

Chris 02:17

Thanks for having me.

The complexity of semiconductor supply chains

Fin 02:19

Well, I look forward to talking about the history of semiconductor manufacturing, but maybe a place to start is just to get a sense of how incredibly complex the industry is today. So I wonder if you could share some details about just how impressive and how complicated the semiconductor industry and supply chain is right now.

Chris 02:41

So making an advanced semiconductor requires hundreds or in many cases, several thousand process steps. And advanced semiconductors today, like the one you’d find in your smartphone, for example, or in a data center, will have billions of transistors on them. And the most advanced transistors are the smallest transistors. And today transistors can be made the size of a virus by the billions. And so the specialized and ultra purified chemicals and gases are needed are really extraordinary. And the machine tools that can work with the materials by layering chemicals onto silicon wafers just a couple of atoms thick or by etching canyons in the silicon just a couple of atoms wide and do so with basically perfect accuracy a billion times an hour is really an extraordinary feat of manufacturing.

Chris 03:37

We don’t normally think about computing as resulting from manufacturing prowess but in fact all of the chips that produce computing power and remember data for us are all manufactured using precision manufacturing techniques.

Fin 03:51

Right. One fact I enjoyed from your book was that there is one plant operated by TSMC in Taiwan which over the course of a year, maybe it was 2021 manufactured more transistors than all the goods sold across all the industries in all of human history, which really put things in perspective.

Chris 04:08

Just to add on that point, acquiring the machinery, acquiring the chemicals, acquiring the materials involved is something that is impossible for a single company to do. We talk about TSMC being the world’s most advanced chipmaker and that’s true. But they buy machine tools from the world’s most advanced machine tool makers. And there’s five companies that play a really substantial role in the tooling segment. They buy chemicals and gases from a pretty small number of companies as well. And then they acquire chip designs from also a relatively small number of customers that make up the bulk of their revenue. Companies like Apple, Nvidia AMD and others.

Chris 04:51

So across the chip industry there’s a set of sort of oligopolistic markets that produce the precision and the capabilities that chipmaking relies on which on the one hand has made possible this extraordinary specialization which is what’s needed to produce advanced chips. But it also adds to some of the risk in the supply chains because you don’t have lots of producers at most steps of the production process. You’ve got a small number of companies and often a small number of facilities that are actually producing the necessary tools or materials.

Why is the supply chain dominated by a few firms?

Luca 05:23

Can you maybe spell out a little bit as to why, as you said, a lot of this supply chain seems to be so monopolized or dominated by a select like number of few firms? Like, what are the kind of drivers of that?

Chris 05:36

I think there are two basic reasons behind it. One is that many parts of the chipmaking process are extraordinarily capital intensive. That’s true for chip fabrication where a new chipmaking facility can cost $20 billion. So your startup is not going to raise $20 billion. And even the comparatively less capital intensive parts are still pretty capital intensive. So it’s not like software where you can start with a pretty small budget and then grow over time for a lot of segments of the chip industry. You just need massive capex at the outset and so that’s a barrier to entry. But next to that, I think, is the specialized knowledge that is involved in a number of key parts of supply chain is something that is learnable only in that segment of the industry.

Chris 06:23

So a lot of the really unique material science that’s involved is something that you can study the background principles of in a PhD program of material science but you can’t actually get the really nuanced understanding that you need to mass produce it unless you’ve worked for a relevant company. That means that if you’re thinking about potential competitors or new entrants you’re often thinking about new entrants that are founded by someone who’s already in the industry and is able to raise a substantial amount of capital and that just limits the ability of new entrants. Now there are some we do see new chip design firms in particular which is the least capital intensive of all of the chip market.

Chris 07:01

And in other places, like in the photoresist which is one of the types of chemicals you need to make chips, there are some new entrants in that market but in aggregate you see a lot less entrance than exit than you might expect just given those two pretty substantial barriers.

What would it take to dislodge a monopoly?

Luca 07:17

Yeah. And how do these monopolies, if at all, ever get dislodged? You mentioned kind of new entrants there but if we’re looking at the firms that are kind of dominating these supply chains today and we compare it to the 1950s when a lot of this story began, what would it take for any monopoly to get dislodged, if at all?

Chris 07:35

We do have a couple of examples of firms with really substantial market positions getting dislodged. And the key thing is the key driver is really big technological shifts. I think an example of that might be intel right now, which for a long time computing both in the PC and in data centers was really happening around the X 86 architecture which is one of the ways you can architect the chip which intel was really the biggest player in and only had one competitor, AMD. And over the past decade there’s been a really big shift, especially in data centers in terms of the focus of high performance computing. It’s moved away from the architecture that intel specializes in towards alternative ways of designing a chip.

Chris 08:19

Look at GPUs for example, have played a much bigger role in data centers and that’s not because intel did a bad job at CPU design per se, although you could argue about their market surveys to be AMD. But the big shift is that new types of chips emerge to play a bigger role and so that’s where you get pressure on existing major players is when the technology shifts in a way that is unanticipated by most of the market, including the incumbents. But in the manufacturing process you had less of that because chip manufacturing has really operated in the same basic way since 1960. It’s just gotten better and better. But you’re still buying the same types of tools, just tools that are a thousand times more advanced than they were.

Chris 09:03

So there haven’t really been any major disruptions to the manufacturing process itself.

Semiconductors in military systems

Fin 09:08

So lots of countries clearly have an interest in owning and controlling parts of the ship supply chain. I’m curious how much of that interest comes from national defense and military considerations.

Chris 09:24

That is clearly a major driver of why governments are interested. But economics is also a big driver as well. So today, 98 or so percent of chips end up in civilian use cases, whether that’s smartphones, PCs, civilian focused data centers. And it’s really only a small percentage of chips that get consumed by governments for security purposes. And so that means that if you want to make money in chips, the easiest way to do it is selling to the smartphone market or selling the data center market rather than military use cases. But for militaries, computing power is absolutely critical. And it’s both semiconductors that are in data centers that are training new systems, and it’s also lots of semiconductors that are distributed across military systems and sensors processing data that systems are acquiring and then often communicating it back to a more centralized data center.

Chris 10:17

And so militaries, more than ever before, are dependent on semiconductors for their edge. And that’s why, although militaries have been interested in chips since the first ones were invented in the late 1950s, it’s increasingly the case that defense planners are thinking hard about their ability to access the most advanced semiconductors and then quickly apply them to their own military systems.

Luca 10:40

When we’re thinking about especially leading edge chips being used in defense, what kind of image should I have in my head? Are these like chips on missiles? Are they kind of computer simulations? Like what kind of use cases?

Chris 10:53

You do find leading edge chips on military systems themselves. So on an airplane, for example, you can find some leading edge chips. But I think the right thing to think about is data centers, because it’s data centers that are going to train the AI systems that will increasingly play a bigger role in defense applications. And so this could be drones learning to fly autonomously or semi autonomously, they’re trained in data centers. Or it could be electronic warfare systems. So if you want to jam your opponent’s communications and not have yours be jammed, you need to understand what parts of the spectrum they’re jamming. Jump your communications to free parts of the spectrum. And that’s increasingly done not by humans trying to figure out what’s being jammed, but it’s just all automated. And so that type of training also happens in data centers.

Chris 11:40

And so in the past, in the origins of the chip industry, chips were being plugged into missiles to guide the computers more accurately. But now military systems are being trained in data centers to work more accurately. And so when we talk about chips for military systems at the cutting edge, what I think people don’t think about but should be thinking about, is the data centers that could be used to train a civilian system or a military system just as effectively. But militaries are, like the rest of the economy, turning to data centers to hone their performance. And that’s where the real security risks come in, because the same types of concerns about do we understand what our data centers are training? Do we understand the systems that are emerging is particularly important when those systems have explosives at the end of them.

Chris 12:29

And so we really want to be careful that we understand exactly how they operate.

How did TSMC become so dominant?

Fin 12:33

So you mentioned earlier that TSMC is one of the biggest players in the chip supply chain. If I remember right, they make approaching half of all the consumer chips in the world. Can you say something about the story that led up to that point? How do they become so dominant?

Chris 12:49

Yes. He was founded in 1987 by a businessman named Morris Chang, who had spent most of his career in Texas at Texas Instruments. He was the runner up to become CEO in the mid 1980s, passed over the CEO job and then left and was approached by the government of Taiwan, where he’d spent some time before as a Texas Instruments executive. And the Taiwanese government wanted to build their own chip industry and gave him essentially a blank check to invest in Taiwan and build up TSMC. And he had a new idea for structuring a business that revolutionized the industry. And that idea was the foundry model of production. So before this time, almost all chips were designed and manufactured by the same companies. But his intuition, which proved absolutely correct, was that if you focus solely on manufacturing, you could end up manufacturing more chips.

Chris 13:44

You could serve multiple different customers, and you could get economies of scale that let you both drive down costs and hone your production processes. They became more advanced because you were learning from your chipmaking process over the production of a larger volume of chips than your competitors. And so from that point, TSMC has become both the world’s most advanced producer of processor chips and also the world’s largest chipmaker, producing more chips than any of its rivals and far more today than, for example, intel, which still mostly produces chips that itself designs in house.

Luca 14:18

So Taiwan seems to be a case study where a country or government was able to, in large part, intentionally capture a big part of the semiconductor supply chain.

Chris 14:28

But there also seemed to be a.

Why did the Soviet Union fail to build a viable semiconductor industry?

Luca 14:29

Lot of failures of other countries trying to do the same. And in your book, the Soviet Union seems to be like one particular case study here. Can you explain a little bit about what happened there and in particular why it is so hard actually doing this.

Chris 14:42

Well, I found the Soviet case study fascinating and it started really as a puzzle to me as to what went wrong in the Soviet Union because if you had to sort of from a first principles perspective say what do you need to build a chip industry? You’d say well, you need a lot of smart physicists and scientists. Well Soviet union of that. In fact they had physicists who won Nobel rises in semiconductor physics. So from the science perspective they had what you needed. You need a lot of capital investment while a centrally planned economy that was spending a ton on its military certainly checked the box when it came to capital investment you need a baseline understanding of technology. Well, the Soviets had that too.

Chris 15:17

The Soviet Union invented their own first integrated circuit just two years after it was measured in the United States. So they were not far behind by any metric. But there were two things that went wrong in the Soviet Union. The first was that from the beginning there was a focus on trying to replicate what US firms were doing and to electric and European firms were doing. And replication was a good strategy in many ways and in terms of thinking about catch up growth and catch up technological progress, replication often works quite well up to a point. But in this industry it worked very badly because the Soviets were quite effective at replicating. But they were replicating at a pretty long time lag and thanks to Moore’s Law, that left them quite far behind the cutting edge.

Chris 16:06

The second challenge they faced was that they could never develop any sort of international market or international supply chains. And that meant that the market to which they were selling was tiny relative to the market that US earlier Japanese ship firms were selling to. And the component purchase process was very complicated because if you were at Texas Instruments you could buy components from anywhere in the United States or Japan or the key industrialized countries of Western Europe and so had a vast product market to choose from. Whereas if you’re in the Soviet Union you have to make it all yourself. You’d buy from East Germany, you could buy from Hungary to a certain extent, but it was a really small component base.

Chris 16:46

And so that meant as you tried to specialize and tried to advance in terms of your chipmaking, you were selling to a small market and therefore your capital expenditure to revenue ratio was way off. And simultaneously you had to spend more money on the investment side because you couldn’t simply buy premade materials or premade equipment off the shelf to the extent that western firms could. And so those two dynamics, the copying plus the stunted market for components and market in terms of end sales really caused huge problems for Soviet firms and meant that despite all the capital investment, despite all the brilliant physics that were already in the Soviet Union, the country was never able to build anything close to a viable chip industry.

Chris 17:29

And therefore it never really was able to build a viable computer industry either, to the extent that it was importing mainframe computers from IBM throughout the Cold War.

Will China catch-up?

Luca 17:39

When we’re looking forward to now, the key question around kind of catching up to the leading edge or even overtaking now seems to be around China. And I wonder how you take the case of China to feature on these two criteria points you mentioned there.

Chris 17:52

Yes, I think that there’s two questions to be asked here. One is what was China doing 15 years ago or ten years ago? And there’s two, what is China going to be doing over the next ten or 15 years? Because I think you’ve seen a really big shift driven both by politics in Beijing but also by politics in Washington. For most of the last 15 years, china’s electronics industry was one of the most internationally connected of all the industries in China. If you think of all the tech supply chains that intersect China, if you think of all the Western venture capital money that funded Chinese startups, there was just a ton of international interconnection between the Chinese tech sector and the Chinese electronics assembly sector and the rest of the world. And that enabled really substantial advances across China’s tech and electronics space.

Chris 18:43

But I think what we’ve seen over the last five to seven years is a meaningful shift driven by politics towards a much more bifurcated tech supply chain, with a Chinese focused supply chain emerging and then a US Japan, Korea, Taiwan supply chain that’s emerging on the other side. And simultaneous to that, you see a pretty active US Japanese, Taiwanese effort to cut off transfers of advanced technology to China with the goal of holding China back. And the combination of this means that I think the outlook for Chinese tech over the next 1015 years is going to be quite different from the last ten or 15 years because the factors that were driving technological advancement are increasingly under threat.

Chris 19:34

And the strategy of the Chinese government, as I understand it today with regard to the tech sector is to have state directed capital investment try to fill the gap that international connection and the increasingly increasing kind of threat to international connections has left. And so what that means is that the Chinese government is going to bet that they’re going to spend more on the chip industry and hope that this is going to solve their technological challenges. And I think that’s a mediocre bet to be making. It might work, but I think the track record suggests that capital investment is a necessary but not a sufficient factor to make advances.

Chris 20:19

I think if you talk to people in the Chinese chip industry, many of them are actually quite pessimistic proud of their advances in recent years, but also looking forward to a much more uncertain decade ahead.

The CHIPS Act and export controls

Fin 20:31

So we’ve recently also seen a big move in US policy with the Chips Act and the Fabs Act. On the one hand, this is trying to perhaps clamp down on technology transfers and also it’s trying to onshore production back to the US. Ultimately.

So I wonder if you could just briefly explain what these acts are saying. And I’m also curious whether you view this as a break from historical US semiconductor policy.

Chris 21:03

So there were two big policy changes last year in the US. The first was the Chips Act, which allocates around $50 billion in funding in the next five years to a mix of incentives for US and foreign companies to build chipmaking facilities in the US. And then a portion of that money is also going to fund a longer term R and D. So that’s the Chips Act. And then almost simultaneous to that, the US imposed new export controls that limit the transfer of GPU chips, as well as a number of types of chipmaking tools to China. I think you need to look at these two moves as related but distinct. The export controls were primarily designed to hold back China’s progress.

Chris 21:53

And if you listen to US officials like Jake Sullivan, the National Security Advisor, he was very explicit in a speech in September right before these controls were rolled out, saying our goal is to grow the gap between US and Chinese capabilities by pushing US capabilities forward and holding China’s back. So it’s a pretty zero sum view of the world technology landscape and a change from the last couple of decades of US tech policy. The Chips Act I think, is a little bit different in its focus. I think the Chips Act is really best read as an insurance policy in case of a war on the Taiwan Straits.

Chris 22:30

I think the combination of growing Chinese military capabilities, visa the US coupled with the outbreak of the Russia Ukraine war have sensitized, I think, US policymakers, and not only US policymakers, to the risk that something could go wrong in the Taiwan Straits. And given TSMC’s absolutely critical role in producing the chips that global manufacturing output relies on, there’s real concern that even if your percentage likelihood of a China Taiwan war in the next ten years is not that high, the expected value is still already huge because the cost would be enormous. And so Chips Act is really an insurance policy around that. And so it’s going to have an effect not a dramatic effect, I don’t think, but an effect increasing investment in chip fabrication in the US.

Chris 23:18

And similar programs in Japan, in Europe, india and elsewhere, I think will also, on the margin, have an impact. Taiwan’s critical role is still going to remain, but there will be a bit less concentration in Taiwan than in the past. Is this a change from prior policy? In some ways, yes. It’s certainly a change from policy over the last 20 years where the US. Government and most governments didn’t think that hard about semiconductors. But in some ways it’s a return to the historical norm because the US. Government has always seen semiconductors and computing more generally as a source of strategic advantage in terms of intelligence collection and in terms of military systems.

Chris 24:06

And I think we’re basically reverting to the norm of the 1960s terms of treating this type of technology not simply as something that goes into consumer goods but also something that is really critical to determining the military balance.

Luca 24:26

I guess one question I have especially about export controls is just why would companies comply with this? Presumably or what’s the either carrot or the sticker? Presumably there is a lot of money to be made in either transferring technology or selling things to China. And if you talked about a lot of the companies here that are of interest aren’t just companies based in the US. But companies based in Taiwan or elsewhere around the world.

Chris 24:50

Yeah, it’s a fascinating question. And one of the unintended consequences of the internationalization of supply chains is that more and more supply chains cross through more and more countries. And so what that means is that for the bigger economies and for the US. Above all, that actually gives the US. More influence over more supply chains than it used to have. Because if you look, for example, at TSMC, it’s the world’s most advanced ship maker. It’s absolutely irreplaceable but it can’t produce advanced ships without relying on US. Tools. Simply impossible. Maybe it’d be possible in 20 years time, but it’s not possible today. And so TSMC, for a variety of reasons is quite responsive to US. Export controls. Partly because Taiwan is in a unique position visa US. But also because they need US. Tools to produce.

Chris 25:40

And Samsung is the same and every chip company in the world is basically in a similar position of requiring some US. Tools to produce chips. Now, as you go to more lower technology chips but lagging edge chips, that becomes less true as you get into pretty low tech chips. But if you’re anywhere close to cutting edge you can’t do without US. Technology. Similarly, if you look at the companies that make chip making tools in Japan or the Netherlands, what you often find there too is that they rely on a lot of US. Technology too because the R and D processes are often happening in a multinational fashion. Many companies have R and D in Europe and the US. In Japan and have production in multiple countries as well.

Chris 26:23

And so even if you look, for example, at the Lithography tools that Asml produces, the most advanced Asml systems are assembled in the Netherlands using components from Germany but also from the US. And so Asml can’t produce its most advanced photography tools without using really substantial amounts of US components too. So it’s a Dutch company and they’re Dutch made tools, but with components from other countries. And so that’s another reason why they would also be responsive to US export controls. Actually, they’re restricted from exporting EV tools by Dutch export controls, but if they weren’t restricted by Dutch export controls, one could hypothesize that the US. Might also impose controls.

Luca 27:05

Do you have a sense of how far the US. Can push this before there might be kind of, like, resistance or something and I guess, like, kind of in turn seeing more of a fracturing of the global supply chain, not just in terms of US. Versus China, but also in terms of just, like, international companies being, oh, wow, we’re really heavily reliant on some US inputs and therefore really heavily reliant on US foreign policy. It might be good if we start looking to sourcing things elsewhere.

Chris 27:32

Yeah, it’s a really interesting question, and it’s hard to know where the line is when you have companies begin to take costly steps to do so. I think one of the disincentives for that is that if you look at the customer base for advanced chipmaking materials, tools, software, it’s mostly in Taiwan, South Korea. The US. Are the three major customers. So for all of the big tool makers, the three big customers on the logic chipmaking side are Samsung, TSMC, and intel. And in terms of memory chips, it’s Hynix South Korean firm, samsung South Korean firm, and Micron, a US. Firm. And so you’ve got a really interesting overlap between major chipmaking countries and US military allies. And for countries like Taiwan and South Korea, the defense commitment from the US is a very politically important thing.

Chris 28:30

You can’t, I don’t think, simply interpret the chip industry, even though people in the industry like to interpret it just through the context of business relationships. When politics comes into play, there’s a security aspect as well. And so I think if you look at the export controls the US is pushing right now, one of the things that will make them relatively more effective and we’ll see how effective they’re going to be. But I think one of the factors that is going to support their efficacy is that the countries the US is asking to come along board are Japan, which has basically the same threat perception as the US. Does on China taiwan, which is in a really complicated place, but quite reliant on the US.

Chris 29:10

And so those two countries, I think you should expect to be pretty compliant with US desires in large part because there’s a lot of agreement at the outset. I think if you look at countries that have somewhat different threat perceptions visa vis China than the US. Korea or the Netherlands, you see slightly different approaches on these issues. There is an overlap between, I think, the political views of governments and the unwillingness to follow export. Controls. But if you step back and look at the chip industry, what you find is that the US. Is still the biggest player by far from design tooling end use, and China is still a small market at the end of the day, and that really shapes the industry, makes decisions.

What would it take to dislodge the US?

Fin 30:00

I mean, a natural question there is what you think is most likely to change that, if anything. For instance, one example might be if a new bottlenecking technology comes along like the next Euv, and for instance, China becomes a leader. Is that plausible?

Chris 30:17

It’s certainly plausible. The question is, what would that technology be? And the answer is nobody knows. I will say that on the lithography side, if you think of the Euv, we’ve got a pretty clear pathway for the next five or ten years of lithography development, and it’s hard to find someone who would bet against asml leading that charge. And I think the same is basically true with the other key process steps, deposition, etching. Et cetera. I think bigger risks to the status quo actually come from not from the chip making process, but from the importance of chips. And if you look, for example, at all of the growth right now that’s coming from data centers. One key question is to what extent does computing power remain the really limiting factor in data centers?

Chris 31:06

Or can we get algorithms that use less data and therefore require less computing power? Is there a way that kind of computing architectures change in a way that makes us less reliant on chips? Sort of skeptical of that thesis, but that would certainly be a major risk that could upset the current structure of the chip industry.

Outro

Fin 31:24

Got it. Okay. Chris Miller, thank you very much.

Chris 31:27

Thank you for having me.

Luca 31:30

That was Chris Miller on the history of semiconductors, TSMC and the chipsack. If you find this podcast valuable in some way, then one of the most effective ways to help us is just write a review wherever you’re listening to this, be that Apple podcast spotify. Wherever, you can also follow us on Twitter, where at hear this idea. I also mentioned that we still have a feedback form on our website. We read every submission and you’ll receive a free book for filling it out. You can also find that on our website, which fin hearthisidea.com. Big thanks as always to our producer Jessun for editing these episodes. And thanks very much to you for listening.