Kopi Time E108 - Chris Miller on Chip War

Coy Time. Chris Miller is associated professor of International History at the Fletcher School at Tufts University. He published Chip War last year and it has been a major success winning the Financial Times best business book award for 2022. Chris Miller, a warm welcome to Cope time. Thank you for having me. So great to have you. Uh Chris, I'll begin by playing the devil's advocate. Uh the nuclear era quantum mechanics, DNS double he legs man on the moon. Even in my world, the futures and

options in finance. All of these civilization, transforming advances took place in the 20th century with little or no shape, processing power behind them, modern cities, industrial agriculture, amazing bridges and tunnels, aviation. All these predate the computer age. And yet in your book, you state that today's military, economic and geopolitical power are built on a foundation of computer chips. Are you overseeing the case?

in your refrigerator in your dishwasher in your microwave. Uh And almost every device you touch, there are lots and lots of semiconductors and they are so ubiquitous around us that we don't even think about them. Uh But in fact, as we learned uh during the pandemic era, shortages, modern economy simply can't function without them.

And so in that sense, I think they're the most economically critical technology because during the pandemic, the largest dollar value uh disruption to manufacturing uh in terms of component shortages came from uh semiconductor shortages. And so I think that was one very brief and relatively small illustration of just how critically economically important semiconductors have become

And just like in cars or consumer devices, the more advanced the communication, the more advanced the sensing, the more advanced the computing, the more chips are inside the same is true of defense systems as well. Which is why today every major military is thinking harder than ever about its access to semiconductors, both in terms of assuring that it can access the chips that it needs. But also trying to find new ways to apply the capabilities that chips enable the defense systems.

If you talk to defense ministries, you'll hear a lot today about autonomy about applying A I to defense systems and intelligence capabilities. And all of that means using more and more semiconductors for military purposes.

just fantastic first. So 67 chapters on the history of chip making. And we go all the way back to the forties and fifties, the uh the noise and Shockley of the world. And by the way, I did my phd at Urbana Champagne, Illinois. I did not know about John Barin till I read your book. And now I know why there's a Barine quad at Illinois. Um So looking at at that uh and, and the incredible success that Silicon Valley has had over the last 50 60 years in terms of pushing uh innovation.

So help us sort of walk through the critical ingredients behind this continuous series of innovation in the area of computer chips. Well, the the the

have a very small number of transistors on it. Today, your phone has billions of transistors, but just uh six years after the first ship was invented, that he saw a pattern of doubling every year or two and predicted that that rate of doubling would continue for some time. He thought continue for at least a decade. It turns out it

continued all the way up basically to the present. And what that's meant is that the chip industry has improved its capabilities more rapidly over a longer time horizon than any other type of technology in all of human history. Nothing else comes remotely close to that rate of technological improvement, exponential growth for uh more than half a century. It's something that the rest of the economy can only dream of in most companies if sales grow by 5% or if

technology improves by 10% it's a great success. And in the chip industry, that's an extraordinary failure. And so III I like to compare that to other types of veggies, aviation, for example. Um you know what if planes flew twice as fast every year while also being reduced in cost on a regular basis? Will they be flying uh many times faster than the speed of light at this point? If that rate of growth were sustained over half a century. And so it's it's physically impossible for

that to happen. But the chip industry has done something similar and and when you look at the proliferation of chips across society and across economy, that's been made possible because Moore's law has driven down the cost of computing. You want a data point that the first chips that were available commercially had just four transistors on them today. You can go to Amazon and buy a, a thumb drive with a billion transistors or 10 billion transistors

for about $10. Which means that ballpark, the cost of computing has fallen by roughly 1 billion fold uh since uh since the 19 fifties. And that's all possible. Thanks to Moore's Law,

not go on forever and what that means. But uh before we go in that direction, uh Chris, you spent a lot of time in your book talking about the immense complexity and the cost associated with producing chips uh share with us some idea of how, how complex and how coffee that is.

So there's no other industry that combines the scale manufacturing transistors by the billions with the level of precision required. And and doing this is it's, it's simply the most complex uh manufacturing that we do. And as a result, it requires the most complex supply chain that exists. And so today, it's fundamentally impossible, simply impossible to create an advantage ship without drawing on machine tools, software materials, um software from multiple different uh companies in

multiple different countries. And there are several that stand out. Uh the US Japan, the Netherlands, uh Taiwan and Korea are the five big players uh in the chip industry. Uh They're the only five countries that have um uh a substantial share of the chip industry's revenue in double digit terms. Uh And you can't produce a cutting edge uh chip without drawing on expertise from uh each of those countries.

facility is just to pay for the machine tools. So these, these tool tools that can lay down thin films and materials, just four atoms thick with basically complete deformity or carve canyons into silicon, just a couple of atoms wide. This is the most complex machinery humans have ever made that requires the supply chain that stretches from the Netherlands to the US to Japan. And you can't make advanced ships without

these tools. And these tools themselves are are not just made exclusively in these three countries. They're also made by a tiny number of firms. Uh And in the case of some of the tools, there's just one firm that knows how to make them. And so you mentioned the lithography tools when it comes to advanced lithography, a SML and the Netherlands has 100% market share uh of the tools that are necessary to make cutting edge ships.

this amazing analogy. As you say that the Carl Zeiss lens or mirror that is required to reflect the laser in the in the uh fat process for Eu V. It's like uh the, the perfection in that is like, you know, trying to hit a golf ball on the moon from here like the laser has to be that accurate or the pure on the class has to be like if there's one millimeter of impurity, it, it's like the size of Germany, that's the size of the mirror and that's a tolerance level,

which is absolutely mind boggling. But I, I really like the way you portray it because I think we sort of understand the word complexity, we're really understanding it. And I think that sort of analogy really, really helped. Uh definitely it helped me. Um OK, so it's a truly globalized process, all sorts of companies have very specific and world dominating roles in it. So I take it you don't believe the Chinese can do it all by themselves.

answering your question requires specifying the time horizon. Do I think China can indigenously develop cutting edge capabilities soon using solely domestically produced technology? I doubt it. Uh just because I don't think any country can do it uh soon. Uh And you know, if I look at the largest producers of uh in the semi hunter industry, the US, uh uh Taiwan, Japan, Korea, none of them have any sort of remotely plausible pathway over the next 10 years

to be completely self sufficient ship making. And so it seems highly implausible to me that China which is today still a small player in revenue terms, um, has a plausible pathway of its own.

uh about five years ago uh in Taiwan. Um It's a process that is now used in in mid and low range smartphones, newer smartphones will have uh tsm CS most advanced process, uh five nanometers uh in them. Um We don't know about the efficiency of the manufacturing yet. So Huawei's manufacturing partner, Smi um hasn't released any data nor do I think they will

any time soon. So uh whether this chip is being produced at huge inefficiency or relatively efficient, we're still going to find out um what we do know is that it, it was uh produced largely using imported tools. Um because until recently, SMI was able to buy the most advanced as well as the sort of second generation most advanced tools from the US, from Japan uh and from the Netherlands. Uh And so it had purchased a large quantity of those tools and is now still able to

use uh at least some of those. Uh and this ship was made largely using those tools. So, you know, in some ways, it's, it's certainly uh evidence of incremental progress in the ship industry. It's probably not evidence of catching up because TS MC is also moving forward. And so the gap between TS MC and China's capabilities, uh it doesn't appear to be decreasing. Um Is it e evidence of self sufficiency for China? Not really because again, the machines that were producing these chips

were uh largely imported uh machines. So it's, it's, it's progress of a sort, but I, I wouldn't call it a breakthrough.

having errors. But whenever any company brings online, a new manufacturing process more advanced with smaller transistors, there's always a very uh significant learning curve that has to be overcome and yields always start low and then they're always intended to end uh very high in the 90% level um uh as close to 100 as possible. Um And the, the, the pro the process of improving yields is therefore critical

to financial viability for companies. And so a key question in your ability to price uh effectively, if you're a chip maker is what yield you could produce at. Because if you're producing a 50% yield. In other words, you gotta spend um amount of, you know, you, you need to, you need to, to fund the production of twice as many chips as you actually get. In terms of working chips, you're gonna be a lot less

competitive than a company with 90% yield. Uh And so we, we don't have good data uh about yields from companies. They're very secretive about their yield data. Um But we know that the the leading chip makers uh end up with very high yields uh in the 90% range um for their, their, their mature processes. And so if you look at the new Huawei phone, you know, a fundamental question is, what's the yield? And I don't think we're gonna know the answer to that question any

time soon. But any assessment of how viable a processes is or not has to uh has to try to get us a sense of the yield.

right now. Are we saying that China is restricted from importing any lithography machines going forward or is the idea that, you know, they'll just get stuck in whatever they were allowed to import and they can keep on importing that but nothing else after that.

nanometer production. So China has been able in the pass to buy lots of tools and has bought huge volumes of tools that used for 14 nanometers, 10 nanometers and seven is the the note after that which is the note that the Huawei phone was uh produced on and so until recently and actually the the Dutch regulations don't come into force until the end of this year. So it's still the case that um that Chinese buyers

can buy uh certain tools from the Netherlands. But once all these restrictions are enforced, the uh the the net effect will be to limit China's purchase of Chinese firms purchase of any new tools that can be used primarily for sub 14 nanometer uh production. Now most chips by volume are actually produced at larger geometries. In other words, older manufacturing technologies at 40 nanometers or 100 and 30 nanometers. But in dollar terms, the most expensive chips are the ones that are produced

at more advanced technologies. And so inside your smart, the main application processor is produced at something pretty close to cutting edge inside your PC. Inside data centers, the key chips are uh are are are also pretty close to the cutting edge. And so for a lot of key technology areas, there are huge performance gains to be had by producing on a cutting edge technology or something very close to it.

to China, sub 14 nanometers. And so today, China imports almost all of its sub 14 nanometer chips from Taiwan, from Korea, from the US uh from from elsewhere, which is the reason that China spends as much money each year importing chips, it spends importing oil, it's largely smartphone processors, PC processors. And so it's actually not much of a change to China's chip buying patterns to have it keep buying chips from abroad. Um In so far as the regulations uh by limiting China's domestic

production force China to buy chips uh abroad. There are also as you mentioned, specific US rules that ban the transfer of GP US or chips used to train A I systems to China. Uh but these are very, very specific rules governing just one type of chip. And so the US has via its rules kept wide open the transfer of any chip for smartphones, any chip for autos, any chip for P CS. Um almost all types of chips except GP US. So China can still buy very advanced ships from abroad for smartphones.

The only restriction is when it comes to the chips that are used to train A I systems. And those are the chips called GP US that are largely produced by NVIDIA.

continue it. But I wonder if it will even intensify and if China is willing to spend more money on self sufficiency. In other words, produce more chips at a very inefficient manner domestically, you could end up buying fewer chips from abroad. And if China starts buying fewer chips from abroad, it also loses some of its leverage because right now, one of the factors limiting foreign restrictions on China is that foreign and chip firms still sell a lot of chips

to China. But if they begin to be restricted from market access by China's self sufficiency drive, then you'll have less lobbying in Washington, in Tokyo uh in European capital saying be careful with your restrictions, you're gonna hurt our market because if that market's already declining, that'll become much less powerful argument. And that's I think a very tricky balance for China.

It wants self sufficiency for sure, but it also values the fact that because its market is valuable, it gives it some bargaining power in its relations with the West.

how risky it is. Is it for like one company at this day and age being responsible for 100% of something that the world needs critically?

MC because of the risks to Taiwan. Uh that I think most people assess have grown somewhat in, in recent years. The, the, the Taiwan question I think is, is important, not just because of TSM CS dynamic as being the central player of in producing advanced ships, but also because it has become this critical producer of advanced ships precisely as the us ability to credibly defend Taiwan has deteriorated.

I think that's the second half dynamic. It's not just a chip story, it's also a story of the military balance shifting in Taiwan's favor. And so for most of the last 50 years, uh tech executives, investors, um electronics firms didn't have to think that hard about the military balance because it was obvious if there were a war, everyone knew who would win, everyone knew which would win. Beijing and Washington and Taipei everywhere. Today. If there were a war, there's a lot of

uncertainty about what the outcome would be. And I think that has inspired uh an increased sense of uncertainty about whether China might be willing to escalate. And if so what the ramifications would be for the supply chain. And so that's really the reason why we're talking about uh Taiwan risk is that the military balance has shifted dramatically in China's favor.

very instructive. Ok. I want to come back to something we talked about earlier, Gordon Moore and his incredible Christian call 55 years ago that, you know, chip speed processing speed can double 18 to 24 months. How long can that last Chris?

early two thousands. Gordon Moore himself said that he couldn't fathom ways that the industry could keep Mori's law alive for more than and here we are two decades later. Uh and it's still alive. So I think we should be very cognizant of the fact that past predictions of Moore's Law's failure have always been wrong. Uh And that should I think sensitize us to the fact that there are lots of smart people trying to find ways to keep it alive.

I think also it's the case that we, we have a pretty clear line of sight. If you look at the road maps uh that have been put out by TS MC or Intel uh over the next decade, how we're gonna get at least another decade of transistor shrinkage beyond that point, nobody knows. But it's always been the case that more than a decade out, nobody knows. It's just too far out to, to, to realistically plan,

to provide better semiconductors um with the same size transistors. So, for example, there's new ways to design semiconductors and GP U is the best example. What makes GP U a great chip for A I is not that the transistors are any smaller than the chip inside of your iphone. It's that they, they're designed in a different way. And so I think we're still in the very early stages of exploring ways chip design can produce uh performance enhancements.

And then second, there's a whole new trend in the industry called advanced packaging, which sounds very boring, but actually, it is going to be revolutionary in terms of letting us put different types of chips together in different ways, which will provide huge performance improvements in terms of speed of data transfer, in terms of managing heat dissipation, huge issue when you've got lots of electrons flying around.

Um and also in terms of cost, letting you combine, for example, a low cost ship and a high cost ship in a way that brings down your overall device cost. And so we're in the early stages of of this trend as well. And so when I look at the trajectory we have of at least another decade of shrinking transistors further lost the design renaissance is underway. Plus the advanced packaging shifts. I'm pretty optimistic that we've got a long pathway for further advances.

um Chris, we talked about military uh issues earlier. I want to bring that back. Uh Even in your book, uh from the very beginning when rockets were being developed, uh military uh experts saw the application of chips in the guidance system and you have this uh really cool chapter in your book where you talk about the early nineties during the Gulf War. First time we saw the application of, you know, advanced guidance systems. Um So military applications are always, always at the forefront

of ship demand. Is that a good thing?

is is being used. For example, militaries have computer vision algorithms in their intelligence systems. So take a satellite photo and differentiate what's a tank and what's a truck, not like be done automatically. That's, that's pretty simple. Um But uh more decisions about um communications. For example, if your adversary is jamming your frequencies of your communication system, can you use an A I to rapidly jump between frequency in a way that's faster than your adversary system?

Uh These are some of the use cases of, of a, of, of A I and military contacts. And then I think if you look at the role that drones have played in the Russia Ukraine war again, that provides a very clear, I think worrisome but clear pathway for how militaries will begin to deploy increasing degrees of autonomy in drones as well. And so we're gonna, I think have a very complex ethical debates over the next several decades about in what ways are we comfortable deploying

autonomy in military systems? But just as we're running at autonomous cars and just as autonomy is coming to all other, the range P applications. So too, it's already coming to military systems. Uh And there is a, I think very real race between the world's leading militaries and above all between China and the US to harness this technology for defense purposes.

joining the front. Um Chris, you know, I mean, this is just a stream of consciousness observation while the Ukraine conflict shows all sorts of cutting edge application, it still looks like a very first world war type of uh terrain. When we look at the videos trenches and something called dragon's teeth, which is basically, you know, concrete boulders to stop tanks from going over. Um So it seems like, you know, we have a lot of shiny new toys. Uh but the adaption of them, the

full usage of them, it takes a while, right? I mean, would you agree if I say that the sort of compute power that we have right now, even the existing computer power, forget about the stuff that's coming. We're barely using a fraction of that.

Um We collectively expected that Russia would launch devastating cyberattacks and knock out Ukraine's cyber infrastructure ali accounts, but that didn't happen in no small part because Ukraine was defended by Google and Microsoft and some of the key Western tech giants and,

and their computing capabilities are key to cyber defenses. And it, you know, we've talked about autonomy and uh in, in many parts of the economy, well, autonomy in both cyber offense and cyber defense is coming soon, it's actually already

already arrived. And so uh that will be a AAA new frontier of the application of, of A I. When it comes to the intelligence that Ukraine has received from the West, we've already uh we've got many examples of um Ukraine receiving very accurate targeting data from the West and Signals Intelligence is one of the early use cases uh of

the application of A I to military systems. And so we don't have uh great visibility into all of the specific platforms that the US is using to uh give Ukraine uh targeting data, for example, for its high on rockets. But I don't think you can

um safely. Uh I think you have to assume that uh there's a whole lot of computing uh involved in the very rapid identification of targets uh transmission of those targets, Ukrainian forces and the Ukrainians uh our system, if you just press a button and the rocket flies automatically it hit its targets. So those are just two ways in which computing is made.

I think a real difference on the battlefield. But you know, I think you're right that, uh, in some ways this is a 21st century war, in other ways, it's 1/19 century war. Uh, and I, I think that speaks to the, uh, the, the fact that regardless of the technologies or in some ways because of the technologies, uh, warfare remains a very brutal business.

have written so articulately about. But, uh, just give me a sense, uh, we've talked just a little bit about the technology around the Ukraine conflict. But as a historian, you know, what's your take on the course the conflict has taken and what would be a path to resolution?

And I think there's no pathway in sight for that issue to be resolved. And people speculate, welcome, you know, this territorial settlement, that territorial settlement resolve the conflict. But I think the core question is um is Russia going to continue uh fighting to keep Ukraine or try to keep Ukraine in its sphere or is it willing to let Ukraine not be uh in its sphere? And that's, I think the key uncertainty that uh we just don't know uh Russia's willingness to fight

on to pursue that aim. And the Ukrainians will say we're not going to do a cease fire now because if we do, the Russians will start fighting again after rebuilding for a couple of years. And you know, I understand why they think that because that's been the track record of, of recent history, there was a cease fire in 2014, the Russians rebuilt and uh attacked again. And so I, I think that that that key question of, of ultimately, is Ukraine an independent country

or not? And does Russia recognize it is independent or not? That's what's at stake? Uh And I don't see right now um any clear pathway to getting a resolution of, of that question that both Ukraine and Russia could agree on,

that there's a war going on. Uh when you walk the streets of Moscow and when you go to electronic stores that used to be, say, you know, filled with Apple and Microsoft products, now they're filled with uh made in China electronics products and they look really good and they're shiny and people are buying those things. So this sort of, you know, pivot toward Chinese tech. I mean, can Russia sort of, you know, withstand Western sanctions substantially better because of that.

large budget deficit now. Um but that is produced a sort of Keynesian impact on the Russian economy, uh which I think ought to have been expected. But most people myself included didn't fully uh expect in the early days of the war. And so rather than unemployment or declining incomes. We actually have very rapid income growth driven by very low unemployment in Russia because of labor shortages now. So that's, that's one surprise to the war.

I think the second surprise that is about the West's willingness to impose sanctions because one of the really striking acts of the war is that uh the West at first started to impose very tough oil sanctions on Russia and then immediately walked them all back through the price cap scheme. Um And so now Russian oil exports are basically unchanged since before the war. Um precisely because the West has collectively decided that it

doesn't actually want to sanction Russian oil exports. It wants more Russian oil exports to keep prices as low. That is possible. And so if you look at the Russian economic situation, that's the key dynamic is that a, the domestic economy is being supported by deficit spending domestically and b the external accounts are um they look relatively stable because oil exports and prices fall on oil. But uh volumes are uh almost as high as ever,

then that's where the revenue is coming from. But going back to the tech issue, um the, the view is that, you know, if you don't have access to Western technology, if the Apple I OS is not, you know, upgrading and you're not getting the Microsoft 3 65 support that day to day commerce and inventory management. Everything sort of comes to a standstill, but clearly, that's not the case for Russia. So I'm assuming that they're not getting those regular updates from Western tech.

And still, I don't hear any logistics failure or any difficulty in running power plants or for that matter, running the military in the case of Russia. So it's basically what smuggling is that is that the way they're sort of keeping themselves going. Well, I think for, for

Um But that's, that's pretty manageable. Um I think for, for enterprise software, one of the surprising things has been how long many enterprise software firms kept servicing Russian customers. Um And uh there's, it's been a, a matter of some debate, I think in the West as well. That was the right approach, but many enterprise software companies, um, essentially kept Russian clients um, up to date with functional software for a very long time, some even up to the present

uh day. So I, I think on the software side, Russia has actually had very few difficulties um in accessing maybe for certain niche types of industrial software. But generally speaking, it hasn't, doesn't appear to have been a problem um for, for tech components, you know, there have been open discussions in the Russian

press about certain types of component shortages. Um but the Russian manufacturing base is actually relatively small um outside of the auto sector which has gotten uh clobbered by um by sanctions of export controls. The other big sector for manufacturing is the military. Um And the military has got the budget, it needs to uh import components um from abroad. It probably does

sort of an inefficient manner. It probably faced supply chain difficulties due to sanctions, but it's been able to import at least some volume because we know that multi production continues. Um And so I think those are kind of the autos in the in the the defense industry, the two sectors of Russian manufacturing that are most reliant on uh foreign components.

requirements as well. Uh And so III I would say that Singapore's chip making has not been cutting edge, but I wouldn't say that that means it has been excellent. Um because there are many different ways you can actually define what's a profitable business. What's a um what's a, a technologically viable uh uh product? And I think Singapore is likely to remain an important hub of chip making. Uh and also uh a hub of assembling some of the tools that are used to make chips,

And as a result, there's not nearly as much differentiation between uh the firms and the pricing dynamics are also quite different profitability is uh generally lower. Um But Korea is by far the world's largest producer of uh memory chips and Korean firms, Samsung and Hex are are absolutely critical suppliers of of of memory chips when it comes to Japan, Japan is actually not primarily important as a chip maker. There, there are chips made in Japan, especially

for auto and industrial uses. Um But Japan is is really unique as a producer of highly specialized chemicals used in ship making and also some of the machine tools needed to manufacture advanced ships. And that's really where Japanese firms have uh pretty unique uh market positions.

Um Chris uh I believe in your book, you had mentioned as far as the memory chips, you know, the, the DRS and the NANS are concerned that is one area where the Chinese can probably eat Korea's lunch. Well, it's, it's,

And so actually, um I think one could argue that Korean firms have seen a potential major competitor really uh impacted by these controls in a way that might protect their market share uh for some time to come. So there are lots of unexpected um knock on effects of the controls that have interesting implications uh for for a number of different countries. But Korea being one of them,