The Economist on Silicon Supremacy

A couple of weeks ago, the cover story of The Economist was Chip Wars: China, America and silicon supremacy. For the last few years it has been the biggest story in the semiconductor industry. You may already know the incredible fact that the value of semiconductors imported by China is greater than the value of all the oil imported by China.

PCAST covered a similar topic. Although their report was officially titled Ensuring Long-Term US Leadership in Semiconductors, it was not the challenge of Japan or Europe that was the focus of the report. It was China. For more details on that see my post PCAST: The President's Council of Advisors on Science and Technology.

As it happened, on September 29th, the same day that the Economist published (officially it dates its covers with the Saturday of the week, but in fact each edition goes online the preceding Thursday), the Hoover Institution up the road on the Stanford Campus published Chinese Influence & American Interests: Promoting Constructive Vigilance.

Obviously, in semiconductors, it is a hot topic. In fact, it has been hot for a couple of years. At SEMICON West in San Francisco in 2017 I titled my overview blog post of the conference SEMICON: China, China, China.

DARPA's Electronic Resurgence Initiative meeting took place in San Francisco this July. I covered various aspects of the meeting but the overview is my post The DARPA Electronic Resurgence Initiative (ERI) which covered Bill Chappel's introduction to the ERI, which he runs. In that post, I said:

The US government has a particular problem, with no access to leading-edge processes for semiconductor fabrication and state-of-the-art assembly, due to "our own regulations." I'm assuming this means things like fanout 3D packaging. However, I assumed that they had access to 14nm at GlobalFoundries' Fab8 in New York, but apparently that is too entangled with Samsung since it is their process.

Of course, to make it worse, 14nm is no longer the true leading-edge process, that would be 7nm. But Global Foundries recently refocused their effort and decided not to pursue 7nm (and EUV lithography that goes with it). For more on that, see my post GTC: GlobalFoundries Pivots. Intel has 7nm (they call it 10nm) although they have had widely-publicized problems getting it to yield.

Trade

It is difficult to understand trade in any context, let alone in a market as globalized as semiconductors. Let me start with what seems to be the economists' view (with a small "e", meaning most academic economists, although the magazine basically espouses the same position, as it has since its founding in 1843 to advance the repeal of the British corn laws, the big trade dispute of the day).

The economists' view starts with exchange. If you have a car to sell, and I want a car, then I can give you some money and drive off. You wanted the money more than the car. I wanted the car more than the money. So we both benefit from the exchange. If not—because I won't pay your price, or I don't like your car—then the exchange doesn't happen. It doesn't really matter who the two parties are. Safeway wants $5 more than a gallon of milk, and I want the milk more than the money, again we both benefit. From a trade point of view, it really doesn't matter if the two parties are in the same city, the same state, or even the same country. I'm unlikely to buy a used car or a gallon of milk from China, but Société in France wants $10 more than it wants a pound of Roquefort, and I'm in the mood for blue cheese more than $10. Again, we both benefit from the exchange. It doesn't really change anything that there are other companies involved along the path of the exchange, and they are in France and I'm in California.

This applies to semiconductors too. Apple wants a wafer of A12 chips more than it wants a few thousand dollars. TSMC wants a few thousand dollars more than a wafer. Both benefit from the exchange.

So the economists' view is that all exchange like this benefits both parties and thus anything inhibiting it causes a loss of value. Tariffs, to the economists, are "throwing rocks in your own harbor."

The really big gains to exchange come from the fact that it enables specialization. Cadence doesn't run its own cafeteria, it gets Guckenheimer to do it for us. Guckenheimer doesn't need to make their own cellphones since we play a part it getting that done for them. This works fine across country boundaries too. Fabless semiconductor companies in the US don't need to make their own wafers since foundries in Asia do that for them. Correspondingly, foundries can specialize in manufacture and don't need to design their own chips, since fabless semiconductor companies do that "for" them.

Another good rule in economics is to take the perspective of the consumer. By this standard, the traditional economists' view is correct, and voluntary exchange is always a good thing for both parties. Indeed, in that model, there isn't even such a thing as a "country", they are just lots of individual producers and consumers.

When, above, I quoted Bill Chappel saying that the US government has no access to leading-edge processes "due to our own regulations" he wasn't really being critical of the regulations. For obvious reasons, the US is not going to buy chips from China and then put them in its latest jet fighters or ships. They have the ultimate in tariffs, they are infinite. There is no price, even if each chip came with a twenty-dollar bill, at which the DoD is going to buy their chips from China. It is probably true that this causes a loss of value, in the simple financial sense that the DoD could buy chips from overseas more cheaply, This limits the amount of specialization that can occur, which increases costs. Apple can have phones "designed in Cupertino and assembled in China" but the DoD cannot. They care about more than plain economic gains from exchange.

Scaling up to Countries

I thought about this quite a bit after I went to hear Clayton Christensen talk about his forthcoming new book. For details, see my post Clayton Christensen on the Prosperity Paradox. He pointed out that individual companies doing what is best for them doesn't automatically end up with doing what is best for the country. So every electronics company making a rational decision to buy chips from a handful of foundries outside the US, also leads to the decision for the US to abandon leading-edge semiconductor manufacturing. This happened without anyone making a decision that this would be a good idea, or even realizing it was happening until it was too late to do anything about it. You can argue, if you like, that this doesn't matter, since we can buy wafers more cheaply than we can make them, this is all good. Or you can, especially if you are the US DoD, argue that this is a big problem since they need leading-edge chips, and now they can't get them.

But, for better or worse, the result is that the US is largely out of advanced semiconductor manufacturing.

It is not just semiconductor, but a lot of other areas. Furniture is one I read about recently. The US used to have a big furniture industry but between 2001 and 2012, over 60,000 furniture factories closed in the face of Asian imports. The US steel industry declined since the big users of steel, such as the automotive industry, can buy it cheaper from China and India. A similar thing happened with rare earths, where from the 1960s to 1990s most of the world's rare earths were supplied by the Mountain Pass mine in California, but it became uneconomic and was closed due to environmental restrictions and competition from China. Now the market is almost entirely supplied by China. Bur rare earths are very important in magnets and so are important in cellphones, windpower generation, electric traction in cars, and more.

In all of these cases, the economists consider this a good deal for the US, since we can buy semiconductors, steel, furniture, and rare earths more cheaply than we can make them (or mine them for the rare earths). It is only when you look outside the simple economic arguments that there is any reason you might worry. For example, around 2010 China was rumored to have decided to restrict exports of rare earths to some countries, or to give priority to Chinese companies. It's not just a one-way thing, in 2015 the US decided not to let Intel sell high-end chips to the builders of Chinese supercomputers. Of course, in both these cases, the economists' argument is that these are mutually beneficial exchanges that are not taking place, and so an unambiguously bad thing.

That Economist Article

I'm perhaps making it sound like everything is bad for US manufacturing and great for China. But, as The Economist (capital "E") said in their leader about the semiconductor industry:

Firms from outside America and its allies, such as South Korea and Taiwan, dominate the most advanced areas of the industry. China, by contrast, remains reliant on the outside world for supplies of high-end chips.

Actually, I don't know what The Economist means by "by contrast". It seems to me that both the US and China are mostly in a similar position, "reliant on the outside world" (meaning suppliers outside their country) for "supplies of high-end chips." Intel builds its own microprocessors of course. Micron builds some memories. In mobile, Apple mostly it buys from Taiwan. But Huawei, ViVo, Oppo, Lenovo, and the rest are the same. They either buy chips from the same foundries, or they use chipsets from Qualcomm or Mediatek that come from the same foundries in Taiwan and Korea (Samsung does have a big fab in Austin TX that runs foundry, so maybe some come from there).

Later in the piece, The Economist indulges in another bit of wishful thinking:

Today America has the edge over China in designing and making high-end chips.

I am sure the US probably has a greater number of high-end design groups, but there are islands of excellence in China, even if fewer in number. For example, Huawei/HiSilicon designs their own mobile applications processors. When the US stopped Intel selling high-end microprocessors to China, they designed their own chips. But China graduates a lot more engineers and computer scientists that the US, so that will probably change over time. It is perhaps technically true that America has the edge over China in manufacturing chips, since the US has Intel and China has nothing equivalent. Intel has fabs in other countries such as Ireland and Israel too. But it's a bit like deciding whether Malaysia or Thailand has the edge in automotive manufacturing. All the leaders in manufacturing are elsewhere—for leading-edge semiconductors, in Taiwan and Korea.

A third bit of The Economist leader that I'm critical of is in its conclusions. The US needs:

...to prepare for a world in which Chinese chips are more powerful and pervasive. That means, among other things, developing proper testing procedures to ensure the security of Chinese-made products. And tightening up on data-handling standards so that information is not being sprayed about so carelessly.

In other words, to get all the advantages of exchange that go with the economists' (small "e" again) model requires proving that chips not only do what they are meant to, but don't do anything evil too. I don't believe anyone has a clue how to do this, certainly not starting from the chips themselves, as opposed to design data like the RTL. As for the last sentence on spraying data around carelessly, I don't even know what it means. If it just means we need better security, then nobody is going to argue.

Trust

All transactions depend somewhat on trust. If I buy that gallon of milk from the supermarket, I assume that they are selling me real milk, not just some white fluid. If you buy a semiconductor chip, you are assuming it does what you expect. Despite what The Economist suggested, you can't actually check it completely. If you are buying from some future hypothetical large leading-edge Chinese semiconductor company, then you have to trust them. Otherwise, buy from someone else.

The US DoD doesn't trust just anyone. They only trust you if you follow their rules, manufacture in the US, perhaps using only US citizens, perhaps with security clearances. Automotive companies won't buy semiconductors from just anyone, they audit the procedures at the manufacturing plants and qualify the reliability of the semiconductor process.

If you want to read something scary about trust, then you should take to heart the lesson of Turing Award winner Ken Thompson, who shocked the entire computer science community by showing that you can't even trust a program even if you have the source code. I summarized his scary acceptance speech in Why You Shouldn't Trust Ken Thompson. 

If you use Cadence tools for designing your chips (and you should!) then you are trusting Cadence. Of course, in reality, we have a huge team of engineers doing what they can to make your design successful. But you are implicitly taking that fact on trust. An evil Cadence could be adding gates to your chip. An evil Intel could sell you chips that leak your information to the management processor. An evil AWS...you get the idea. You can check some things, but checking has a cost too.

You don't run lab tests on your milk before putting it in your coffee. It might seem like talking about milk is just a joke. But back in 2008, China had a scare about contaminated baby formula. As I said above, if you can't trust your supplier, get a new supplier, in this case Hong Kong (HK). People from mainland China stripped the shelves in HK to such an extent that HK mothers couldn't find formula at all, and so HK enacted a rule that people could only export a limited amount out of the territory. Of course, this meant there was beneficial exchange that was not taking place. The economists would say that the correct solution was that the price of formula in HK should have risen to the market-clearing rate so that there was enough on the shelves, which would also have had the side effect of people from other places (US?, Australia?) stepping up production to supply the newly attractive high-priced market. But at some point, if HK can't even feed its own babies, the pure financial nature of transactions is not the only thing to consider.

The big question behind all these reports and articles is whether semiconductors are like baby formula, where we need to feed our own babies whatever happens geopolitically, or whether it is like furniture, where all Americans can benefit from cheap furniture (paradoxically often sold by a Swedish company) at the cost of a relatively small number of non-strategic jobs.

More

How this all plays out over the coming years is one of the biggest stories in the semiconductor industry. I quoted from The Economist's leader but their main article on the topic is The semiconductor industry and the power of globalisation (probably need a subscription).

Sign up for Sunday Brunch, the weekly Breakfast Bytes email.