USA vs. China – Road to Nowhere

Multi-patterning is an economic dead end.

• Huawei and SMIC appear to be in a position to mass manufacture leading-edge semiconductors but flooding the market with cheap semiconductors will require them to manufacture chips economically, which looks impossible given the techniques that they are using.
• The Financial Times (see here) has an exclusive with multiple sources claiming that SMIC is setting up new production lines that will enable it to manufacture silicon chips at 5nm and 7nm.
• This capacity will be largely absorbed by Huawei to make 5nm applications processors for its premium smartphones as well as 7nm AI training chips for the cloud.
• The new production lines will use US equipment with the lithography equipment having already been purchased from ASML.
• This confirms that the lithography equipment being used will be older generation deep ultraviolet equipment (DUV) that has been available for many years and, of which, China has a significant stockpile.
• DUV uses an argon-fluoride excimer laser that produces light with a wavelength of 193nm which has for years been effectively used to produce semiconductors down to a line width of 10nm.
• Going past 10nm has been extremely difficult as one has to figure out how to draw a line that is 7nm wide with a pencil that is 193nm (28x thicker).
• In order to achieve this, a technique called multi patterning is employed which is capable of producing 7nm and 5nm lines, but it does so at low yields.
• This means that a high proportion of the chips printed on the semiconductor wafer are defective and have to be discarded meaning that the cost of making the wafer is divided amongst fewer chips making them much more expensive to make.
• This is the heart of the issue because it means that while SMIC might be able to make 7nm and 5nm chips, it is unable to do so economically.
• This is why the entire industry has switched to lithography that uses extreme ultraviolet lasers that produce light at 13nm and this is currently the only way to get high yields at these nodes.
• One can see hints of this problem in the sources that the FT has quoted but I suspect that the FT is underestimating the degree to which this activity is a financial black hole.
• The FT’s sources admit that SMIC’s yields are way below TSMC’s at “below 30” and that it has to charge “40% to 50%” more than TSMC but I suspect that this is understating the problem.
• Both TSMC and Intel, who are two of the best manufacturers of semiconductors in the world failed to get multi-patterning to work economically for 7nm and in Intel’s case, the decision to go down this route was the beginning of all of the problems that it is still dealing with.
• Hence, I suspect that I am pretty safe ground in believing that SMIC is faring far worse and that the economics for SMIC in making chips at both 7nm and 5nm are truly awful and that yields are below 10%.
• This means that as SMIC makes more of these chips, it is either going to lose more money or it will have to charge Huawei more money for the working chips that it makes.
• I suspect that Huawei paying more is the most likely scenario by which the Chinese state will finance this economically ruinous strategy as SMIC is a Hong Kong-listed company while Huawei is not.
• Either way, vast amounts of money are going to be needed to keep this up which I do not think is sustainable especially given the weakness of the Chinese economy and the other crises that the Chinese state is having to deal with.
• Consequently, I think that there is no danger whatsoever of the Chinese competing with leading-edge semiconductor makers outside of China and I am sceptical whether they will be able to do so at any node below 20nm.
• However, at nodes above 20nm where there are no restrictions on what technology Chinese semiconductor makers can access, I think that they could become extremely competitive.
• A large factor in the total cost of ownership (TCO) of a semiconductor fab is the subsidy that its owner receives for building its fab in that government’s or state’s jurisdiction.
• A recent report by Boston Consulting Group concluded that on economics alone, making semiconductors in China is cheaper than anywhere else due to the higher level of state support offered compared to anywhere else.
• Hence, if China ramps up its lagging edge capacity and the current subsidies remain in place, it is likely to put significant pressure on pricing at these nodes.
• Customers will have to pay more for the same product if they want one manufactured outside of China which is precisely what Chris Miller argues in his piece in the FT (see here).
• However, I think the media is not being clear in terms of where the real ability of the Chinese semiconductor industry to compete lies.
• I put this at everything above 20nm, very little at 20-10nm and nothing below 10nm.
• Hence at the nodes below 20nm and especially at the leading edge, China is unlikely to make any headway and I suspect that even at home, it will be unable to continue meeting the financial requirements of such uneconomic manufacturing.
• That being said, China’s efforts on this front have reaped untold value in terms of public relations and propaganda as it is widely seen as being able to defy the efforts of the West to contain it even if in reality, it is unable to do so.
• I think that it will be the Chinese state’s perception of how valuable this publicity is to its ambitions that determines how long it will continue to finance a manufacturing system that will never be economically viable.