Design a site like this with WordPress.com
Get started

As the Fog Lifts: Reflections on the Chip War’s Impact after One Month–Chip Fabrication

Four weeks into America’s declaration of war on China’s chipmaking future, today I would like to assess the impact of the policy for limiting China’s advances in fabrication from what we know thus far.

Will this Policy Work?

I have been very sympathetic to concerns that a broad unilateral set of controls would work at cross-purposes to maintaining an American technological edge over China––look no further than my last post­––but I think this unilateral approach has such a finely woven net that it cuts off many of the corporate countermeasures I worried would occur with broad controls.

Bateman in a recent piece in Foreign Policy voiced similar concerns about pushing allies past their breaking point (link) where they decide to buck American controls, but the calculations under these new regulations are not so simple.  Previously, the most likely, fastest scenario to de-Americanizing fab equipment involved American firms pursuing offshoring [assuming non Foreign Direct Product Rule( FDPR)] or somehow designing out American-origin  tech when under the FDPR (link).  The former was reportedly done by some firms––equipment from American vendors made in Southeast Asia was utilized for controlled nodes by China’s SMIC legally without the need to apply for a license––and various American vendors did scenario-planning about the latter. Admittedly, even a lawyer who formerly worked for the Bureau of Industry and Security could not fully explain how this de-Americanization under FDPR could be pulled off (at what point was zero American technology content reached). At the same time, the same lawyer regarded this scenario as a real possibility.  The new US Persons controls (yes yes they were in Export Control Reform Act of 2018 but not yet activated––probably more accurate to say detonated), which include American firms as well as real live American citizens and permanent residents, effectively block American firms from evading the controls.  Even if an American firm somehow manages to produce equipment offshore with zero American content, this US Persons control means such a savvy firm still cannot sell equipment to or even “facilitate” fabs for equipment meeting these metrics (16/14-nm logic; 18nm half pitch for DRAM; 128 or more layers for NAND) without a license.  Thus, there is no surprise that no FDPR is imposed for this equipment since it is not necessary.

The only other plausible scenario is one in which Japanese vendors de-Americanize fabs with or without ASML’s participation. To be clear, here we are talking about a scenario where these firms are displacing American equipment for not quite cutting-edge fabs using DUV rather than EUV equipment since pre-October 7 controls with Dutch acquiescence already foreclosed the possibility of creating advanced fabs with non-American/ASML equipment any time soon.  However, with “eleven or so” different types of controlled equipment monopolized by American firms (if we accept Under Secretary Estevez’s numbers link), the Japanese would face very difficult technical barriers to create such a de-Americanized fab any time soon.  Thus, the fact that Tokyo Electron dominates certain types of equipment, such as photoresist processing, does not mean creating 14nm FinFET fabs without American equipment will be easy, contra Nikkei Review’s recent piece suggesting the desperate need to get allies on board (link).

Concerns about the controls being overly broad to include activities that the US government was not aiming to limit have been addressed. BIS clarified on October 28 that it is targeting advanced equipment and the servicing of such equipment when it comes to US Persons.  There are clear carve-outs for less advanced fabs and back-end assembly and testing.  Moreover, as Under Secretary Estevez emphasized on October 27, licenses can be issued so for non-Chinese companies any overly broadly written regulations concerning metal deposition, for example, can resolved through granting licensing or even refining the regulations through revisions.

US National Security Council’s Tarun Chhabra and Commerce’s Alan Estevez are going later this month to the Netherlands to talk further controls with ASML, meaning presumably controls for DUV equipment that can be used for FINFET process. These controls affect 40% of ASML sales to China and sales to China are 16% of revenue so affects only over six percent of ASML’s overall revenue (link).  In the context of a large order backlog and the prospects of controlled technology fabs not being able to function any time soon without American equipment anyway, ASML will probably prove to be cooperative.

The incentives of Japanese capital equipment producers and ASML have to be placed in the global market context. There has been a growing support for investing in fabrication capacity in Japan, the EU and the US, so the current fab equipment market that falls under controls, which amounts to only 5% (2021) and 4% (2022) of world fab equipment, simply does not represent an enticing enough prize to do something both so financially and politically costly. After all, some of that controlled Chinese demand for equipment will be made up by increased demand elsewhere.  Yes, there is a burgeoning market downturn but beyond the short-term, the controlled technology fabrication capacity that could have been built in China will be built elsewhere where licenses aren’t need to sell equipment.

Thus, we do not actually have to accept Under Secretary Estevez’s assurances from October 27 that our allies are itching to get on board and magically transform these unilateral actions into multilateral ones.  The costs of creating an alternative would be high and with a low bang for buck ratio.  Calculations might be different if licenses weren’t forthcoming for sales to the Chinese fabs owned by allies.  However, one-year licenses have been issued for just these fabs (e.g., SK Hynix and Samsung) and the comments by Undersecretary Estevez suggest some openness to further licensing for non-Chinese firms going forward.

In sum, the players that really matter in capital equipment are the Netherlands, Japan and the US.  The US Persons decree effectively means that the US firms even with offshore production are thoroughly blocked from facilitating Chinese fabs.  The options for de-Americanized fabs at the controlled levels of technology in the near to medium-term are economically and politically unattractive for the major non-American capital equipment makers from Japan and the Netherlands. While Japan probably will not formally adopt the same export control regime as the US, it does not matter in the near to medium-term. Furthermore, the proposed Chip 4 Alliance of Japan, Korea, Taiwan and the US is basically irrelevant to achieve this outcome since only one of America’s East Asian allies is a big producer of capital equipment.

Beyond being a carefully crafted policy, we can also dismiss scaremongering around the economic cost of this policy.

Exaggerated Threat: Huge Economic Costs!

There are two major fallacies in thinking about the economic cost.  The first, already addressed above, is that the fabrication capacity affected by the controls is a huge part of the world market and thus the policy is an utter economic disaster (link).  At the height of China’s share of global fabrication equipment purchases in 2021, China’s 28 percent market share only translated into demand by local Chinese firms for controlled equipment of five percent of global market demand.  On top of this relatively paltry figure, Bernstein Research estimates half of that five percent will be compensated for by increased demand for equipment outside of China. Two to three percent of global demand is not a catastrophe.

The second fallacy is that all the additional capacity that was planned to be built was sustainable.  Eventually, some of the planned fabs would be cancelled or there would be excess capacity for years.  The past two decades have been a story of subsidized capacity increases in China, Korea and Taiwan forcing adjustment (i.e., loss of relative fabrication market share) on the other existing producers, Japan, the EU and the US, which were less generous in subsidizing fabrication.  The new era where the latter three woke up to the fact that fabrication, especially advanced fabrication, would be disappear in their economies without more proactive support (i.e., subsidies) has meant a new game of chicken.  Either every major producer subsidized until widespread overcapacity and concomitant asset destruction was reached or some producer(s) flinched before that outcome.  This new policy basically forces the adjustment back on China before it all ends in tears and mutual recrimination among excess capacity.  Of course, it still may end up in excess capacity and tears (recrimination against the US by Korea and Taiwan for its daring to do it what they do has already started), but this policy has set everything on a somewhat more sustainable path by removing one of the most generous and least efficient subsidizers from the equation.

Of course, just because the policy works to limit China’s fabrication and economic costs are exaggerated does not mean there is nothing objectionable about it.

Overreaction to China’s Perceived Threat?

Even while recognizing that this policy will most likely achieve its objectives of limiting China’s fabrication advances even in not-cutting-edge fabrication in the short to medium-term, there are still several vantage points from which to criticize or raise concern about the policy.

First, as I have raised repeatedly in this blog and elsewhere (link), which Reva Goujon also brings up in an interesting discussion on China Talk (link), broad controls do spell the end of fostering continued Chinese dependency on American technology.  And a critical part of that dependency was the fostering of commercial allies within China of the connection to American technology.  I’ll delve into more details on this topic in the next posting where I discuss the FDPR controls on supercomputing and AI chips.

The US government’s retort would of course be that China needs to be cut off from chips and semiconductor-related inputs vital to its military-civil fusion program and weapons.  There are two issues with this stance.  First, how reliably can the US cut off Chinese access to chips for its various weapons systems? I suspect that Jay Goldberg from the same China Talk episode linked above is correct in pointing out that many of these chips can be acquired through market channels.  Of course, the number of chips needed for which weapon and how many weapons are needed for the extreme contingency of war all are factors that need to be considered when assessing the efficacy of these controls.  For example, if a large number of very specialized chips (i.e., chips that are easier to track through channels) are needed to create an adequate supply of a certain weapon, then a control regime seems more feasible given the heightened ability to track and the need for the control to be very porous in order for China to get its hands on sufficient chips to make them militarily useful.  If the chips are not very specialized/easy to track or not needed in large numbers, then such a control regime probably won’t work. The second and related issue regarding MCF concerns is that many of the targeted technologies seem to make for a less than compelling case in terms of MCF/weapons concerns.  The policy targets DRAM and NAND flash, which are widely available from non-American producers, and are commodity chips i.e., sold in large numbers.  It is hard to imagine that cutting off Chinese fabs from producing these chips will prevent China from using gray channels to procure such chips. 

A second major objection to this new policy is that parts of it seem to be based on overreactions to Chinese purported developments.  Insiders at YMTC report its Xtacking to be far from mass production ready.  The US government’s targeting of NAND flash at 128 layers and placing YMTC on the Unverified List (what is functionally just the waiting list for the Entity List) suggest just such an overreaction to unsubstantiated reports about YMTC’s technological “leapfrogging”.  The fact that NAND flash is a commodity chip begs the question of what is really driving policy decisions here.

Similarly, targeting logic at 16/14 nm FinFET makes sense in terms of substantial issues of equipment re-use all the way down to 10 nm.  According to Intel and TSMC, something along the lines of 90-95 percent of all equipment at the least advanced FinFET nodes can be reused at the most advanced nodes.  However, this policy may be more driven by reports of SMIC’s “success” in 7 nm process technology than on concerns of what chips are produced on a 14 nm process. No capital equipment engineer, including many selling into SMIC and other Chinese companies, believes that SMIC’s 7 nm process jerry-rigging DUV lithography equipment is achieving anything but appallingly bad yields.  If the American national security concerns are based on 7 nm process technology being in the hands of Chinese companies, then targeting 16/14 nm process technology is misplaced.

One final point is that China has been doing a bang-up job of not catching up in semiconductors even before the US announced this new set of policies in October. Some, such as Doug O’Laughlin of the rightfully well regarded Fabricated Knowledge substack, assume that China’s semiconductor industrial development was upwards and onwards so China could have caught up to the US if not for the restrictions the US placed on China in this industry over time.  He invokes the lesson of Tiger Technology, a book published way back in 2000 about how Korea, Malaysia, Singapore and Taiwan developed their semiconductor industries (see his discussion in the same China Talk podcast mentioned above) through proactive industrial policy.  The problem is that China had already deviated from the upwards and onwards trajectory of Korea and Taiwan even before the American turn away from engagement. China has had ambitious semiconductor industrial policies since the 1990s and yet China’s own institutional features have always limited the technological catch up from these lavishly funded ambitions as I’ve documented here, here and here.  China’s major problems in industrial policymaking have been: 1. severe information asymmetries between firms, local governments and the central state 2. a pronounced bias towards SOEs in financing and policymaking and 3. use of state procurement, which can be easily gamed, as opposed to export performance as a major vehicle for industrial promotion.  Under Xi, the bias towards SOEs and the related phenomenon of the state advances the private sector retreats (国进民退) has only gotten worse, especially in the semiconductor sector.  Given this trajectory, the smartest American policy may have been to let Xi get hoisted by his own industrial policy petard.  Of course, the main exception to this downward trajectory has been the chip design sub-sector which we will discuss in the next post.

Advertisement

2 thoughts on “As the Fog Lifts: Reflections on the Chip War’s Impact after One Month–Chip Fabrication

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s

%d bloggers like this: