On May 15, the world’s largest contract chipmaker announced plans to open a production plant, or “fab,” in Arizona, US. If you know the industry, it doesn’t seem to make business sense: the Taiwan Semiconductor Manufacturing Company (TSMC) will build a 5nm fab in Phoenix, Arizona and start churning out chips by 2024, with a target of processing 20,000 wafers per month. The chipmaker plans to invest $12 billion through 2029.
First, the Arizona fab will be small, and not at all leading-edge by TSMC’s standards. A 5nm fab will be mid-range, at best, in 2024. Right now, TSMC itself already produces Apple’s A14 processor on 5nm nodes for the upcoming iPhone 12. Qualcomm, AMD, and Nvidia are working closely with TSMC to ensure high volume production of their 5nm chips by 2021. If TSMC sticks to its plan, the company will start high volume production of 3nm chips in 2022. Also, 20,000 wafers per month is a tiny amount—“like dipping your toes in the water”—compared to the 2.5 million wafers per month TSMC currently processes.
Second, it will be expensive. In a recent investors’ call, TSMC itself said that between a fab in the US and one in Taiwan, “there is a cost gap, which is hard to accept at this point.” In a nutshell, the world’s largest contract chipmaker just announced that they will open a small, mid-level fab in the US for $12 billion. Why would they do that?
Jan-Peter Kleinhans is the director of the project Geopolitics & Technology at Stiftung Neue Verantwortung, an independent, charitable, non-partisan tech policy think tank in Berlin.
Tune in to our Tech After Hours webinar discussion with the author, Jan-Peter Kleinhans, on “The Great Chip War: Can China achieve semiconductor independence?” tomorrow on May 28, 8pm (GMT+8). Space are limited so register now.
Because TSMC, maybe better than anybody else, knows that the semiconductor value chain is the football in the US-China competition over tech. And the contract chipmaker is right in the middle of this.
Even though their planned Arizona fab may not make a lot of sense economically speaking, it is an understandable long-term business decision to stay in the US government’s good graces. Especially since both the US Department of Commerce and the US Department of Defense have pushed TSMC for quite some time to open a fab in the United States.
Value chain chokepoints
If the US government perceives a foreign ICT (information and communication technology) vendor as a threat to their national or economic security, they will go to great lengths to curtail this vendor’s technological advances. Huawei is learning this lesson the hard way. The Chinese telecoms giant is the target of several export control measures by the US government.
Even though export controls are rather crude and antiquated policy tools, they are still effective if the market is highly concentrated. This is the case for the semiconductor value chain, and that is why the US government will most likely continue to utilize export control measures to cut Chinese companies off.
The global semiconductor value chain relies on a handful of US companies for certain production steps, most importantly electronic design automation (EDA) software and semiconductor manufacturing equipment (SME). EDA software is necessary to design any type of chip, and EDA vendors have close connections to both foundries, such as TSMC or Samsung, and SME vendors to integrate the next generation of production lines into their software. Right now, and for the foreseeable future, there is no way around Synopsys, Cadence, or Mentor Graphics EDA software if you want to design modern chips—and all three are US companies.
The market for SME machines, which foundries such as TSMC must buy to produce chips for their clients, is less concentrated but still largely under US control. Five companies dominate the world market: Applied Materials (US), ASML (NL), LAM Research (US), KLA-Tencor (US), and Tokyo Electron (JP).
In 2019, China’s largest foundry, the Semiconductor Manufacturing International Corporation (SMIC), tried to buy extreme ultraviolet (EUV) lithography equipment from ASML. EUV lithography is necessary to produce any leading-edge chip. But EUV equipment falls under existing export control regimes and so far, the Dutch government has denied the necessary license to sell the equipment to SMIC, partly due to pressure from the US government. Since ASML is the sole supplier of this kind of EUV equipment, SMIC will not be able to produce any chips that are 7nm or smaller.
By focusing on EDA software and SME machines, the US government has found a chokepoint in the industry. With the recent US export control measures, Huawei and its chip design subsidiary Hisilicon cannot use US EDA software to develop chips, and foundries are not allowed to use US manufacturing equipment to produce chips for Huawei.
The US approach only works because both EDA and SME are highly concentrated markets dominated by US companies. For the few suppliers outside of the United States, such as ASML, the US government uses diplomatic pressure to ensure that Chinese chipmakers are cut off from leading-edge technologies. That means, no matter how much funding SMIC can secure, if they cannot buy leading-edge manufacturing equipment, they will not be able to produce leading-edge chips anytime soon and will fall even further behind.
Of course, TSMC knows these chokepoints within the global semiconductor value chain intimately. That is why their move to open a small, financially unjustified foundry in the United States makes perfect sense from a business continuity perspective. TSMC’s commitment to the US can be interpreted as a trust-building measure, providing the company with potential leverage in future negotiations.
The truth is, TSMC’s fab will make very little difference to the US’s strategic position. That is why the US government is not united on this issue. Right after TSMC’s plans were announced, some US senators wrote a letter to Trump criticizing lack of transparency over potential subsidies and tax breaks for TSMC as well as national security concerns. They argue that the US government should instead invest in US semiconductor companies, such as Intel, to strengthen their national industry.
In times of geopolitical uncertainty, both governments and companies are trying to diversify their semiconductor supply chain. Because the semiconductor value chain is highly concentrated, US export control measures are highly disruptive.
Huawei’s latest announcement of collaboration with the French-Italian chipmaker STMicroelectronics is one way for Huawei to lessen their reliance on US-origin technology and get on better terms with European governments.
Meanwhile, Intel has just invested in a Chinese EDA company, potentially signaling to the Chinese government that they support a healthy Chinese semiconductor ecosystem.
While the US government is trying to isolate China’s semiconductor industry and prevent some of China’s national champions from making technological advances, corporations are continuing to make investments in a globalized industry.
Many suppliers within the semiconductor value chain will struggle not to fall victim to the tech rivalry between China and the US. This value chain is highly efficient, but not at all resilient, and every new export control measure the US government implements will be met by the Chinese Communist Party with accelerated efforts for self-sufficiency.