Pages

14 December 2020

Taiwan, Chips, and Geopolitics: Part 1

By John Lee and Jan-Peter Kleinhans

As tensions have risen across the Taiwan Strait, the prospect of a Chinese attack on Taiwan has loomed larger. The need for countries worldwide to consider their position in the event of a cross-strait crisis has risen accordingly. One aspect of these calculations is Taiwan’s outsize role in the global semiconductor industry, and especially that of the computing chip manufacturer Taiwan Semiconductor Manufacturing Company (TSMC).

Semiconductors are the backbone of modern society, and the foundation of emerging technologies including artificial intelligence (AI), autonomous vehicles, and quantum computing. The U.S. export control measures targeting Huawei illustrate how the transnational semiconductor value chain can be “weaponized.” Vulnerabilities stemming from this value chain’s highly concentrated nature mostly work in favor of the United States and its allies, save in one respect: China’s ability to threaten Taiwan militarily.

In this context, TSMC’s dominance in global semiconductor production makes for a volatile mix with the geopolitics around Taiwan. It is gaining attention in government, the media, and online discussion as a factor in decision-making for capitals worldwide about the right stance toward Taiwan’s de facto independence and potential military measures that Beijing might take to reduce it. As debate unfolds over whether the United States should end its policy of “strategic ambiguity” on this issue, calls are being made for an express defense commitment to Taiwan that is justified by TSMC’s international importance.

On closer inspection however, TSMC appears unlikely to be a decisive factor shaping either Beijing’s calculus over the “Taiwan problem” or that of third-party governments. TSMC’s role in the global semiconductor industry does not change the fundamental guidelines shaping either China’s Taiwan policy or its technology policy. It neither gives foreign governments sufficient incentive to change their basic calculations vis-à-vis Beijing in the event of a war in the Taiwan Strait, nor leaves them without options to mitigate risks from losing access to TSMC’s services, at least over the long term.

The Semiconductor Value Chain

The global semiconductor industry relies on high transnational divisions of labor stemming from the increasing complexity of chips and the economic pressure to innovate. While the average desktop processor in 1998 consisted of roughly 7 million transistors, smartphone processors in 2018 consisted of 7 billion transistors. Among different types of semiconductors, logic chips in particular – such as processors or AI accelerators – depend on “node shrinkage,” improvements in the fabrication process allowing manufacturers to squeeze more transistors onto a square millimeter of silicon wafer. But not many chip manufacturers survived this competition for ever smaller feature sizes, dubbed the “More Moore race”: while in the late 1990s more than 20 companies operated 180 nanometer (nm) fabrication plants or “fabs,” today only TSMC in Taiwan and Samsung in South Korea have the capabilities to successfully run cutting-edge 5 nm fabs.

Why did the global market consolidate? Today’s advanced logic fabs operate on the atomic level, with hundreds of process steps, and rely on highly specialized suppliers for silicon wafers, chemicals, and manufacturing equipment. Equipment in particular has become increasingly complex and expensive, to the point that a single machine may consist of 100,000 components, weigh 180 tons, and cost over $120 million. With equipment making up 80 percent of the cost of fabs, TSMC estimates that its next generation 3 nm fab will cost close to $20 billion.

To run a cutting-edge logic fab, companies need extensive process knowledge, close research collaborations with their suppliers, deep pockets to constantly invest in new equipment, and substantial government backing. These drivers explain why the number of firms operating cutting-edge logic fabs went from more than 20 to just two within two decades.

They also gave rise to the “fabless” business model – only designing chips and relying on contract foundries to manufacture them. Alibaba, Apple, Hisilicon, Tesla, and others all design their own chips but rely on TSMC as their contract chipmaker. Furthermore, companies need to decide which foundry’s manufacturing process the chip design will be based on, creating strong lock-in effects. Switching foundries, for example from TSMC to Samsung, means almost complete re-design of the chip, implying hundreds of millions of dollars wasted.

Why TSMC Is Critical

TSMC is the largest contract chip maker in the world, with a market share of more than 55 percent. It is often the only viable option for chip design companies who want to develop cutting-edge logic chips – processors used in data centers, laptops, smartphones, automobiles, and military applications. TSMC’s dominance of advanced logic semiconductor manufacturing will most likely not change in a decade and perhaps longer, for several reasons.

First, Samsung – TSMC’s only competitor for 10 nm processes and below – is not a “pure-play foundry” but also designs and sells its own logic chips. This means that Samsung’s competitors will be wary to employ Samsung’s foundry business, given the close collaboration that the manufacturing process requires. Additionally, Samsung’s foundry business so far seems to be mainly focused on high volume contracts, rather than accommodating emerging players who want to bring new chip designs to market.

Second, the number three and four players in the contract chipmaking business – GlobalFoundries and UMC, the latter a Taiwanese firm – both dropped out of the “More Moore race” in 2018 at 14 nm, focusing instead on alternative manufacturing processes. These processes may eventually erode the market dominance that TSMC has built on extant silicon-based processes, but such an outcome is unlikely to play out on a significant scale until many years in the future.

Third, SMIC – the largest foundry in China and number five in the world by revenue – has struggled to catch up. To date, SMIC has introduced new process nodes roughly four years later than the market leader TSMC. Because of this, fabless companies rely on SMIC only for simpler, lower-end chips. Additionally, with U.S. export controls recently introduced against SMIC, it is unclear how fast the firm will be able to upgrade its capabilities to manufacture at nodes below 10 nm – necessary for advanced logic chips for cloud, high-performance computing, and smartphones.

Managing Dependencies

In this context, the importance of TSMC and of Taiwan as a chip manufacturing location becomes clear. Even if TSMC builds fabs outside of Taiwan, they will be trailing-edge: TSMC’s 5 nm fab in Arizona will only be operational in 2024, by which time the company will already be mass-producing next generation 3 nm chips in Taiwan. The business ecosystem within Taiwan that has facilitated TSMC’s rise is also unlikely to be replicated overseas quickly under market conditions, even with the cooperation of friendly governments such as in the United States.

Against this, it must be observed that many chip design companies do not require cutting-edge fabs. But for those that do need the smallest feature sizes, TSMC is usually the only viable partner. Examples include Tesla for electric and self-driving cars, Amazon for cloud computing, Alibaba and Intel for artificial intelligence chips, Qualcomm and Unisoc for mobile chipsets, and Xilinx for chips used in F-35 fighter jets. With ongoing digitalization of the world economy, an increasing number of industries depend on access to TSMC’s capabilities.

These capabilities might seem to provide strong additional incentives for China to forcefully seize Taiwan, given Beijing’s ambitions for technological independence and lack of other leverage points over the semiconductor value chain. This in turn might appear to raise incentives for third parties to make a stronger commitment to Taiwan’s independence. But TSMC’s undoubted importance must be weighed against various other factors. Once considered in context, it is unlikely to tip the scales in Beijing’s overall decision-making calculus or that of other governments, as our next piece will explain.

John Lee is a senior fellow at the Mercator Institute for China Studies (MERICS).

Jan-Peter Kleinhans is the project director for technology and geopolitics at Stiftung Neue Verantwortung (SNV), focusing on semidconductors as a strategic asset.

No comments:

Post a Comment