Semiconductors represent perhaps the world’s most-important industry, as they are the foundation of a wide array of products and services. Moreover, they play a key enabling role in emerging technologies such as artificial intelligence (AI), high-performance computing (HPC), 5G, the Internet of Things, and autonomous systems, among others.
Unlike industries in which China has already gained significant global market share—including high-speed rail, solar panels, and telecom equipment—China’s global market share and competitiveness in semiconductors, especially with regard to Chinese-headquartered firms, is still quite modest, with the global leaders largely based in Europe, Japan, South Korea, Taiwan, and the United States.
It is because of this that China has targeted the industry for global competitive advantage, as detailed in a number of government plans, including “Made in China 2025.” China has taken a wide range of steps to propel itself into becoming a major global semiconductor competitor. However, while some of these policy actions are fair and legitimate, most are not and are “innovation mercantilist” in nature, seeking to unfairly benefit Chinse firms at the expense of more-innovative foreign firms.
Competition can drive innovation, but only if it is market-based. When Apple came out with the iPhone and helped drive Blackberry from the market, this spurred innovation, because it was based on consumer demand for a better product, with innovation driving the change. In contrast, Chinese semiconductor firms lag significantly behind the global leaders, usually by two generations of chip development, and Chinese firms patent less than the global leaders. As such, Chinese chip sales largely depend on unfair support from the Chinese government; and each sale reduces the pace of global semiconductor innovation by taking market share and revenue away from more-innovative non-Chinese firms. In fact, this report estimates that without Chinese innovation mercantilist policies in the semiconductor industry, there would be more than 5,000 additional U.S. patents in the industry annually than there are now.
This report provides an overview of the semiconductor industry and the innovation dynamics driving it, including an explication of why innovation mercantilist policies harm innovation. It then describes Chinese innovation mercantilism in the sector and examines the deleterious effect of China’s policies on global innovation in the sector. Finally, it provides policy recommendations for how policymakers can address these challenges.
The term “semiconductor” refers to a solid substance—such as silicon or geranium—which has electrical conductivity properties allowing it to be used either as a conductor or an insulator. Semiconductors, also referred to as integrated circuits (ICs), constitute the brains powering electronic equipment, providing the computational and storage capacity underpinning digital computing. Semiconductors pack as many as 30 billion transistors onto a chip as small as the size of a square centimeter, with circuits measured at the nanoscale (“nm,” a unit of length equal to one-millionth of a meter) level, with the very-newest semiconductor fabrication facilities producing semiconductors at 5ânm and 3ânm scales. Leading-edge semiconductors contain transistors that are 10,000 times thinner than a human hair, operating at tolerances smaller even than the size of the coronavirus. In 2019, the global semiconductor industry generated $412 billion in revenues and shipped over 1 trillion semiconductors. Analysts expect the industry to grow to $730 billion by 2026.
In 2019, U.S.-headquartered semiconductor enterprises held a 47 percent market share of global semiconductor industry sales (down about 5 percent from the 51.8 percent share they held in 2012), followed by South Korean firms with 19 percent, Japanese and European firms with 10 percent each, Taiwanese firms with 6 percent, and Chinese enterprises with 5 percent. (See figure 1.)
However, production shares are different, as many U.S. semiconductors are produced in places such as Taiwan and China. In fact, as of 2019, the United States possessed just 11 percent of global semiconductor fabrication capacity, whereas South Korea held 28 percent, Taiwan 22 percent, Japan 16 percent, China 12 percent, and Europe 3 percent. (See figure 2.) China’s share of global semiconductor fabrication capacity doubled from 2015 to 2019. As of year-end 2020, there were just 20 semiconductor fabrication facilities (“fabs”) operating in the United States.
Figure 1: 2019 Global semiconductor industry sales market share by nationally headquartered company
Figure 2: Global semiconductor fabrication capacity by country
The four main types of semiconductor are logic chips, memory (usually dynamic random-access memory (DRAM) or NAND) chips, analog chips (those which generate a signal or transform signal characteristics, and which are especially prevalent in automotive and audio applications), and discrete chips (ones designed to perform specific electronic functions). In terms of market share by location of company headquarters for each major segment of the global semiconductor industry, in 2019, the United States led clearly in logic and analog, South Korea led in memory, (followed by the United States), and European firms led in discrete semiconductors. Chinese-headquartered companies registered but a 9 percent share of the logic market sector and a 5 percent share of the discrete sector. Moreover, as this report shows, if not for massive Chinese government interventions to stand up a domestic industry, Chinese-headquartered companies likely wouldn’t even have had these shares. (See figure 3.)
Figure 3: Semiconductor industry market share, by segment, by nationally headquartered company, 2019
In 2019, the logic segment accounted for $107 billion in global sales, memory for $106 billion, and analog for $54 billion. Intel is the world’s leader in logic chip market share. Texas Instruments, Analog Devices, and Infineon account for the market leaders in the analog chip market, with market shares of 19, 10, and 7 percent, respectively, as of Q1 2020. Samsung and SK Hynix (both headquartered in South Korea) followed by Micron (United States) lead the world in DRAM production, accounting respectively for 44 percent, 29 percent, and 21 percent of global market share as of Q1 2020. Intel, Samsung, TSMC, SK Hynix, and Micron led competitors in final forecasted 2020 semiconductor sales. (See Table 1.) This table also shows the strength of the U.S. fabless sector: U.S.-domiciled semiconductor design companies accounted for 65 percent of fabless global sales in 2019, with leaders including Qualcomm. Broadcom, NVIDIA, Apple, and AMD.
Table 1: Top 15 2020 forecast semiconductor global sales leaders
As Table 1 illustrates, the industry is highly globalized, with a large number of countries fielding enterprises competing across multiple facets of semiconductor production, from semiconductor design, to manufacturing, to ATP (assembly, test, and packaging) activities. In fact, each segment of the semiconductor value chain has, on average, firms from 25 countries involved in the direct supply chain and firms from 23 countries involved in support functions. Over 12 countries have enterprises directly engaged in semiconductor chip design and 39 countries have at least 1 semiconductor fabrication facility, while over 25 countries have enterprises engaging in ATP activities. (See figure 4.) A considerable amount of value gets created along each segment of the semiconductor production process, with the U.S International Trade Commission (U.S. ITC) estimating that 90 percent of the value of a semiconductor chip is split evenly between the design and manufacturing phases, with the final 10 percent of value production accounted for through ATP activities.
Figure 4: Number of countries with enterprises participating in various phases of semiconductor production activity
The global breadth and scale of the value chains needed to produce some of the world’s most sophisticated technological devices have given rise to a number of business/operating models in the industry. (See figure 5.) Historically (dating back to the 1950s and 1960s), the semiconductor industry consisted primarily of integrated device manufacturers (IDMs)—that is, firms which conduct all key facets of semiconductor manufacturing, especially design and fabrication, internally. Firms such as Infineon, Intel, Micron, Renesas, Samsung, SK Hynix, and Texas Instruments remain leading IDM players to this day.
Figure 5: Operating models in the semiconductor industry
In 1987, Morris Chang founded the Taiwanese firm TSMC (Taiwan Semiconductor Manufacturing Company), which pioneered the foundry business model, concentrating on contract manufacturing for other semiconductor companies, often ones focused on designing semiconductors for application-specific purposes such as AI, wireless communications, or HPC uses. This essentially represents outsourced manufacturing, or “manufacturing as a service”—and it revolutionized the industry, giving rise to a host of new players beyond TSMC, including America’s Global Foundries, China’s Semiconductor Manufacturing International Corporation (SMIC), and Taiwan’s United Microelectronics Corporation (UMC).
The advent of foundries in turn supported the rise of the fabless industry; that is, companies which focus on semiconductor chip design, such as (now-fabless) Advanced Micro Devices (AMD) (chips for AI, HPC, and graphics), NVIDIA (graphics chips), and Qualcomm (5G and other wireless chips). Collectively, this is referred to as the “fabless-foundry” model.
Outsourced ATP (which may also be referred to as outsourced assembly and test, or OSAT) is performed by a number of global players, including Amkor (United States), ASE Technology (Malaysia), J-Devices (Japan), Power-Tech (China), and Siliconware Precision Industries (Taiwan). On the front end of the process are companies and consortia that focus on semiconductor research and development (R&D) activity, such as CEA-Leti (France), Imec (Belgium), ITRI (Taiwan), SEMATECH (United States), and the Semiconductor Research Corporation (United States).
Whereas almost 30 companies manufactured integrated circuits at the leading-edge of technology 20 years ago, only 5 do so today (Intel, Samsung, TSMC, Micron, and SK Hynix).
Also of high import are another set of companies—notably Applied Materials (United States), ASML (the Netherlands), KLA Tencor (United States), and Lam Research (United States)—which manufacture the machines and tooling equipment that run semiconductor fabs. The global semiconductor manufacturing equipment industry generated $62 billion in revenues itself in 2020, with revenues expected to rise at a 9 percent compound annual growth rate (CAGR) to $96 billion by 2025. Finally, a number of enterprises, especially ones from Japan, South Korea, and Taiwan, manufacture chemicals and components essential to the semiconductor manufacturing process. For instance, fluorinated polyimides, a group of specialty polymers that provide physical strength and heat resistance, are produced by Daikin Chemical (Japan), DuPont (United States), Kaneka Asahi Kasei (Japan), and Taimide Technology (Taiwan).
In other words, a key driver of the global semiconductor industry has been specialization, as enterprises—and indeed entire industrial ecosystem clusters within nations—have elected to concentrate their competitive energies on mastering key facets of the semiconductor production process (for instance, Dutch strength in extreme ultraviolet (EUV) lithography, Japanese strength in chemicals and production equipment, Korean strength in memory chips, Taiwanese strength in foundries, or Malaysian and Vietnamese strength in ATP activities).
There’s been no industry with an innovation dynamic so clearly and effectively codified as the semiconductor industry’s Moore’s Law. Coined by Intel co-founder Gordon Moore, the “law” (which technologically speaking refers to “process-node scaling”) represents the notion that the number of transistors on a microchip doubles about every two years, effectively meaning a semiconductor’s capability in terms of speed and processing is doubled, even though its cost is halved. Moore’s prediction has proven remarkably prescient—and up to the present day at least, highly reliable. The innovation process captured by Moore’s Law has delivered tremendous improvements in semiconductor performance and cost: The number of transistors per wafer has increased by a factor of almost 10 million (since Moore articulated the law in 1975), yielding a 100,000-fold gain in processor speed and a cost reduction of more than 45 percent per year for comparable performance. (See figure 6.) Moore’s Law has also offered somewhat of a guiding innovation pathway for the industry, providing an orientation for the efforts of pre-competitive research consortia and their development of long-term industry roadmaps. However, while some have come to take Moore’s Law for granted, one study found that the number of researchers required to achieve Moore’s Law today is more than 18 times larger than the number required in the early 1970s.
Figure 6: Increasing performance, decreasing relative costs of semiconductors over the past half-century