Contents
Introduction
Semiconductors represent one of the world’s most important industries, the core technology that powers the modern digital world.[1] Recognizing this vital role, China’s government has prioritized the sector, investing hundreds of billions of dollars to catalyze the development of an indigenous semiconductor ecosystem and to ideally cultivate globally competitive semiconductor firms across virtually all segments of the semiconductor value chain, from semiconductor design and fabrication to assembly, test, and packaging (ATP).
Thus far, those efforts have met with uneven success. With regard to the fabrication of leading-edge logic semiconductor chips, China’s flagship competitor, the Semiconductor Manufacturing International Corporation (SMIC), likely stands about five years behind global leaders such as the Taiwan Semiconductor Manufacturing Company (TSMC). Chinese companies As G. Dan Hutcheson, vice chair of research firm TechInsights, explained, “Ten years ago, [Chinese companies] were two generations behind. Five years ago, they were two generations behind, and now they’re still two generations behind.” Chinese firms Chinese competitors are even further behind with regard to semiconductor manufacturing equipment (SME), such as the lithography tools that make semiconductors: One commentator noted that Chinese firms might be as many as five generations behind in this field. The best machinery a Chinese company can produce makes chips that are 28 nanometers wide; the industry’s cutting-edge equipment can make 2-nanometer chips. As one analyst explained, “The best machinery a Chinese company can produce makes chips that are 28 nanometers wide; the industry’s cutting-edge equipment can make 2-nanometer chips.”
That said, Chinese semiconductor firms appear to be catching up in certain pockets: for instance, industry analysts view the design attributes and features of Huawei’s Mate 60 Pro smartphone as within 18 to 24 months of competitors’ versions. China has also made inroads in the production of legacy semiconductors (those greater than 28 nanometers), although Chinese firms appear to be competing in this sector on a more price- than innovation-intensive basis. In total, China continues to lag behind global leaders in most facets of semiconductor design and fabrication, but its firms’ intellectual property (IP) and innovation capabilities are accelerating rapidly as China pursues an aggressive whole-of-society strategy in an intense state-directed effort to achieve domestic semiconductor self-sufficiency.
A Brief Overview of the Global Semiconductor Industry
Modern semiconductors contain billions of transistors on a chip the size of a square centimeter, with circuits measured at the nanoscale (“nm,” a unit of length equal to one billionth of a meter). The very newest semiconductor fabrication facilities, which can cost over $30 billion to construct, produce semiconductors at 3 or 2nm (and even sub-2 nm) scales.6
The semiconductor sector is a $527 billion global industry that’s expected to become a trillion-dollar one by 2030.[7] Over 70 new semiconductor fabs are expected to be constructed worldwide by 2030 to satisfy this growing demand.[8] In short, semiconductors represent the commanding heights of the modern global digital economy, and this explains why leadership in the sector is so fiercely contested among nations, not least the People’s Republic of China (PRC), the European Union nations, Japan, South Korea, Taiwan, and the United States.
Computer circuit board technology (with processor), close up. Modern semiconductors contain billions of transistors on a chip the size of a square centimeter, with circuits measured at the nanoscale.
The semiconductor production process represents perhaps the most complex engineering task humanity undertakes. When all production phases are considered, the entire semiconductor production process extends from raw material procurement to end-product manufacturing.9 (See figure 1.) The core (or “narrow”) steps of the semiconductor production process include chip design, chip fabrication, and back-end ATP, with these steps supported by key inputs such as electronic design automation (EDA) software and SME such as lithography, deposition, and etch tooling.
Semiconductors represent one of the world’s most important industries, the core technology powering the modern digital world and empowering innovation and productivity growth across every industry.
Lithography, wherein a chip wafer gets inserted into a lithography machine and exposed to deep ultraviolet (DUV) or extreme ultraviolet (EUV) light and a pattern is printed onto a chip’s resist layer through a photomask, in particular represents a crucial step in the chipmaking process. EUV represents the latest and most sophisticated lithography technology, which Dutch firm ASML having invested €6 billion ($6.5 billion) to innovate over the past 17 years.10 The October 2022 export controls the United States placed on China included restrictions on equipment that can manufacture chips below 20 nm (impacting both DUV and EUV).11 In June 2023, the United States brokered a deal with the Netherlands to restrict exports of leading-edge EUV equipment to China.12
Figure 1: Facets of the semiconductor value chain[13]
The four most prevalent types of semiconductors are logic chips, memory chips (usually dynamic random-access memory (DRAM) or NAND “flash”), analog chips (those which generate a signal or transform signal characteristics, and are especially prevalent in automotive and audio applications), and power chips (those used as a switch in power electronics). “Advanced” or leading-edge logic chips are generally viewed as being sub-14 nm, while “legacy” (often called “mature-node”) chips refer to those manufactured using 28 nm or larger technology processes. Legacy chips are especially common in automobiles, medical devices, household appliances, energy, infrastructure, and aerospace products.
Lastly, several key business models define the industry. Integrated device manufacturers (IDMs) characterize firms—such as Infineon, Intel, Micron, Renesas, Samsung, SK Hynix, and Texas Instruments—which conduct all key facets of semiconductor manufacturing, especially design and fabrication, internally. Foundries such as Global Foundries, SMIC, and TSMC specialize solely in semiconductor manufacturing, often of the chip designs developed by “fabless” companies that specialize in designing application-specific chips, such as Advanced Micro Devices (AMD), chips for AI, high-performance computing (HPC), and graphics, Apple (mobile devices), NVIDIA (AI chips), or Qualcomm (5G and other wireless chips). [14]
Background and Methodology
The common narrative is that China is a copier and the United States is an innovator. That narrative often supports a lackadaisical attitude toward U.S. technology and industrial policy. After all, America leads (almost by right) in innovation, so there is nothing to worry about. But, first, this assumption is misguided because it is possible for innovators to lose leadership to copiers with lower cost structures, as has been the case in many U.S. industries, including consumer electronics, solar panels, telecom equipment, and machine tools.[15] Second, it’s not clear that China is a sluggish copier that’s always destined to be a follower.
To assess how innovative Chinese industries are, the Smith Richardson Foundation provided support to the Information Technology and Innovation Foundation (ITIF) to research this question. As part of this research, ITIF is reviewing particular sectors, including semiconductors.
To be sure, it is difficult to assess the innovation capabilities of any country’s industries, but it is especially difficult for Chinese industries. In part, this is because, under President Xi Jinping, China discloses much less information to the world than it used to, especially about its industrial and technological capabilities. Indeed, as The Economist wrote, “China’s chip industry operates under a shroud of secrecy. Breakthroughs and setbacks are often deemed to be state secrets, the divulgence of which can result in arrest.” [16]
Notwithstanding this, ITIF relied on three methods to assess Chinese innovation in semiconductors. First, we conducted in-depth case study evaluations of several Chinese semiconductor companies randomly selected from companies listed on the “2023 EU Industrial R&D Investment Scoreboard.” Second, we conducted interviews and held a focus group roundtable with global experts on the Chinese semiconductor industry. And, third, we assessed global data on semiconductor innovation, including scientific articles and patents. This report provides ITIF’s assessment of China’s current level of innovativeness in the semiconductor sector and offers a forward-looking perspective based upon rapidly evolving trendlines.
Importance of Semiconductors and the U.S. Role
The United States invented the semiconductor industry, harkening back to 1947, when Bell Labs’ John Bardeen, Walter Brattain, and William Shockley invented the transistor, a semiconductor device used to amplify or switch electronic signals and electrical power. In the mid-1950s, Jack Kilby at Texas Instruments and Robert Noyce and a team of researchers at Fairchild Semiconductor pioneered the integrated circuit (IC), placing multiple transistors on a single flat piece of semiconductor material, giving rise to the modern visage of a “semiconductor chip.”[17]
From 1990 to 2021, the U.S. share of global semiconductor production fell by 70 percent, from 37 percent to 12 percent.
U.S. enterprises continue to lead the world in semiconductor design (e.g., AMD, Apple, Qualcomm, etc.), and are highly competitive in semiconductor manufacturing (e.g., Intel and Micron) and SME tooling (e.g., Applied Materials, Lam Research, etc.) Indeed, as the Australian Strategic Policy Institute (ASPI) wrote, “The US excels in the design and development of the most advanced semiconductor chips and has a research lead in the technology areas of high performance computing and advanced integrated circuit design and fabrication.” [18] However, while the emergence of the fabless chip design ecosystem has preserved America’s leading revenue share in the global semiconductor industry, where the United States has severely faltered is in its extent of semiconductor manufacturing. In fact, from 1990 to 2021, the U.S. share of global semiconductor production fell by 70 percent, from 37 percent to 12 percent. [19] Conversely, over that time, China’s share rose from virtually nil to 12 percent. (See figure 2.)
Figure 2: Global semiconductor manufacturing capacity, 1990–2030 forecastFigure 20
The significantly declining U.S. share of global semiconductor manufacturing activity was the catalyst for America’s major recommitment to revitalizing the competitiveness of its semiconductor industry when Congress passed, and the Biden administration signed, the 2022 CHIPS and Science Act. The legislation authorized a 25 percent investment tax credit (ITC) and appropriated $52.7 billion to support the industry, including $11 billion for R&D activities and $39 billion for a “CHIPS for America Fund” to bolster U.S. semiconductor manufacturing by providing financial incentives for building, expanding, and equipping domestic fabrication facilities.[21] The United States (like Europe) seeks to double its share of global semiconductor manufacturing activity over the coming decade—objectives that will abut China’s efforts to also similarly increase its share in the years ahead.[22] That U.S. semiconductor companies depend on the Chinese market for 36 percent of their sales represents a significant concern as China seeks to grow its domestic industry in a quest for autarkic self-sufficiency.[23]
Overview of China’s Semiconductor Industry
As noted, China lags behind the global innovation frontier in most aspects of semiconductor design and fabrication, including with regard to key inputs such as EDA software and lithography equipment. However, Chinese enterprises such as Huawei’s HiSilicon and Biren have innovated increasingly competitive logic chips, especially those powering mobile devices and graphical processing units (GPUs) servicing AI applications. China has also demonstrated an ability to cost-effectively manufacture less-technologically complex mature-node logic chips. Chinese memory chip manufacturers Yangtze Memory Technologies Co. (YMTC) and ChangXin Memory Technologies (CXMT) seemed to be making significant strides up until about 2020, but appear to have fallen off the innovation pace relative to global leaders since.
Nevertheless, China is rapidly closing the gap across many facets of the semiconductor production process and is developing genuine IP and innovation capabilities across the board. In January 2024, Intel CEO Pat Gelsinger asserted that, despite China’s ongoing efforts to advance its semiconductor industry and design more sophisticated chip manufacturing tools, the country still lags behind the global semiconductor industry by approximately 10 years.24 While there’s no question that China’s behind, the real gap, as noted, is probably half that now, or about five years—at least for the design and fabrication of leading-edge logic chips. China continues to plough hundreds of billions of dollars into its semiconductor industry in an effort to close that gap. Moreover, over the long term, as one observer commented, “The likelihood of China developing advanced chip-making capabilities is almost certain.”25
China is rapidly closing the gap across many facets of the semiconductor production process and is developing genuine IP and innovation capabilities across the board.
Ever since the 2013 Third Party Plenum, the Chinese government “has made semiconductors the country’s top industrial innovation priority.” [26] China’s “2014 National Guidelines for Development and Promotion of the IC Industry” (often called the “National IC Strategy”) called for $150 billion in investments to establish a fully “closed-loop” semiconductor ecosystem in China. The plan unabashedly called for eliminating China’s trade deficit in ICs by 2030 and making China the world’s leader in IC manufacturing by then. [27] In 2015, China released its “Made in China 2025” (MIC 2025) strategy, which refined some of these targets, setting a goal of achieving 40 percent self-sufficiency in semiconductors by 2020 and 70 percent by 2025. [28] In reality, China is likely to only achieve 30 percent self-sufficiency by the end of 2025. [29] RAND’s Jimmy Goodrich has noted that China would probably need to invest at least an additional $1 trillion more to achieve true self-sufficiency in the industry. [30]
Nevertheless, these strategies make clear that China recognizes that semiconductors represent a foundational technology that underpins its economic and national security wherewithal, and that the country is willing to undertake the long-term investments necessary to reach its goal of achieving self-sufficiency and reducing dependence on foreign technologies in this critical sector. As Goodrich explained, “Semiconductors are integral to China’s goal of developing a complete industrial system. This means having capability up and down the supply chain, with the goal that, in the event of a conflict, China can sustain its industrial system and its economy.”[
](#_edn31)
China’s efforts have paid dividends. For instance, from 2001 to 2016, China’s share of global value added in the semiconductor industry grew almost fourfold, from 8 to 31 percent, while the United States’ share fell from 28 to 22 percent, and Japan’s share fell by over two-thirds, from 30 to 8 percent. (See figure 3.)
Figure 3: Country share of value added in global semiconductor industry[32]
Nevertheless, according to data from the Semiconductor Industry Association (SIA), considering global semiconductor industry market share, China commanded only 7.2 percent of the global market in 2023, compared with 7 percent for Taiwan, 9 percent for Japan, 12.7 percent for the EU, 18.8 percent for South Korea, and 50.2 percent for the United States. (See figure 4.)
Figure 4: Global semiconductor industry sales market share by nationally headquartered companies[33]
Despite the significant decline in the share of global semiconductor manufacturing the United States has experienced, it has maintained its leadership in overall global semiconductor market share thanks especially to its strengths in EDA and semiconductor design as well as to world-leading companies in the logic, memory, and analog market segments.
Japan, once the world leader in semiconductors, has experienced significant decline over the past three decades. Since 1988, Japan has gone from holding 52 percent of the semiconductor market share to just 7 percent, a decrease of 87 percent. This major decline can be attributed to an inability or unwillingness of Japanese firms to adapt during the 1990s and early 2000s as other significant semiconductor foundries were established. This lack of innovation, research, and development, coupled with falling Japanese electronics sales, made adaptation and competition in the semiconductor industry difficult for legacy Japanese companies.[34]
One of these new semiconductor manufacturers founded during the 1990s was TSMC. Now the largest manufacturer of semiconductors in the world, TSMC produces the majority of semiconductors made in Taiwan. The semiconductor industry accounts for 15 percent of Taiwan’s gross domestic product (GDP) while producing 60 percent of the world’s semiconductor chips, a large percentage of which are highly advanced.35 Though Taiwan accounts for just a small share of the global market (7 percent), its output relative to its size makes Taiwan the strongest relative performer. This can be seen using location quotient (LQ) information from ITIF’s Hamilton Index, which assessed 10 strategically important industries. An LQ is calculated as a country’s share of global output in an industry divided by the country’s overall share of the global economy.36
Figure 5 shows LQ data for the Hamilton Index category of the computers and electronics industry, a strategically important industry that includes semiconductors. Taiwan tops the list with an LQ of 8.79, exceeding close competitors such as South Korea and Singapore with LQs of 6.17 and 4.23, respectively. With LQs greater than 1, all three countries are overperforming, producing an output in the computer and electronics industry greater than the global average. China’s LQ of 1.56 means that the computers and electronics industry contributes 50 percent more to China’s economy than it contributes to the global economy. The United States’ LQ of 0.99 means that the computers and electronics industry contributes about as much to the U.S. economy as it contributes relatively to the global economy. Japan lags behind here with an LQ of 0.88. (See figure 5.)
Figure 5: Relative performance in computers and electronics industries[37]
SubSector Analysis of China’s Semiconductor Industry
This section analyzes China’s innovativeness by industry subsector, considering China’s strengths in EDA, semiconductor design and fabrication—of leading-edge and mature node logic chips, and memory chips—as well as the semiconductor ATP and SME/tooling sectors.
Electronic Design Automation
EDA refers to the software, hardware, and services that assist in the definition, planning, design, implementation, verification, and subsequent manufacturing of semiconductors.38 Historically, EDA has “constitute[d] a weak link in the PRC’s efforts to build a leading domestic semiconductor value chain.”39 As such, China has targeted progress in EDA since MIC 2025 and the 14th Five Year Plan, with the efforts now “logging some initial successes.”40
In March 2023, Huawei announced that it had achieved a number of breakthroughs in the development of EDA software, which it suggested would free China’s industry from reliance on foreign suppliers of those tools when producing semiconductors of 14 nm or more.[41] Huawei claimed it had substituted 78 (foreign) software and hardware items over the previous three years.[42] These breakthroughs likely came in partnership with Beijing-based Empyrean Technology, China’s leading EDA player. Empyrean claimed last year that it could fully support 7 nm digital processes and 5 nm analog processes.[43]
Beyond Empyrean, other leading Chinese EDA companies include GWX Technology, Primarius, Semitronix, Shenzhen Giga Design Automation, UniVista, and X-EPIC.[44] By 2020, Chinese companies had captured 11.5 percent of the domestic EDA market (up from 6.2 percent in 2018), with that share expected to increase to 14 percent by 2025.[45]
In June 2023, in Nanjing, China, Southeast University launched a National EDA Innovation Center, the first national technology innovation center in China dedicated to IC design (and which had been under development since September 2020). Yang Jun, a professor at Southeast University tasked with leading the center, noted its mission would be to “undertake the critical work of breaking the United States’ chokehold in EDA software,” not just for mature-node chips, but also ultimately for advanced chips including gate-all-around (GAA) transistors, which the United States has subjected to export controls.[
46 ](#_edn46) The Center is coordinating the work of multiple laboratories, facilitating the establishment of an independent EDA ecosystem by coordinating the participation of companies such as Empyrean, and promoting EDA competitions, such as the Integrated Circuit EDA Elite Challenge, with the goal of “increasing domestic market share of Chinese EDA players.”[
47 ](#_edn47)
Despite the progress, Chinese EDA enterprises have a long way to go, and even Empyrean in a recent prospectus noted that it “could not yet cover the full digital circuit process on its own.” [48] Observers believe that “it is more likely in the medium term that the PRC will make headway in EDA tools for processes at the 14 nm level or above.” [49]
Design and Fabrication
Semiconductor chip design and semiconductor fabrication are clearly separate steps in the semiconductor production process (and ordinarily would be evaluated separately), but since China’s progress in one is so intimately linked to its progress in the other, this section will consider both largely simultaneously, but segment by advanced-node and mature-node logic chips and then memory.
Semiconductor Design
As one report notes, “China’s design industry has rapidly increased in size since 2015, driven by factors such as the widespread availability of capital (including from both the government and private sector), government support, a desire to localize the industry, demand growth, the acquisition of foreign firms, and downstream users designing their own semiconductors.” [50] In fact, from 2010 to 2022, the number of semiconductor design firms in China increased nearly sixfold, from 582 to 3,243. (See figure 6.) The Chinese IC design industry’s total sales reached 557.4 billion yuan ($76.7 billion) in 2023. [51] Still, Chinese design firms accounted for only 8 percent of global design revenue in 2022, with no Chinese firms among the top 25 global design firms. [52]
Figure 6: Number of semiconductor design firms in ChinaFigure 53
RISC-V is an open-source instruction set architecture used to develop custom processors for a variety of applications, from embedded designs to supercomputers.[54] RISC-V essentially represents a set of computing standards that provide a common language for designing the processors found in devices such as smartphones, tablets, and Wi-Fi routers.[55]
China is actively embracing and innovating with RISC-V technology.[56] And as industry analyst Paul Triolo wrote, “[E]ager to wean Chinese firms off proprietary Western chip IP and architectures, such as x86 and Arm, Beijing is also fully supporting the development of the RISC-V reduced instruction set architecture. Chinese officials, and industry leaders such as Alibaba and its chip-design arm T-Head, have embraced the RISC-V approach over the past three years.”[57] Indeed, more than 100 “significant” Chinese companies are designing chips with RISC-V today, as are at least 100 more start-ups, according to industry analyst Handel Jones.[58] The top 10 RISC-V star-tups in China have secured funding totaling over $1 billion.[59] As one report notes, while many RISC-V “applications are in fairly mundane consumer products,” engineers believe “the technology will eventually take over more demanding tasks.”[60] For example, on November 24, 2023, DAMO Academy (Alibaba’s research division) unveiled three RISC-V-based processors. Elsewhere, “Chinese aerospace scientists have proposed using RISC-V to develop high-performance spaceborne computers,” while Alibaba’s T-Head has designed a RISC-V chip to operate a cloud-styled computing service.[61]
Advanced-Node Logic Chips
As The Economist wrote, “China’s chip industry [remains] far from the technological frontier. Even if Huawei and SMIC eventually succeed at producing 5nm chips, they will remain well behind Samsung, a South Korean tech giant, and TSMC, a Taiwanese foundry, both of which began mass-producing 3nm chips as far back as 2022.” [62] As G. Dan Hutcheson, vice chair of research firm TechInsights, explained, “Ten years ago, [Chinese companies] were two generations behind. Five years ago, they were two generations behind, and now they’re still two generations behind.” [63] However, this doesn’t mean China isn’t making considerable progress in certain pockets.
In August 2023, Huawei released the Mate 60 Pro smartphone, shown, which deployed the 7 nm Kirin 9000S chip and which was manufactured by SMIC using its SMIC N+2 process technology.
For instance, in August 2023, Huawei released the Mate 60 Pro smartphone, which deployed the 7 nm Kirin 9000S chip (which Huawei’s HiSilicon design arm architected) and which was manufactured by SMIC using its SMIC N+2 process technology “with capabilities that shocked the world in terms of its performance.” [64] As Dylan Patel of SemiAnalysis and Doug O’Loughlin of Fabricated Knowledge commented, “Comparing the MatePro’s [chip] to Qualcomm’s chip, made by Samsung, it is probably only 18 months behind and in some respects it’s actually just as good. … People didn’t think China would be capable of producing this.” [65]
Patel and O’Loughlin observed that the Mate Pro’s network performance (upload/download) speeds were on par with that provided by Qualcomm’s chips (and probably superior to those of the latest iPhone) and that its GPU/AI features (supporting phone features such as gaming, videos, and cameras) were mostly on par with those of competitors."66" They commented that “it’s a really good chip, with both hands tied behind their back [meaning it was manufactured without leading-edge EUV lithography equipment, but with older DUV equipment]” and “at worst it’s 18 months behind [global leading edge].”"67" As Patel concluded:
Put simply, [the] Kirin 9000S is a better designed chip than the West realizes. It has solid power and performance. Even with the lackluster export controls, this is a leading edge chip that would be near the front of the pack in 2021, yet was done with no access to EUV, no access to cutting edge US IP, and intentionally hampered. We cannot overstate how scary this is.[68]
Because SMIC could not make Huawei’s Kirin processor using the latest EUV equipment due to the aforementioned export controls, it used existing DUV equipment it possessed (pre-export controls) to employ a technique called “double patterning”—a process that uses multiple laser passes to etch IC designs at sub-20 nm resolutions—to manufacture at 7 nm.[69] However, this process is time consuming and expensive, making it difficult to manufacture using this process at volume scale, which explains why analysts expected Huawei to ship only 7 million of the MatePro phones in 2023, and perhaps 40 million in 2024.[70] Nevertheless, Patel has contended that SMIC’s N+2 process represents a “real, high volume production process technology” with strong yields.[71]
The SAQP case reflects that Chinese semiconductor developers are attempting to be as innovative as they can possibly be at process innovation, at least with the assets and technologies available, even if these are clearly behind the global frontier currently.
On Friday, March 22, 2024, Huawei (presumably with SMIC) announced that it had filed for patents for a technology called “self-aligned quadruple patterning,” or SAQP: a technique for etching lines on silicon wafers multiple times to increase transistor density.72 The SAQP method described in Huawei’s patent application involves etching lines on silicon wafers multiple times to boost transistor density, reduce power consumption, and potentially increase performance.73 Already, SiCarrier, a Chinese state-backed chipmaking developer, had been issued a SAQP-related patent in late 2023 that illustrates how to employ DUV and SAQP processes to achieve certain technical thresholds evident on 5 nm chips. The SAQP technology represents a similar approach to one Intel explored when looking to move to sub-10 nm processes, but as Anton Shilov wrote, “In the Huawei and SMIC case, quadruple patterning is the only technique that increases transistor density using the tools that the contract chipmaker already has.”74 In other words, the case reflects that Chinese semiconductor developers are attempting to be as innovative as they can possibly be in process innovation, at least with the assets and technologies available, even if these are clearly behind the global frontier currently. It’s also evidence that China is trying to reach a state of “good enough” to have sufficient domestic technologies to produce information communications and telecommunications (ICT) goods servicing domestic markets.
SMIC is currently developing its SMIC N+3 “5 nm” process that should feature 130 million transistors per square millimeter (mm^2). As Patel has written, “If nothing changes with current restrictions, we expect Huawei and SMIC to have a true 5nm-based chip in 2025 or 2026 with large scale AI chips not so long after.” [75] Ultimately, Patel has argued that “SMIC is at most only a few years behind Intel and Samsung” and “at worst only a handful years behind TSMC.” [76] However, as noted, most analysts would peg this gap at closer to five than just a couple years.
Elsewhere, a plethora of Chinese competitors are designing AI chips, including Huawei (with its Ascend chip), Biren, Tencent, Alibaba, Baidu, and MetaX. Patel has asserted that these firms “will soon be able to deliver on chips that are on par with Nvidia’s A100 [using] SMIC 7nm in 2 years at significant volumes.”[77] Commenting on Huawei’s Ascend GPU chips, one observer (requesting anonymity) noted that Huawei’s highest-end product is likely comparable to an NVIDIA H800 in most features save for energy efficiency. Huawei has developed a proprietary software platform, called “CANN,” that helps developers use its chips to build AI models.
In the GPU space, Biren Technology’s BR100 processor competes in the market against NVIDIA’s H100. The H100 features 80 billion transistors on the TSMC N4 process node whereas the BR100 is only 3 billion transistors behind the 7 nm process node.[78] Hong Zhou, cofounder and CTO of Biren Technology, asserts that the BR100’s architecture optimizes the data flow in depth, solving the bottleneck of data migration and insufficient parallelism through six technical characteristics, which means the BR100 chip has “achieved leapfrog progress” in performance and energy efficiency.[79] While the BR100 delivers powerful performance speeds, industry observers note that NVIDIA’s CUDA ecosystem, which tightly integrates hardware and software, is likely more attractive and has made NVIDIA an essential partner for developers and researchers.[80]
Chinese central processing unit (CPU) chipmaker Loongson recently launched its new 3B600 and 3B700 processors, which the company claims match Intel’s 10th-generation chips in single-core performance.[81] The company claims its latest iterations have seen up to a 20-fold gain in single-core capabilities, though company vice president Zhang Ge admits the company’s “chips lag mainstream offerings for multi-core” processors.[82] Nevertheless, China’s government is eyeing the company’s homegrown solutions as attractive offerings for the country’s lucrative education and government markets. Elsewhere, Wuhan Xinxin is building a factory that will be able to produce 3,000 12-inch high-bandwidth memory (HBM) wafers a month to expand China’s domestic base of AI-capable microchips.[83]
Larger-Node Chips
As noted, larger-node chips, also known as legacy or mature chips, are 28 nm in size or larger.Legacy chips But as one report notes, “Despite the name, legacy chips are not stale technology. The connotations associated with terms like ‘mature,’ ‘older,’ and ‘legacy’ are misleading because these categories of chips are constantly being refined for new requirements and applications.”Another report Rather, larger-node chips are foundational to many markets, including the industrial, aerospace, and defense sectors. Innovations in these chips include the use of wide bandgap materials such as silicone carbide (SiC) and gallium nitrate (GaN), important in clean energy applications. Elsewhere, as one commentator noted, “The main innovation for the auto industry is happening on mature nodes. They need to be very energy efficient and safe.”Commentary
China is likely to become an ever-more significant player in this sector of the market, but its basis for competitive advantage here is likely to be more predicated on massive scale (supported by large, state-driven industrial subsidization) that facilitates price-driven, not innovation-driven, competition.87
China will add more chipmaking capacity than the rest of the world combined in 2024, with 1 million more wafers a month than in 2023.
China will account for the most significant share of new semiconductor capacity coming online over the next several years. Indeed, analysts expect that China will add more chipmaking capacity than the rest of the world combined in 2024, with 1 million more wafers a month than in 2023. China’s share of global mature-node production is expected to grow from 31 percent in 2023 to 39 percent in 2027.[88] China currently commands 27 percent of global production capacity for chips in the 20–45 nm range, and 30 percent of global production capacity in the 50–180 nm range. Moreover, analysts expect China to build the most new fabs or major expansions in the 2022 to 2026 time period, with China bringing 26 new facilities online, and Taiwan 19. (See figure 7.) China’s IC output surged 40 percent to 98.1 billion units in the first quarter of 2024, driven primarily by production of legacy chips.[89]
Figure 7: New fabs and major expansions expected to come online, 2022–2026[90]
The growth of China’s semiconductor industry—especially in the legacy chip segment—has been driven considerably by massive industrial subsidization designed to help its companies reach economies of scale in production. Thus, as one report notes, “Chinese firms—supported by lower costs in China due to government subsidies and other factors—are able to offer significantly lower prices.” [91] For instance, China’s microcontroller processor manufacturer Giga Device was offering its products at prices that were 20 to more than 30 percent lower than non-Chinese competitors, such as the French firm ST Microelectronics, across most of 2022 and 2023. [92] Chinese subsidies allow Chinese semiconductor firms to compete in markets without having to earn market-based rates of return, and thus they can sell their products at much lower prices, which places firms that do have to earn market-based rates of return at a significant disadvantage while also disrupting the economics of innovation in the industry, as companies depend on the profits from one generation of semiconductor products to finance the R&D expenses that go into innovating the next generation. This has been a significant challenge for global memory chip manufacturers as well—and this dynamic threatens to extend to every semiconductor subsector which China aggressively subsidizes.
China’s aggressive subsidization of larger-node chips contributes to overcapacity, artificially lowers prices, and disadvantages firms that must earn market-based rates of return, significantly disrupting the economics of innovation in the global semiconductor industry, with deleterious downstream ramifications not just for mature-node chipmakers, but for chipmakers at all node sizes.
Memory Chips
The semiconductor memory industry has long been a strategic priority for China’s economic development.93 YMTC, China’s leading NAND maker, is a Chinese state-controlled joint venture launched by the National IC Industry Investment Fund, the erstwhile state university-controlled fabless semiconductor firm Tsinghua Unigroup, and the Hubei Science and Technology Investment Group, supported by $24 billion in initial government funding allocated for its initial Wuhan factory alone.94 Launched in 2017, CMXT is another Chinese-government-created semiconductor manufacturer, focusing on DRAM technology. CMXT was established as a project co-led by the local state-owned Hefei Industrial Investment Fund (HIIF) and GigaDevice Semiconductor Beijing (a Chinese designer of flash memory chips), with HIIF kicking in $8 billion to launch the effort.95
China’s aggressive subsidization of larger-node chips contributes to overcapacity, artificially lowers prices, and disadvantages firms that must earn market-based rates of return, significantly disrupting the economics of innovation in the semiconductor industry.
As Triolo wrote, “Memory is a very different sector than logic, highly commoditized and competitive, with no legacy nodes, requiring companies to constantly upgrade toward the most advanced processes.” [96] As he noted, “YMTC had moved rapidly up the NAND manufacturing curve, producing 128-layer NAND and moving rapidly toward more advanced processes, at 232 layers and above.” [97] A TechInsights article finds that YMTC introduced “the first 200+ layer 3D NAND Flash” on the market, ahead of rivals Samsung, SK Hynix, and Micron. [98] YMTC grew incredibly fast after its launch, reaching 5 percent of global market share by 2021, on a track to surpass 10 percent by 2027. [99] In 2022, Apple considered acquiring YMTC’s NAND memory chips for use in iPhones and iPads to be sold in China, and reportedly wanted to increase the order up to 40 percent of the chips required for all iPhones. [100] However, in December 2022, the U.S. Department of Commerce’s Bureau of Industry Security (BIS) added YMTC to its “Entity List”—a trade “blacklist” that subjects listed entities to export licensing requirements—and this action appears to have significantly stymied the company’s progress. In the months that followed, YMTC laid off 10 percent of its staff and reduced its equipment procurement and expansion plans. [101]
For its part, CXMT has became the workhorse DRAM company in China, but “the company faces significant challenges in the execution of its roadmap” and analysts don’t regard it as competing at the leading edge.[102] The company is “racing to produce China’s first domestic high bandwidth memory, a critical component in artificial intelligence computing.”[103] However, as Counterpoint research semiconductor analyst Dan Wang noted, “When your DRAM technology already lags behind global rivals, that puts your HBM technology at a disadvantage to be competitive in a fully commercial market … It’s not that easy for China’s national champion to break the international dominance in the sector. Its primary objective remains fulfilling China’s domestic demand.”[104]
Ultimately, as Triolo noted, while “key players like memory makers YMTC and CXMT [are] hobbled by lack of access to cutting-edge tools, [they are] still capable of producing useable memory” [105] In fact, TrendForce projects that the Chinese industry’s share of the NAND flash market will have peaked at 31 percent in 2023 before declining to 18 percent by 2024, while its share of the global DRAM market will fall from 15 percent in 2022 to 12 percent by 2025. [106] Nori Chiou, an investment director at White Oak Capital and a former analyst who looked at the IT sector, estimates that Chinese chipmakers lag their global rivals by a decade in HBM. [107] To be sure, YMTC (like CXMT) has shown itself to be an innovative firm; however, its lack of access to advanced tooling raises questions as to whether it can maintain its innovative wherewithal.
Semiconductor Manufacturing Equipment
Lithography represents a crucial step in the chipmaking process, wherein a chip wafer gets inserted into a lithography machine and exposed to DUV or EUV light and a pattern is printed onto a chip’s resist layer through a photomask. In the subsequent process step, “etching,” the wafer is baked and developed, and some of the resist layer is washed away to reveal a 3D pattern of open channels.[108] The essential point is that the lithography process determines just how small the transistors on a chip can be, which is why lithography innovations have been every bit as foundational as those in design to moving the industry toward lower process nodes.
As such, catching up in lithography has long been a central focus of China’s chip aspirations, with the efforts going back to China’s “Project 02” launched in 2008, called “The Project of Manufacturing Complete Sets of Technology for Very Large Integrated Circuits.” [109] Despite these efforts, as Triolo wrote, “Chinese lithography companies appear to be years behind industry leader ASML and Japanese firms such as Nikon and Canon, making lithography one of the critical technology bottlenecks.” [110] For this reason, Triolo observed that “officials in Beijing are developing new approaches to public-private collaboration to push innovation on key technologies, such as advanced lithography … easing the transfer of advanced state-backed R&D to designated private sector companies, by pushing companies to work together on critical technologies, and by pursuing approaches that have been successful in other sectors.” [111]
Shanghai Micro Electronics Equipment (SMEE) Group, China’s leading lithography developer, claimed in December 2023 to have successfully developed a 28 nm lithography machine, the SSA/800-10W.[112] (By comparison, TSMC was manufacturing 28 nm chips by 2011, and Intel was by at least 2014. Intel and TSMC are now, or imminently about to be, delivering chips at the 2 nm or even sub-2 nm level.)[113] Nevertheless, as one observer commented, “This advancement represents a major leap in China’s quest to close the technological gap in the global chip industry.”[114]Despite the claimed breakthrough, analysts have raised questions about how long it will take SMEE to produce such machines in bulk. SMEE’s current SSA600 series can utilize 90 nm, 110 nm, and 280 nm processes.
Naura Technology Group represents China’s largest chip production equipment manufacturer, its tooling focused on the etching process, followed by AMEC, which makes deposition equipment. Overall, the domestic market share of Chinese producers of wafer-fabrication tools rose from 4 percent in 2019 to an estimated 14 percent in 2023. (See figure 8.)
Figure 8: Chinese chip-fabrication equipment makers’ share of Chinese market [115]
Semiconductor Assembly, Test, and Packaging
After front-end chip fabrication, wafers are typically sent to other facilities for back-end manufacturing activities such as ATP. At this step, chips are cut from the silicon wafer, tested for performance, and packaged to protect the chip and allow for its integration into finished electronic devices by attaching electrical interconnections.116 Semiconductor ATP generally occurs through one of two business models: 1) as in-house ATP services performed by IDMs and foundries after fabrication, or 2) by outsourced assembly and test (OSAT) firms, which perform ATP activities for third-party customers.117
Figure 9: Number of ATP facilities per country/region, 2021[118]
As of 2021, China accounted for 27 percent (134) of the world’s 484 ATP facilities. (See figure 9.) By August 2023, Chinese ATP firms commanded 38 percent of the market, with the five largest OSAT players—JCET, HT-Tch, TF, LCSP, and Chippacking—all being Chinese.[119]
ATP has historically been viewed as labor intensive and lower value added than design and fabrication, explaining why, historically, firms have set up ATP facilities to a larger extent in developing countries.[120] However, one observer noted that “packaging is the new pillar of innovation in the semiconductor industry—it will change the industry dramatically” as new technologies enable chips to be combined and stacked, and their performance enhanced.[121] Packaging matters for the kind of high-power semiconductors needed for AI applications. Indeed, a shortage of a particular type of packaging known as Chip on Wafer on Substrate, or CoWoS, has been a key bottleneck in the production of Nvidia Corp’s AI chips.[122]
Indeed, as Triolo noted, “Packaging is now becoming a key part of overall production, with back-end packaging being designed into the entire production process, from EDA tools, to integration of IP from companies such as ARM, to 3-d packaging designs that enable greater functionality with a mix of semiconductors at different levels of complexity.” [123] He further observed, “When it comes to design and packaging, some Chinese firms are using chiplet design already, which is a design approach that integrates chips produced using different processes on one substrate, along with advanced packaging technologies, including 2.5- and 3-d packaging.” [124]
However, industry observers at an ITIF roundtable contended that “while China has made some progress, it really doesn’t have that capability on the leading edge” of semiconductor ATP.[125]
Advanced Semiconductor Research
Chinese researchers stand at the forefront of innovating some semiconductor technologies, although the question is the extent to which these can be commercialized and scaled by industry. For instance, Professor Liu Kaihui of Peking University has developed new wafers for semiconductors that are just one atom thick (thereby termed “2D”). These new 30.5 centimeter wafers are thinner and more efficient and could “cheaply and potentially revolutionize the semiconductor industry,” its creators claim."126" Elsewhere, researchers from China and the United States have jointly created a new type of stable semiconductor graphene, which exhibits performance 10 times higher than silicon and 20 times larger than the performance of the other two-dimensional semiconductors."127"
China’s semiconductor industry is not nearly as R&D-intensive as other leading nations’ semiconductor sectors.
In March 2024, at the IEEE International Solid-State Circuits Conference in San Francisco, Professor Zhou Jun and his team from the University of Electronic Science and Technology of China (UESTC) unveiled what’s been billed as “the world’s most energy-efficient AI chips for mobile devices.” [128] The chips’ design leverages a novel architecture that addresses AI chips’ thirst for power through multiple optimizations, including dynamic computation engines, an adaptive noise suppression circuit, and an integrated keyword and speaker recognition circuit. A UESTC press release noted, “The chip achieves a recognition energy consumption of less than two microjoules per instance, with an accuracy rate exceeding 95 per cent in quiet scenes and 90 per cent in noisy environments, setting new global benchmarks for both energy efficiency and accuracy.” [129] At the conference, the UESTC team also unveiled a chip that helps detect seizures in individuals with epilepsy that achieved a detection accuracy rate of over 98 percent. [130]
In July 2024, Chinese scientists announced development of what could be the fastest analogue-to-digital converter (ADC) for military use. The device can reduce the time delay of electronic warfare receivers from nanoseconds to picoseconds, or one-trillionth of a second. The chip technology would make radar signal detection and responses 91.5 percent faster than currently deployed technologies, nearly doubling the speed of combat to potentially give the Chinese military a critical edge. The chip is based on mature 28 nm process technology."131"
Analysis of Innovation Inputs to China’s Semiconductor Sector
This section examines indicators assessing China’s semiconductor competitiveness at the industry level, considering such factors as R&D intensity, scientific publications, and patenting levels.
R&D Intensity
Broadly, China’s semiconductor industry is not nearly as R&D-intensive as other leading nations’ semiconductor sectors. In fact, in 2022, China’s semiconductor-sector R&D intensity of 7.6 percent was just 40 percent of America’s 18.8 percent, and well below the EU-country average of 15 percent, or Taiwan’s 11 percent (South Korea and Japan recorded R&D intensities of 9.1 and 8.3 percent, respectively). [132] (See figure 10.) China’s 7.6 percent semiconductor-sector R&D intensity in 2022 was actually down slightly from the 8.4 percent it had invested in 2018. [133]
Figure 10: Select nations’ semiconductor industry R&D expenditures as a percentage of sales, 2022[134]
In terms of semiconductor companies’ levels of R&D investments (as reported in the “2023 EU Industrial R&D Investment Scoreboard”), Huawei led all Chinese semiconductor-related firms with an R&D intensity of 25.2 percent (although it produces a range of products beyond semiconductors), which was competitive though still below that of global leaders Intel, AMD, and MediaTek, which recorded R&D intensities of 29.6 percent, 25.9 percent, and 25.7 percent, respectively. (See Table 1.) Huawei was the only Chinese semiconductor company in the top 13 on this measure in the study.
Table 1: Leading semiconductor investors on the “2023 EU Industrial R&D Investment Scoreboard” [135]
Company
Headquarters
R&D Investment (Billions)