Two years have passed since US President Joe Biden’s administration pushed through the CHIPS and Science Act, which allocated US$52 billion in subsidies to encourage semiconductor manufacturers to build up their capacity within the United States. Over this period, the US also imposed broad restrictions on the export of chip-related technology to China. The goal of these measures was to achieve US self-sufficiency in microchips and to impede China’s quest to achieve the same. Have they succeeded?

US efforts to limit China’s access to chipmaking technologies focus largely on the most advanced integrated circuits (ICs), which account for over 80% of the global semiconductor market. There are three main types of ICs: logic chips (used for data processing in mobile phones and personal computers), Dram (a type of random-access memory commonly used in PCs and servers), and Nand  (flash memory).

China produces all three, with logic chips coming from Semiconductor Manufacturing International Corporation (SMIC), Dram from ChangXin Memory Technologies (CXMT) and flash memory chips from Yangtze Memory Technologies Corp (YMTC). Though these firms are not industry leaders, they have been catching up with their competition, both in terms of technology and market share.

But Chinese firms target the low-end segment of their markets, and breaking into the higher-end segments is becoming increasingly difficult. After all, the cost and complexity of chip fabrication is rising fast, and US sanctions severely restrict Chinese firms’ access to the technologies needed to produce the most advanced semiconductors – specifically, logic chips with node sizes below 14 nanometres (nm), Dram below 18nm, and Nand  memory chips of 128 or more layers.

For example, the most advanced extreme ultraviolet (EUV) photolithography systems – US$200 million “photocopiers” used to transfer chip-design patterns – are produced exclusively by one Dutch company, ASML. Taiwanese chip giant Taiwan Semiconductor Manufacturing Company (TSMC) has more than 200 EUV sets – its cutting-edge 3nm logic-chip fabrication alone requires more than 50 – and South Korean behemoth Samsung Electronics has about 50. China, by contrast, has none, owing to US sanctions.

Whereas advanced logic and Dram (12 to 14nm) chips require EUV systems, Nand chips – which comprise hundreds of layers with the same geometry – do not. And this is the segment where Chinese firms are nipping at the heels of global leaders like Samsung. YMTC’s 128-layer 3D Nand product, rolled out in 2021, was on par with its competitors’ offerings. As of October 2023, the world’s most advanced 3D Nand memory chip was found in a YMTC-made solid-state drive that was launched quietly in July of the same year.

China might soon be able to make similarly deep inroads into the Dram market, thereby advancing its quest for microchip self-sufficiency. The process of making high-bandwidth memory (HBM) chips, a type of advanced Dram chip, does not necessarily require cutting-edge EUV photolithography, and HBM chips are generally easier to produce than traditional Dram chips with fewer than 10nm nodes. This means that China can make its own versions (17 to 19nm) without the latest equipment. China is already strong in advanced packaging technologies, such as through-silicon via, that HBM requires.

HBM chips are in high demand. The technology overcomes constraints on memory-transfer speeds and allows for greater energy efficiency, so it is emerging as the preferred solution for artificial-intelligence models and high-performance computing. South Korea’s SK Hynix currently produces the most advanced HBM chips – and has benefited massively from the AI boom. But Chinese firms are working hard to catch up: while China lags behind South Korea and Taiwan in US-registered patents related to traditional Dram and logic chips, it leads the way in HBM patents (see chart).

Leading Chinese firms Cambricon Technologies and Huawei have reportedly made some strides in AI processors. To accelerate progress, China’s government is discouraging companies from purchasing Nvidia’s H20 chips, which were designed for the Chinese market in response to US sanctions.

The Chinese government’s push for chip self-sufficiency is also driving Chinese chipmakers to turn to local firms for the equipment they need. As a result, equipment producers like Naura Technology Group and Advanced Micro-Fabrication Equipment have recorded remarkable surges in revenue and profits. Meanwhile, the localization rate of chip equipment and components in China has risen sharply, from 21% in 2021 to more than 40% late last year.

South Korea and Taiwan – which currently dominate chip markets – have responded to all this by increasing their domestic capacity, especially in the high-end segment, while maintaining lower-capacity production abroad. Samsung produces its core product – Dram chips – in South Korea, operates a Nand factory in China, and is building a logic-chip foundry in the US.

For its part, TSMC produces lower-end logic chips for Chinese customers in China and high-end chips for US firms – such as Apple, Advanced Micro Devices and Nvidia – in Taiwan, where all four of its “giga-fabs” are located. The company is developing fabs in the US, Europe and Japan, but the capacity of each will be equivalent to just 3-5% of its giga-fabs. In 2028, when all the planned factories are complete, their combined capacity will still amount to just 20% of the company’s total.

TSMC’s foreign fabs will also be one to two generations less advanced than its giga-fabs at home. For example, though the fabs TSMC is building in Arizona will manufacture 2nm chips, they will be behind the frontier by the time production starts.

So, the US remains far from self-sufficient in microchips and is unlikely to become so any time soon. The same is true for China – particularly when it comes to the most advanced chips – though the country is managing to overcome significant headwinds to strengthen its position in the industry.

Keun Lee is a distinguished professor of economics at Seoul National University, a fellow at CIFAR, an editor at Research Policy and a former vice-chair of the National Economic Advisory Council for the President of South Korea.

Copyright: Project Syndicate