In October 2022, the US instituted new, wide-ranging export controls aimed at controlling a set of “chokepoint” technologies in the global semiconductor supply chain. China is one of the leading countries in AI, but AI advancements have historically run on compute, and China’s access to cost-competitive machine learning (ML) compute — the Chinese AI industry currently relies in large part on AI chips designed and fabricated abroad, and in particular Nvidia GPUs — is now in question.

The US export controls prevent the export of high-performance AI chips to China, as well as the export of advanced tooling and materials used to make such chips. Still, one way for China to gain better access to cost-competitive ML compute is to indigenously produce high-quality AI chips. In 2015, China set out an ambitious plan aiming to bring the proportion of chips bought by Chinese firms that is domestically produced from 15% to 70% by 2025; as of 2021 that number had gone up to 24%. With the October restrictions — and with US allies like Taiwan, Japan, South Korea, and the Netherlands enacting their own controls — Chinese firms now face substantial challenges in their attempts to indigenize semiconductor production.

Some of the most important inputs into the chip-making process is semiconductor manufacturing equipment (SME) – highly complex and specialized tools essential to the chip-making process. (Other important inputs include specialized materials, software, and more generally capital, talent, and know-how.) Many of these tools — including most tools necessary for producing the most advanced chips — are prohibited from being exported to China.

In particular, the most important type of SME is photolithography machines, which are some of the most complex machinery ever made by humans. These machines expose a photomask (which holds the pattern that needs to be etched onto the silicon wafer) to light; the patterned light then reaches a photoresist material on the wafer. The exposed parts of the photoresist dissolve or harden, and the remaining parts can be removed to yield a mask for the next deposition or etching steps that form the chip’s circuits. The photolithography industry is dominated by the Dutch company ASML.

China’s major photolithography maker is Shanghai Micro Electronics Equipment (SMEE), founded in the early 2000s. SMEE markets krypton fluoride (KrF) and argon fluoride (ArF) dry steppers for mature process nodes, but these either aren’t ready for mass production and/or are far behind (>15 years) Dutch and Japanese KrF and ArF machines in terms of resolution, throughput, and quality. SMEE does not market any more advanced argon fluoride immersion (ArFi) machines, though it is reportedly “on track to reveal its first scanner capable of producing chips on a 28nm process technology by the end of 2023”.

For more information, I have written a primer on indigenously made Chinese AI chips, intended to provide a useful introduction and reference for anyone interested in forecasting or learning more about China’s ability to indigenously produce AI chips. It explains some key concepts, provides an overview of the chip-making process, provides an overview of relevant export controls, and describes key inputs, context, and organizations relevant to understanding Chinese chip-making progress.