Lithography is a crucial process in semiconductor manufacturing, using light to print intricate patterns onto silicon wafers. These wafers become the foundation for computer chips powering modern electronic devices, including AI technologies. As advanced chips drive global AI competition, control over lithography technology has become a key geopolitical concern.
We've released two forecast questions about lithography technologies developed with Ying Yi Dang, technology fellow at RAND. Below, we asked Dang about how crowd forecasts can help her team’s investigation and eventual policy recommendations for maintaining U.S. leadership in AI:
Q: What are you looking to understand from these questions, and why are they important?
A: We’re trying to understand China's potential strategic pathways in their quest for global AI leadership, given the recent U.S. export controls. The semiconductor ecosystem is complex and globally interconnected, with China holding strong positions in various segments of the value chain. The long-term effects of U.S. export controls on China are still debatable.
This project starts with the broad questions about the effects of the U.S. export controls and narrows down to China’s short-term strategic options. Specifically, I’m focusing on whether China can produce its own lithography machines by a certain date, as those machines are explicitly targeted by U.S. export controls.
By framing this as forecast questions, we can better understand China's semiconductor ecosystem, particularly in lithography production – a major point of tension in the U.S.-China AI race.
Q: How will these questions inform your research analysis? How are you planning to use the outputs?
A: These questions will be a way to draw out important insights about key factors in understanding China’s potential pathways for semiconductor indigenization. Indigenization refers to China creating domestic, localized capabilities to fabricate semiconductor manufacturing equipment needed to manufacture high-end semiconductor logic and memory chips to meet 100% of China’s advanced AI capability needs.
We will be comparing the crowd forecasts with forecasts from RAND analysts, and the outputs will help inform proposed policy actions as part of a broader strategy for U.S. leadership in AI.
Q: Are there specific resources or information you think would be valuable for forecasters participating in these questions?
A: There are two main pieces for forecasters to understand:
- The technical challenges of making lithography machines, and
- China’s semiconductor ecosystem factors.
For the first point, I’d recommend gaining an understanding of how these tools function, such as this resource from Intel. Really interested forecasters might also read Focus: The ASML Way by Marc Hijink.
For the second point, it is important for forecasters to understand potential approaches that China can take to indigenize the production of advanced semiconductors (see examples below), as well as the varying levels of indigenization they might reach (see graphic).
- Developing their own equipment: China may invest heavily in R&D to build semiconductor manufacturing equipment themselves, in order to reduce dependence on foreign suppliers and possibly make a game-changing technology innovation.
- Collaborating externally: China might seek partnerships with non-restricted international companies and form joint ventures to access critical semiconductor technologies.
- Poaching talent: China might actively recruit skilled semiconductor engineers from foreign countries with expertise in advanced chip design or manufacturing processes.
- Reverse-engineering existing equipment: China may acquire foreign technology in order to reverse-engineer and replicate advanced semiconductor machinery.
- Acting covertly or illegally: China may steal equipment or intellectual property, use front companies or diversion tactics like venture capital and private equity firms, or conduct hostile takeover of foreign entities to gain access to advanced semiconductor technologies.
Q: What forecast questions are part of this effort?
A: Forecasters can participate on two questions under the “China Lithography” tag on RFI:
- Will a Chinese organization produce DUV (ArFi) photolithography machines that are capable of high-volume manufacturing (at least 140 wafers per hour) before 1 January 2026?
- Will a Chinese organization produce or sell EUV photolithography machines that are capable of high-volume manufacturing (at least 90 wafers per hour) before 1 January 2030?
These questions will accept forecasts through November 30, 2024, and will not be scored due to stakeholder interest in a longer time horizon.