Future TechnologyThe 12-inch photomask is what makes ASML's $400 million machine worth buying
Key takeaways
- ASML has agreement from TSMC, Samsung and Intel to move to 12-inch photomasks, up from a 6-inch standard that has held for decades
- The change raises throughput on high-NA EUV systems by roughly 40 percent, which is what makes a $400 million machine pay for itself
- Larger masks remove reticle stitching, where a chip too big for one exposure field has to be printed in pieces and joined
- The timetable is long: a pilot mask line around 2031 and production readiness by 2033, with Samsung using high-NA for DRAM by 2028 and TSMC on advanced nodes from 2030
ASML has secured agreement from TSMC, Samsung and Intel to move the industry from 6-inch photomasks to 12-inch ones. The change raises throughput on its newest high-NA EUV systems by roughly 40 percent.
No new machine was announced. This is a standards agreement between three companies that compete on almost everything else, and it is the part of the lithography roadmap that decides whether the machines already being installed are economically sensible.
Why the mask size became a problem
High-NA EUV uses anamorphic optics. The trade for the higher numerical aperture that makes finer features possible is that the exposure field projected onto the wafer is halved.
Under the 6-inch mask standard, which has held for decades, that halved field is smaller than a modern AI accelerator die. So the chip pattern has to be split across two separate masks and joined on the wafer, a process called stitching.
Stitching works, and it costs. It adds masks, alignment steps and yield risk at exactly the point in the process where a defect is most expensive. A 12-inch mask restores a field large enough to hold the whole die, and the stitching step disappears.
Throughput, not resolution, is the economic question
A high-NA EUV system costs around $400 million. At that price, resolution alone does not justify the purchase. Wafers per hour does.
A 40 percent throughput improvement changes the arithmetic on every one of those tools. It is the difference between a machine that can only be justified for the highest-margin leading-edge parts and one that makes sense across a broader slice of a fab's output.
That is also why the agreement needed all three of ASML's largest customers rather than one. Photomasks are made by a separate supply chain, with its own blanks, writers, inspection tools and pellicles. None of that retools for a single buyer.
The timetable is measured in nodes, not quarters
Samsung plans to use high-NA EUV for DRAM by 2028. TSMC is targeting advanced logic nodes from 2030. The 12-inch masks themselves are aimed at a pilot line around 2031 and production readiness by 2033.
Those dates are worth sitting with. A decision announced in September 2026 lands in volume production seven years later, and the mask supply chain has to be rebuilt in the interval. This is why lithography roadmap news tends to be under-covered relative to its importance: nothing visible happens for years, and then every process node after it depends on the choice.
The part that is hard to copy
The Decoder reports Huawei is running a parallel effort through equipment maker Yuliangsheng and a set of domestic suppliers, aimed at reducing China's dependence on ASML lithography.
Building a competitive EUV scanner is the widely discussed obstacle. It is not the only one, and arguably not the hardest. A scanner needs a mask supply chain, a pellicle supply chain, inspection and metrology tools, and a set of customers who have all agreed to the same physical format at the same time.
That last item is the one that resists a national programme, and this is analysis rather than reported fact. Three commercial rivals coordinating a format change on a seven-year horizon is a governance achievement as much as a technical one, and it is the kind of thing that is slow to build and slow to replicate. Huawei's effort is real and is making progress on individual components. The coordination layer is a different problem.
What this actually signals
The useful reading of this week's agreement is not that high-NA EUV just got better. It is that three companies with every reason to hedge have now committed to the same roadmap, which removes a large piece of uncertainty from ASML's order book and from everyone planning fabs around it.
For anyone tracking AI compute supply, the mask decision sets an outer boundary on how large a single die can economically get, and when. Chiplets and advanced packaging exist precisely because that boundary has been binding. A larger exposure field relaxes it, seven years from now.
