EUV Lithography: How a Chip Is Printed with a Drop of Tin and Some Mirrors

Printing a chip is, at bottom, photography. A silicon wafer is coated with a light-sensitive layer and the pattern on a mask is projected onto it with light, reduced four times: where the light lands, the layer changes; where it doesn’t, it stays. Layer by layer, tens of billions of transistors are printed this way. The process is called lithography, and its hardest form, EUV, is done today by exactly one company. This page is about what happens inside that machine, and how one Dutch town became the narrowest point in the whole industry.

Why ordinary light is too coarse

Lithography’s limit is physical: a wave of light struggles to draw a line much finer than its own wavelength. This is diffraction; squeezed through a small opening, a wave spreads sideways, and the image blurs. For decades the industry dealt with it by shortening the wave, from visible light to ultraviolet and finally to 193 nanometres. That “deep ultraviolet”, DUV, still prints most chips today, and engineers squeezed an astonishing amount out of it: they shone the light through a film of water, they printed the same layer twice and then four times.

By the late 2010s transistors were so small that those tricks were raising cost and time faster than they raised density. A far shorter wave was needed. EUV, “extreme ultraviolet”, has a wavelength of 13.5 nanometres, fourteen times shorter than 193. The problem is that light like that does not exist in nature in any usable quantity.

Light made from drops of tin

Inside an ASML machine, 13.5-nanometre light is made like this. In a vacuum chamber a generator fires droplets of molten tin, each about 25 micrometres across, a third of the width of a human hair, at 50,000 a second and 70 metres a second. Each droplet is hit by two laser pulses. The first, weak one flattens it into a disc, so the second has more surface to strike. The second, a powerful CO₂ laser, vaporises the disc into plasma at, by Zeiss’s figure, nearly 220,000 degrees, some forty times hotter than the surface of the Sun. That plasma is what emits extreme ultraviolet.

This happens 50,000 times a second, and the laser has to hit each droplet to within less than a micrometre, or there is no light, or not enough of it. The power of the light decides directly how many wafers the machine can print in an hour. Sources in production today run at 600 watts; in February 2026 ASML showed a 1,000-watt one, which uses 100,000 droplets a second and an extra shaping pulse.

Mirrors instead of glass

Light at 13.5 nanometres has one awkward property: almost everything absorbs it, air included. So the whole path, from source to wafer, runs in vacuum, and the lenses that ordinary lithography relies on are useless here, because glass does not pass this light, it swallows it. Mirrors take the place of lenses, and ordinary mirrors will not do either. The mirrors, made by Zeiss in Germany, are built from alternating layers of molybdenum and silicon, up to a hundred of them, each a few atoms thick, spaced so that the reflections from every layer reinforce one another. Even so, the best such mirror reflects only about 70 percent of the light. The projection optics alone hold six mirrors, and each loses a third: only a small fraction of the source’s light reaches the wafer. That is why the laser’s power matters so much.

A machine that arrives on three aircraft

One EUV machine weighs up to 180 tonnes and, by ASML’s account, takes about forty containers, twenty trucks and three cargo aircraft to deliver. The current generation, the NXE:3800E, prints 220 wafers an hour. ASML does not publish a price; press estimates put it around €180 to 200 million.

The next generation, High-NA, the EXE:5000 and EXE:5200B, uses bigger mirrors: the optics’ numerical aperture rose from 0.33 to 0.55, which draws 8-nanometre lines in a single exposure instead of 13. That has a cost. The larger mirrors throw light onto the mask at angles where ordinary reduction no longer works, so the optics are anamorphic: they shrink the image four times in one direction and eight times in the other. The exposure field is halved as a result, and to keep throughput up the wafer stage has to accelerate at 8g and the reticle stage at 32g, four times the previous generation. The EXE:5200B’s specification is 175 wafers an hour; in September 2026 an ASML executive said it was doing about 135 in practice, with 175 on the roadmap. One machine, by 2024 reports, costs about $380 million. Intel took the first in December 2023; by September 2026 ten were running at four customers, and Intel has used High-NA on some layers of its 18A process, the first high-volume use.

How one company became the industry’s narrowest point

ASML was founded in 1984 as a joint venture of Philips and ASM International, and its first product failed: the PAS 2000 moved the wafer by oil pressure, and the oil leaked, in a clean room where a speck of dust is a catastrophe. The company had no money to build everything itself the way its Japanese rivals did, and that became its strategy: ASML assembles and integrates, and buys the parts from specialists, its optics from Zeiss, of whose semiconductor division it has owned 24.9 percent since 2017. A modular machine can be taken apart and fixed on site, and the PAS 5500 of 1991 made the company profitable for the first time.

The move to EUV was an existential risk, and ASML did something unusual: in 2012 it asked its customers to pay for the research. Intel put in €3.3 billion, for 15 percent of the shares and research funding; TSMC took 5 percent, Samsung 3. The world’s three largest chipmakers became part-owners of their supplier, and in 2013 ASML also bought Cymer, the maker of its laser source, for €1.95 billion. Today ASML is the only company in the world that builds EUV machines. In 2025 its sales were €32.7 billion, its net income €9.6 billion, and it sold 48 EUV systems. For 2026 it expects sales of €43 to 45 billion, and on 25 September its market value was about $666 billion, the highest in Europe.

Who may buy one, and who may not

When one campus governs the most advanced chips in the world, it is politics too. China has never received an EUV machine, because a licence has never been granted. Since September 2023 the Netherlands has required licences for the most advanced DUV machines as well; the list widened in 2024, measuring and inspection equipment was added from April 2025, and in November 2025 the controls went onto the EU’s own list. In the United States a bill introduced in April 2026 would ban the sale and servicing of every DUV immersion machine to China; it is not yet law.

The other side is not idle. In 2023 ASML itself disclosed that a former employee in China had taken technical data. In December 2025 Reuters reported that in a closed laboratory in Shenzhen former ASML engineers, some under false names, had assembled an EUV prototype: it produces extreme ultraviolet light but has not printed a chip, lacks optics of Zeiss’s class, and is aimed at 2028, with 2030 the realistic date according to the sources. ASML’s reply was one line: replicating the technology is “no small feat”.

What comes next

Three things are on ASML’s roadmap. A source of 1,000 watts, giving more than 330 wafers an hour, before 2030. High-NA at full throughput and 90 percent availability, by the end of 2026. And Hyper-NA, at a numerical aperture of 0.75, for lines of a few nanometres, “around 2030” and still at the research stage. Each step is dearer and slower than the one before, and every one of them runs through the same town.

If you are reading this on a phone, the chip showing you the text was almost certainly drawn with light from a machine assembled in Veldhoven. In Veldhoven, hardly anyone talks about it.