How does EUV light print a chip?

A chip machine is a projector that shrinks. A mask holds the pattern for one layer of a chip, and the optics project it four times smaller onto a silicon wafer. The finest detail it can print is set mostly by the light.

ASML writes the limit as the Rayleigh criterion: CD = k1 · λ / NA. CD, the critical dimension, is the smallest feature. λ is the wavelength. NA, the numerical aperture, is how much light the optics can collect. k1 covers everything else in the process, and ASML puts its physical limit at 0.25.

Why the wavelength? Light spreads as it passes the edges of a tiny pattern, and the lens catches only part of that spread. Resolution is limited by the wavelength and by how much of the diffracted light the lens captures. So a shorter wavelength, or optics with a larger NA, prints finer patterns.

The numbers show what EUV buys. The best deep-ultraviolet machines use 193 nm light with an NA of 1.35 and print about 38 nm in one exposure, Wikipedia’s comparison shows. EUV light is 13.5 nm, about 14 times shorter. With an NA of only 0.33, it prints about 13 nm. The picture at the top draws three waves to scale: 39 EUV waves fit in one wave of green light.

With a laser and a stream of tin. Trumpf, which builds the laser, describes the source: a generator fires 50,000 droplets of tin a second through a vacuum chamber, and a CO₂ laser with more than 30 kilowatts of average power hits every droplet twice. ASML says the molten drops fly at 70 meters a second.

Why twice? A droplet is smaller than the laser’s focus spot, so a single hit could not pass on the pulse’s full power. A weak pre-pulse comes first and spreads the tin out; ASML says it flattens each drop into a pancake. The main pulse follows close behind and strikes with full power. The tin becomes a plasma hotter than the surface of the Sun.

That plasma is the light source. Tin ions missing 8 to 13 electrons give off light in peaks around 13.5 nm, while neutral atoms and solid matter cannot emit EUV at all. A collector mirror gathers the light and sends it on into the machine.

Because almost everything absorbs EUV light. ASML says it is absorbed even by air, so the whole light path, from source to wafer, sits in high vacuum. Glass absorbs it too, so there are no lenses. Every optic in the machine is a mirror, and so is the mask.

They are not ordinary mirrors. Each is a stack of alternating layers of molybdenum and silicon, and the small reflections from the many layers add up through interference. Each mirror still absorbs around 30% of the light that hits it, Wikipedia notes. After the collector, the beam meets four mirrors in the illuminator, then the mask, then six mirrors in the projection optics. With 11 reflections, only about 2% of the source light reaches the wafer: 0.7 multiplied by itself 11 times is about 0.02.

The mirrors also have to be almost perfect. Zeiss, which makes them, says that scaled up to the size of Germany, the largest unevenness on one of its EUV mirrors would be a tenth of a millimeter.

It prints one layer of a chip. The wafer is first coated with photoresist, a film that reacts to light. The mirrors project the mask’s pattern onto it, shrunk four times, one field at a time; on most EUV machines a field is 26 × 33 mm. ASML’s wafer stage places the wafer to within a quarter of a nanometer for each exposure.

Where the light lands, the resist changes. In the most common kind, positive resist, the exposed parts become soluble in a developer and wash away. The resist that is left protects the wafer while an etch removes material from the open areas. That is one layer’s pattern. A wafer may go through the cycle as many as 50 times, one layer after another.

At these sizes, chance starts to matter. EUV printing is particularly sensitive to random, or stochastic, defects, and Wikipedia notes that the odds of a failure rise steeply as features shrink.

High-NA works on the bottom of the formula. A larger NA means optics that collect more light, and a smaller CD. ASML’s High-NA machines raise the NA from 0.33 to 0.55, and their resolution goes from 13 nm to 8 nm, Wikipedia’s comparison lists.

When ASML showed the machine in February 2024, Reuters reported that it costs about $350 million and is the size of a double-decker bus. Its larger mirrors are made by Carl Zeiss. ASML said it lets chipmakers shrink the smallest features by up to 40% and nearly triple the density of transistors. The bigger optics have a cost: they shrink the pattern 4× one way and 8× the other, so each field is 26 × 16.5 mm, half the area of before.

The next step is more light. In February 2026, ASML told Reuters its researchers had found a way to raise the source to 1,000 watts, from 600 watts now. They double the tin drops to about 100,000 a second and shape each with two smaller laser bursts instead of one. That should let customers process about 330 wafers an hour on each machine by the end of the decade, up from 220 now.

1. 1 See

   Open his EUV machine. Under Follow, tap Tin, then set Time to Slow-mo. It runs 40 times slower, so you can watch each droplet hit twice: a weak pre-pulse flattens it, then the main pulse turns it into plasma.

2. 2 Change

   Tap Light and follow the beam from the plasma through eleven mirrors to the wafer. Reaches the wafer reads 1.4%, because in his model each mirror reflects about 68%.

3. 3 Understand

   Tap All, then switch Optics from 0.33 NA to High-NA. Smallest line falls from 12.3 nm to 7.4 nm: the Rayleigh criterion with a larger NA. Throughput falls too, because his model gives High-NA a little less light at the wafer and more time stepping between its smaller fields.

His model runs in the browser and works best on a larger screen. Split cuts the machine in half and Exploded pulls every module apart. Its numbers are his model’s, and the next section says where they differ from the companies’.

His notes say it plainly. It is “a generic EUV scanner, not a specific product,” built from public figures, and his post on X adds: “It is a model, not ASML data.” The droplets, the plasma and the beam are drawn much larger than real, and the internal layout is simplified.

Some of his numbers differ from the companies’. His plasma is about 220,000 K; ASML’s page says about 500,000 K. His mirrors reflect about 68% each and leave 1.4% of the light at the wafer, where Wikipedia’s round 70% leaves about 2%. His throughput comes from what his notes call “our model”: about 160 wafers an hour from a 250 W source, while ASML’s machines run at about 600 W and 220 wafers an hour, Reuters reported in February 2026. His October 13 post, and the video the film excerpts, give other figures again: eleven mirrors that each keep 71%, and one photon in 148 reaching the wafer.

So read his model for the path the light takes and why every step costs light, not for a real machine’s numbers. The film uses short excerpts of his video and of ASML’s own footage, credited on screen. Its narration takes its figures from ASML, Trumpf and Wikipedia, not from his panels.