How EUV Lithography Works, in Plain Terms

Every advanced chip is, at its core, a pattern printed onto silicon. Lithography is the printing step, and extreme ultraviolet lithography, or EUV, is the technology that lets the most advanced fabs print features small enough for today's leading nodes. In mid-2026 that means the 2nm-class chips now ramping into volume production. The basic idea is simple. The execution is closer to controlled science fiction.

Printing with light#

Chip manufacturing builds circuits layer by layer. For each layer, the wafer is coated with a light-sensitive chemical called photoresist. Light is then projected through a patterned mask, the photomask, onto the resist. Where the light lands, the resist changes chemically, so it can be washed away or kept in the next step. That reveals a pattern, which then gets etched into the material underneath. Repeat this dozens of times, with different masks and materials, and you build up the transistors and the wiring that connects them.

The limit on how small a feature you can print is tied to the wavelength of the light: shorter wavelengths resolve finer detail. For years the industry printed with deep ultraviolet, or DUV, light at a 193nm wavelength and leaned on clever tricks, immersing the last lens in water and printing the same layer in several passes, to push it well past its natural limit. Those tricks eventually ran out of room.

Why 13.5nm light is so hard to use#

EUV light has a wavelength of 13.5nm, roughly fourteen times shorter than the old 193nm light. That short wavelength is exactly what makes it useful, and also what makes it brutally difficult to work with.

The core problem is that EUV light is absorbed by almost everything, including air and ordinary glass lenses. You cannot focus it the way you focus visible light. This single fact forces a cascade of unusual engineering choices.

Here is the consequence that ties it together. Because every surface reflects only part of the light, and EUV has to bounce off roughly ten mirrors plus the mask before it reaches the wafer, only a small fraction of the light that starts out ever arrives. That relentless loss is the reason the light source has to be so absurdly bright.

Making the light is the wild part#

Generating EUV at the needed intensity is the part that sounds invented. The common production method is called laser-produced plasma, and it works roughly like this. Tiny droplets of molten tin are fired across a chamber, tens of thousands of times per second. A high-power laser hits each droplet, and in most designs it hits twice: a first, gentler pulse flattens the droplet into a wider, better-shaped target, and a second, powerful pulse from a high-energy laser vaporizes it. The vaporized tin becomes a superheated plasma, and that plasma emits the 13.5nm EUV light. A large collector mirror gathers that light and funnels it into the rest of the machine.

This repeats continuously, droplet after droplet, to produce a source steady and bright enough to expose wafers at a commercial pace. Catching and reusing the tin debris, keeping the collector mirror clean, and stabilizing the plasma are all part of why these machines are so complex, and why, to date, only one company, ASML, builds them.

Why the industry needed it#

The payoff is resolution, and simplicity per layer. Before EUV, printing the finest layers with 193nm light often required multiple patterning: exposing and processing the same layer several times, with several masks, to build up one fine pattern. That added cost, steps, and chances for error. EUV can print many of those fine layers in fewer exposures, because the light itself resolves smaller features. It does not make chips cheap, since the machines are among the most expensive tools in any factory, but it keeps the smallest layers manufacturable at volume. The 2nm-class parts shipping in 2026 lean on EUV for exactly those critical layers.

DUV, EUV, and High-NA, side by side#

High-NA is a refinement of the same core approach, short light, mirrors, vacuum, a reflective mask, but the larger optics print a smaller area per exposure, so it is being adopted selectively at first. In mid-2026 Intel became the first to run High-NA EUV in high-volume logic production, on selected layers of its 18A node, the chips it sells as Core Ultra Series 3 and codenames Panther Lake. TSMC, by contrast, has stayed with standard-NA EUV plus multiple patterning for its initial 2nm generation. Both approaches print leading-edge chips; they are simply different bets on cost, yield, and complexity.

What to keep in mind#

A few honest caveats round this out. EUV is not used for every layer of a chip. The coarser layers are still printed with older, cheaper DUV light, and EUV is reserved for the critical layers that need its resolution. Numerical aperture, the figure that jumps from 0.33 to 0.55 in High-NA, is just a measure of how wide a cone of light the optics can gather, and a wider cone resolves finer detail.

The takeaway is that EUV is not magic. It is a chain of extreme engineering choices that all follow from one fact: the light is so short that everything normal about optics stops working, and the machine is built to cope. That is also why so few companies can print at the leading edge, and why a single toolmaker sits at the center of the whole industry.