A laser pulse vaporises a 20-micron droplet of molten tin into a plasma hot enough to emit 13.5 nm extreme-ultraviolet light. That wavelength is absorbed by ordinary glass, so every optical element downstream — collector, relay mirrors and the patterned mask itself — has to be a reflector: multilayer Mo/Si mirror stacks tuned to reflect close to 70% of the EUV light per bounce, with real systems chaining 8–11 of them before the wafer. The reflected pattern is demagnified and projected onto photoresist, where exposed regions become soluble and are etched away, leaving the chip's nanoscale relief.
resolution = k1 · λ / NA
EUV: 0.40 · 13.5nm / 0.33 ≈ 16 nm
DUV: 0.40 · 193nm / 0.33 ≈ 234 nm
- Light source — switches λ between 13.5nm (EUV, tin-droplet plasma) and 193nm (DUV, ArF excimer laser). Everything else held equal, the shorter wavelength alone shrinks the smallest resolvable feature by roughly 14×.
- Numerical aperture (NA) — how wide a cone of light the final mirror/lens can capture; a larger NA sharpens the printed pattern but shrinks the usable focus depth.
- Process factor k₁ — bundles everything real fabs do to beat the naive diffraction limit (off-axis illumination, phase-shift masks, multiple exposures); it can't go below about 0.25 for a single exposure.
- The wafer grid below downsamples the same fixed mask pattern to the current resolution — watch fine circuit traces blur into blocky, merged shapes as the feature size grows.
Real-world relevance: EUV is why modern chips can pack tens of billions of transistors on a fingernail-sized die — the jump from 193nm immersion DUV to 13.5nm EUV around 2019 was what let leading-edge fabs keep shrinking transistors after multi-patterning DUV hit its economic limit.