wafer map
at the wafer · 13 nm lines
CD — · LER — · defects —
drag to orbit · pinch or ⌘/Ctrl+scroll to zoom · slow time to catch a pulse
An EUV scanner: a plasma sun on the left, a mirror cathedral in the middle, nanometre choreography at the bottom.
Drop the dose and the lines break apart — at 13 nm you are counting photons. That fight is today's lithography frontier.
How an EUV scanner prints
- Smaller features need shorter light. A projector cannot draw lines much finer than its wavelength: resolution ≈ k₁·λ/NA. Deep-UV at 193 nm was stretched for two decades with immersion and multi-patterning; EUV jumps to λ = 13.5 nm — over 14× shorter — and 13 nm lines print in a single exposure.
- Nothing transmits EUV. At 13.5 nm every material absorbs — air, glass, water, everything. So the entire light path lives in vacuum, and there are no lenses anywhere: every optic is a mirror, coated with 40–50 alternating molybdenum/silicon layers ~6.9 nm thick that reflect by interference, like an iridescent beetle shell engineered to one wavelength.
- The light source is a star in a bottle. 50,000 times a second, a 27 µm droplet of molten tin falls through the source vessel. A pre-pulse flattens it into a pancake; ~3 µs later the main pulse of a CO₂ laser ignites it into a plasma tens of times hotter than the Sun’s surface. Highly charged tin ions radiate at 13.5 nm; the ellipsoidal collector throws the flash to the intermediate focus. (Drawn violet and red here — both beams are really invisible.)
- The photon budget is brutal. Each multilayer mirror keeps only ~70% of the light. Collector, four illuminator mirrors, the reflective mask, six projection mirrors — twelve bounces — so roughly 1.4% of the EUV ever reaches the wafer. Watch the drawn beam dim at every mirror: that fade is the real number, 0.7¹².
- The flattest surfaces ever made. The projection mirrors are figured to picometre-class accuracy. Scaled to the size of Germany, the tallest bump on one would stand less than a millimetre high. Their arrangement here is representative — the real prescription is one of the most closely held designs in industry.
- The mask is a mirror too. The reticle carries the chip pattern at 4× size in an absorber layer on top of the same Mo/Si stack, hung face-down and hit at a 6° angle so the reflection can escape. A thin pellicle membrane keeps particles out of focus.
- Why it scans. The optics correct aberrations well only inside a thin arc-shaped field, so the machine sweeps that slit across each die: the reticle scans through the fixed slit at exactly 4× the wafer’s speed, in the opposite direction, while ~260 pulses accumulate the dose at every point.
- Two wafers in flight. While one wafer is exposed, the other is measured — height map and alignment — on the second stage. Then the chucks swap. The stage must place a moving wafer to sub-2 nm overlay, at metres per second: the most precise mass-produced motion system in existence.
- Printing means counting photons. At 30 mJ/cm², a square nanometre of resist absorbs only ~20 photons. The inset simulates this honestly with Poisson statistics: lower the dose and shot noise visibly breaks the lines apart — the stochastic defects that are modern lithography’s hardest fight.
- The absurd numbers. ~180 tonnes, shipped in ~40 freight containers; over $150 M each; ~125 wafers per hour; 50,000 plasma flashes a second; mirrors aligned to picometres inside a machine the size of a bus. It is the most precise machine humans have ever built — and a museum that opened with a pendulum closes with it.
Honest exaggerations: EUV drawn violet and the CO₂ laser red (both invisible) · droplets drawn far larger and closer together than reality (27 µm spheres 1.4 mm apart would be invisible) · the pulse-step photon packet rides the mirrors in seconds (really nanoseconds) · geometry representative of the 0.33-NA class, not any proprietary layout · the cutaway itself — the real vessel is sealed, holding vacuum and a whisper of hydrogen.
Try: PULSE STEP through one 20 µs period · drop the dose to 21 and watch the lines shatter · push focus to ±100 nm · X-ray off to see the machine’s skin · Beam off to see what your eyes would really see (nothing) · wait ~25 s for the chuck swap.
Try: PULSE STEP through one 20 µs period · drop the dose to 21 and watch the lines shatter · push focus to ±100 nm · X-ray off to see the machine’s skin · Beam off to see what your eyes would really see (nothing) · wait ~25 s for the chuck swap.