Workings 门道中文版

The Printing Limit

The smallest feature one exposure can print, 1980 to 2038 — and the twelve years it did not move.

Every transistor on a chip has to be drawn by light. The number that governs how small it can be drawn is the critical dimension: the smallest feature a lithography machine can resolve in one exposure. Halve it and four times as many transistors fit in the same square millimetre. For most of the industry's history that single number set the pace — and then, for twelve years, it stopped moving.

Half a century of the smallest printable feature
The critical dimension since 1980, and how pushing the boundary came to rest on one company.
1,000 nm 100 nm 10 nm 1980 1990 2000 2010 2020 2030 smallest printable feature, single exposure · log scale NOW g-line · 436 nm i-line · 365 nm KrF · 248 nm ArF · 193 nm “350” “180” “90” “65” 2007 – 2019 · 40 nm, then 38 nm from 2008 193 nm light + water – the single-exposure limit “45” “32” “22” “14” “10” “7” Node names, plotted at face value The single-exposure limit held near 40 nm for 12 years; repeat passes printed finer – the names ran ahead of both. After 2019, every gain is one company. the wavelength cliff (2019) · the aperture steps (2027 → ) 2019 · the wavelength moves EUV: 193 → 13.5 nm 2027 · the lens widens NA 0.33 → 0.55
† printable ≠ shipped – foundries lag the tool by design and packaging cycles; node names lag further. “45” shipped against dry ArF’s ~58 nm limit; immersion’s applied from “32”.Plain Sight Research

The equation

Resolution comes from an equation borrowed from Lord Rayleigh's 19th century work on telescopes: critical dimension = k1 · λ / NA. Three terms, three ways to print smaller. λ is the wavelength of the light. NA is numerical aperture, how wide a cone of that light the optics can gather and focus. k1 is the process factor – everything the fab does around the exposure, from mask tricks and illumination angles to resist chemistry and computational correction. It starts near 1.0 and has a hard physical floor of 0.25.

Every improvement in lithography, ever, has come from moving one of those three.

Reading the chart

The navy line is the machine's limit: the smallest feature one exposure could resolve, plotted at each generation of light. It falls steadily from about 1,250 nm under g-line in 1980 to 40 nm in 2007, when all three terms ran out at once – k1 a shade under 0.28, argon-fluoride light stuck at 193 nm since the turn of the century, and the last available gain taken by flooding the gap between lens and wafer with water to reach NA 1.35. One further step to 38 nm in 2008, and then nothing for eleven years.

The rust-coloured labels are node names, plotted at face value. They keep falling through the plateau because they had stopped describing the critical dimension: density was being bought with repeat exposures and redesigned cells rather than finer printing. Between Intel's 45-nm node and TSMC's N7 density rose roughly 27-fold while the line above barely moved. The names have since moved on to “ångströms” against features an order of magnitude larger.

The gold segment is 2019, when ASML moved the wavelength from 193 nm to 13.5 nm – a fourteen-fold cut, where no previous change of light had delivered even two-fold. After it, the line is one company's roadmap: a single step when the aperture widens from 0.33 to 0.55, and flat thereafter.

What the chart cannot tell you

Printable is not shipped. A tool's specified resolution arrives years before silicon using it reaches customers, and node names lag further again. The dagger note under the chart carries the two seams that matter: “45” shipped against dry argon-fluoride's ~58 nm limit, and immersion applies only from “32” onward.

Nor is the critical dimension the whole of density. Since the plateau began, most of the gain has come from standard-cell design, stacking and patterning technique – which is why a chart of printing alone understates how much progress the industry made while its central number stood still.

Used in The EUV Gap.
Revised August 2026