Shrink the Line
how chipmakers learned to print lines 8 nm wide.
a story in five chapters: a lens prints fine lines only if it catches the beams they split light into, so fabs kept printing finer by turning the knobs of CD = k1 lambda / NA: bluer light through the eras, water under the lens, then 13.5 nm light from blasted tin, steered by mirrors in a vacuum
A mask of fine lines splits light into beams, the diffraction orders, at angles where sin(theta) = m lambda / p. A lens draws the lines only if it catches at least two of them. With light coming straight down, that is the middle beam and both first beams, so when the lines get too fine the side beams miss its opening and nothing prints: the finest pitch is lambda / NA, where NA, the numerical aperture, is n sin(theta) of the steepest beam the lens takes. That is Ernst Abbe's rule; his resolution formula is engraved on his memorial in Jena. So chipmakers turned the knobs. Bluer light: the mercury lamp's 436 nm line in the 1980s, then its 365 nm line, then lasers at 248 nm in the 1990s and 193 nm in the 2000s (Cadence, History of Lithography; SPIE, Lasers and Moore's Law). A brighter lamp moves no beam, and a sharper mask cannot help on its own, since the lens still has to catch the beams; cleverer masks (phase-shifting, and the next card's corrected shapes) and tilted light belong to k1, the craft. A wider lens: in air NA cannot pass 1, because steeper beams cannot leave the glass (total internal reflection), and dry lenses stopped at about 0.93; water, n = 1.44 at 193 nm, fills the gap and NA reaches 1.35 (Burn Lin, TSMC, C. R. Physique, 2006; ASML NXT:2000i). Then EUV at 13.5 nm: tin drops about 25 um wide (ASML; IEEE Spectrum says 30) are hit 50,000 times a second; a weak laser pulse flattens each into a pancake and a strong one turns it into a plasma that glows at 13.5 nm, hundreds of thousands of kelvin hot (about 200,000 C by IEEE Spectrum, about 500,000 K by ASML), far hotter than the Sun's surface (about 5,800 K). Air and glass swallow that light, so the machine is a vacuum full of mirrors. Light tilted to the lens's edge (a dipole) lets one side beam be enough, which nearly halves the finest line: ArF immersion prints lines 38 nm wide in one exposure (ASML NXT:2000i, k1 about 0.27); High-NA EUV, NA 0.55 (ASML EXE:5000), 8 nm. CD = k1 lambda / NA, where CD, the critical dimension, is the width of the finest line (lettered on the card as width): the light, the lens, and k1, the craft. Worked back from those machines, k1 is 0.266 for ArF immersion (38 x 1.35 / 193) and 0.326 for High-NA EUV (8 x 0.55 / 13.5). Every print here is computed: scalar Abbe imaging of a 1D grating of equal lines and spaces (each order's amplitude sin(m pi/2)/(m pi), passed by the lens's opening, interfering into the image), the resist a threshold at the image's middle, printed when the contrast reaches 0.3; each photon lands where the computed light is bright. The first pages' lens is NA 0.45 at 436 nm with light straight down; the check's two machines are on a dipole. Toy parts: one photon stands for very many; the lens's opening is drawn as a slot; the machine is drawn at 1x (real ones shrink the mask 4x; High-NA 4x one way and 8x the other); the angles on the water page are the real ones but the gap is not to scale; the laser is infrared, drawn red; the mirror's aim is a toy, and nothing is to scale. Only 436 nm is visible: 365, 248 and 193 nm are ultraviolet and 13.5 nm extreme ultraviolet, so violet, cyan and gold are false colours.
Concepts: diffraction, optics, interference, light, scaling.
An idea you find out by playing, on Wundr, made by @neelu.
On the shelf senses, and also on matter.
More on senses
- Adapt: stare at the dot. then trust nothing.
- Apparent: a footstool at your feet. a monolith at the door. the same cube.
- Bend: tilt the beam till the glass traps it whole.
- Colorsmith: two paint chips, nearly the same. which is lighter?
- Flicker: nothing on this screen has ever moved.
- Foil: the rainbow is in the angle, not the ink.