The Extreme Ultraviolet Mirror

How Global Computation Depends on Molten Tin Droplets

Every microprocessor, graphics card, and smartphone chip on Earth is manufactured by projecting a microscopic shadow of a circuit blueprint onto a silicon wafer coated with light-sensitive chemical resist. For half a century, the steady doubling of transistor density predicted by Gordon Moore in 1965 depended on shrinking the minimum printable wire width—the critical dimension $CD$—governed by Lord Rayleigh's optical diffraction equation:

Here $\lambda$ is the exposure light wavelength, $NA = n \sin\theta$ is the numerical aperture of the projection lens system (measuring the widest angle $\theta$ of light rays the lens can gather through a medium of refractive index $n$), and $k_1$ is the physical process factor, bounded for a single optical exposure by the diffraction barrier $k_1 \ge 0.25$.

To print smaller transistors, engineers must widen the numerical aperture $NA$ or shorten the wavelength $\lambda$. From the 1980s through 2001, photolithography machines stepped down the electromagnetic spectrum from mercury-lamp ultraviolet lines (365 nm) to krypton fluoride lasers (248 nm) and argon fluoride (ArF) deep-ultraviolet (DUV) lasers at $\lambda = 193\text{ nanometers}$. At 193 nanometers, the industry hit a wall. Engineers stretched 193 nm DUV light for nearly twenty years by injecting a film of ultrapure water ($n = 1.44$) between the final quartz lens and the moving silicon wafer—boosting the numerical aperture to $NA = 1.35$—and splitting single circuit layers across four overlapping exposures. By the seven-nanometer generation in 2018, printing a single wafer required over eighty mask steps, and nanometer misalignments between consecutive passes threatened to halt silicon scaling.

The only way forward was to leap from $\lambda = 193\text{ nanometers}$ all the way down to Extreme Ultraviolet (EUV) light at $\lambda = 13.5\text{ nanometers}$—a fourteen-fold wavelength reduction on the boundary of soft X-rays. Yet that leap broke every rule of classical optics: 13.5 nm radiation (91.8 electron-volts per photon) is immediately absorbed by air, water, and every transparent glass lens in existence.

Inside the cleanrooms of Veldhoven in the Netherlands, the semiconductor equipment company ASML spent twenty-five years and nine billion euros engineering the Twinscan NXE EUV lithography machine—a 180,000-kilogram apparatus built from over 100,000 components that requires forty freight containers and three Boeing 747 cargo jets to ship. Because no laser can directly emit continuous high-power light at 13.5 nanometers, the machine manufactures its own light inside a sealed high-vacuum chamber using a Laser-Produced Plasma (LPP) source developed by Cymer in San Diego.

At the top of the vacuum source chamber, a pressurized nozzle heats ultrapure tin ($\text{Sn}$) above its $232^\circ\text{C}$ melting point and fires 50,000 microscopic droplets per second—each a liquid sphere 27 micrometers in diameter—falling at seventy meters per second. Beneath the chamber floor sits a 25-kilowatt pulsed carbon dioxide ($\text{CO}_2$) drive laser operating at an infrared wavelength of 10.6 micrometers. As each falling tin droplet reaches the focal zone, the $\text{CO}_2$ laser shoots it twice in rapid succession. First, a low-energy pre-pulse strikes the 27-micrometer sphere, using photon radiation pressure to flatten the droplet into a concave 200-micrometer pancake tilted toward the collector optics. Microseconds later, a high-energy main pulse slams into the flattened tin disk, stripping up to fourteen electrons off the tin atoms and heating the metal into a 400,000-kelvin plasma—seventy times hotter than the surface of the sun. As those ionized tin atoms ($\text{Sn}^{8+}$ to $\text{Sn}^{14+}$) recombine and drop energy levels, they radiate a burst of 13.5 nm EUV photons, repeating this dual-laser explosion fifty thousand times every second.

Because glass lenses would swallow the 13.5 nm beam immediately, the entire optical path operates in vacuum using curved mirrors rather than refractive lenses. Even polished metal absorbs most 13.5 nm light, so the German optical firm Carl Zeiss in Oberkochen manufactures Distributed Bragg Reflector mirrors coated with forty to fifty alternating atomic layers of pure molybdenum ($\text{Mo}$) and pure silicon ($\text{Si}$), each bilayer exactly 6.9 nanometers thick (half the 13.5 nm wavelength). When EUV light strikes the stack, each atomic boundary reflects a few percent of the wave, and the forty reflections interfere constructively in phase ($2 d \sin\theta = m\lambda$). Polishing the underlying substrate to an atomic smoothness below 0.1 nanometers—equivalent to leveling a mirror the width of Germany until no bump exceeds a tenth of a millimeter—allows each Zeiss mirror to reflect $70\%$ of the 13.5 nm light while absorbing the remaining $30\%$ as heat. After bouncing off the collector mirror, the reflective circuit mask, and ten consecutive Zeiss Bragg mirrors ($0.70^{11} \approx 0.02$), barely two percent of the plasma light reaches the silicon wafer—just enough to etch transistor gates thirteen atoms wide for every advanced artificial intelligence processor and smartphone on Earth.