Before EUV Light Can Print a Transistor, It Has to Bounce Off a Mirror That Can't Have a Single Flaw. Two Companies Make Nearly All of Them.
It's a six-inch square of glass coated with roughly forty paper-thin layers of metal, and before ASML's record-priced lithography machines can print a single transistor, every advanced chip in the world has to start here. A speck of dust caught under those layers doesn't go away — it rides straight to the top and gets printed onto every wafer that blank ever makes.
Grow the mirror, then see if the flaw survives
A simplified cross-section of an EUV mask blank's multilayer mirror, viewed edge-on. Deposit all 40 layers over a single contaminant particle and watch whether its bump is smoothed away or still pokes through the top — which is what decides whether that blank gets used or scrapped. Layer thicknesses, bump sizes, and the smoothing rate are exaggerated and simplified for legibility, not a precise model of real deposition physics.
The plain version
Every advanced chip in the world — the ones inside your phone, your car, and the servers training AI — starts, improbably, with a mirror. Not a bathroom mirror; a mirror built from roughly forty pairs of alternating molybdenum and silicon layers, each only a few atoms thick, stacked onto a six-inch square of ultra-pure glass. That stack is called an EUV mask blank, and it's the raw material that chipmakers' own mask shops etch a circuit pattern onto before it ever goes near one of ASML's famous extreme-ultraviolet lithography machines.
Why a mirror and not a lens? The light used to print today's smallest transistors — extreme ultraviolet, or EUV — is so energetic that it gets absorbed by almost everything, including ordinary glass and even air. Nothing can focus it the way a camera lens focuses visible light. So engineers reflect it instead, bouncing it off that precisely tuned multilayer stack the way you'd bounce a flashlight beam off a mirror.
Here's the catch: if even one microscopic speck of dust lands on the glass before the first layer goes down, every layer deposited on top of it conforms around that bump — and the bump doesn't fully disappear. By the fortieth layer, it can still be poking up just enough to get printed onto every single chip made with that blank, forever. There's no fixing it after the fact; the blank just gets thrown away.
Making that defect-free glass at scale turns out to be absurdly hard — hard enough that only two companies, Japan's AGC and Hoya, reliably do it at volume, together supplying an estimated 90%-plus of the world's qualified EUV mask blanks. A South Korean challenger, S&S Tech, is trying to break in, but qualifying can take over a year. Before anyone worries about chip tariffs or GPU shortages, this is where the chain actually starts.
The expert version
EUV lithography uses 13.5 nm light, a wavelength absorbed by virtually all materials, so refractive optics are not viable — the entire optical path, from source collector through projection optics to the mask itself, is reflective. The mask (reticle) is built on a blank: a low-thermal-expansion-glass substrate roughly 6 inches square, coated with on the order of 40–50 alternating bilayers of molybdenum (~2.1 nm) and silicon (~4.1 nm), engineered so each interface reflection constructively interferes near-normal incidence — a Bragg reflector tuned to 13.5 nm that reaches roughly 65–70% peak reflectivity. A capping layer and an absorber pattern sit on top; the absorber is what actually gets patterned to form the circuit.
The defect problem is structural, not incidental. A particle or pit at the substrate, or trapped mid-stack during sputter deposition, propagates upward through every subsequent bilayer as a "phase defect" — a local disruption of the reflected wavefront rather than a simple blocked spot, which is harder to compensate for in mask design than an absorber-layer defect. Because the multilayer is buried under dozens of subsequent depositions, it can't be inspected or repaired mid-process; a blank found defective after full deposition is scrapped outright. Historical SEMATECH-era programs targeted zero "killer" defects down to tens of nanometers in diameter for leading nodes — a yield bar that for years outpaced available substrate-polishing and deposition tooling.
That bar is also why blank supply has stayed concentrated. AGC and Hoya are, by their own disclosures, the two qualified volume suppliers to the industry; AGC describes itself as the only maker handling the full process in-house, from glass substrate through multilayer coating, and both companies have added capacity specifically for EUV blanks (Hoya's Singapore plant, AGC's dedicated blank lines) rather than the larger, more fragmented DUV mask-blank market. Independent trade-press estimates put the two combined at more than 90% of qualified EUV blank supply industry-wide. South Korea's S&S Tech, sampling blanks to Samsung since March 2026, is the most advanced attempt to add a third qualified source, with reported qualification cycles of six months to over a year.
Why it matters for tech + supply chain: Everyone talks about ASML's EUV machines as the chokepoint under every advanced chip. There's a quieter one underneath that one — two companies' glass.
Why it matters for tech + supply chain: Mask-blank qualification, not scanner output, is increasingly the real ceiling on EUV-node capacity — and it's a bar only two companies have cleared at volume.