The Insane Engineering Behind EUV Lithography: How a Laser, Flying Tin Droplets, and the Worlds Most Perfect Mirrors Rule the Semiconductor Industry

A deep engineering exploration of Extreme Ultraviolet (EUV) lithography — the technology that prints billions of transistors onto silicon chips. Covers the…

·

The Insane Engineering Behind EUV Lithography: How a Laser, Flying Tin Droplets, and the World's Most Perfect Mirrors Rule the Semiconductor Industry

There is a machine in a factory in Veldhoven, Netherlands, that fires a 25-kilowatt carbon dioxide laser at a stream of molten tin droplets traveling at 70 meters per second — 50,000 times per second. Each droplet, struck by the laser, explodes into a plasma that glows at 200,000°C, emitting light at a wavelength that doesn't exist naturally on Earth. That light bounces off mirrors so flat that if you scaled one to the size of Germany, its tallest mountain would be 1 millimeter high. The reflected light etches patterns smaller than 20 silicon atoms wide into a wafer of pure crystalline silicon.

This machine costs $380 million. Only one company on Earth can make it. And without it, your smartphone, your laptop, every AI data center on the planet, and the entire global economy as we know it would stop functioning within approximately three years.

This is Extreme Ultraviolet Lithography — the most complex manufacturing tool humanity has ever built.

I've written about SpaceX landing rockets with convex optimization and hypersonic retropropulsion. I've covered 3D printing, sheet metal design, and microcontrollers. But EUV lithography exists on an entirely different plane of engineering complexity. It combines plasma physics, high-power laser optics, ultra-high vacuum technology, atomic-scale precision manufacturing, computational lithography, photochemistry, and mechatronics — all operating simultaneously at the bleeding edge of what's physically possible.

This guide will take you through every major subsystem of an EUV lithography machine, explain the physics that makes each one nearly impossible, and show why ASML's monopoly on this technology makes it the most geopolitically significant company in the world.

1. Why We Need EUV: The End of Small

To understand why EUV is necessary, we need to understand the fundamental limit of optical lithography: resolution.

The smallest feature you can print with light is given by the Rayleigh criterion:

Resolution (R) = k₁ × λ / NA

Where:

To print smaller features, you have three options:

  1. Reduce the wavelength (λ) — go from visible light (435 nm) → UV (365 nm) → Deep UV (248 nm → 193 nm) → Extreme UV (13.5 nm)
  2. Increase the numerical aperture (NA) — make the optics collect light from wider angles
  3. Reduce k₁ — use computational tricks like optical proximity correction, phase-shift masks, and multiple patterning

The semiconductor industry exhausted option 3 first. Between 2005 and 2018, chipmakers used Deep UV (DUV) at 193 nm with increasingly elaborate tricks:

These techniques pushed 193 nm DUV from ~65 nm features down to ~7 nm features — an extraordinary achievement. But each additional patterning step adds cost, cycle time, and defect risk. By 2018, the most advanced chips required 40+ lithography steps for a single layer that could be done in one EUV exposure.

The only sustainable path forward was to reduce the wavelength by a factor of 14 — from 193 nm to 13.5 nm. And that required solving a series of physics problems that each, individually, would be considered the engineering achievement of a decade.

2. The Light Source: Firing Lasers at Flying Tin

The Wavelength Problem

EUV light at 13.5 nanometers is extreme in the literal sense: it has a photon energy of approximately 92 electron volts (eV) . For comparison:

Light Source · Wavelength · Photon Energy

Visible (green) · 532 nm · 2.3 eV

DUV (ArF excimer) · 193 nm · 6.4 eV

EUV · 13.5 nm · 92 eV

Silicon bond energy · — · ~4 eV

Carbon bond energy · — · ~3.6 eV

At 92 eV, every EUV photon breaks chemical bonds — which is exactly what you want for the photoresist, but it also means the light is absorbed by everything: air, glass, water, most metals, and all organic materials. EUV light travels less than a millimeter through air before being completely absorbed. This single fact drives the entire machine architecture.

Creating 13.5 nm Light: The Tin Plasma Source

How do you generate light at 13.5 nm with sufficient power to expose wafers at commercial throughput? The answer, arrived at after decades of research, is tin plasma.

The physics: when tin atoms are heated to approximately 200,000°C (roughly 30 times hotter than the surface of the Sun), they become fully ionized — stripped of multiple electrons. The resulting plasma contains tin ions in charge states Sn IX through Sn XIV (tin atoms missing 8 to 13 electrons). When these highly charged ions capture electrons and transition between specific energy levels — specifically 4p⁶4dⁿ → 4p⁵4dⁿ⁺¹ + 4dⁿ⁻¹4f — they emit photons at precisely 13.5 nm. This isn't one sharp spectral line but a broad "unresolved transition array" (UTA) spanning roughly 13.3–13.7 nm, centered at 13.5 nm.

The practical implementation is the most metal thing I've ever described:

  1. Melt tin (melting point: 232°C) in a heated reservoir inside the machine
  2. Force it through a 30-micron nozzle at high pressure, producing a continuous stream of droplets
  3. Modulate the stream with a piezoelectric actuator to break it into precisely spaced droplets — 50,000 droplets per second, each ~30 microns in diameter (roughly half the diameter of a human hair)
  4. Fire droplets through a vacuum chamber at 70 m/s (250 km/h — about the speed of a racing drone)
  5. Hit each droplet with two laser pulses:

- Prepulse (~1–2 kW, wider beam): Hits the droplet, flattens it into a pancake-shaped target, creating a diffuse cloud of tin vapor - Main pulse (~25–30 kW, focused): Fires into the pancake ~1–2 microseconds later, heating the expanded tin to plasma temperature

  1. Collect the 13.5 nm photons emitted by the plasma
  2. Capture the spent tin debris with a collector mirror and hydrogen gas flow
  3. Repeat 50,000 times per second

The CO₂ laser used for this is not your average laser cutter. It's a master oscillator power amplifier (MOPA) configuration producing pulses at 50 kHz with pulse energies exceeding 500 millijoules. The beam is focused to an intensity of roughly 10¹¹ W/cm² at the tin target — sufficient to create the plasma in ~10–20 nanoseconds.

The engineering challenges here are staggering:

The Illuminator: Collecting and Shaping the Light

The tin plasma emits EUV light in all directions. The first optical element — the collector mirror — is a large ellipsoidal multilayer mirror that surrounds the plasma and gathers roughly 5–6 steradians of emitted light. It focuses the collected light into the illuminator, a series of mirrors that:

The illuminator is entirely reflective — there are no lenses in the entire EUV optical path.

3. The Mirrors: Atomic-Scale Perfection

Why Mirrors Instead of Lenses

EUV light is absorbed by every material. You cannot make a lens for 13.5 nm light — it would be opaque. Instead, EUV optics use multilayer Bragg reflectors: alternating thin-film layers of two materials with different refractive indices, where each interface reflects a tiny fraction of the light, and all the reflections add up coherently.

The standard EUV mirror coating is molybdenum/silicon (Mo/Si) — alternating layers of molybdenum and silicon, each precisely 6.9 nm thick (one quarter of the EUV wavelength in that material). A typical EUV mirror has 40–60 bilayer pairs, totaling roughly 280–400 nm of coating thickness.

The physics: each Mo/Si interface reflects approximately 0.7% of the incident EUV light. With 40 interfaces, the constructive interference of all reflections produces a theoretical peak reflectivity of ~70% at exactly 13.5 nm — and this is per mirror.

There are at least 11 mirrors in the optical path of a modern EUV scanner:

Total light transmission through the system: (0.70)¹¹ ≈ 2% . Of the EUV light generated by the plasma, only 2% reaches the wafer. This is why the light source must produce so much power — to get 250 W at the wafer, you need ~12,500 W of generated EUV at the plasma.

The Flattest Surfaces in the Universe

The mirrors are manufactured by Carl Zeiss SMT in Oberkochen, Germany — and they represent the pinnacle of optical fabrication.

Surface figure error: The deviation of the mirror surface from its ideal mathematical shape. For EUV projection optics, the allowable figure error is approximately 50 picometers RMS (0.05 nm). That's roughly one-quarter the diameter of a silicon atom.

To put this in perspective: if you scaled an EUV mirror to the size of Germany (roughly 1,000 km across), the 50 picometer tolerance at full scale would be equivalent to a height variation of 1 millimeter — the tallest "mountain" on this country-sized mirror would be the thickness of a credit card.

Achieving this requires:

  1. Substrate material: Ultra-low expansion (ULE) glass or Zerodur — glass-ceramics with a coefficient of thermal expansion near zero (CTE < 10 ppb/K). Any thermal expansion would distort the mirror shape by far more than 50 pm.
  1. Ion beam figuring (IBF) : After initial polishing, the mirror is measured by interferometry at the operating wavelength. A focused beam of argon or krypton ions is then rastered across the surface, sputtering away individual atoms where the surface is too high. This process removes material atom-by-atom in a controlled fashion, correcting errors to the picometer level. Each mirror can take months to figure.
  1. Multilayer deposition: The Mo/Si multilayers are deposited by ion beam sputtering or magnetron sputtering in ultra-high vacuum. Each layer must be uniform to within ~0.1 nm across the entire mirror diameter (typically 200–600 mm). Any thickness variation shifts the reflected wavelength locally, introducing aberrations.
  1. Metrology: How do you measure a surface to 50 picometers? The answer is interferometry at the operating wavelength — using a synchrotron light source or a dedicated EUV interferometer that compares the wavefront produced by the mirror to an ideal reference. This is a measurement technology that exists essentially only at Zeiss and a handful of national laboratories.

The projection optics assembly — the set of mirrors that images the mask pattern onto the wafer — is a single aligned unit weighing several hundred kilograms, with each mirror positioned to sub-nanometer precision and maintained at a stable temperature within millikelvins. If you dropped a grain of sand on one of these mirrors, you'd crater a surface that took months to perfect.

4. How EUV Exposure Actually Works

The Mask (Reticle)

The photomask — the "negative" that carries the circuit pattern — is itself an EUV mirror. It's a low-thermal-expansion substrate coated with 40–60 Mo/Si bilayers, then patterned with an absorber layer (typically tantalum-based) where the circuit features should block light.

The mask is used in reflection: EUV light from the illuminator hits the mask at an angle (typically 6° off-normal), reflects off the multilayer in the open areas, and is absorbed by the tantalum in the dark areas. The reflected light then enters the projection optics, which demagnify the pattern by 4× (a feature on the mask is 4× larger than the corresponding feature on the wafer).

Mask defects are a multi-billion-dollar problem. Any particle landing on the mask — a single dust grain, a stray tin atom from the plasma source, a flake of resist — will print onto every wafer exposed with that mask. This is why EUV masks use pellicles: ultra-thin (50–100 nm) membranes suspended a few millimeters above the mask surface, acting as a particle shield. Any particle that lands on the pellicle is out of focus (the depth of focus in EUV is only ~50–100 nm) and won't print. But the pellicle itself absorbs ~10–20% of the EUV light and must survive 250 W of EUV illumination without distorting or breaking — a materials engineering challenge in its own right.

Photoresist Chemistry

EUV photoresist doesn't work like conventional photoresist. At 92 eV, the photon energy is so high that the primary interaction isn't direct chemical bond breaking in the resist molecule. Instead, the process works through secondary electrons:

  1. An EUV photon is absorbed by an atom in the resist (or the underlying substrate), ejecting a photoelectron with ~80–90 eV of kinetic energy
  2. This high-energy electron travels through the material, colliding with other atoms and generating secondary electrons (typically 5–20 eV each)
  3. Each EUV photon produces roughly 4–10 secondary electrons
  4. These secondary electrons trigger the chemical reactions that change the resist solubility (making it either soluble or insoluble in developer, depending on whether it's a positive or negative tone resist)

The problem: secondary electrons are stochastic. An EUV photon that lands in one spot might produce 4 electrons; another photon in an identical spot might produce 10. The electrons travel random distances (1–6 nm on average) in random directions. This stochastic noise is the fundamental resolution limit of EUV lithography — and it becomes dominant at the smallest feature sizes.

At 3 nm and 2 nm process nodes, the number of EUV photons used to define a single transistor feature is small enough (tens to low hundreds) that statistical shot noise — the random variation in photon arrival times and locations — creates visible patterning errors:

Managing stochastic effects is now the central challenge of EUV lithography, driving research into higher-sensitivity resists, post-exposure bake optimization, and computational correction techniques.

5. High-NA EUV: The Next Frontier

The first generation of production EUV tools (ASML NXE:3400 and NXE:3600) used a 0.33 numerical aperture (NA) projection system. Combined with 13.5 nm wavelength and a k₁ factor of ~0.4, this gives a resolution of:

R = 0.4 × 13.5 / 0.33 ≈ 16 nm half-pitch

To print features smaller than this (which is necessary for 3 nm, 2 nm, and beyond), you need either multiple patterning (expensive) or higher NA.

The next generation — High-NA EUV (ASML EXE:5000) — increases the numerical aperture to 0.55. This reduces the printable half-pitch to approximately 8–10 nm, enabling single-exposure patterning at the most advanced nodes.

But increasing NA from 0.33 to 0.55 comes with enormous consequences:

Anamorphic Optics

In a conventional lithography system, the demagnification is the same in both the X and Y directions (usually 4×). But at 0.55 NA, the projection optics would need to accept light at such extreme angles that the mask would need to be impractically large — the beam footprint on the mask would exceed commercially viable mask blank sizes.

The solution: anamorphic optics. The projection system demagnifies by 4× in the Y-direction (scan direction) but only 8× in the X-direction (perpendicular to scan). This:

Tighter Depth of Focus

Depth of focus scales inversely with NA²:

DOF ∝ λ / NA²

Going from 0.33 NA to 0.55 NA reduces the depth of focus by a factor of (0.33/0.55)² ≈ 0.36 — meaning the wafer must be positioned to within roughly 50 nanometers vertically across the entire exposure field. This requires wafer stages with sub-nanometer positioning accuracy and real-time focus control that measures the wafer surface topography and adjusts the lens-to-wafer distance continuously during the scan.

Larger, Heavier, More Expensive

A High-NA EUV machine weighs approximately 150 metric tons (about the weight of a Boeing 747 without engines), stands over 3 meters tall, and requires 5–6 months to install in a specially reinforced cleanroom. Each machine costs approximately 380–400 million. A single leading-edge logic fab might install 10–20 of these machines — a capital investment of 4–8 billion just in lithography tools.

6. The Contamination Problem: Fighting Physics at Every Turn

EUV lithography is an endless battle against contamination, because the very physics that makes it work also creates conditions that degrade the machine:

Tin Redeposition

The tin plasma source continuously produces tin vapor and nanoparticles. Despite hydrogen gas scavenging, some tin inevitably deposits on the collector mirror and illuminator optics, gradually reducing their reflectivity. The collector mirror must be replaced every 6–12 months, at a cost of hundreds of thousands of dollars — and a machine with a degraded collector produces fewer wafers per hour, directly impacting fab economics.

Carbon Growth

Residual hydrocarbon molecules in the vacuum chamber (from outgassing materials, vacuum pump oil backstreaming, or resist outgassing) are cracked by EUV photons into reactive carbon species that deposit as amorphous carbon films on the mirror surfaces. Even a few nanometers of carbon significantly reduces EUV reflectivity. Countermeasures include:

Resist Outgassing

When EUV light hits the photoresist on the wafer, it ejects not just electrons but also volatile organic fragments from the resist chemistry. These fragments can:

The solution is a dynamic gas lock — a differentially pumped region between the last projection mirror and the wafer, with a continuous flow of clean gas (typically hydrogen or helium at low pressure) that sweeps contaminants away before they reach the optics.

7. The Business and Geopolitics of EUV

The ASML Monopoly

One company — ASML, headquartered in Veldhoven, Netherlands — manufactures 100% of the world's EUV lithography machines. This isn't because no one else tried. Canon and Nikon, the giants of DUV lithography, attempted EUV development and abandoned it. Intel, Samsung, and TSMC all contributed billions in R&D funding to ASML to ensure the technology would exist for their fabs.

ASML doesn't actually build most of the components themselves. Their supply chain is a carefully orchestrated network of monopolies-within-a-monopoly:

Component · Sole Supplier · Country

Optics (mirrors, projection system) · Carl Zeiss SMT · Germany

EUV light source (CO₂ laser + tin plasma) · Trumpf + ASML (Cymer) · Germany / USA

Wafer stages (nanopositioning) · ASML (in-house) + VDL ETG · Netherlands

Vacuum systems · Edwards / Pfeiffer · UK / Germany

Photoresist · JSR, TOK, Shin-Etsu, DuPont · Japan / USA

Mask blanks · Hoya, AGC · Japan

Pellicles · Mitsui Chemicals, Asahi Kasei · Japan

The geopolitical leverage this creates is extraordinary. The United States, through export controls, can prevent ASML from selling EUV machines to China. Since China's semiconductor industry has no domestic EUV capability (as of 2026), this effectively caps Chinese chip manufacturing at roughly the 7 nm node (achievable with DUV multi-patterning) — approximately two generations behind the cutting edge. The chip sanctions are enforced not at the chip level, but at the tool level. Control the lithography, control the industry.

The Economics of EUV

The economics of EUV are brutal but unavoidable:

Item · Approximate Cost (2026)

EUV lithography machine (0.33 NA) · $180–200 million

High-NA EUV machine (0.55 NA) · $380–400 million

Annual service contract · $20–30 million per machine

EUV mask set (single chip design) · $10–30 million

Photoresist (per liter) · $5,000–10,000

Collector mirror replacement · $200,000–500,000

Electricity (per machine, annually) · $2–3 million

A leading-edge logic fab with 15 EUV machines and 5 High-NA machines represents roughly 6–8 billion in lithography capital alone. This is why only three companies — TSMC, Samsung, and Intel — operate at the absolute leading edge. The barrier to entry is not technical skill or even manufacturing capability; it's the ability to write a check for 8 billion and wait 3–4 years for the machines to arrive.

Future Generations

Beyond High-NA (0.55 NA), the roadmap gets speculative:

8. The Machine That Saved Moore's Law

In 1965, Gordon Moore observed that the number of transistors on a chip was doubling roughly every two years. For five decades, the semiconductor industry met or exceeded this pace through a combination of wavelength reduction, numerical aperture improvement, and process control advances.

By the early 2010s, Moore's Law was dying. 193 nm immersion DUV had been extended as far as physically possible through multiple patterning tricks. Further shrinks required a wavelength jump that had been under research since the 1980s — and which most experts believed would never be commercially viable.

ASML's EUV program burned through an estimated $10–15 billion in R&D before shipping the first production-worthy machine. At multiple points, the project was nearly cancelled. The light source wasn't powerful enough. The mirrors weren't flat enough. The photoresist wasn't sensitive enough. The pellicles kept breaking. The debris management didn't work. The uptime was too low to be economical.

Each of these problems was solved — not through a single breakthrough, but through thousands of incremental improvements across plasma physics, materials science, precision engineering, computational lithography, and vacuum technology. The machine that emerged is the most complex piece of equipment ever manufactured, and it quite literally saved the trajectory of semiconductor progress.

The next time you hold a smartphone with a 3-nanometer processor containing 20 billion transistors, remember: each one of those transistors was printed by light from exploding tin droplets, reflected off mirrors flatter than continents, in a machine that costs more than a 747 and is harder to build than a spacecraft.

References: ASML public technical documentation; "EUV Lithography" (SPIE Press, 2018, ed. Vivek Bakshi); Chris Mack, "Fundamental Principles of Optical Lithography" (Wiley, 2007); Zeiss SMT technical presentations; MIT Technology Review, "Inside the Machine That Saved Moore's Law" (2021); SPIE Advanced Lithography conference proceedings (2018–2025).

Last updated: July 2026

More FabFlow blog posts