Why China Can't Copy ASML — shantum's notes

ELI5/TLDR

ASML makes the one machine on Earth that prints the most advanced computer chips. It costs up to $380 million, weighs as much as two airliners, and shoots microscopic tin droplets with a laser 50,000 times a second to create a tiny sun that spits out a very specific kind of light. China cannot copy it — not because they lack money or brains, but because the machine is the endpoint of 40 years of global collaboration across German optics, American lasers, and Dutch precision engineering, and most of the real know-how lives inside the heads of veteran technicians, not in any blueprint.

The Full Story

The wall the industry hit

For decades, chipmakers printed circuits using deep ultraviolet light — a 193 nanometer wavelength. Think of light as a brush, and the wavelength as how thick the bristles are. A 193 nm brush could paint details down to about 10 nm by using clever tricks. The best trick was immersion lithography: shine the laser through a thin layer of ultra-pure water between the lens and the wafer, and the water bends the light to sharpen the line.

But once the industry needed to draw 7 nm features, the 193 nm brush was hopelessly clumsy. The workaround — “multi-patterning” — meant running each wafer through the machine three or four times, etching a little, baking it, washing it, coating it again, lining it up to the previous pass with sub-nanometer precision. Every extra pass multiplies the chance of a ruinous defect. The industry had reached a breaking point. It needed a scalpel, not a sledgehammer.

The scalpel: a miniature supernova, 50,000 times a second

The scalpel is 13.5 nm extreme ultraviolet light (EUV). This wavelength does not occur naturally on Earth and no normal lamp or laser can produce it. ASML’s solution is, as the video puts it, closer to science fiction than manufacturing.

Imagine a vacuum chamber. A machine fires microscopic droplets of molten tin straight down at 250 km/h. As each droplet falls, sensors track it. A first laser pulse — a weak one — hits the droplet and flattens it from a sphere into a pancake, which is easier to vaporize. A microsecond later, the main laser pulse obliterates the pancake, creating a plasma hotter than 220,000°C that briefly glows with 13.5 nm light.

Controlling light beams and physical matter with such localized spatial and temporal accuracy is akin to shining a laser pointer from the Earth and successfully hitting a specific coin resting on the surface of the moon.

This happens 50,000 times per second. And of the light produced, only 5.5% is the useful wavelength. Incinerating 50,000 tin drops a second also creates a constant debris storm, which ASML manages with a carefully shaped flow of hydrogen gas — a microscopic hurricane that sweeps stray tin atoms away from the optics without disturbing the light path.

The mirrors, because glass won’t do

EUV light is so fragile it cannot pass through normal glass lenses. It must be bounced off mirrors, inside a vacuum, the whole way. These mirrors are up to a meter across and polished so smooth that any bump on their surface is measured in picometers — trillionths of a meter. Each mirror is then coated with about 100 alternating thin layers of silicon and molybdenum. Each layer is a few atoms thick, and the stack has to reflect a specific wavelength via Bragg reflection — essentially, tuning the mirror like a musical instrument so that the light waves bouncing off each layer reinforce each other.

Even then, each mirror only reflects 70% of the EUV light. The machine uses 11 of them in sequence. Less than 2% of the light ASML generates actually reaches the wafer. The laser needs to pump in tens of thousands of watts to deliver a tiny fraction of a watt where it matters.

It’s not one company, it’s a global orchestra

Even if you perfectly mapped every one of the 100,000 parts inside an EUV scanner, you couldn’t build one. ASML is a conductor. The orchestra has about 1,200 hyper-specialized suppliers:

The carbon dioxide laser that vaporizes the tin — built by Trumpf in Germany, a piece of optical engineering that took them decades to perfect.
The droplet generator and the plasma physics — pioneered by Cymer in San Diego (now owned by ASML).
The picometer-precise mirrors — made exclusively by Carl Zeiss SMT in Germany, in a partnership so tight that Zeiss literally cannot sell these mirrors to anyone else.
The wafer handler — built with VDL in the Netherlands, using magnetic levitation stages that accelerate at over 3G and stop with sub-nanometer accuracy.

China, to clone the machine, would have to simultaneously rebuild the German optics industry, the American laser industry, and the Dutch mechatronics ecosystem — while being legally cut off from all three.

The deepest moat: tacit knowledge

Here is the part that cannot be stolen. Knowledge splits into two kinds. Explicit knowledge is what can be written down — dimensions, chemical compositions, wiring diagrams. Tacit knowledge is the unwritten intuition technicians build over decades. It is the feel for how a Zeiss polisher knows when a mirror is almost ready, or how an ASML engineer coaxes the twin laser pulses into landing on a falling tin drop.

You could exfiltrate every blueprint and still not build the machine, because the machine is built by hands that have failed in specific ways for thirty years.

What this forces China to do

Blocked from EUV, and since 2023-2024 blocked from the advanced DUV machines too, China’s leading foundry SMIC is stuck doing 7 nm chips the hard way — multi-patterning on old DUV machines. The economics are brutal. Every extra pass adds a chance of a dust speck or a misaligned layer killing the chip. The mature industry expects yield rates north of 80%. SMIC’s on advanced nodes is reportedly much lower, which means SMIC charges 40-50% more per chip than TSMC does for the same class of silicon. And TSMC produces 100,000-120,000 advanced wafers a month. SMIC’s advanced output sits around 3,000-4,000.

For state-security projects that don’t care about cost, this is livable. For competing globally on phones and cloud compute, it’s a structural disadvantage.

The real moat

Limitless capital cannot purchase a time machine.

China is not trying to copy a machine. It is trying to compress 40 years of compounded global collaboration — failures, iterations, cross-border research, decades of technician hands — into one decade, alone, in the dark.

Key Takeaways

EUV lithography uses 13.5 nm light, generated by laser-vaporizing tin droplets 50,000 times a second at 220,000°C.
Only 2% of the generated EUV light reaches the wafer after bouncing off 11 mirrors, each with picometer-level surface accuracy.
ASML is not one company — it coordinates ~1,200 specialized suppliers (Trumpf lasers, Zeiss mirrors, Cymer plasma sources, VDL wafer handlers).
Zeiss and ASML have an exclusive alliance: Zeiss cannot sell the mirrors to anyone else, ASML cannot buy them elsewhere.
The deepest moat is tacit knowledge — decades of calibration and troubleshooting intuition that lives in veteran technicians’ hands, not in any blueprint.
SMIC makes 7 nm chips on older DUV machines using multi-patterning — 3-4 passes per wafer, yields well below the 80%+ industry norm, prices 40-50% higher than TSMC.
Volume gap: TSMC ~100-120k advanced wafers/month vs SMIC ~3-4k/month.
US export controls expanded in 2023-2024 to block even advanced DUV machines from China.

Claude’s Take

This is a solid explainer. The physics is accurate, the numbers check out, and the core framing — tacit knowledge as the real moat — is the right one and not said enough. The “laser pointer hitting a coin on the moon” quote is a favorite among ASML executives and is genuinely helpful for grasping the precision involved.

What’s missing or soft: the video doesn’t mention that ASML already owns Cymer (has since 2013), which slightly weakens the “global orchestra” framing — ASML has been vertically integrating its most critical suppliers for years. It also glosses over China’s actual workarounds, which are more interesting than “they can’t do it.” SMIC did manage 7 nm chips for the Huawei Mate 60 Pro in 2023, which surprised a lot of observers. The direction of travel matters — the gap is huge, but it’s not stationary. SMEE, China’s domestic lithography company, has roadmaps for 28 nm DUV and is reportedly working on their own immersion system. None of that catches up to EUV, but “can’t copy” and “can’t ever approach” are different claims, and the video blurs them.

The yield-rate and wafer-volume comparisons feel directionally right but are based on leaked and reported figures. Take the exact numbers with a grain of salt — SMIC and the Chinese state are not exactly forthcoming.

Score: 8. Gets the hard parts right, slightly overclaims the “impossibility” angle, worth watching for anyone who wants to understand why this single Dutch company sits at the center of the 21st century’s most important supply chain.