TSMC is building the machines that build the

I lost a weekend to semiconductor manufacturing. Not building semiconductors, obviously. Reading about how they’re built. Falling down a rabbit hole that started with a casual question (“how small is 7 nanometers, actually?”) and ended with me sitting in the dark at 2am, staring at a diagram of an extreme ultraviolet lithography system, feeling something close to religious awe.

Let me try to explain.

The scale of what we’re talking about

A human hair is about 80,000 nanometers wide. TSMC is now manufacturing transistors at 7 nanometers. That means you could line up more than 10,000 of these transistors across a single strand of hair.

But here’s the part that broke my brain: these transistors aren’t just small. They’re precisely small. Every single one has to be exactly right, positioned within fractions of a nanometer of where it should be, on a silicon wafer that contains billions of them. Not thousands. Not millions. Billions. And if the process drifts by even a tiny amount, the whole chip is garbage.

Imagine drawing a detailed map of a city, every building and street and fire hydrant, on a grain of sand. Now imagine doing it perfectly, every time, thousands of times per day. That’s roughly what TSMC does for a living.

What is 7 nanometers?

I keep trying different ways to make this number real. Here are my best attempts.

A red blood cell is about 7,000 nanometers across. A single transistor at the 7nm node is a thousand times smaller than a red blood cell. DNA is about 2.5 nanometers wide, so we’re in the neighborhood of individual molecules. A silicon atom is about 0.2 nanometers, which means a 7nm feature is roughly 35 atoms across.

Thirty-five atoms.

If you messed up the placement of one layer by just a few atoms, the transistor wouldn’t work. The tolerances are measured in fractions of nanometers. This is engineering at a scale where quantum mechanical effects start mattering, where electrons tunnel through barriers they shouldn’t be able to cross, where the distinction between “solid material” and “empty space” gets philosophically blurry.

And TSMC does this. Billions of times. On every chip. On millions of chips per year. In Hsinchu, Taiwan, right now, as I type this.

How light becomes circuitry

The basic idea of chip manufacturing hasn’t changed since the 1960s. You take a silicon wafer. You coat it with a light-sensitive material called photoresist. You shine light through a pattern (the circuit design). Where the light hits, the resist changes. You wash away the changed parts and you’ve got a pattern etched into silicon.

Simple in concept. Absolutely mind-bending in execution.

The problem is physics. Specifically, diffraction. Light has a wavelength, and you can’t print features smaller than that wavelength. It’s like trying to paint a miniature with a house brush. For decades, chipmakers used deep ultraviolet (DUV) light at 193 nanometers. To print features smaller than 193nm, they developed tricks: immersion lithography (putting water between the lens and the wafer, because light has a shorter effective wavelength in water), multi-patterning (printing the pattern in multiple passes), optical proximity correction (deliberately distorting the pattern so diffraction distorts it back to what you wanted).

These tricks worked. Brilliantly. They pushed 193nm light far beyond what anyone thought was possible. 14nm. 10nm. 7nm. Features forty times smaller than the light printing them, pulled into existence through physics tricks and engineering stubbornness.

But the tricks are getting harder. Each node requires more passes, more corrections, more complexity. The cost per wafer goes up. The yields go down. The whole system is straining under its own cleverness.

Which brings us to EUV.

The light that changes everything

EUV stands for extreme ultraviolet. The wavelength is 13.5 nanometers. Compared to the 193nm light used in current lithography, that’s like going from a house brush to a single-hair paintbrush.

But here’s why I spent all weekend reading about this: you can’t make EUV light the normal way. You can’t just get a brighter lamp. Nothing emits 13.5nm light in useful quantities through normal means.

So ASML, the Dutch company that makes lithography machines, came up with a process that sounds like science fiction. They take tiny droplets of molten tin and fire them through a chamber at high speed. Then they hit each droplet with a laser. Not just any laser. A high-powered CO2 laser. The laser vaporizes the tin, creating a plasma that emits EUV light. This happens 50,000 times per second.

Fifty. Thousand. Times. Per. Second.

Each tin droplet is about 25 microns across (smaller than a grain of pollen). Each one gets hit by a laser with the precision of a few microns. At fifty kilohertz.

The light that’s generated then bounces off a series of mirrors (you can’t use lenses at this wavelength because EUV light is absorbed by basically everything, including air, so the whole system operates in a vacuum). The mirrors are polished to a smoothness of less than an atom’s width. I read that twice. Less than an atom’s width. If one of these mirrors were scaled up to the size of Germany, the tallest bump would be less than a millimeter.

And then this impossibly precise light, generated by laser-vaporized tin in a vacuum, bounced off atomically smooth mirrors, shines through a mask onto a silicon wafer to print features smaller than a virus.

I mean, come on.

The clean room

I should mention the environment this happens in.

A TSMC fab is a clean room classified at ISO Class 1. That means there are fewer than 10 particles larger than 0.1 microns per cubic meter of air. For comparison: the air you’re breathing right now, assuming you’re in a normal room, has about 35 million particles per cubic meter. A hospital operating room has about 3,500. A TSMC clean room has 10.

The workers wear full-body suits called bunny suits. Not because the chips are dangerous to humans. Because humans are dangerous to chips. A single flake of dead skin, a single fiber from your clothing, a single droplet of moisture from your breath, any of these could land on a wafer and destroy a million transistors.

The air in the fab is filtered and recirculated constantly. The temperature is controlled to within a fraction of a degree. The humidity is controlled. Vibration is controlled (the fabs are built on massive foundations designed to absorb any movement from nearby roads or construction). Even the chemical purity of the water used in processing is controlled to parts per trillion.

It’s an environment more controlled than any natural environment on Earth. More controlled, in some ways, than the International Space Station. And it exists so that light can be aimed at silicon with atomic precision.

The cathedral nobody visits

Here’s what gets me. I read all of this on IEEE Spectrum and AnandTech and TSMC’s own publications, and I kept thinking: how is this not front-page news? How is this not something people talk about at parties?

This is, without any hyperbole, the most complex manufacturing process humans have ever developed. Nothing else comes close. Not aircraft carriers. Not particle accelerators. Not even the space shuttle. The precision, the scale, the physics involved in making a modern chip is staggering.

And it happens in near-total obscurity. TSMC’s fabs run 24/7 in Hsinchu, Taiwan. They produce the chips that power your phone, your laptop, the servers running the internet. Every piece of digital technology you touch has, at some point, passed through a process like the one I just described.

But nobody thinks about it. The chip is a black box. A commodity. A thing that’s “in there somewhere.”

The geopolitics I can’t ignore

Something else has been nagging at me. TSMC is in Taiwan. The most advanced semiconductor manufacturing on the planet, the kind that goes into every high-end phone and computer and server, happens on an island with a complicated political relationship with mainland China.

I’m not a foreign policy expert. I’m barely a foreign policy amateur. But I’ve been reading enough to know that this concentration of manufacturing capability in one place, on one island, is something that keeps national security people up at night. If TSMC’s fabs were disrupted for any reason (natural disaster, conflict, supply chain break), the world would feel it within weeks. Not just tech companies. Everyone. Because chips go into everything: cars, medical devices, power grids, communication systems.

It’s strange to think that a factory in Hsinchu, Taiwan, that most people have never heard of, is one of the most strategically important locations on Earth. But I think it is. And I think the fact that so few people know this is itself a kind of vulnerability.

I’m probably being dramatic. Maybe I’m reading too much geopolitics late at night. But when I learned how concentrated the leading-edge chip supply chain is, how few fabs can produce chips at 7nm, how one company in one country dominates the market, something in my gut said: this matters, and people should know.

Why this keeps me up at night

I think there are two kinds of impressive human achievements. The first kind is the kind we celebrate: going to the Moon, building the pyramids, writing Beethoven’s Ninth. These are spectacular, visible, singular events.

The second kind is the kind we ignore: manufacturing. The daily, invisible miracle of making billions of incredibly complex things, reliably, at scale, forever. Nobody writes songs about yield optimization. Nobody makes movies about photoresist chemistry. But without these things, the modern world doesn’t exist.

TSMC processes millions of wafers a year. Each one goes through hundreds of steps. The machines cost hundreds of millions of dollars. The clean rooms are thousands of times cleaner than a hospital operating room. A single particle of dust, smaller than you could ever see, can ruin everything.

And they do it. Every day. Quietly.

There’s a term I’ve been thinking about: “the invisible cathedral.” The great cathedrals of Europe took generations to build. They were the most complex structures of their era, built with the best technology available, by the most skilled craftspeople alive. People traveled from across the continent to see them.

TSMC’s fabs are our cathedrals. But nobody visits. Nobody even knows they’re there.

I think that’s a shame. I think if people understood what it takes to make the chip in their pocket, they’d feel the same awe I felt at 2am on a Saturday, reading about tin droplets and laser plasma and atomically smooth mirrors.

Maybe I’m wrong. Maybe manufacturing will always be invisible. But I’m going to keep reading about it, because right now, somewhere in Hsinchu, a machine is using laser-vaporized tin to print circuits smaller than a virus onto a silicon wafer the size of a dinner plate, and that feels like something worth paying attention to.