Step 1 of 8 · The machine

A printer the size of a bus

Inside a spotless chip factory stands a machine the size of a bus. Its job: print the patterns of the chip in your phone, using light.

An EUV machine shines a pattern of light onto a silicon disc, again and again, to build the billions of tiny switches in a chip.

1. Huge machine. This is an EUV lithography machine, built by the Dutch company ASML. It weighs about 180 tonnes and has over 100,000 parts. It's the only kind of machine that can print today's smallest chip patterns. Like the world's most precise photocopier, as heavy as a blue whale.

   It's so big that it travels to the chip factory in pieces: about 40 shipping containers, 20 trucks and 3 cargo planes. Engineers then spend months putting it back together.

   Each one costs well over 150 million euros. Only a handful of companies in the world, like TSMC, Samsung and Intel, buy them.

   If a single one stops for a day, a factory can lose thousands of chips' worth of production, so engineers keep watch on them around the clock.

2. Cleanroom. People here wear full-body suits, and the room is cleaner than a hospital operating theatre. One speck of dust is bigger than the patterns being printed, so it could ruin a chip. Like baking a cake where one grain of sand ruins the whole thing.

   The air is filtered all the time and blown down through the floor, so any dust is carried away. The perforated floor tiles let that air flow out.

   The room glows yellow on purpose. The chemicals used to print chips react to blue light, so the lights have their blue filtered out.

3. Wafers in. Thin round discs of pure silicon, called wafers, slide in at one end. Each wafer is 30 cm across and will be cut into hundreds of chips at the very end.

   Wafers arrive in sealed pods so they never touch dirty air. A robot arm lifts each one into the machine.

   The machine handles about 200 wafers an hour. That's a new wafer every 18 seconds, each printed with about a hundred copies of the pattern.

   So what's inside the box? Let's x-ray it and follow the light.

Weight

≈180 t

Parts

100,000+

Did you know? Engineers started working on EUV in 1986. It took more than 30 years before the first machines printed chips in real factories.

Read more

Every chip, from your phone's processor to a laptop's memory, is made of billions of tiny electric switches called transistors. To make them, factories print patterns onto silicon, layer by layer, a bit like printing a photo.

This printing is called lithography, which means 'drawing on stone'. Light shines through a stencil with the chip's pattern, and the pattern lands on a silicon wafer coated with a light-sensitive layer.

The smaller the pattern, the more switches fit on a chip, and the faster and more power-efficient it gets. But to print smaller things, you need light with a shorter wavelength, and that's where EUV comes in.

EUV stands for extreme ultraviolet. Its waves are just 13.5 nanometres long. A nanometre is a millionth of a millimetre.

This one machine is built from parts made by hundreds of suppliers across the world. No single country can make one alone.

Step 2 of 8 · The machine

The light's journey

Let's x-ray the machine and follow the light, from where it's born to where it lands.

Light is made in one corner, bounced off a chain of mirrors and a patterned stencil, and lands on the wafer.

1. Light source. In this corner, a powerful laser blasts tiny drops of tin. Each blast makes a flash of extreme ultraviolet light, too short-waved for our eyes to see. That flash is where every chip pattern begins. Like a camera flash firing 50,000 times every second.

   The laser is so big that it has its own room under the factory floor. Its beam travels up into the machine through pipes.

   We'll zoom right into this source in a moment. It's one of the hardest bits of engineering in the whole machine.

2. Mask. Near the top, the light bounces off the mask: a plate carrying the chip's pattern. Bright areas reflect the light and dark areas soak it up, so the light picks up the pattern. Like a stencil, but it works by reflecting light instead of letting it through.

   The mask is a square plate of special glass, 15 cm across, coated with a mirror and then a patterned dark layer on top.

   Its pattern is four times bigger than what ends up on the wafer. The mirrors after it shrink the image down, so tiny flaws on the mask shrink too.

3. Wafer. At the bottom, the shrunken pattern lands on the wafer. The whole route, from flash to wafer, takes a few billionths of a second and bounces off about ten mirrors.

   Every mirror loses some light, so only a few percent of what the source makes ever reaches the wafer. That's why the source must be so bright.

   The whole light path is in a vacuum: almost all the air is pumped out, because even air would swallow EUV light.

Mirrors

≈10

Inside

Vacuum

Did you know? The light inside the machine is invisible. If you could stand inside (you can't, there's no air), you would see nothing at all.

Read more

The machine has three big jobs: make the light, carry it to the mask and shape it, then shrink the mask's pattern onto the wafer.

The first group of mirrors, the illuminator, spreads the light evenly so every part of the mask gets the same amount. The second group, the projection optics, focuses the pattern onto the wafer four times smaller.

Because light is lost at every bounce, engineers use as few mirrors as they can. Each extra mirror would cost about a third of the remaining light.

The light travels several metres through the machine, yet the pattern lands in the right place to within a few atoms.

Step 3 of 8 · Making the light

Blasting tin with a laser

It all begins in that corner. Let's zoom into the light source, where the flash is made.

Tiny falling drops of tin are hit twice by a laser, and each one bursts into a flash of EUV light.

1. Tin droplets. A nozzle at the top shoots a stream of molten tin drops downwards. Each drop is about a third as wide as a hair, and they fly at 70 metres per second, faster than a car on a motorway. Like a dripping tap, but the drips are metal and faster than a racing car.

   Tin melts at 232 °C, so the droplet maker keeps it hot and liquid. A vibrating nozzle breaks the stream into perfectly even drops.

   Any drop that isn't hit falls into a catcher at the bottom, and the tin is collected.

   Why tin? When tin gets hot enough, it glows strongly at exactly 13.5 nanometres, the wavelength the mirrors are built for.

2. Two laser hits. As a drop passes, a weak laser pulse squashes it flat, like a pancake. A split second later, a much stronger pulse hits the pancake. Flattening it first gives the main pulse more tin to hit. Like flattening dough before you bake it, so it cooks evenly.

   The laser is a carbon dioxide laser. Its light is infrared, which is invisible to us; here we draw it red.

   The timing is incredible. The laser must hit a drop smaller than a hair, flying at 70 m/s, twice in a row, 50,000 times a second.

   If the laser misses, there is no flash for that drop, and the wafer gets a little less light. Sensors watch every drop and adjust the aim constantly.

3. Plasma flash. The main pulse heats the tin to about 220,000 °C, 40 times hotter than the Sun's surface. The atoms lose some of their electrons, turning into a glowing gas called plasma. That glow is the EUV light. Like a firework: a burst of heat, then a flash of light.

   Atoms are made of a heavy middle (the nucleus) with lighter electrons around it. In plasma, it's so hot that some electrons are knocked loose.

   When electrons fall back into place around the tin atoms, they give off energy as light. For tin, much of that light comes out at 13.5 nm.

Plasma

≈220,000 °C

Drop speed

70 m/s

Did you know? The tin droplets are hit 50,000 times every second. That's about 3 million flashes every minute, nonstop.

Read more

Light is a wave, and the distance between two wave peaks is its wavelength. Visible light has waves of about 400 to 700 nanometres. EUV light's waves are only 13.5 nanometres.

The shorter the wave, the finer the details it can draw. The older machines used 193 nm light. Switching to 13.5 nm is more than 14 times shorter.

Making EUV light is hard because nothing simple gives it off. You need matter heated to hundreds of thousands of degrees, so the machine makes a tiny star, 50,000 times a second.

The laser fires two pulses because a round drop is a small target. The first pulse spreads it into a thin disc. The second, stronger pulse then turns that disc into plasma much more evenly.

To work out how to make EUV, engineers spent years testing other materials, like xenon gas. Tin won because it gives the most light at 13.5 nm.

Step 4 of 8 · Making the light

Catching the flash

That flash sprays light in every direction. Something has to catch it and send it on its way.

A big curved mirror behind the flash catches the light and focuses it into a single point, the doorway to the rest of the machine.

1. Collector mirror. Behind the flash sits a dish-shaped mirror about 65 cm wide. Its curve catches light flying out in many directions and sends it all towards one point. Like a satellite dish, catching weak signals and focusing them on one spot.

   The laser comes through a hole in the middle of the dish to reach the drops.

   The dish's surface is coated with dozens of super-thin layers. Later we'll dive right into those layers to see how they reflect.

2. Into the dark. The light travels through a vacuum, a space with almost no air. EUV is so easily swallowed that a few centimetres of normal air would stop it. The light reaches a focus point and moves on into the scanner.

   Big pumps keep removing air from the machine all the time. Only a tiny bit of hydrogen gas is allowed in, on purpose.

   The focus point is called the intermediate focus. It's the doorway between the light source and the rest of the machine.

3. Hydrogen clean-up. Each flash also sprays tiny bits of tin. A steady flow of hydrogen gas sweeps across the mirror, catching the tin before it can stick and dim the mirror. Like windscreen washers keeping your view clear in the rain.

   Hydrogen reacts with tin to make a gas called stannane, which pumps can suck away. Without this, the mirror would be coated in tin within hours.

   Even so, collector mirrors wear out slowly and are swapped for fresh ones during maintenance.

Dish

≈65 cm

Air inside

None

Did you know? The collector mirror sits just a few centimetres from a plasma 40 times hotter than the Sun's surface, and still has to stay perfectly shaped.

Read more

A flash of plasma sends light out in all directions, like a light bulb. To use it, the machine must gather as much as possible and point it one way.

The collector is shaped like part of an egg (an ellipse). Light from one special point of an ellipse always bounces to a second special point. The flash sits at the first point; the doorway to the scanner sits at the second.

The whole source works in a vacuum with a whiff of hydrogen. Hydrogen barely absorbs EUV, but it slows down flying tin and cleans the mirror.

Without the hydrogen flow, the precious collector mirror would go dull very quickly, so the gas is its constant cleaner.

Step 5 of 8 · Printing the chip

The stencil and the shrink

The focused light leaves the source and travels on through the machine. Let's follow it along to its next stop: the stencil with the chip's pattern.

The light bounces off the patterned mask, then curved mirrors shrink the pattern four times on its way to the wafer.

1. The mask. The mask hangs upside down at the top. Its surface is a mirror covered with a dark pattern: where it's shiny, light bounces down; where it's dark, the light is soaked up. Like a shadow puppet, but made with reflected light.

   Each chip layer needs its own mask. A modern chip may need dozens of different masks, used one after another.

   Making one mask can cost hundreds of thousands of euros, because every line on it must be perfect.

2. Dust shield. Just below the mask is a pellicle: a see-through sheet thinner than a soap bubble. Dust lands on the sheet instead of the mask. It's out of focus there, so it doesn't print. Like a dusty window: you look straight through it at the view outside.

   The pellicle is only a few tens of nanometres thick, so it lets most of the EUV light through. It gets very hot, so it must be made of tough materials.

   Without it, a single speck of dust on the mask would print onto every chip on every wafer.

3. Shrinking mirrors. Below, curved mirrors focus the pattern down onto the wafer, four times smaller. A mistake on the mask shrinks by four too, which makes the job a little easier. Like a magnifying glass used backwards, making things smaller instead of bigger.

   These are the projection optics. Each mirror is curved with a precision of a fraction of a nanometre.

   The newest High-NA machines have bigger mirrors that catch light at steeper angles, so they can print lines down to about 8 nm.

Shrink

4×

Mask size

≈15 cm

Did you know? The mask is only about 15 cm across, but the pattern on it is enough to print the whole chip, over and over.

Read more

A chip is built up in layers, like a skyscraper with many floors. Each floor has its own pattern, so each one needs its own mask and its own trip through the machine.

EUV masks reflect light rather than letting it through. They're made of ultra-flat glass with the same layered mirror coating, plus a patterned dark absorber on top.

The projection mirrors shrink the image four times. So a line 52 nm wide on the mask becomes a 13 nm line on the wafer.

The light only uses a narrow slit of the mask at a time. The whole pattern is printed by sliding mask and wafer past that slit.

Step 6 of 8 · Printing the chip

Printing in a sweep

The shrunken pattern now reaches the wafer. Let's watch how it gets printed.

The pattern is printed through a thin slit of light while the mask and wafer slide past it, perfectly in sync.

1. Slit of light. The light doesn't print the whole chip at once. It forms a thin strip, the slit. The mask and wafer slide past it, so the pattern is printed one sweep at a time. Like a scanner at home: a bar of light passes over the page, line by line.

   The mask moves four times faster than the wafer, because its pattern is four times bigger. And they move in opposite directions, because the mirrors flip the image.

   One sweep prints one chip area, called a field. Then the wafer jumps to the next field and the machine sweeps again.

2. Floating stage. The wafer sits on a stage that floats on magnets, with no friction. It speeds up harder than a race car, then stops dead in the exact right spot, again and again. Like a maglev train for a single wafer.

   Magnets lift and push the stage, just like the maglev train in our other story. With nothing rubbing, it can move fast and stop precisely.

   Lasers measure where the stage is thousands of times a second, and the magnets adjust instantly.

   If the stage wobbled even slightly, every line would blur. So the whole machine sits on its own shock-absorbing frame.

3. Perfect alignment. Each chip layer has to land exactly on top of the last one, within about 1 to 2 nanometres. That's about the width of a few atoms. Like stacking transparent drawings so every line matches up perfectly.

   Before printing, sensors measure small marks on the wafer to find exactly where earlier layers are.

   The wafer can even warm up slightly and stretch from the light. The machine measures this and corrects for it as it goes.

   But all of this depends on mirrors that can bounce EUV at all, which is a puzzle in itself. Let's dive into one.

Alignment

1–2 nm

Speed

≈200 wafers/h

Did you know? The machine prints a wafer every 18 seconds or so, about 200 wafers per hour.

Read more

A machine that sweeps the pattern instead of flashing it all at once is called a scanner. It lets the mirrors stay small, because they only need to print a thin strip perfectly.

Moving the mask and wafer in perfect step is a huge challenge. If they drifted by even a nanometre relative to each other, the lines would blur.

After each field, the stage jumps to the next spot and sweeps again. A full 30 cm wafer has about a hundred fields.

While racing back and forth, the stage always knows where it is to well under a nanometre, measured by laser beams.

Step 7 of 8 · Down to the atoms

A mirror made of layers

We keep saying "mirror". But EUV gets swallowed by almost everything, even glass. So how can a mirror bounce it? Let's dive into the surface of one of those shrinking mirrors.

Each mirror is coated with dozens of atom-thin layers, and each layer reflects a tiny bit of light, all perfectly in step.

1. Light goes in. Here's the EUV light, drawn as a wave, arriving at the mirror's surface. A normal mirror would absorb almost all of it. So this one is built differently. Like a ball that would sink into mud instead of bouncing.

   A bathroom mirror uses a shiny metal layer behind glass. At EUV wavelengths, both the glass and the metal would soak up the light.

   So engineers stopped trying to find one perfect reflector, and stacked lots of weak ones instead.

2. Stacked layers. The mirror is coated with 40 to 50 pairs of layers: one of a metal called molybdenum, one of silicon. Each pair is about 7 nanometres thick, only a few dozen atoms. Like a stack of LEGO plates, each just a few atoms tall.

   Each layer is laid down atom by atom by a process called sputtering, with almost no bumps. Here we draw only 14 pairs so you can see them.

   The spacing is matched to half the wavelength of the light. If the layers were even a bit too thick or thin, the echoes would fall out of step.

3. Echoes in step. Each time the light meets a metal layer, a small echo bounces back. Because the layers are spaced just right, all the echoes leave in step and add up into one strong reflection. About 70% of the light comes back. Like a crowd clapping in rhythm: many quiet claps make one loud beat.

   This is called constructive interference: waves whose peaks line up add together into a bigger wave.

   70% sounds good, but after about ten mirrors, 0.7 × 0.7 × ... leaves only a few percent of the light. That's why fewer mirrors are better.

Layer pairs

40–50

Reflects

≈70%

Did you know? If one of these mirrors were blown up to the size of Germany, its biggest bump would be only about 0.1 millimetres tall.

Read more

Light is a wave. When two waves meet with their peaks lined up, they add together into a bigger wave. When a peak meets a dip, they cancel out.

Each boundary between molybdenum and silicon reflects only about 1–2% of the light. But with dozens of boundaries, and each echo lined up with the others, those tiny bits add up to about 70%.

The mirrors are made by ZEISS in Germany. Their shape must be accurate to well under a nanometre across the whole surface, or the pattern on the wafer would blur.

Some EUV mirrors take months to polish and coat, and they are among the smoothest objects humans have ever made.

Step 8 of 8 · Down to the atoms

Light turns into a pattern

So that's how the light is steered. Staying at this tiny scale, let's slide over to where it finally lands: the light-sensitive coating on the wafer.

The wafer is coated with a light-sensitive layer; where light hits, chemistry changes it, and those spots are washed away.

1. EUV photon. Light comes in tiny packets of energy called photons. Each EUV photon carries about 40 times more energy than a photon of visible light, so it hits the wafer's coating, the resist, like a hammer. Like a single hailstone hitting a car roof, not a drizzle.

   Because each photon is so powerful, only a small number arrive at each tiny spot. That makes the printing a bit uneven, a problem engineers call shot noise.

   When a photon is absorbed, it knocks loose a few electrons (drawn in blue), which spread out and trigger the chemistry.

2. Acid forms. The resist is a plastic made of long chain molecules, with special yellow molecules mixed in. When the electrons hit those molecules, they turn into acid. The wafer is then warmed, and the acid snips the chains around it. Like scissors cutting a long necklace into loose beads.

   One acid molecule can cut many chains, one after another. This 'chemical amplification' means fewer photons are needed.

   But the acid can also drift a little, which blurs edges. Engineers tune the warming time and temperature to keep it in check.

3. Pattern appears. A liquid called the developer washes the snipped plastic away. What's left are sharp lines of resist, a copy of the mask's pattern. Next, the bare silicon gets etched or filled with metal, and the whole thing repeats for the next layer.

   The leftover resist acts as a stencil for the next step, where chemicals or plasma carve the pattern into the material below.

   A modern chip has dozens of patterned layers. Many of the most detailed ones are printed with EUV.

   So from a flash of exploding tin, through ten flawless mirrors, the light has finally become part of a chip.

Lines

≈13 nm

Photon

≈92 eV

Did you know? The lines printed here can be about 13 nanometres wide. A human hair is about 80,000 nanometres wide.

Read more

Resist is a thin coating, only a few dozen nanometres thick, spread evenly over the wafer before it goes into the machine.

In most EUV resists, light makes a tiny amount of acid. During a short bake, that acid breaks the long molecules, so the exposed spots dissolve in the developer.

After developing, the wafer has a stencil of resist on top. Etching then carves the pattern into the layer beneath, and the resist is cleaned off.

A finished chip has gone through hundreds of steps like this, building billions of transistors and the wires that connect them.

Engineers are designing new metal-based resists that soak up EUV better, so even fewer photons are needed for each sharp line.