Inside a chip factory, a machine the size of a bus drops balls of molten tin through a vacuum at 70 metres per second. Each ball is about 25 micrometres wide, a third the width of a human hair. A laser hits every falling droplet twice: a weak pulse flattens the droplet into a pancake, and a strong pulse blasts the pancake into a glowing plasma, many times hotter than the surface of the Sun.

The machine repeats the double hit 50,000 times every second. Each blast gives off a flash of light no eye can see, and that light prints the switches inside the chip in your phone. The finest lines it prints are several times narrower than a flu virus.

Only one company in the world, ASML in the Netherlands, builds the machine. Each one costs a couple of hundred million dollars (the price depends on the model and the year), and Intel says delivering one takes 40 freight containers and three jumbo cargo jets.

Start with what the light prints: switches. What is a switch on a chip, and why would anyone want billions of them?

Switches made of silicon

A chip is a city of tiny switches called transistors. A transistor has no moving parts. A small voltage on one terminal, the gate, opens a path for current between the other two terminals. Remove the voltage, and the path closes.

Wire a few switches together and the switches make decisions. A few dozen can add two numbers (our page Adding With Nothing but Switches builds an adder step by step). A phone chip that plays video and recognises your face needs billions of transistors on a piece of silicon about the size of a fingernail.

So how small does each switch have to be?

Shrink the switch

Your goal: drag the slider to shrink every switch, until your 1 cm chip holds about 19 billion switches, as many as a 2023 iPhone chip.

Each time a switch gets half as wide, four times as many fit, because the switch shrinks in both directions. Ten halvings fit about a million times as many switches.

Smaller switches also flip faster and use less energy per flip. And packing more switches onto the same silicon made each switch cheaper, decade after decade.

In 1965, Gordon Moore, who later co-founded Intel, noticed that the number of transistors on a chip was doubling every year. In 1975 he changed his estimate to every two years. The prediction became known as Moore's law, and the industry kept roughly to that pace for about 50 years.

To shrink a switch, someone has to draw the switch smaller. What can draw a line 20 nanometres wide?

Drawing with light

Chips are printed a bit like photographs, a process called lithography. A mask holds the pattern for one layer of the chip, like a stencil. Light passes through the mask, and a lens shrinks the pattern four times onto a silicon disc called a wafer.

The wafer is coated with a film that changes wherever light hits the film. Washing clears away the lit parts, and the bare silicon underneath can be etched or filled. A modern chip repeats the print-and-etch steps for dozens of layers, each layer lined up with the one below to within a few nanometres.

So light is the pen. How thin a line can a pen made of light draw?

Light is a fat pen

Light is a wave, and the distance from one crest to the next is called the wavelength. A wave spreads out as the wave passes an edge, so the image of a sharp edge comes out blurred. Drawing chip lines with light is like writing tiny letters with a thick marker.

Your goal: pick a light, then drag the line width down until the printed lines fall apart. Which light draws the thinnest clean lines?

Blue-violet light from a lamp printed chips in the 1980s. Deep ultraviolet light at 193 nm, from a laser, still prints most layers of a chip today. Extreme ultraviolet light, or EUV, at 13.5 nm, prints the thinnest lines of the three.

If shorter waves draw thinner lines, why did EUV take decades to work?

Light that air swallows

EUV light is absorbed by almost everything: glass, lenses, even air. So the whole light path sits in a vacuum, and every lens is replaced by a mirror. Even the mask is a mirror.

An ordinary mirror swallows EUV too. An EUV mirror is a stack of about 40 pairs of alternating layers of molybdenum and silicon, each layer a few nanometres thick. Each layer reflects a little light, and the small reflections add up.

Zeiss, the German company that makes the mirrors, says that if one mirror were blown up to the size of Germany, the biggest bump on the mirror would be about 0.1 mm high.

Losing 97% of the light means the source must be fiercely bright. How do you make bright EUV light?

Exploding tin

No bulb makes 13.5 nm light. A tin plasma does: a gas of tin so hot that the tin atoms lose many of their electrons. Those tin atoms glow strongly at almost exactly 13.5 nm.

Hence the falling droplets. A weak pulse flattens each droplet, and a carbon dioxide laser built by the German company Trumpf blasts the flattened droplet into plasma: about 220,000 °C, by Trumpf's figure. A curved collector mirror gathers the flash.

Your goal: press Fire when a droplet crosses the dashed laser beam. Try to hit five droplets in a row.

The real machine hits 50,000 droplets a second, hour after hour. A stream of hydrogen gas sweeps away tin dust before the dust can coat the collector mirror.

All of this, just to make a flash of light. How did one company end up as the only maker of the machine?

Thirty years to work

In the mid-1980s, Hiroo Kinoshita, a researcher at the Japanese phone company NTT, printed patterns with extreme ultraviolet light. Many experts doubted the idea would ever work in a factory. In the late 1990s, Intel and other chipmakers teamed up with US national laboratories to push EUV forward.

The light source was the hardest part. Nikon and Canon, ASML's big Japanese rivals, eventually dropped EUV, and ASML kept going. In 2012, Intel, TSMC and Samsung invested in ASML to help pay for EUV. In 2013, ASML bought Cymer, the San Diego company building the light source. The first phone chips with EUV-printed layers went on sale in 2019.

Veritasium tells the whole story, with the engineers who built the machine, in a long video. The video can't let you print a line or fire the laser, so try those in the widgets above.

So shorter light prints thinner lines. Exactly how thin can a line be?

How thin a line can be

Chip engineers use one short formula. The smallest line width depends on the wavelength \(\lambda\), on how wide a cone of light the lens can catch, called the numerical aperture (NA), and on a factor \(k_1\) for how clever the factory is:

Where it shows up

• Your phone and laptop. The main chips in recent flagship phones and laptops have EUV-printed layers.
• AI chips. The graphics chips that train AI models are among the largest chips made, some with more than 100 billion transistors.
• Memory. Some memory chips (DRAM) use EUV for their finest layers.
• Not everything. Chips in cars, appliances and chargers mostly use older, cheaper processes with no EUV at all.

Only a few companies make the most advanced chips, mainly TSMC, Samsung and Intel, and all of them depend on ASML for EUV machines.

Your chip, explained

Now you can explain the chip in your pocket. The chip holds billions of switches, and halving a switch's width fits four times as many: you built a phone chip from a grid of sand-sized switches.

Thin lines need short waves: you watched 193 nm light blur where EUV still printed clean. EUV dies in air and in glass, so the light travels through a vacuum off layered mirrors that lose about 97% of the light.

To make up for the loss, the machine blasts 50,000 tin droplets a second, a job you tried by hand. That chain of hard problems is why one company builds the machine.

Where to go next

EUV opens a lot of other questions:

• Is Moore's law over, or just slowing down?
• What happens when a transistor is so small that electrons leak straight through the transistor's walls?
• If switches can't shrink much more, how will chips keep getting faster?
• Why are most of the world's most advanced chips made in Taiwan?
• How does a chip factory keep out dust that would ruin a 20 nm line?

Ask Phy

• Walk me through every step of making one layer of a chip, from a bare wafer to an etched pattern.
• Why does putting water between the lens and the wafer let 193 nm light print thinner lines?
• What is a gate-all-around transistor, and why are chipmakers switching to it?