Growing the wafer
Every chip starts as something humble: ordinary sand, which is mostly silicon. But sand isn't pure enough. To make electronics, you need silicon so clean that out of a billion atoms, only one is a stranger. Imagine an Olympic swimming pool of water in which a single drop of ink would be too much pollution — that's roughly the purity we're after.
Once you have ultra-pure silicon, you melt it and grow it into a single giant crystal. A small seed crystal is dipped into the molten silicon and slowly pulled upward while spinning, and the melt freezes onto it atom by atom — like pulling a glassy candle out of a vat of wax, except the whole thing locks into one perfectly ordered lattice. The result is a shiny cylinder called an ingot, often as tall as a person and as heavy as a sack of cement.
The ingot is then sliced into thin discs, like cutting a salami into rounds. Each disc is a wafer. It's polished until it's flatter and smoother than any mirror you've ever seen — flat to within a few atoms across the whole surface. This blank, gleaming wafer is the canvas. Everything that follows is about painting circuits onto it.
Photolithography — printing with light
Here's the central magic trick of the whole fab. You can't draw circuits one wire at a time — there are billions of them, far too small to touch. Instead you print them all at once, with light. The process is called photolithography, which literally means "writing on stone with light."
Think of an old film photograph. You shine light through a negative onto light-sensitive paper, and wherever the light lands, the paper changes. A fab does the same thing, but in reverse and in miniature. First the wafer is coated with a thin film of light-sensitive goo called resist. Then light is shone through a mask — a glass plate carrying the pattern for one layer of the circuit, like a stencil. Wherever light hits the resist, the resist's chemistry flips, so it either hardens or washes away.
- Coat the wafer with a thin, even layer of light-sensitive resist.
- Shine light through the mask, projecting the layer's pattern onto the resist.
- Wash away the exposed (or unexposed) resist, leaving a stencil of the pattern on the wafer.
- Etch: a gas or acid eats into the silicon wherever the resist no longer protects it, carving the pattern in.
- Strip off the remaining resist, and the carved pattern stays behind.
Layer by layer
One round of printing and etching makes just one layer. A real chip is built from dozens of layers, stacked and aligned on top of each other with breathtaking precision — like a city built floor by floor, where the plumbing on the tenth floor has to line up perfectly with the pipes left on the first.
Three kinds of step do the building. Doping sprinkles in tiny amounts of other elements to change how silicon conducts — this is how you create the working parts of a transistor, the tiny switches that do all the thinking. Deposition lays down fresh thin films of material, like spraying on a new coat of paint. And etching, which you just met, carves shapes back out. Repeat, repeat, repeat.
The bottom layers hold the transistors. The layers above them are the wiring — many levels of tiny metal wires (usually copper) that connect the transistors into circuits, stacked like the road network of a vertical city: local streets near the ground, big highways higher up. A modern chip can have fifteen or more wiring levels threaded over the transistors below.
Process nodes (the 'nm')
You've probably seen chips advertised as "7nm" or "3nm." That number is the process node, and roughly speaking it's a label for how small and tightly packed the features are. Smaller features mean you can cram more transistors into the same square of silicon — more switches, doing more work, using less energy per switch. A nanometer is a billionth of a meter; for scale, a single atom of silicon is about a quarter of a nanometer across, so these features are now just a handful of atoms wide.
To print features that small, ordinary light is too clumsy — its waves are simply too fat to draw such fine lines, like trying to sign your name with a paint roller. The newest chips use EUV (extreme ultraviolet), a kind of light with a far tinier wavelength. Making EUV is wild: a laser blasts droplets of molten tin tens of thousands of times a second to create a glowing plasma, and the machines that do this are among the most complex objects humanity has ever built.
Yield & testing
When your features are only a few atoms wide, a single speck of dust is a boulder. One stray particle landing on the wafer can short out wires and kill a chip outright. That's why fabs are cleanrooms thousands of times cleaner than a hospital operating room, and the workers wear head-to-toe "bunny suits" — not to protect themselves, but to protect the wafer from the flakes of skin and lint that you and I shed constantly.
Because defects are unavoidable, not every chip on a wafer works. The fraction that come out good is called the yield. A single wafer holds hundreds or thousands of identical chips (each one is called a die), and the bad ones are simply discarded. Early in a new process the yield might be dismal; over months of tuning it climbs, and good yield is the difference between a profitable chip and a money pit.
Before any chip leaves, it's tested — needle-fine probes touch the die and run millions of electrical checks while it's still on the wafer. The ones that pass aren't all equal, either: some run faster or sip less power than others. So they get sorted, a practice called binning — the fastest, best dies might be sold as a premium part, while perfectly good but slower ones become a cheaper model. Same recipe, different grades, like sorting apples by size.
Packaging
At last the finished wafer is diced — sawn or laser-cut along the grid lines into individual dies, like breaking a sheet of brownies into squares. The bad ones, already marked during testing, are thrown out. Each good die is far too tiny and fragile to use on its own, so it gets a home: it's bonded onto a small base, its connection points wired or bumped to the base's pins, and then sealed inside a protective package of plastic or ceramic.
That package is the black rectangle with metal legs you actually recognize as a chip — what engineers casually call a chip. Inside that humble shell sits the real marvel: a sliver of silicon holding billions of transistors, all printed by light. The package just protects it, carries away its heat, and gives it sturdy pins so it can plug into the wider world of a circuit board.