The Invisible Trillion · Sand, Light, and the World’s Most Precise Factories

Step 5: The Recipe

Inside the cleanroom, the wafer will go through somewhere between 500 and 1,000 process steps before it becomes a finished chip. That’s not a typo. Five hundred to a thousand.

Each step does one small thing. Deposit a thin layer of material. Etch a pattern into that layer. Implant some atoms. Clean the surface. Inspect. Deposit the next layer. Etch the next pattern. And so on, for months.

The whole sequence is called a “process flow” or “recipe,” and it is the single most valuable piece of intellectual property at a chip company. The recipe for a cutting-edge chip is the collective work of thousands of engineers over years and costs hundreds of millions of dollars to develop. A rival company that stole the recipe could save itself a decade of work.

I am going to describe the three most important kinds of steps, because once you have them, you have the basic building blocks of everything a fab does.

Step 6: Photolithography (The One That Matters)

If you are going to understand only one thing about chip making, make it this.

Almost every step in the recipe depends on patterning — on making a specific, precise shape on the wafer. The transistors are shapes. The wires connecting them are shapes. The insulation between them is shaped. Every layer of the chip is a shape, and the shapes have to line up, layer after layer, to within a few nanometers of accuracy.

The way you make these shapes is called photolithography, and it works like this.

You start with a wafer coated in a thin layer of a chemical called photoresist. Photoresist is goo that changes its chemistry when light hits it. You’ve used something a bit like it if you’ve ever been to a summer camp that made sun-print T-shirts.

You take a mask — a quartz plate with an incredibly precise pattern etched into it in chrome — and you shine a specific kind of light through the mask onto the photoresist-coated wafer. The light passes through the clear parts of the mask and is blocked by the chrome parts. Where the light hits the photoresist, it changes the resist’s chemistry. Then you rinse the wafer with a developer solution, which washes away either the exposed parts or the unexposed parts (depending on which kind of resist you used), leaving behind a precise pattern.

That pattern is now the template for whatever you do next. If you etch the wafer, the etching happens only where the resist is gone. If you deposit metal, the metal sticks to the exposed silicon and sits on top of the resist. And so on.

This process — expose, develop, etch/deposit — gets repeated dozens of times per chip, with different masks for each layer. One modern chip might go through 80 different lithography steps, each with its own mask, each of which has to line up with the ones before it to nanometer precision.

Here is the catch. The smallest feature you can make with photolithography is roughly the wavelength of the light you’re using. Visible light has wavelengths between 400 and 700 nanometers, which means you can’t pattern features smaller than a few hundred nanometers with visible light. The chip industry passed “a few hundred nanometers” in the 1990s, which means for the last 30 years the industry has been desperately trying to use shorter and shorter wavelengths of light to make smaller and smaller features.

Now we need to talk about the hero of this whole book.

All 19 chapters of The Invisible Trillion · Library · All courses