Wafer

A wafer is a round, very thin disc made of silicon. Silicon is the material that sand largely consists of as well, only for wafers it is cleaned with extreme effort. The disc is typically 30 centimeters wide and less than a millimeter thick, and its surface is polished smooth on an atomic level. Computer chips are manufactured on this surface: the tiny components that handle computing in phones, cars, and data centers. In the process, not just one chip is created per disc, but many at once, neatly arranged side by side in a grid. Only at the very end is the disc sawn apart, and each individual rectangle becomes a finished chip.

Why disc size determines chip price

Manufacturing a wafer is expensive, but the price barely depends on how many chips end up on it. A factory processes the disc as a whole, step by step. The more usable chips a disc yields, the cheaper the individual chip becomes. That’s why the industry has shifted from small discs to ever larger ones over decades. Today’s standard is a 300 millimeter diameter.

Also decisive is the yield. This refers to the proportion of chips on a disc that actually work. A single speck of dust or a flaw in the material can render a chip unusable. With very modern manufacturing processes, a significant portion of chips often fail initially. If the yield rises later, the cost per functioning chip drops considerably, without much changing in the process itself.

This is relevant for investors and news because large chip companies state their capacity in wafers per month. A factory that manages 100,000 wafers a month is a different league than one with 20,000. When Nvidia or Apple need more chips, they have to secure wafer capacity from a contract manufacturer like TSMC. This capacity is scarce and is often booked years in advance.

From crystal to sawn disc

At the beginning there is a single large silicon crystal. It is slowly pulled upward out of a glowing melt and grows into a cylinder about one meter long in the process. This cylinder is sawn into discs, ground, and polished. The result is the bare wafer, on which no circuitry is yet visible.

Afterward, hundreds of processing steps follow in the factory. Again and again, an extremely thin layer is applied, a pattern is exposed onto it with light, and the excess material is etched away. You can imagine this like printing very many layers on top of each other, where each layer must fit exactly onto the one below. This is how the conductive tracks and transistors of a chip are created, layer by layer.

Finally, each chip on the disc is tested electrically. Defective specimens are marked and later sorted out. Then a machine saws the wafer into the individual rectangles, which are called dies. Each good die goes into a package with connectors and thus becomes the chip you see on a circuit board.

Wafers in news about chip factories and AI

The term mostly comes up when it’s about supply bottlenecks or new factories. When a report speaks of a factory for 40,000 wafers per month, it’s describing its production volume. Subsidies such as the European Chips Act are also frequently measured in planned wafer capacity. That is the industry’s common currency, not the number of finished chips.

In the AI world there is a special case that often causes confusion. The company Cerebras does not saw apart its disc, but instead builds a single giant chip out of a complete wafer. This Wafer Scale Engine is roughly the size of a dinner plate and intended for training AI models. This is the exception that shows just how unusual normal sawing apart actually is.

A common misconception is equating wafer and chip. The wafer is the workpiece during manufacturing, the chip is the end product made from it. Those who distinguish between the terms will find it much easier to understand reports about the semiconductor industry.