Why manufacturing matters
A modern chip is not assembled from parts. It is made by building up thin patterned layers on a flat disc of crystal, so that many identical circuits are produced at once. The history of the chip is therefore as much a history of manufacturing methods as of circuit design. Who invented the first integrated circuit is a separate story; this article follows how the making of silicon chips developed afterward.
From sand to wafer
Silicon is common in the Earth’s crust, mostly bound up in sand and rock. For chips it has to be purified to an extreme degree and grown into a single crystal. A widely used method, the Czochralski process, pulls a seed crystal slowly out of molten silicon so that a large cylindrical ingot grows. The ingot is sliced into thin discs, the discs are polished flat and these become wafers.
Early wafers were small, roughly an inch or two across. Over the decades the standard grew, and a diameter of 300 millimeters is common in modern fabrication plants. A bigger wafer holds more chips, which generally lowers the cost per chip, but it also demands equipment and handling to match.
Silicon takes over
The first integrated circuits of the late 1950s varied in material and layout, but silicon became the dominant choice. One reason was that silicon forms a stable oxide layer when heated in oxygen. That oxide can insulate and protect, and it can be etched in patterns. In 1959, Jean Hoerni at Fairchild Semiconductor developed the planar process, which built transistors in a flat layer beneath a protective oxide. It made production more reliable and suited it to the repeated, step-by-step methods that chipmakers still use.
Photolithography
The central step is photolithography. The wafer is coated with a light-sensitive material called photoresist. Light is shone through a mask that carries the pattern for one layer of the circuit, and the exposed or unexposed areas are then washed away, depending on the type of resist. The remaining pattern guides the next operation, such as etching away material, adding dopant atoms to change how silicon conducts electricity, or depositing metal.
A finished chip is built from dozens of such layers, each one aligned precisely with the one beneath. Early lithography used contact masks, which could wear out. Later systems used projection optics and, from the 1970s onward, equipment that exposed the pattern one small field at a time and stepped across the wafer. Over time, manufacturers moved to shorter wavelengths of light to print smaller details, and more recent plants use extreme ultraviolet light for the finest layers. The tools involved are among the most complex and expensive machines in industry.
Cleanrooms and yield
At these scales, a speck of dust can ruin a circuit. Fabrication plants, often called fabs, are built as cleanrooms where air is filtered and workers wear full coverings. Not every chip on a wafer works. The share that does is called yield, and improving yield is a constant concern, because it directly affects cost. Wafers are tested, cut into individual dies, and the good ones are packaged with connecting pins or pads.
The doubling of transistors
In 1965, Gordon Moore, then at Fairchild and later a cofounder of Intel, noticed that the number of components on a chip had been roughly doubling every year and expected the trend to continue for some years. In 1975 he revised the pace to about every two years. The statement became known as Moore’s law, but it is better described as an observation about industry progress than a law of nature. It held as a rough guide for decades because the industry organized its research around it, not because physics guaranteed it.
For scale, Intel’s 4004 microprocessor of 1971 is usually described as having a little over two thousand transistors. Processors of the 2020s contain billions. The numbers are not directly comparable across designs, but the growth is real.
Limits and new directions
As features shrank, manufacturers ran into heat, electrical leakage and the cost of new plants. The names given to modern manufacturing generations, quoted in nanometers, are now marketing labels that no longer correspond neatly to a single physical dimension. To keep improving, designers have changed transistor shapes, stacked layers vertically and combined several small chips in one package. Whether the pace of the past can continue is uncertain, and experts hold different views.
Why it matters
Cheap, dense chips made possible everything from the quartz watch on a wrist to smartphones and GPS chipsets. The underlying lesson of the chip story is that a manufacturing process, repeated millions of times with tiny improvements, can change what technology costs and what it can do.