Making of Microchips

From Sand to Silicon: Making of Microchips

Microchips are everywhere. They control our phones, cars, computers, medical equipment and industrial machinery. Artificial intelligence depends on thousands of them working together in enormous data centres. A modern processor can contain tens of billions of transistors, packed onto a piece of silicon barely larger than a fingernail. That sounds impressive. How we actually manufacture those transistors is even more remarkable.

It Starts With Silicon

Microchips are usually made from silicon, an element found abundantly in the Earth’s crust and, among other things, in sand. But you cannot simply melt some beach sand and make an iPhone processor. The silicon used by the semiconductor industry has to be purified to an extraordinary degree. It is then melted and formed into a large single crystal, from which extremely thin discs called wafers are cut. These wafers are polished until their surfaces are almost perfectly smooth. Hundreds of chips can eventually be produced on a single wafer.

Why Silicon?

Silicon is a semiconductor. Copper conducts electricity easily. Glass generally does not. Silicon is more interesting because engineers can manipulate its electrical behaviour. By adding tiny quantities of other elements — a process called doping — regions can be created in which electrical current behaves differently. This property makes it possible to build the transistor. A transistor is essentially an incredibly small electronic switch. It can control whether current flows and thereby represent the zeros and ones on which digital computing is based. One transistor does very little.

Billions working together can run an operating system, calculate a weather model or generate an answer from an AI model.

Printing Something Almost Unimaginably Small

The most fascinating stage of chip manufacturing is lithography. A light-sensitive material is applied to the silicon wafer. A highly sophisticated machine projects a pattern onto it. Selected areas can then be removed, modified or covered with another material. The procedure is repeated over and over again. Imagine constructing a city by repeatedly printing maps on top of each other. One map defines the buildings, another the roads, another the electricity network and another the communications infrastructure. Now imagine doing this at a scale measured in billionths of a metre.

That is roughly what happens inside a semiconductor factory.

EUV: When Ordinary Light Is Too Large

Making transistors smaller eventually created an unusual problem: the wavelength of conventional light became too large to produce the required patterns efficiently. The industry’s answer was Extreme Ultraviolet lithography, or EUV. EUV systems use light with a wavelength of about 13.5 nanometres. Producing that light is an engineering achievement in itself. Microscopic droplets of tin are fired through a vacuum and struck by powerful laser pulses. The resulting plasma produces EUV radiation. The light then has to be directed using extremely precise mirrors because ordinary lenses cannot simply be used in the same way as with visible light.

These lithography machines rank among the most complicated machines humans have ever built.

A Chip Is Built, Not Carved

A processor is not simply etched out of a piece of silicon. It is constructed layer by layer. Materials are deposited, patterns are exposed, parts are etched away and selected atoms are implanted into the silicon. Then the process starts again with another layer. An advanced processor can require hundreds or even thousands of manufacturing steps. The transistors themselves are also no longer simple flat structures. Modern designs increasingly use complex three-dimensional shapes that give engineers better control over the flow of electrons. Above them are multiple layers of microscopic metal connections. Think again about a city. The transistors are the buildings, while the metal connections are the roads connecting everything together.

Except this city contains billions of buildings.

One Piece of Dust Can Ruin a Chip

This extraordinary scale explains why semiconductor factories look so unusual. A particle of dust that would be completely harmless in an ordinary factory can destroy microscopic structures on a chip. Manufacturing therefore takes place inside extremely clean environments known as cleanrooms. The air is continuously filtered. Temperature and humidity are tightly controlled. Workers wear protective clothing covering almost their entire bodies. The clothing is not primarily there to protect the workers.

It protects the chips from the workers. Humans constantly release microscopic skin particles, fibres and other contamination — exactly what a semiconductor factory does not want.

Not Every Chip Works

Even with all this precision, manufacturing is never perfect. After production, the wafer is tested to determine which individual chips work correctly. This introduces one of the most important concepts in semiconductor economics: yield. Suppose a wafer contains 500 processors but only 300 work. The manufacturer cannot simply charge customers for the other 200.

Improving the yield from 60 to 90 percent can therefore make an enormous difference to production costs. It is one reason why new generations of chips are initially so expensive. Manufacturers gradually learn how to produce them with fewer defects.

The wafer is eventually cut into individual chips, which are tested again and packaged.

Why Chips Are Becoming More Than One Chip

For decades, progress largely meant putting more transistors onto a single piece of silicon.

That strategy is becoming increasingly difficult and expensive.

One solution is the chiplet.

Instead of manufacturing one enormous processor, several smaller specialised chips can be combined inside one package. A manufacturer might use its most advanced manufacturing technology for processor cores while producing less demanding components using an older and cheaper process. Advanced packaging is therefore becoming almost as important as manufacturing the transistor itself.

This is particularly important for artificial intelligence. AI processors perform enormous numbers of calculations simultaneously and constantly exchange data with memory. The distance between processor and memory, the speed of those connections and even the ability to remove heat can limit performance.

The fastest transistor is of little use if it spends much of its time waiting for data.

Are We Reaching the Limit?

For decades the semiconductor industry followed the general direction described by Moore’s Law: more and more transistors could be economically placed on a chip. But there is an obvious problem with making things smaller forever. Eventually you encounter atoms. At these dimensions, quantum-mechanical effects become increasingly important. Electrons can behave in ways that engineers designing larger electrical circuits never have to worry about.

This does not mean that progress is about to stop. Instead, the nature of progress is changing.

Future computers will increasingly gain performance from three-dimensional structures, chiplets, specialised processors, faster memory, new materials and more sophisticated packaging rather than simply shrinking every transistor generation after generation.

A Global Technological Puzzle

There is another remarkable aspect of the microchip. No single country really makes an advanced processor entirely by itself. A chip can be designed in the United States, using specialised software developed by another company, manufactured in Taiwan or South Korea, using lithography equipment from the Netherlands containing components and technologies from several countries, and finally packaged elsewhere in Asia. Modern semiconductor production is therefore not merely an engineering achievement.

It is one of the most complicated international supply chains ever created. And that explains why microchips have become strategically important to governments as well as technology companies.

The Invisible Technology Behind Modern Life

We hardly notice microchips anymore. That may be their greatest achievement. A car can contain hundreds or even thousands of semiconductor devices. Our phones contain several highly specialised processors. Data centres contain vast numbers of chips operating continuously. Behind every one of them lies an extraordinary combination of physics, chemistry, optics, materials science, mechanical engineering and computer science.

We started with something as ordinary as silicon.
We learned how to manipulate matter at almost atomic dimensions.

And from that we created machines containing billions of switches capable of performing billions or trillions of operations.

The microchip may be tiny.
The technology behind it certainly isn’t.