Every modern microprocessor consists of millions of basic electronic components — transistors. Over time, these transistors have become smaller and smaller, enabling companies like Intel to squeeze more of them inside modern microprocessors.
Consider this: Back in 1971, the Intel 4004 processor had 2,300 transistors. Forty-four years on, the 6th Generation Intel Core Skylake CPU has over 1.3 billion. It’s thanks to this ever-increasing transistor count that we have computer technology that literally fits in our pockets.
As you’d imagine. Building these advanced CPUs is a complicated process. In this article, we take a look inside Intel’s factories (aka ‘Fabs’) to see how they work.
The preparation process
Intel and other chip makers buy their basic materials from specialised producers, who supply them with silicon-based, quartz sand. Silicon, after oxygen, is the second most common element on Earth.
The material is fashioned into silicon cylinders (Monocrystalline Silicon Ingots), 11.8 inches in diameter, which are purified so that there is less than one alien atom per billion. This feat is achieved by gradually melting the cylinder and pulling the impurities in a vertical direction from the bottom to the top.
Imagine a wax candle enveloped in a heating spiral — like the element in an immersion heater. The spiral moves up and down the length of the candle from the bottom to the top and, by gradually melting it, the imperfections get washed out, like skimming the foam off the surface of a soup.
Once an ingot has solidified, it’s cut into 11.8 inch slices. These are then mechanically tested in a vibration box to ensure they will be able to survive being transported, be it by air or land, to the next stage of the production process.
The manufacturing process
During the manufacturing process, the silicon slices have photolithographic masks of the individual layers applied. This is followed by exposure and etching processes (the latter process is repeated multiple times), up to the point where processors with millions, or even billions of transistors, can be etched onto the silicon slice.
These transistors make up the individual flip-flop circuits, which make up a logical circuit. Another level is the programmable processor with a high level of self-intelligence. This, in the context of the manufacture of silicon plates, is still called a die (i.e. a coreless processor with no contact areas.)
The individual processors from the slices undergo further testing. Silicon wafer manufacture is such a delicate process that errors, particularly in separate processors, simply will appear.
High manufacturing cleanliness
Semiconductor production requires strict cleanliness. Inside Intel’s Fab 24, for example, it is 10,000 times cleaner than an operating theatre. A level of pressure is generated inside the fab, which ensures that no impurities are able to get inside — anything undesirable gets blown out through little crevices, particularly at the doors.
The only workers not required to wear protective suits were those of Fab 10, who used to assist in some of the manufacturing operations.
When a higher than usual error rate was recorded in this factory (at that time, the Fab was still using submicrometer technology), analysis showed this was due to pollen grains penetrating the manufacturing area during the spring pollen season.
Just to be sure, Intel introduced a new rule requiring smokers to wash their mouths after taking a smoke break because even smoke particles could potentially increase the error rates.
In Fab 24, there are no people at all. Everything is fully automated. The only human presence is the occasional device tester with a laptop, enveloped in a protective suit.
Semiconductor production also demands that a factory is divided into two halves — a “no copper” and a “copper” half. Copper is considered “poisonous” in silicon production, as it could damage the p-n junctions (the interfaces between different types of semiconductor material) inside a semiconductor.
A human hair has a diameter of 75-100 micrometers (100,000 nanometers). The smallest particle the human eye can see is no bigger than 50 micrometers. The 45 nm technology factory needed to be certified for 0.5 micrometer construction accuracy. The p-n junction width of the 45 nm technology semiconductor is 20 nm, which is less than the size of the influenza virus (including swine flu).
The path to PC manufacturing
Upon testing of the individual cores, the silicon slices are again tested for mechanical endurance. They are packed into a special container and transported to Singapore for another process of production. Here, they get cut into separate processors, tested further, repacked and ultimately dispatched to PC manufacturers.
New 10 nm technologies and 3D transistors
Intel is already utilizing 14 nm technology in its 2015 6th Generation Core processor line-up, with 10 nm chips incorporating 3D transistors already on the horizon. Intel will continue to improve its manufacturing processes, either by building new factories or by retooling existing facilities.
Intel has Fabs situated all around the world to ensure continuous chip production in case of any localised natural disasters. Intel doesn’t start production in its new factories until it achieves more demanding goals in yield, performance, and more.
It should be very interesting to watch how much further the limits on processor manufacturing can be pushed, what their roadmap will look like, and what next steps will Intel’s “tick-tock” strategy follow.