The integrated circuit: printing circuits with light

The full story

This is the episode's narration, word for word. Headings jump to that point in the video.

The question 0:00

Last time, we met the MOSFET: a switch whose gate holds it on or off for almost no power, and which, for decades, got better as it shrank. So here's the next problem. By 1958, a useful machine needed thousands of parts, all wired together by hand. Modern chips need vastly more, and nobody could solder that many. So how do you put them together? The answer is that we stopped building circuits and started printing them, with light.

The tyranny of numbers 0:34

In 1958, a Bell Labs vice-president, Jack Morton, described the problem in print. Digital systems needed hundreds, thousands, sometimes tens of thousands of devices, and each one had to be made, tested, shipped, and then wired in, one at a time. Such systems, he wrote, suffered from what had been called the tyranny of numbers. Every hand-soldered joint is another chance to fail, and a single bad joint could knock out a whole module. More parts meant more wiring, and more wiring meant less reliability.

One material for everything 1:13

That summer, an engineer called Jack Kilby had just joined Texas Instruments, and most of his colleagues were away on the company holiday. Working in a quiet lab, he had a simple, radical idea. Everyone else was shrinking separate parts. Kilby asked: why not make every part, not just the transistor, out of the same piece of semiconductor? A resistor can just be a strip of doped material. A capacitor can be a junction, like the ones from episode three.

A sine wave on a sliver 1:46

On the twelfth of September 1958, Kilby showed his managers a bar of germanium about eleven millimetres long and under two millimetres wide. On it, he'd etched a transistor, a capacitor and resistors, wired up as an oscillator. He pressed a switch, and a steady wave appeared on the oscilloscope. It was the first working integrated circuit.

The flying wires 2:13

But look closely at the famous photo, and you'll see the catch. The parts were connected by fine gold wires, arching over the surface, bonded on by hand. The components were integrated. The wiring wasn't. So the tyranny of numbers had been shrunk, not beaten, and the Computer History Museum calls the approach impractical for mass production. To beat it, the wiring itself had to become part of the chip.

Leave the glass on 2:43

That came from Fairchild Semiconductor. Transistors then were made as little plateaus, with their delicate junctions exposed at the edges, and the oxide layer was normally stripped off. The physicist Jean Hoerni did the opposite. He left the oxide on. Dopant goes in through a window in the oxide, but it also spreads a little sideways, so the junction comes to the surface tucked under the glass. Sealed in from the start, the junctions leaked far less. With everything on one flat face, he called it the planar process.

Print the wiring too 3:21

Fairchild's patent lawyer pushed them to find more uses for it, and in January 1959 Robert Noyce wrote down the big one: make many parts on one piece of silicon, and make the connections as part of manufacturing. The trick, spelled out in his patent that July: if the oxide is an insulator, you can lay metal strips right on top of it. Open small windows wherever a wire needs to touch the silicon, and the metal can cross over a junction without shorting it. In practice the metal was aluminium, and it was deposited, not soldered. The wiring had become one more layer you could print.

Keeping neighbours apart 4:03

In 1960, Jay Last's team at Fairchild got the first monolithic chips working. One was a flip-flop, with four transistors and five resistors on a single piece of silicon. One more problem: put lots of parts in one slab of silicon, and they're all touching through it. Fairchild settled on an idea from Kurt Lehovec at Sprague Electric: surround each part with a junction held in reverse, so it blocks current, like the one-way street from episode three.

Who invented it? 4:37

So who invented the integrated circuit? Kilby filed his patent in February 1959, and Noyce in July. Their companies fought for about a decade. The case turned on wording: Kilby's patent said metal could be laid down on the insulation, and Noyce's said it adhered to the oxide. In 1969 an appeals court sided with Noyce. But by then it hardly mattered, because in 1966 the two companies had agreed to license each other.

Shared credit 5:11

History now credits both. Kilby built the first working circuit, and Noyce made it something you could manufacture. In 2000, Kilby shared the Nobel Prize in Physics for his part in the invention. Noyce couldn't share it. He had died in 1990.

Turning the microscope around 5:30

Now, the printing. At a US Army lab in the mid-1950s, Jay Lathrop had spent years using microscopes to make small things look big. He and James Nall turned that around, using the optics to make a big pattern small. They coated germanium with a light-sensitive chemical from Kodak, the camera company, and named the process photolithography: printing with light. At Bell Labs, Jules Andrus and Walter Bond had been adapting printing-industry tricks to silicon since 1955.

Coat 6:06

Here's how it works, step by step. Start with a wafer wearing that layer of silicon dioxide. Drop on a liquid called photoresist, and spin the wafer at a few thousand revolutions a minute, so it spreads into a thin, even film. This is why the lithography rooms in chip factories glow yellow: the resist reacts to ultraviolet and blue light, so those colours are filtered out of the room lights.

Mask and expose 6:35

Next comes the mask: a plate of quartz with a pattern in chromium, like a photographic negative. Shine violet and ultraviolet light through it. Where the chrome blocks it, the resist is untouched. Where the light gets through, the resist changes chemically, and the most common kind becomes soluble.

Develop, etch, dope 6:57

Then you develop it, like a photo, and the soluble resist washes away, leaving a stencil. An etch eats through the oxide wherever it's bare, opening precise windows. Strip off the rest of the resist, and now the oxide itself is the stencil. Dope through the windows, exactly as in episode two, or fill them with metal. Light only changed the resist. Chemistry did the cutting.

Layer by layer 7:26

Then you do it again. Each layer, whether it's a doped region, a gate or a level of wiring, gets its own mask, lined up precisely on the layers beneath. A complex chip can go through this cycle as many as fifty times. And every pass prints every transistor on that layer together, not one by one. Printing a thousand connections takes no more steps than printing one. That's how the tyranny of numbers was finally broken.

How small can light draw? 7:57

So how small can you print? Light spreads out when it squeezes through a tiny gap, so edges blur once features get near the wavelength. The rule of thumb says the smallest feature is proportional to the wavelength, divided by how much light the lens can gather. That's why chipmakers keep moving to shorter wavelengths. First came mercury lamps at four hundred and thirty-six and three hundred and sixty-five nanometres, then ultraviolet lasers at two hundred and forty-eight and one hundred and ninety-three. Lathrop's first strips were about two hundred micrometres wide; features below a tenth of a micrometre are now routine. And the next step, extreme ultraviolet at thirteen and a half nanometres, is a story for episode ten.

Wafer to die 8:47

You don't print one chip at a time. The machine prints the same pattern again and again across the wafer, stepping from one patch to the next. Each copy is called a die. Before the wafer is cut, needles probe every die, and the failures are marked. Then a diamond saw, cooled with water, slices along thin lanes between them.

Why big chips cost more 9:13

And not every die works. A speck in the wrong place usually kills the die it lands on, so a bigger die is more likely to catch one. In one university course's worked example, quadrupling the die area drops the share that work from about eighty-five per cent to about half. Fewer big dies fit on the wafer, too, and the edges waste more. So the cost of each good chip goes up about seven times, not four.

Why dust matters 9:42

Which is why fabs are obsessed with dust. A particle on the wafer casts a shadow during exposure, or blocks dopant atoms on their way in. People shed skin and salt, so workers dress head to toe. Ordinary room air can hold tens of millions of dust specks per cubic metre. A typical cleanroom standard allows just three and a half thousand, and the cleanest classes far fewer. That's about ten thousand times cleaner than room air.

Moore's observation 10:14

In 1965, Gordon Moore noticed that the number of components on a chip was doubling every year, and guessed it might reach sixty-five thousand by 1975. In 1975 he revised it to doubling every two years. Someone else, Carver Mead, named it Moore's law. It was never a law of physics. It was an observation about what printing with light, and money, could keep doing.

What do you do with them? 10:45

So now we can print switches by the thousand, then by the million, and wire them up in the same breath. But a switch only knows on or off. How do you get on-and-off switches to add, compare and decide? That's logic: gates, CMOS and the 555 timer. That's next time.

Sources

Every factual claim in the episode is tied to one of these. Spotted an error? Tell us.

1. IEEE / ETHW: "Milestones: First Semiconductor Integrated Circuit (IC), 1958" (plaque;… — ethw.org
2. G. Freeman, "Invention of the Integrated Circuit", IEEE Pikes Peak Section talk (May… — site.ieee.org
3. Wikipedia: "Tyranny of numbers" — en.wikipedia.org
4. National MagLab, Magnet Academy: "Jack Kilby" — nationalmaglab.org
5. Computer History Museum, Silicon Engine: "1958: All Semiconductor 'Solid Circuit' is… — computerhistory.org
6. Computer History Museum, Revolution: "The Integrated Circuit" (Dummer 1952 quote;… — computerhistory.org
7. IEEE Spectrum: "July 1958: Kilby Conceives the Integrated Circuit" — spectrum.ieee.org
8. Wikipedia: "Invention of the integrated circuit" — en.wikipedia.org
9. Wikipedia: "Jack Kilby" (Nobel citation; demo description) — en.wikipedia.org
10. J. S. Kilby, US Patent 3,138,743, "Miniaturized Electronic Circuits", filed 6 Feb… — patents.google.com
11. Computer History Museum, Silicon Engine: "1959: Invention of the 'Planar'… — computerhistory.org
12. J. A. Hoerni, US Patent 3,025,589, "Method of Manufacturing Semiconductor Devices",… — patents.google.com
13. Computer History Museum, Silicon Engine: "1959: Practical Monolithic Integrated… — computerhistory.org
14. Computer History Museum blog: "The Relics of 'St. Bob'" (Noyce notebook 23 Jan 1959;… — computerhistory.org
15. R. N. Noyce, US Patent 2,981,877, "Semiconductor Device-and-Lead Structure", filed 30… — patents.google.com
16. Semiconductor History Museum of Japan (Shimura): "Kilby vs. Noyce" — shmj.or.jp
17. Nobel Prize Outreach: "The Nobel Prize in Physics 2000" summary (Kilby "for his part… — nobelprize.org
18. Linda Hall Library, Scientist of the Day: "Jack Kilby" — lindahall.org
19. MIT Technology Review (23 June 2023): "These chip-patterning machines will shape the… — technologyreview.com
20. Computer History Museum, Silicon Engine: "1955: Photolithography Techniques Are Used… — computerhistory.org
21. Arizona State University NanoFab: "Techniques: Lithography" — cores.research.asu.edu
22. MicroChemicals GmbH: "Lithoprotect yellow-light filters" (‹500 nm blocked) — microchemicals.com
23. Wikipedia: "Photolithography" — en.wikipedia.org
24. TU Wien, Institute for Microelectronics: R. Minixhofer, PhD thesis, §"Lithography"… — iue.tuwien.ac.at
25. ASML: "Lithography principles: the Rayleigh criterion (CD = k1 • λ / NA)" — asml.com
26. Halbleiter.org: "DUV vs. EUV" — halbleiter.org
27. Wikipedia: "Wafer testing" (probe card; ink dot → wafermap; before packaging) — en.wikipedia.org
28. Wikipedia: "Die preparation" (dicing; diamond saw; scribe lines) — en.wikipedia.org
29. M. Mudawar, KFUPM COE 501 Computer Architecture, "IC Manufacturing, Cost, Power, and… — faculty.kfupm.edu.sa
30. NPTEL (Govt. of India), course 103106075, Lecture 36 "Yield" (Poisson model; area vs… — archive.nptel.ac.in
31. Halbleiter.org: "Wafer cleaning" (particles shadow exposure, block implantation;… — halbleiter.org
32. imec: "Inside the cleanroom" — imec-int.com
33. Cleanroom.net: "ISO 14644-1:2015 classes of air cleanliness by particle concentration" — cleanroom.net
34. Production Automation Corp (PAC): "ISO 14644-1 Cleanroom Classifications" (vendor… — gotopac.com
35. Computer History Museum, Silicon Engine: "1965: 'Moore's Law' Predicts the Future of… — computerhistory.org

Researched and scripted with AI assistance, fact-checked claim by claim, with synthetic narration and diagrams drawn in code. How we make episodes.