OUTTAKE: The Heart of the Information Economy
An outtake left on the cutting room floor from the ms. of "Slouching Towards Utopia?: An Economic History of the Long 20th Century
One set of lenses with which to view the technological core of modern economic growth brings into focus General Purpose Technologies: those technologies where advances change, if not everything, a lot, as they ramify across sector upon sector.
Steampower was the first. Early machine tools which embodied in their design and construction so much technological knowledge about how to shape materials was the second. Then came telecommunications, materials science, organic chemistry, internal-combustion engines, the assembly line, subsequent machine-tool generations, and electricity. We know that these are GPTs because, eventually, it made no sense to speak of them as discrete “sectors” apart from the rest of the economy.
After the first two of these, the rest make up Robert Gordon’s “one big wave” of technological advance that he sees transforming the global north over 1870-1980, and then ebbing.
In the 1950s there came another GPT. Electricity was no longer a sector, for it was everywhere. But there emerged something called “microelectronics”: electrons made to dance Not in the service of providing power but rather of assisting and amplifying calculation—and communication.
Take sand, which is mostly very finely grained particles of the rock quartz. Purify it by removing things like pieces of shell. Heat it to more than 1700°C—that is 3100°F. Add carbon. The carbon will then pull the oxygen atoms out of the quartz to make carbon dioxide, and leave behind pure silicon. Cool the silicon to about 1400°C—that is, 2550°F. Then drop a small seed crystal into the barely liquid silicon, and pull up the seed crystal as the surrounding silicon cooling into solidity and attaches itself to it. If you have done this right, you will then have a cylinder of pure monocrystalline silicon. Slice it finely. Each slice is a “wafer” of pure silicon crystal.
A pure silicon crystal will not conduct electricity. Of the 14 electrons associated with each silicon atom, two are tightly bound in the quantum states chemists call the 1s orbital, one doing something that our feeble East African Plains Ape brains metaphorically try to understand by saying that it is “spinning” with its axis of rotation pointing up, one pointing down. (But that name “orbital” is wrong: they really do not really “orbit”, although Niels Bohr a century and more ago thought they did. But he did not have it right. Schrödinger put him straight.)
It is a fundamental property of electrons that they cannot crowd each other into the same quantum state, and so there is only room for two—one spin-up, one spin-down—in the 1s orbital. The next eight electrons must find quantum states with higher energy: two electrons in the 2s orbitals—one spin-up, one spin-down—and six in the 2p orbitals. (Why six? Because p electrons “revolve” in the same sense that all electrons “spin”; because space is three-dimensional tending to revolve from front to back, from left to right, and from side-to-side are different quantum orbital states; and because a spin-up electron can share the rest of its orbital quantum numbers that specify its state with a spin-down one.)
That leaves four electrons remaining. They, in a silicon crystal are in quantum states that chemists call 3sp orbitals. They are all tightly shared by their home nucleus and the nuclei of the four silicon atoms that are its close neighbors in the crystal. These electrons are locked into their quantum orbital states: it requires a lot of energy to knock them out and make them move. Thus silicon cannot conduct electricity. Being a conductor requires that it be easy to knock electrons out of their ground states and into conduction-band orbitals quantum states, in which they are only very loosely bound to any one particular atomic nucleus.
However, suppose you replace one of the silicon atoms in the crystal with a phosphorus atom, with its not fourteen but fifteen electrons. If there are not too many such phosphorus atoms—one in every 10,000 atoms is more than enough—each phosphorus nucleus will simply slot into and fill the place of the silicon nucleus it replaces. Fourteen of the phosphorus atom’s electrons will act like the silicon atom’s electrons: locked into place, Tightly bound in their 3sp orbital to both their home nucleus and to the four neighboring nuclei. But the phosphorus atom’s fifteenth electron cannot fit. It finds a higher energy band in which it is only loosely bound to any one nucleus: the conduction band. Silicon doped with phosphorus thus becomes a conductor. And if you do something to create a current that pulls those 19th electrons away, it becomes a non-conductive insulator instead.
Thus by applying or removing small voltages of electrical current and electromagnetic pressure to a duped region of a silicon crystal, we can turn that region into a switch, and then we can turn the switch on and off as we choose, and let the current flow or not as we choose.
Right now, in the semiconductor fabricators of the Taiwan Semiconductor Manufacturing Company on the island of Taiwan, the machines that it has bought from ASML and Applied Materials and programmed are carving 13 billion such semiconductor solid-state switches with attached current and control paths—we call them —onto a crystal silicone chip About 2/5 of an inch wide and 2/5 of an inch tall. TSMC’s marketing materials imply that the smallest of the carved features is only 25 silicon atoms wide. (In actual fact, the features are more like ten times that size—but, still, that means that there are only about 1250 free electrons able to move easily in a cross section.) These 13 billion transistor switches switch on and off, synchronously, 3,200,000,000 times a second. If all 13 billion components of this small chip of rock made from sand were carved correctly ,and the chip passes its tests, it will then be an Apple M1 microprocessor, like the one at the heart of the machine I am typing these words on.
These small chips of silicon rock doped with phosphorus (and boron, and other things) carved into microscopic paths are the single most essential pieces of our modern information-technology industries.
The electric telegraph is really the start of our information technology industries. In 1774 George-Louis Le Sage used electric wires to successfully send alphabetical signals from one room of his house to another. In 1830 William Ritchie demonstrated that you could Use magnets to send weak information-carrying electrical signals from one end of a lecture hall to another, making the use of electricity for a near-instantaneous long-distance communication device not just a theoretical curiosity but a potentially practical possibility. And in May 1844 Samuel Morse sent his message “what hath God wrought?” over the Baltimore Washington DC telegraph line.
In 1906 Lee De Forest—who claimed that he never understood why it worked, he just knew that it did—built the first triode: the first device in which, without relatively bulky magnets involved, current would or would not flow through a circuit to another depending on whether or not another control voltage was on or off. The era of vacuum tubes and radio and all of its spinoffs plus stereo amplifiers and the earliest electronic computers had begun. But it was solid-state microelectronics that enabled the flowering.
William Shockley, John Bardeen and Walter Brattain are the three credited with the first transistor at Bell Telephone Laboratories in 1947. It was an industrial research lab: that means that in reality hundreds if not thousands contributed. The hanging of individuals’ names on such discoveries and inventions is much more a reflection of our being designed to think in personal narratives then of the scientific and technological reality. Dawon Khang and Mohamed Attila (and the Bell Labs team) built the first metal oxide semiconductor field effect transistor—The one that works so much better than earlier concepts that as of now More have them have been made by humans than any other device by past orders of magnitude. It was Jay Last’s group building on the ideas of Robert Noyce and Jean Hoerni of Fairchild Semiconductor who, in 1960, Built the first operational solid state integrated circuit. By 1964 General Microelectronics was making and selling a 120-transistor integrated circuit.
Vacuum-tube electronic switching elements were four inches—100 millimeters. Transistors in 1964 were packed 1 millimeter apart in early integrated circuits: 100 times smaller, enabling 10,000 times as much computation power to be packed into the same space, with orders of magnitude less power consumption. They were made using a 50μm—50 micron, 1/20 of a millimeter—photolithography process, etching boron- and phosphorus-rich regions onto the silicon substrate to make gates and circuits.
In 1965 Fairchild Semiconductor’s Gordon Moore observed that the number of solid-state microelectronic circuit elements in frontier integrated circuits had grown from 1 to 100 in the seven years since 1958. He made a bold ten-year highly speculative forecast. He looked forward to continued increases in density and a future of “component-crammed equipment”, projecting a 1975 in which a 100 square-millimeter silicon chip would have 65,000 components on it. That would allow for “electronic techniques more generally available throughout all of society, performing many functions that presently are done inadequately by other techniques or not done at all”. Moore forecast, back in 1965:
Home computers—or at least terminals connected to a central computer—automatic controls for automobiles, and personal portable communications… [plus] integrated circuits in digital filters [to] separate channels on multiplex equipment… telephone circuits and… data processing. Computers will be more powerful, and will be organized in completely different ways… <https://newsroom.intel.com/wp-content/uploads/sites/11/2018/05/moores-law-electronics.pdf>
By 1971 integrated-circuit semiconductor fabricators had taken four steps downward to a finer, 8μm lithography process. The first microprocessor, the Intel 4004, packed 20,000 transistor into a square milimeter—features were 200 microns apart. By 1985 the process node in the semiconductor fab was down to 1.5μm, and the Intel 80386 microprocessor had shrunk that average feature-plus-separation distance down to 20. In 2005 the process node size was down to 90nm—nanometes—and the Intel Pentium 4’s average feature-plus-separation distance was down to 0.7—700 nanometers. Today’s Apple-TSMC M1 packs 16 billion transistors onto a 133 square-millimeter silicon die, with an average feature-plus-separation distance of 90 nanometers, and TSMC’s labelling the process as the equivalent of 5nm. The 90 nanometers of average feature-plus-separation distance are only 450 silicon atoms across. Today’s microelectronics are thus, in a sense, 115,000 times smaller than the transistors of 1985, and can pack thirteen billion times as many possible switching elements into the same space.
In 1979 to execute 1 MIPS—million instructions per second—required 1 watt of power. By 2015 1 watt could drive more than 1,000,000 MPIS. As components became smaller, they became faster, at least up until the late 2000s. Halve the size of the feature, and you can run it twice as fast—up to a point. Before 1986 microprocessor clock speed quadrupled every seven years. Then with the coming of the simplicity of reduced instruction sets came seventeen years in which each quadrupling of speed took three years rather than seven. Then, after 2003, the quadrupling time went back to seven years, until further speed improvements hit a wall around 2013.
At its most rapid pace during the information-technology revolution, the company at the heart of the innovation economy, microprocessor designer and manufacturer Intel, was tick-tocking—tick, improving the microarchitectural details of its microprocessors so programs can run faster; tock, improving the fine resolution of its manufacturing so that it can make the features and thus the microprocessor smaller—and completing a full cycle in under three years. With microprocessors doubling in speed every two years, and with the information-technology sector taking full advantage, measured economy-wide productivity growth after 1995 rose again, and came close to its golden-age immediate post-WWII pace—until the Great Recession disruption came in 2008…
(Remember: You can subscribe to this… weblog-like newsletter… here:
There’s a free email list. There’s a paid-subscription list with (at the moment, only a few) extras too.)