Microprocessor

A complete central processing unit — the part of a computer that fetches instructions, performs arithmetic and logic, and directs everything else — shrunk onto a single integrated-circuit chip. Before it, a processor was a cabinet of wired-together circuit boards; the microprocessor put that whole apparatus on a fingernail of silicon that could be stamped out by the thousand and reprogrammed, in software, to do almost anything.

Computation stopped being a room and became a component — a part you could buy for a few dollars and drop into anything with electricity. Because one general-purpose chip could be reprogrammed rather than rewired, a single mass-produced design could run a calculator, a car engine, a phone, or a spacecraft. Anything with power could now also think a little, and the silicon economics that had made transistors cheap now made whole computers cheap.

In 1969 the Japanese calculator maker Busicom hired Intel to build a custom twelve-chip set for a desktop calculator. Intel's Ted Hoff judged the design too complex and proposed the reverse: one general-purpose programmable processor whose behavior lived in software rather than in fixed wiring. Federico Faggin — applying the silicon-gate MOS technique he had helped pioneer at Fairchild, with Stan Mazor and Busicom's Masatoshi Shima — turned that idea into working silicon across 1970–71, packing roughly 2,300 transistors onto a 4-bit chip. Intel released it as the 4004 in November 1971, after buying back the commercial rights from Busicom.

The personal computer, the embedded systems that now hide inside almost every powered object, and eventually the servers and routers that form the internet's physical substrate. Each generation put more of the same idea onto denser integrated circuits, riding Moore's-law scaling from 2,300 transistors to billions. The smartphone — and the data-center hardware that trains modern neural networks — sits at the far end of the line that begins with the 4004.