From a Sliver of Germanium to a Phone: Four Fissions of the Chip

In 1947 computing still ran on vacuum tubes — ENIAC used 18,000 of them: hot, burning out, filling whole rooms. Bell Labs' solid-state physics group, led by theorist William Shockley (37), aimed to build a solid amplifier needing no heated filament. Two doers anchored it: theorist John Bardeen and master experimentalist Walter Brattain.

On December 16, 1947, Bardeen and Brattain — using two gold contacts placed extremely close on a germanium crystal — produced the first device that could amplify a signal, the "point-contact transistor," demonstrated to management on the 23rd. Shockley was proud yet jealous: the breakthrough wasn't his own hands' work. Within weeks, working alone, he designed the sturdier, more manufacturable junction transistor. In 1956 the three shared the Nobel Prize in Physics.

The same tension replays today: whether a foundation model ships as open weights or a closed API decides not just one firm's moat but the evolutionary speed of an entire ecosystem.

Shockley returned to California in 1956 to found Shockley Semiconductor, recruiting top young talent. But he was paranoid, suspicious, and chaotic as a manager. In September 1957, eight core researchers — led by Robert Noyce (29) and Gordon Moore — resigned en masse. Shockley branded them the "traitorous eight."

The eight secured backing from Fairchild Camera and founded Fairchild Semiconductor. Two breakthroughs followed: in 1959 Jean Hoerni invented the planar process, letting transistors be mass-printed onto a silicon surface; almost simultaneously, Noyce conceived of wiring multiple components onto one slab of silicon — the integrated circuit (arrived at independently and near-simultaneously by Jack Kilby at Texas Instruments). A single chip could now hold an entire circuit.

The strongest organizations may not retain people but keep "fissioning" out new ones. From the PayPal Mafia to the recent OpenAI exodus, high-frequency talent flow is itself an engine of innovation.

In 1965, while still at Fairchild, Gordon Moore predicted in a short article that the number of components on a chip would double every year (later revised to roughly every two). This became Moore's Law. In July 1968, Noyce and Moore left Fairchild too, founding Intel.

In 1971 Intel engineers Ted Hoff and Federico Faggin built the 4004 — the first commercial microprocessor, squeezing a computer's central processing unit onto a chip the size of a fingernail. Moore's Law then became the industry's metronome: equipment makers, design houses, and customers all planned and drew roadmaps along that curve — an empirical observation turned into an industry-wide self-fulfilling promise.

AI's "scaling laws" play a similar role today — at once an observation and the shared belief that lets the whole industry dare to bet hundreds of billions.

As process nodes approached physical limits, competition shifted to another axis: the instruction-set architecture — the language a chip "understands." In 1990, Britain's Acorn, with Apple and VLSI, founded ARM, specializing in low-power reduced-instruction-set (RISC) design. ARM builds no chips; it licenses the blueprints.

This "asset-light" model met two big events: in 1987 Morris Chang founded TSMC, pioneering the pure foundry and fully separating design from manufacturing; and the mobile era arrived, where power efficiency trumped all, putting ARM's low-power architecture into nearly every phone. In 2010, David Patterson's team at UC Berkeley released RISC-V — a fully open-source, free-to-use instruction set that directly challenged the licensing walls of ARM and Intel.

This mirrors the "control plane vs. data plane" split in distributed systems — real power moves from execution (making chips) to the party that defines the protocols and interfaces (the architectural standard).