Fairchild Semiconductor

Contribution to AI

The problem the planar process solved was manufacturing, not logic. Jean Hoerni's 1959 method — oxidise a silicon wafer, open windows in the oxide photographically, diffuse dopants through them — meant a transistor was made by patterning a flat surface rather than by handling a part, and Robert Noyce realised days after seeing it that the same oxide could carry evaporated metal wiring between neighbouring devices. That is the ancestry of every chip an AI system runs on: not the first integrated circuit, which Jack Kilby had built at Texas Instruments months earlier, but the first one that could be printed in batches. Micrologic reached the market in 1961 at prices only missile and Apollo work would pay; by 1965 the company was selling chips for a dollar, below what they cost to build, on the bet that volume would follow, and it did. Gordon Moore, running the research laboratory, drew a line through five data points in an April 1965 article for Electronics and produced the doubling curve that became the industry's planning assumption — the reason anyone could sensibly promise that a computation too expensive this year would be cheap in five. The other output was people. Staff left to found Intel, AMD, National Semiconductor and hundreds of lesser firms, and two alumni started Kleiner Perkins and Sequoia, so the valley took both its silicon and its financing habits from a single payroll.

Drafted with AI and edited by hand (claude-opus-5, reviewed 2026-08).