Hewlett-Packard

Contribution to AI

Two conditions had to be met before machine learning could be an experimental practice rather than a theoretical one: computers had to sit next to the experiment, and academic engineering had to be able to turn itself into a business. Frederick Terman's advice to two of his Stanford students — build your own firm rather than take salaried posts back east — settled the second, and the Addison Avenue garage became the template against which every later university spinout was measured. The 2116A of 1966 addressed the first. Built to drive the company's own programmable test instruments, it interfaced with a wide range of laboratory equipment and made the computer a piece of bench apparatus rather than a scarce shared resource; customers found other uses for it immediately, which is how the firm ended up in computing at all. HP Labs later reopened a question the field had left dormant, showing in 2008 that memristance arises naturally in nanoscale devices and so giving Leon Chua's 1971 fourth circuit element a physical form. The dot-product engine work that followed used a memristor crossbar to perform a vector–matrix multiplication in a single analogue step, without shifting weights across a bus, and most subsequent in-memory and neuromorphic accelerator research is an argument with that design. At the opposite extreme of scale, the Cray business absorbed in 2019 produced Frontier, the first machine measured above an exaflop, and El Capitan after it.

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