Inside the magnetic core memory of Spacelab's French-built flight computer

Ken Shirriff dissects the core memory of the Mitra 125 MS, the French minicomputer that ran the European-built Spacelab laboratory flown in the Space Shuttle’s cargo bay. Rather than trust the Shuttle’s own IBM AP-101 systems, Spacelab carried three identical Mitra machines — one for the lab, one for experiments, and one hot spare — each holding just 128 KB of storage. That storage was not silicon but magnetic core: a bit per ferrite ring, packed into a seven-board stack (two driver boards, four core-plane boards, and an interface board) that formed roughly a third of the computer and bolted to a detachable side panel for conduction cooling.

The piece doubles as a clear primer on how core memory actually worked and why it dominated the 1950s–1970s. Coincident-current addressing let a grid of X/Y wires select a single core by exploiting magnetic hysteresis — each wire carried half the current needed to flip a bit, so only the intersection switched. Reads were destructive, requiring a rewrite cycle, and inhibit lines let individual planes in a stacked word hold their value. A diode matrix made the whole thing economical, letting N drivers per side address N⁴ cores while diodes blocked current from sneaking through unintended paths.

The history is a reminder of how contested this now-obsolete technology once was. Shirriff credits Jay Forrester (Whirlwind, 1953) as the key figure but notes the tangle of competing patents that forced IBM to pay $400,000 to An Wang — seed money for Wang Laboratories — and $13 million to MIT. For technical readers, the article is a well-illustrated look at the electromechanical ingenuity that kept spaceflight-grade computers running before flash and DRAM.