Magnetic Core Memory

Core memory works because of one unusual material property. Measure it before relying on it.

1. Wind a few turns of wire through a ferrite toroid as a drive winding, and a few more as a sense winding.
2. Drive it with a slow triangular current and plot drive current against the integrated sense voltage on the oscilloscope in X-Y mode.
3. You are drawing the hysteresis loop of the material.
4. Compare a ferrite intended for memory with an ordinary transformer core.

The memory ferrite gives a nearly RECTANGULAR loop: below a threshold current almost nothing happens, and above it the core flips fully. An ordinary core gives a lazy S-shape.

That squareness is what makes half-current selection possible. Half the switching current must do essentially nothing, or every core on the driven row would flip. A material with a rounded loop would half-flip hundreds of cores and destroy the data.

Build the smallest plane that demonstrates coincident selection. It is fiddly and it is the whole point.

1. Mount sixteen cores in a 4 by 4 grid, each tilted 45° so the wires can pass cleanly.
2. Thread an X wire through every core in each row, and a Y wire through every core in each column.
3. Thread a single SENSE wire diagonally through all sixteen.
4. Drive one X and one Y line with half the switching current each and observe the sense wire.

Only the core where the two driven wires cross receives full current and flips; every other core on those lines sees half and stays put. The sense wire picks up a voltage pulse from the one core that switched.

Core has one property that nothing has beaten, and it kept it flying for decades.

1. Consider what physically stores the bit: the direction of magnetisation of a ferrite ring.
2. Now consider what a passing charged particle does to that.
3. Compare with a semiconductor cell, where the bit is a small packet of charge on a capacitor or a node.

A cosmic ray can dump enough charge into a semiconductor cell to flip it; it cannot realign a magnetised ferrite ring. Core memory is essentially immune to single-event upsets, and it is non-volatile, so a power transient loses nothing.

That is why Apollo used core, why the Space Shuttle's early computers used it, and why some spacecraft carried it long after it was commercially extinct. The Apollo Guidance Computer's programs were in core ROPE memory — wires woven through or around cores by hand, unalterable once woven, and utterly reliable.