Remembering Magnetic Memories and the Apollo AGC

Remembering Magnetic Memories and the Apollo AGC

4th March 2026

Memories, both human and magnetic

In the 1960s NASA engineers made a technical decision that would define the relationship between astronaut and spacecraft: when complex decisions, precise coordination and fast reactions were required, how should a spacecraft be controlled? They decided, in anticipation of developments in so many areas of society, that the best results would be achieved with digital execution subject to human oversight; the human brain would be paired with an artificial memory comprised of wires and magnets.

The evolution of digital computing in the middle of the 20th century placed challenging demands on memory technologies. Engineers required ever-more memory at a lower cost, faster access speed, higher density and with greater reliability than previous technologies. This provided the conditions for a creative period of iterative invention and development of magnetic memory technologies which slowed only when semiconductor-based memories came to dominate from the 1970s.

It's these magnetic memories that I decided to explore, understand (at least superficially) and illustrate.

Computers in spaceflight

In 1962, President John F. Kennedy famously committed NASA’s Apollo mission to landing a man on the moon before 1970. As NASA engineers scrambled to turn this objective into an achievable technical program, they came to understand the complexity of the task. Spacecraft would exceed the speed of sound, entering a new aerodynamic paradigm – and that before exiting the Earth’s atmosphere entirely and travelling by gravitational influences and thrust reaction alone. Astronauts could simply not react fast enough or in such abstract terms to control a craft travelling over 11 km/s and accurately and fuel-efficiently reach a target 384,000 km away.

Flying such a craft, they reasoned, would be more akin to extremely rapid control-feedback and solving complex equations than piloting a conventional plane. Even the lunar landing, a short period of manual control, was in fact mediated by the guidance computer, which operated feedback loops between the pilot’s controls, sensors and the craft’s actuators.

Assigning these tasks to the guidance computer led to the development of a specification; the Apollo Guidance Computer (AGC) needed to be able to make observations about the spacecraft’s state (e.g. velocity, position) at a suitable rate and then apply known information and calculations to decide which actions should be made to keep the craft on course.

The 'DSKY' (pronounced diss-key) or display and keyboard was the main user interface for the apollo AGC. To execute a command the astronaut entered a 'verb' using a 2-digit code, followed by a 'noun' (also 2-digits).

The AGC therefore needed to store the data and programs required to complete mathematical calculations and to interface with sensors, actuators and the user interface. Executing a program thus required read-only memory (ROM) to be loaded and commands to be executed using sensor data – which had to be held in temporary random-access “erasable” memory.

What made the AGC’s memory cores so specialised was not so much the quantity of data it stored or the calculations it would run but the constraints that applied to it. The memory needed to operate reliably while withstanding vibration, temperature and radiation, all while being as physically small and lightweight as possible.

Remembering Magnetic Memory

• Memories can be read/write or read-only. Read/write (or “erasable” in NASA terminology) memories are those which permit data to be changed/re-written as part of normal use. Read-only memory (ROM) cannot be easily changed.
• Memories can be sequential or random access. If data has to be read in address sequence the data is sequential, otherwise if data can be read regardless of where in the memory it is stored the technology allows random access.
• Memory can be volatile or non-volatile. Volatile memory is lost when the system loses power, non-voltage memory is retained on power-off.
• Memory can feature a destructive-read, where data is removed from memory when read and must be re-written if still of use.
• In terms of other factors, the 1984 Intel Memory Components Handbook summarises an engineer's requirements as access time, cost, memory size, power consumption and environmental considerations.

TROS (Transformer Read Only Storage)

Esoteric rating: hobby project Complexity: punch-cards on PCBs Read-only Random access

When was it developed, and by who? First developed by T. L. Dimond in 1945, Bell Laboratories. Used by IBM in their early mainframe computers. It surely complemented IBM’s punch-card computing machines by retaining a similar programming workflow.

How does it work? A single conductor encodes a word by passing inside or outside numerous ferrites (the number of ferrites determining the word length), each with a sensing coil looped around to form a transformer. When a particular conductor (corresponding to a specific memory address) is energised an output current is induced on each ferrite that the conductor passes through, signifying a “1” bit. Ferrites that the conductor passes outside of do not have a current induced and thus signify a “0” bit.

In IBM’s implementation for the System/360 mainframe computer the ferrites are rectilinear “U” shapes and arranged in pairs in two rows, with a bar fixed on top to complete the flux loop. This allows the ferrites to be conveniently opened.

Data was encoded on mylar (a trade-name of PET, a useful electrical insulator that would have been fairly novel at the time, being patented in 1948). The mylar sheets had copper tracks which passed both inside and outside rectangular cutouts for the cores.