Every modern computer stores information in silicon chips, microscopic transistors etched onto wafers of crystalline material. The process is so refined, so miniaturized, that a chip the size of a fingernail can hold billions of bits. But the first electronic computers had no silicon. They had no transistors. When the engineers building the earliest stored-program computers needed a way to hold data in electronic form, they turned to an unlikely material: liquid mercury.
The story of computer memory is a story of extraordinary ingenuity under severe constraints. The engineers of the 1940s and 1950s knew exactly what they needed (fast, reliable, rewritable storage) but had almost nothing to build it with. The solutions they invented, including mercury delay lines, cathode ray tubes, magnetic drums, and magnetic cores, were brilliant improvisations that kept computing alive until semiconductor technology matured. Each technology had strengths and weaknesses, and each left its mark on how we think about memory today.
The Problem Von Neumann Identified
In his 1945 “First Draft of a Report on the EDVAC,” John von Neumann described the architecture of a stored-program computer: a machine that holds both its instructions and its data in the same memory. The architecture was elegant in theory, but it created an immediate practical problem. The computer needed a memory device that could store thousands of numbers, retrieve any one of them in microseconds, and allow the stored values to be changed at will.
In 1945, no such device existed. Vacuum tubes could store a single bit each but were expensive, power-hungry, and unreliable (a computer with thousands of vacuum tube memory cells would burn out constantly). Mechanical devices (relays, wheels, drums) were too slow. Photographic film was permanent and could not be rewritten. Punched cards could store data but not at electronic speeds.
The solution came from an unlikely source: wartime radar technology.
Sound Waves in Mercury
During World War II, radar engineers needed to compare successive radar pulses to distinguish moving targets from stationary ones. They developed the mercury delay line, a device that stored an electrical signal by converting it into a sound wave, passing the sound wave through a tube of liquid mercury, and converting it back into an electrical signal at the other end. Because sound travels much more slowly than electricity, the signal was delayed by a predictable amount of time. The delayed signal could then be compared with the next incoming pulse.
J. Presper Eckert, one of the designers of ENIAC and EDVAC, recognized that this delay could be used for memory. If a train of electrical pulses (representing binary data) was converted into sound waves and sent through a mercury tube, the data would take a fixed amount of time to travel from one end to the other. At the receiving end, the signal could be amplified, reshaped, and fed back into the input end, creating a continuous loop. The data would circulate endlessly through the tube, preserved as long as the circuit remained active.
A single mercury delay line tube, about 1.5 meters long and a few centimeters in diameter, could store approximately 1,000 bits of data. The EDVAC used 128 mercury delay lines, giving it a total memory of about 1,024 words of 44 bits each. By the standards of the time, this was enormous.
Elegant but Impractical
Mercury delay lines worked, but they had serious drawbacks:
- Mercury is toxic. Handling it, filling tubes with it, and cleaning up spills were occupational hazards that the early computer builders accepted with a nonchalance that would horrify modern safety engineers.
- Temperature sensitivity. The speed of sound in mercury changes with temperature, which means the timing of the delay changes. The tubes had to be maintained at a constant temperature, often by immersing them in temperature-controlled baths.
- Sequential access only. Data circulating in a delay line could only be read when it arrived at the output end. If the bit you needed had just passed, you had to wait for it to circulate all the way around again. This made access times unpredictable, ranging from nearly instantaneous (if the desired data happened to be at the output) to the full circulation time (about 1 millisecond).
- Physical bulk. The mercury tubes were heavy, fragile, and took up considerable space. A memory system based on delay lines was a piece of industrial plumbing as much as an electronic device.
Despite these limitations, mercury delay lines were used in several pioneering computers: the EDVAC (completed 1951), EDSAC at Cambridge (1949), UNIVAC I (1951), and the ACE at the National Physical Laboratory (based on Alan Turing’s design). For a few critical years, mercury was the material that made stored-program computing possible.
The Williams Tube: Memory on a Screen
A competing approach emerged from the University of Manchester, where Frederic Williams and Tom Kilburn developed a way to store data on the screen of a cathode ray tube (CRT), the same technology used in television sets and oscilloscopes.
The Williams tube exploited a subtle physical effect. When an electron beam strikes a phosphor-coated screen, it creates a small charged spot. The charge pattern can be detected by a metal plate placed near the screen. By writing dots and dashes at specific positions on the screen, the Williams tube could store binary data as a pattern of charges. The data could be read by detecting the charge at each position and refreshed by periodically rewriting it (the charges faded over time).
The great advantage of the Williams tube was random access. Unlike a mercury delay line, where data had to circulate to the output, a Williams tube could read any bit immediately by directing the electron beam to the appropriate position on the screen. This made it much faster for programs that needed to access data in unpredictable patterns.
The Manchester Baby (officially the Small-Scale Experimental Machine), which ran the first stored program on June 21, 1948, used a Williams tube for its memory. It stored just 1,024 bits (32 words of 32 bits each), but it proved the concept. The Manchester Mark 1 and its commercial descendant, the Ferranti Mark 1 (1951), used Williams tubes as their primary memory. IBM’s early scientific computers, including the 701 and 702, also used Williams tubes.
The disadvantages were reliability and maintenance. Williams tubes were sensitive to electrical interference, temperature changes, and the gradual degradation of the phosphor coating. They required constant adjustment and recalibration. Engineers spent significant time tuning their CRT memories to keep them functioning correctly.
Magnetic Core: The Memory That Conquered Computing
The technology that finally solved the computer memory problem was magnetic core memory, developed in the late 1940s and early 1950s by several researchers, most notably Jay Forrester at MIT.
Magnetic core memory used tiny rings (cores) of ferrite material, each about 1 millimeter in diameter, threaded onto a grid of wires. Each core could be magnetized in one of two directions, representing a 0 or a 1. By sending electrical currents through the wires threading a particular core, the computer could set or read its magnetic state.
Core memory had extraordinary advantages over its predecessors:
- Non-volatile. Unlike mercury delay lines and Williams tubes, core memory retained its data when the power was turned off. The magnetic state of each core persisted indefinitely.
- Random access. Any core could be read or written independently, without waiting for data to circulate.
- Reliable. Ferrite cores were physically robust, insensitive to temperature, and did not degrade over time. A core memory plane could operate for years without maintenance.
- Fast. Access times of a few microseconds were typical, fast enough for the computers of the era.
The disadvantage was manufacturing complexity. Each core had to be threaded by hand onto a grid of wires, a painstaking process performed (primarily by women workers in electronics factories) under magnifying glasses. A memory plane containing thousands of cores required thousands of individual threading operations. Despite this, core memory was so superior to its alternatives that it dominated computing from the mid-1950s through the early 1970s.
The Apollo Guidance Computer, which navigated the spacecraft to the Moon, used magnetic core memory for both its erasable memory (2,048 words) and its fixed memory (36,864 words of rope core). When astronauts landed on the Moon in 1969, their computer’s memory was woven from tiny magnetic rings on copper wires.
From Cores to Chips
Magnetic core memory was finally displaced by semiconductor memory (RAM) in the early 1970s. Intel’s 1103 dynamic RAM chip, introduced in 1970, could store 1,024 bits on a single silicon chip. It was smaller, cheaper (in mass production), faster, and easier to manufacture than core memory. Within a few years, semiconductor RAM had replaced core memory in virtually all new computers.
The transition happened with remarkable speed. In 1970, core memory was the standard. By 1975, it was obsolete. The semiconductor revolution, driven by Moore’s Law (the doubling of transistor density approximately every two years), has continued ever since. A modern smartphone contains more memory than all the computers in the world possessed in 1960, and it fits in your pocket.
The EDVAC’s Legacy
The mercury delay lines that gave the EDVAC its memory are long gone, replaced by technologies that Eckert and von Neumann could not have imagined. But the architecture they served, the stored-program design described in von Neumann’s 1945 report, survives in every computer built today. The Kronecker Wallis edition of the EDVAC Report, printed on blue Fabriano paper with metal screw fasteners, is a physical echo of the industrial era in which tubes of toxic liquid mercury were the most advanced information storage technology on Earth.
The history of computer memory is a reminder that great ideas often arrive before the technology needed to implement them. Von Neumann’s stored-program architecture required a fast, reliable, rewritable memory. In 1945, no adequate memory technology existed. Engineers improvised with mercury, phosphors, and ferrite rings, each solution good enough to keep the idea alive until something better came along. The architecture outlasted all of them. It is still waiting, patiently, for whatever comes next.
No products in the cart. 