Magnetic Drum and Core Memory (pre-RAMAC storage)

Between 1946 and 1956 computer storage was built from two magnetic ideas that were not disks: the drum, a rotating cylinder with one fixed read/write head for each track, and the core, a ring of ferrite switched by currents in crossing wires. The Computer History Museum dates the first working drum stores to 1946 (Booth, London) and 1947 (ERA, Minnesota), the first delivery of a drum computer to 1950, and the first demonstration of core memory to August 1953. IBM's RAMAC of September 1956, usually treated as the first disk drive, still contained a drum and a core register in its processing unit, so the disk stack took over bulk storage rather than replacing every earlier memory.

1932 and the patent record. The Museum says Gustav Tauschek (1899-1945), working at an IBM punched-card subsidiary in Sommerda, Germany, demonstrated and patented a prototype magnetic drum in 1932, with fixed heads on a ferromagnetic-coated cylinder holding 500,000 bits, and cites German patent DE 643,803 (filed 1 July 1933, issued 17 April 1937). This card opened one Tauschek patent, US 2,080,100, a continuation in part of an application of 4 August 1932. It describes magnetising points on groups of steel strips to hold account balances, selected by a dial and read back into number wheels. It does not describe a rotating drum or a 500,000-bit capacity. The drum attribution therefore rests on the German patent, which was not opened here; the US patent is evidence only that Tauschek was recording numbers magnetically in 1932-1933.

1946-1948: Booth in London. The Science Museum Group holds an experimental drum store that it dates to about 1946 in London, devised by Andrew Donald Booth as a memory device. Booth's own autobiography, reproduced by the IEEE Computer Society, says the drum was a 2 inch diameter brass cylinder plated with 0.0005 inch of nickel, with 21 parallel channels plus a clock channel, storing 256 words of 21 bits for his Automatic Relay Computer, which was demonstrated on 12 May 1948. The Museum and the IEEE page give the drum as 2 inches in diameter and 2 inches long at 10 bits per inch. Booth also reports that his first attempt, a spinning 10 inch recording disc with a head held at a fixed distance by the Bernoulli effect, failed because the disc flapped, and that a small firm run by his father made drums, largely for export to the US. He says T. Kilburn of Manchester visited on 2 November 1948 and took away a sample of his read/write heads, which he later found had been copied by the Manchester group and by Ferranti. Those statements are Booth's recollection and no second source was found for them.

1947-1950: ERA in Minnesota built a different kind of drum. The Museum says William Norris led a group at Engineering Research Associates (including John Hill and Robert Perkins) that cemented 3M ferric-oxide-coated Mylar tape onto a cast aluminium cylinder for a US Navy cryptographic unit in 1947. In 1950 ERA produced an 8.5 inch drum with 200 stationary heads and 16,384 words of 24 bits for the Atlas 1 code-breaking computer, which became the Univac 1101 after Remington Rand bought ERA. The company's 1951 brochure (vendor claim) says the first 1101 was delivered to the US Government in 1950 and ran eight days later, that marks are stored at 1,280 per square inch in parallel peripheral tracks with one head per track, that a revolution takes about 17 milliseconds, and that access time is 32 microseconds to 17 milliseconds. The VIP Club, a veterans' group, says Arnold Cohen and Sidney Rubens are credited with patenting the rotating drum and that a prototype description was presented as a paper in 1947. It also claims the ERA drum was the world's first computer hard drive; this card does not adopt that claim.

Drums become the memory of the early 1950s. The Museum lists the Harvard Mark III (1949), Manchester Mark I (1949), SWAC (1950) and MIT Whirlwind (1951) among early machines with drum memory, and the Univac 1101 (1951), IBM 650 (1953) and English Electric DEUCE (1955) among commercial ones. The VIP Club quotes a 14 October 1952 ERA newsletter saying MIT staff were running acceptance tests on ERA magnetic storage systems for Whirlwind; that is a year later than the Museum's 1951 date and may describe a later unit, and the card does not reconcile them. IBM says it announced the 650 on 14 July 1953. Its drum held 20,000 digits in the first version and up to 40,000 later, in 20 or 40 bands of 50 words, spun at 12,500 rpm, and gave an average access of 2.4 milliseconds. IBM also says the first 650 was delivered in December 1954, nearly 2,000 were installed, and production ended in 1962 (vendor history).

Core memory. The Museum says Frederick Viehe filed a core patent in 1947 and An Wang in 1949, with RCA's Jan Rajchman filing in 1950 and MIT's Jay Forrester in 1951, and that other important contributors include E. Albers-Schonberg, J. P. Eckert and M. K. Haynes. Forrester's patent 2,736,880, Multicoordinate Digital Information Storage Device, was filed in May 1951 and granted on 28 February 1956 (Museum; Centre for Computing History). The Whirlwind project, a US Navy real-time flight simulator, replaced a troubled electrostatic CRT memory with a 32 by 32 array of 1,024 cores and demonstrated it in August 1953 (Museum). The MIT course page adds that IBM then licensed the technology, that Ken Olson, later founder of Digital Equipment Corporation, was the liaison between MIT and IBM, and that a lawsuit over the patent rights produced an unusually complete archive. The Museum says IBM announced its first commercial core unit, the Model 737 with 4,096 words of 36 bits, in 1954, and that cost fell from about 1.00 dollar to one cent per bit before semiconductor memory replaced cores in the 1970s.

What this meant for disk drives. Once cores held the working data, drums and then disks took the role of bulk storage. IBM's archive page describes the 305 RAMAC processing unit as containing a magnetic process drum and a magnetic core register, with the 350 disk storage unit attached: 50 disks, 50,000 sectors of 100 characters each (5 million characters), 1,200 rpm, 20 tracks per inch recorded at up to 100 bits per inch, typical head-to-disk spacing of 800 microinches and an average seek of about 600 milliseconds, with the first model announced on 4 September 1956. A Computer History Museum profile by A. S. Hoagland says the RAMAC lab's insight was to use a stack of disks instead of the magnetic drums then in use, because a disk stack offers far more recording surface for its volume, and that the air-bearing head, spin-coated iron-oxide disks and the moving access arm were new. IBM's own history says the team tried rods, strips, tapes and flat plates before settling on aluminium disks coated with iron oxide paint, and quotes Reynold Johnson as saying the goal was half a second to reach any track and that about 800 milliseconds was achieved. The Museum says drums persisted in specialised uses into the 1980s, and the VIP Club says Univac's Fastrand drums were a mainstay of mass storage in the 1960s and 1970s.

What the sources do not settle. No source opened here gives a production count for magnetic drums across all makers, and the card does not state one. The 1932 drum claim is unverified beyond the Museum's statement. Booth's account of the copying of his heads is a single recollection, and the ERA veterans' claim to the first computer hard drive comes from an interested party.