IBM 3340 and the Original 'Winchester' Drive: Low-Mass Landing Heads, Lubricated Disks and the 3348 Data Module (1969-1976)

Short answer: the IBM 3340 was a disk drive for low-end System/370 computers whose real novelty was a head and disk pairing. A very light three-rail ferrite head rested on a lubricated disk in a landing zone when the disk was stopped and lifted off on an air bearing when it spun, so no mechanism was needed to load and unload heads; the head carriage and disks were packed together in a removable 3348 Data Module of 35 or 70 megabytes. The Computer History Museum says the project began in 1969 under Kenneth E. Haughton at IBM San Jose and that the drive began shipping in November 1973. 'Winchester' was the project's code name, taken, according to Haughton, from an early requirement for two spindles of 30 megabytes each, a '30-30' like the rifle cartridge. The name later stuck to the technology and to disk drives in general. The module itself was not repeated: the 3350 that followed made the head-disk assembly non-removable, and the sources read give 1975 or 1976 for it.

Why IBM built it, 1969. The museum's catalogue text says San Jose management wanted a lower-cost storage subsystem for its coming low-end System/370 models and to answer plug-compatible disk makers, and chose to answer with new technology instead of a cut-down 3330. Mulvany's 1974 paper says the performance and cost per megabyte were to be similar to the 3330's, aimed at small and intermediate systems. The same paper explains that the main cost of a removable disk pack lies in aligning the drive's heads to tracks written on any drive, typically within about 20 percent of a track width, and that the aim was to remove that adjustment and the head loading mechanism. The museum text adds that Haughton and the San Jose management liked the idea of a fixed-media drive because the head that wrote the data would always be the head that read it, that the original plan included a later fixed drive code-named 'Weatherbee', then 'Madrid', which became the 3350, and that sales staff resisted fixed packs with the message that disks 'gotta be removable because they always have been' (quoting Haughton, 2004 panel, via the museum text). That is why the 3340 kept removability, by moving the heads into the removable part.

Where the head came from. Two IBM papers and the museum text agree that the idea of heads that start and stop in contact with a lubricated disk was already being tried inside IBM. The 1981 retrospective says IBM's Los Gatos laboratory built a fixed-head file in the late 1960s, led by J. T. Ma, as a refresh buffer for a high-resolution display, and that this was the first file to use lubricated iron-oxide particulate disks and start-and-stop-in-contact heads, with heads developed by Data Disk Corporation for a plated disk. The museum text adds that Ma built it first for the Rand Corporation and then as the IBM Aries. The Data Disk head had three small pads in a triangle, with the magnetic element at the apex. Mulvany says a version of it, made of barium titanate with a ferrite element and two ferrite pads, flew in the IBM 3735 Programmable Buffered Terminal, and that W. S. Buslik was developing a sealed five-megabyte single-head drive whose small, self-contained form encouraged the data-module idea. The museum text says the 3340 project started with the Data Disk tri-pad head and that it proved too costly to make and, because of its asymmetry, troublesome for the fast access servo, so the team, mainly Mike Warner, designed the head that is now called the Winchester head. The 1981 paper dates the start of that development effort to 1969.

The tri-rail slider. The patent that the museum cites, US 3,823,416 by Michael W. Warner, was filed on 1 March 1973 and granted on 9 July 1974, so it was filed after the project began in 1969 and before the first shipments. It describes a ferrite slider with three parallel rails: the two outside rails form the air bearing, and a C-shaped ferrite core is glass-bonded at the trailing end in line with the narrow centre rail, so that the recording gap sits at the roll axis. Outside rails have a taper at the front and a length-to-width ratio of at least five, preferably ten, and the air that bleeds from the sides produces two pressure peaks along each rail, which Mulvany's paper and the museum text both describe as giving a stable head with a small load. The patent says the prior art needed loads of 300 to 400 grams and head-unloading mechanisms, and it gives a preferred body 0.25 inch long and 0.150 inch wide, with outside rails 0.025 inch and a centre rail 0.005 inch wide and a head assembly mass of 0.25 gram. The patent says that at rest the transducer is 'in a landed condition' on the disk, that the front taper builds pressure as the disk begins to turn so that the leading edge lifts 'rapidly and smoothly', and that on slowing the trailing portion touches down first. Mulvany's Table 2 lists the 3348 head at 10 grams of load and 0.25 gram of suspended mass (the printed table is partly garbled and its figures for the other heads are not used), and his text says the single-taper 2305 head, whose ferrite and glass construction the design borrowed, needed a load of more than 1.2 kilograms. The museum text says the batch method used for the 2305 head, a 'stiff finger' process of slicing and lapping many parts at once, could be adapted for the new head.

Disk lubrication and the landing zone. Contact at start and stop required a surface that would not wear. The 1981 retrospective says the first lubricant was a 0.75 percent solution of Dow Corning RDC-200 in a freon solvent, applied with a soft absorbent cloth, that this disk proved very reliable, and that it led to disk products built on a lubricated surface with start-and-stop-in-contact heads. In the 2004 oral history Chris Coolures says the challenge was to get the right amount of lubricant and to know what became of it after a year of spinning, and names the risk that the head would stick to the disk (stiction); Warner adds that the lubricant remained a problem in later generations (their recollections). Mulvany's paper lists a 0.51-centimetre stroke to move the heads from the data zone to the landing zone among the actuator requirements, and the museum catalogue text says the heads started and stopped in contact with the disk on a dedicated landing zone. The 1981 retrospective adds that, because the head slider was cheap, two heads per surface could be used, which halved the stroke length and lowered the cost of the carriage and actuator; Mulvany lists the same two-heads-per-surface decision.

The 3348 Data Module and what 'sealed' meant. The museum text credits Richard B. Mulvany and Rudolf W. Lissner with inventing the data module, and Mulvany's paper says Lissner's contribution was recognised by a joint IBM Outstanding Invention Award in 1974 and that Haughton started the task force that developed the module concept. Each module held the disks, spindle and bearings, head carriage and head-arm assemblies. Mulvany describes a polycarbonate cover with a flexible roll-top door that the drive opened and closed, and says that at the end of loading the cover was sealed to an air interface and filtered air was introduced; he also lists a read-only interlock in the carrying handle. The museum catalogue text says the module is often called 'sealed' but 'was not in a conventional sense sealed', and that it was removable from the start, contrary to what it calls an urban legend, quoting Haughton on this in 2007. The Datapro report of August 1973, writing from IBM's announcement, describes the heads and disks as 'all sealed into a removable cartridge' in an 'air-tight environment' with a closed-loop air filtration system in the drive. The two descriptions are compatible if 'sealed' means closed to the room when mounted, but they are not the same claim, and the card shows both.

The numbers. Mulvany's paper records how the specification moved. The first program called for a dual drive with 30 megabytes per cartridge at 66 tracks per centimetre (167 per inch), 2,400 rpm, 806 kilobytes per second and an average access time of 25 milliseconds. Track density was then raised to 118 tracks per centimetre (300 per inch), rotation to 2,964 rpm and the data rate to 885 kilobytes per second, with bit density reduced from 6,276 to 5,636 bits per inch (the text gives 2,465 to 2,180 bits per centimetre, while Table 3 prints 2,216, an inconsistency inside the paper), giving about 12 megabytes per surface; the modules were finally set at 35 and 70 megabytes, with a 70-megabyte module weighing 8.6 kilograms and a 35-megabyte module 7.7 kilograms, and a later Model 70F with 0.5 megabyte served by fixed heads. The 35-megabyte size was chosen in part to hold the data of one 2314 or 2319 pack, according to Mulvany. Datapro, reproducing IBM figures, gives 348 cylinders and 34.9 million bytes for the Model 35, 696 cylinders and 69.8 million bytes for the Model 70, 12 tracks per cylinder, up to 8,368 bytes per track, an average rotational delay of 10.1 milliseconds and 885,000 bytes per second, and says the recording density was more than 1.5 million bits per square inch, twice the 3330's and nearly eight times the 2314's (vendor claim). Multiplying Mulvany's 5,636 bits per inch by 300 tracks per inch gives about 1.69 million bits per square inch; that is the card's own calculation, and it is consistent with Datapro's figure. The 1974 paper, the museum text and Datapro all give 25 milliseconds as the average access time.

Defect skipping. The museum text credits Donald Frush with another 3340 first, defect skipping, covered by US patent 3,997,876 (granted 14 December 1976). Mulvany explains that on the 3330 the strategy was an error-correcting code able to correct errors up to 11 bits long, so a disk with a longer defect or more than one defect on a track was rejected at manufacture. For the 3340, defects found in track-by-track tests during manufacture were recorded in the home address, and data were formatted so that a single defect on a track is skipped, by splitting a field or extending a gap. The paper says 128 bytes were allocated on every track of a 3348 disk for this and that a defect of up to 16 bytes was allowed. Datapro says the 3340's error-correction code could correct an error up to 3 bits long and detect one up to 11 bits long in each record.

Dates, announcement and shipment. The museum catalogue text says the 3340 was announced in May 1973 and began shipping in November 1973, was the only disk drive for the System/370 Model 115, and was also offered on larger systems. Datapro says IBM 'unveiled' the Winchester disk concurrently with the Model 115 announcement, which it dates to 13 March 1973, and says customer shipments would begin in November 1973 for System/370 Model 125 systems and in March 1974 for other models; the March and May 1973 dates are shown as a conflict that these sources do not settle. In the 2004 panel Warner first said the machine shipped in November 1972 and Harker corrected him to November 1973, which agrees with the museum and Datapro.

What happened next. The museum text says the combination of the low capacity point, the high cost of the data module and a complex module load mechanism let the other makers answer with conventional, cheaper disk-pack drives, and names Control Data's SMD line as the main non-IBM product until the early 1980s; it says only Control Data and Nippon Peripherals built a drive able to use 3348 media, that several media makers, Memorex among them, made compatible modules, and that there was no later data-module product. Datapro had predicted in 1973 that independent makers would rush to design plug-compatible replacements and that retooling to build data modules would be formidable for pack manufacturers. The museum text adds that the 3340 did not gain significant volume on higher-end IBM systems and that neither the 3340 nor the IBM low-end systems 'were big sellers', calling any link between the two speculation; Harker says in the 2004 panel that a low-volume first product was perhaps good strategy because the technology was proven before the 3350. The 3350 turned the data module into a fixed head-disk assembly of 317 megabytes (317.5 in the panel), which the museum says some observers call 'the real Winchester' and which it says remained the basic packaging idea; the Storage Engine page dates it 1975 while the 2004 panel abstract and a panel exchange say it shipped in 1976. The same text says the rest of the industry did not immediately follow IBM to fixed media, citing the 1975 CDC 9676 and the 1983 DEC RA60 as removable packs. In the 2005 ferrite-head oral history, quoted by the museum, Warner says the head technology was soon adopted outside IBM for eight-inch and five-and-a-quarter-inch drives with one head per arm and that IBM was slower to do so; this is one participant's recollection. The museum text says the low-mass, low-load, inductive-head concept lasted into the 1990s until much smaller heads with a different air bearing and transducer replaced it. IBM's 3370 of 1979 used thin-film heads, per a 2005 oral-history quotation in the same text.

What the sources do not settle. Whether the 3340 was announced in March or May 1973; the nominal flying height (18 microinches in the museum text citing Haughton's 2007 oral history, about 20 microinches in the 1981 retrospective, and 0.51 micrometre, about 20 microinches, in Mulvany's paper); the head load (under 20 grams in the museum text and the 1981 retrospective, 10 grams in Mulvany's table and as the design target in the museum catalogue text); whether 'sealed' fairly describes the module; and the first shipment year of the 3350. The Computer History Museum pages cite oral histories with Haughton (2007), a panel on advanced ferrite disk heads (2005) and Al Shugart (2000, in Computer Technology Review); the two used for quotations here were taken only as the museum quotes them, and the Haughton and ferrite-head transcripts were not opened. Not covered: the 3330 and 2305 in detail, servo and actuator design beyond a mention, manufacturing and yield, and the competing products of Control Data, Memorex and others. The IBM Archives page for the Winchester drive, which the existing timeline and hdd cards cite, returned a not-found error on 11 October 2026, so it is not used here.