Voice Coil Actuators in Hard Disk Drives: From the IBM 2310 and 3330 to Rotary Arms (1965-1990s)

Short answer: the voice coil actuator is the part that replaced hydraulic and stepper positioning in disk drives with a coil and a magnet, and the sources read tie its history to IBM in three steps. The Computer History Museum says the IBM 2310 cartridge drive of 1965 used IBM's first commercial voice coil actuator and that it eliminated the fluid leaks of hydraulic actuators. The IBM 3330 of 1971 combined a voice coil motor with a servo surface that gave the first track-following feedback loop in a production disk file (Harker et al., 1981). In the mid-1970s IBM's British laboratory produced a rotary moving-coil version with one moving part, which is the arm shape later drives use (Harker et al.; Abramovitch and Franklin, 2002).

Before the voice coil, 1956 to 1963. Harker et al. describe the IBM 350 positioning two air-bearing heads through two counter-rotating magnetic powder clutches and a cable, first vertically to one of 50 disks and then along a radius, with a mechanical detent for the final position, 20 tracks per inch and an average seek of 600 milliseconds. The 1301 introduced a comb of heads, one per surface, moved by a hydraulic actuator with two sets of cylinders for coarse and fine positioning, 165 milliseconds average seek and 50 tracks per inch, later 100. The 1311 pack drive used a cheaper two-speed hydraulic actuator and a detent, with an average seek of 150 milliseconds, and a later change to a three-speed actuator cut the average seek from 150 to 60 milliseconds (the paper does not say which product first had it). The 2004 panel's Porter gives the 2314's average seek as starting at 75 milliseconds and later versions reaching 60; the two accounts differ on the starting figure and are shown as such.

1965: the first voice coil in a drive. The museum says the IBM 1130 system of 1965 included the 2310 drive, known as the Ramkit, with a removable single-platter cartridge and 1,500 revolutions per minute, and that it used IBM's first commercial voice coil actuator, a coil moving linearly in a fixed magnetic field from a permanent magnet. Abramovitch and Franklin say the first linear voice coil actuator was developed by IBM in 1965 and that the early comb actuators were hydraulic until the first voice coil motor in a drive in 1965; Disk/Trend lists the 2310 as the first voice coil actuator drive. These three agree but are not independent: the museum page and Abramovitch and Franklin both cite Stevens, and the Disk/Trend list gives no sources. The Harker et al. paper, read in full, describes the voice coil motor in its 3330 section and does not mention the 2310 in connection with it, and the museum's own 3330 page says the 3330 coupled its servo with 'the voice-coil actuator of the IBM 2310'. Abramovitch and Franklin add that the 1965 voice coil motors ran open loop until 1971.

1971: the 3330 and the closed loop. Harker et al. say the IBM 3330, first shipped in 1971, was the first production disk file with a track-following feedback control system, and that Hoagland had shown the idea feasible in the early 1960s. The system had one dedicated disk surface with prerecorded servo tracks, a servo head, electronics that turned the position error into drive current, and a motor that produced force in direct proportion to current: a wire-wound bobbin on a carriage in a loudspeaker-like magnetic structure, which is why the paper says it is called a voice coil motor. Seeks were controlled by monitoring servo track crossings; in track-following mode the loop corrected radial runout and temperature effects. With other tolerance improvements it reached 192 tracks per inch and a 30 millisecond average seek. The museum page agrees on the 30 millisecond access time and the closed loop, says the servo reference was recorded on one of the 20 surfaces in the pack, and adds that Shugart later called the feature one of the four most significant technical developments in mass storage. In the 2004 panel Santana says the 3330 closed the loop on one surface in the middle of the pack to minimise tolerances and that the move from hydraulic to electric positioning was a major innovation; Massaro calls closed-loop servos 'clearly a disrupting technology', and contrasts the 2314, an open-loop servo with a detent, with the 3330, where the head followed tracks written on the disk. Porter's figures in the panel are 19 data surfaces and one servo surface for the 3330 against 20 data surfaces for the 2314, 3,600 against 2,400 revolutions per minute, and 30 against 60 to 75 milliseconds average seek.

1973: the 3340 and how seeks were controlled. Harker et al. say the 3340 put two heads on each surface because its heads were cheap, which halved the stroke and lowered the cost of the carriage and actuator, and that it reached 300 tracks per inch and a 25 millisecond access time. Abramovitch and Franklin say the control system of the 3340 was the first in which all the pieces of a disk-drive control loop were in place, credit Dick Oswald's 1974 paper with a seek method that approximated bang-bang control without the control signal chattering, and say this was later codified as the proximate time-optimal servomechanism. They also describe the seek sequence that most servo systems still follow in some form: a seek algorithm, then a velocity profile into the target, then a settle mode, then track following, with an extra integrator switched in at the end. Oswald's paper itself was not opened; these statements rest on that one secondary article. The same article reports, from a private communication, that pivot friction was first noticed on the Winchester's linear actuator when a technician set the lower limit of a swept-sine test at 10 hertz instead of 100; that is a single-source anecdote.

The rotary actuator, 1975 onward. Harker et al. describe a moving-coil rotary actuator developed at IBM's development laboratory at Hursley, England, with pivoting data arms and, in their words, only one moving part. They say it gave most of the advantages of the linear voice coil actuator at very low product cost, that it was combined with Winchester heads and a non-removable disk in a sealed enclosure for small-system drives (the paper's figure caption for the rotary actuator names the IBM 62PC), and that the 3310 used a rotary actuator with a dedicated servo surface and sectored servo that reduced temperature effects. Their reference list names the actuator patent as US 3,849,800 (Cuzner, Dodman, Heath and Rigby, 1974) and a 1976 IBM paper by J. S. Heath on a swinging-arm actuator; neither was opened. Disk/Trend lists the IBM 62GV as the first drive with a rotary actuator in 1975. Abramovitch and Franklin say rotary actuators were first designed at IBM's 'Winchester Labs (in Winchester, U.K.)', which names the laboratory differently from Harker et al. and is shown as a naming conflict. They also say the first rotary actuators for 8-inch drives were relatively large truss structures with the suspension turned sideways, and that as drives shrank the pivot moved to a corner of the enclosure so the arm swept the disk in a single line.

The low-cost alternative: stepper and band. The 1982 Seagate manual for the ST-506 and ST-412 5.25-inch drives describes heads on a ball-bearing carriage positioned by a band wound on a stepper motor shaft with open-loop stepping, one track in or out per step pulse. It specifies, as vendor claims, a 3 millisecond track-to-track time, an average access of 85 milliseconds with the fast-seek algorithm, 15 milliseconds settling, 3,600 revolutions per minute, 255 tracks per inch and 153 cylinders for the ST-506 and 345 tracks per inch and 306 cylinders for the ST-412. Disk/Trend lists the ST-506 as the first 5.25-inch rigid drive in 1980 and the Conner CP340 as the first 3.5-inch drive with a voice coil actuator in 1986; the second item is a single-source list entry and no other opened source confirms the date. The step from stepper to voice coil in small drives is therefore only partly documented by the sources read.

What a rotary arm changed for the servo. Abramovitch and Franklin say that virtually all drives of their time used sectored servo, with position information multiplexed with user data, and that the control law was a state-space regulator for track following and a trajectory-following controller for seeks. They add that dedicated servo was suited to files with many surfaces and a poor choice for single-disk drives, and that as track densities rose the thermal offsets between heads in a stack became too large for dedicated servo. They report sample rates of 6 to 14 kilohertz and tracking bandwidths of 500 to 1,000 hertz at the time of writing, three dominant actuator resonances (torsion and sway modes), and pre-2002 proposals for dual-stage actuators whose adoption they called inevitable; the dual-stage forecasts were predictions in 2002 and are not treated as history here. These figures describe drives of around 2000 and are not specifications of current drives.

What the sources do not settle. Whether the 2310 or the 3330 deserves the label 'first voice coil motor' in a disk file: the museum, Abramovitch and Franklin and Disk/Trend say 1965, while Harker et al. describe the voice coil motor only in the 3330 section; the starting average seek of the 2314 (75 or 60 milliseconds); the name of the British laboratory; which IBM drive is called Piccolo, since Abramovitch and Franklin say 'IBM 3350' and Disk/Trend lists the 62PC as Piccolo and the 3350 as Madrid; and when low-cost 5.25-inch and 3.5-inch drives adopted voice coils. Abramovitch and Franklin also write that the 3340's control paper started the architecture of future servos, which rests on the authors' own judgement and a paper not opened here.