How Hard Drive Read/Write Heads Work

A hard drive read/write head is a thin-film electromagnetic transducer mounted on a slider that flies 5 to 10 nanometers above a spinning platter. It reads data by detecting changes in magnetic polarity on the platter surface and writes data by applying a focused magnetic field to flip the polarity of individual magnetic domains. Each platter surface has its own dedicated head, and all heads are connected to a single actuator arm assembly driven by a voice coil motor. Head stack replacement is one stage of our complete hard drive data recovery workflow.

Voice Coil Motor and Actuator Assembly

The actuator assembly is the mechanical structure that positions the read/write heads over the correct track on the platter. It pivots on a bearing located near the center of the drive. The voice coil motor (VCM) drives this pivot by passing current through a coil suspended between two permanent magnets. Reversing the current direction moves the actuator in the opposite direction.

Modern drives abandoned stepper motors for head positioning in the early 1990s. Stepper motors move in fixed increments, which limits track density. Voice coil motors provide continuous, proportional positioning with sub-micron accuracy. A servo feedback loop reads positioning data from servo wedges embedded on the platters and adjusts coil current thousands of times per second to keep the heads centered on the target track.

When external power is lost, the spindle motor acts as a generator. The rotational momentum of the still-spinning platters produces back-EMF, which is routed to the VCM coil to actively sweep the heads off the platters before the air bearing collapses. Most modern drives retract the heads to a parking ramp at the outer edge of the platter stack; older drives used a landing zone near the inner diameter. Once the heads reach the park position, a magnetic latch holds them in place.

Head Slider Aerodynamics and Fly Height

The read/write element is not a standalone component. It is fabricated onto a ceramic slider, typically made of alumina-titanium carbide (AlTiC). The bottom surface of the slider is the air bearing surface (ABS), which is etched with a pattern of rails and channels designed to generate aerodynamic lift from the thin layer of air dragged by the spinning platter.

At 7,200 RPM, the platter surface moves at roughly 80 to 120 km/h depending on the radial position. This airflow creates a pressurized air cushion that lifts the slider to a fly height of 5 to 10 nanometers. For context, a smoke particle is roughly 250 nanometers in diameter, which is why opening a drive outside a particulate-controlled environment risks contamination.

These dimensional disparities dictate strict cleanroom controls during any drive opening. Corneocytes (skin flakes) are typically around 1,000 nanometers thick, roughly one hundred times the air gap. A bacterium is 1,000 to 5,000 nanometers long. Any of these particles caught between the slider and the platter acts like a wedge forced through a bearing at surface speeds of 80 to 120 km/h. The slider ramps over the obstruction, scores the magnetic coating, and sprays fresh debris across the remaining surfaces. Our 0.02 micron ULPA-filtered clean bench keeps airborne particulates out of the head-disk interface during head stack work.

The fly height must remain constant across the entire platter radius despite variations in air velocity. The ABS geometry is engineered to compensate: negative pressure regions at the leading edge balance positive pressure at the trailing edge where the read/write element sits. Some modern drives use thermal fly-height control (TFC), where a small heater embedded in the slider expands the tip by a few nanometers to fine-tune the gap during operation.

Read Sensor Technology: MR, GMR, TMR

The read element and the write element on a modern head are two different structures stacked on the same slider. The write element is a tiny electromagnet; current through a copper coil drives a focused field through a write pole to flip magnetic domains on the platter. The read element is a passive sensor whose electrical resistance changes in response to the magnetic field coming off the platter surface. The sensor technology has gone through three generations.

MR (magnetoresistive)

First introduced by IBM in the early 1990s. An MR sensor is a thin strip of nickel-iron alloy whose resistance changes by a few percent when a magnetic field crosses it. MR heads enabled the jump past 1 Gb per square inch. They were replaced once areal density outpaced what the anisotropic magnetoresistive effect could resolve.

GMR (giant magnetoresistive)

Dominant from the late 1990s through the mid-2000s. A GMR sensor is a sandwich of two ferromagnetic layers separated by a non-magnetic conductive spacer. When the field polarity on the platter aligns the two layers, resistance drops sharply. The effect is large enough to resolve smaller magnetic domains, which is what carried drives from single-digit Gb per square inch into the hundreds.

TMR (tunneling magnetoresistive)

Standard on current drives. The conductive spacer of a GMR stack is replaced with a thin insulating barrier, typically magnesium oxide. Electrons tunnel through the barrier in quantities that depend on the relative alignment of the two ferromagnetic layers. The resistance swing is larger than GMR, which is why TMR heads are required for modern areal densities above roughly 500 Gb per square inch.

From a recovery standpoint, the sensor generation affects donor matching. A TMR read element is more sensitive to electrostatic damage during handling and more intolerant of fly-height excursions than the MR sensors on older drives. The parameters the drive firmware uses to bias the sensor, calibrate write current, and interpret the read channel output are stored in adaptive tables in the System Area. Those adaptives are tied to the specific head stack that was calibrated against them at the factory. A donor head stack with a different sensor variant or a different preamp will not read the platters correctly until the adaptives are either re-used from the original PCB or translated through PC-3000 utilities.

Preamp Chip and Signal Path

The signals produced by the read element are measured in microvolts. Sending these tiny signals down a flex cable to the PCB would introduce noise that overwhelms the signal. To solve this, every modern hard drive has a preamplifier (preamp) chip mounted directly on the head stack assembly (HSA), millimeters from the read/write elements.

Preamp

Amplifies the microvolt read signal to millivolt levels before sending it to the read channel chip on the PCB. Also drives the write current to the write element. Each head pair (read + write) has its own preamp channel.

Flex Cable

A thin polyimide ribbon connects the preamp to the PCB connector. It carries amplified read signals out and write signals in. Damage to the flex cable produces symptoms identical to head failure.

Read Channel Chip

Located on the PCB, the read channel chip converts the analog signal from the preamp into digital data. It applies signal processing algorithms (partial response maximum likelihood, or PRML) to extract data from noisy signals.

The preamp chip is the single most common point of failure in the signal path. Electrical surges, flexing of the HSA during impact, and thermal stress can all damage preamp channels. A preamp failure on one channel disables the corresponding head. If the failed head is needed to read data, a head swap is the only path forward.

Head Failure Modes and Symptoms

Head failures are the most common mechanical failure in modern hard drives. The symptoms vary depending on how the heads failed and how many heads are affected.

Head Configurations by Capacity

The number of heads in a drive depends on the number of platters and whether data is written on one or both surfaces. Manufacturers configure the head count to hit a target capacity at a given areal density.

For example, the Seagate Rosewood family (ST1000LM035, ST2000LM007) ships in multiple head configurations under the same model number. Because Seagate fills 1TB orders using binned or degraded 2TB platters, an ST1000LM035 can appear in the wild as a 2-head build (1 platter, both surfaces), a 3-head build (2 platters, one surface unmapped), or a 4-head build (2 platters, both surfaces mapped). The 2TB ST2000LM007 uses 4 heads across 2 platters.

Head configuration matters for recovery because a head swap requires a donor drive with the same head count, head map, and preamp compatibility. A 2-head, 3-head, and 4-head Rosewood all use different head stack assemblies and are not interchangeable, even though they share the same model number.

Donor Head Matching Criteria

A donor head stack assembly is not interchangeable by model number alone. Two drives with identical part numbers on the label can carry incompatible head stacks if they were built in different production runs. Before a donor is pulled from the shelf, its head stack must match the patient drive across several dimensions.

Preamp Revision and Family

The preamp chip mounted on the HSA carries a revision code that the manufacturer uses to track silicon generations. A donor HSA with a different preamp revision drives the read and write channels with different bias currents, different write compensation curves, and different channel gain. The patient firmware's adaptive tables expect the original preamp's response curve; swapping in a mismatched preamp causes the drive to clock in but read garbage or fail calibration during the initial ID read.

Head Stack Assembly Family

HSA families are tracked by the manufacturer with internal part numbers distinct from the drive model number. Within a model family, the HSA generation changes as the head wafer changes. A Seagate Barracuda built in week 32 of a given year may carry a different HSA generation than the same model built in week 38, with different magnetic properties on the read element. Donor identification requires cross-referencing the patient label codes (site code, firmware revision, factory date code) against documented HSA family maps.

Micro-Jog Tolerance

Micro-jog is the small radial offset between the read element and the write element on each head, calibrated at the factory. Each head carries its own micro-jog value stored in the System Area. A donor head with a different micro-jog will write to one location and read from a different location, producing track misregistration. On drives where micro-jog values can be read from the patient adaptives, PC-3000 utilities can apply the patient's micro-jog profile to the donor heads. On drives where micro-jog is tied to the HSA ROM, the donor must already carry compatible values.

Firmware Adaptives and ROM Data

Adaptives are the per-head calibration tables the firmware uses to bias the read element, set write current, and tune the read channel. They are generated on the factory test track and stored in the System Area and sometimes the PCB ROM. After a head swap, the original adaptives must be preserved; either by retaining the patient PCB or by reading the patient adaptives out through a PC-3000 terminal session and writing them back into the assembled drive before the first spin-up with the donor heads.

Head Map and Surface Pairing

Multi-platter drives assign each logical head to a specific platter surface (upper or lower). The patient head map defines which physical heads are enabled and how they translate to logical head numbers. Donor HSAs pulled from drives with a different head map (for example, a 4-head donor for a 3-head patient built on the same model number) cannot be installed directly; the surface alignment will not match the patient adaptives.

Mismatches across any of these axes produce distinct symptoms after a swap. A preamp revision mismatch usually shows as a drive that spins up, identifies, and then fails to read the System Area or reads it with heavy retries. A micro-jog mismatch produces track-by-track bad sectors that move when the drive is power cycled. A true donor head failure (dead donor, damaged in storage) looks identical to the original head failure. Diagnosing which category a post-swap failure falls into is why donor selection uses multi-axis matching rather than model-number matching.

PRML and Viterbi Read-Channel Tuning

Once the signal path is intact, the read channel still has to turn a noisy analog waveform into bits. Modern drives do this with partial response maximum likelihood (PRML) detection. The read channel oversamples the preamp output, equalizes it through an analog and then digital FIR filter, and feeds the resulting samples into a Viterbi detector that traces the most likely bit sequence through a trellis of possible states. Extensions like EPRML (extended PRML) and NPML (noise predictive maximum likelihood) add higher-order targets and noise-whitening filters to keep bit error rates manageable at current areal densities.

When heads are degraded but not dead, the drive's own error recovery procedures (ERP) run a fixed sequence of read retries with altered channel parameters. In production use the ERP is tuned for throughput, not salvage; it gives up quickly on hard sectors to keep the host operating system responsive. For recovery, the ERP is the wrong policy. A drive with weakening heads will burn through the remaining head life running aggressive retries on sectors that a patient, parameter-aware imager could skip and revisit.

DeepSpar Disk Imager addresses this by controlling the drive through its SATA interface at a level below the host OS. It issues raw ATA read commands, caps retry counts, skips bad sectors on the first pass, and returns to them on later passes with altered read offsets. If a head is fading unevenly across the radius, DeepSpar can image the healthy zones first, preserve that data, and come back to the marginal zones under tighter retry budgets. This is the difference between a one-pass image that dies halfway through with the marginal zones never read and a multi-pass image that drains the readable sectors first before the head finally stops responding.

PC-3000 Portable III and PC-3000 Express go a step further on supported drive families by communicating with the drive through its terminal (TTY) or factory port, below the normal ATA command interpreter. Inside PC-3000's Data Extractor module, a technician can bias the read channel parameters directly: FIR filter coefficients, Viterbi branch metric thresholds, channel gain, and per-head read retry strategies can be overridden on a head-by-head basis. On a weak head with low read signal amplitude, raising the channel gain and relaxing the Viterbi threshold may recover sectors that the stock firmware rejected. On a head with high noise but adequate amplitude, tightening the FIR filter and enabling NPML targets can recover sectors the stock configuration dropped.

These adjustments are not universal. Each drive family exposes a different subset of channel parameters through PC-3000, and each patient drive responds differently based on the state of its heads, platter surfaces, and adaptives. The tuning is iterative: image a pass with the stock configuration, review the per-head bad-sector distribution, adjust the channel parameters for the worst-performing head, image another pass. Temperature also enters the picture; FLIR thermal monitoring during long imaging runs catches preamp heating that shifts the read signal amplitude and forces a parameter re-tune. The full tool chain and pricing tiers for this work are documented on our hard drive data recovery service page.

Why Head Failures Require Lab-Level Recovery

Data recovery software cannot compensate for a physical head failure. The drive firmware needs functioning heads to read the servo wedges, access the System Area (where firmware modules are stored), and read user data tracks. If the heads cannot read, the drive cannot initialize, and no software on the host computer can bypass that.

A head swap involves opening the drive in a particulate-controlled environment (laminar flow bench with 0.02 micron ULPA filtration), removing the failed head stack assembly, and installing a matched donor HSA. After the swap, the drive's ROM chip data and adaptive parameters must be transferred from the original PCB to ensure the new heads can initialize with the drive's existing calibration data. PC-3000 handles this firmware alignment step. The full procedure, tool chain, and donor-match criteria are described in what a head swap involves.

Frequently Asked Questions