When can I recover data with software?
Software recovery is safe only when the drive spins up cleanly, the host BIOS or UEFI detects the drive at the correct factory capacity, and the failure is logical (deletion, format, RAW filesystem) or limited to a small, non-escalating bad-sector count. If any of those conditions is missing, software cannot help and will likely accelerate the damage.
Imaging a failing drive on Windows is the most common avoidable mistake. Windows auto-mounts the volume, walks the file system to build thumbnails and search indexes, and runs SMART polling. Each of those triggers random non-sequential reads across the platter; a healthy drive shrugs them off, but a failing drive responds by hammering the read/write heads across the surface and burning service life it does not have. Image from a Linux live USB with auto-mount disabled, or pull the drive and image it from a separate machine.
The canonical ddrescue command sequence
GNU ddrescue is the open-source standard for imaging a drive with bad sectors because it does not stop at the first read error and because every pass it runs is recorded in a mapfile. The mapfile records which logical block addresses have been read, which failed, and which are still untried, so any pass can be paused and resumed without re-reading sectors that are already in the image.
First pass: forward, no scrape, secure the easy data first.
ddrescue -n /dev/sdX /mnt/safe/image.img /mnt/safe/image.mapfile
The -n flag (no-scrape) tells ddrescue to skip past slow or failing zones on the first pass. The goal of pass one is to copy every healthy sector before the drive deteriorates further. Slow zones get marked in the mapfile and are revisited later.
Second pass: bounded retries on the bad zones, with direct I/O.
ddrescue -d -r3 /dev/sdX /mnt/safe/image.img /mnt/safe/image.mapfile
The -d flag (--idirect) bypasses the kernel page cache so the read goes straight to the drive instead of waiting on the kernel block layer to time out. The -r3 flag bounds the retry count at three; an unbounded retry loop on a dying drive is how recoveries die mid-job.
Third pass: reverse direction against stubborn sectors.
ddrescue -d -R -r3 /dev/sdX /mnt/safe/image.img /mnt/safe/image.mapfile
The -R flag reverses the direction of the entire pass. Asymmetric platter damage and head suspension issues sometimes read better from the opposite physical direction; the reverse pass is the cheapest tactic for picking up the last few percent of unread blocks.
Recovery work runs against the image file. The original drive should never be the source for data carving, filesystem repair, or test mounts. If the imaging stalls or SMART begins escalating Current Pending Sector Count or Reallocated Sector Count during a pass, stop the imaging job and ship the drive. HDDSuperClone (now OpenSuperClone) replaces ddrescue when the drive needs ATA pass-through and per-head skipping logic; the tool-by-tool comparison lives at ddrescue vs HDDSuperClone for failing drives. The full DIY recovery guide has the longer walkthrough.
What disqualifies a drive from DIY recovery?
Any acoustic anomaly, any electrical anomaly, or any firmware-level misidentification disqualifies the drive. The shared property of every signal in this list is that continued power draws contamination or scoring across the magnetic media; once the recording surface is gouged, the data under the gouge is gone and no software pass brings it back.
Acoustic signals that mean stop
- Clicking or ticking. The voice coil actuator is sweeping the head stack to the limiter and back because the heads cannot lock onto the embedded servo bursts. The most common cause is preamp failure or physically damaged sliders.
- Beeping or buzzing. The spindle motor is drawing current and refusing to spin. Either the heads are stuck to the platter surface in a stiction event, or the motor bearing has seized. Continued power damages the motor coil windings.
- Grinding or scraping. The slider air-bearing has collapsed and the head is dragging through the magnetic recording layer. Every additional rotation deepens the scratch and aerosolizes more media debris into the chassis.
- Spin-up followed by silent spin-down. The drive is failing its initialization sequence; it cannot read its own Service Area firmware modules. This is a firmware repair job on a PC-3000, not a software imaging job.
Electrical signals that mean stop
No spin, no vibration, no LED activity means the PCB is not delivering power to the motor or preamp. Surge events typically blow the TVS diodes on the 5V or 12V rails; they short to ground deliberately to protect the rest of the board. Swapping a PCB from a parts donor without transferring the original ROM chip will brick the drive because modern PCBs hold per-drive adaptive parameters that the heads were calibrated against. ROM transfer requires lab-grade equipment.
Firmware and SMART signals that mean stop
- Wrong reported capacity. A 1TB drive that enumerates as 0 GB, 32 MB, or as a factory model name (for example a WD ROM model string) has corrupted translator or firmware modules. The drive is in a degraded factory mode and cannot be addressed by user software.
- Not detected at all. If the BIOS or UEFI cannot see the drive, no recovery utility can either. Cause is usually firmware corruption, head failure preventing SA module read, or PCB failure.
- Current Pending Sector Count or Reallocated Sector Count escalating during a read. Bad sectors growing in real time means the magnetic surface is still degrading. The heads are likely contaminated, and every additional pass adds debris to the chassis.
A good rule of thumb: power the drive once for a careful diagnosis, capture the acoustic signature and SMART output, and then leave it powered off until a decision is made. Repeated power cycles in the hope the drive will mount "just one more time" are the single most common way recoverable drives become unrecoverable.
How does a head swap actually work?
A head swap is a physical transplant of the entire head stack assembly from a precisely matched donor drive into the patient drive, performed inside a particle- filtered enclosure with non-magnetic precision tooling. The procedure has four hard prerequisites: a clean bench, slider-on-ramp parking discipline, a duralumin head comb, and a donor that matches at the firmware-revision and head-map level.
The clean bench
The slider flies above the platter on an air bearing measured in single-digit nanometers. Ambient room air carries skin cells, smoke residue, and dust particles that are orders of magnitude larger than the flight height. The Austin lab performs all open-chassis work inside a 0.02 micron ULPA-filtered laminar flow clean bench so that opening the hermetic chassis does not deposit a particle that the slider will hit on the next rotation. Shop fans, household HEPA units, and ad-hoc plastic enclosures do not meet this requirement.
Slider-on-ramp parking discipline
Modern 3.5-inch and 2.5-inch drives use a load/unload ramp at the outer edge of the platter; the head stack parks on this ramp when the drive spins down. The transplant procedure keeps the sliders on the ramp, or on a separate removal fixture, from the moment the chassis opens until the head stack is seated in the donor. The read/write elements are mounted on spring-loaded suspensions that bias toward the platter; if the heads come off the ramp without tooling holding them apart, the gimbals collapse and the sliders clap together. A single clap shatters the read and write elements and turns the patient or the donor into scrap.
Duralumin head combs
A head comb is a precision tool that slides between the suspension arms to hold each slider air-bearing surface clear of the platter while the head stack is moved between chassis. The combs the lab uses are CNC-machined from duralumin, an aluminum-copper alloy. Three properties matter: duralumin is non-magnetic, so it does not interact with the actuator magnets; it is rigid enough to resist the spring tension of the suspensions without deflecting; and it is soft enough that an accidental tap against a platter does not score the magnetic surface. Plastic and resin combs flex under the same spring load, the sliders shift during the transfer, and the heads either touch the platter or clap together. The duralumin geometry is matched per-platform; one comb does not fit every drive family.
For the full mechanical walkthrough, see what a head swap involves.
Donor matching at the head-map level
Buying the same model number from the same manufacturer guarantees nothing. Hard drive vendors revise heads, preamps, and platter coatings inside a single product line and ship the result under the same SKU. A valid donor matches on:
- Firmware revision. The four-character Seagate firmware code (for example SN03) or the Western Digital firmware band must align so the read channel calibration is compatible.
- Site code. The factory of manufacture maps to the chassis geometry and the suspension supplier. Seagate site codes such as WU (Wuxi), SU (Suzhou), and TK (Thailand) are not interchangeable.
- Physical head map. Drives in the same product line can ship with different active head counts; the donor must have the same active surface configuration as the patient.
- Preamp IC revision. The preamp is the analog amplifier on the actuator arm. Vendors swap preamp suppliers and revisions during a production run; mismatched preamp revisions deliver the wrong gain to the donor heads and the read channel returns garbage.
- DCM string (Western Digital). The Drive Configuration Matrix on the WD label encodes head and platter variants; specific characters must align between donor and patient.
- PCB revision and manufacture date window. Board revisions ship with different preamp drivers and ROM adaptive layouts even within a single firmware band. The lab cross-checks the PCB silkscreen revision and the date code on the patient label against the donor pool, and rejects donors whose manufacture date falls outside the window where the same head and platter assembly was in production.
The full donor matching procedure, including how the lab sources matching units, is documented at how donor drives are matched.
Why a close donor still fails: micro-jog calibration
Even with a clean transplant and a tight donor, the patient drive can refuse to calibrate. Modern read and write elements are physically offset on the slider by a sub-track distance called the micro-jog. The factory measures this offset for every individual head and stores it in the drive's adaptive parameters in the Service Area or the ROM. When a donor head sits on the patient's arm, the patient's logic board still applies the original micro-jog values to a head that was calibrated against different platters. If the offset between the donor heads and the patient's stored adaptives exceeds the hardware tolerance, the read channel cannot lock onto the servo bursts. The drive clicks, sweeps, or returns garbage even though the mechanical work was perfect.
Resolving a micro-jog mismatch is firmware work, not mechanical work. The PC-3000 connects through the drive's diagnostic terminal, reads the Service Area, and rewrites the relevant adaptive modules so the patient's read channel is tuned against the new heads. After the firmware is re-tuned, sector-by-sector imaging can begin. For the underlying physics of the failure mode that drives most head swaps, see what happens during a head crash. Service detail and pricing live on the hard drive data recovery page.
