NVMe SSD Controller Failure: What Recovery Actually Involves

In order, before anything else:

• Power it down and keep it down. If the drive is the boot drive and Windows or macOS is hanging on it, do not keep retrying the boot. Shut down properly, or hold the power button if you must.
• Do not run recovery software against a failing drive. A scan reads the whole surface at speed, which is the heaviest possible load on a controller that is already unstable.
• Do not update or reflash the firmware. Vendor tools rewrite the drive's own service area. If the fault is in that area, a failed flash can turn a recoverable drive into a very difficult one.
• Do not keep swapping enclosures and adapters. One clean test in a known-good slot is diagnosis. Ten is wear.
• Skip the freezer, the oven and the reballing videos. None of these apply to NAND flash, and heat cycling a board you cannot re-image afterwards removes options.
• Write down what happened. Power cut, failed update, dropped laptop, liquid, sudden disappearance, gradual slowdown. The sequence of events genuinely narrows down where the fault sits.

Then get it in front of someone who can read it below the operating-system level. Diagnosis at our laboratory is free, and you get a fixed written quote before any work starts. You can tell us what happened here or bring the drive to us beside Tai Seng MRT.

An SSD is not a simple box of memory. It is a small computer. The controller is a multi-core processor running its own firmware, and it sits between your operating system and the NAND flash chips soldered either side of it.

Your computer thinks in logical block addresses: block 1, block 2, block 3, in a neat line. The NAND does not work that way. Flash is written in pages and can only be erased in much larger blocks, and every erase wears the cells out a little. So the controller keeps a translation layer, a constantly updated map that records which physical page on which chip currently holds each logical block.

That map is the drive. Without it, the NAND is a pile of numbered pages with no index.

The controller also runs error correction on every read, manages several NAND channels in parallel for speed, performs garbage collection in the background, and on nearly all modern drives encrypts everything on the way in. On DRAM-less designs it borrows a slice of your system memory to hold part of the map. All of this is vendor-specific, undocumented, and different between one manufacturer's model and the next.

Flash cells have a finite number of program and erase cycles. If a drive wrote your Outlook file to the same physical cells every time you saved, those cells would wear out years before the rest of the drive. So the controller deliberately spreads writes around, a technique called wear levelling, and it moves cold data out of lightly-used blocks so those blocks can take their share of the work too.

The practical consequence matters enormously for recovery. A single 40 MB presentation does not sit in one tidy place. Its fragments are interleaved across multiple dies and multiple channels, deliberately, for both endurance and speed. Two saves later, the newer fragments are somewhere else again and the old ones are marked stale, waiting for garbage collection to wipe them.

This is also why deleted files behave very differently on an SSD than on a mechanical drive. On a healthy, powered SSD, TRIM tells the controller which blocks the file system no longer needs, and garbage collection erases them, often within minutes. That is a genuine, physical erase. If you have deleted something from a working NVMe drive, speed matters far more than it does with a hard disk, and we explain the realistic limits on our deleted file recovery page.

This is the single most misunderstood point in SSD failure, and it is the reason a drive that looks completely dead is often not.

Typical signs of a controller-side fault rather than worn-out flash:

• The drive disappears from the BIOS or from Disk Management entirely, usually without warning
• It is recognised but reports the wrong capacity, or 0 MB
• It enumerates and then drops off the bus under sustained load
• It comes up read-only, or in a vendor safe mode after an unexpected power loss
• The whole machine hangs or takes several minutes to POST with the drive fitted

Common causes are power loss during a write, a corrupted service area, a failed firmware update, a power-delivery component on the M.2 board giving up, physical or liquid damage to the board, and long-term heat stress.

In all of these, the NAND is frequently untouched. Every page of your data is still sitting there. What has been lost is the machinery that knows how to find it. That is a very different problem from a drive whose flash has genuinely worn out, or one that has been secure-erased or reformatted and then left powered on while garbage collection did its work. In the second group, the data really is gone, and any lab that tells you otherwise is guessing.

Our approach on SSD and NVMe recovery is always to try to bring the drive's own controller back into a state where it can read for us, using vendor-level technical modes. Any board-level or component work we do is done solely to make the media readable long enough to image it. You receive your recovered files, not a repaired device.

When the controller cannot be persuaded to talk under any conditions, the remaining route is chip-off. Most shops that advertise SSD recovery cannot do this in-house and subcontract it, often overseas. We do it on our own premises, and nothing leaves Singapore.

The physical part is the easy part. The NAND packages are desoldered under controlled heat, cleaned, and read on a specialised reader that produces a raw dump of every die. That dump is not files. It is not even close to files.

The real work is everything after:

• Descrambling. Controllers whiten data with a pseudo-random pattern before writing it, to keep charge distribution even. That pattern has to be identified and reversed.
• Error correction. Each page carries spare bytes holding an ECC code. The scheme is vendor-specific, and modern TLC and QLC dumps arrive with a real error rate that must be corrected before anything parses.
• De-interleaving. Pages and blocks are distributed across channels, dies and planes in a pattern that has to be worked out empirically for that controller family.
• Rebuilding the translation layer. The map has to be reconstructed from metadata scattered through the dump so logical blocks can be reassembled in order, and only then can a file system be mounted.

Where it stops: chip-off is destructive, so the drive does not go back together afterwards. Some designs place controller and flash in a single package, which removes the option. Newer high-density 3D NAND with proprietary firmware is genuinely harder than the drives of five years ago. And then there is encryption, which is a wall of a different kind.

We will be blunt, because this is where people get misled. Nobody can read encrypted data without the owner's own password or key. Not us, not anyone. We do not bypass encryption, passcodes or account security, and any lab that claims otherwise is not telling you the truth.

Three situations come up constantly:

• BitLocker. If the Windows volume was protected, the raw NAND is ciphertext. Recovery needs the 48-digit recovery key or the password. That key may be in your Microsoft account, your organisation's Entra ID, a printout or a saved file. Find it before you do anything else, and see our BitLocker recovery page for where to look.
• Self-encrypting drives. Most modern NVMe SSDs encrypt everything at hardware level by default, whether or not you ever set a password, and the media encryption key lives inside the controller. If that controller is dead, a chip-off dump is encrypted noise even though you never switched anything on.
• Apple silicon Macs. The flash is soldered to the logic board and the keys are held by the Secure Enclave in the SoC. The chips on their own mean nothing. Recovery has to involve the original board, which is why we handle these differently on our Mac data recovery service.

This is why we ask about encryption at intake rather than after the invoice. It changes whether the job is worth attempting at all, and you deserve to know that on day one.

Gen 4 and especially Gen 5 controllers move far more data and draw noticeably more power than the Gen 3 drives most people still picture. That energy comes out as heat, concentrated in one small chip. Fitted without a heatsink, tucked under a graphics card, or squeezed into a thin laptop or a small-form-factor case, a Gen 5 drive will throttle to protect itself. Throttling is the drive behaving correctly. The concern is what years of sustained heat and repeated thermal cycling do to the controller and its solder joints, and what storage heat does to charge retention in NAND that is sitting unpowered.

Three practical habits: fit the heatsink and give it airflow, do not treat a fast NVMe drive as a cold archive left in a drawer, and be careful with cheap plastic USB enclosures, where the bridge chip and the drive cook each other. Enclosure failures are extremely common and often look exactly like drive death, which is why we test the drive separately from its caddy on every external drive recovery job.

As for expectations: no honest laboratory will quote you odds over the phone for an NVMe drive it has not seen. What we will commit to is the process. Diagnosis is free. You get a fixed written quote before any work begins. You see the list of recovered files and confirm it before you pay, and on standard cases it is no recovery, no fee. Everything is handled in line with Singapore's PDPA, and NDAs are available on request. Typical ranges for other work are set out on our data recovery cost page.

We have been doing this in Singapore for more than 20 years, across 140,000 recoveries, from our own cleanroom on our own premises. If your drive has gone quiet, leave it powered off and get in touch.

• Data Recovery Singapore
• Data Recovery Cost in Singapore
• Hard Drive Recovery in Singapore
• SSD Data Recovery in Singapore