Nothing else in a computer stops this suddenly. There is no warning noise and no slow decline, so the whole of the notice you get is a machine that was fine yesterday reporting an error today where the drive used to be. M.2 sticks and PCIe cards arrive from households, from gaming builds put together in spare rooms, and from the offices and workshops around Leicester and Loughborough. Behind nearly all of them is a controller that has stopped talking rather than memory that has worn out.
Every NVMe that arrives here is examined at no charge. The figure quoted afterwards is put in writing and settled before anybody reaches for a screwdriver: £300 + VAT on any one hard drive or SSD, an NVMe blade counted the same.
Logical recoveries carry no fix, no fee. The named exceptions to it are electronic and mechanical failures, chip-level work, DVR jobs and forensic jobs; physical work takes 50% up front. The five bands are set out in full on the data recovery cost page.
Tracing a symptom back to the failure underneath it is where the real work begins, and these thirty account for all but a handful of the boxes opened at the Cambridge bench. A fault missing from the list is not an unfamiliar one — describe it on the telephone and you will get a straight view of the odds before you spend anything on postage.
Present at lunchtime, missing by the evening, with nothing in between to explain it. That shape of failure belongs to the controller or the mapping layer, not to worn-out memory. The module goes quiet on the PCIe bus and the flash behind it stays exactly as it was written.
Invisible at both levels means the module is not negotiating a link at all. Power delivery, a controller stuck in reset, or a firmware area that will not load are the three candidates. None of the three involves the NAND, which is why a module nobody can see is a better prospect than it sounds.
The controller has lost its translation table and is falling back on a default identity. Rebuilding that table restores both the real size and everything addressed through it. This is one of the more common arrivals and one of the more straightforward outcomes.
Bare modules throttle when they get hot, and past a certain point they stop answering altogether. Thin laptops and small-form-factor builds produce a steady flow of exactly this through the summer months. Repeated thermal cycling on a module that is already unstable leaves the mapping tables in a worse state each time.
The controller ends up holding two halves of two different versions and refuses to come up. Retrying the update on a module in that condition rarely helps and occasionally finishes the job off. Service-mode entry restores a coherent firmware state, and the image comes before anything else is attempted.
The boot partition sits in a small handful of blocks and a fault at that end removes the whole volume from view. The temptation is to reinstall Windows so the machine can be used again. That writes over the exact region a recovery would begin from, so the module comes out first.
TRIM operates on NVMe exactly as it does on SATA flash, and it operates whenever the module has power. Every minute the machine spends switched on after a deletion is another minute of blocks being genuinely erased. Shut it down fully, not to sleep and not to hibernate, and leave it off until the module is posted.
Heavy internal error correction on nearly every read. The module is still handing over correct data and working hard to do it. That is the ideal moment for a full image and the worst possible moment to carry on using the machine normally.
Storage fixed permanently to a logic board is outside what is done here, and it is better to know that before the postage than after. Apple Silicon machines, a number of thin Windows ultrabooks and most tablets fall into this category. A standard M.2 module held by one small screw is an entirely different proposition.
Marginal power delivery or a controller struggling to initialise. Each attempt has a chance of succeeding and each attempt also carries a risk. Rather than gambling on the tenth try at home, it gets imaged on the first successful initialisation at the bench, with the retries controlled in hardware.
M.2 slots come in SATA-only, NVMe-only and dual varieties, and a module that has never worked in a particular machine may simply be in the wrong socket rather than faulty. It costs nothing to have that established, and it is one of the few outcomes here that ends with the customer keeping their money.
A module that negotiates at reduced width, or fails to negotiate at all, is usually looking at a damaged contact or a marginal capacitor rather than dead memory. Cleaning, reseating and inspection under magnification settle it, and a full image follows immediately if the link comes back.
The bridge chip in the caddy is a separate component with its own failure modes, and a dead bridge presents as a dead drive. Sending the module together with its enclosure lets the bridge be ruled in or out in ten minutes rather than guessed at.
The commonest genuine hardware fault on this form factor. What has stopped is the part that speaks for the flash. Where the manufacturer's service mode can still get a word out of it, the module is read normally. Where it cannot, the packages come off the board and are read directly.
Data is interleaved across every package on the stick, so a single failed package removes a share of nearly every file rather than one file in eight. Internal error correction covers some of that where the controller still runs. Where it does not, the reconstruction has to reproduce the interleave precisely.
The 2280 form factor is a thin card held at one end, and it does break. What matters is whether the fracture ran through the packages or through empty board. The assessment will say which within the two working days, and it will say it in plain terms rather than as a maybe.
A shorted rail on the mainboard, or a power supply that failed badly, can put the wrong voltage down the slot. The controller takes it and the NAND usually does not. Microscope rework and a chip-off read cover this, and electronic and chip-level work both fall outside no fix, no fee.
Many NVMe modules encrypt everything as a matter of routine, holding the key in the controller. That is invisible while the drive works and decisive when it does not, because reading the raw packages then produces perfectly recovered ciphertext. If the controller can be revived the files come back. If it cannot, that is said honestly.
Both commands change the internal key rather than wiping cells, and the effect is immediate and total. If the command ran to completion the answer is no, and no laboratory will tell you differently. If it was interrupted early there is occasionally a partial result, and establishing which happened costs nothing.
Disk Management proposing to write a fresh signature and an empty partition scheme. The module answering at all is good news, because the electronics and the firmware area are clearly alive. What is missing sits further in and gets rebuilt from the file system structures rather than replaced by a new one.
Mapping tables are held in volatile memory and written back in batches, and losing mains during that write leaves the module unsure where anything is. Consumer modules rarely carry the capacitors that would let them finish the write. The volume comes back once the tables are reconstructed on the bench.
Counterfeit sticks are shipped with the firmware reporting a size the flash cannot physically hold. Everything behaves until the writes pass the real limit, at which point the earliest data is quietly overwritten. What comes back is whatever genuinely fitted, and the rest never existed on the module at all.
Modules configured as a cache in front of a larger volume are not holding whole files in any readable arrangement. Recovery has to take the cache and the backing store together, so both need to arrive, and it needs to be said clearly on the booking form which is which.
Sustained writes at high temperature in a case with poor airflow are hard on a bare module. Corruption with no pattern to it, appearing under load and clearing after a reboot, is the classic sequence. It gets imaged before it deteriorates further, and the image is what everything else is worked from.
Software striping across a pair of NVMe drives is quick and unforgiving, because losing either one loses the volume. Both modules have to arrive and both get imaged before the stripe is reassembled from the copies. Send them labelled with which slot each came out of, since that saves guessing at the order.
Expansion storage from a recent console is standard NVMe wearing a heatsink, and it is read here in the ordinary way. What is stored on it is frequently encrypted against the console itself, so the honest position is that the data comes off the flash and what can be done with it afterwards depends entirely on that.
Firmware compatibility between certain controllers and certain mainboards is genuinely patchy, particularly on older boards that gained NVMe support through an update. Before anyone concludes the module has failed, it gets tried on known-good hardware, and a fair share of these end there.
Charge leaks out of flash cells over time and leaks faster from a module that has seen heavy writing. Unpowered storage is not archive storage. Error correction on the bench recovers a good deal of what has drifted, but the result depends on how long it sat and how hard it worked before it did.
Feature updates occasionally move a user profile, leave the old one in place under a different name, and present the machine as empty. Nothing has been deleted. This is one of the more cheerful outcomes on the list and it is resolved by examining the image rather than by anything done to the module itself.
A laptop that will not power on tells you nothing about its storage. Where the module unclips, it travels on its own in a small padded envelope and the rest of the machine stays with you. That is the cheapest parcel this laboratory receives and it accounts for a good number of the happiest outcomes.
An NVMe drive talks to the processor straight down the PCIe lanes with nothing mechanical in the way. No spindle, no arm, no platter. That is where the speed comes from and it is also why the ending is so abrupt. Inside the package a controller runs a large firmware layer over stacked memory and maintains a running record of which physical block currently holds which part of your data. Break either the controller or that record and the module stops answering, while everything written to the memory carries on sitting exactly where it was put. Getting back to it takes one of two routes. Most controller families still respond to the diagnostic mode their manufacturer uses on the production line, and that is always tried first because it leaves the drive in one piece. Where the controller has gone past answering anything, the memory packages are lifted off and read on their own, after which the page order, the interleave and the error correction the original controller applied all have to be reproduced before a single file appears. Neither route is helped by another twenty rounds of switching the machine on and off, however tempting that is at eleven at night.
The rule here is physical rather than technical. Anything that comes out of a slot and fits in a padded envelope gets worked on, and that covers 2280, 2260, 2242 and 2230 sticks, add-in cards for a full PCIe slot, modules living in external caddies and console expansion drives, in every generation from Gen3 to Gen5 and under any badge you care to name. What is turned away, without exception, is flash soldered onto a logic board. Heat deserves a paragraph of its own because it causes more of these failures than wear does. A bare module driven hard with nothing sitting over it throttles first and drops off the bus second, and a stick that has been round that cycle a few dozen times tends to arrive with its mapping tables in poor order. Compact builds and thin laptops produce a steady dribble of exactly that between June and September. One more thing, and it is the most time-sensitive line on the page: TRIM runs on NVMe whenever current is going in, so if something has been deleted, shut the machine down fully and leave it off until the module is in the post.
Modules arrive in an envelope and get treated the same way a full-size drive would, which means read-only handling from the first contact onwards. The benches below cover everything from a stick that simply needs coaxing back onto the bus to one where the packages have to come off the board.
Phison, Silicon Motion, Marvell, Samsung and Realtek controllers all have a production diagnostic path, and getting a stalled module back onto the bus through it is the least invasive route available. It is tried first every time, because it leaves the module whole.
2280, 2260, 2242 and 2230 sticks, add-in cards, Apple blade connectors and the external USB bridges. A module that will not talk to your mainboard frequently talks perfectly well to a bench adapter, and establishing that takes ten minutes.
Modules that drop out when warm are imaged with active cooling and in short passes rather than in one long run. It is unglamorous and it is often the whole difference between a partial image and a complete one.
Where the controller cannot be revived, the packages are removed under controlled heat and read individually, and the page order, XOR scheme and error correction are reproduced in software. Chip-level work sits outside no fix, no fee and takes 50% up front.
Cracked substrates, lifted pads and burnt rails repaired at magnification, enough to bring a module back up for as long as it takes to produce an image. Electronic work is one of the published exclusions from the guarantee.
Every adapter and every port has blocking in front of it. On flash this matters more than on a spinning disk, because an operating system that mounts a module read-write can begin trimming blocks the moment it recognises the file system.
OEM modules such as the Samsung PM981 and the Kioxia BG4 outnumber retail sticks on this bench, simply because that is what laptop makers fitted at the factory. Any module that unclips is worked on: 2280, 2260, 2242 and 2230, Gen3 through Gen5, in a laptop, a desktop, an external caddy or a console. What is refused, every time, is flash soldered onto a logic board. The price is 300 pounds + VAT for one module, exactly as for a SATA SSD or a hard drive, with the assessment free and closing two working days after arrival. Chip-off work takes 50% up front and falls outside no fix, no fee. The one piece of advice worth repeating: if files have been deleted, switch the machine off properly and leave it off, because TRIM only works while there is power going in.
The post office does most of the work of getting a job here. A drive that is already unwell travels better boxed and insured than rattling around a car for a day of errands, and something dropped into a Leicestershire postbox this afternoon is generally logged in at Cambridge tomorrow.
The general rule is the drive travels and the machine stays behind — out of the laptop, out of the tower, out of the iMac, out of the recorder under the counter. This bench does not dismantle equipment, and a repair shop will do it while you wait. Three things are the other way round, and getting them wrong costs you the recovery: an external drive stays sealed in its own case, a NAS comes as a complete unit, and a WD My Passport or My Book travels whole with its cable, because on those the encryption key is held on the bridge board rather than on the disk — separate the two and the data becomes unreadable even to us. A Fusion Mac needs both of its drives, each labelled. The one thing nobody can work round is flash soldered onto the mainboard, as on Apple Silicon machines: if it will not come off, there is nothing to post.
↓ Print the shipping & booking-in form (PDF)
Address it to Cambridge Data Recovery. It is about seventy miles from Leicester if you fancy driving it — M1 south to Junction 19, then the A14 east — and the lab is two minutes off Junction 32 with parking at the door. Posting costs you a stamp and a day instead. Whichever you choose, you hear from us the moment it is booked in, and the free diagnostic closes two working days after that.
Not certain what belongs in the box? Ring 0800 689 0668 before you tape it up, or let the free online diagnostic ask the questions for you.
The examination is free, one written figure follows it, and the band covering this page is £300 + VAT on any one hard drive or SSD, an NVMe blade counted the same.