USB flash drives and memory cards are among the most widely used storage devices in the world — carried in pockets, slotted into cameras and distributed across offices and homes without a second thought about their longevity. When they fail, they do so abruptly and without warning, and the nature of flash storage means that the usual recovery approaches for hard drives and SSDs often do not apply. Understanding how these devices are constructed, why they fail and what recovery is realistically possible is important for anyone storing data they cannot afford to lose.

Why Flash Media Fails Differently From Hard Drives

A hard drive contains spinning platters and a read/write head on a mechanical arm. When it fails, there are physical symptoms — clicking sounds, spinup failures, head crashes — that give a technician meaningful diagnostic information before any recovery attempt begins. The failure mechanism is mechanical, and the data pathway through the platters remains structurally the same whether the drive is working or not.

Flash media works on a completely different principle. USB drives, SD cards, microSD cards and compact flash cards store data as electrical charges in NAND memory cells — silicon structures that hold a charge state representing one or more bits. There are no moving parts, no heads, no platters. The devices are silent, shock-resistant and small. But the trade-off is a failure model fundamentally unlike spinning storage.

Every flash device contains two distinct functional layers: the NAND memory chips, which physically store the data, and a controller chip, which manages all communication between the NAND and the host device. The controller handles wear levelling, error correction, bad block management, encryption and the flash translation layer — the mapping system that tells the device where to find data inside the NAND. When the NAND fails, the data is physically gone. When the controller fails, the data is often completely intact inside the NAND chips but entirely inaccessible through normal means, because nothing can communicate with the memory correctly.

This distinction — between NAND failure and controller failure — is the single most important concept in flash media recovery. The two failure types require entirely different approaches and produce very different outcomes.

NAND Flash Types and Endurance: SLC, MLC, TLC and QLC

Not all flash memory is equal. NAND flash is manufactured in four configurations based on how many bits are stored in each memory cell, and the number of bits per cell directly determines endurance, speed and failure behaviour.

Consumer SD cards — including most sold under trusted brands at standard retail price points — use TLC or QLC NAND. For a photographer or videographer who regularly deletes footage and shoots new content, write cycles accumulate quickly. For a 128GB TLC card used intensively in a professional video workflow, total write endurance may be exhausted within one to three years of heavy daily use. Budget and no-name flash products frequently use lower-grade NAND than their labelled specifications suggest, or source NAND from rejected wafers that fell below quality thresholds in manufacturing. These drives tend to fail earlier and more catastrophically than branded alternatives.

How USB Flash Drives Fail: The Four Main Failure Modes

USB flash drives are the most mechanically vulnerable flash storage format in common use. The USB-A connector — the standard rectangular plug found on most drives — is rated for a finite number of insertion and removal cycles, typically 1,500 to 10,000 depending on manufacturing quality. Drives that are regularly plugged and unplugged, or that are subject to lateral stress while inserted in a port, will eventually develop connector damage. This is the leading physical failure mode and accounts for a significant proportion of drives arriving at recovery labs.

Beyond connector damage, USB drives fail in three other primary ways. Controller failure — caused by power surges, manufacturing defects, or voltage instability during write operations — renders the drive completely unresponsive to the host OS. The drive may not appear in Device Manager at all, or may appear briefly as an unknown device before disconnecting. The NAND chips in these cases are often completely intact and fully readable with the right equipment. NAND wear and cell degradation occurs as cells are written and erased repeatedly, gradually losing their ability to hold a charge reliably. The controller’s error correction compensates for increasing bit errors, but once error rates exceed the ECC threshold, data integrity breaks down — files fail to open or return corrupt data even though the drive appears to mount normally. Logical corruption — caused by unsafe removal, interrupted writes, malware, or sudden power loss — damages the partition table, FAT or directory structure without touching the underlying data. The drive may be fully readable at the block level; the data is present but mismapped or inaccessible through the file system.

SD and MicroSD Cards: The Monolithic Construction Problem

Standard SD cards, in their original PCB-based construction, contain a separate controller chip and one or two discrete NAND chips soldered to a small circuit board inside the plastic housing. The NAND can, in principle, be removed and read externally if the controller fails — the same chip-off process used for USB drives and compact flash cards — giving a recovery lab meaningful options when the controller is the problem.

MicroSD cards, and most modern high-capacity full-size SD cards, are built differently. They use monolithic construction: the controller and NAND memory are integrated into a single inseparable die, sealed in one package. There is no separate chip to desolder and read. This architecture is cheaper to manufacture and produces a smaller device, which is why it dominates the microSD market — but it creates a fundamentally different and far more difficult recovery problem when the card fails.

Compact Flash and CFexpress: Professional Camera Cards

Compact Flash cards, which dominated professional camera storage for two decades before the transition to newer formats, use a PCB-based design with discrete NAND chips and a separate controller. The larger physical format and more robust construction make CF cards more mechanically durable than SD or microSD cards, and their internal architecture gives recovery labs more viable pathways when they fail.

CF cards typically contain one to four NAND flash chips in BGA152 or TSOP48 packages, soldered to a PCB alongside the controller. When a CF card fails — whether due to controller failure, firmware corruption, or contact pin damage — chip-off recovery is the standard approach. The NAND chips are removed using a precision hot-air rework station and read individually using a NAND programmer. Because CF cards may contain multiple NAND chips, the raw data from each must be reassembled in the correct sequence, with XOR and scrambling parameters determined by the specific controller model.

CFexpress cards — the current professional standard in Canon, Nikon and Sony mirrorless cameras — use an NVMe interface over a PCIe bus rather than the PATA interface of older CF cards. Their internal construction is closer to an M.2 SSD than a traditional compact flash card. Controller failures on CFexpress cards present similar recovery challenges to NVMe SSDs: hardware-level encryption, TRIM-like zeroing behaviour, and complex FTL mappings that vary by manufacturer. For professional photographers and videographers relying on CF or CFexpress cards, a Lexar CF or CFexpress card that fails to mount requires professional diagnosis before any recovery attempt — the correct pathway depends entirely on whether the failure is logical, controller-based, or physical.

When Recovery Software Works — and When It Does Not

Consumer data recovery software — tools like Recuva, PhotoRec, EaseUS Data Recovery Wizard and TestDisk — operates at the file system level. These tools scan for recognisable file system structures, partition tables, deleted file entries, and file signatures in order to reconstruct what was lost. They are genuinely effective in one specific scenario: logical failure where the underlying storage media is physically healthy and fully accessible to the operating system.

If an SD card mounts on a computer, shows a drive letter, and the problem is accidental deletion or a quick format where no new data has been written to the card, consumer software can recover a meaningful proportion of the lost files. File carving — searching raw sectors for file header signatures regardless of the file system — is particularly effective for photo recovery from cards that have been reformatted without new data written to them.

However, consumer software fails entirely in several common flash failure scenarios. If a drive is not detected by the operating system — either because the controller has failed or the connector is damaged — software tools have nothing to scan. If a drive appears in Device Manager as 0 MB or reports an obviously wrong capacity, controller firmware corruption is likely; software that attempts to scan a 0-byte volume returns nothing useful. Once data has been deleted and the freed space overwritten — which modern OSes can do near-instantly on flash media through TRIM-equivalent operations — the data in those sectors is gone. Unlike magnetic HDDs, where overwritten data may leave residual traces, zeroed NAND cells contain no recoverable information.

Running recovery software on a device with NAND cell degradation or a deteriorating controller also carries real risk. Each scan cycle subjects the failing hardware to additional read stress, potentially triggering further failures in cells that were marginal but still readable. Professional labs image the device first — working from the image rather than the original — to protect whatever data remains from the recovery process itself.

Chip-Off Recovery: Bypassing the Controller

When a device’s controller has failed but the NAND chips are physically intact, chip-off recovery is the primary lab technique available. The process involves physically removing the NAND chip or chips from the PCB and reading them directly using specialised NAND programmer hardware, bypassing the failed controller entirely.

NAND chips are removed using a precision hot-air rework station — applying carefully controlled heat to reflow the solder connections while lifting the chip cleanly from the board. The chips must be removed without overheating: excessive heat damages the NAND cells themselves and can permanently destroy the data they contain. Once removed, residual solder is cleaned from the chip’s pads and the chip is placed in a socket adapter matched to its specific package type — TSOP48, BGA152, or another form factor depending on the device.

Reading the raw data from the chip is the straightforward part. The difficult step is reconstruction. The controller’s flash translation layer applies a mapping between the logical addresses the OS uses and the physical pages in the NAND. Without the controller, this mapping is lost. Additionally, most controllers apply XOR scrambling and sometimes proprietary data interleaving across multiple NAND chips. A recovery engineer must identify the scrambling parameters, page layout, block size, and interleave pattern for the specific controller model in order to reassemble the raw NAND dump into a coherent disk image. Professional platforms such as PC-3000 Flash contain databases of controller profiles that automate much of this reconstruction for known controller families. Unrecognised controllers require manual analysis — more time-intensive and with less certain outcomes.

Monolithic Flash: The Hardest Recovery Problem in Flash Media

Monolithic flash devices — the majority of microSD cards and many modern full-size SD cards — present the hardest recovery challenge in the flash storage category. Because the controller and NAND are integrated into a single sealed package, there is no chip to desolder and read externally. The only path to the data is through the NAND’s own test interface, buried beneath the epoxy encapsulant.

The recovery process requires grinding away the protective epoxy using a micro-milling tool or abrasive — under a microscope — to expose the test point pads that provide access to the NAND flash interface. The positions of these pads vary by card model and are not published by manufacturers. Labs working on monolithic devices rely on accumulated documentation of pad layouts for known card variants, as well as X-ray imaging to locate components inside cards where the layout is unknown.

Once the pads are exposed, hair-thin wires are micro-soldered to the test points and connected to a flash reading device. The raw NAND data must then be reconstructed using the same FTL analysis required for standard chip-off recovery. Monolithic recovery is expensive, time-consuming and not guaranteed. Critically, the process is largely irreversible: once the epoxy has been ground away and wires attached, the card’s physical structure is permanently altered. This makes it essential that all other recovery pathways are exhausted before monolithic access is attempted.

What This Means for Data Recovery

The practical implication of everything above is that flash media recovery is highly outcome-dependent on two factors: whether the NAND chips are physically intact, and whether the failure was caught before additional writes occurred. Of these two, the second is entirely within the user’s control.

The single most important action when a USB drive, SD card or CF card stops working is to stop using it immediately. Do not attempt to reformat it, run disk repair utilities on it, or copy files to it. Every write operation — including what the OS writes invisibly when it mounts a volume or runs chkdsk — risks overwriting data that would otherwise be recoverable. For logical failures, an untouched device that has had nothing written to it since the failure is significantly more recoverable than one that has been through multiple repair attempts.

For physical failures — broken connectors, water damage, or controller failure — professional intervention is required regardless. No consumer tool addresses these at the hardware level. Attempting to force a drive with a damaged connector into a port, or using a bent SD card in a reader with poor contact, risks cracking the PCB or NAND die and converting a recoverable situation into an unrecoverable one.

SouthBit recovers data from USB flash drives, SD and microSD cards, compact flash cards and CFexpress cards. The PC-3000 Flash platform covers a wide range of controller families — including Phison, Alcor, Silicon Motion, Innostor, JMicron and others — for both logical recovery and FTL reconstruction after chip-off. Monolithic microSD recovery is available on a case-by-case basis depending on card variant and failure type. See our pricing page for full details on flash media recovery costs. As with all our services, assessments are free and no fee is charged if the data cannot be retrieved.

Book a Free Assessment

Sources

• Kingston Technology — Flash Memory Guide: NAND Types and Endurance
• PITS Data Recovery — Micro SD Chip-Off Recovery: Retrieving Data from Damaged Cards, 2025
• Gillware Data Recovery — SD Card and MicroSD Card Data Recovery
• mdrepairs.com — What Is Chip-Off Data Recovery and When Is It Needed?, 2026
• Recover My Flash Drive — SD Card Recovery and Repair: Monolithic and PCB Variants