Open a microSD card (conceptually, please) and you won’t see any components, just one thing, the “monolith”: controller and flash memory all fused into a single resin-sealed slab. This is why they can survive washing machines, but when they do die, there are barriers that traditional recovery will not get past; only chip-level work does. It was in this lab that we helped pioneer chip-level work in the UK, and here’s what it actually involves.
Inside each microSD is a single resin-sealed sliver. Here is essentially what you're doing when you read it "directly".
In most cases when a person says they have "dead" cards, they don't; instead, the card's file system was corrupted (for whatever reason) but the pictures remain as full images on the card, and they can be restored at the file system level. When a card actually dies, either the electronic portion of the memory controller has failed or the outer plastic housing has cracked open, and the card is essentially dead as far as a reader, a computer, or any type of software is concerned, since the machinery used to present the chip to the outside world is destroyed. But although the electronics that made the card "alive" are lost, very often the flash chips themselves survive: the passive nature and ruggedness of NAND chips means that while everything else is failing around them, those little chips are still happily storing data within the resin until some other method of accessing the data becomes available.
The other way in is the monolith's own service interface: micro test points underneath the resin that were installed at manufacturing so that a factory can program them. Chip level recovery maps these test points, sanding off layers of coating under an optical microscope to identify the pinout for that monolith family, and then reads the raw NAND, completely bypassing the dead controller. This information is designed to be impossible to read because the controller scrambles the data across the flash with wear levelling, interleaves data together, adds XOR patterns and adds vendor specific error correction codes. To complete the process, we must reverse engineer what the controller did and mathematically undo the scrambling, recreating the logic behind how the controller scrambled the data until the files can be rebuilt. This is the same type of work done when recovering the Vaio SSD at a much larger scale, and it has been in our toolkit since the technique's early UK days.
Three takeaways are travelling home from this. The snap of a card is an inquiry, not a verdict: all that really matters is where the card broke, and that's what your free look will tell you. "Nothing can read it" is simply to say that the regular avenues have been depleted, not the extraordinary ones: second opinions exist precisely for the gap between shops and benches. Lastly, prevention (in this case backing up) continues to be the cheapest option, since cards are the only type of media most people carry without back-ups during events such as weddings, holidays and once in a lifetime events. The memory card page ends every answer the same way this post does: back up the card tonight.
An unreadable card? Snapped, silently ignored or non-existent? It’s free to diagnose; the quotation for the work is fixed and always written down, and honest advice can be had by contacting us (no charge) on 0131 202 0491.