I have deployed industrial microSD cards, swissbit offers some for which you can choose if they should run the flash cells as mlc or slc trading capacity for endurance.
When you deploy some controller module in a water pumping station in a mountain, you gladly pay 25€ for a better card to make sure you don't have to fly out a technician who knows what to do to restore that module because the flash thing failed because of logging data.
SD card flash quality and specs are a huge rabbit hole, no info about tlc/qlc or static/dynamic wear leveling is available for consumer cards, its pretty annoying.
Wow! Only about 100X more expensive than consumer grade.
A simple quality control test I do with any new flash memory device is to reformat it. If it has a serious problem, it's better to find out up front.
Most cheap, generic memory cards from China fail this test --- often because half the flash memory advertised by the onboard software is physically missing.
For example, a card that advertises 32Gb to the OS may only have 16Gb of actual flash on board. It will work until the real physical capacity is exceeded.
Some newer cards have gotten sneaky enough to fake the formatting somehow *and* fake the labeling to mimic a name brand. Apps are available to extensively test by writing random data and reading it back.
Simply buying from Amazon offers somewhat of a credibility buffer. Vendors selling fake product are quickly excluded due to the sheer volume of returns/chargebacks.
The testing I did back when I did this sort of thing was to put on a filesystem stress test automatically on boot, and then plug the whole system into a lamp timer set to hard cut the power every 15 minutes. I went through _dozens_ of cards, until I finally bit the bullet and went with the “100x more expensive than consumer grade”-grade from digikey. Every other option eventually failed, in the system-no-longer-detects-any-storage-in-the-slot sense of failure.
It sounds like the abrupt power cut test eventually produced a voltage spike that physically damaged components on the card.
Ever consider that your system's power or card interface design might be to blame as much as the card?
For a lot of applications, spike suppression hardware components or a simple UPS might be a more cost effective investment that could also provide other ancillary benefits.
Perhaps what we have here is some enterprising company that has incorporated a few cheap spike suppression components costing pennies each inside the card itself --- and now sells the cards as "industrial grade" for 100X as much. I would bet money that they haven't developed their own "industrial" flash memory chips.
They don't publish specs on consumer cards, so it's all speculation without heavy reverse engineering.
There seems to be people saying that it's not just voltage spikes, it's power failure corrupting data at the FTL layer, that the card wasn't designed to recover from.
If it were up to me, SD based devices like the RasPi would have a tiny capacitor, diode, and supervisor circuit, so that the SD could never lose power until at least a few ms after the CPU did, but I'm not the one in charge of such things, all I can do is spec a UPS or an industrial card.
Or more likely, just keep the logs in RAM as much as possible and use fatrace to hunt down various card destroying daemons.
“The flash memory devices we studied in this work demonstrated unexpected behavior when power failure occurs. The error rates do not always decrease as the operation proceeds, and power failure can corrupt the data from operations that completed successfully. We also found that relying on blocks that have been programmed or erased during a power failure is unreliable, even if the data appears to be intact.”
Yes, the hardware it's plugged into could also solve (or at least alleviate) the issues. It's not really useful when you're provided the hardware and told to make it work. “Get the manufacturer to acknowledge the issue, release a new revision, and then start replacing installed hardware in the field in 6 months when you start receiving it” is not preferable to “spend an extra 100 dollars on a 25 thousand dollar electrical/instrumentation installation, on an sd-card that has a demonstrated ability to deal with the hardware we have today”. I've literally spent more money on a simple push button, of which I needed dozens on such an installation (for example: https://www.digikey.ca/en/products/detail/schneider-electric... ). :)
I'm sorry, are you suggesting that extensive testing and certification for aerospace/medical/military/etc. use can be replaced by 'just formatting' the cards?
I'm suggesting that only those who really have a need for it should pay 100X the price.
For most others, buying reputable consumer grade and testing yourself might be a more reasonable option to eliminate a significant source of failures. You be the judge.
I’ve been learning recently about flash memory and NAND in particular, and I was surprised to learn that SD Cards contain software, that different memory types (SLC vs MLC vs others) have very different life span and all these variables is what makes an “industrial” flash so. Not just higher temperatures tolerances, but the flash and software itself.
When you deploy some controller module in a water pumping station in a mountain, you gladly pay 25€ for a better card to make sure you don't have to fly out a technician who knows what to do to restore that module because the flash thing failed because of logging data.
SD card flash quality and specs are a huge rabbit hole, no info about tlc/qlc or static/dynamic wear leveling is available for consumer cards, its pretty annoying.