Longevity of SD data

[Connecteur]

Looking through some old stuff today, I found my old music player, a Panasonic SV-SD50.  I don't exactly know how long it was sitting idle, but I think it was at least 20 years.  It had in it, one of the biggest capacity SD cards I could afford at the time, 256 MB.

I've always been curious about just how long an SD card will hold data; I'd heard 10 years mentioned several times, but I'm sure that was mostly a guess.  Anyhow, I put an AAA battery into the player, and all the music was still there, intact.  I couldn't say if such longevity applies to more modern SD and micro-SD cards, but they did use larger dies back then, so I couldn't say for sure.  But I was surprised at how long this one retained data.

[Whales]

10 years is a common minimum spec.

Inside the SD card is NAND flash.   I presume this now means they use TLC and QLC?

[Connecteur]

This may or may not be interesting, but the player wouldn't turn on until I did several things.  The play button normally starts the player, but it did nothing. None of the other buttons seemed to work either.  But holding the play button while sliding the hold switch back and forth is what finally turned it on, and restored all controls.

[Infraviolet]

That weird turn on might well indicate nothing to do with the flash on which the song files are stored, but instead give some indications of the power switching having its state stored elsewhere. A lot of devices have "soft" power switches which act to switch on and of the main power via some intermediary logic or even full on microcontrollers, so if these were all fully drained by having a long dead battery then its not surprising it might take several cycles of actual power before anything would happen.

I assume after that unusual startup procedure, it has subsequently turned on and off entirely normally?

[amyk]

256MB is certainly going to be SLC flash, rated for 10 years of retention (at 25C) after 100K P/E cycles.

Then various generations of MLC came and it was 5 years after 10K, 5K, or 1.5K cycles depending on the generation (65nm, 50nm, 34nm, 25nm, etc.)

These days, it's around 1 year after a few hundred cycles for QLC.

[ataradov]

The specified value is usually conservative and the longevity is specified over the whole temperature range. Operating at a room temperature, most devices would easily retain 10x the specified value.

[Connecteur]

That weird turn on might well indicate nothing to do with the flash on which the song files are stored, but instead give some indications of the power switching having its state stored elsewhere. A lot of devices have "soft" power switches which act to switch on and of the main power via some intermediary logic or even full on microcontrollers, so if these were all fully drained by having a long dead battery then its not surprising it might take several cycles of actual power before anything would happen.

I assume after that unusual startup procedure, it has subsequently turned on and off entirely normally?

It appears to operate normally now, but it didn't want to turn on at first. It was found with no battery, so I assume it was removed before it went dead.  I don't know what it may have had to do with the SD card.

[Connecteur]

I don't believe this SD card had a large number of read/write cycles.  As I recall, I only filled it once, and played it a few times before I began using other devices.

I believe I bought the device around 2002, and used it for a year or two before I put it away.  I do recall that the Panasonic software converted music into a proprietary format, that can't be shared on any other device.  As far as I know, both the software and the proprietary music format are long gone.  I was unable to access any of the music files on the SD card.

[tom66]

Longevity of flash memory is strongly influenced by write cycles.

If you only write a few times to an SD card it is highly likely the data will be there in 50 years time (at least for a modern card kept at normal temperatures.)

As cells as written they wear and they leak charge.  This effect is cumulative.  For modern QLC cards, the charge leakage only needs to be very small to corrupt a cell.  You are then relying on the ECC to recover data for you.

A modern card will refresh flash pages that it detects are weak even if those pages are not written to, but this is no good if the card is not connected to power. 

[Connecteur]

Just to check the content integrity, I went through all the tracks on the player.
A few tracks wouldn't play, and the player skipped them.
I couldn't say if they were corrupted from the beginning, or if it happened over time.

[Infraviolet]

"f you only write a few times to an SD card it is highly likely the data will be there in 50 years time"
Does that mean 50 years of the card just sitting at room temperature in a low humidity environment with no sources of serious electrical noise nearby, or would that need the card to be connected to power every year, or more often, as a means to let the onboard "top up the charge" behaviours rewrite any weakened pages?

[ataradov]

Cards will not perform any rewrites on their own. The only way to force a rewrite is to actually trigger it over the interface.

Storage in an unpowered state is better for the longevity, since simply being powered on may result in trapping of the charges and data corruption.

Even better is to power up it once a year (or month, or week, depending on how paranoid you are) and do a full rewrite. This will force all the wear leveling algorithms to reallocate the data. And writing even weekly will not result in a significant wear.

[wraper]

I have 64GB Plextor Patriot SD card I bought a decade ago, and it starts corrupting the data after about a year since it was written. And it has basically no wear, I don't think I wrote its full capacity even once.

[Connecteur]

I think there's a difference between longevity of the data, and longevity of the storage medium.
I don't expect that the SD flash memory would degrade over time, and I didn't read and write to that card a bunch of times either.  Far as I know, I bought the card some time back in the early 2000s, wrote the playlist to it, and then it sat in storage for about 20 years.

I was wondering if anyone else had any experience or information about long term data storage on SD flash memory cards.  I know that capacities have increased by orders of magnitude over the intervening years, so I wonder if the newer one are likely to be better of worse than the card in question.  I'm not going to assume that newer is always better.

[tom66]

In general, smaller flash cells as used in modern flash memory will be less reliable for long term storage.   It is possible that some aspects of the technology have improved with respect to leakage.  But you have a fundamental physics problem with a small floating gate.  You are storing not that many electrons on that gate, and it doesn't take much leakage for those electrons to change the apparent state of the cell.

This is especially problematic for technologies that use the cells to store more than one bit.  The majority of consumer SD cards are at minimum TLC but most are QLC. So you need to distinguish between 8-16 different levels on the cell.  It takes even less leakage to change a cell from 1111 to 1110 than it does to change a cell from 1 to 0.

The ECC algorithm will correct the odd error, but it is only going to be able to detect so many errors.  It will eventually run out of redundant bits and corrupt data will be output, or even worse, the card may prevent you reading that sector altogether.

The other factor is how many writes have been done to the media.  If the cells have only been written once and then the card is stored I would expect the data to last longer than if the cells were written hundreds of times (e.g. digital camera operation) then the card is stored.  The SD controller will balance writes across the card reasonably well but this will mean if any data corruption does occur it will be likely to occur equally across the card.

[amyk]

Temperature is also a huge factor in data retention. The relationship is exponential. The hotter the storage temperature, the more quickly the electrons will leak out.

PLC (32 voltage levels!) flash is apparently starting to appear in the cheapest memory devices, but the manufacturers are obviously very secretive about it.