It would seem that some of my ancient 10+ year old SSDs in the 60-128gb range are experiencing sudden death after sitting around for years. They're basically bricked and can't be written to.

EDIT: This should be a PSA. SSDs are not for cold storage.

@Lydie
Have they been unpowered? I never quite managed to find a definitive answer on if/how often modern flash needs to be refreshed, and if/when the controllers actually do it.

@srtcd424 Yup, unpowered. Just old and in a box. I'm sure not every drive would brick, but certain controller types may if the NAND is too dirty. So PSA stands, because you can't know about your particular units.

@Lydie
Yeah, I think one of the studies I did manage to find said the problem mainly manifested with very 'worn' cells. Turns out us old timers were right about magtape being the only true long term storage, huh? :)

@Lydie
SSDs are rated to retain data unpowered for six months for consumer grade devices, and for three months for enterprise grade. They need to stay powered up so that they can scrub (refresh) memory cells that lose their charge over time. This is because each data bit is actually stored by a surprisingly small number of electrons in modern NAND flash devices.
Spinning drives, despite their potential mechanical failure modes, are actually better for cold storage than SSDs.

@brouhaha

Oh wow. I've been suspecting something like that, I had no idea the interval was so short.

I've made sure that one of my machines has a spinny rust drive. I will probably add one more.

Thanks for that!

@Lydie

@tomjennings @Lydie
As I just wrote in another reply, that's the rated minimum over the full temperature range, so in reality it generally isn't quite that bad. It's sort of the storage industry's deep dark secret. It's not actually a secret, but the manufacturers definitely do not want to call attention to it.
NOR flash has historically been rated for decades of retention, so many of us incorrectly assumed that ultra-high-density NAND would be similar, but unfortunately it isn't.

@Lydie

How strange.

I would expect the charges to have dissipated and reset everything after 10 years of cold storage…but I would expect that to leave the SSD blank, not unwritable.

@brouhaha @Lydie

Wow, this will have huge ramifications for historical data retention. Worlds will disappear, mostly small scale and at-home stuff.

@tomjennings @brouhaha @Lydie a lot of random other flash just degrades too. USB sticks, SD cards, even on device firmware slowly dribbles it's brains out.

@highvoltage @brouhaha SLC NAND is pretty darn resilient!!

@argv_minus_one @Lydie There are descriptors stored in flash, and if those become unreadable, there's no way for the controller to know what it's working with; I remember when somebody read raw flash chips from some SSD that the descriptor page was repeated 512 times to minimise the risk of it being completely unreadable, but it still happens.

@jernej__s

Then should it not still allow you to SECURITY ERASE UNIT and write fresh descriptors in the process?

@Lydie

@Lydie

Meanwhile, a hard disk drive from 20 years ago probably still works perfectly. Ever since the stiction problem was solved, HDDs have been way more reliable than such a delicate and precise machine has any right to be.

@argv_minus_one @Lydie Depends on whether what's left of the firmware's enough to make the drive visible on the bus.

@jernej__s

Wait, the firmware is also stored in flash?

Because if the firmware goes corrupt, then yeah, drive is kaput. But this somehow doesn't seem to apply to HDDs, which last a stupidly long time.

@Lydie

@argv_minus_one @Lydie Of course the firmware's stored in flash (on hard drives it used to be stored on disk itself; I'm not sure if this is still the case).

@jernej__s

Why the hell are mask ROMs not still a thing? 🤦‍♂️ They last for like a century.

@Lydie

@argv_minus_one They're all good! We're all 44 years old, functions better than me! 😆

@Lydie

Beautiful and impressive!

How'd you keep them from sticking? Spin them up once a year?

@argv_minus_one @Lydie Because the firmware has to be upgradable.

@jernej__s

On an SSD? What in the world for?

@Lydie

@etchedpixels @tomjennings @brouhaha @Lydie We've been discovering at the museum that it's often harder to get old computers from the 90s running than stuff from the 60s or 70s, because of all the batteries and nvram everywhere.

@argv_minus_one @Lydie I remember security fixes from several vendors (for self-encrypting drives), reliability updates (these were quite important when TLC drives were new), recently there were updates for several WD models that outright didn't work with Win11 24H2, or where 24H2 caused excessive wear.

@jernej__s

Self-encrypting is pointless. Trusting the security of a self-encrypting drive is like trusting the durability of a Cheeto serving as a door-locking bolt. Encryption belongs on the CPU.

I guess reliability updates are well and good, but putting the firmware on flash is itself bad for reliability, so…

What exactly was the problem with 24H2? Was it really a drive firmware bug, or did Microsoft screw up?

@Lydie

@argv_minus_one @Lydie (speaking of WD/Sandisk, they still haven't fixed their DRAMless drives in native 4k sector mode)

@etchedpixels @tomjennings @brouhaha @Lydie petition to make "dribbles its brains out" an industry standard term for bitrot of unused devices

@etchedpixels

I hadn't considered usb sticks; I've always thought of those as short term transport but I bet, post your post, lots of people don't think that.

@brouhaha @Lydie

@argv_minus_one @jernej__s there is an important table of defects stored on the platter in HDDs.

@Lydie

Can the drive not rebuild the defect table, if necessary, by scanning the surface?

@jernej__s