How long could an SSD for read only purposes last?

[Kippa]

I am aware that constantly writing to an SSD will reduce the lifespan of the drive. I was just wondering what if you purely used an MLC based SSD for purely archiving and specifically for read purposes, how long do you think a statistically average MLC drive could last? Are there any official numbers on how long the drive would last for purely read purposes?

Lets say the drive has a warranty to for 3 years. That has got to factor writing and reading. If it is just 100% reading could the drive realistically last 5+ years? Maybe longer?

 

[Cerb]

Not really.

With a good PSU, the capacitors would likely be the first to go. But, due to size, it's now common that SSDs not use electrolytics. Multilayer film, tantalum, etc. last a very long time (multilayer film types can pretty much last forever).

Consumer MLC is rated to minimum 1 year of offline retention at ~70F, after having been written to as many times as its p/e rating. Some makers spec for longer durations than JEDEC's minimum.

So, if you read everything off of the disk on a somewhat regular basis, assuring that it will get all your data and its metadata blocks checked out, it should last a very, very, very long time. The older such a pieve of electronics gets, the more that factors unrelated to how it was made matter, like power cleanliness, temperature and humidity changes, contact oxidation, etc..

 

[Phynaz]

Longer than you'll be able to find something to read it with.

 

[Kippa]

Some people have been saying that if you leave an SSD unpowered completely for say 3 months+ that you might get loss of data. Is there any truth to that belief? If this is true, if you powered the SSD every few weeks should it last a very long time (purely for archiving)?

 

[BrightCandle]

I have heard that its anywhere from about 3 months to a 1 year that the drive could loose some data if its unpowered. I have a Intel SSD that has been off 14 months which I tested and its all 100% fine so its probably a bit pessimistic and based on bad conditions but the charge can and does leak away over time. But once its powered the drive should rewrite itself and refresh the data if its been there a long time.

So if you wrote the entire contents of the disk just once and ensured it was powered sufficiently how long could it last? Well about 750 -3000 years. It needs to be rewritten (assume the disk does this automatically) every 3-12 months and you get about 3000 writes to an MLC cell (at least). Its most likely that something else in the drive will go before then, like not being able to find a motherboard with a SATA3 port in 2314! Other parts of the drive might fail but SSDs are solid state devices so they could go for 20+ years if they are made well.

 

[Cerb]

Some people have been saying that if you leave an SSD unpowered completely for say 3 months+ that you might get loss of data. Is there any truth to that belief?

Flash binned and sold to JEDEC standards should last a very long time--a year at its wear point at room temp. It can't be said for sure what it would be like, and would vary quite a bit between SSDs, for any longer duration at a much lower cycle count, but likely it would be in the decades range, if not written to all that much.

3 months is for enterprise flash at its p/e rating, 1 year is for consumer flash at its p/e rating, both are minimums at about room temp, both vary by temperature, and both will vary between individual flash chips, as far as how much better than the minimums they are.

Basically, if written to very little, they will last a very very long time, such that nobody has any useful specs for it.

 

[Elixer]

Consumer MLC is rated to minimum 1 year of offline retention at ~70F, after having been written to as many times as its p/e rating. Some makers spec for longer durations than JEDEC's minimum.

I have a 830 that was offline for about 4 1/2 months, and when I pulled it out to use on another system, it had corrupted files.
This makes me believe that you should apply power to the device monthly at least.

Can I prove it was because it was offline for that long period of time that it was the root cause of the corruption ? No, but I am strongly leaning that way, since it hasn't acted up again with constant use.

 

[rob426]

On the points raised in this thread, regarding SSD’s being unpowered for a long time may cause corruption

Does this apply to other types of memory, SD cards, Memory sticks, tablets etc. or only SSD’s ?

 

[Elixer]

It would apply to anything that uses NAND technology, so you should always make backups.

 

[jimhsu]

Something I've always wondered: is degradation a Poisson process, or does it result in catastrophic failure? I'm referring to a particular usage app that has recently gained popularity: storing private keys (cryptocurrencies, etc) in vaults for decades by writing the key hundreds of thousands of times over the drive.

 

[Cerb]

A single bit "lost" loses you at least a flash page (typically 8K to 16K, today). The flash has ECC, and errors during writing are expected, to a point.

The loss of raw bits should be gradual, and fit a Poisson distribution, but in any page/block losing strength of data, it will read fine until too many bits are lost, after which point the entire page, at least, will not be readable, which very well may ruin an entire file, all at once (IE, catastrophic loss). An active SSD should be able to read and move data in weakened areas.

 

[sheh]

I have a Intel SSD that has been off 14 months which I tested and its all 100% fine

What model is it? How many GBs were written to it?

In general I suspect newer models might have worse data retention, just like their raw write performance (or erase?) is worse than previous generations. But if there's correlation between the P/E cycles rating and data retention, perhaps recent SSDs are equal to the previous generation. But I don't know if there is correlation. It's also possible the P/E rating stayed static as of late not because of improved flash properties, but just because there's more error correction.