SSD Unpowered Data Retention

[BFG10K]

I’ve had a Samsung 830 Pro (MLC) unpowered in my drawer for around 9 months which contained a backup of my Windows partition. I booted it up now to test it and it worked fine. I’ve now updated the image and powered it down again.

SSD unpowered data retention information is really hard to come by so I thought I’d share my anecdotal evidence. Feel free to share yours.

 

[Elixer]

For what it is worth, when my SSD (MX100) was unpowered for ~8 months, it decided to enter into some kind of a special state which required PSID reset before I could get it working again.

http://www.portvapes.co.uk/?id=Latest-exam-1Z0-876-Dumps&exid=threads/crucial-mx100-lost-the-security-features.2482065/

 

[Mr Evil]

There was an article on AT a couple of years ago discussing some interesting numbers that appeared in a JDEC presentation by Alvin Cox. They suggested that you should expect >1 year under typical conditions.

 

[BSim500]

I’ve had a Samsung 830 Pro (MLC) unpowered in my drawer for around 9 months which contained a backup of my Windows partition. I booted it up now to test it and it worked fine. I’ve now updated the image and powered it down again.

SSD unpowered data retention information is really hard to come by so I thought I’d share my anecdotal evidence. Feel free to share yours.

Thanks for posting this. I had a good experience with an 830 too as well as Crucial's MLC drives. Sadly unpowered data retention is one of the least tested "features" of modern flash-based storage (both SSD & USB sticks). In theory the JEDEC numbers that regularly do the rounds "guarantees" it for a year. In practise, those figures came out years ago based on Intel +20nm MLC tech. Since 16nm TLC, we've already seen slowdowns on both Samsung 840's and BX200's even in daily powered drives. A lot of people using them as system drives will not see data loss problems if the firmware constantly rewrites the data to hide voltage drift issues.

But for unpowered drives, that's obviously not possible and there's a serious problem when people start recommending them as HDD spinner replacements for stuff written once and then stored for +2 years, eg, long term archives of wedding photo's, etc, typically viewed only years after being written. It's actually quite sad just how poor many modern SSD reviews infomercials are compared to some of the in-depth durability testing done a few years back.

 

[corkyg]

I have 5 SSDs I use in my laptop. I only use it for travel, so a once a month power up and rotation through all (2 830s, 2 850 EVOs, 1 MX500) has never indicated a trace of a problem. The laptop has the battery removed and I only run it on A/C.

 

[Billy Tallis]

For what it is worth, when my SSD (MX100) was unpowered for ~8 months, it decided to enter into some kind of a special state which required PSID reset before I could get it working again.

Unexpected power loss and hot swapping will also frequently trigger this behavior on Crucial MX series drives, and occasionally on Samsung SATA SSDs as well.

 

[corkyg]

That may have something to do with it - my power losses are never unexpected, and I never hot swap in a laptop - it is not really practical to do that.

 

[Elixer]

Unexpected power loss and hot swapping will also frequently trigger this behavior on Crucial MX series drives, and occasionally on Samsung SATA SSDs as well.

@Billy Tallis, while I never hot swapped a SSD or had any power loss issues (use a UPS), since hot swap is part of the SATA spec, does the lack of having bigger CAPs on SSDs part of the problem, or are these issues just firmware bugs?

I noticed "dumb" drives, like USB pen drives or SD/microSD cards don't seem to care about hot swapping or power loss. Obviously, I don't mean writing to it when that happens, but, I mean stick it in, pull it out, and so on. If I did that with a SSD, the SMART parameters for unexpected power loss would skyrocket...

 

[JimmiG]

Thanks for posting this. I had a good experience with an 830 too as well as Crucial's MLC drives. Sadly unpowered data retention is one of the least tested "features" of modern flash-based storage (both SSD & USB sticks). In theory the JEDEC numbers that regularly do the rounds "guarantees" it for a year. In practise, those figures came out years ago based on Intel +20nm MLC tech. Since 16nm TLC, we've already seen slowdowns on both Samsung 840's and BX200's even in daily powered drives. A lot of people using them as system drives will not see data loss problems if the firmware constantly rewrites the data to hide voltage drift issues.

But for unpowered drives, that's obviously not possible and there's a serious problem when people start recommending them as HDD spinner replacements for stuff written once and then stored for +2 years, eg, long term archives of wedding photo's, etc, typically viewed only years after being written. It's actually quite sad just how poor many modern SSD reviews infomercials are compared to some of the in-depth durability testing done a few years back.

It seems logical that data retention would get worse with the finer manufacturing processes. There are simply fewer electrons, thinner barriers etc.
My newest SSD uses 15nm TLC, and I wouldn't trust it with any important data whatsoever (it's just a Steam library drive). I probably trust my mechanical HDD's more. The 850 Evo on the other hand was made with the 40nm process, so it should be able to keep those electrons trapped for longer.

 

[Glaring_Mistake]

I’ve had a Samsung 830 Pro (MLC) unpowered in my drawer for around 9 months which contained a backup of my Windows partition. I booted it up now to test it and it worked fine. I’ve now updated the image and powered it down again.

SSD unpowered data retention information is really hard to come by so I thought I’d share my anecdotal evidence. Feel free to share yours.

I'm running tests on a number of drives regarding voltage drift and data retention so I've some experience with that.
No drive has failed data retention tests so far, even those with 2D TLC NAND have passed despite being tested under some challenging conditions so I'm not surprised that your 830 worked even after being unpowered that long.
Have seen read speeds drop however, including for some drives with larger margins (like larger lithography and/or MLC NAND).

Have tested the PM830 for example (mSATA-version of the 830) and it held up better under challenging conditions than I expected (though since the top read speed is limited to around 400MB/s it is possible that drops in read speed may not be noticed as fast as for some other drives).

Since 16nm TLC, we've already seen slowdowns on both Samsung 840's and BX200's even in daily powered drives.

Funny thing is that for the BX200 read speeds can actually drop faster if it is powered on than if left unpowered.
But that is under specific conditions though so most of the time it is probably better to have it powered on rather than unpowered.

But for unpowered drives, that's obviously not possible and there's a serious problem when people start recommending them as HDD spinner replacements for stuff written once and then stored for +2 years, eg, long term archives of wedding photo's, etc, typically viewed only years after being written.

Not sure I would recommend like for example the BX200 for that but I think it would be possible to do that with some other drives, even some using 2D TLC NAND actually.

It's actually quite sad just how poor many modern SSD reviews infomercials are compared to some of the in-depth durability testing done a few years back.

Those tests concentrated more on endurance (or how many writes the drive could withstand before giving up) than retention however.

It seems logical that data retention would get worse with the finer manufacturing processes. There are simply fewer electrons, thinner barriers etc.
My newest SSD uses 15nm TLC, and I wouldn't trust it with any important data whatsoever (it's just a Steam library drive). I probably trust my mechanical HDD's more. The 850 Evo on the other hand was made with the 40nm process, so it should be able to keep those electrons trapped for longer.

Well, I think lithography is a part of it but not all of it.
For example I think that when Toshiba/SanDisk went from 19nm to 15nm TLC retention improved (going by the results of my tests at least).
In fact I think that their 15nm TLC is probably the best 2D TLC NAND in terms of retention.

It's another story with the SM2256(s) that they use in the WD Green since it is not really the best controller when it comes to managing voltage drift, though with well-written firmware and decent NAND it is not that bad.

 

[bononos]

.......
Not sure I would recommend like for example the BX200 for that but I think it would be possible to do that with some other drives, even some using 2D TLC NAND actually.
.....

2yrs sounds like along time to leave a TLC ssd unpowered. How long do you have to power on such an ssd after that amount of time. I imagine the drive needs to alot of rewriting.

 

[Glaring_Mistake]

2yrs sounds like along time to leave a TLC ssd unpowered. How long do you have to power on such an ssd after that amount of time. I imagine the drive needs to alot of rewriting.

Agree that it is a long time but I have seen one drive using 2D TLC NAND able to keep its read speeds up even under challenging conditions, better in fact than some drives with larger litography and/or using MLC NAND.
So I don't find it that farfetched that it would be able to retain data after being unpowered for that long if under more favourable conditions.

Since it is dependent on several factors it is difficult to say how much it would need to rewrite after two years.
If it is necessary to perform rewrites however then the drives I'm thinking of would probably make it a priority (even if they may not rewrite all files in need of it).
Also you may reboot the computer a few times since that seems to get several drives to realize that they should rewrite some files.

However I think that if you do leave it unpowered that long it may be a good idea to run DiskFresh on it rather than to rely on the drive's own rewrite function.

 

[bononos]

Agree that it is a long time but I have seen one drive using 2D TLC NAND able to keep its read speeds up even under challenging conditions, better in fact than some drives with larger litography and/or using MLC NAND.
So I don't find it that farfetched that it would be able to retain data after being unpowered for that long if under more favourable conditions.
.......

It would be strange for a 2d tlc drive to be faster unless its rewriting data more often since tlc drive would rely more on complex error correction methods to read more marginal data.

 

[Glaring_Mistake]

It would be strange for a 2d tlc drive to be faster unless its rewriting data more often since tlc drive would rely more on complex error correction methods to read more marginal data.

Not the first thing you'd expect but even a drive using 2D TLC NAND can have read speeds that are pretty stable.
And in fact it is possible for it to maintain high read speeds better than some drives with larger litography and/or using MLC NAND depending on the conditions.

 

[bononos]

Not the first thing you'd expect but even a drive using 2D TLC NAND can have read speeds that are pretty stable.
And in fact it is possible for it to maintain high read speeds better than some drives with larger litography and/or using MLC NAND depending on the conditions.

When you say "maintain high read speeds", are you talking about measures taken by ssd's to long term voltage drift? If so what is going on since afaik, there isn't much manufacturers can do but to rely more on more complex error correction.

Long term unpowered performance degradation aside, I find 2d tlc drives pretty bad in performance. There is a quick drop off in performance once their small slc/dram caches fill up and then they perform alot closer to spinning drives.

 

[Glaring_Mistake]

When you say "maintain high read speeds", are you talking about measures taken by ssd's to long term voltage drift? If so what is going on since afaik, there isn't much manufacturers can do but to rely more on more complex error correction.

Well, I don't know quite how it manages to correct for voltage drift and still not suffer from clear slowdowns.
But I think that the reason why it managed to outperform some drives (and not just ones also using 2D TLC NAND) is due to its controller which seems to be pretty good at adjusting for a number of factors.

Long term unpowered performance degradation aside, I find 2d tlc drives pretty bad in performance. There is a quick drop off in performance once their small slc/dram caches fill up and then they perform alot closer to spinning drives.

Agreed, they tend to have that problem though Samsung is pretty good at getting drives with 2D TLC NAND to perform well (their issues with read speeds dropping due to voltage drift aside).
And considering that it is getting more and more common with DRAMless drives performance may drop even further.

 

[BFG10K]

I decided to plug it in again to make a new OS image, and just by sheer coincidence it's been exactly a year unpowered. The old data seemed to be fine.