• Some users have recently had their accounts hijacked. It seems that the now defunct EVGA forums might have compromised your password there and seems many are using the same PW here. We would suggest you UPDATE YOUR PASSWORD and TURN ON 2FA for your account here to further secure it. None of the compromised accounts had 2FA turned on.
    Once you have enabled 2FA, your account will be updated soon to show a badge, letting other members know that you use 2FA to protect your account. This should be beneficial for everyone that uses FSFT.

SSD NAND decay?

Status
Not open for further replies.
V

Valnar

Supreme [H]ardness
Joined
Apr 3, 2001
Messages
4,554
Does anyone have any articles you can point me to about how long SSD Flash will last when the data is stagnant? I'm interested in the life expectancy difference between 50nm MLC vs 34nm vs the new 25nm coming out.

I know manufacturers are going to use tricks to constantly read/write and refresh the data so this doesn't happen, but what if the PC is not on and the SSD is not being used regularly?

I have a seldom used PC and I may need to jump on a 34nm MLC SSD now instead of waiting for the newer, and less stable 25nm gen3.
 
While the SSD is powered on it should have algorithms which keep the data moving around to refresh it. An unpowered SSD has a data retention limit of ~10 years for old-school SLC, 90/65 nm Flash. For MLC 34 nm it's around a year. 25 nm MLC will probably be around 8 months or so, though I don't have numbers on that one.
 
Elledan said:
While the SSD is powered on it should have algorithms which keep the data moving around to refresh it. An unpowered SSD has a data retention limit of ~10 years for old-school SLC, 90/65 nm Flash. For MLC 34 nm it's around a year. 25 nm MLC will probably be around 8 months or so, though I don't have numbers on that one.
Click to expand...

Damn, that's pathetic. I was afraid of that. Maybe I'll just buy a fast HDD instead.

Any linkage?
 
Just solidifies my choice on a HDD. Thanks.

I'm glad I bought the Intel X25-E for my main rig.
 
Robstar said:
Can anyone explain why this "decay" (when powered off) happens?
Click to expand...

Flash memory works by trapping electrons. Over time these electrons leak away, until the charge is too small for the data to be read any more. With smaller feature sizes (34 nm instead of 45 or 65 nm) this leakage is more significant and fewer electrons can be stored per bit, thus the time during which the stored value can be maintained is decreased.
 
Do all cells leak at exactly the same rate? If not, can't even not-powering the SSD take it's toll at partial rates of decay?
 
Any mass storage device that can't retain 100% of it's data after a year is bad in my book. I have crappy brand CDR's that do better than that.

Maybe I'll save up and buy another X25-E while I still can for my second PC.
 
You're probably gonna have a hard time finding anything concrete on 25nm since it's not in widespread use yet (Intel/Micron only announced they were ramping up production a couple months ago no?)... Why exactly do you need a SSD on a seldom-used system anyway? Just curious, legitimate question.

Valnar said:
Any mass storage device that can't retain 100% of it's data after a year is bad in my book. I have crappy brand CDR's that do better than that.

Maybe I'll save up and buy another X25-E while I still can for my second PC.
Click to expand...

SSD aren't exactly mass-storage devices... Most people use 'em as OS/boot drives, daily usage for performance, not for storage. AFAIK the day when HDDs will become obsolete for data storage are still nowhere to be seen (or predicted) since our data consumption is still growing (photos, music, non-lossy formats, HD video, etc) and the price per GB of HDDs is still falling. I still don't see why you need an X25-E (as opposed to a more mainstream drive, X25-M or w/e), 'least not from a data retention perspective. /shrug

I'm not sure what kinda system you leave turned off for an entire year at a time, and why said kind of system would ever need a SSD, but o-kay... I guess maybe if you're deployed in the military or working overseas or something, in which case, backups are your friend regardless.
 
I get that people like the speed of SSD's. I like that too. But don't rationalize away that its okay the data starts deteriorating the minute the drive has no power. It might be acceptable for 50nm or even 34nm. It won't be for 3-bit 25nm NAND. After all, what's to say the stagnant data (like Windows itself) won't get the refresh if you don't scrub through the SSD enough times during uptime?

I don't accept mediocrity as an excuse and certainly don't justify it even for a speed increase. The same is true for these ever expanding hard drive sizes. 1.5TB and 2TB drives are notorious for problems. Would you recommend I back up a 25nm SSD with a 2TB hard drive? I've been bit by bad 2TB drives. I fear that the manufacturers pushing the limits will compromise all our data.

/pissed at manufacturers for cheating us
 
Last edited:
Most of those questions will be answered by the engineers who're currently experimenting w/25nm flash and working on future controllers... Right now all you're doing is speculating tbh. /shrug Most people don't/won't accept a drive w/major debilitating flaws, hence the reason most people held off on SSDs before TRIM for instance... I'd recommend you backup your SSD to w/e you're comfortable with, heh... What 2TB drive have you had issues with? Saying that 1.5TB and 2TB drives are "notorious for problems" is a big generalization. Seagate had issues with their firmware for the 7200.11 line but other manufacturers haven't had any major issues AFAIK.

Those 2TB Hitachi drives seem to be pretty darn reliable, there's a big thread about 'em here on the boards (people seem to prefer 'em over WD drives even tho it's a 5-platter drive because of cost, RAID issues w/WD, etc.). Backups are all about redundancy tho, by nature... If you're super paranoid then backup to more than one drive, if you're super super paranoid then backup to two different brands and spend 2-3x on an SLC SSD drive, by all means.

P.S. You were proclaiming how durable your CD/DVDs are earlier, but they're also prone to decay over time (physical decay), just as anything else... If you put your baby's pictures on a DVD when he's born, I wouldn't expect to be able to retrieve them intact when he's 30 years old for instance... Or even 20. Nothing last forever, that's why we create backups, and these days it's far easier to do w/hard drives even if they're less reliable by nature (moving parts, etc.).
 
Oh and if you really think hard drives are, on balance, less reliable now than they were 5-10 years ago you've either been very lucky or you haven't been around long... There's always been issues w/new drive lines, 'specially when they push the limits. IBM's infamous 75GXP line (which included the first 75GB drive) was plagued with way more severe issues than anything you've seen of late, I'm pretty sure they lost the class action suit brought against 'em over it... And storage space cost a lot more back then, relatively speaking.
 
There seems like a lot of speculation on here by people not involved in the industry. So let's clarify a few points.

Every vendor of Nand flash must meet JEDEC specs in order to qualify it. The overall spec is JESD47F. Google it for more information. But the specifications do not change for smaller lithographies. But manufacturers may qualify up to a certain limit.

Within this specification, there are data retention requirements (JESD22-A117 and JESD22-A108) which must be met. To translate these: the data retention requirement for the component Nand to be at least one year at room temperature at the maximum cycle count. Moreover, the data retention must be roughly inversely proportional to the cycle. For example, at roughly 1/10th the cycle count, the data retention should be 10X.

The specs also specify the way this is to be measured and modeled at different temperatures.

For an 80GB drive, written at 20GB a day (an extreme client usage model), assuming a worse-case write amplication factor of 2:
20GB user writes per day * 2 = 40GB nand writes per day = 0.5 cycles per day = ~550 cyles in three years

Assuming this was an X-25M G2 and that the 34nm MLC flash is qualified out to 5K cycles, then you should be guaranteed 9 years of room temp data retention in this situation.

Know that the industry players are busy readying a JEDEC spec for SSDs that will cover some of these same questions.
 
NandFlashGuy said:
Every vendor of Nand flash must meet JEDEC specs in order to qualify it. The overall spec is JESD47F.
Click to expand...

Thanks for injecting some solid facts into this discussion, and for the reference. That is a long document, but for those who want the details, I excerpted a part about cycling and data retention testing:

Code: 
Non-Volatile Memory Cycling Endurance (NVCE)
JESD22-A117
3 Lots / 77 units
25 °C and Tj ≥ 55 °C
Up to Spec. Max / Cycles per note 

NVCE − See Figure A for linked flow for NVCE, HTDR, and LTDR. Half
of the devices are cycled at room temperature and half at elevated
temperature. Cycling should be performed to the max spec. cycle count,
but for large memories doing so on all cells may take an impractically
long time. In that case, total cycling time is to be 500 hours, with
separate blocks or sectors cycled to 1%, 10%, and 100% of the maximum
specification. One-third of the cycling time should be devoted to each
of these three cycle counts, so that in a many-block device the numbers
of blocks cycled to 1%/10%/100% of max. spec. are generally in the ratio
of 100:10:1. For multi-block memories, at least one block of each device
must be cycled to the max. spec. cycle count, regardless of the time
required. For products specified to be used with wear leveling, which
ensures that all cells receive comparable cycle counts in system use,
all cells should be uniformly cycled for the 500 hours. The supplier
may specify that cycling not exceed a certain rate per day, to avoid
overstress due to unrealistic conditions. For high-temperature cycling,
the supplier may specify that delays or bakes are to be inserted between
cycles, to emulate delays expected in intended application. These are
acceptable provided that the delays plus the cycling time itself do
not add up to more than 500 hours at 85 °C, 96 hours at 100 °C, or
10 hours at 125 °C, and that they are distributed per the guideline
in JESD22-A117. These delays do not necessarily demonstrate the effect
that would be seen with a particular use condition. For example, with
apparent activation energy of 1.1 to 1.2 eV for dielectric charge
detrapping, the delay durations are equivalent to about 1.5 years of
cycling at 55 °C. An application condition with less delay would be
more severe than is represented by the qualification delays specified
above. If application use conditions deviate considerably from the cycle
counts or equivalent times described above, then an application-specific
qualification methodology can be pursued per JESD94.



Data Retention for Non-Volatile Memory: High Temperature (HTDR)
JESD22-A117
3 Lots / 39 units
Option 1: Tj = 100 °C
  Cycles per NVCE (≥55°C) / 96 and 1000 hrs / 0 Fail / note
Option 2: Tj ≥ 125 °C
  Cycles per NVCE (≥55°C) / 10 and 100 hrs / 0 Fail / note 

HTDR − See Figure A for linked flow for NVCE, HTDR, and LTDR. The NVCE
devices cycled at elevated temperature are placed in high-temperature
retention bake. Two options are given, either of which is acceptable
for qualification, and for each option two bake durations. The longer
of the two durations is to be applied to the blocks cycled to ≤ 10%
of the max. spec. cycles. The shorter of the two is to be applied
to blocks cycled to 100% of max. spec. cycles. For example option 2
requires that blocks cycled to ≤ 10% of max. spec. cycles retain data
for 100 hours of 125 °C bake, and blocks cycled to 100% of max. spec.
cycles must retain data for 10 hours of 125 °C bake. The durations
listed are generally acceptable for qualification but do not necessarily
demonstrate the retention requirement for a particular use condition,
which depends on failure mechanisms and application environments. For
example, with activation energy of 1.1-1.2 eV for dielectric charge
detrapping, 125 °C stress temperature (option 2) and 55 °C use
temperature, the acceleration factor (Arrhenius equation) is 939 to
1750. Bake time is then equivalent to 11-20 years for ≤ 10% of max.
spec. cycles and 1-2 years for 100% of max. spec. Retention lifetime
necessary in use will be less than total product lifetime, because the
HTDR requirement is a sequential reliability stress that is preceded
by up to one lifetime’s worth of endurance cycling (NVCE). If the
application requirement does not match these retention values, or the
technology has different activation energy, then a knowledge- based
qualification should be followed (see JESD94).



Non-Volatile Memory Low-Temperature Retention and Read Disturb (LTDR)
JESD22-A108
3 Lots / 38 units
Ta = 25 °C
Cycles per NVCE (25°C) / 500 hrs / 0 Fail

LTDR − See Figure A for linked flow for NVCE, HTDR, and LTDR. The NVCE
Devices cycled at room temperature are placed into room-temperature
operating-life stress which sequentially performs dynamic read accesses
on all memory addresses. 25 °C stress temperature is used to determine
sensitivity to non-temperature-accelerated retention failure mechanisms,
or to mechanisms that can entirely recover at high temperatures, such
as the SILC mechanism. Biased life stress is performed to detect
voltage-induced disturbs due to random bit accesses, in addition to
unbiased data retention mechanisms which occur when a bit is not being
accessed. Inserted bakes as described for NVCE are not acceptable for
the 25 °C cycling condition used prior to LTDR. If the cycle counts
from note (e) or the retention lifetimes of 500 hours are insufficient
to meet a specific application requirement, or if bit accesses in
application are expected to be highly concentrated on specific bits,
then knowledge- based qualification methods using special techniques
should be used (see JESD94)
 
Does the JEDEC spec guarantee that in 1 - 10 years (depending on cycle usage) electron leakage won't happen to a non-powered SSD? Manufacturers do admit that they use tricks like shuffling to keep data intact.

Are you saying that JEDEC specifies that inert data on a non-powered SSD will last years, even with 34nm and 25nm NAND?
 
Looking at the JEDEC spec he referenced, it appears to apply to NAND flash memory chips. The chips themselves. So I do not see how it could rely on "shuffling" in an SSD.

If you look at the excerpt that I posted, the high temperature data retention (HTDR) test is meant to show that 55C data retention is at least 1 year. And that is at max cycles. At less than 10% max cycles, the HTDR test is meant to indicate 55C retention of at least 10 years. Since they obviously cannot measure a 10 year data retention directly, that lifetime is extrapolated from a high temperature bake. The example they give is that if the activation energy for dielectric charge detrapping is 1.1 to 1.2eV, then the acceleration factor going from 55C to 125C is 939-1750, call it 1000. Then a 100 hour bake at 125C is equivalent to a 100,000 hour lifetime at 55C, which is about 11 years.

I do hope that as time goes on, they re-run some tests with a longer duration. Hopefully IMFT has some max cycle test lots that are aging for 1 year at 55C, and some min cycle test lots that are baking for 1 year at whatever temperature they calculate will give an acceleration factor of 10.
 
Last edited:
I'll probably have a new model in a year anyway. At the rate they are progressing why keep one for any longer.
 
@Valnar & john4200:


The JEDEC JESD47F is applicable only to the semidonductor memory components, regardless of lithography. For qualication purposes, the overall cycling capability may be limited by the ability to retain data. So it's conceivable that a 25nm litho qualifies to a lower cycle count than 34nm, for example, because it can guarantee 1 year of data retention only up to some lower cycle count.

Currently, there is no SSD equilvalent JEDEC spec regarding cycling, data retention and the methods for proper sampling. But the spec is being actively developed by JEDEC members, which include semiconductor and hard-drive manufacturers as well as software companies.

There are system-level solutions that can be employed, but as you mention, this are only applicable when the drive is powered on. Moreover, solutions like this must be correctly accounted for in the modeling that is part of the JEDEC spec.

For example, a poor shuffling algorithm could cause a large number of cycles, resulting in the drive reaching the cycling limit faster than if no shuffling solution was in use.

As the theme of the tread implies, currently there is a lack of clear standards and experimental methods to measure the advertised capabilities. This is probably responsible for some of the wide gap in quality seen in SSDs to date.
 
One year later and I want to bring this topic back up again.

Does anyone have more information about NAND decay for stagnant data? I've dropped my desire to use an SSD to store data as a backup device (when not powered up), but I'm still on the fence about a powered-up-but-mostly-read device. Some of you have suggested that it will refresh and shuffle the data as needed, but I can't understand how it can do that when the SSD has no idea if I would be powering off at any moment.

I plan to use it in a MAME arcade machine where the arcade ROMs are pretty much stagnant. They will be read-only for years to come. Would this still be okay with an SSD?
 
I think the issue the OP has with SSDs stems from the perception consumers have about SSDs.

I personally have never viewed SSDs as a proper replacement for magnetic media or hard drives. While this may change in 10-20 years, right now, SSDs are pretty much a gap filler between main storage memory and hard drives. Historically speaking, there has always been a clear division between volatile and non-volatile storage. However, (current) SSDs are straddling that line. The current tech behind them really places them in a semi-volatile category (the same goes for all flash-memory based devices).

Even more recent efforts such as Intel's Smart Response have pushed the SSD abstract into a more memory-like cache role. Cache inherently is volatile. But it can also be shown how multi-roled and versatile SSDs can be, especially when speed matters.

Also, it's becoming more evident that SSD's selling point is speed above all else, with size becoming a closing (but still distant) second. As NAND and similar flash memory processes continue to get cheaper, there may eventually be a shuffling around of priorities. But data retention is largely dependent on the tech involved. You have to refresh memory cells with power. That power decays over time.

A somewhat temporary solution might be for mfgs to incorporate some kind of interchangable battery for "offline" cell refreshing. But this adds extra cost (and space) and shifts the SSD marketing towards are more specialized segment.
 
Absalom said:
I think the issue the OP has with SSDs stems from the perception consumers have about SSDs.

I personally have never viewed SSDs as a proper replacement for magnetic media or hard drives.
Click to expand...

While the issue of NAND discharge over time is real, and interesting discussion, it also seems to be purely academic when it comes to typical use in a "main rig". My main rig is on and in use every day unless we're on family vacation, and the performance boost I get from an SSD OS/boot drive far outweighs any "made up" concern I may have about possibly loosing some data if the drive is left powered off for as little as 8 months. Making sure that cells get exercised an refreshed may be a good reason to be sure to buy an SSD from an industry leader (like Intel) instead of whatever-is-cheapest.

I make semi-regular backup images of my SSD, just as I did when I used an HDD for boot/OS. Failures are possible for both technologies and a simple backup provides protection. Cost alone should (today at least) make a pretty clear conclusion that SSDs are not full HDD replacements as bulk data or archival devices.

I'm pretty sure that the OP (with his X25-E) and I (with my X25-M) are probably equally likely to suffer NAND discharge related data loss on our "main rigs" - namely zero. I'm glad that he feels "glad" he spent 10x what he needed to prevent a problem he's almost certain to never have. Maybe the -E drives keep tigers away too... I don't feel that the industry has bamboozled or misled the public in any way and I'm very happy to see this technology go almost-mainstream. I've said before the performance increase is fantastic even over a reasonably fast HDD. I maybe off by a model or two, but I used to run a 7200.9 for boot/OS and the upgrade to X25-M was fantastic. Just a couple of weeks ago I did a SB upgrade and repurposed the old parts as a second gaming rig. With exactly the same system core I freshly installed Win7 x64 on a 7200.11 and the slowness compared to the SSD is readily apparent even though all the other specs are the same. The gains are certainly not imagined, or limited in scope. Performance gains are even greater on laptops where 5400 or 4500 rpm drives are the norm.

I know that Intel was running ads about using SSDs for permanent archives. I'm not sure who the target market is with the $/GB being what it is for Intel SSDs but this degradation issue would certainly make me think twice about using SSD as an archival or near-line storage option that is usually powered off. As already stated, the cost would be far more of a factor than the NAND discharge.
 
In case it was missed, I brought this thread back to life only to ask about my current question posted today. Will static data, even if powered on, remain intact for a long period of time?

Forgive the resurrection of the old thread - my reason was to bring in the people who were discussing it a year ago.
 
That question was already answered long ago -- the data retention time starts at 10 years and drops to 1 year as all of the P/E erase cycles are used up.
 
That is the JEDEC JESD47F standard for data retention for non-volatile memory. I doubt a major flash vendor would sell any flash that did not meet the spec, but of course you would have to check with each manufacturer if you want to be certain.
 
john4200 said:
That is the JEDEC JESD47F standard for data retention for non-volatile memory. I doubt a major flash vendor would sell any flash that did not meet the spec, but of course you would have to check with each manufacturer if you want to be certain.
Click to expand...

I have seen data retention specified for NAND 90 nm chips as 10 years, and <1 year for current generation MLC NAND chips. This by manufacturers themselves. Write cycles themselves have dropped for MLC from 10,000 to 3,000 in that same time frame.
 
Elledan said:
I have seen data retention specified for NAND 90 nm chips as 10 years, and <1 year for current generation MLC NAND chips. This by manufacturers themselves. Write cycles themselves have dropped for MLC from 10,000 to 3,000 in that same time frame.
Click to expand...

Right....which is why I wanted to know if I setup a second partition on an SSD (call it D: drive) and put all my static zip files there, will they deteriorate?
 
Valnar said:
Right....which is why I wanted to know if I setup a second partition on an SSD (call it D: drive) and put all my static zip files there, will they deteriorate?
Click to expand...

Unless there's some kind of controller-based algorithm which periodically refreshes cells and there is power to the SSD regularly, the contents will deteriorate as the captured electrons will gradually leak away. I'm not aware of any SSD doing this.

It might be fun to use EEPROMs for this purpose, though, but that's just the geek in me talking :D
 
I think I can guarantee a well-treated hard drive will retain its static data for many years, so I think I will go that route for this project.
 
Elledan said:
I have seen data retention specified for NAND 90 nm chips as 10 years, and <1 year for current generation MLC NAND chips. This by manufacturers themselves. Write cycles themselves have dropped for MLC from 10,000 to 3,000 in that same time frame.
Click to expand...

You have "seen"? Where? Link?
 
i don't get what you're trying to do OP; the reason people pay nearly $2.00/GB for an ssd is mostly because of it's awesome performance not to archive data that's gonna be sitting around for a long time.
 
john4200 said:
You have "seen"? Where? Link?
Click to expand...

http://techon.nikkeibp.co.jp/article/HONSHI/20090528/170920/

NAND Flash memory quality is also beginning to drop. Chips manufactured using 90nm-generation technology in 2004-05, for example, were assured for about 100,000 rewrites and data retention of about a decade. As multi-level architecture and smaller geometry are introduced, quality is showing a sharp decline. The 30nm 2-bit/cell chips expected to enter volume production in 2009-10 may well end up with a rewrite assurance of no more than 3,000 cycles, and a data retention time of about a year. The first 3-bit/cell chips are hitting the market now, with only a few hundred rewrites.
Click to expand...

'About a year' turns out to be closer to about 8 months, and this number will keep dropping. I'd worry about keeping static data on an SSD for longer than half a year at this point. They're little more than semi-permanent storage at this point, basically as the linked article already says. A buffer, maybe, but not storage like a HDD.

According to Akihito Nishikawa, senior manager, Memory Division, SSD Application Engineering Dept of Toshiba, "The memory cell is assured for about 3,000 rewrites. If the target SSD capacity is 128GB, then the total bit capacity is 128GB times 3,000 cycles divided by 1.5 (our figure for rewrite efficiency), or 256TB. Toshiba research indicates that actual PC users generally don't rewrite more than 20GB, max, per day. If these numbers are used to estimate the SSD service life, it works out to 256TB divided by 20GB divided by 365 days, or about 35 years. Considering average user access conditions, quality is more than sufficient to assure that."
Click to expand...

Better hope you're staying close to that 20 GB/day limit. Upcoming NAND Flash chips with even smaller feature sizes will have even fewer rewrite cycles. I know that for a video editing station writing single videos of multiple GBs is common, so they'd blow through a few hundred GB a day with ease, up to a few TB in the case of final renders.
 
Elledan, that is not a statement from the major flash manufacturers like you claimed, that is just an old trade journal article. You made it sound like Intel, Micron, Toshiba or Samsung were saying that their flash chips do not meet the JEDEC JESD47F spec.

If you have a credible source that says that current SSD-grade flash memory from one of the major manufacturers does not meet JESD47F, then I would certainly like to know about it. If you do not, then it is irresponsible to continue posting FUD.
 
Last edited:
Valnar said:
I plan to use it in a MAME arcade machine where the arcade ROMs are pretty much stagnant. They will be read-only for years to come. Would this still be okay with an SSD?
Click to expand...

Why not just get a cheap HDD? Why waste your money on a SSD when you obviously don't need the speed and are basically going to be archiving the data. That makes no sense.

SSDs are meant for speed and maximum performance, not maximum stability and storage quality.

For long term, I would always use HDDs, tapes, or optical media. Never would I store long term data on any flash/memory chip device.
 
Red Falcon said:
Why not just get a cheap HDD? Why waste your money on a SSD when you obviously don't need the speed and are basically going to be archiving the data. That makes no sense.

SSDs are meant for speed and maximum performance, not maximum stability and storage quality.

For long term, I would always use HDDs, tapes, or optical media. Never would I store long term data on any flash/memory chip device.
Click to expand...

Well, it is a theoretical question as much as practical. Besides, hard drives are notoriously unreliable these days compared to a few years ago. I've had two 2TB drives die in less than a month.
 
Status
Not open for further replies.
Back
Top