Next generation of SSDs to have a shorter lifespan?

Elledan said:
Did you ignore my post with links to interviews with people who work in the actual industry?
Click to expand...
I would bet you he/she did not read any of it (or at least comprehend it).

It is a scary thing to think that logical evaluations aren't worth much these days.
Some people are waiting for us to show them tests with numbers like in benchmarks.
If principles based on atomic physics/chemistry are speculations, then there is very little left to debate. :rolleyes:
 
Lost Circuits has an excellent series of objective articles on the positives and negatives of NAND technology for the more intellectually curious. And Lost Circuits itself has intimate knowledge of SSD, being as the owner of the site works for OCZ.

NAND is very much still an immature technology. The silicon was actually developed with far different, less critical applications in mind than storing your data indefinitely. The difficulty remains in getting NAND to adapt past major limitations.
 
No, I read it. I did not see any facts about data retention time that were directly relevant to the current generation or next generation of SSDs, which is what I am interested in.

I have no doubt that the device physics get more difficult as the size is reduced. But I am more interested in the performance of actual SSDs than I am in the difficulty of making them.

If it is indeed the case that the 34nm SSDs have a data retention time of 6 months to a year, as someone in your referenced article speculates, then where are the articles showing problems with the Intel X25-M G2 drives? Or other SSDs using 3x nm NAND flash? Since I have not seen any articles about data retention problems with these SSDs, it seems to me that the people who engineered these devices found a way to make the data retention time much longer than the naysayers fear.

As long as I have been following the electronics industry (more than 30 years), there are always naysayers -- "this cannot be done" or "no one will be able to do better" or "we have reached the limit". But it has always turned out that some clever people found a way to do it. So I tend to ignore the naysayers unless they have actual evidence, which in this case would be SSDs that have been tested and found to have data retention time of less than a year.
 
john4200 said:
If it is indeed the case that the 34nm SSDs have a data retention time of 6 months to a year, as someone in your referenced article speculates, then where are the articles showing problems with the Intel X25-M G2 drives? Or other SSDs using 3x nm NAND flash? Since I have not seen any articles about data retention problems with these SSDs, it seems to me that the people who engineered these devices found a way to make the data retention time much longer than the naysayers fear.
Click to expand...

Then you didn't get what's being said. The data retention issue occurs when the drive is inactive (not powered) for the data retention limit. This means that the SSD would have to sit there unused and unpowered for a year or longer before the data on it expires. When the drive is powered, the static data, the very data which would otherwise expire is constantly moved and rewritten to keep it 'fresh'. SSDs just lying there is not a common scenario at this point (who pays hundreds of dollars for an SSD, puts data on it and then doesn't use it?), but could be an issue with USB Flash sticks, SD cards and such which are generally used in an off-line fashion and don't use the static data reshuffling algorithms (or rarely at least). How many of your SD cards do you leave lying around unused for a year? What if half a year becomes enough to make bits go poof?
 
No, that is not the way failures work. There is some probability distribution to failures. If the mean data retention time is 1 year, there will be some proportion of failures at less than 1 year. Also, there is accelerated stress testing (often done with higher temperatures).

Where are the articles about SSDs or flash drives losing data after 6 months or a year? Even if the most common usage scenario is not to leave the flash unpowered for long periods of time, there will be people who have. And there will also be people who specifically test that sort of thing for whatever unusual application they are looking at. Where are the articles?
 
From the article you claim to have read, "The role of wear leveling is to ensure that all blocks are subjected roughly to the same number of write / erase cycles. This is important since the write endurance of a typical MLC block is around 10k program/erase cycles (50 nm process) and every shrink in process geometry will shorten the average life span of a block. That is, at a 3x (30-39nm) process geometry the write endurance is declining towards 3-4000 P/E cycles."
 
but could be an issue with USB Flash sticks, SD cards and such which are generally used in an off-line fashion and don't use the static data reshuffling algorithms (or rarely at least)
Click to expand...

I just got a 16MB flash card out the loft that i havent used in years, the last datestamp on a file is august 2005, but everything still works on that.
 
Toytown said:
I just got a 16MB flash card out the loft that i havent used in years, the last datestamp on a file is august 2005, but everything still works on that.
Click to expand...

Your 16MB doesnt have the newer 25nm process which is what is being discussed here. I would estimate that traditional, older flash memory chips have data retention in the order of 10-20 years.
 
Wow really interesting stuff.
I had not hear about the data retention issues with different manufacturing processes.

Maybe i should swap my G2s for some G1's
 
john4200 said:
No, that is not the way failures work. There is some probability distribution to failures. If the mean data retention time is 1 year, there will be some proportion of failures at less than 1 year. Also, there is accelerated stress testing (often done with higher temperatures).
Click to expand...
I didn't say the data will vanish instantly, but that it's the point where data retention can no longer be guaranteed. Slowly bits will start to go poof after this point.

Where are the articles about SSDs or flash drives losing data after 6 months or a year? Even if the most common usage scenario is not to leave the flash unpowered for long periods of time, there will be people who have. And there will also be people who specifically test that sort of thing for whatever unusual application they are looking at. Where are the articles?
Click to expand...

I don't know why there aren't any articles on this (as far as I know, but I haven't searched for them). The loss of data with Flash memory is a fact, however, as this type of memory is based on the retention of electrons (electron trap). Since no kind of retention is perfect the electrons will slowly work their way out of the trap. Once the number of electrons drop under the read limit, the data will be gone. The speed with which this limit will be reached depends on the number of electrons stored initially and the number of electrons needed for a successful read. With newer, smaller process sizes both numbers become smaller, this is why data retention with old SLC (lots of electrons in a big, relatively electron-tight trap) is about 10 years, whereas 32 nm MLC Flash, where far fewer electrons are used in a far more leaky trap since the structures are smaller and thus leakage is higher, data retention is much shorter, much like current leakage is a problem with CPUs and similar as they're made smaller.

In short, the fact that Flash memory has a limited data retention is very well established at this point. What the exact limit is per Flash chip differs, of course, but the fact remains that it isn't going up while process sizes are going down.
 
Elledan said:
I didn't say the data will vanish instantly, but that it's the point where data retention can no longer be guaranteed. Slowly bits will start to go poof after this point.
Click to expand...

Sure, but that is not the information that we need. All kinds of data storage have BERs that increases with time. The practical questions are how much, and how much can be done to mitigate it?

The claims being thrown around here are that data retention time for the current (34nm) or the next gen (25nm) NAND flash devices could be 1 year, 6 months, or less. The problem is that these claims are ambiguous. As I've mentioned before, failures are probabilistic, so it is not enough to say that the data retention is 1 year -- there needs to be some sort of range or distribution specified.

Another problem is that several people seem to be making claims about the individual NAND gates, whereas what is important to computer users are the retention times for the consumer products that use NAND flash. While it is certainly true that the physics of the smaller devices lead to difficulty with long-term data retention, the practical question is whether the consumer products can do enough to mitigate these difficulties.

But answering that question requires a lot of information. What are the BERs vs. time for the NAND flash at various temperatures? How broad are the distributions? How much ECC is required to mitigate those BERs? What other innovative techniques can be used to correct errors (eg. RAID/RAISE)?

If no one has reported seeing a lot of SSD or flash drive data retention failures after 6 months or a year, the question is, why? I'll certainly grant you that it is possible that no one has yet put in the time to test the latest generation of products carefully and write up their results. But I think the more likely answer is that the producers have used innovative techniques to mitigate the problem, and that the practical unpowered data retention times are longer than the naysayers are claiming. If that is not the case, then we should be seeing a lot of complaints over the next year or two, particularly from people with 34nm NAND flash drives who find they have lost data.
 
I think that to some extent the hype surrounding Flash as the cool new storage method (SSDs especially) is making it hard for criticisms to be heard and be taken seriously. It also negates the need for manufacturers to publish numbers on data retention and such things. Only mission-critical SLC SSD drives are guaranteed for a certain lifespan, beyond that you'd have to find datasheets for the controller and Flash chips used in a particular SSD, SD card and such to get an idea of what the expected lifespan is.

Considering that Flash is still a relatively immature technology there are still a lot of questions to be answered, that's the one thing I hope we can all agree on. I also agree that data retention and write endurance issues could become a big issue during the coming years. We should also begin to see more and more Flash-based devices produced years ago to begin to die.
 
Right now Samsung is having major difficulties with their transition to a mere 32nm. They are way behind their own production timelines. We're talking about the number two player in the industry. The problems? Samsung's 32nm NAND is significantly slower and less reliable than what's currently available.

Samsung wants us to think the problem with their 32nm NAND lies with the controller. They don't want investors to be aware of NAND limitations. And guess what? They will still be selling these chips for less demanding embedded devices. We obviously shouldn't expect manufacturers to be honest about or within their own industry. I laugh whenever Intel quote IOPS specs for desktop computers.

Enthusiasts need to read between the lines and connect the dots. Read more trade papers and technical news than marketing pamphlets. Otherwise they act as extensions or arms of marketing departments. All I see lately are zealous enthusiasts buying into hype (and underdeveloped products) and giving manufacturers free passes.

NAND process shrinkage initially means a shortening of write endurance of a block. Eventually as the process matures so does the lifespan. NAND shrinkages benefits the manufacturer more, NAND maturity benefits the consumer.
 
john4200 said:
...
If no one has reported seeing a lot of SSD or flash drive data retention failures after 6 months or a year, the question is, why? I'll certainly grant you that it is possible that no one has yet put in the time to test the latest generation of products carefully and write up their results. But I think the more likely answer is that the producers have used innovative techniques to mitigate the problem, and that the practical unpowered data retention times are longer than the naysayers are claiming. If that is not the case, then we should be seeing a lot of complaints over the next year or two, particularly from people with 34nm NAND flash drives who find they have lost data.
Click to expand...

Why? Because most everyone is using devices where the issue is less problematic.
In fact, the issue is so less pronounced with those devices that they will likely fail from other causes before they get to the point where what's being debated is an issue.

The point of this discussion: whereas data retention and write-erase cycle endurance were lesser issues with higher manufacturing processes, they are bound to be major issues as the processes shrink.
Hence, whereas most old USB thumb drives are good for 10-20 years, new drives based on a 25nm process might get as little as a year.

Frankly, the data retention issue is the lesser of the issues I am trying to raise here.
If it gets too short, then SSDs will be limited to online mode only and might sport a battery pack (to keep static data shuffling running) if data retention goes as low as a few days.
The decreased write-erase cycle endurance is the bigger issue along with the increasing bit errors that have to be handled.

Pair 1000 write-erase cycles with a 2MB block (block sizes are getting larger too), and you can wear out a 2TB SSD drive with 1GB worth of bytes written individually.

Want proof? Buy me an SSD and I will prove it. ;)
 
This thread is sounding off red alerts with no fire.

when this thread is dead I will comeback and post how my 160 MLC drive is doing. I bought it Dec 24th, 2009.

Note to myself make backups of my backups cause you never know when the sky is gonna fall :)
 
Elledan said:
I think that to some extent the hype surrounding Flash as the cool new storage method (SSDs especially) is making it hard for criticisms to be heard and be taken seriously.
Click to expand...

I keep asking for serious criticisms with evidence, but you have not supplied any. Where are the people with 34nm flash drives reporting data loss due to gate leakage or whatever?

It is one thing to say that there might be a problem with data retention in current or future flash drives, without presenting any evidence. It is entirely another thing to claim that it is a fact that there is a problem with no evidence to back up the claim. That's just FUD.
 
vista_blista said:
All I see lately are zealous enthusiasts buying into hype (and underdeveloped products) and giving manufacturers free passes.
Click to expand...

All I see are fear-mongers making unsubstantiated claims. How hard is it to present some actual evidence if the problem is anywhere near as severe as you claim?
 
spectrumbx said:
Hence, whereas most old USB thumb drives are good for 10-20 years, new drives based on a 25nm process might get as little as a year.

Pair 1000 write-erase cycles with a 2MB block (block sizes are getting larger too), and you can wear out a 2TB SSD drive with 1GB worth of bytes written individually.
Click to expand...

That is speculation. When someone has convincing evidence that their flash drives are losing data after 6 months or a year, then I will be interested.

As for the write endurance issue, I am not as concerned as you seem to be. If I have an application where I need extremely high write endurance, I will choose the best type of storage for the application. If the manufacturers have not made 25nm MLC SSDs with sufficient write endurance, then I will choose something else. I think this will not be an issue for the vast majority of consumers.
 
nitrobass24 said:
Wow really interesting stuff.
I had not hear about the data retention issues with different manufacturing processes.
Click to expand...

I agree. I hadn't even considered this. While I think there's no need to be too alarmed, it's obvious that this is a real issue that SSD engineers are facing (or will have to face relatively soon).

Is it enough to simply have the SSD powered on, or do the cells actually need to be refreshed to ensure data retention? It wasn't really clear to me whether supplying the cells with power would prevent against data loss, or if the data needed to be "reshuffled" as Elledan put it. Since the problem seems to stem from the loss of electrons, it would seem that simply supplying the device with power every so often would be enough to re-energize the cells.

nitrobass24 said:
Maybe i should swap my G2s for some G1's
Click to expand...

I was going to say the same thing, lol. I guess that's the one advantage G1 owners can claim over us! ;)
 
imyourzero said:
...
Is it enough to simply have the SSD powered on, or do the cells actually need to be refreshed to ensure data retention? It wasn't really clear to me whether supplying the cells with power would prevent against data loss, or if the data needed to be "reshuffled" as Elledan put it. Since the problem seems to stem from the loss of electrons, it would seem that simply supplying the device with power every so often would be enough to re-energize the cells.
...
Click to expand...
No, just having the drive under power is not sufficient. Else, it would really be a non-issue.
The block has to be "reset" with an erase and re-written.
Luckily, drives like those made by Intel have a background process that re-shuffles the static data.
Again, data retention issues are not that big with current drives.
The concern is with future drives and the ever shorting erase-write cycles.

Makaveli@BETA said:
This thread is sounding off red alerts with no fire.

when this thread is dead I will comeback and post how my 160 MLC drive is doing. I bought it Dec 24th, 2009.

Note to myself make backups of my backups cause you never know when the sky is gonna fall :)
Click to expand...

Feel free to wait until the flames engulf you to paid attention to the warnings.
For me, early smoke detection will do.

Ps. I too own the Intel 160GB G2.
For my usage, I think it will do fine for up to 3 years.

Backup is not going to save you when you start getting massive bit errors beyond the ECC fault tolerance implemented.
Go ask those that used the iRAM (data corruption issues killed it).

If you really want to dare me, send me your drive and I'll prove it. :)

john4200 said:
That is speculation. When someone has convincing evidence that their flash drives are losing data after 6 months or a year, then I will be interested.

As for the write endurance issue, I am not as concerned as you seem to be. If I have an application where I need extremely high write endurance, I will choose the best type of storage for the application. If the manufacturers have not made 25nm MLC SSDs with sufficient write endurance, then I will choose something else. I think this will not be an issue for the vast majority of consumers.
Click to expand...
Yeah, I can already see people tripping over themselves to go destroy their drives just to convince you. :rolleyes:

Again, the current drives are going to be just fine for the average user looking for a 3-5 years usage.
Future drives will wear out much faster.
 
Last edited:
vista_blista said:
Enthusiasts need to read between the lines and connect the dots. Read more trade papers and technical news than marketing pamphlets. Otherwise they act as extensions or arms of marketing departments. All I see lately are zealous enthusiasts buying into hype (and underdeveloped products) and giving manufacturers free passes.

NAND process shrinkage initially means a shortening of write endurance of a block. Eventually as the process matures so does the lifespan. NAND shrinkages benefits the manufacturer more, NAND maturity benefits the consumer.
Click to expand...

My SSD(s) work great, one of the best upgrades I've ever made, I could care less what happens to them in 5 years... I'm pretty sure most enthusiasts or early adopters would have the same frame of mind, and let's face it, they're the main market for SSD right now. I don't see what the big deal is. Are there hurdles to overcome (or sidestep) with future products, smaller processes, or as current products age? Sure, but for the products that are out right now it's not a big deal, so I don't see why you're trying to paint this picture of hype and misleading marketing being used to deceive people. /shrug

Would we benefit from more information and testing on the subject? Absolutely, but as you said so yourself... If what you're doing is really important, you should be using SLC drives anyway. I mean hell, even if enthusiasts ignore these topics entirely, most of them are still not gonna be using flash for any sort of long-term storage anyway. It's just not practical with the current low cost of HDDs (and that's likely to stay true for a few years to come).
 
More blocks allow for more wear leveling. So if each blocks have 100 thousand erase cycles and there are twice as many of them it its a wash if they are twice as large.
 
drescherjm said:
More blocks allow for more wear leveling. So if each blocks have 100 thousand erase cycles and there are twice as many of them it its a wash if they are twice as large.
Click to expand...

Only if you get twice as much space for the same money and if the decrease in the write-erase cycle is linear.
Have you ever wondered why SLC gets 100K while MLC gets 10K (at best)?

Further, we need to stop quoting 10K for MLC since that's only true for 90nm MLC cells.
A 25nm MLC cell is likely going to be as low as 1k.

The only good news here is that we might start seeing some 160GB 25nm SLC drives that should last longer than current generation MLC drives.
 
I just read this whole thread and I had no idea that the smaller die size was tied into worse longevity. Wow.....eye opener.

So the idea of storing data indefinitely on a USB thumb drive is a bad idea? Does anyone know what nanometer process size is used in the various USB Flash drives from years past? 1GB, 2GB, 4GB, modern 8GB - 32GB, etc?

Also, based on the fear being produced in this thread, would it not be better to splurge on an SLC based Intel X25-E which will last for many years vs buying *several* MLC based drives over the course of the same timeframe?

I was thinking of buying an SSD for an OS drive, and the last thing I want to worry about is having to replace it every 1-2 years simply because manufacturer XYZ decided to use 25nm MLC NAND. As was pointed out, the core Windows files never change.
 
You could always image your drive, wipe it, and restore the image to refresh/recharge the cells (sector-based copy, not file-based). That may seem like a hassle, but you wouldn't have to do it often. I can't say that I use the same installation of Windows for years at a time anyway.
 
imyourzero said:
You could always image your drive, wipe it, and restore the image to refresh/recharge the cells (sector-based copy, not file-based). That may seem like a hassle, but you wouldn't have to do it often. I can't say that I use the same installation of Windows for years at a time anyway.
Click to expand...

Yes, everything you said is correct, just like everything I said is correct. But the difference is putting a price on peace of mind. The infamous car analogy comes into play. Buy a new car for more money but have less problems? Or buy a cheaper used car, but it'll have higher maintenance.

If 50nm SLC is the reliability sweet spot - and we'll never see it again, I think I may end up with the X25-E unless I'm missing something. :confused:

spectrumbx said:
Further, we need to stop quoting 10K for MLC since that's only true for 90nm MLC cells.
A 25nm MLC cell is likely going to be as low as 1k.
Click to expand...
Yep... I believe that illustrates my point even more.
 
USB Flash sticks generally use older Flash chips, so they could retain data for longer than new MLC Flash SSDs. I wouldn't count on it, though. Buying those bargain bin USB sticks may be the safe option, however ;)

Data-retention isn't an issue as long as you keep the drive powered and it has the reshuffling algorithm (check the manual). With regards to durability, your needs depend on how much you write to disk on a daily basis. It's around the 20 GB figure I quoted earlier in this topic, it shouldn't be an issue. If it's considerably more, an SLC SSD would be the way to go.
 
Valnar said:
Yes, everything you said is correct, just like everything I said is correct. But the difference is putting a price on peace of mind. The infamous car analogy comes into play. Buy a new car for more money but have less problems? Or buy a cheaper used car, but it'll have higher maintenance.
Click to expand...

That analogy appears to make sense when discussing cars, but not so much when you think of it in terms of this thread. The newer drives are supposedly the ones that will require more maintenance. You could say, "If I buy a new car and let it sit in the garage without ever driving it, it won't last as long as my old car which also sits in the garage and isn't ever driven." If the information in this thread is to be believed, a new SSD would actually require more maintenance than an older one since it would have to be refreshed (or "driven") more often than an older one in order to ensure the data was retained.

As far as your thoughts on the older SLC flash, I agree...it seems to be the sweet spot as far as speed an longevity. The older SLC thumb drives are highly sought after by some, and I certainly wouldn't mind picking up an SLC SSD if provided peace of mind as far as data integrity.
 
Last edited:
My analogy was more in reference to the cost. Paying more up front vs paying more later - or dealing with "maintenance".
 
I just ordered the Intel X25-E. I wanted to get one while I still can. I anticipate 64GB will be enough for a C: drive for years to come, considering my current C: partition has 39GB used.
 
No one's even making drives w/25nm Flash yet, splurging on an X25-E SLC for an OS drive on a home PC seems rather silly to me, if you're just doing it over data retention paranoia, but knock yourself out... An X25-M or comparable drive would've easily lasted you 5 years. /shrug
 
Impulse said:
No one's even making drives w/25nm Flash yet, splurging on an X25-E SLC for an OS drive on a home PC seems rather silly to me, if you're just doing it over data retention paranoia, but knock yourself out... An X25-M or comparable drive would've easily lasted you 5 years. /shrug
Click to expand...

You don't know that. I hope for everyone else you're right though.
 
SLC is more of a known tested storage quantity having been around a long time in enterprise applications. MLC is a relatively unknown storage quantity having been around for approximately two years. SLC of the same process will have more than 10x the number of write endurance and data retention as MLC. So if your data is critical SLC well worth the money. Nothing to sigh or shrug about if you can afford it.
 
Valnar said:
You don't know that. I hope for everyone else you're right though.
Click to expand...

Intel rates them for 100GB a day for five years, I tend to trust their massive R&D... Most of the speculation in this thread is about drives and flash that aren't even in use yet. Even if it lasts 2-3 years, at worst, it's still larger than an X25-E and about 1/3rd the cost, I'm not the 'least bit concerned... It's not like I store priceless data on it, and I keep several backups of my OS install (freshly installed and several up-to-date revs). Beats the heck out of paying $700+ for something that's gonna be outdated in a year anyway. For enterprise applications, servers, etc. I think it's absolutely worth it, but for a personal or home system? Please...
 
Except the speculation about 25nm flash had been previously stated by Micron and some industry insiders.

Is NAND flash about to hit a dead end?
Some believe the ubiquitous chips have five years left — at most.
February 4, 2010
http://www.computerworld.com/s/artic...it_a_dead_end_

"Computerworld - When IM Flash Technologies (IMFT) announced this week that it's manufacturing a 25-nanometer (nm) NAND flash chip, the company also admitted that shrinking the technology much further may not be possible because of problems with bit errors and reliability.

If that turns out to be true, one of the basic building blocks used for storage in solid-state drives (SSDs) and memory cards may be nearing a dead end.

"I think in the next four years or five years, it's probably going to be the case" that NAND will no longer be the storage medium, said Gregory Wong, a flash memory analyst with market research firm Forward Insights. "Everybody's looking at alternatives."​


BTW, Intel's SSD's are far from perfect.... even the enterprise SLC


Pillar kicks Intel's SSD to the curb, upgrades storage array
It blames the solid-state drive for operations time-outs
http://www.computerworld.com/s/arti...ntel_s_SSD_to_the_curb_upgrades_storage_array
By Lucas Mearian
September 16, 2009 03:03 PM ET

"Pillar Data Systems Inc. said this week that it's replacing Intel's X25-E solid state disk (SSD) drive as an option for its storage arrays with a drive from STEC Inc. because of firmware problems with the Intel's drive that lead to performance slowdowns."​


Now that the bloom is off of the elusive enthusiast rose, we can maintain pro/con critical thought past the marketing literature.
 
Oh, btw, Tom's Hardware actually made itself useful by reiterating this about major manufacturers in a recent SSD Q&A with Kingston.

"Getting new information from a major vendor isn't always easy. Once you get through the marketing pages and white papers, it can feel like everything has all been said before. And maybe it has. But by getting information straight from the horse's mouth, in a candid, informal conversation, sometimes new facts emerge or get said in new ways that make them more intelligible or pertinent."​

So perhaps there's hope for enthusiast websites after all. In fact, I encourage everyone to read the article. It's actually pretty good.

http://www.tomshardware.com/reviews/kingston-ssdnow-ssd,2550.html
 
Next generation SSDs will of course have higher lifespan because they are greater capacity / more flash cells. That means wear leveling has much more cells to play around with and the average predictable lifespan of the unit will increase.

Better yet, when an SSD has exceeded its write cycles; it will turn read-only. So while it would mean you lost hardware, you don't lose your data.
 
You must log in or register to reply here.
Back
Top