Next generation of SSDs to have a shorter lifespan?

spectrumbx

[H]ard|Gawd
Joined
Apr 2, 2003
Messages
1,647
Reading this AnandTech's article on Intel and Micron next generation SSDs built on the new 25nm process, there is a paragraph with important implications:

The other major change to the 25nm NAND is an increase in the page size. At 50nm and 34nm, IMFT’s page size was 4KB. At 25nm on the 8GB device the page size is now 8KB. Block size has also gone up from 128 pages to 256 pages. This will obviously have performance implications and require some firmware reorganization, but given that Intel has known this was coming for some time now I would expect that its 3rd generation drives will be optimized for 25nm.

This means that the block size is now 256 pages x 8KB = 2MB. :eek:
Remembering that in order to overwrite data you need to erase a full block, these new drives will not be good for small write applications.

People doing things like video editing or using SSDs for mostly read operations will have nothing to worry about, but people using SSDs for things like databases and others might want to stick to SLC or the current generation MLC.

Normally one would think Intel and Micron would bring out new controllers that efficiently address this issue. However, increasing the page and block size is inherent to controller limitations. Better controllers would simply do without that increase.

Since we have a full year before we start seen these new drives, we can only hope that better controllers will come into the picture.
 
Yea but I dont really see it being a problem.
The current gen SSDs will outlive any other component in a computer.

What the mfgr's want to do is increase space and decrease cost/price so there will be more demand and thus higher profits.
Also will mean better GB/$

Plus chances are we will see improved controllers
 
lol, this is all so hypothetical at this point. Wait until they hit the market before making such claims.
 
Theoretically 25nm NAND will indeed have a shorter lifespan than 32nm. Of course, attempts will be made to improve the controller and process to somewhat bridge this lifespan gap. There's also a chance on paper that 25nm NAND could be initially slower in speed than what's available today.
 
Number of write cycles decrease with each process shrink as well, together with the data retention. Currently data retention for SLC Flash is about 10 years, for these new MLC Flash it could be as low as 8-12 months.

It's been known for a while what the limits of Flash are, and we're getting pretty close already. Unless there's some miraculous breakthrough at some point, all one can do is mask these limitations through various tricks. It'll be unlikely we'll see Flash progress beyond 25 nm any time soon.
 
We could see a theoretical limit of 22nm with the current process. But yeah, 25nm is already pretty close to the breaking point. If anything the shrink will be more advantageous to manufacturers bottom line. I wouldn't even be surprised if 32nm remains high-end and 25nm bargain brand at first.
 
Mission critical enterprise stuff should be running on SLC drives already anyway...
 
lol, this is all so hypothetical at this point. Wait until they hit the market before making such claims.

Err... nothing hypothetical here... just facts.

Yea but I dont really see it being a problem.
The current gen SSDs will outlive any other component in a computer.

What the mfgr's want to do is increase space and decrease cost/price so there will be more demand and thus higher profits.
Also will mean better GB/$

Plus chances are we will see improved controllers

Well, I see some serious problems.
Shorter data retention and increasing write amplification aren't good if you care about your data.
Everybody is focusing on benchmark numbers when there are more important issues.

Yes, the next area for lifespan improvement will be on the controller.
However, all the tricks to be had have already been exhausted at this point.
The only area left will be to reduce the block size in order to reduce write amplification.

It is true that it is not all bad news since capacity is increasing, but let's all know and remember the implications here so that we can make better choices.
People like myself that are after high write endurance at an affordable price are not hearing what we want to hear.

Further, the focus on die shrinkage is a little obsessive.
How about focusing on producing quantity of dies at any reasonable manufacturing process?
I will take a more reliable 3.5" drive over a 2.5" one for the same capacity.
 
I don't think a 2.5" form factor vs 3.5" would make much of a difference towards the reliability or cost of current SSD... If anything it's probably easier to just manufacture and market a single line that they can pitch to OEMs (for laptops) and enthusiasts alike. What kinda database are you running at home that you're so worried about a SSD's lifespan anyway? :p
 
I'm not worried. Just as with every process shrink in the CPU space, Intel (and others) will find a way to mitigate the negative effects of smaller transistors
 
I don't think a 2.5" form factor vs 3.5" would make much of a difference towards the reliability or cost of current SSD... If anything it's probably easier to just manufacture and market a single line that they can pitch to OEMs (for laptops) and enthusiasts alike.

My comments on the form factor were less about the form factor itself and more on the manufacturing process shrink focus.
To fit more capacity on a 2.5" form factor, a die shrink is needed.

My comments were about unbinding ourselves with form factors and focusing on packing as many reliable dies as it takes to reach a given capacity.

It costs money for R&D and to create new fabs for each die shrink.
If each die shrink brings more issues, then there should be a cut off point where mass manufacturing becomes the greater focus.

What kinda database are you running at home that you're so worried about a SSD's lifespan anyway? :p
Top secret. ;)
 
Number of write cycles decrease with each process shrink as well, together with the data retention. Currently data retention for SLC Flash is about 10 years, for these new MLC Flash it could be as low as 8-12 months.

That's the insane part about these new MLC's. The cells holding the charges are so small that the electrons are leaking out faster than ever. If you don't refresh your drive on a 6 month cycle you may see files rarely written to like MBR's and OS files evaporate on you.
 
ARe holding the 25nm parts in your hands? Unless you do, all you are doing here is just discussing theory. You have no idea what Intel/Micron gonna do in reality.

You need to read more on the technology and understand the implications being discussed in this threads.
Until then, it is theory to you and fact to the rest of us. ;)
 
That's the insane part about these new MLC's. The cells holding the charges are so small that the electrons are leaking out faster than ever. If you don't refresh your drive on a 6 month cycle you may see files rarely written to like MBR's and OS files evaporate on you.

Yeah, people rarely seem to realize the impact of decreased data retention rates. Lots of people have files on their drives which they rarely access. I know I have cruft on my HDDs which I haven't touched in years. If I had used an MLC SSD instead, I would have lost this data already unless it happened to be on some blocks which got refreshed due to unrelated writes, or some algorithm moved static data around. The latter decreasing the write cycles even more, of course.

Flash seems like a losing technology battling against impossible odds. Its only chance of survival at this point is at staying big (65 nm and up) and just cramming as many ICs in an SSD as possible. I'm surprised there don't appear to be any 3.5" SSDs.
 
Where do you guys get these numbers from? 6 months?

There was an article on Nikkei's Tech-On! a while ago with interviews featuring people in the industry who basically described this decrease in data retention.
 
Flash seems like a losing technology battling against impossible odds. Its only chance of survival at this point is at staying big (65 nm and up) and just cramming as many ICs in an SSD as possible. I'm surprised there don't appear to be any 3.5" SSDs.

You can pry this losing technology from my cold dead hands tho... :p It's not like HDDs are any more reliable, if anything SSD decay will probably be more predictable than that of HDD's, and people will just keep using SSD for their OS drives and HDD for data (the big losers in that scenario are people who depend on a laptop exclusively, in which case you'd have to deal w/larger dual-bay laptops or external drives).
 
Last edited:
You can pry this losing technology for my cold dead hands tho... :p It's not like HDDs are any more reliable, if anything SSD decay will probably be more predictable than that of HDD's, and people will just keep using SSD for their OS drives and HDD for data (the big losers in that scenario are people who depend on a laptop exclusively, in which case you'd have to deal w/larger dual-bay laptops or external drives).

Problem is with retention so short it is not safe to use SSD for OS as there are many files that are static. In fact most would be outside of log and temp files so you will need to reinstall everything (or ghost and copy back) every few months to prevent the data evaporating.

The issue of retention was never raised before as they used to be in the decades range.
 
Problem is with retention so short it is not safe to use SSD for OS as there are many files that are static. In fact most would be outside of log and temp files so you will need to reinstall everything (or ghost and copy back) every few months to prevent the data evaporating.

The issue of retention was never raised before as they used to be in the decades range.

Yeah, with original SLC having 10+ years retention it was pretty much a non-issue unless you were thinking of using it for archiving (a terrible idea with Flash no matter what). Now that 1 year appears to be the new point where data begins to get lost, things do become serious. That's why 25 nm MLC is going to be a very risky thing, as the tolerances have become so small and write damage so much more harmful that it may very well begin to backfire. Long term I wouldn't be surprised to hear reports from people having lost data stored on some USB Flash stick or so.

I have got floppy disks around from 10 years ago which I can read just fine. It'd be sad if that kind of data retention became a thing of the past, especially with optical media being not much better.
 
Sometimes I wonder if it would be more worth while in the future to jump straight to ram drives (those adata's). At least that keeps you on your toes.

It's entirely possible for intel within the next year start having their firmware secretly shuffle data around the drive every few months to refresh them. If they don't want reports of drives auto destructing after a year it would very much be mandatory.
 
Sometimes I wonder if it would be more worth while in the future to jump straight to ram drives (those adata's). At least that keeps you on your toes.

It's entirely possible for intel within the next year start having their firmware secretly shuffle data around the drive every few months to refresh them. If they don't want reports of drives auto destructing after a year it would very much be mandatory.

It seems to me that a new technology like PCM (Phase Change Memory) may end up replacing Flash before long. It'd be faster, easier to manufacture, denser, truly random access (reads and writes) and basically be closer to DRAM than Flash, yet since it uses the physical state changes of some material, it has an in theory indefinite data retention plus no limit to the number of writes.
 
You need to read more on the technology and understand the implications being discussed in this threads.

I understand the implications and will accept them as facts when I see people losing data due to said implications.



I see speculation that we may have data loss....
Now that 1 year appears to be the new point where data begins to get lost
or
maybe will need to reinstall everything (or ghost and copy back) every few months to prevent the data evaporating.
.

Maybe we'll get some facts when it starts happening?
 
I've no idea if that's really how 25nm flash durability will pan out... But I seriously doubt Intel would ever put out a drive to market that would only last a year, I can tell ya that much. The algorithms on these things are pretty advanced, and if that optimizations on that end alone won't cut it then I'm sure we just won't see 25nm flash SSD in 2010. Personally I could care less if I had to restore an image backup of my OS every 6 months, but that would obviously never fly w/the average consumer and the last thing SSD manufacturers want is to push back the adoption rate of SSD w/a shoddy gamble on smaller flash.

I don't see anything terribly wrong w/settling at 35nm for a while anyway, increased demand will eventually drive prices down substantially regardless of whether they shrink it further or not... 'Course that's an enthusiast's pov, I can deal w/small-ish drives, even on my portable systems. The average consumer's not gonna bite until he can get at 'least 120GB w/o paying too much of a premium for it... +$200 over the cost of a system w/o a SSD, tops, that's where I think we'll see 'em really hit their stride w/the mainstream market.
 
...

It's entirely possible for intel within the next year start having their firmware secretly shuffle data around the drive every few months to refresh them. If they don't want reports of drives auto destructing after a year it would very much be mandatory.

This is exactly what Intel is currently doing in its firmware.
In addition to free space, static data is also shuffled around often enough that for the average user, the risk is properly addressed.
The point remains that not everyone is your average user as some people are using SSDs in pure read-only mode. I am not sure if Intel's GC/background routine properly shuffles the data around when a drive is being used in mostly read-only mode (it should).

I understand the implications and will accept them as facts when I see people losing data due to said implications.

I see speculation that we may have data loss.... or .

Maybe we'll get some facts when it starts happening?
The key term in the title of the thread is "shorter lifespan".
There is absolutely nothing speculative about that. It's a fact.

Now, whether it is an alarming concern can be up to debate.
When someone tells you his/her USB thumb drive died or is exhibiting read/write issues, what do you think is happening?
If you have never heard of these issues, please Google them.
There is nothing speculative about data loss on NAND devices.
You might not have experienced it, but plenty of people have.
 
The key term in the title of the thread is "shorter lifespan".
There is absolutely nothing speculative about that. It's a fact.

Let's see some links to people who have tested this fact. All I am seeing is some people saying over and over that data retention is less than a year for new SSDs, but not providing any evidence.
 
When someone tells you his/her USB thumb drive died or is exhibiting read/write issues, what do you think is happening?

I dunno what to think but I don't jump to a conclusion that the drive is exhibiting data loss by being old and worn out.

Or is that the only reason USB drives die?
 
Personally, I don't know very many geeks that have 10-year old flash drives with dying flash cells (presumably they'd be like 64MB large and the owner would still have a reason to use it?)... :p I do have a 7 year old 256MB Creative MuVo NX (a flash drive / MP3 player combo) that still works as good as the day I bought it tho, I started using it to exercise with a few years ago figuring if I smashed it no worries (plus I can use it w/rechargeable AAA batteries), still haven't killed it, heh.

That's not here nor there, I'm just saying, I think manufacturers would know better than to put out a product that would literally die soon as the warranty wears out... We're not talking about Detroit here! :D
 
My comments on the form factor were less about the form factor itself and more on the manufacturing process shrink focus.
To fit more capacity on a 2.5" form factor, a die shrink is needed.

More capacity than what? The limiting factor for most people today when it comes to SSD isn't the amount of apace one can fit in a 2.5-drive, but price.
 
Let's see some links to people who have tested this fact. All I am seeing is some people saying over and over that data retention is less than a year for new SSDs, but not providing any evidence.

The fact (once again) is that the new SSD drives as referenced in my first post will have a shorter lifespan.
Subscribe and read this doc if you want someone with higher credentials to tell you why.

Any date that has been thrown should be regarded as speculative... yes.
However, I find it to be valuable speculation driven by intelligent guesstimates.
The fact is, for a given drive, some of the cells could fail within a few days/weeks/months (and the block containing them will be marked as a bad block as it's done with HDs) while the rest of the cells could endure for years.
SSDs could even be much faster if it wasn't for the fact that every write is a write-test (write and test if the write succeeded in entirety - else, mark as bad block and try somewhere else).

I dunno what to think but I don't jump to a conclusion that the drive is exhibiting data loss by being old and worn out.

Or is that the only reason USB drives die?

Not that it would be the only reason, but one of the reasons; and hence, data loss due to data retention issues being real issues.
 
but one of the reasons; and hence, data loss due to data retention issues being real issues.

Yes but you have no stats to say how much can be attributed to data retentation.

The fact is that you are speculating about the severity of the problem and admit this yourself...
Any date that has been thrown should be regarded as speculative... yes.

It's been a nice topic of conversation but hardly alarming and only reinforces what a long term data saver should know....don't put all your eggs in the same basket and check it from time to time. :)
 
It's been a nice topic of conversation but hardly alarming and only reinforces what a long term data saver should know....don't put all your eggs in the same basket and check it from time to time. :)

So you mean I should back up my important data regardless of what sort of drive I'm storing it on? No way :p
 
So you mean I should back up my important data regardless of what sort of drive I'm storing it on? No way :p

I really hate bringin' up these new fangled ideas....there's always soo much opposition! :D
 
The fact (once again) is that the new SSD drives as referenced in my first post will have a shorter lifespan.
Subscribe and read this doc if you want someone with higher credentials to tell you why.

Thanks for the link, but that article is from December, 2003. The research would be 7 or more years old. That is a long time in electronics development. There have been scientists and engineers saying that we have reached the limit of transistor miniaturization for the last 20 years. But it seems that clever people keep inventing ways to make electronics smaller and better.

If that article is your main evidence, then I must point out that it is likely that multiple solutions to those problems have been discovered in the past 7 years.

If you have other, more recent evidence, then please post a link.
 
I found this series of articles to be very enlightening:

http://techon.nikkeibp.co.jp/article/HONSHI/20090528/170917/ (Part 1)

Part 3 talks about reliability: 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.

Also a telling quote:
The biggest difference between SSDs and HDDs, as a source at a major manufacturer of SSD controller IC explains, is that "The number of rewrites, data retention time and other characteristics are limited. SSDs cannot be used in the same way as HDDs, which have always been considered as effectively unlimited." It would be best, it seems, not to think of them merely as HDDs without any moving parts.

Then there are those who don't see an issue:
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."

Which of course kind of falls flat for markets where people will write more than 20 GB/day. There's also the reshuffling of static data to be kept in mind, as part of the SSD controller to combat data retention failure. Simplistic calculations rarely work in the real world. A serious video editing system, for example, would burn through hundreds of GBs/day at the very least. Would this mean that SSDs would only be suitable for netbook users who do barely more than browse and chat on IM clients?
 
So you mean I should back up my important data regardless of what sort of drive I'm storing it on? No way :p

Back it up on SSDs and you will need to back up your backup.
Will backups guarantee you better uptimes? :rolleyes:

Thanks for the link, but that article is from December, 2003. The research would be 7 or more years old. That is a long time in electronics development. There have been scientists and engineers saying that we have reached the limit of transistor miniaturization for the last 20 years. But it seems that clever people keep inventing ways to make electronics smaller and better.

If that article is your main evidence, then I must point out that it is likely that multiple solutions to those problems have been discovered in the past 7 years.

If you have other, more recent evidence, then please post a link.

Has anything changed from what it is talking about? Have you read it?
What fundamental problem does it state that you know of has been addressed since then?

State of denial? :confused:

I can't copy the content of the article. So, you will have to read it for yourself.

As NAND cells shrink, they contain less atoms (and electrons) to stabilize a charge leading to a greater occurrence of charge loss/gain over time (Google on the stability of atoms).
Normally, ECC is used when reading data off NAND cells, but ECC has a fault tolerance limit.

Now read this: http://features.techworld.com/storage/3212075/is-nand-flash-about-to-hit-a-dead-end/?intcmp=ft-hm-m
 
In other words, no you do not have any more recent factual articles. Only the speculative one.
 
In other words, no you do not have any more recent factual articles. Only the speculative one.

Did you ignore my post with links to interviews with people who work in the actual industry?
 
You can lead an enthusiast to facts but you can't make him/her think. But you can sell them plenty of bleeding edge peripherals if you make it look glossy enough and grease the palms of as many reviewers and bloggers as possible. Anyhow the facts are out there. It shouldn't be anyone's responsibility to lead anyone to them.
 
Back
Top