Djembe

The Future of SSDs: How do semiconductor process shrinks affect the de

Recommended Posts

I've been interested in SSDs for the last couple of years and have eagerly read everything I could find on the subject, including many different reviews, Storage Search articles, and Intel IDF presentations. My question to the members here is as follows: how long will Solid State Drives be a viable storage option for consumer computing? Storage Search predicts that SSDs will be around for a long time and grow until they have capacities in the petabytes, but the editor there has a strong focus on the enterprise sector, which has different needs.

The constraining issue I see is an inevitable reduction in reliability with each new generation of SSDs. SSD amounts of storage available per drive and drive speeds seem to increase and prices per gigabyte seem to decrease as semiconductor process technology shrinks, which gives both manufacturers and consumers reasons to want ever smaller process technology in SSDs. However, SSD reliability in the form of writes allowed also lowers with each process shrink. MLC which used to be able to handle 10,000 writes per block at 50nm can now only take half or less of that amount at 32nm or 25nm. Intel estimates 1000 writes per block as the necessary endurance required for the vast majority of SSD users, so what happens when the process tech shrinks to a point where that is no longer possible? There is also some evidence that process tech cannot continue to shrink indefinitely due to quantum tunneling which would prevent data from being able to be stored for any reasonable length of time.

Keeping these process shrink factors in mind, will SSDs survive into the next dominant storage medium as Storage Search predicts or will they end up being a shorter-term stopgap tech that never achieves dominance because a newer tech supersedes it? I'm torn about this and would be interested in your opinions. Also, feel free to bring up areas I haven't thought of.

Share this post


Link to post
Share on other sites

The 25nm NAND even posts 3k write cycles for the Tier 1 NAND, so that shouldn't be a problem, at least for now.

SSDs are going to be around for a long time, especially if you count in the hybrid solutions as well, which look to be gaining steam. There's nothing else that will be commercially viable in the next many years to compete with flash or HDDs.

Share this post


Link to post
Share on other sites

I agree that there doesn't seem to be anything else commercially viable now and that currently SSDs are in great shape. My concern or question was about what will happen in say 5 years when it's entirely possible that process shrinking can either hit a limit, bring the write limits to where they're impractical for a 5-year life cycle, or both. Will many people be kicking themselves for buying what they did not realize were unreliable products or will there be technology to mitigate the effects? And if such mitigating tech will keep NAND viable for many years to come (say 20+), what would it look like or what would it do?

Share this post


Link to post
Share on other sites

Depending on how processes or materials improve, things could get smaller and more reliable... who knows what is going to happen in 5 years. One important way to look at whats going on right now is depending on how the manufacturer designs the product and prices it, you could hit a certain mark where the device has X capacity, has plenty of overprovisioning to handle the decreased write-life, and offer higher total capacity for the same price as before. The problem we are running into right now is manufacturers are adapting new technology in older platforms and taking a hit on performance, size, and capacity without passing any savings onto the consumer.

Share this post


Link to post
Share on other sites

I agree that there doesn't seem to be anything else commercially viable now and that currently SSDs are in great shape. My concern or question was about what will happen in say 5 years when it's entirely possible that process shrinking can either hit a limit, bring the write limits to where they're impractical for a 5-year life cycle, or both. Will many people be kicking themselves for buying what they did not realize were unreliable products or will there be technology to mitigate the effects? And if such mitigating tech will keep NAND viable for many years to come (say 20+), what would it look like or what would it do?

hopefully as years go by, better caching mechanisms would be employed :)

Share this post


Link to post
Share on other sites

You can correct me if I'm wrong on this, but I thought caching was primarily done to improve or maintain performance, not to boost reliability. I don't see any kind of performance limitations in the future of SSDs, but I do see the possibility of reliability problems as smaller processes are used.

As an example, I just read a review of an SSD with Sandforce's new controller. It adds on additional ECC data and greater overprovisioning due to reduced reliability of smaller-process NAND. From what I've read elsewhere, that kind of additional ECC data especially needs to increase exponentially with each process shrink to maintain the same level of data integrity. Doing this vastly increases the complexity (and price) of the controller when you go out 2-3 SSD generations from now and the required overprovisioning to maintain a reasonable life cycle goes from 20 to 30 to 50% or more of the total NAND on the board.

Unless some very significant techniques are developed, I do not currently see how the trend of increasing SSD capacity and performance and lowering price per gigabyte can continue in a long-term scenario. However, since I am a fan of the tech and would like it to increase in market penetration, I'm hoping someone has an idea of something that would make SSDs truly viable in the long run.

Share this post


Link to post
Share on other sites

You can correct me if I'm wrong on this, but I thought caching was primarily done to improve or maintain performance, not to boost reliability.

The fact is that "write caching" enhance SSD reliability because rewriten sectors stay in the cache. So there are less NAND writes. The same "write caching" feature does not improve HDD reliability but it does improve SSD reliability.

...adds additional ECC data and greater overprovisioning due to reduced reliability of smaller-process NAND.

...required overprovisioning to maintain a reasonable life cycle goes from 20 to 30 to 50% or more of the total NAND on the board.

NO. Overprovisionning to more than 20% is an economical and pratical nonsense.

Current overprovisionning rates are here to totally avoid the risk of freezing at garbage collecting time by providing free space "for sure" (Even when TRIMing features can't be enabled because of RAID usages or legacy hw or sw)

I read that 25nm NAND supports between 3000-5000 write cycles which are more than enough to support your personal usage for years

Share this post


Link to post
Share on other sites

Ah, yes. Thank you for reminding me about write caching. You're right, that does increase reliability with most data loads.

You are also correct regarding overprovisioning requirements and write cycles in 25nm NAND. My question is what happens with 16nm or 11nm NAND. As I see it, in order to continue increasing the capacity and performance of SSDs while continuing to lower price per gigabyte (which typifies the current SSD market trends), the companies that make flash memory will have to use smaller and smaller processes. My concerns are that the upcoming smaller processes will carry with them reliability challenges that may be difficult to overcome.

Share this post


Link to post
Share on other sites

I'm not sure about reliability, I'm more concerned about endurance. But either way, your concern is valid. The endurance issue was paramount when SSDs first came out, now we're seeing 25nm NAND that's dropped to 3000 write cycles form the 5000 write cycles of 32-34nm NAND. Will anyone notice? Probably not, as few consumers ever see even 1000 write cycles in the 5 year life expectancy of an SSD. Anyway, we're years from even contemplating another die shrink - but your concerns and questions are still valid.

Share this post


Link to post
Share on other sites

After a short conversation in a different thread, I realize that process shrinks are not always understood by everyone interested in SSDs. As a result, I'll try to clear up what they are and why they're done.

Since SSDs are built on computer chips (unlike hard drives, which are built on spinning magnetic disks), they are subject to the same types of technology advances as processors and graphics cards. The big chip manufacturers realized a long time ago that in order to improve performance of their chips without making the chips themselves much bigger, the components and the connections that compose the chips must be made smaller. The process of making computer chips from smaller components and connections is called a process shrink. Chips are typically labeled by how far the space is between 2 of the same type of component on the chip (this explanation could get much more technical, but I'm trying to keep this as simple as possible). Since flash memory is made of computer chips, it is also labeled by this distance.

As processing chips use smaller processes, they get more powerful, are less expensive to produce, use less electricity, and produce less heat. All in all, process shrinks are great for processors and graphics cards. Flash storage (what SSDs are made of), also known as NAND, also benefits from process shrinks. It allows more data to be stored in the same amount of space, can move that data at faster speeds, and is less expensive to produce. However, unlike processors, NAND also has a down side to process shrinks. The smaller the process used, the harder it is to keep data intact. For more information about exactly how this works, I would recommend reading Anandtech's SSD information articles and Intel IDF presentations such as this one. Since SSDs have a limited number of times they can write data to each block of NAND, making it harder to maintain data integrity means that number is reached faster on each successive smaller process that is used. Basically, they wear out faster.

This is why SSDs have controllers that manage how data is read and written to the computer chips inside. In another thread, a forum member said that the reduction in write cycles (faster wearing out) of SSDs using smaller-process NAND makes the flash inferior. However, since it is also faster, has a higher capacity, and costs less, I believe this argument is overly simplistic.

Currently, SSDs use NAND that can use between 3000 and 10000 write cycles before being unwritable. According to Intel, the overwhelming majority of SSD users do not require more than 1000 write cycles per block of NAND, which means that today's SSDs should be more than capable of satisfying anyone's data usage patterns. However, the number of write cycles will continue to diminish with each process shrink. These will also continue to give SSDs higher capacity and faster access and transfer speeds while lowering prices, and are therefore highly likely to continue. This thread is about discussing ways SSD and controller makers can keep the benefits of process shrinks and lessen the effects of the detriments.

Share this post


Link to post
Share on other sites

That still gives you an inferior product.

By an "old" 80GB SSD and buy a "new" one - the new one will fail sooner because the Flash takes fewer writes.

-> i.e. inferior product

And the argument "most people will only ever need" - the only response I can give to Intel with respect to that is "shut up you capitalist idiots".

Of course if you expect everybody to buy a new SSD every 2 years you can lower the write cycle - but to most people a SSD should replace a HDD, and most people would possibly even hope to use it for a longer period of time because it doesn't wear from little shocks as a HDD would.

Edited by DetlevCM

Share this post


Link to post
Share on other sites

That still gives you an inferior product.

By an "old" 80GB SSD and buy a "new" one - the new one will fail sooner because the Flash takes fewer writes.

-> i.e. inferior product

And the argument "most people will only ever need" - the only response I can give to Intel with respect to that is "shut up you capitalist idiots".

Of course if you expect everybody to buy a new SSD every 2 years you can lower the write cycle - but to most people a SSD should replace a HDD, and most people would possibly even hope to use it for a longer period of time because it doesn't wear from little shocks as a HDD would.

I can't agree...the early and still current generation SSDs were made with huge write cycles partly because that's what it took to get people to feel comfortable buying SSDs. To continue with progress, we're giving up a few write cycles that people don't use anyway, I don't see why you're grumpy about this.

Share this post


Link to post
Share on other sites

I can't agree...the early and still current generation SSDs were made with huge write cycles partly because that's what it took to get people to feel comfortable buying SSDs. To continue with progress, we're giving up a few write cycles that people don't use anyway, I don't see why you're grumpy about this.

Because you loose more than you gain.

Intel right now pretty much guarantees I think 3 or 5 years at 5 or 6TB per year - something I actually managed to write to my SSD so far.

-> Now if I would be shooting more photographs, maybe even more videos and did more Photoshop editing (rather than CameraRAW) then that's quite easy to exceed. And what I've done with videos was only reencoding so far - i.e. not that write intesive.

Now on the positive sides those tend to be more sequential writes.

If you look at small random writes, then they can wear out a SSD significantly faster - it was only a couple of TB, I think 15 or 16TB for the Intel SSD.

-> I honestly don't want to be forced to replace a SSD every few years because the manufacturer can gain a larger profit by selling less durable equipment.

If you buy a HDD for your desktop PC it can pretty much last forever - yes, at some point the motor will possibly fail, but I have not heard that anybody has ever managed to wear out the data layer on a HDD (and I don't mean headcrashes or similar).

Why should we put up with SSDs that are less durable when the first models were more durable?

If I sell you a car and say it'll do 200.000km then fail, don't worry, most people buy a new one after 100.000km - would you buy that?

I wouldn't.

And yes, you can buy SLC SSDs, but they are more expensive and tend to have a smaller capacity - not really an option for the average consumer.

Share this post


Link to post
Share on other sites

And yes, you can buy SLC SSDs, but they are more expensive and tend to have a smaller capacity - not really an option for the average consumer.

What you seem to be missing, in my opinion, is that the above is exactly why process shrinks must happen and are going to continue. Right now, you're happy with a 160GB SSD with roughly 240/100 MB/s read/write speeds that cost you $400 or so. Will you still be happy if you buy a new computer in 5 years and are still paying the same amount for the same capacity at the same speed? Most people wouldn't be happy with that because they expect technology to keep improving, not stay the same. If you stop process shrinks, then you're not going to see higher capacity SSDs, you're not going to see faster SSDs, and you're not going to see less expensive per gigabyte SSDs. Essentially, in order to maintain the number of read/write cycles, you'd also have to stop everything else. Would you really be satisfied with that?

Share this post


Link to post
Share on other sites

What you seem to be missing, in my opinion, is that the above is exactly why process shrinks must happen and are going to continue. Right now, you're happy with a 160GB SSD with roughly 240/100 MB/s read/write speeds that cost you $400 or so. Will you still be happy if you buy a new computer in 5 years and are still paying the same amount for the same capacity at the same speed? Most people wouldn't be happy with that because they expect technology to keep improving, not stay the same. If you stop process shrinks, then you're not going to see higher capacity SSDs, you're not going to see faster SSDs, and you're not going to see less expensive per gigabyte SSDs. Essentially, in order to maintain the number of read/write cycles, you'd also have to stop everything else. Would you really be satisfied with that?

If it is clear as to why I am paying the same for the same performance and space, then yes.

And I am pretty sure that you can even make SSDs a bit faster using the "old" technology - I don't think it's performance maximum is quite reached yet.

What do I gain from a cheaper product when I have a higher change of having to replace it? -> Nothing.

Maybe it would be nice to have two lines - one with the "old" MLC flash for reliability, one with the new one.

I suppose the problem is really today's throwaway society - people will not buy quality they can keep, they'd rather buy junk 30 times per year even if it costs more in the end...

Share this post


Link to post
Share on other sites

Detlev, did you see the Micron data I posted this morning? They're showing endurance of 72TB over 5 years. That's not a step backward - 40GB per day of writes...there's no way any normal person will hit that.

Share this post


Link to post
Share on other sites

Detlev, did you see the Micron data I posted this morning? They're showing endurance of 72TB over 5 years. That's not a step backward - 40GB per day of writes...there's no way any normal person will hit that.

40GB per day is easy to hit in:

Music production

Video Editing

Professional Photography - especially if you work with Photoshop edits (i.e. not just CameraRAW)

Share this post


Link to post
Share on other sites

Fair point, but would you consider that normal usage for the entire client computing user base? And, if you're doing all that on an SSD so be it - you're a heavy user. The time savings still make it better than an HDD.

Share this post


Link to post
Share on other sites

Fair point, but would you consider that normal usage for the entire client computing user base? And, if you're doing all that on an SSD so be it - you're a heavy user. The time savings still make it better than an HDD.

Normal usage - no, possibly not. On the other hand - aside from the "hardware addicts" those users are the prime customers for a SSD.

In an ideal world we'd have both... but we live in a profit driven world... :(

Share this post


Link to post
Share on other sites

Agreed - I would say though that the way you're using a drive, either you *might* burn through them faster, or you consider going with an enterprise class drive that has more over provisioning. The cost of those drives isn't a whole lot more in some cases. I still think you're somewhat unique in the way you're using the SSD.

Share this post


Link to post
Share on other sites

Agreed - I would say though that the way you're using a drive, either you *might* burn through them faster, or you consider going with an enterprise class drive that has more over provisioning. The cost of those drives isn't a whole lot more in some cases. I still think you're somewhat unique in the way you're using the SSD.

-> Well, I'm just doing normal photo editing. Just CameraRAW and I went through the 20GB per day without too much trouble... -> and that's as a hobbyist who was doing a lot of shooting in summer.

-> So not that unique ;) and the professionals will easily surpass me.

Share this post


Link to post
Share on other sites

Has anyone announced what the long term cost savings from the recent 32/34nm to 25nm die change is going to be yet? I know 10-15% has been thrown around, but would that savings increase as the capacity increases? The last drop in write cycles was 40%, but what if as the prices equalize this next year, the cost of a 240GB SSD is roughly the same as a 120GB drive right now? When you look at the overall picture, individual cell life went down, but total capacity lifespan increased given the same price. So each generational change still brings greater benefit to basic and enthusiast users when you look at it from the device level.

Now none of this will hold true if prices only drop 15%, since you lost a higher percentage of write cycles... but if we see another surge in capacity for the same pricepoints it might work out well long term.

Share this post


Link to post
Share on other sites

Price drops will continue to be a slow slide from what I can tell. 240GB for the 120GB price today isn't happening any time soon, but 240GB today is pretty tolerable considering the prices when SSDs started.

Share this post


Link to post
Share on other sites

It's also important to consider that some SSD manufacturers are starting to develop larger profit margins to make their business more profitable like what OCZ did with the switch to 25nm NAND. It cost them less and they tried to slide it under the noses of their customers while charging the same amount. If they hadn't had to use more overprovisioning, they might have gotten away with it, too. This kind of thing is another reason why choosing a manufacturer is important. Most companies are not as honest as Intel and Western Digital about their SSDs.

Share this post


Link to post
Share on other sites

It's also important to consider that some SSD manufacturers are starting to develop larger profit margins to make their business more profitable like what OCZ did with the switch to 25nm NAND. It cost them less and they tried to slide it under the noses of their customers while charging the same amount. If they hadn't had to use more overprovisioning, they might have gotten away with it, too. This kind of thing is another reason why choosing a manufacturer is important. Most companies are not as honest as Intel and Western Digital about their SSDs.

Very true. At least from what came out last week, out of the big key players, it really looks like only OCZ tried to take that path. Its hard to say what some of the smaller vendors are doing since their turnover isn't as high as the big guys, but if they are smart they won't try and do the same thing. This is especially true for SandForce licensees which offer similar/same configurations as other licensees and try sell by brand name, trust, and price. If you are sold on Intel or Crucial and they did something like this, you need to find a whole new platform... if you dont like one of the SandForce licensees you just look for another one and get very similar performance.

Share this post


Link to post
Share on other sites

Create an account or sign in to comment

You need to be a member in order to leave a comment

Create an account

Sign up for a new account in our community. It's easy!

Register a new account

Sign in

Already have an account? Sign in here.

Sign In Now