Lifespan of Solid State Drives

M

maverick786us

2[H]4U
2FA
Joined
Aug 24, 2006
Messages
2,425
We all know that Solid State Drives will soon replace the conventional Hard Disks because of its Speed, Silent Operation, Lower Power Consumption, High Mechanical Reliability etc.

However the biggest drawback that would concern anyone about SSDs is have limited lifetimes and will often wear out after 1,000,000 to 2,000,000 P/E cycles. Since my knowledge of internal structure of storage system is sluggish, what does it means? how much would be the lifespan of an SSD if we compare it with conventional hard disks?
 
We all know that Solid State Drives will soon replace the conventional Hard Disks because of its Speed, Silent Operation, Lower Power Consumption,
Click to expand...

That is very doubtful to me. In some cases yes but hard drives are here for at least 10 more years.

However the biggest drawback that would concern anyone about SSDs is have limited lifetimes
Click to expand...

Theoretically Intel SSDs have a longer lifespan than hard drives. I mean they specify that you can write 20GB/day every single day for 5 years. Its doubtful that most users will write anywhere near that so in theory they will last longer. At home I expect my intel SSD to be working in 2020 with all of my 10 hard drives long since dead.
 
drescherjm said:
That is very doubtful to me. In some cases yes but hard drives are here for at least 10 more years. .
Click to expand...

i guess the only reason conventional hard disks are still alive is because of lower cost per GB. Otherwise what benifits do conventional hard drives hold over solid state drives? Since its been more than an yr that solid state drives have reached consumer. Its prices should gi down rapidly. Because it does'nt seem that theie production cost is that much compared to conventional hard disks
 
Its prices should gi down rapidly. Because it does'nt seem that theie production cost is that much compared to conventional hard disks
Click to expand...

It is just the opposite. SSDs are made with many expensive parts while hard drives are made with few cheap parts. SSDs can only decrease the cost by shrinking the process node in the same way CPUs have done this, The problem is that with the current technology and materials shrinking the nodes causes SSDs to become less reliable.
 
Cost per GB is about the only major advantage that I can see, but it's a big one.

As for production costs, ever priced out a 25nm semiconductor fab lately?
 
There is one other advantage. SSDs can not be used for data archive. I believe current 34nm NAND only has a 2 year data retention if left unpowered. Please correct that if I am wrong.
 
Hard drives have a huge cost/gig advantage. Of course SSD prices are dropping but then so will regular hard drives. I don't see SSD drives hitting the current 1-2TB spot in the near future. (1-2 years out). While HDs are going to push past there soon. Seagate already announced a 3tb model, correct? While the current 2TB are slowly breaking the sub $100 point. I'm sure we'll see more deals before year's end at that price point.

There are definite advantages to SSD with speed being the largest case. But not everything needs that kind of $ for performance. Bulk storage will be around for quite awhile longer and unless there's some major innovations for SSD, HDs will be around for quite awhile longer.
 
drescherjm said:
There is one other advantage. SSDs can not be used for data archive. I believe current 34nm NAND only has a 2 year data retention if left unpowered. Please correct that if I am wrong.
Click to expand...

This appears to be the major problem with the higher density flash chips, and it looks like it's only going to get worse with smaller process geometries and more bits crammed in per cell. That said, I've heard doubts expressed about the longevity of super-high-density magnetic platters as well (though probably not as bad as flash). I don't think hard drives have ever been considered a good archival storage medium for a number of reasons.
 
It also depends upon your usage patterns. SSD do not respond to random writes as well as they do to sequential writes. for instance the write capacity of a x25 is:

X25-M 160GB - 100% random host data 4k writes = 15TB
X25-M 160GB - 100% sequential writes = 370TB


as you can see the different write usage patterns will correlate directly to the lifespan of your product.
 
The longevity of the hardware parts depends on each brand and model, it could theoretically be dozens of years depending on the quality of the components and manufacturing. Intel stated some of its X25 SSDs data longevity at 10 years.

As for the durability of the data, manufacturing, MLC vs SLC, NAND cells wear and loss of charge are the main factors. The twice cheaper MLC cells can be erased safely only 10K times, while SLC is about 100K times before the magnetic signal starts wearing off.

This is combined with more and more efficient wear-leveling algorithms to push SSD longevity further. A bigger free space would also reduce the number of write operations on the NAND cells for temporary files, although a really efficient wear-leveling algorithm would rotate all files around, including the static OS and program files. Sequential long writes vs. random small size write operations play a role too.

Over time, we should see more SSD-specific utilities, for instance one that can analyze your disk and notify you if it's time to replace it, one that will periodically read/rewrite the data to avoid signal loss for low usage disks, although at the current cost per GB, nobody in his right mind would under-use SSDs. The constant technological advances on this rather new field will also increase data longevity on the newer models.
 
Last edited:
I agree entirely Chimel, we really do not even have a file system that is geared completely towards SSD used as a standard right now. everything is written and coded to optimize the platter based drives. as companies embrace this tech more, and start to actually make programs for it specifically, the lifespan should increase dramatically.
it is like multi-core cpu's. they are everywhere, but there are precious few programs that utilize it efficiently. we could do far more with correctly designed software/games etc.
 
Regardless of what the math works out to on paper, I am not expecting this initial round of SSD's to last very long. My guess is 3 years.
 
well that may be up for debate, of course, only time will tell. I know a bencher who holds many hwbot points who has systematically raped his ssd's repeatedly, over the course of five years, even at one point fashioning an LN2 pot (GPU) onto one of his ssd's to attempt to speed it up, and they are still kicking. alive and well!
as a matter of fact, he just used his array in GOOC this week.
 
william_fontaine said:
No way HDDs will be used in 20 years except for huge storage, and maybe not even that.
Click to expand...

I highly doubt that NAND chips made using silicon will be the bases of what ever storage technology that replaces hard drives. There are too many problems with shrinking this technology to reduce the cost and increase the density and you can only shrink the process a few more iterations.
 
Last edited:
Pfft, so many advances already in the 10 minutes since my last comment: MLC NAND endurance has been increased to 50K write cycles per cell, TDK already rotates your static OS files around to ensure identical wear across all cells, Hynix started mass producing 20nm NAND chips, Foremay has a 2TB SSD drive in 3½" form factor, 1TB in 2½"...
 
Chimel said:
Pfft, so many advances already in the 10 minutes since my last comment: MLC NAND endurance has been increased to 50K write cycles per cell, TDK already rotates your static OS files around to ensure identical wear across all cells, Hynix started mass producing 20nm NAND chips, Foremay has a 2TB SSD drive in 3½" form factor, 1TB in 2½"...
Click to expand...

So much BS in such a short post. As an example, Hynix isn't producing 20nm nand. That's all marketing and spin. They are producing 20nm ***CLASS*** nand which by definition can mean the process can be as large as 29.9999999999999999999999999999999 nm. They are playing the same marketing game that samsung played back in April when they tried to snooker the press that they were producing 20nm nand. As of this point the smallest confirmed nand process is the IMFT 25nm.

And no, MLC NAND endurance hasn't been increased to 50K erase cycles. About the best eMLC you can buy is around ~30k erase cycles with general purpose MLC between 3k-10k cycles. All Anobit, which is what I assume you are refering to, is doing is applying heavier weight correction codes at the sacrifice of capacity. They are effectively relying on statistical probability that the error density within a read/write block can be overcome with the amount of HC/CRC bits they are using so that they can mathematically reconstruct the data. At some point they are probably going deep into probabilistic recovery which depending on their computational resources will either slow the data reconstruction time or result in increase power use.

And foremay? Really? you are going to buy your storage from a company that thinks silicon valley is in socal? No seriously, they are all into claiming their headquaters are in sic "Silicon Valley, CA, USA" but all their contact data points to Pasadena, CA. Last time I checked the only area considered silicon valley was in nocal. From all appearances they are more marketing than reality. I'd much rather trust my data to STEC or bitmicro in all honesty. You know, companies that actually design something instead of releasing PR about something.
 
drescherjm said:
There is one other advantage. SSDs can not be used for data archive. I believe current 34nm NAND only has a 2 year data retention if left unpowered. Please correct that if I am wrong.
Click to expand...

It was ~10 years for 90 nm SLC. 34 nm MLC is a year max. 25 nm will be around 6-8 months at the current rate data retention is decreasing.
 
They better solve that soon. I cringe at the thought of what the data retention on an ~11nm MLC would be.
 
drescherjm said:
They better solve that soon. I cringe at the thought of what the data retention on an ~11nm MLC would be.
Click to expand...

Elledan continues to post speculation as fact. The fact is that the JEDEC data retention specification for NAND flash is 1 to 10 years. 10 years when the chips are new, and 1 year after they have been cycled 90% of their rated lifetimes. I posted some details of this specification some time ago in another thread.

There is no special exemption for the 25nm or smaller chips. To meet the JEDEC spec, they will also have a data retention of 1 to 10 years.

At some point, the chip fab wizards will run out of tricks and will no longer be able to produce smaller NAND flash that meets the JEDEC spec. But it has not happened yet. Anyone who says otherwise is just speculating.
 
john4200 said:
Elledan continues to post speculation as fact. The fact is that the JEDEC data retention specification for NAND flash is 1 to 10 years. 10 years when the chips are new, and 1 year after they have been cycled 90% of their rated lifetimes. I posted some details of this specification some time ago in another thread.

There is no special exemption for the 25nm or smaller chips. To meet the JEDEC spec, they will also have a data retention of 1 to 10 years.

At some point, the chip fab wizards will run out of tricks and will no longer be able to produce smaller NAND flash that meets the JEDEC spec. But it has not happened yet. Anyone who says otherwise is just speculating.
Click to expand...

10 years at 10% used of the write cycles, 1 year at 100% written. Unfortunately the time at which it could do more than 1 year keeps dropping rapidly: http://en.community.dell.com/dell-b...idn-t-know-about-solid-state-disks-ssd-s.aspx

They got a really nice graph on that page with a handy explanation. One should also keep in mind that data retention is affected significantly by temperature. I.e. at 25C it could be 1 year, but at 50C it'd be shorted by a few months. That's another factor which makes it really hard to just blurt out numbers like a fixed number of years, as it's meaningless without a temperature qualifier.
 
Elledan said:
10 years at 10% used of the write cycles, 1 year at 100% written. Unfortunately the time at which it could do more than 1 year keeps dropping rapidly: http://en.community.dell.com/dell-b...idn-t-know-about-solid-state-disks-ssd-s.aspx

They got a really nice graph on that page with a handy explanation. One should also keep in mind that data retention is affected significantly by temperature. I.e. at 25C it could be 1 year, but at 50C it'd be shorted by a few months. That's another factor which makes it really hard to just blurt out numbers like a fixed number of years, as it's meaningless without a temperature qualifier.
Click to expand...

That graph is rather misleading, since IMFT has much better NAND flash chips than the ones shown in the graph. IMFT's 34nm MLC chips range from 5K to 30K erase cycles for endurance. The thing that the referenced article does not seem to take into account is that some fabs are better than others, and even the best get better with time. That graphs seems to plot some of the worst parts, and is far behind state of the art.

As for temperature, the JEDEC spec I was referring to:

JEDEC JESD47F
Data Retention for Non-Volatile Memory: High Temperature (HTDR)
JESD22-A117

assumes a use temperature of 55C. So that is 1 to 10 years of data retention at 55C. At lower temperatures, data retention would be even longer.
 
The JEDEC spec is nice, of course, but what is to keep a manufacturer from playing things a bit more loose with its cheaper NAND Flash chips? We have had reports of Kingston using sub-par Flash memory in its SD cards, so apparently there are enough managers out there who think that the average consumer won't know the difference between good and poor Flash memory, yet more than willing to pocket the difference.
 
If that is a concern, then it would be important to know which flash NAND chips are used in the SSD you are looking at. Micron (Crucial) and Intel SSDs use IMFT chips, and Micron has stated that those chips meet the JEDEC spec. It is probably more difficult to determine the flash NAND chips used in other SSDs.
 
IMO, the SSD drives we have are going to outlive the equivalent mechanical disks or at least have far less failures. Simply not having a weakness in shock damage is a major plus.
 
MaxBurn said:
IMO, the SSD drives we have are going to outlive the equivalent mechanical disks or at least have far less failures. Simply not having a weakness in shock damage is a major plus.
Click to expand...

What? You keep kicking your PCs? :)
 
MaxBurn said:
Think laptop, gone through about three hard drives at work due to shock damage in the last six years. Also people get a lot of bad drives after shipping.
Click to expand...

You'd have to drop a laptop badly enough to damage other parts of it as well to truly harm the HDD these days. I'm not sure the serious price premium of an SSD is worth the 1-2 laptop HDDs which might die over a few years time.

I have been building my own systems and other people's for quite a few years and never encountered a DOA HDD. Of course, improper packaging can cause this but companies in this country seem to be good at making stuff not die during transport :)

I'm still at zero dead HDDs among myself, friends and customers.
 
I have been building my own systems and other people's for quite a few years and never encountered a DOA HDD.
Click to expand...

For a while there with newegg every work hard drive order there was at least 1 DOA drive. I can not buy from newegg anymore for work purchases and I have not had a DOA drive since.
 
Elledan said:
I'm still at zero dead HDDs among myself, friends and customers.
Click to expand...

I can't believe your luck. I'm averaging about one a year across the twelve drives total I use. There is a google article somewhere detailing dead drives they replace in the vast quantities they use and I can't remember the number but it was an informative article on what you can really expect. My point is that replacing a mechanical HDD with a SSD is going to significantly raise the reliability.

Found the article in PDF:
http://static.googleusercontent.com...abs.google.com/en/us/papers/disk_failures.pdf
 
Elledan said:
I'm still at zero dead HDDs among myself, friends and customers.
Click to expand...

WOW. We have a 1% to 5% death rate per year at work. I have done 5 RMAs of less than 3 year old drives this year and had 3 other deaths (4 to 6 year old) and 1 more 2 year old drive will die soon (increasing reallocated sectors daily). Since I have between 100 and 200 drives it looks like this year we are going to be higher than 5% hard drive death rate..
 
william_fontaine said:
No way HDDs will be used in 20 years except for huge storage, and maybe not even that.
Click to expand...

drescherjm said:
I highly doubt that NAND chips made using silicon will be the bases of what ever storage technology that replaces hard drives.
Click to expand...
this^^
i didnt say we will be using regular hard drives for the next 50 years, i just said they will not be simply replaced by SSDs in their current form any time soon. the current tech has too many pitfalls to be a direct replacement.
 
So what about PCM? It provides much higher data retention that is not tied to endurance, and can theoretically be produced at smaller geometries than NAND, and is apparently a lot faster.

I haven't heard much about PCM since Micron bought up Numonyx and Samsung released their 512Mb MCP.... any news?
 
Elledan said:
You'd have to drop a laptop badly enough to damage other parts of it as well to truly harm the HDD these days. I'm not sure the serious price premium of an SSD is worth the 1-2 laptop HDDs which might die over a few years time.

I have been building my own systems and other people's for quite a few years and never encountered a DOA HDD. Of course, improper packaging can cause this but companies in this country seem to be good at making stuff not die during transport :)

I'm still at zero dead HDDs among myself, friends and customers.
Click to expand...

You are in the minority.

I'm positive towards the lifespan on HDD's, but even I know they can fail at any time for any stupid reason. Big spinning magnetic platters with a needle reading off of them, yikes. Just asking for trouble. The vast majority of HDD's will be just fine. I've never personally had any problems *HUGE knock on wood* but I've met many people who have.
 
Jon55 said:
You are in the minority.

I'm positive towards the lifespan on HDD's, but even I know they can fail at any time for any stupid reason. Big spinning magnetic platters with a needle reading off of them, yikes. Just asking for trouble. The vast majority of HDD's will be just fine. I've never personally had any problems *HUGE knock on wood* but I've met many people who have.
Click to expand...

Storing data by pushing some electrons on the floating gate of a transistor seems pretty silly to me too.

I think that one reason why I don't see many dead HDDs at least for myself is that I treat them with great care (including wearing an ESD wrist strap when handling them) and rarely make use of their start-up cycles, which is the only thing about a HDD which isn't theoretically unlimited due to wear on the bearings. My HDDs spin 24/7 except for the occasional reboot (every few months for my workstation, my servers stay online for years). HDDs love spinning, and hate starting up. It's why I don't make my HDDs go into standby all the time as well.

I still had the Seagate 20 MB HDD from my 386SX in working condition until a few years ago when I gave it away :)


On PCM, I absolutely agree that it'll most likely replace Flash memory. It is already in the process of replacing NOR Flash as it is far superior over this type (PCM is faster and more resilient, mostly). Once the storage density of PCM starts ramping up it'll be only a matter of time before NAND Flash is pushing daisies.
 
Yes and only going to get better from here on out. They will figure out whatever it takes to get the chips to last longer and longer.
 
You must log in or register to reply here.
Back
Top