SSD Data Retention

Denis247 wrote:

I've been reading about various methods of backups, and in particular about the use of SSD's.

There was mention that when powered down, data could start to be lost after only 3 momths or so.

Has anyone any experience of such data loss after extended periods of power-down in anything using SSD's?

This would probably have to exclude laptops as I'm sure provision is made to keep them supplied with a standby power, but what about removable SSD's?

If you look at the spec sheets for SSDs, you'll find something called "Data Retention". That specification is for how long memory cells can maintain their state without any power applied to the drive.

Virtually all current model MLC based SSDs are rated for 3 months data retention. SLC based drives (extremely expensive) may have longer data retention.

For example, see the spec sheet for this Samsung 840 Pro Drive, and you'll see 3 months shown in the warranty section on page 5:

http://www.samsung.com/us/pdf/memory-storage/840PRO_25_SATA_III_Spec.pdf

You'll see 3 months shown for Data Retention on page 3 of this specs sheet for Seagate SSDs, too:

http://www.seagate.com/www-content/product-content/ssd-fam/600-pro-ssd/en-us/docs/100727924b.pdf

Or, see the specs for this Intel Drive, and you'll see 3 months listed for data retention in Table 15. Note that it states once the drive has reached maximum endurance.

http://www.intel.com/content/dam/www/public/us/en/documents/product-specifications/ssd-dc-s3700-spec.pdf

So, how much the drive has been used also comes into the equation, and specs usually assume worst case for data retention (that cells have reached their rated P/E cycles before being removed from power).

If cells have never been written to before you back up something to it and put it into storage, then data retention may be years in a controlled environment (temperature can impact data retention, too). The more P/E (Program/Erase) cycles a memory cell has been through, the lower the data retention time

When powered up, the firmware monitors the state of cells and refreshes them as needed. But, without power, charge leakage from memory cells can lead to data loss in as little as 3 months for a drive that's been used for a while.

Jim Cockfield wrote:

Denis247 wrote:

I've been reading about various methods of backups, and in particular about the use of SSD's.

There was mention that when powered down, data could start to be lost after only 3 momths or so.

Has anyone any experience of such data loss after extended periods of power-down in anything using SSD's?

This would probably have to exclude laptops as I'm sure provision is made to keep them supplied with a standby power, but what about removable SSD's?

If you look at the spec sheets for SSDs, you'll find something called "Data Retention". That specification is for how long memory cells can maintain their state without any power applied to the drive.

Virtually all current model MLC based SSDs are rated for 3 months data retention. SLC based drives (extremely expensive) may have longer data retention.

For example, see the spec sheet for this Samsung 840 Pro Drive, and you'll see 3 months shown in the warranty section on page 5:

http://www.samsung.com/us/pdf/memory-storage/840PRO_25_SATA_III_Spec.pdf

You'll see 3 months shown for Data Retention on page 3 of this specs sheet for Seagate SSDs, too:

http://www.seagate.com/www-content/product-content/ssd-fam/600-pro-ssd/en-us/docs/100727924b.pdf

Or, see the specs for this Intel Drive, and you'll see 3 months listed for data retention in Table 15. Note that it states once the drive has reached maximum endurance.

http://www.intel.com/content/dam/www/public/us/en/documents/product-specifications/ssd-dc-s3700-spec.pdf

So, how much the drive has been used also comes into the equation, and specs usually assume worst case for data retention (that cells have reached their rated P/E cycles before being removed from power).

If cells have never been written to before you back up something to it and put it into storage, then data retention may be years in a controlled environment (temperature can impact data retention, too). The more P/E (Program/Erase) cycles a memory cell has been through, the lower the data retention time

When powered up, the firmware monitors the state of cells and refreshes them as needed. But, without power, charge leakage from memory cells can lead to data loss in as little as 3 months for a drive that's been used for a while.

jim, i used to work for intel as an engineer in flash r&d department. one of my sub- project was data  retention at room temp. the project went on for 10 years and there was no data loss, the project was still going on as far as P/E was concerned, 100K cycles was the spec, which was the minimum. flash memory has circuits in place to insure taking care of any charge loss. and as far as flash memory refresh is concerned, it doesn't need to be refreshed, i think you are mistaken flash memory for DRAM, which needs to be refreshed while power is on. just look at your flash thumb drives, have you ever lost data in one of those after leaving them out there without using them for a long time? i haven't experienced any data loss in my flash drives after a year or 2 not using them

cheeerz.

rebel99 wrote:

jim, i used to work for intel as an engineer in flash r&d department. one of my sub- project was data retention at room temp. the project went on for 10 years and there was no data loss, the project was still going on as far as P/E was concerned, 100K cycles was the spec, which was the minimum. flash memory has circuits in place to insure taking care of any charge loss. and as far as flash memory refresh is concerned, it doesn't need to be refreshed, i think you are mistaken flash memory for DRAM, which needs to be refreshed while power is on. just look at your flash thumb drives, have you ever lost data in one of those after leaving them out there without using them for a long time? i haven't experienced any data loss in my flash drives after a year or 2 not using them

You were likely working with SLC. Some years ago, 100,000 P/E cycles was the norm for SLC based NAND flash.

Virtually all consumer SSDs on the market today use MLC versus SLC NAND Flash. For a given die size, SLC memory is rated for approximately 10 times the P/E cycles compared to MLC, and data retention is also much longer.

But, SLC is cost prohibitive with today's higher capacity consumer drives. So, manufacturers now use MLC instead. Or, in the case of some of the entry level Samsung Drives, TLC is being used, which is only rated for 3,000 P/E cycles.

As die size has continued to shrink to allow denser designs with more storage in the same chip size, P/E cycles have continued to fall. 25nm MLC is rated at around 10,000 P/E cycles (as compared to approx. 100,000 P/E cycles for SLC memory using the same die size); and 25nm MLC is the type of memory you found in some of the consumer SSDs around 5 years ago.

But, in order to increase capacity, the industry has moved from 25nm to 22nm, etc., and we're seeing 15nm designs in some of current drives, with P/E ratings as low as 2,500 cycles for 15nm designs, with only a few electrons being stored to maintain the cell state.

Because the memory cells are becoming much smaller as time passes with shrinking die sizes, data retention is also decreasing, because you're going to have charge leakage after more P/E cycles break down the insulation, and with only a few electrons per cell being stored, it doesn't take as long for data loss.

With current MLC based drives, the industry standard is 3 months if power is not applied. Those are manufacturer's specifications (which is why I linked to the specs showing data retention from several current drive models in my first post to this thread, including an Enterprise Class Intel Drive), not something I'm pulling out of my rear. 

You'll also find a number of white papers discussing the data retention problem with today's newer MLC based drives, as die sizes continue to shrink over time.

Now, you may see years of data retention for a drive that hasn't seen a lot of P/E cycles, as those specs assume worst case data retention. But, the memory cells we're seeing in today's drives are much smaller than what you were probably working with, and manufacturers have switched from using SLC to MLC based NAND flash in order to keep cost down for typical consumer workloads. Because very few electrons are stored per cell, with insulation breaking down due to the erase portion of the cycles to reuse a cell, data retention time has become lower and lower with each new generation of MLC memory used in SSDs.

The most interesting new technology is 3D NAND (a.k.a., V-NAND), and Samsung just announced an 850 Pro Line of SSDs using this newer design. They're the first manufacturer to launch products based on that technology, but other manufacturers like Intel, Micron and Toshiba are likely to move to the same type of NAND Flash soon, because it's just not practical to continue using the older designs due to reliability issues with shrinking die sizes in order to allow higher capacity SSDs at an affordable price point.

You'll find a number of new reviews of the new Samsung 850 Pro line (as the drive was just launched this month), and the new type of NAND Flash being used can handle more P/E cycles compared to MLC or TLC based drives.

It looks like anandtech has the new V-NAND memory used by the 850 Pro estimated at 6,000 P/E cycles (double what most other current SSDs based on the latest MLC memory can withstand).

http://www.anandtech.com/show/8239/update-on-samsung-850-pro-endurance-vnand-die-size

Here's a page from an 850 Pro review explaining why the older MLC and TLC memory types are no longer suitable due to reliability issues because of shrinking die sizes in order to produce higher capacity drives at an affordable level, since you may only have around 3 electrons per cell being stored to maintain the cell state with some of the current consumer SSDs.

http://www.anandtech.com/show/8216/samsung-ssd-850-pro-128gb-256gb-1tb-review-enter-the-3d-era/2

What happens with for example a laptop that comes with SSD standard but is not sold immediately? let's say it stays in stock somewhere longer than that.

exumaparadise wrote:

What happens with for example a laptop that comes with SSD standard but is not sold immediately? let's say it stays in stock somewhere longer than that.

A new laptop using an SSD should not have any issues with data loss for a long period of time (probably years) if sitting on a shelf without any power applied, because there have been no P/E (program/erase) cycles to break down the insulation within memory cells like you'd have with a drive being used more.

Basically, they're just going to load a disk image file to the SSD in that laptop, without any activity that uses erase cycles for reusing cells over and over as time passes, so you can expect that loaded disk image on the SSD to be fine for a long time to come.

The problem comes in if the drive is under heavy use, where the erase portion of the P/E cycles allows break down of the insulation within the memory cells over time, making it easier for charge loss to occur (since only a few electrons per cell are being stored in today's newer SSD designs using MLC or TLC based cells)

So, I wouldn't worry about that kind of thing if buying a new laptop that's been sitting on the shelf for a while. The problem with data retention comes in with an SSD that's been used for a while, with more P/E cycles as cells have been updated due to OS upgrades, data updates, etc., then has power removed from it over an extended period, as newer firmware on the SSDs controller monitors the cell state and refreshes cells (or moves the data to spare cells) from time to time that have errors detected, as long as power is applied.

Of course, the more free space on a drive, the fewer P/E cycles each cell will incur. So, it's a good idea to buy a larger drive that needed and don't fill it up all of the way, since sophisticated wear leveling algorithms in today's SSDs will try to spread updated data more evenly so that each cell has viewer P/E cycles (whereas if you keep an SSD almost full, you'll have more P/E cycles per cell over time).

Newer SSD designs also set aside memory cells as spares, and the more spares allocated, the longer lifespan you can expect. With my Samsung 830 series SSDs, I increased the amount of spares for that purpose (to help get longer life from them), as their software allows you to tweak that kind of thing.

Most consumer SSDs should last years with typical use patterns, as long as they're not removed from power for an extended period of time (think several months) after heavy use. With a drive that has not been used much (as in a new laptop that's been sitting on a retailer's shelf for a while with virtually no P/E cycles on memory cells), you should not see any issues.

The 3 month Data Retention number you'll find in the specifications for current SSDs using MLC memory is considered to be a worst case spec, where memory cells have incurred a lot of P/E cycles over time.

But with a a new laptop with an SSD, there have been no P/E (Program/Erase) cycles to update data like you'd have with typical usage patterns, and the erase portion is what breaks down the insulation within cells over time that can lead to premature charge loss, as a cell must be erased before new/updated data is written to it.

Since the laptop manufacturer just writes a disk image to the drive (meaning no erase cycles to write updated data to cells has occurred) it would probably be fine for years without any power being applied.

Jim Cockfield wrote:

rebel99 wrote:

jim, i used to work for intel as an engineer in flash r&d department. one of my sub- project was data retention at room temp. the project went on for 10 years and there was no data loss, the project was still going on as far as P/E was concerned, 100K cycles was the spec, which was the minimum. flash memory has circuits in place to insure taking care of any charge loss. and as far as flash memory refresh is concerned, it doesn't need to be refreshed, i think you are mistaken flash memory for DRAM, which needs to be refreshed while power is on. just look at your flash thumb drives, have you ever lost data in one of those after leaving them out there without using them for a long time? i haven't experienced any data loss in my flash drives after a year or 2 not using them

You were likely working with SLC. Some years ago, 100,000 P/E cycles was the norm for SLC based NAND flash.

Virtually all consumer SSDs on the market today use MLC versus SLC NAND Flash. For a given die size, SLC memory is rated for approximately 10 times the P/E cycles compared to MLC, and data retention is also much longer.

But, SLC is cost prohibitive with today's higher capacity consumer drives. So, manufacturers now use MLC instead. Or, in the case of some of the entry level Samsung Drives, TLC is being used, which is only rated for 3,000 P/E cycles.

As die size has continued to shrink to allow denser designs with more storage in the same chip size, P/E cycles have continued to fall. 25nm MLC is rated at around 10,000 P/E cycles (as compared to approx. 100,000 P/E cycles for SLC memory using the same die size); and 25nm MLC is the type of memory you found in some of the consumer SSDs around 5 years ago.

But, in order to increase capacity, the industry has moved from 25nm to 22nm, etc., and we're seeing 15nm designs in some of current drives, with P/E ratings as low as 2,500 cycles for 15nm designs, with only a few electrons being stored to maintain the cell state.

Because the memory cells are becoming much smaller as time passes with shrinking die sizes, data retention is also decreasing, because you're going to have charge leakage after more P/E cycles break down the insulation, and with only a few electrons per cell being stored, it doesn't take as long for data loss.

With current MLC based drives, the industry standard is 3 months if power is not applied. Those are manufacturer's specifications (which is why I linked to the specs showing data retention from several current drive models in my first post to this thread, including an Enterprise Class Intel Drive), not something I'm pulling out of my rear.

You'll also find a number of white papers discussing the data retention problem with today's newer MLC based drives, as die sizes continue to shrink over time.

Now, you may see years of data retention for a drive that hasn't seen a lot of P/E cycles, as those specs assume worst case data retention. But, the memory cells we're seeing in today's drives are much smaller than what you were probably working with, and manufacturers have switched from using SLC to MLC based NAND flash in order to keep cost down for typical consumer workloads. Because very few electrons are stored per cell, with insulation breaking down due to the erase portion of the cycles to reuse a cell, data retention time has become lower and lower with each new generation of MLC memory used in SSDs.

The most interesting new technology is 3D NAND (a.k.a., V-NAND), and Samsung just announced an 850 Pro Line of SSDs using this newer design. They're the first manufacturer to launch products based on that technology, but other manufacturers like Intel, Micron and Toshiba are likely to move to the same type of NAND Flash soon, because it's just not practical to continue using the older designs due to reliability issues with shrinking die sizes in order to allow higher capacity SSDs at an affordable price point.

You'll find a number of new reviews of the new Samsung 850 Pro line (as the drive was just launched this month), and the new type of NAND Flash being used can handle more P/E cycles compared to MLC or TLC based drives.

It looks like anandtech has the new V-NAND memory used by the 850 Pro estimated at 6,000 P/E cycles (double what most other current SSDs based on the latest MLC memory can withstand).

http://www.anandtech.com/show/8239/update-on-samsung-850-pro-endurance-vnand-die-size

Here's a page from an 850 Pro review explaining why the older MLC and TLC memory types are no longer suitable due to reliability issues because of shrinking die sizes in order to produce higher capacity drives at an affordable level, since you may only have around 3 electrons per cell being stored to maintain the cell state with some of the current consumer SSDs.

http://www.anandtech.com/show/8216/samsung-ssd-850-pro-128gb-256gb-1tb-review-enter-the-3d-era/2

jim,

your info is misleading and incorrect. when i was working for intel, MLC was storing 2 bit in the same transistor, now it is storing 3 bit in the same transistor by having 3 floating gates and applying different voltage to each gate activates flow of currents through PN material, thus generating 3 bits. intel used to manufacture NOR falsh about 7 years ago, which was being used specifically on cellphones because it had ability of fast "read" feature. on other hand, NAND flash was moving ahead due to capability of mass data storing "write" characteristic, which was needed for music, movies, and so forth devices. as cellphones became "smart", so did capabilities of storing lots of data. so, the outcome was obvious and intel joined the NAND camp as well. intel always had stringent quality control that the rest of the industry could not afford. that is why intel SSDs are still so expensive! however, i am not sure if intel still follows the same quality control as they used to. as i mentioned in my original massage, NAND flash memories have capability of a long R/W/E (read/write/erase) cycles and a 100K+ wouldn't be surprising. if you can, don't turn on your SSD "C" drive for a few years, and then turn it on again and you'll find out that all your data will still be there. you can do the same experiment with small flash drive and you'll get the same result. good designed flash ICs have redundancies that if one transistor fails, failed address will be routed to the redundancy address without rendering the whole IC useless.

as far as refreshing flash memory is concerned, obviously you have no clue how flash works. let me give you one, today, there is not many gadget or vehicle, or airplanes (you name it) that doesn't use flash memory and the reason why this ubiquitous memory is being used is because it has capability that it doesn't need to be refreshed or lose data while not being used. and as far as retention is concerned, if even one of the address fails in your IC, the whole chip will be rendered useless and will not work again. i looked at the intel .PDF you had posted and there was no mention of "RETENTION" you are talking about in there, or, may be i missed it hope this will give you some basic idea of what is going on in the world of flash memory! there is no need for you to disseminate this "16 day retention" rumor (BS) around and cause a false alarm!

cheerz.

rebel99 wrote:

jim,

your info is misleading and incorrect. when i was working for intel, MLC was storing 2 bit in the same transistor, now it is storing 3 bit in the same transistor by having 3 floating gates and applying different voltage to each gate activates flow of currents through PN material, thus generating 3 bits. intel used to manufacture NOR falsh about 7 years ago, which was being used specifically on cellphones because it had ability of fast "read" feature. on other hand, NAND flash was moving ahead due to capability of mass data storing "write" characteristic, which was needed for music, movies, and so forth devices. as cellphones became "smart", so did capabilities of storing lots of data. so, the outcome was obvious and intel joined the NAND camp as well. intel always had stringent quality control that the rest of the industry could not afford. that is why intel SSDs are still so expensive! however, i am not sure if intel still follows the same quality control as they used to. as i mentioned in my original massage, NAND flash memories have capability of a long R/W/E (read/write/erase) cycles and a 100K+ wouldn't be surprising.

It's very well known that 25nm MLC using 2 bits per cell is rated at 10,000 (not 100,000) P/E cycles, as I've mentioned in previous posts.

You'll find *MANY* articles, specs and white papers on the subject.

Heck, here's one from Intel I found with a quick google search. Note that SLC is rated at 100,000 P/E Cycles and MLC using 2 bits per cell is rated at 10,000 P/E cycles. As I previously mentioned, SLC memory is good for around 10 times the P/E cycles compared to SLC at the same die size.

http://download.intel.com/pressroom/kits/vssdrives/Nand_PB.pdf

But, with today's denser 20nm and 15nm MLC, P/E cycles are even lower (down to around 2,500 to 3,000 P/E cycles for the 15nm MLC memory used in the latest entry level consumer SSD models now).

There are *many* tech papers and articles on the subject. If you've kept up with trends in MLC since leaving Intel, you'd see that kind of thing fairly often in articles discussing each newer generation of SSDs.

if you can, don't turn on your SSD "C" drive for a few years, and then turn it on again and you'll find out that all your data will still be there. you can do the same experiment with small flash drive and you'll get the same result. good designed flash ICs have redundancies that if one transistor fails, failed address will be routed to the redundancy address without rendering the whole IC useless.

Actually, I left an EeePC with an SSD powered off for some months, and the drive was "trashed" when I attempted to power it back up again, with half of the memory cells totally useless (even after attempts to zero fill it). So, I've seen that kind of thing myself with SSDs using cheaper MLC NAND flash. I should have known better than to use a journalling file system with it (ext3 is my case), as the error correction, sparing and wear leveling algorithms were not very good with some of the early consumer SSD models, and I should have taken steps to reduce the amount of write activity and P/E cycles needed when using it.

But, again, the number of P/E cycles impact data retention, too. You'll find many white papers and technical articles explaining why that's the case, including docs from major NAND flash makers. I even pointed to one page about how more P/E cycles break down resistance within cells leading to charge leakage in my previous posts to this thread.

as far as refreshing flash memory is concerned, obviously you have no clue how flash works. let me give you one, today, there is not many gadget or vehicle, or airplanes (you name it) that doesn't use flash memory and the reason why this ubiquitous memory is being used is because it has capability that it doesn't need to be refreshed or lose data while not being used. and as far as retention is concerned, if even one of the address fails in your IC, the whole chip will be rendered useless and will not work again. i looked at the intel .PDF you had posted and there was no mention of "RETENTION" you are talking about in there, or, may be i missed it

Try page 13 (second part of table 15, where I mentioned that info would be) in the specs I linked to for an Intel Enterprise Quality drive using 25nm MLC. There is a Data Retention spec shown in that page. Here's a quote from it:

"3 months power-off retention once SSDreaches rated write endurance at 40 °C"

As mentioned, that's even an Enterprise level drive, and it's using 25nm MLC memory. The latest consumer drives tend to use 20nm or even 15nm MLC anymore, with far fewer P/E Cycles available (typically around 3,000 with the latest MLC NAND memory, versus around 10,000 with 25nm MLC).

You'll see the same kind of thing in the other drive specs I linked to from Samsung and Seagate in my first post to this thread (3 months data retention, as that's the current industry standard for the latest drives once they've reached their write endurance, and around 3,000 P/E cycles is typical for the latest consumer SSD models using 15nm to 20nm MLC.

Again, as previously mentioned, data may last for years if a drive has not incurred a higher number of P/E cycles. But, the industry standard for a drive using MLC memory that has reached it's rated P/E cycles is currently 3 months for Data Retention, as you'll find in the specifications from drives made by a number of manufacturers (and I linked to specs from newer Samsung, Seagate and Intel SSDs showing 3 months data retention in my first post to this thread).

Like it or not, manufacturers can only do so much to maintain reliability as the need for higher density memory marches on (hence why we've been seeing shrinking die sizes, where we had 25nm memory as commonplace not long ago, and now we're down to 15nm MLC in some of the newest drives, to try and market higher capacity drives to consumer at a lower price point.

The most promising new technology for now is the latest 3D NAND (or what Samsung refers to as V-NAND), which allows around 6,000 P/E cycles per cell according to estimates made by review sites, and allows for much higher capacity in smaller chip sizes.

See the link to a technical discussion about why we've reached the limits of current MLC based NAND technology in my earlier posts that's included in a review of the latest Samsung 850 Pro. This page:

http://www.anandtech.com/show/8216/samsung-ssd-850-pro-128gb-256gb-1tb-review-enter-the-3d-era/2

The following page explains how the new 3D NAND (a.k.a., V-NAND as Samsung calls it) can help get around the current limitations, since consumers are constantly wanting more storage at a lower price point:

http://www.anandtech.com/show/8216/samsung-ssd-850-pro-128gb-256gb-1tb-review-enter-the-3d-era/3

Jim Cockfield wrote:

As die size has continued to shrink to allow denser designs with more storage in the same chip size, P/E cycles have continued to fall. 25nm MLC is rated at around 10,000 P/E cycles (as compared to approx. 100,000 P/E cycles for SLC memory using the same die size); and 25nm MLC is the type of memory you found in some of the consumer SSDs around 5 years ago.

Here's a page from an 850 Pro review explaining why the older MLC and TLC memory types are no longer suitable due to reliability issues because of shrinking die sizes in order to produce higher capacity drives at an affordable level, since you may only have around 3 electrons per cell being stored to maintain the cell state with some of the current consumer SSDs.

I think you are using the term "die size" when you really mean to say "geometry size".  The die is the actual working silicon area on the wafer, which is then cut, and bonded into a chip package.  So geometry size shrinking is what allows denser designs on the die, and what degrades reliability.  The density is independent of the die size, you can have the same density on a large die or small die.

dwalby wrote:

Jim Cockfield wrote:

As die size has continued to shrink to allow denser designs with more storage in the same chip size, P/E cycles have continued to fall. 25nm MLC is rated at around 10,000 P/E cycles (as compared to approx. 100,000 P/E cycles for SLC memory using the same die size); and 25nm MLC is the type of memory you found in some of the consumer SSDs around 5 years ago.

Here's a page from an 850 Pro review explaining why the older MLC and TLC memory types are no longer suitable due to reliability issues because of shrinking die sizes in order to produce higher capacity drives at an affordable level, since you may only have around 3 electrons per cell being stored to maintain the cell state with some of the current consumer SSDs.

I think you are using the term "die size" when you really mean to say "geometry size". The die is the actual working silicon area on the wafer, which is then cut, and bonded into a chip package. So geometry size shrinking is what allows denser designs on the die, and what degrades reliability. The density is independent of the die size, you can have the same density on a large die or small die.

True... but the point is that manufacturers are trying hard to increase storage capacity within the same wafer size by using a smaller "geometry" (as you put it), going from 25nm MLC memory a few years back, down to 15nm to 20nm MLC now within the same chip size.    But, the downside of that technique is fewer P/E cycles for the latest [smaller] MLC memory cells.

In typical use scenarios, that's still not a huge problem, as with typical consumer workloads, an SSD can last for years.  But, data retention time is also impacted [negatively], as this thread is discussing.

Denis247 wrote:

Are they the same components in USB drives?, ie are the arguments re SSD's also applicable to USB drives ?

And would you recommend not using SSD's and/or USB drives as long-term backup storage devices?

What are you trying to backup?

Some of the higher end memory cards and flash drives from manufacturers like Sandisk are using SLC versus MLC memory cells in order to get the fastest write speeds and greatest reliability. That's one reason the higher end models (Sandisk Extreme Pro and similar using SLC memory) tend to cost a lot more.

Heck, Sandisk even markets what they call a "Memory Vault" for long term storage of images. It's a USB attached "write once" device using larger SLC memory cells, with no ability to erase/reuse any of it's storage space, that is rated at up to 100 years storage. It's write once, because once you start erasing and reusing cells, data retention time is negatively impacted. But, it's fairly expensive for the amount of storage you get with it (only available in up to 16GB size, and again, it's a write once device).

As pointed out, the more P/E cycles (Program/Erase cycles) a memory cell has endured, the lower the data retention rate, especially when referring to today's shrinking MLC memory. But, if you're taking a brand new memory card, flash memory drive, etc., for writing backup files to, you can expect years of data retention from what I've seen mentioned in white papers for 25nm MLC based devices using higher quality NAND Flash cells, or much longer for SLC memory based devices (a lot more costly compared to MLC).

But, if you're using a card (or flash drive) for a while, expect much shorter data retention times (as low as 3 months once memory cells get nearer to their rated P/E cycles if it's using MLC versus SLC memory.

I'd give more info on what you want to store and how much data you want to store for more informed responses. Personally, for larger amounts of data, I use traditional physical hard disks drives (magnetic storage), making sure to have more than one drive with the same data stored.

Then, just copy the data to newer drives every few years or so. That's a more cost effective way to approach it, as other media types (SSDs, Flash Drives, etc.) tend to have a dramatically higher price/GB for storage space, and you can see much lower data retention times if flash media has been reused often (erasing current data, and updating it with new data).

Jim Cockfield wrote:

dwalby wrote:

Jim Cockfield wrote:

As die size has continued to shrink to allow denser designs with more storage in the same chip size, P/E cycles have continued to fall. 25nm MLC is rated at around 10,000 P/E cycles (as compared to approx. 100,000 P/E cycles for SLC memory using the same die size); and 25nm MLC is the type of memory you found in some of the consumer SSDs around 5 years ago.

Here's a page from an 850 Pro review explaining why the older MLC and TLC memory types are no longer suitable due to reliability issues because of shrinking die sizes in order to produce higher capacity drives at an affordable level, since you may only have around 3 electrons per cell being stored to maintain the cell state with some of the current consumer SSDs.

I think you are using the term "die size" when you really mean to say "geometry size". The die is the actual working silicon area on the wafer, which is then cut, and bonded into a chip package. So geometry size shrinking is what allows denser designs on the die, and what degrades reliability. The density is independent of the die size, you can have the same density on a large die or small die.

True... but the point is that manufacturers are trying hard to increase storage capacity within the same wafer size by using a smaller "geometry" (as you put it), going from 25nm MLC memory a few years back, down to 15nm to 20nm MLC now within the same chip size. But, the downside of that technique is fewer P/E cycles for the latest [smaller] MLC memory cells.

agreed, I wasn't challenging your main premise, I was simply pointing out that the phrase "die size has continued to shrink" is incorrect terminology for what you're describing.  The correct terminology is "device geometry has continued to shrink".  25/20/15nm is a measure of device geometry.  To increase storage capacity, the die size would typically stay the same, while shrinking the device geometry to fit more logic gates into the same die area.

rebel99 wrote:

Jim Cockfield wrote:

rebel99 wrote:

jim, i used to work for intel as an engineer in flash r&d department.

You were likely working with SLC. Some years ago, 100,000 P/E cycles was the norm for SLC based NAND flash.

as far as refreshing flash memory is concerned, obviously you have no clue how flash works.

Actually, Jim is correct.   With the push to shrink process sizes and cram more and more cells onto a die, the bits are being retained by fewer and fewer electrons.   Although flash memory is designed to be non-volatile, in actual fact the static charge that stores data bits does leak out of the cells over time, and the smaller the cells are and the more "worn" they are (due to lots of P/E cycles), the faster that occurs.   The 3-month figure is for the worst-case scenario where the cells have been programmed and erased the maximum number of times.

To mitigate risk, flash memory controllers use ECC codes just like disk drives do.   And just like disk drives, if the controller detects correctable ECC errors for a block of data it will take the corrected data and re-write it back to a flash memory page again.   Note that a page that starts to loose data because of static dissipation is not necessarily a damaged page - static dissipation is a normal, expected behaviour given a long enough time.   So the data can often be written right back to the very same page, recharging each cell with a full static charge (or whatever charge the cells can accept if they've been "worn out" because of a lot of P/E cycles).

This is the kind of "refresh" that's being discussed for flash memory, not the type that's used with DRAM.   The two are completely different animals.  DRAM refresh is there by necessity and is an essential part of the way DRAM works even when there are no memory errors (it also occurs hundreds to thousands of times per second).   The way that flash drives "refresh" data from degraded memory pages by writing it back again is a data reliability technique that's only needed when the device starts to loose the ability to correctly read the data.

This is why the 3-month limit is specified for "powered off" drives.   When the drive is powered on, the controller periodically scans all of the blocks to verify the ECC codes.   Any suspect blocks are re-written again.   So if the drive is powered on the 3-month limit isn't a serious concern.

Jim Cockfield wrote:

Most consumer SSDs should last years with typical use patterns, as long as they're not removed from power for an extended period of time (think several months) after heavy use. With a drive that has not been used much (as in a new laptop that's been sitting on a retailer's shelf for a while with virtually no P/E cycles on memory cells), you should not see any issues.

The 3 month Data Retention number you'll find in the specifications for current SSDs using MLC memory is considered to be a worst case spec, where memory cells have incurred a lot of P/E cycles over time.

I'm going to avoid the silly cock fight here, but I have to point out that you contradict your dire warning with the above.    My desktop drive (a 500gb 840 Pro) has been in use for nearly a year and I still haven't make it to 1TB in writes.   A device used for weekly backups won't see a huge number of cycles either.    Since we're already established it takes 15 years at a fairly aggressive write pace to hit a threshold, I don't see the threat for a few years of backup duty.   The 5 year service life standard should apply.   And if he's going to update it far less often, how is his backup any different at all from a system install on a laptop that sits in the warehouse waiting to be sold?

As the other person noted, people store info on USB sticks all the time and leave it alone.   And operating systems don't spend their time rewriting blocks of storage media - that would in fact exhaust them quickly.

kelpdiver wrote:

Jim Cockfield wrote:

Most consumer SSDs should last years with typical use patterns, as long as they're not removed from power for an extended period of time (think several months) after heavy use. With a drive that has not been used much (as in a new laptop that's been sitting on a retailer's shelf for a while with virtually no P/E cycles on memory cells), you should not see any issues.

The 3 month Data Retention number you'll find in the specifications for current SSDs using MLC memory is considered to be a worst case spec, where memory cells have incurred a lot of P/E cycles over time.

I'm going to avoid the silly cock fight here, but I have to point out that you contradict your dire warning with the above. My desktop drive (a 500gb 840 Pro) has been in use for nearly a year and I still haven't make it to 1TB in writes. A device used for weekly backups won't see a huge number of cycles either. Since we're already established it takes 15 years at a fairly aggressive write pace to hit a threshold, I don't see the threat for a few years of backup duty. The 5 year service life standard should apply. And if he's going to update it far less often, how is his backup any different at all from a system install on a laptop that sits in the warehouse waiting to be sold?

As for the laptop sitting on a shelf scenario, the manufacturer is just writing a disk image to a new drive. So, it hasn't gone through any additional p/e cycles to reuse the cells that can break down the insulation and allow more charge leakage. So, sure, I'd expect it to be fine for years without any issues.

We're now seeing memory cell sizes as small as 15nm, where only a few electrons leaking can result in read errors. Heck, the TLC memory used in the latest Samsung EVO models only store 3 electrons per cell, so I'd expect charge leakage to be an even bigger problem with them.

Here's a good discussion of how charge leakage can be a problem because of the smaller and smaller cell sizes being used by the latest consumer SSD models. It's from the new Samsung 850 Pro review and discusses why we're already at the practical limit of reducing cell sizes with latest SSDs, which is why manufacturers will be moving to the use of newer 3D NAND (or V-NAND as Samsung calls it). I linked to it earlier during this discussion. But, here it is again:

http://www.anandtech.com/show/8216/samsung-ssd-850-pro-128gb-256gb-1tb-review-enter-the-3d-era/2

"The way NAND is programmed and erased is also its Achilles' Heel. Because such high voltage is needed, the insulators around the floating gate (i.e. ONO and tunnel oxide) wear out as the NAND goes through program and erase cycles. The wear out causes the insulators to lose their insulating characters, meaning that electrons may now be able to escape the floating or get trapped in tunnel oxide during a program or erase. This causes a change in the voltage state of the cell."

Again, that page goes into a lot more detail about why the current cell sizes present a problem with charge leakage (since very few electrons are being stored in the latest smaller memory cells),

As for your comments on the post you quoted, what Contradiction?

From my very first post, I've explained that 3 months data retention when powered off is a worst case specification for a drive, assuming it's reached it's rated P/E cycle count. I've said the same thing in other posts to this thread, too.

Here's a quote from my first post to this thread, after pointing to the specifications from several different drive models that had Data Retention shown as "3 Months Typical without Power":

http://www.dpreview.com/forums/post/54040003

"So, how much the drive has been used also comes into the equation, and specs usually assume worst case for data retention (that cells have reached their rated P/E cycles before being removed from power).

If cells have never been written to before you back up something to it and put it into storage, then data retention may be years in a controlled environment (temperature can impact data retention, too). The more P/E (Program/Erase) cycles a memory cell has been through, the lower the data retention time

When powered up, the firmware monitors the state of cells and refreshes them as needed. But, without power, charge leakage from memory cells can lead to data loss in as little as 3 months for a drive that's been used for a while."

As the other person noted, people store info on USB sticks all the time and leave it alone.

So do I. I keep linux distros on USB Flash Drives that I use for various purposes. I've also seen errors requiring me to reinstall those distros to Flash Drives in the past, too. I've had a couple I've needed to do that with lately.

The memory cell size and memory cell type (SLC versus MLC) also comes into the equation, with higher quality flash drives using SLC, and lower quality flash drives using MLC.

And operating systems don't spend their time rewriting blocks of storage media - that would in fact exhaust them quickly.

The Operating System doesn't do it. The SSD Firmware performs memory scrubbing when it's powered up, refreshing cells as needed. Just look at articles about the controllers used in modern SSDs and you'll find memory scrubbing as one of their characteristics.

Given that you may have very few electrons being stored to each memory cell as manufacturers reduce cell size to allow higher capacity at a lower cost, a few lost electrons may result in read errors trying to determine the state of a given cell.

Here's a quote from the spec sheet from a Seagate 600 Pro SSD (marketed for Enterprise versus consumer use). Note that a brand new drive may store data for up to 10 years in this model's case (or course, that's "up to" and assumes there have not been any P/E cycles to break down the insulation within memory cells). Newer consumer models are using even smaller memory cells than this model, with fewer electrons being stored per cell. Bold added for emphasis, since we're discussing why the specs show such a low "typical" (typical is the word used in some of the drive specs) data retention time when a drive is removed from power).

" As NAND Flash devices age with use, the capability of the media to retain a programmed value begins to deteriorate. This deterioration is affected by the number of times a particular memory cell is programmed and subsequently erased. When a device is new, it has a powered off data retention capability of up to ten years. With use the retention capability of the device is reduced. Temperature also has an effect on how long a Flash component can retain its programmed value with power removed. At high temperature the retention capabilities of the device are reduced. Data retention is not an issue with power applied to the SSD. The SSD drive contains firmware and hardware features that can monitor and refresh memory cells when power is applied."

http://www.seagate.com/www-content/product-content/ssd-fam/600-pro-ssd/en-us/docs/100727924b.pdf

Again, bold added for emphasis, as memory scrubbing is a feature found in modern controllers used by SSDs.

I don't know how many times I have to say that the "3 Months typical" specification without power applied is the worst case specification for a drive that's reached it's rated P/E cycles; as I've repeated that multiple times in this thread.

Denis247 wrote:

Thanks Jim, and thanks everyone for such expert responses.

Jim, I use removeable hard drives, spinners, stored off site, and I would add that it's only personal data, photo's, music etc. I use 2, and backups are taken quite regularly. So I'm quite confident in my own backup system.

But another member of my family, with not so many photo's etc needed to be backed-up, uses USB sticks. So when I saw the article I referred to in my original post, I thought I'd better find out a bit more about their reliability data-retention wise. I was also asking about using SSD's just out of curiosity.

Although there have been no issues with data loss so far, on the USB sticks, the answers here confirm that using them will be ok with perhaps the proviso that good quality and relatively new sticks should be used, and that backups are taken regularly, and to use at least 2 seperate sticks.

Small number of photos, huh?

Sandisk released some interesting products a few years back. One was called the "WORM (Write Once Read Many) Secure Digital Card", and the other was called the "Memory Vault"

Both are write once devices (you can't erase anything you write to them or overwrite the data), and Sandisk claimed up to 100 years data retention. Here's a page that still works on Sandisk's site about the Memory Vault Product:

http://www.sandisk.com/go/preserve/

But, links from that page are broken, since it's a discontinued product now.

The Memory Vault was relatively expensive when launched in 2011. For example, the 8GB Memory Vault listed for $49.99, and the 16GB Memory Vault listed for $89.99.

Sandisk has never released many details on the type of technology used to achieve that data retention time (other than to say they're using proprietary ASICs, etc.).

I had speculated that they might have been using higher grade SLC memory with larger cell sizes and better insulation properties to avoid charge leakage, and only allowing one write to to each cell via the controller firmware to achieve the longer retention time.

But, doing some more digging around, some of the engineering sites have speculated that they used a new type of non-volatile OTP (One Time Programmable) memory developed by Matrix Semiconductor, which Sandisk acquired some years back. See this article on the subject:

http://www.engineering.com/ElectronicsDesign/ElectronicsDesignArticles/ArticleID/4477/Safeguarding-digital-archives-Part-1--SanDisk-Memory-Vault.aspx

You can still find them in stock at some vendors (even though the Memory Vault has been discontinued for a while), at much lower price points from vendors that still have any remaining stock of them. For example, I see Adorama has a listing for the 16GB model for only $24.95 now (when it listed for $89.99 when released):

http://www.adorama.com/IDSMV16GB.html

Some of the merchants listing products on Amazon have them at pretty good prices, too. Since it's been discontinued for a while, Amazon doesn't have any in stock directly from Amazon. But, some of the merchants listing items on amazon.com still have them (including Adorama). Here are some search results for Sandisk Memory Vault:

http://www.amazon.com/s/ref=nb_sb_noss_1?url=search-alias%3Delectronics&field-keywords=sandisk%20memory%20vault

So, that may be something to consider for a small number of images, given Sandisk's claims about data retention times.

Of course, why they were discontinued is a question in my mind (unless they just were not selling well due to the very high price point as compared to other types of storage media, versus reliability issues found later).

In any event, they're available at much lower prices now from some vendors that still have them in stock).

But, I'd stick with a reputable vendor like Adorama (just in case they've been counterfeited, as is common with Sandisk Products).