How bad will next gen NAND P/E cycle count will be?

[taltamir]

http://www.anandtech.com/show/4284/sandisktoshiba-take-back-the-crown-with-a-different-kind-of-nand

Like IMFT, they expect to see similar endurance as their 24nm products, around 3,000 program/erase cycles. This is owing to ever improving ECC and wear leveling algorithms, ensuring as few wasted p/e cycles as possible

By definition, wear leveling and ECC has no effect on P/E cycle count. It has an effect on the SSD's lifespan, which is determined by P/E cycles * size / write amplification. Wear leveling algorithms lower the write amplification, thus increasing lifespan on a same P/E cycle count NAND.
The thing is, wear leveling is already very very low... And ECC compensates for lowered reliability (an increase in bit errors during writing, again, due to the lowering quality of NAND as it shrinks; requires sacrificing space to compensate for quality loss).

I don't know why anandtech didn't point out that this is bad spin, but I am concerned about the issue

 

[bulanula]

Finally somebody that is smart enough to realize that this technology is dead in its tracks and is not going anywhere soon ( except to the trash maybe ).

You just cannot scale this technology any further. Graphene or another next-gen material is the answer.

Soon enough you will see 1000 cycles NAND. What a total complete joke. There is only so much lifespan that the controller can save up before the reality that NAND has too few cycles begins to set in and people see their drives dying and becoming read only in a couple of months.

Interested to see how 1000 cycles 20nm NAND will cope. It's basic physics. They claim that 20nm NAND will be the same as 25nm NAND. Only a clever marketing spin, I believe.

50nm NAND > 34nm NAND > 25nm NAND > 20nm NAND
SLC > MLC

It's just the way it is.

Take the Intel x25-e as an example.

50nm SLC NAND >>> 25nm MLC NAND in the Intel 320 series junk disposable SSD.

 

[Ancalagon44]

Its one of few examples of technology in which an improvement introduces an unavoidable negative side effect.

 

[taltamir]

i am not going so far to so it would be "trash". I am expressing my disbelief in their claims and curiousity as to how bad it will be. So far all we have to go by is:
1. previous generations have been progressively worse.
2. The spin machine is already in full swing. trying to sell us on blatant lies (same P/E count via ECC and wear leveling).

even if we maxed out the shrinkage and sub 25nm never makes it to market you can still develop manufacturing to reduce costs and stack more die together for quite a lot of time... Speed and size can be vastly increased without NAND improvement. And this could continue until one of the many alternative SSD technologies under development matures.

 

[imaheadcase]

I see the future without hardrives in general, just upgradeable like ram on motherboard. You can fit lots of SSD chips on motherboards now. Soundcards went that way, LAN went that way, its a natural progression.

 

[Mark R]

i am not going so far to so it would be "trash". I am expressing my disbelief in their claims and curiousity as to how bad it will be. So far all we have to go by is:
1. previous generations have been progressively worse.
2. The spin machine is already in full swing. trying to sell us on blatant lies (same P/E count via ECC and wear leveling).

It's been known since the inception of flash memory, that cell size is flash's biggest enemy. For that reason, physicists have been predicting the demise of flash for some time.

It was only about 5-6 years ago that there were scientific papers coming out stating a minimum usable flash cell size of 22 nm - with estimates of barely 1000 cycles for SLC, and a few hundred for MLC.

Since then, there have been some innovations, with improved dielectrics and better control of oxide thickness, etc. These have made MLC practical at 25 nm, whereas 5-6 years ago, most physicists would have suggested that MLC would not be practical at 25 nm given flash technology at the time.

However, these are only incremental changes, and it seems implausible that flash, even with further incremental advances, is likely to be usable beyond 20 nm.

 

[Voo]

Well a current controller with compression should easily be fine with 1k PE cycles for 2years, which is longer than the upgrade cycles for most other parts in a machine around here.

Considering that we have no idea how much PE cycles the current flash memory really has - the published numbers are obviously conservative (otherwise you hardly would get the same numbers for both 34 and 25nm), flash should be fine for the next few die shrinks at least.

And to forecast development in such a fast changing field over a longer timespan hasn't worked especially well in the past so I think I won't even try. So many people claimed that 20nm processes wouldn't be feasible in the 90s and well, see how that turned out. Well actually I think the latest claims are that you can't scale below 16nm because of quantum tunneling effects, oh well.

 

[groberts101]

not to mention that huge increases in capacity can buy back some of the lifespan loss.

Plus if it's cheap enough, it can be stacked up enough and larger amounts put into reserve(to protect from the idiots that fill them up as if it's HDD) that we'll be able to get by with lower PE/c nand. In fact, that's what the Sandforce controller was initially designed for and will go much easier on the space hit at the physical level.


Eventually, these things will require a 3.5 casing just to physically fit all the extra nand chips in there and laptops will be limited in the max capacity they can use due to the resrictions of smaller cases. We're already there actually.

 

[VirtualLarry]

There was one place that specializes in flash drive recovery, and they claimed that the limit on 3LC, was 10 (yes, ten) writes. After that, you started to see bit errors.

A different thread here on AT, mentioned that Sandisk USB flash drives were already using 3LC, because they had advanced controller chips.

Scary, really.

I'm glad I still have a 16GB Sandisk Cruiser, with the older style casing (rounder, not squarish), which should still be 2LC.

Edit: Update, I recently purchased a 32GB Sandisk Cruiser (BestBuy $40), and while attempting to use it, it hung several times. I wiped it completely with vconsole.com 's USB flash drive tester, and then after reformatting it in windows, it had lost a bunch of sectors. Finally, I tried using it again for what I intended it for, and it worked. It's a bit slow (4MB/sec write, 16MB/sec read).

 

[LokutusofBorg]

Given the failure rate I recall with 3.5" floppies, I'll take 10 writes on my thumb drive. I'd like to see a reference/link on that though, as 10 is absurdly low, even for 3LC. All storage has bit errors. And bit errors have little to do with NAND write limits, no?

Back to the OP, the quote that stuck in my mind when IMFT made the same claim is that when they were at the same point in refining the manufacturing process with their older NAND as they are now with their newer NAND their stress testing was showing the same write limits. So they expect the same rating once they've refined the process. They are definitely bumping up against physical limits, but the processes are evolving in ways that push that physical limit out. Note the recent article about 3D transistors. Advances at that level are still possible.

And the article you quote there wasn't written by Anand, so I wouldn't put as much weight in the specifics of how he worded things as I would if Anand had written it himself. I agree ECC and wear leveling have nothing to do with write limits. That article needed a good edit pass from Anand.

 

[Dark Shroud]

I'll be buying on of the large 320 drives when the price comes down a bit more. I want the 300gig model but I don't want to pay more than $500. I think $450 is actually more more fair price at $1.50 per gig.

The real question is if Intel 20nm NAND can maintain the 5,000 write cycle the same as the 25nm NAND.

 

[bulanula]

The real question is if Intel 20nm NAND can maintain the 5,000 write cycle the same as the 25nm NAND.

It cannot. It is just basic physics. No matter how much you "refine the process". Just get one of the good old trusty G2 drives before they get discontinued.

 

[taltamir]

I'll be buying on of the large 320 drives when the price comes down a bit more. I want the 300gig model but I don't want to pay more than $500. I think $450 is actually more more fair price at $1.50 per gig.

The real question is if Intel 20nm NAND can maintain the 5,000 write cycle the same as the 25nm NAND.

5k is 32nm nand. 25nm nand from intel is 3k

 

[VirtualLarry]

Given the failure rate I recall with 3.5" floppies, I'll take 10 writes on my thumb drive. I'd like to see a reference/link on that though, as 10 is absurdly low, even for 3LC. All storage has bit errors. And bit errors have little to do with NAND write limits, no?

http://www.recovermyflashdrive.com/articles/29F64G08CFAAA-29F64G08FAMCI-3Bit-MLC-high-failure-rate

 

[Voo]

http://www.recovermyflashdrive.com/articles/29F64G08CFAAA-29F64G08FAMCI-3Bit-MLC-high-failure-rate

Well considering that the amount of people exhausting their PE cycles and then posting here so far is: 0. So we have a few possibilities: a) this information is incorrect (at least the other number of PE cycles stated for MLC is clearly wrong) or b) nobody uses that flash for SSDs (although the article claims it) or c) anyone who happened to got such a drive decided to just say nothing.

So in 2 of 3 cases we can reason that the article isn't correct and the 3rd one? Really? People love to complain about anything, especially on the net, so that sounds highly unlikely.

 

[LokutusofBorg]

I still say bit errors of the kind being referred to in that "article" have little to nothing to do with write limits. When NAND wears out due to write limits, it becomes read-only. I don't think a data-recovery company is going to be called in to deal with a NAND device that has switched over to read-only... Those devices are defective. That's a completely different situation.

 

[DirkGently1]

I don't think endurance is going to be an issue.

http://www.xtremesystems.org/forums/showthread.php?t=271063

 

[frostedflakes]

I don't think endurance is going to be an issue.

http://www.xtremesystems.org/forums/showthread.php?t=271063

Really interesting link, thanks for posting it. Am anxious to see the results. I always wondered why you don't see tests like this more often, especially with smaller SSDs that have become relatively affordable. No better way to answer the durability question than to pound the disk with writes and see how long until it fails.

 

[aceO07]

Great link. I'm looking forward to updates from the guy with the Intel 320.

 

[Voo]

Yep always nice to have someone who's got no problem ruining his nice 40gb SSD. For science!

 

[Mark R]

Given the failure rate I recall with 3.5" floppies, I'll take 10 writes on my thumb drive. I'd like to see a reference/link on that though, as 10 is absurdly low, even for 3LC. All storage has bit errors. And bit errors have little to do with NAND write limits, no?

Bit errors are everything to do with NAND write limits. The cycle life is determined as how many cycles, on average, it takes to degrade the flash to where the bit error rate becomes unacceptable. Bit error rates of up to 8 bits per 8 kB sector are generally considered acceptable, and for 3-bit MLC, error rates of 10x that may be expected during use.

All flash, be it SLC, 2-bit MLC, or 3-bit MLC, tends to have a low error rate when fresh (apart from manufacturing defects), and the error rate gradually rises as the wear increases. Not only that, but the data retention time also decreases as wear rises (this is what eMLC is - it's just 2-bit MLC, but the flash is only specified to have a data retention time of 3-6 months. Normal MLC is specced for 10 years, but after about 5k cycles for 32 nm, it can't achieve that. If you respec it for 6 months, then you can get another 10k cycles before the flash can't meet the data retention specification).

 

[lol123]

It's really amazing how bad of a technology flash memory is. I for one can't wait until there's a reasonable alternative, like MRAM, PRAM etc. to go into SSDs.

 

[DirkGently1]

I don't think endurance is going to be an issue.

http://www.xtremesystems.org/forums/showthread.php?t=271063

The latest graph..

One quick calculation based on the above curve and my Intel 320 will last me 136 years with my current daily usage :thumbsup:

 

[frostedflakes]

Interesting how the 25nm based 320 seems to be fairing better than the 34nm X25-V.

 

[DirkGently1]

Interesting how the 25nm based 320 seems to be fairing better than the 34nm X25-V.

The prevailing theory seems to be that even with the lower nand endurance, the controllers are improving enough with each generation to negate the issue. Even a couple of years ago people in the know were saying controllers would be the biggest point of failure rather than the nand. So far that has proven to be correct.

I think it's finally safe to say that people can stop worrying about 'wearing out' their SSD drives. What we need is 22nm and beyond architecture, higher capacities and lower prices. It would be nice to have a 1TB SSD storage drive!