25nm NAND Flash

[her209]

Intel and Micron Double Flash Storage Capacity With 25-Nanometer NAND
http://www.pcworld.com/businesscent...h_storage_capacity_with_25nanometer_nand.html

 

[Engineer]

Intel and Micron Double Flash Storage Capacity With 25-Nanometer NAND
http://www.pcworld.com/businesscent...h_storage_capacity_with_25nanometer_nand.html

Shame..Shame...You're supposed to link the AnandTech article too!

 

[RU482]

Shame..Shame...You're supposed to link the AnandTech article too!

give him...sorry, her....a break, there is mourning on the (late night) passing of Conan O'brian to be doing

 

[bradley]

On the flip side of Intel's profit equation, the 25nm NAND shrinkage also means decreased longevity. So perhaps you'd be paying less for less. Whether that means anything in the real world is anyone's guess ATM.

 

[RU482]

On the flip side of Intel's profit equation, the 25nm NAND shrinkage also means decreased longevity. So perhaps you'd be paying less for less. Whether that means anything in the real world is anyone's guess ATM.

I'm listening...please elaborate.
I hope you are not assuming that heat is a constant

 

[frostedflakes]

So in theory, 80GB SSDs should now be as cheap to produce as 40GB ones were. Let's see if the prices actually follow. 80GB for ~$100 might be good enough to get a lot more people to adopt the technology. 40GB just doesn't cut it these days for most people, IMO 64GB/80GB is the bare minimum for an OS/apps drive.

 

[Emulex]

40gb for $110 is worth it.

backup times way faster (think about backing up 50 machines overnight)
full virus scan + active scans

window of time it takes to do those is so much smaller it is a sys admin dream.

 

[IntelCeleron]

There is an interesting discussion going on about this over at HardForum. Specifically, the data retention length and the lifespan of the drive.

http://hardforum.com/showthread.php?t=1492711

 

[Brian Stirling]

There is an interesting discussion going on about this over at HardForum. Specifically, the data retention length and the lifespan of the drive.

http://hardforum.com/showthread.php?t=1492711

Which begs the question ... what is the best strategy for users of SSD's to protect there data. If the data retention thing implies that you need to refresh the entire SSD periodically would you:

1. Make backup of SSD

2. Full clean erase of SSD

3. Restore backup to SSD


If so, what are the implications for partition offset etc.?


Brian

 

[frostedflakes]

As far as the larger page and block sizes, it shouldn't be an issue as long as the controller has a large enough write cache to prevent excessive erase/write cycles. I think the Indilinx and Sandforce drives have write caches on the order of 32MB and 64MB, so there should be plenty of space to cache data and write it in 2MB blocks. No different than previous drives with write caching, you just need more cache to accommodate blocks than you would with 50nm or 34nm flash.

 

[IntelUser2000]

According to Anandtech, Sandforce controllers have no dedicated DRAM, but a large controller cache.

 

[Emulex]

cache ram or dram? semantics.

if you don't have a battery or capacitor you best be using the ram for something other than write-back caching.

 

[cbn]

On the flip side of Intel's profit equation, the 25nm NAND shrinkage also means decreased longevity.

Why does the shrinkage mean less longevity?

 

[Mark R]

cache ram or dram? semantics.

if you don't have a battery or capacitor you best be using the ram for something other than write-back caching.

The sandforce controller has power-fail detection and is designed for use with a supercapacitor. The seagate uses this, as does the OCZ Vertex2. They can therefore perform generous write caching, combining and abandonment.

The older indilinx controller, as used in older drives has been fitted with up to 64MB of cache specifically to reduce random write penalty - however, there is no capacitor backup. That's a serious fail.

 

[Mark R]

Why does the shrinkage mean less longevity?

Smaller cells hold fewer electrons, so the signal is weaker, and can be disturbed more easily. You also get thinner insulation around the flash cell.

This is worse in MLC Flash, where the flash cell holds an analog voltage, the memory contains a 2 bit or 3 bit ADC to go back to digital, making MLC much more sensitive to fluctuations in flash cell voltage. SLC is pure digital so is very robust against fluctuations in stored voltage.

The problem with flash is that you need to get electrons in and out of a fully insulated cell. This is done by using high voltage to "tunnel" electrons through the insulation. However, the tunnelling process damages the insulation by degrading its crystal structure. When the insulation is damaged, the electrons can leak out of the flash cell.

Smaller flash cells have less margin, and MLC have even less, so you get 2 problems:
1. Fewer erase cycles before the insulation is critically damaged.
2. More leakage (relative to size of cell) so the data 'fades' faster. Some engineers predict data may only last months on 25nm before it fades so much to be unreadable (although this is a theoretical estimate).

 

[cbn]

Smaller cells hold fewer electrons, so the signal is weaker, and can be disturbed more easily. You also get thinner insulation around the flash cell.

This is worse in MLC Flash, where the flash cell holds an analog voltage, the memory contains a 2 bit or 3 bit ADC to go back to digital, making MLC much more sensitive to fluctuations in flash cell voltage. SLC is pure digital so is very robust against fluctuations in stored voltage.

The problem with flash is that you need to get electrons in and out of a fully insulated cell. This is done by using high voltage to "tunnel" electrons through the insulation. However, the tunnelling process damages the insulation by degrading its crystal structure. When the insulation is damaged, the electrons can leak out of the flash cell.

Smaller flash cells have less margin, and MLC have even less, so you get 2 problems:
1. Fewer erase cycles before the insulation is critically damaged.
2. More leakage (relative to size of cell) so the data 'fades' faster. Some engineers predict data may only last months on 25nm before it fades so much to be unreadable (although this is a theoretical estimate).

From #2 (in your post): Data fading within months? I was hoping these SSDs would be more durable than magnetic drives.

So how long till someone fixes this problem? I was really hoping the price on these SSD devices would fall pretty quickly.

 

[Mark R]

From #2 (in your post): Data fading within months? I was hoping these SSDs would be more durable than magnetic drives.

So how long till someone fixes this problem? I was really hoping the price on these SSD devices would fall pretty quickly.

First off, this is a theoretical estimate - no one is producing 25 nm flash in big volumes yet, so we don't know for sure how stable 25 nm flash is.

The problem with flash is that you have a insulated cell which stores electrons. All insulation leaks at this nano-scale, but with big 50 nm cells and their thick insulation, the leakage is minute and the cells hold so many electrons, that you can leave the cells untouched for 10-20 years before the cells deplete enough to misread.

At smaller sizes, your cells are smaller and your insulation is thinner - it leaks faster even when in perfect condition. So you get a double whammy effect - faster leakage, and cells that are more sensitive to leakage.

Not only that, but the thinner insulation is more fragile, so it suffers relatively more damage from electron tunneling. So you get fewer erase-write cycles until it hopelessly damaged. It's now a triple whammy!

Perhaps someone will come up with a way to 'refresh' flash memory, in the same was as DRAM can be refreshed. DRAM works by storing electrons in a capacitor - however, unlike in flash, the capacitor isn't fully insulated, so the capacitor discharges over a period of a few miliseconds. To get around this, about 4000 times a second, a 'refresh' circuit is activated which reads the whole RAM bank and resets the capacitors to the appropriate voltage. I don't see why you couldn't do this with flash - but I suspect that even such a refresh would still cause wear, because it would still need to tunnel electrons into the flash cells. And it wouldn't do anything when the flash is powered off.

No doubt the flash manufacturers are working on clever technologies to try and improve the performance of their memory - e.g. better insulation layers, etc. However, these are only likely to be minor upgrades.

It may be that flash is getting close to its limits. The current theory of flash cell design puts a limit of about 20nm at flash cell size. Smaller than this, the cells work so badly as to be practically useless.