backtop


Print 25 comment(s) - last by PrezWeezy.. on Oct 21 at 1:25 PM

Fewer reasons remaining against SSD adoption

Limited write endurance is one of the factors that detractors bring up with regards to solid state drives (SSDs). Most NAND flash chips using multi-level cell (MLC)technology in SSDs have a write endurance of around 10,000 cycles. That isn't as great a problem in SSDs greater than 120GB that use wear-leveling technology, but smaller sized SSDs have less capacity and will reach the upper limit much quicker.

That issue is why almost all SSDs aimed at the corporate and enterprise market use Single Level Cell flash chips, which typically have a write endurance around the 100,000 cycle mark. These include Intel's X25-E, OCZ's Vertex EX and Agility EX series, and Super Talent's MasterDrive RX series.

Micron Technology is one of the key partners in IM Flash Tech along with Intel Corporation. IMFT produces the 34nm NAND flash used in Intel's second generation X25-M SSDs using 2-bit-per-cell MLC chips. Micron and IMFT have been working on improving the write endurance of their NAND chips, and they have now reached a breakthrough.

“By leveraging our mature 34nm NAND process, Micron has developed Enterprise NAND products that support customers’ high-endurance requirements. These products ensure that enterprise organizations have a highly reliable NAND flash solution – be it MLC or SLC – for design into the broader enterprise storage platform,” said Brian Shirley, Vice President of Micron’s memory group.

The company’s new 32Gb MLC Enterprise NAND devices achieve an impressive 30,000 write cycles. They are also introducing a 16Gb SLC Enterprise NAND device that achieves 300,000 write cycles. The new chips also support the ONFI 2.1 synchronous interface, making them easier to integrate into new products.

Both of these new chips are built on the 34nm process which IMFT introduced last year, and can be configured into multi-die, single packages supporting densities of up to 32GB for MLC NAND and 16GB for SLC NAND.

Micron is now sampling its Enterprise NAND products with customers and controller manufacturers, and is expected to enter volume production at the beginning of 2010.



Comments     Threshold


This article is over a month old, voting and posting comments is disabled

Cycle counting
By martinw on 10/19/2009 12:57:36 PM , Rating: 2
Anyone know what the metric really means? Does it mean that after 30k cycles 50% of all such devices will have failed? And what is the distribution of failures? Is the profile gaussian? If so, what is the standard deviation?




RE: Cycle counting
By ilkhan on 10/19/2009 1:49:04 PM , Rating: 2
AFAIK its the number of writes per cell before failure. Its a range of course, but probably an average with a fairly low SD.

Regardless, by the time my G2 wears out I'll have moved to a new drive.


RE: Cycle counting
By menting on 10/19/2009 6:46:57 PM , Rating: 2
the standard deviation is probably a trade secret, but the 30k cycles is probably the average, and it's probably not gaussian.
30k cycles is quite a bit in an SSD, with a "good" write leveling algorithm. Basically it means if you write your HD over once every day, it'll still take about 30k cycles on average before fails come in. And the SSDs probably have extra NAND chips to replace those cells marked as bad.


RE: Cycle counting
By PrezWeezy on 10/19/2009 8:31:44 PM , Rating: 2
The way Intel does wear leveling is by actually putting 80 GB on the drive, then telling the OS it has ~74 GB. It's the old 1000 vs. 1024 standard that they use to calculate the amount of extra chips. IE if you had a rotational "80 GB" drive you end up with ~74 GB as calculated by the OS. So the Intel SSD tells the OS it has the 74 GB usable, then uses their wear leveling algorithm to decide where to write the data the most efficient way.


RE: Cycle counting
By jordanclock on 10/20/2009 3:02:16 AM , Rating: 2
Not quite. What SSD manufacturers do for wear leveling is basically intentionally fragment the drive. Since random access time is the same as sequential, fragmentation doesn't matter much. So with that in mind, they spread out the writes to all parts of the drive. This means that no one area is likely to fail before the rest.

As for the 80GB drive formatting to 74GBs, yes. It's the classic issue of different ideas of kilo/mega/giga. Also, SSDs have sort of "extra" cells. They aren't normally available, but as cells fail, they are activated for use by the firmware. This means storage capacity remains about the same as cells fail for a bit, instead of dropping.


RE: Cycle counting
By SAnderson on 10/20/2009 3:59:42 PM , Rating: 2
Correct. How else do you get 80/160GB from a 16/32 Gb Chip? Its one of the way Intel designs the controller to help deal with write alg. and the pesky way Flash deletes data. Extra space is available for the controller to use but only 80GB is available. And yes once some cells fail, you will still have 80GB of good cells to use, thus GREATLY increasing the average cycling per cell. Say you have 90GB of cells worth of cycling but with only 80GB of data spread across it.


RE: Cycle counting
By PrezWeezy on 10/21/2009 1:23:37 PM , Rating: 2
I think if you use 5x16GB chips you end up with 80 GB...And if I'm not mistaken 10x16 still equals 160.


RE: Cycle counting
By PrezWeezy on 10/21/2009 1:25:19 PM , Rating: 2
Go read the SSD Relapse at Anandtech. he explains it very well how the wear leveling works. It is, in fact, an 80 GB drive. It does report 74 GB to the OS and it uses the other 6 GB for it's wear leveling and "extra" cells you are talking about.


RE: Cycle counting
By motigez1 on 10/20/2009 3:27:22 PM , Rating: 2
it is very simple.
10K cycles means you can fill the media 10K times, i.e. a 64GB SSD, you may write 10000 times, or 64GB *10000 = 640 terabytes, let's assume efficiency is not perfect so you can write only a third ~200TB.
how much do you need?
an average client user with windows OS will write 4-5GB/day, assume 365 working days, so in the entire year you will write 5*365 = 1.8TB
which means you can use the drive for 200TB/1.8TB ~ 100years!!! you would probably replace your NoteBook in less than 5yrs,
now, you should tell me if 500 cycles are not enough!
Micron does not enable 30K for us, the end users, but for the data center storage. we should be safe with only 500 Cycles! (assume a decent SSD design)


RE: Cycle counting
By SAnderson on 10/20/2009 4:02:43 PM , Rating: 2
These will not be priced for normal users, just the normal 1k/5k cycle chips. Its basically the same product with a few small process differences to earn a larger margin. The server market makes the big bucks for chip manufactures.


"Nowadays you can buy a CPU cheaper than the CPU fan." -- Unnamed AMD executive














botimage
Copyright 2014 DailyTech LLC. - RSS Feed | Advertise | About Us | Ethics | FAQ | Terms, Conditions & Privacy Information | Kristopher Kubicki