SandForce Demos 19nm Toshiba & 20nm IMFT NAND Flash
by Anand Lal Shimpi on June 8, 2012 4:08 AM EST- Posted in
- Trade Shows
- Storage
- SSDs
- SandForce
- IMFT
- NAND
- Computex 2012
SandForce's controllers have fairly broad compatibility with NAND available on the market today. It shouldn't be a surprise that the first demo we saw of Toshiba's 19nm and Intel/Micron's 20nm NAND was at the SandForce/LSI suite in the Grand Hyatt in Taipei.
Even though 19nm/20nm 64Gbit devices aren't very different from their predecessors, they still require custom firmware support for many reasons including dealing with different program times. Intel and Micron have told us that they expect similar endurance from their 20nm NAND as we currently see with their 25nm offerings.
The demo we saw was a simple Iometer test on both platforms. SandForce built SF-2000 drives using NAND from Intel, Micron and Toshiba. The Intel part number hasn't changed much from the 25nm generation. The table below highlights the differences:
| Intel MLC NAND Part Number Comparison | ||||
| 25nm 64Gbit | 20nm 64Gbit | |||
| Intel | 29F64G08ACME2 | 29F64G08ACMF3 | ||
The last F actually indicates that we're looking at a 20nm part, while the 3 actually refers to generation which seems to have been incremented along with the process node identifier. We should see the first 19nm/20nm based SSDs this year, but keep in mind these aren't dramatically different architecturally from the 24/25nm parts on the market today.
When we get to 128Gbit 20nm die from IMFT next year we will see some major changes including a faster interface (ONFi 3.x) and larger page sizes (16KB).
Facebook
Twitter
RSS
20 Comments
View All Comments
meloz - Friday, June 8, 2012 - link
>We should see the first 19nm/20nm based SSDs this year, but keep in mind these aren't dramatically different architecturally from the 24/25nm parts on the market today.I -and I suspect most consumers- don't want more performance compared to current generation SSDs, but hopefully these NAND flash will be will be more economical per gigabyte. And thus result in cheaper SSDs.
drumm_22 - Friday, June 8, 2012 - link
Will they be more stable? And how much longer will they last if they even last longer?Kristian Vättö - Friday, June 8, 2012 - link
Moving to a smaller process node does not result in an immediate price drop. The R&D behind a die shrink isn't exactly cheap and yields will be lower at first. Once the process has matured and the supply is good, prices start to fall.I would say the decline in NAND prices is pretty linear, although I don't have any concrete data to support that. However, overnight price drops are unlikely.
drumm_22 - Friday, June 8, 2012 - link
how much longer do you think before these new SSD's will drop in price?LordConrad - Saturday, June 9, 2012 - link
I would rather keep the current pricing than accept a reduction in program/erase cycles. I don't care what Intel says, smaller node processes result in fewer p/e cycles.drumm_22 - Saturday, June 9, 2012 - link
SO are you saying that these smaller node processes will result in SSD's that wont last as long?Nihility - Saturday, June 9, 2012 - link
They might improve the write amplification to compensate for the decreased write cycles, but in general you can expect a decrease in life expectancy as the flash gets smaller.phillyry - Tuesday, July 31, 2012 - link
See above.phillyry - Tuesday, July 31, 2012 - link
But so far p/e cycles haven't been the factor limiting getting these things in people's hands. It's been price and controller reliability.I'm pretty sure that while p/e cycles have gone down with each step down to a smaller cell size, price has gone greatly down, longevity based on p/e cycles has never been an issue in the consumer marketplace, and the controllers have managed to compensate for the decreased p/e cycles by using algorithms that decrease the amount of writes that are actually made to the NAND. Hence, the term wear leveling (http://www.google.com/search?q=wear%20leveling%20w... The idea, as Anand has clearly illustrated in his many SSD articles, is that the controller compresses the data and also works to minimize the write amplification that would occur as each block is moved and re-written.
Overall, improve tech via controller. Decrease cost via MLC and decreasing cell size (x nm). And each gen. we end up with larger, faster, cheaper, more reliable SSDs.
drumm_22 - Friday, June 8, 2012 - link
With these new 19nm/20nm drives coming out would it be best advised to wait to get one of these new SSD's or are the ones on market today reliable enough and worth the money?