What flash needs is a little TLC
Two bits good, three bits better
Comment The breakthrough when flash becomes affordable is thought by many people to be TLC (triple-level cell) – 3-bit multi-level cell NAND – which adds a third more capacity to flash cells. But how far are we with TLC implementation, and when will we see it in mainstream IT?
According to Jim Handy of Objective Analysis, SanDisk is the TLC market leader. Handy says: "I think 3-bit might be over half of their gigabyte shipments and 19nm accounts for a significant part as well. It has been maybe a year since SanDisk started to ship production volumes of 3-bit flash. It is in products that you may already own."
In an earnings call, SanDisk CEO Sanjay Mehrotra said: "We are on track to ramp production of our 19-nanometer technology this quarter using both 2-bit-per-cell MLC as well as 3-bit-per-cell X3 memory, which we expect will sustain our cost leadership in 2012."
Three-bit MLC is known as X3 flash, and also TLC flash, with 2-bit MLC described as X2 flash.
The background is that SanDisk's mainstream NAND production uses a 24nm process. It is developing a smaller 19nm process which increases the number of flash dies obtainable from a wafer and so lowers their unit cost.
Combine 3-bit cells with 19nm processes and you have a recipe for flash endurance that is quite short, over in a flash as you might say.
Although SanDisk is shipping TLC product, we don't see it being used in mainstream IT products such as tablets, notebooks, desktops, servers and storage arrays. Handy more or less agrees with this scenario: "TLC (X3) is shipping in USB flash drives, camera cards, and MP3 players. I don't believe it's yet being used in the markets you mention, but I am not certain."
There are several reasons for this, including the points that flash endurance - its working life - reduces with every process shrink, and also reduces with every added cell bit. Thus 19nm NAND will have a shorter working life than 24nm NAND, and 2-bit MLC flash won't last as long as single-level cell flash. Combine 3-bit cells with 19nm processes and you have a recipe for flash endurance that is quite short, over in a flash as you might say.
SanDisk and Pliant
The antidote to this is the use of cleverer controllers that reduce the number of writes, have spare capacity set aside to replace dead cells, and are better at getting usable bit value information from cells that are wearing out. SanDisk has bought Pliant, a supplier of mainstream IT flash controllers, to develop an in-house controller capacity, but Handy doesn't think the SanDisk Pliant product line will use SanDisk's 19nm or 3-bit flash any time soon.
He said: "Since SanDisk doesn't ship raw NAND to the flash chip market, SanDisk flash would not have been available to Pliant prior to the acquisition, and it's too soon for Pliant to have converted. If I remember correctly, Samsung is a financier of Pliant, making it most likely that Plaint drives use Samsung SLC flash. Samsung is the supplier of the majority of the world's SLC flash, and SanDisk doesn't make any SLC as far as I know.
"It's also probable that Pliant SSDs use 55nm flash. Enterprise SSDs tend to use flash whose attributes are very well understood – in other words, time-tested (and nearly obsolete) process geometries. Since Samsung is the technology laggard in NAND flash, then Samsung 55nm is the product that Pliant is most likely to be using."
"The most probable scenario will be that Pliant SSDs will migrate to SanDisk flash over the following year or two, but will continue to use the process that SanDisk is phasing out. I would expect Pliant to convert to MLC flash over the next year, and by early 2013, this is likely to be SanDisk's 24nm X2 flash."
Mainstream IT use of TLC NAND
Thus Pliant-controlled product will enter the 19nm flash era probably in 2014. El Reg thinks it could have TLC 24nm flash products towards the end of 2012. SanDisk shares its flash foundry capability with Toshiba, and that company has an interest in Violin Memory, the developer of networked flash memory arrays.
We might see TLC product coming from SanDisk in SSD and PCIe card formats by the end of 2012, and from Violin Memory at the same time. If both companies manage to have controllers that provide acceptable endurance, say a nominal five years, and product that is substantially cheaper than 2-bit MLC capacity equivalents then they could expect to sell exabytes of the stuff.
The other flash foundry owners such as market leader Samsung, Intel and Micron, and others will have to up their game to compete. HP, with fab parter Hynix, is making aggressive noises about its Memristor technology attacking mainstream flash markets by mid-2013.
TLC hindered by resistive RAM
Handy is hesitant about HP's chances: "I am wary of the prospects for the Memristor ... HP has some aggressive claims, but the company has no existing memory business to propel this effort and is currently re-evaluating every one of its cost centres, so we will have to wait and see what the new management has to say about funding the development of the Memristor. Although Hynix is a very capable company, I don't expect for it to take the Memristor technology in its current state and bring it to the market by itself."
Could IBM's racetrack memory initiative affect the fortunes of flash in the next two or three years?
"Racetrack's very existence was disclosed in 2008, and I compare that to Ramtron's FRAM which was disclosed around 1988, and is still only in limited production. It's a cool technology, but we don't know that it will be able to unseat flash as the low-price-leader any time within the next few process generations, and in memory price is everything!"
"MRAM, PCM, and FRAM all have a lot more experience than racetrack, but I wouldn't dare to discount IBM's deep pockets and the company's ability to fund this technology so well that the others are left behind. On the other hand, IBM's Crocus licence indicates that Racetrack isn't the only horse the company is betting on."
Future NAND closed door
What looks possible, even probable, is that TLC NAND makes it to market first and is then faced with resistive RAM – memristor, phase-change (PCM), MRAM, etc – competition. Whether that future product can deliver the goods of combining DRAM speed, non-volatility and NAND capacity in a TLC NAND-beating way is a big, big question. It does seem likely though that 4-bit NAND won't happen and neither will sub-10nm NAND because resistive memory technology will have progressed enough by then to close the door on such developments.
El Reg thinks that the tender loving care SanDisk is paying to TLC flash is about to pay dividends and that Violin Memory could be looking at TLC as well. ®
Sponsored: Hyper-scale data management