Flash NAND capacities are increasing briskly as wafer die size continues to shrink multi-level cell (MLC), now...
as small as 14 nm. Price per gigabyte is falling rapidly. Error detection and correction keeps getting better. Venture capitalists are throwing money at startups, with recent funding announcements for Pure Storage, SolidFire, Nimble Storage, GreenBytes, Astute Networks and Starboard Storage, and this isn't slowing down. This is because flash storage companies are being rapidly swept up. EMC, IBM, LSI and OCZ have made acquisitions, and rumors are afoot about other potential flash storage vendor acquisitions.
Times are good for the flash market.
But with all this activity and excitement in the flash storage market, why is there is so little talk, buzz or even mention about TLC flash or triple-bit-per-cell flash NAND? Where is it? What's going on?
Remember, the flash SSD market started with single-level cell (SLC). Performance was outstanding, whereas up-front costs were breathtaking, and not in a good way. Market pressure to reduce costs led to the development of MLC.
In a previous tip, I explained that MLC is the flash storage industry's answer to the cost and capacity issues of SLC. TLC reduces costs substantially below MLC's, with potentially higher capacities. MLC has four states (00, 01, 10, 11) and SLC has two (0, 1), whereas TLC has eight (000, 001, 010, 100, 011, 110, 101, 111), or twice the density or capacity of MLC.
TLC should logically be emerging now as the lower-cost alternative in flash SSDs. It has not. There are several reasons why.
The first reason is short lifecycles known as "program/erase (P/E) cycles." TLC flash has significantly shorter P/E cycles than MLC. MLC NAND has an approximate lifecycle range of 5,000 to 10,000 P/E cycles. Clever flash controller wear-leveling and error correction code (ECC) algorithms are now extending that MLC lifecycle to as much as 100,000 P/E cycles. TLC P/E cycles, on the other hand, are an order-of-magnitude less and range from 800 to 1,000 P/E cycles. This is OK for some consumer products and totally unacceptable for any business class of storage systems.
What about those clever flash controller algorithms? If they can increase that lifecycle by 10 times, as they can do with MLC, it would put TLC into the low range of native MLC lifecycles. Regrettably, this is a much more difficult task than it is with MLC. Having twice the number of states in TLC makes it far more difficult to get a positive value determination in each cell. Doing so requires more power. More power generates more heat (aka, waste heat). More heat requires more cooling. More power and cooling cut into the flash SSD TCO value proposition. That difficulty in determining the cell's state also means slower performance, higher electron leakage and greater corruption. Ultimately, the controller wear-leveling and ECC algorithms must provide a leap in robustness just to get it into the low end of native MLC lifecycles.
Another reason is the rapidly decreasing MLC die size. Reducing the die reduces the cost. Reducing MLC costs reduces the market pressure for a lower-cost alternative such as TLC. At press time, 14 nm MLC is already being shipped in limited quantities. One SSD vendor is sampling 800 GB SSDs based on this form factor at below $2 per gigabyte to the OEMs.
Yet, as positive as this is for MLC prospects, the road ahead becomes a bit fuzzier. Shrinking below 14 nm becomes increasingly more difficult. At some point in the near future, reduced die size gains will no longer be economically feasible. To meet market expectation of continuously decreasing cost per gigabyte and increasing capacities will force flash NAND manufacturers to increase the bits per cell (density) or, in other words, TLC.
TLC will become increasingly viable as the flash industry's lower-cost alternative. In fact, high-density, small-die-size TLC will rival high-density hard disk drives (HDD) in cost per gigabyte and rack real estate. It may, in fact, have lower capital expenditures (CapEx) and much lower operating expenditures (OpEx) for real estate, power and cooling.
Several vendors are currently working on TLC storage products. OCZ announced a TLC controller this year and strongly hinted at TLC SSDs. Samsung, Micron, Intel, LSI and others are working on TLC SSDs. Earlier this year, GreenBytes announced a flash storage system product called "Solidarity," which is allegedly based on TLC.
The conclusion is that TLC flash is a work in progress. It's coming. Realistically, TLC will enter the market first for the consumer market, perhaps a bit later for the small office home office (SOHO) and small and medium-sized (SMB) markets. Eventually, it will appear in the enterprise markets. When it does, the issue of storage tiering will move front and center once again. TLC creates the same disparity in price/performance for SSDs (SLC and MLC) that exists today with HDDs between SAS and SATA, or between SAS and nearline SAS. It will be déjà vu all over again.
About the author
Marc Staimer is the founder, senior analyst and CDS of Dragon Slayer Consulting in Beaverton, OR. Marc can be reached at firstname.lastname@example.org.