I attended the Flash Memory Summit last month in Santa Clara, and although it was primarily a vendor fest, the few customers that attended were very enthusiastic about their experiences with Flash. The vendors were exhibiting the latest in flash technology like higher density 3D NAND and other NVM memory technologies like ReRAM. Samsung and Toshiba announced 16 TB SSD’s using 3D NAND flash technology. Toshiba predicted that they could push capacities to 128TB by 2018! Now flash drives will have more capacity than the largest hard drive, but at what cost and when will they be available for the enterprise market?
Instead of doubling bit densities to increase capacities as we do with disk drives, flash vendors are increasing densities by going to multiple layers (32 to 128). With conventional planar or 2D NAND (MLC and TLC), densities are increased by shrinking the memory cell. However, as the memory cell shrinks it becomes harder to retain electrons in the floating gate and the durability of flash degrades. By stacking the cells vertically, 3D NAND vendors are not only able to increase capacity by multiple layers of cells, but are also able to increase durability by relaxing the cell size to store more electrons.
There is no common approach to 3D NAND and different vendors have different structures. In order to commercialize this technology, manufacturers will have to build new Fab facilities, which will cost billions of dollars. In order to recover their investment, the vendors will need to displace the 2D NAND flash drive market to drive their volumes. That market is currently being served by MLC (multi-level) and TLC (triple level). The window for 3D NAND may also be very short since other types of technologies like ReRAM, with higher nanosecond speeds, orders of magnitude more endurance, and no need for preformatted cells, are waiting in the wings. Here is an example of the different structures, which will be competing for this market.
Also at the Flash Memory Summit were TLC and MLC flash array vendors. Prices for those technologies are eroding much faster than we thought a few years ago. While 3D NAND flash technology increases capacity by stacking cells in layers, MLC and TLC, use one 2D layer and vary the charge levels to create different states. MLC has two charge levels, which provides 4 states while TLC has 3 levels, which provides 8 states.
MLC states: 00 01 10 11
TLC states: 000 001 010 011 100 101 110 111
Distinguishing the intermediate charge levels and managing interference from adjacent cells, degrades the write performance and durability of the flash device. Where an MLC cell may be good for 20,000 to 30,000 writes, TLC, may only be good for 200 to 300 writes. However, TLC can double the capacity without a major new investment in the MLC fab process. While endurance was a major concern when flash drives were introduced, experience has shown that MLC drives, with wear leveling, have not been a problem from that perspective. Some vendors will be introducing TLC flash as a lower tier of storage for less frequently written data at a lower price per capacity than hard disks. One presentation at the Flash Memory Summit advocated TLC for long-term archive with an SSD controller that could manage the endurance cycle of TLC. The advantage of a TLC archive would be faster search and access to data with lower environmentals than disk.
The TCO for MLC flash is already better than 15K high performance disks for active data. We will see some TLC (triple level) flash appearing on the market in the near future in multi-TB capacities at prices that will be competitive to high capacity enterprise disks. They will be offered as a tier of storage for less writes combined with an MLC flash tier for the more intensive write applications. I think it will be a few years before we see the 16 TB 3D NAND flash drive available for enterprise class storage arrays. Competitive pricing versus MLC/TLC could be a challenge for adoption of 3D NAND.
So what does this mean for hard disks and hybrid flash arrays? AFA, all flash array vendors, Violin and Pure Storage (who filed for an IPO last month) are proclaiming that disk is dead. While everyone agreed that flash was displacing high performance disks for active data, many customer and analysts still saw the need for disks for overall TCO and long-term durability. Many also saw the need for hybrid flash arrays that had the enterprise scalability, availability, and management that are lacking in midrange AFAs. In fact we have customers who have virtualized AFA vendors like Pure, Violin, and XtremeIO behind our G1000 and VSP in order to get the availability and management functions that they are lacking.
Hitachi Data Systems storage arrays can be configured as all flash arrays with SSDs or with our purpose built Flash Module Drives (FMD), which has a current capacity of 1.6 and 3.2 TB. We have customers who have all flash arrays as well as hybrid arrays. In a following post I will talk about these two use cases. As far as whether the HDD is dead, I agree that the high performance HDD is being replaced by flash, but the high capacity HDD still has some legs in terms of TCO, especially if we go to erasure coding which enables the use of commodity disks at a much lower cost than enterprise disks. (See how we use erasure coding and commodity disks in our HCP S10: The Advantages and Benefits of Erasure Coding in the HCP S10).
Our customers have huge investments in disk systems that they will need to transition to new storage technologies. A hybrid storage array with the ability to virtualize and transition storage from current disk or flash technologies to whatever new storage technology that we have in the future must be a key part of your overall storage strategy. It is inevitable that Non-Volatile Memory will eventually replace disks, but the current variations of Flash technology, may not be that technology. While flash is beginning to replace high performance disks, it is premature to say that the hard disk is dead.
What do you think?