The future of flash memory: speed and price reductions
Smartphones, tablets, USB flash drives, digital cameras, video recorders and SSDs all rely on fast, rugged, nonvolatile NAND flash memory. Gartner predicts that yearly NAND sales will reach 200 petabytes by 2016, up from just 50 petabytes in 2012. Much of the memory will go into the SSDs for servers and desktop PCs, as well as laptops, tablets, and other mobile devices.
The difference in speed between an SSD and a fast hard drive is obvious even to the untrained eye. In PCWorld's December 2012 roundup, the fastest consumer SSDs read at almost 500 MBps and wrote at over 600 MBps. Meanwhile, a high-end, 10,000-rpm hard drive averaged around 200 MBps reading and writing. That's a three-fold advantage in performance, and enterprise-grade SSDs are even faster.
With NAND's limited number of write cycles rendered moot by advanced operating system support and techniques such as wear-leveling, SSD technology's only constraints are cost and capacity.
Current consumer SSDs top out at 512GB, and each gigabyte costs 10 times what a gigabyte of a platter-based hard drive costs. The prices of both storage technologies are dropping at similar rates (20 to 25 percent a year), and this 10-to-1 ratio should remain about the same for at least the next five years according to numbers Gartner shared with us.
Traditionally, NAND has suffered a limited number of Program/Erase (P/E) cycles. Each cycle reflects an erasure and write on a memory cell. Fortunately, though, the P/E problem is being rendered moot by advanced operating system support and firmware techniques such as wear-leveling, so now SSD technology's only constraints are cost and capacity.
Currently, faster and more expensive 100,000 P/E, SLC (single-layer cell, 1-bit) NAND is used in enterprise-level drives, with 10,000 P/E, MLC (multilevel cell, 2-bit) NAND used in consumer-grade drives. This is likely to remain the pattern for the near future, especially as firmware improves practical longevity even further—and it may have to, because as dies shrink, so do the P/E ratings of NAND. The promise of TLC (triple-level cell, 3-bit) NAND is often bandied about in SSD discussions, but without substantial improvement in lifespan (currently about 1000 P/E cycles) and performance (it's slow), you'll only see it in smaller mobile devices over the next few years.
The biggest challenge for NAND is physics. Greg Wong of Forward Insights says that there is most likely only one die shrink left after Intel's current 20-nanometer process. After that, decreasing endurance and performance will force a halt to further reductions in die size for flash memory. When asked by how much Intel might be able to reduce the die, Wong says he was unsure.
But as many real estate–challenged metropolitan areas prove, when you run out of horizontal space, you can always go vertical. To this end, 3D or layered NAND chips will likely provide the increase in capacity that's required for NAND to continue growing in capacity.
With a large, well-established manufacturing infrastructure, it's unlikely that we'll see serious inroads by competing nonvolatile memory technologies anytime soon. But trust us—various technologies are vying to offer an alternative, including HP's Memristor (memory resistor),Toshiba's ReRAM (resistive memory), and FeTRAM (Ferro-Electric Transistor memory), to name just a few.
Next Up: Optical: On the way out?