This article is more than 1 year old
Life after server-side flash: What comes next?
Dave Cartwright essays six post-NAND competitors
Flash suffers from a steadily shorter working life, slower access speed and shorter working life the smaller the actual cells the NAND become.
It’s generally reckoned that the scaling wall will be hit attempting to shrink cell size beyond 16nm. At that point, to continue increasing non-volatile memory capacity without increasing chip size will mean using new technologies such as 3D NAND, Resistive RAM (ReRAM), Phase-Change Memory or some other technology. In this article, we survey various post-NAND technologies.
What's happening inside that 'Void If Opened' metal casing?
We're just getting used to the idea of shelling out our hard-earned corporate cash on flash (a.k.a. Solid State Drive, or SSD) storage for our data centre SANs. The fact that it's only now starting to catch on demonstrates that the technology is in its infancy relative to traditional storage media … but that doesn't stop us wondering what's coming next.
And this is understandable, because as with any fledgling concept this spanking new technologies has some pretty hard-core limitations. The obvious one is price, but the worst is the fact that the number of times you can re-write a flash cell is not only finite but actually quite small.
It's understandable, therefore, to be asking what's coming next. It means those of us who are holding fire for now are able to watch what's coming over the horizon so we have an idea of when to jump into Flash.
It's clear that today's NAND technology has a limited shelf life, and that it's something of a stepping stone on the path to something with a little more longevity. As one often finds, there's a whole load of competition out there in the non-volatile memory market, the potentially promising ones being:
- Phase Change Memory (“PCM”): Works by changing the state of chalcogenide glass between its crystalline and amorphous states. No, I didn't do A-level Chemistry either, but it basically means changing a specific substance between a state where the atoms are unaligned and random (and high resistance) and one where they're neatly aligned (and low resistance). IBM likes this technology, and it's both fast and long-life.
- Magneto-Resistive RAM (“MRAM”): Been around since the 1990s, so it's well understood (in fact if you remember Core Memory from the days of prehistoric computing, some of the principles are the same). Promising in principle but expensive to make, so it's yet to find favour in a commercial sense.
- 3D NAND: You'll read later about PMR, which in a spinning disk world provides increased data density on a medium by aligning the elements vertically rather than horizontally. Well, 3D NAND is something similar but in a flash world. Toshiba and Samsung are the names in this field.
- Resistive RAM (“RRAM” or “ReRAM”): terabyte-scale storage on a single chip that's already at the prototype stage and is being touted by the likes of Panasonic (which started shipping in 2013) and Crossbar (which has pretty much bet the farm on RRAM being the next big thing).
- Spin-Transfer Torque RAM (“STT-RAM”): A relative of MRAM which is showing promise and has been used to a limited extent in commercial products already. A company by the name of Grandis was the leading light, and guess what: they were acquired and are now a part of Samsung's memory division.
- Graphene memory: Still very much in the lab, but of great interest if you're a Samsung boffin – it has the potential to be both transparent and flexible, and they're already trying out different sizes of “graphene quantum dot” and discovering their relative performance. Don't expect to see it in PC World any time soon though.
Of the above, the current race is generally accepted to be between RRAM and 3D NAND. If one had to guess, the obvious evolution would be from today's NAND drives, via 3D NAND and then onward to RRAM and eventually PCM – and given that it has finger in pretty much every pie, Samsung will be there at the front of the field simply because most of the competitors in the race have a Samsung logo on their shirts! In my experience, though, IBM its likely to be in there too with its PCM offering. The company has a habit of not being first to market in a particular sector but then coming up on the rails and making good: though the early bird catches the worm, the second mouse gets the cheese.
Alongside the development of faster and faster storage technologies, however, one has to remember that the kit doing the actual reading and writing has to be able to keep up. The idiotically fast storage tech will sit in the server, not the SAN, so it can benefit from direct CPU and local memory connectivity.
After all, why bother having something fast in the SAN's disk array when the interconnect can't keep up? Of course, the interconnect guys will continue their race to keep up – we can now do 500Mbyte/sec transfers on 6Gbit/s S-ATA, for instance, and the only way is up for all of these connectivity mechanisms.
But hang on a moment ...
We now have an idea, then, about some of the technologies that might (or might not) be the successor of today's NAND-based flash storage technologies. There's one question remaining to be answered, though: do we actually care?
In the early 2000s, I don't recall many of my colleagues eagerly anticipating the advent of Perpendicular Magnetic Recording (“PMR”) technology, which became a commercial product around 2005 and brought a big hike to the usefulness of spinning disks. Similarly the development of Heat Assisted Magnetic Recording (“HAMR”), helium-filled drives for better cooling (= more platters for the same heat output), and so on. No: all we care about is that as time goes on, the chaps in white coats at the disk vendors' R&D labs continue to do clever stuff that lets us store more in less space, with a longer lifetime and with lower power consumption.
Before I started writing this series of features on flash storage my comprehension of the topic was merely above average – not least because a slightly disturbing proportion of my family works or has worked in the storage industry. I now know a lot more than I did.
But does it matter, in my day job as an IT ops manager, whether I understand the chemistry and nano-physics of how my storage works. Nah: just keep making it faster and better so I can improve my services as I expand, upgrade and replace. ®