While far from a new product announcement, a prototype SSD using a type of resistance RAM (ReRAM) - Phase Change Memory or PCM - as an alternative to NAND flash is welcome proof of industry progress. Because the flash used in today's SSDs has serious problems with reliability and scaling.
Flash in the pan
NAND flash used in SSDs today has a limited future for several reasons. As I noted two years ago in the The dismal science of flash:
NAND flash traps electrons in quantum wells. As feature sizes shrink, so do the quantum wells and, more critically, the number of trapped electrons that store the information. That makes it harder to preserve and read the stored information.
Thus the many contortions - wear-leveling, predictive ECC, DRAM buffers, over-provisioning - that SSDs employ to deliver performance and reasonable life. But the physics of NAND flash are brutal. Storage needs something better.
That something better looks to be ReRAM, which is currently implemented in a variety of technologies, including PCM. The invention of PCM predates flash, but it has been a long, slow slog. Indeed, despite multiple product announcements it isn't clear who, if anyone, is shipping product with embedded PCM.
Researchers at IBM Almaden used a Micron 64GiB PCM SSD based on 45nm technology for system level testing and simulation. They modeled a hypothetical storage device that used PCM SSDs, flash SSDs and hard drives, using an instrumented version of Linux to get nano-second accuracy of I/O latency.
This is important because chip-level performance stats may not reflect what is possible at a storage device level. For example, flash chips have very slow write cycles, but through the use of parallelism, smart garbage collection and DRAM buffers, that can be hidden in most apps.
PCM writes are non-trivial: a bit of material must be heated and allowed to cool for the material to reach either the amorphous or crystalline state that stores the data. This is a costly process, but once written the data is highly stable, unlike flash.
The modeled storage used PCM SSD for hot read-intensive data, flash SSD for hot write-intensive data, and disk drives for cold data. The simulations used storage traces from actual retail, bank and telco applications.
The researchers conclude:
Based on the results above, we observe that PCM can increase IOPS/$ value by 12% (bank) to 66% (telecommunication company) even assuming that PCM is 4× more expensive than flash.
The Storage Bits take
I hope to learn more at next week's File and Storage Technology (FAST '14) conference in Silicon Valley, where the not-yet-publicly-available paper Evaluating Phase Change Memory for Enterprise Storage Systems: A Study of Caching and Tiering Approaches by Hyojun Kim, Sangeetha Seshadri, Clement L. Dickey and Lawrence Chiu of IBM Almaden Research will be formally presented.
But on the face of it this is not a persuasive case for PCM. The performance improvements described and the economic benefits derived fall short of those required, in my experience, to drive commercial adoption.
But the real point is that the industry is actively looking beyond today's flash to tomorrow's underlying storage technology - a technology that, like PCM, should be much more robust than NAND flash. Flash has the considerable economic advantage that comes with enormous volume, but that came only after decades of research and development.
While PCM may not win the race for the next-gen flash, this research gives a more-nuanced approach to how a more robust - and more costly - ReRAM could be incorporated into significant commercial products. It is this kind of decades-long development that creates the next "overnight" sensation in technology as well as life.
Comments welcome, of course. An example of the creative incorporation of a new storage technology can be seen in the architecture of Nimble Storage, described in a video white paper I did for them almost three years ago.