Do you like Flash-based solid state drives? Get your fill now, because in 2025 they may be more fit for the junkpile than for your datacenter.
OK, that may be a bit of a contextless exaggeration, but according to a semi-controversial research paper entitled, "The Bleak Future of NAND Flash Memory," delivered at the 2012 Usenix Conference on File and Storage Technologies earlier this month, given the rate at which SSDs are increasing in density, their quality of performance will fall to the point at which they are potentially less cost-effective than standard HDDs after 2024.
Two of the researchers, Laura M. Grupp and Steven Swanson (who presented a paper a year ago on SSD data remanence), are a graduate student and a professor, respectively, at the University of California, San Diego. A third, John D. Davis, works for Microsoft.
In empirical studies of 45 flash chips, ranging in sizes from 72nm to 25nm, from 6 different manufacturers, the researchers identified a major flaw in NAND-based SSDs -- that as SSD density increases, SSD performance and lifetime dramatically decrease. Latency and bit error rates will skyrocket. Bandwidth, throughput, energy efficiency, and program/erase endurance will plummet.
Industry analysts widely agree that by 2024, NAND flash cells will be only 6.5nm in size. Based on their empirical data, the researchers have predicted that 2024 will represent a crux in NAND-based SSD technology. This is because the way multi-level cell (MLC) and triple-level cell (TLC) Flash chips apply high voltages to move electrons around strains the insulator (known as the "gate oxide") such that eventually, the chip is no longer able to store a charge at all -- which can lead to electron leaks and data errors. The problem is compounded by the thinner cell walls and narrower distributions required as chips get both smaller and denser.
"It's not going to be viable to go past 6.5nm," said Grupp at the conference. "2024 is the end."
It's not all doom and gloom; SSDs will technically continue to offer superior results over HDDs in some areas. For instance, SSDs will still substantially outperform HDDs on throughput (even with halved input/output rates), according to the paper, with the slowest SSD configuration considered breaking 32,000 IOPS -- compared to HDD technology topping out at 200 IOPS.
Nonetheless, the outperformance in some areas simply might not be cost-efficient in a dozen years. The researchers write:
With current trends, our SSDs could be up to 34x larger, but the latency will be 1.7x worse for reads and 2.6x worse for writes. This will reduce the write latency advantage that SSDs offer relative to disk from 8.3x (vs. a 7 ms disk access) to just 3.2x. Depending on the application, this reduced improvement may not justify the higher cost of SSDs.
Don't go writing SSD technology's obituary yet, though. Stephen Gallagher, a software engineer and technology blogger, predicts that some enterprise datacenters may increasingly mitigate the risks and costs of SSD adoption through limited deployment. "I think we'll see a lot of hybrid setups that involve using SSDs being set up as caching drives for much larger traditional hard drives," Gallagher told Enterprise Efficiency. "This way you gain the performance benefits of an SSD for whichever subset of your data is being used 'right now,' while retaining the larger portion of your data on more traditional platter drives and... tape backups."
In any case, as things tend to do in these cases, something better is bound to come along. Robin Harris of ZDNet predicts that Resistive RAM (a.k.a. "ReRAM") "looks to be a good bet" to replace the Flash-based technology of current SSDs in computers within the next 10 years. Other suitable technologies could also step into the picture in the meantime.
The cost efficiency of NAND Flash, however, appears to have a finite lifespan. Plan accordingly.