Bleak Future for SSDs, According to New Research

We are quite used to the fact that technology improves over time, but as many have noted this isn’t always the case with SSDs. In this part of the tech industry, shrinking transistors and higher densities translate to lower costs but also less reliability and in the end, performance. Three researchers from the Department of Computer Science and Engineering at the University of California, San Diego have taken on the unappreciative task of putting this fact on paper. It is called “The Bleak Future of NAND Flash Memory” and is available here. The authors, Laura M. Grupp, John D. Davis and Steven Swanson, point out that if the current downwards trend in flash density continues, all of the advantages of the storage medium will similarly deteriorate.

SSDs will continue to improve by some metrics (notably density and cost per bit), but everything else about them is poised to get worse.

Not surprisingly, this is particularly true for multi-level cell NAND and the even less inspiring triple-level cell NAND.

The research team tested 45 different NAND flash chips based on 72 to 25 nm technology from six different vendors. The test revealed that the write speed of the pages in a flash block part suffered from dramatic but predictable variations in latency. Error rates also varied significantly when the NAND circuits became worn out. Naturally, SLC came out on top in the tests while MLC was worse and especially TLC NAND worse still.


Then they extrapolated heir findings until 2024 when NAND flash is estimated to have come down to 6.5 nanometers. By then the latencies will likely have doubled for MLC flash and increased by a factor of 2.5 for TLC. Error rates are also increased to to a level three times higher than today, which sounds like a very careful estimate.

Site founder and storage enthusiast.

  1. Great article, it properly depicts the realism of the current situation in NAND storage media sector. Nowadays there are only some improvements in costs per gigabyte written only in SLC NAND sector – the reason why i have to purchase expensive SLC SSDs and RAM media for some of my database processing tasks.
    No wonder that PR assholes pay for fake articles to “save” the situation (their interests).

  2. Thanks vt. I hope this paper gets a lot of attention and that it puts puts some perspective on the current tradition of marketing of flaws as advantages (unlikely, but who knows).

  3. Pretty much says what I’ve suspected for some time. SSDs have been nothing but trouble for me. I think they do pretty well in low-write-low-rewrite situations, but in my estimate they are not that great in normal high write frequency applications like as main drive for computers. The failure rate for me is far to high for me to consider them as superior to HDDs for my applications, at least. For HDD, I only usually experience failures if I treat them very poorly. For SSDs you can treat them as poorly as you like and they will be fine. If you actually use them, thats what will kill them! haha

  4. So, it looks like transfer rates will level off at around 1GB/sec for MLC NAND, maybe a little less. Hard drives are never going to catch up. It’ll have to do until we come up with something better.

  5. […] A smaller production process leads to more energy-efficient chips that use less energy and produce less heat to provide the same level of performance (more transistors in the same physical space). By and large it’s a major improvement in most types of circuitry (excluding, perhaps, SSDs). […]

    Leave a reply