Flash memory for storage is a vast improvement on the venerable old hard drive, but even with microsecond access times, flash is slow compared to dynamic RAM.
Flash has the restriction of being block addressable, increasing the minimum transfer to 4 KB and requiring the use of the operating system file stack, thereby adding a lot of CPU overhead. This has resulted in a search for an alternative solid-state offering, and several candidates have surfaced, including phase change memory technology.
Phase change memory (PCM) looks like it might be the most viable alternative to solid-state. This memory operates by changing the conductivity of a cell from low to high (the phase change). This can be achieved by applying a voltage to modify the cell state. This change is very rapid; in the ten nanoseconds range.
Because it is much closer in speed to dynamic RAM (DRAM), phase change memory technology is ideal for both NVDIMMs and NVM Express (NVMe) SSDs. Intel and Micron have both announced versions of 3D XPoint, a product reputed to be PCM-based. While this has seen some delays and performance issues, it looks to be faster than flash in its first generation, and perhaps much faster in its follow-on versions.
PCM is byte-addressable, at least in the NVDIMM versions, allowing direct small transfers to take place. Only writing and reading a word or so compared with a full block makes it possible to read and write directly from CPU instructions that register to memory operations, which are very fast indeed.
Apart from the speed advantage, phase change memory technology is much more durable than flash, so the number of daily writes isn't an issue. There is some discussion about feature size and die capacity being better than flash, too, but the devil is in the details, and the first products will lag well behind 3D NAND drives in total capacity.
Best guesses are that Intel and Micron plan to price between DRAM and flash, placing a premium on the approach, at least in the first couple of years. In addition, phase change memory technology has other hurdles to overcome.
3D NAND is just hitting its stride, so we can expect a dramatic reduction in cost per bit, which will pressure the newbie PCM products. Likewise, capacity per die is expanding rapidly with the 3D approach, and 100 TB 2.5-inch drives are on the 2018 roadmap.
Any 3D XPoint products will need byte addressability to be competitive under these conditions, though Intel is also postulating a cluster model with byte addressability for all its 3D XPoint memory across the cluster using an NVMe fabric.
IBM touts triple-level cell phase change memory innovation
PCM is just one possible NAND replacement
Shortage in NAND flash supply has wide-reaching effect
Dig Deeper on Flash memory and storage
Related Q&A from Jim O'Reilly
OpenStack Cinder has added a revert-to-snapshot function, enabling enterprises to recover from corrupted data sets. However, if the feature falls ... Continue Reading
Don't let backup data encryption fall through the cracks. When encrypting backups, key management and compression are just two of the best practices ... Continue Reading
Replacing all your HDDs with SSDs won't solve the storage issues associated with in-memory databases. Look to hyper-convergence and NVDIMMs instead. Continue Reading