PLC flash: A solid state candidate for analytics and archives?

When QLC flash storage drives arrived in the market last year, it looked like the storage media makers had pushed the manufacturing process as far as it could go.

But now, Intel has signalled that a further iteration of NAND solid state storage – PLC, or penta-level cell – may be on its way.

That could mean flash drives that start to push towards 20TB, and while they will be cheaper to run than equivalent-sized spinning disk media, they will be limited in terms of I/O performance, endurance and therefore use cases.

Let’s recap on what a move to PLC actually means.

Successive generations of flash drive have been developed by building on the number of states – of voltage switching between 0 and 1 – per flash cell.

We started with the now-obsolete single-level cell (SLC), with one 0 or 1 per cell, and added a further possible cell state to that with multi-level cell (MLC), which meant four possible switching outcomes (00, 10, 01, 11).

Then triple-level cell (TLC) added another for seven possible voltage switches, then quad-level cell (QLC) provided 16 states and 15 possible switches. QLC drives are currently available up to about 8TB in capacity.

Now PLC – if it can be productised – will almost double that and allow for 32 states and 31 switching outcomes.

Each of these generations’ increases in cell switching capability can be understood as adding to the capacity of drives in which they are used. So, for example, an MLC drive provides 4x that of an SLC drive, and so PLC drives will nearly double the capacity of QLC.

But every increase in cell state capacity has come with a cost in terms of durability. As each generation has packed more electrical operations into a smaller space, the longevity of cells has been eroded.

That is partly because, with multiple voltages necessary for multiple levels of switching, longer times are required to set and detect the charge. The other consequence of this is longer read and write times.

Manufacturers will try to mitigate these issues by over-provisioning capacity. Western Digital, for example, is working on “zoning”, where data of certain types is restricted to certain areas on the flash die, with the aim of making data as near to write-once-read-many as possible.

High-performance enterprise flash does not adhere to this principle, and flash drive maintenance sees data erased and moved as part of processes aimed at optimising performance.

But the big question is: is PLC flash achievable as a product and what will its characteristics and limitations mean for possible use cases?

Well, QLC is mainly aimed at use cases that see little change in data stored on devices that use it, and PLC will follow suit.

That means lots of random I/O and rapidly changing data are out. Instead, PLC, like QLC, will be best suited to reads, and preferably longer sequential reads.

Micron QLC, for example, advertised 32x more I/O than SATA spinning disk and 2x on throughput. We can expect those to be reduced for PLC.

So, in terms of use cases, that could mean some machine learning workloads and archive usage.