Samsung rings death knell for disk, gears up for QLC flash production

Death knell for disk?

QLC flash will have 33% greater capacity per cell than TLC, hence will be at best 25% cheaper per GB - assuming it costs exactly the same per cell to produce, and doesn't require any additional over-provisioning.

It will also have lower write endurance, lower long-term retention, and higher read error rates.

That makes it pretty unattractive for many applications. Throw-away consumer electronics are probably the most likely place this will appear, unfortunately.

Rocky Horror?

Interesting for the "write once-read many" applications but the Rocky Horror reference?

Death knell my arse

Any way I look at it, it is yet another step up on the ladder going down into the land of diminishing returns.

It delivers less in terms of capacity gain than what you have to over-provision in order to hit the actual MTBF and write cycle numbers even for consumer hardware.

So it is a death knell to rusty spinners some other time.

Isn't it the case that floating gates in multiple level flash are very unreliable leading to corruption over time from drift? So, what we need is better archive storage disks.

QLC vs HDD

QLC will provide a 25% lower price best case when it is released into mass production which will be Sept 2018 best case.

So today the cost of SSDs are 7-8x the cost per GB of HDDs. HDD prices dropped in the last year and SSD prices went up. So QLC will help make it so that SSDs are only 5x more expensive than HDDs

Someday SSDs will be half of all the GBs sold for storage. Today is not that day (HDD out-ships SSD GBs >7:1. 2020 is not that day either. 2025 seems good. HDD will be here for a while.

Some thoughts on QLC vs. TLC

Going from SLC to MLC provided a 100% increase in capacity over SLC.

Going from MLC to TLC provided a 50% increase in capacity over MLC.

In both of these cases the NAND industry was able provide adequate write wear (P/E cycles, DWPD) to allow the NAND media to replace general purpose storage.

Also, each new increase in bit per cell density cannot be implemented on the latest semiconductor device fabrication process. Higher bits per cell do not work as well on higher density semiconductors. As a result, the higher bit per cell technology is more often implemented on lower density semiconductors, reducing the price per capacity advantages of the denser bits per cell. The same is true about 3D NAND vs. Planar NAND--more complicated NAND manufacturing is less suited for the highest bit per cell density.

Going from TLC to QLC will only provide a 33% increase in capacity over TLC. However, this increase may be offset by using older, less dense semiconductor density, and lower layer density 3D NAND, compared to state of the art TLC capability. Add to this a significant reduction in P/E cycles and DWPD, and the price per capacity advantage may be minimal--for now. Over time everything will improve and QLC will become more mainstream.

The real advantage may be a combination of mature QLC backing drives with storage class memory caching. This is the likely solution for both hyperconverged and external storage arrays.