Kioxia Researchers Demo Hepta-Level Cell NAND Flash

Kioxia Researchers Demo Hepta-Level Cell NAND Flash(tomshardware.com)

21 points by coobird 3 years ago | 17 comments

NKosmatos 3 years ago |

Is it only me who doesn’t like 7 bits per cell? I prefer endurance and reliability (SLC, MLC, TLC…even QLC) over capacity (PLC, HLC, 7LC).

phkahler 3 years ago | |

>> Is it only me who doesn’t like 7 bits per cell? I prefer endurance and reliability

No, it's not just you, and judging by how strongly I feel about it, it's not just the two of us.

When I find a flash drive that's been sitting for a few years and I can still read it I always feel a sense of relief. I feel like everything we've got today is too advanced to maintain the slightest glitch.

GauntletWizard 3 years ago | | |

One of the big advantages of early flash storage was read durability, so that even when a drive was past it's useful write lifetime you could still read from it. That's not the case anymore, between wear-leveling technologies and higher bit densities.

I would love large write-once flash disks that would last for 20+ years in storage. There's a ton of archival data out there.

fmajid 3 years ago |

I shudder to think of the P/E cycle endurance. QLC is already abysmal at 100-1,000 or so so if the SLC-MLC-TLC-QLC trend holds, it will be in the order of 10 or so in the first generation.

vlovich123 3 years ago | |

I think those numbers are only if you’re doing sub 64k writes no? The data sheets I had seen have QLC numbers very competitive if you could maintain 64k writes.

rektide 3 years ago |

So each cell is now a x86 cache line wide, that's wild.

I have a hard time imagining this is useful or good for many apications, as I just expect endurance to be awful. But given how much data warehousing we do, there is definitely some potential for this, as a post-nearline storage.

This also seems like something zones storage would be so good for, to help reduce write-amplification concerns. Alas there are still zero drives one can go slap down a credit card & purchase.

gruez 3 years ago | |

> So each cell is now a x86 cache line wide, that's wild.

128 bits =/= 128 levels

You're mixing up the units. 128 bits require 2^128 levels to represent. 128 levels can only represent 7 bits. The only way the two are related is the number "128", albeit with different units.

rektide 3 years ago | | |

Pardon duh. 6 bits is 64 levels (cache width). Very different from 64 bits of course.

gaudat 3 years ago |

The caveat is requiring liquid nitrogen.

so-and-so 3 years ago | |

And a Pro model with helium cooling :)

ksec 3 years ago |

Doesn't really change anything. Current 4TB SSD is about $200. And that is close to selling at cost or lost. Even if they achieve it the price per TB will likely be the same just offering them better margins.

7speter 3 years ago | |

I imagine because this NAND has more layers, its going to be have even less endurance than and write speeds than QLC? It may eventually be cheap but, whew boy whats the use case if the loss in performance as layers increase scales linearly?

tenebrisalietum 3 years ago | | |

I guess overprovisioning? Maybe your 16TB 7LC NAND will have 48TB of actual space.

sitkack 3 years ago | | |

Use it like tape, write once and archive it in a freezer.

kube-system 3 years ago | |

Why would price per TB be the same? The whole point of cramming more levels into the same cells is that you can store more data in the same number of flash cells.

ksec 3 years ago | | |

Because you get Cost efficiency from newer tech, ( cost reduction with adding layers are not linear ) but current SSD price are at zero or negative margin. ( There is an oversupply or under demand of NAND ) Which means should this tech arrive they are only going to gain back their reasonable margin. Hence the 4TB might be priced the same.