Western Digital to Enter Mass Production of 162-Layer NAND This Year

[TheLostSwede]

Last week, Micron revealed its NAND flash plans and so did Western Digital during its investors day. The different NAND flash manufacturers use similar, but far from identical ways to manufacture their NAND flash and WD is working together with Kioxia and the two are using a process called bit column stacked or BiCS NAND. This year the two companies will move to its sixth generation of BiCS NAND which is stacking 162-layers of NAND flash. Most of their competitors are already at 176-layer stacks, although WD and Kioxia still deliver the same capacity NAND chips as its competitors. WD claims that their die size is smaller than the competition, which could be an advantage due to more dies per wafer.

In its presentation, WD listed a single wafer capacity of 100 TB, which is an increase from around 70 TB per wafer in 2020. Although it's not something we can verify, WD claims to have the world's best charge trap cell, which the company claims leads to higher performance compared to its competitors. The slide provided by WD claims a 20 MB/s lead of the competition at 60 MB/s vs. 40 MB/s. Although Western Digital and Kioxia claim to have to invest less to increase their capacity, it would appear that they're behind the competition when it comes to the amount of layers they can stack. The two are only planning what they call BiCS+ at over 200-layers for 2024, which is well over a year behind Micron. The roadmap stretches all the way to 2032, when we should apparently be seeing 500-layer NAND flash, assuming everything goes according to plan. The first BiCS+ NAND will be for datacenter products and WD claims that their first product should see a 60 percent increase in transfer speeds, a 15 percent increase in program bandwidth and a 55 percent increase in bit growth per wafer over the current BiCS6 NAND.



View at TechPowerUp Main Site | Source

 

[AnarchoPrimitiv]

Just wanted to thank theLostSwede for their articles and covering the topics that they do.

 

[TheLostSwede]

Just wanted to thank theLostSwede for their articles and covering the topics that they do.

Do you know something I don't? Am I getting fired?

 

[Chrispy_]

Number of layers is a marketing numbers race that has no impact on actual cost/TB or performance, right?

The quality, performance, and density of each layer is just as important as the number of layers - in the same way that counting the number of cores alone isn't a valid strategy in comparing CPUs. Two phones ago (2015) I had an 8-core CPU. That doesn't mean it was faster than a 6-core Ryzen 5 5600X....

 

[TheLostSwede]

Number of layers is a marketing numbers race that has no impact on actual cost/TB or performance, right?

Well, it should have an impact on performance, as seen by the 176-layer chips, but it's only a small bump rather than a big jump in performance.
It's meant to make it cheaper to make larger capacity chips, but again, if that translates to cheaper consumer products or not, is a different matter.

The quality, performance, and density of each layer is just as important as the number of layers - in the same way that counting the number of cores alone isn't a valid strategy in comparing CPUs. Two phones ago (2015) I had an 8-core CPU. That doesn't mean it was faster than a 6-core Ryzen 5 5600X....

Not a fair comparison. However, without knowing the density per layer, comparing layers is sort of pointless, as someone could have twice the density at half the layers, although that's currently not the case. That said, there are density difference between the various NAND flash makers, which makes direct comparisons a bit tricky. The performance difference is also hard to compare, as it comes down a lot to the actual controller in the case of an SSD and how well it works with a certain type of flash.

 

[DeathtoGnomes]

Do you know something I don't? Am I getting fired?

No, just marking the target of who we can point fingers at.

That roadmap could have been done better, broken down into which quarter or half of the expected year.

 

[thegnome]

Oh god PLC, I really hope high performance NVME ssd's don't move towards QLC and PLC like Samsung moving their pro lineup to TLC. Focus on increasing capacity and not let degradation get any worse than it currently is.

 

[looniam]

woah!

lets keep the violence to a minimum please - until bets gets placed.

 

[Chrispy_]

Oh god PLC, I really hope high performance NVME ssd's don't move towards QLC and PLC like Samsung moving their pro lineup to TLC. Focus on increasing capacity and not let degradation get any worse than it currently is.

I don't think PLC will be used outside of the datacenter once it arrives. The number of write cycles will be very low and the raw write performance is going to be pitiful. It'll basically replace high-bandwidth WORM media. QLC is already pushing the limits of cost-effectiveness because each jump in bits-per-cell yields dimishing returns on capacity but scales controller/firmware complexity up exponentially.

PLC compared to QLC is twice the number of voltage levels (which requires ~4x as much programming precision and long-term stability) and it only yields 25% more capacity. Reductions in write performance and endurance are also side effects that will likely make PLC unfeasible for small capacities and general-purpose workloads. Controller/firmware complexities in handling 32 voltage levels likely mean that the break-even point where capacity gains outweigh the complexity costs probably won't happen until 4-8TB of NAND is used. That probably won't have any consumer market appeal and these will likely be limited (initially) to datacenter markets and possibly some specific enterprise niche applications.

Even QLC is struggling to compete with TLC at the low end. It only really makes a meaningful cost difference once you get to the largest capacity QLC drives and compare them against the equivalent TLC drives.