REGISTER | LOGIN
Breaking News
Blog

Limitations of 3D NAND Scaling

NO RATINGS
View Comments: Newest First | Oldest First | Threaded View
Page 1 / 2   >   >>
Sang-Yun
User Rank
Blogger
Re: ...once NAND replaces HDDs
Sang-Yun   10/17/2017 8:06:45 PM
It is good to hear that HDD is still evolving. I think 3D NAND will be unable to close the gap between SSD and HDD price gap. Even though notebook prefers to use SSD over HDD, other large storage media should be sensitive to $/GB.  

realjjj
User Rank
CEO
Re: ...once NAND replaces HDDs
realjjj   10/17/2017 2:48:40 PM
NO RATINGS
Corporations claim a lot of things. That graph factors in pricing for high perf, high margin SSDs that don't target the same market. As costs decline and bit supply grows, NAND has to attack the nearline market too. To make it worse, new memories take share in high end and NAND has to focus more and more on the lower end. HDD makers lose scale too, maybe 3-4 times lower revenue by 2023. That hits margins and won't leave much for OPEX either so i have huge doubts about their HDD price scaling claims too.

wgt0823
User Rank
Author
Re: ...once NAND replaces HDDs
wgt0823   10/17/2017 1:02:49 PM
NO RATINGS
Have you seen the recent WDC MAMR announcement?  They claim with this technolgogy HDD will maintain its 10+ cost advantage through the 20's.   http://bit.ly/2hN7Go8  

Sang-Yun
User Rank
Blogger
Re: Block staircases
Sang-Yun   10/12/2017 10:14:34 AM
NO RATINGS
Good point! In a given cell area, as 3D NAND adds more layers, WL staircase needs more spaces in the given cell area and just small area left for cells. For example. in case of 32-layer,  WL staircase stretches out 20um from the edge of cell array. So, for 128-layer, it would be 80um. WL staircase drastically increases manufacturing cost, decreases wafer throughput, and reduces effective cell efficiency. 3D NAND should rethink its process architecture.  



resistion
User Rank
Author
Block staircases
resistion   10/12/2017 2:54:12 AM
NO RATINGS
There are thousands of blocks in the entire 3D NAND. Each block, I guess, has its own staircase. Then, when more layers are added, the staircase should grow outward. I think this would be a non-negligible impact.

Sang-Yun
User Rank
Blogger
Re: ...once NAND replaces HDDs
Sang-Yun   10/11/2017 11:55:59 PM
NO RATINGS
realjjj, thank you very much for sharing the info and your inputs. Let's have a lunch together in a couple of years later and see how 3D NAND has evolved for the time being. Though, I still believe that 4Tb and 16Tb 3D NANDs won't be available in the market. If I am wrong, lunch will be on me. It will be exciting to see the future of NAND in few years.

realjjj
User Rank
CEO
Re: ...once NAND replaces HDDs
realjjj   10/11/2017 10:56:07 PM
NO RATINGS
Let's put it this way, you know that 2D NAND prices were at some 20 cents per GB in the first half of last year. You admit that 3D is cheaper to produce now, You know that they can etch 64L and it's unreasonable to expect no progress with HAR etch and thinner layers. We also know that 4 bits per cell, logic under the array and horizontal scaling can help quite a bit.

2018-2019 should be mostly about 96L and QLC, 2019-2020 they would need at least 128L(2x64) and some of the other tricks mentioned. 96Lx2 could come next - 96L in one is pretty certain. 128L is maybe questionable today but there are a few years left to figure that out so I got little doubt about it being achievable. I am assuming both scaling and string stacking being used but maybe some push to 128L in one faster than this so progress could be even better.

To go beyond 128L, maybe a next-gen hard mask, 80:1 etch, III-V materials. ALD. Around 2023 they likely need to slow down bit growth too as demand growth slows so maybe they push margins up and there is a bit of a pause in price declines even if costs do keep scaling. Likely that's achieved by moving more capacity to new NVMs, fewer bits and higher ASPs.

NAND will keep scaling for a while but that doesn't mean that it's invincible and better value solutions that scale even better can't show up. Its vulnerability is that it can't scale much at all on the horizontal and it's already at 4 bits per cell.

Sang-Yun
User Rank
Blogger
Re: ...once NAND replaces HDDs
Sang-Yun   10/11/2017 5:29:26 PM
NO RATINGS
realjjj, semiconductor companies have roadmap for the next 4-5 generations regardless technical feability. For example, planar NAND had roadmap down to 10nm (it stop at around 15nm) and planar DRAM (it stop at around 20nm) also has roadmap down to low 10nm. Though, there has been no progress for the last 2-3 years on device scaling of these products. Question regarding 3D NAND roadmap will be how efficiently NAND vendors could build next generation 3D NANDs. I think bit growth rate of 3D NAND (unlike planar NAND) is somewhat ambiguous becasue manufacturing cost of 3D NAND is much higher than planar NAND. So, high BGR of 3D NAND does not imply low cost, I think.  

realjjj
User Rank
CEO
Re: ...once NAND replaces HDDs
realjjj   10/11/2017 3:01:56 AM
Never said indefinitely, provided a clear time frame.

NAND will not scale forever but it's not quite there yet. NAND makers already have a clear path for the next 4-5 generation and they'll figure out a few more generations. Next year bit growth will be about 50% as 64L keeps ramping- output is still relatively low now. Right now array efficiency is at 70-80% ,depending on die size so there is room for some improvements there with logic under the array. Then there will likely be some horizontal scaling and ofc there is QLC. All that without any more layers can lead to up to a 50% bit cost reduction. More layers in one is not easy but it will go above 64L and there is string stacking too.  Unlike classic scaling a new generation of 3D comes every 12-15 months and not every 2 years, it's important to remember that.

NAND scaling might end before 2030 but that's ok as demand growth will slow down  and new NVMs come from behind.

As for HDDs, they are dead , NAND costs will scale enough to kill HDDs, there is no doubt there.

DRAM is a problem as scaling is minimal and solutions are needed ASAP but NAND has some life left in it.

TanjB
User Rank
Author
Re: ...once NAND replaces HDDs
TanjB   10/11/2017 1:58:33 AM
NO RATINGS
The logic of 40% bit growth assumes buyers and infinite technology. The counter example is DRAM where technology hit a wall and production growth shrivelled to about 15% per generation, with a couple of years between.  New fabs were not built, instead the industry consolidated during the shrink of PC market while neither mobile nor data center consumed as many bits as were disappearing.  Then demand caught up and overshot, with prices trippling in the ast year.  The next fab comes in 2019 - fun times. So you have to look not just at the graph, but what is behind it.  As Sang-yun says, the technology has some possibility to stall just like DRAM did (and for similar reasons).  Meanwhile as even you see, the logic of infinite increase capacity in mobile devices is not at all sure, and the mobile market is cooling as the high end has slow rate of replacement and the growth is mostly low or mid end replacements trading up.  While the dat center market follows the logic I outlined. A lot of the data growth is in junk data - multimedia social, plus data exhaust for everything from jet engines to home thermometers and surveillance.  There is no magic, these data types make roughly $0 per GB and so there is a desperate need to figure out how to store it as close to zero as possible.  HDD and tape are way ahead of Flash on price, and HDD is generally preferred because when you put stuff on tape you need strong constraints on how data will be used. So I would stand your numbers on their head and say the market and technology will not sustain a 40% growth of Flash indefinitely, there will be a correction.  Unless Sang-yun or someone else pulls a rabbit out of the hat and can make a terabit chip for $2.

Page 1 / 2   >   >>
Like Us on Facebook
EE Times on Twitter
EE Times Twitter Feed