Hopefully this ends up desirably priced. It'd be great to see SSD storage overtake Disk for mainstream and value desktop, laptop storage by three years from now.
Having entered into mass production mode, doesn't that mean significant portion of Validation & testing is completed already? (forgive me, noob question!)
Exactly. Mass production is a different scale than what the article describes, at least to me. If this isn't on the market in six months, someone is using a weird definition for mass production.
I agree. Since these chips should be cheaper to produce (smaller process), while holding a significantly greater amount of data, and having at least 2x more reliability, the only reason to delay release of these chips would probably be to get rid of the current stock of SSDs. If the info holds true, I believe current SSDs will render obsolete in price, performance, and capacity. This should also be really bad news for conventional spinning drives.
All depends on pricing though. I don't believe Samsung is stupid enough to sell those for cheap, at least not till they make a good margin on them if they get market lead.
Just because you're producing the NAND (or V-NAND) doesn't mean it's going into fully functional products, let along SSDs. Once the NAND is being mass produced, someone needs to integrate that into a useful product. We might see simple flash drives, SD/Micro SD type devices, etc. at first followed by the integration of the product into a fully functional SSD. That last one is a big step, though, as it needs controller logic, testing, validation, debugging, etc. which is why I think this could be more like 1-2 years out.
Also, 10nm is very cutting edge, unless Samsung is somehow using a different definition than others. IMFT is on 19nm I think, and Intel's CPU stuff is still at 22nm with 14nm coming next year followed by 10nm probably in three years. NAND is easier to manufacture (well, test and validate at least) is my understanding, which is why most first wafers off a new assembly line are NAND wafers. When did the first 1xnm NAND start mass production vs. when did it show up in the first consumer SSDs? I'd say a year at least passed.
Anyway, we'll see what happens. I for one would love to pay twice the cost of HDD storage to get an SSD of similar capacity, where right now it's more like 10x the cost for the same storage capacity. I'm looking forward to the day where I can run a 1TB SSD and not have it cost as much as the CPU, GPU, and motherboard combined.
I think it's just a common misinterpretation for people to read 10 nm class devices as 10 nm devices. 10 nm class is anything from 10-19 nm, and when they say 10 nm class devices, they typically mean the process is actually 19 nm and they're just trying to make it seem more incredible and ahead of the competition than it really is. Woohoo, marketing lingo.
Thanks for that...that was highly useful. There was no "blind rewriting" involved, thanks. It was a minor error that has been corrected, based on their use of "10nm-class" vs. "10nm".
Samsung has several technologies in the development phase. 3D will help, but there are also some more disruptive technologies in the pipeline. For one thing, Samsung has a history of licensing RAM technology from smaller companies such as Symetrix, from where they got the FeRAM technology they are implementing in small very low power MCUs. Symetric also has a memristor type nonvolatile called CeRAM "Correlated Electron RAM" that looks very promising. Unlike other phase-change efforts, CeRAM does not involve inducing transition metal filaments between electrodes, so does not require electroforming and uses a homogenous material. It will be cheap and easy to produce. The advantage is that endurance is orders of magnitude greater than flash, on current is lower, and Set/Reset is far faster (close to DRAM rates). The phase change itself takes only femtoseconds. I think first testing runs are 64M chips at the 22 nm node and already being made.
Only a fool believes new process nodes are actually cheaper to produce than the last. To process shrink you need more Photolithography tools and newer/more expensive ones at that. Each process shrink requires and and more layers printed on immersion (193 wet) and less on 193 dry, 248 and i-line tools. New NXT 193 immersion tools go for like $50mil+ a pop and a fab needs several to run latest gen products.
Maybe at a per Gb level prices will drop but per wafer absolutely not. Looking at per Gb cost doesn't do the cost to end consumers justice as chip density goes up but the chip's Gb in SSDs also go up. So the sq/mm of a wafer used in an SSD will stay much more steady (8 die in a chip and 8 chips on the circuit board) all the while per wafer cost isn't dropping.
Just like Samsung likes to market '10nm class' hoping some clueless people think it's half pitch is actually 10nm Samsung also does the same thing with 'mass production.' News releases like this are well, well ahead of the chips making it into products, lease of all SSDs. SSDs probably have some of the higher stringent qualifications and qualification is many months after wafer starts begin in a fab.
Initial measurements show this being like 20-30% die size reduction. If Samsung wants to market these chips closer to SLC due to the CFT technology the end client SSDs may not see any price reduction at all.
Samsung however is on a progressive kick "SSDs for everyone" so pricing could be very very competitive in an effort to sell mass units at lower costs, rather than fewer at a premium price point, tbh they are the only ones who have the numbers that can either justify or damn certain mainstream marketing technics due to the success of the 830/840 lines.
No wonder Samsung stepped out of the hdd market awhile ago. With the added storage and longevity, looks like they're seeking to make sdds dominate mainstream quicker than most expected ... hmm ...
By the way, does this, in any significant way, affect thickness of a slim smartphone or is the stacking so neglible that it won't matter? Enlightenment, please.
My understanding is that the stacking only adds a very small amount to the z-height of one chip, and it should be easy enough to accommodate. I believe some of the upcoming CPUs (Broadwell or Skylake) are expected to do something similar with stacked DRAM vs. stacked NAND. If each layer of NAND adds even 1000nm to the total height of the chip (which would be pretty extreme I would think -- 50-100nm would seem a more likely target), 24 layers would increase the height of the chip by 24um, or 0.024mm.
If true IMHO this could be one of the most disruptive inventions in the tech sector lately. The only problem I see is a lot higher price compared to normal NAND and hence less of an actual benefit. And then my media files are perfectly fine on a slow ass green drive. So if this new SSD would only cost double, well it's still a lot more given one does not need the performance.
To be fair, Crossbar showed much more than HP ever did - a taped out chip produced at 25 nm and being significantly smaller than NAND at this node. That's something companies can actually use and test - and if it's as good as they someone will start building devices out of it. At least OCZ, they're always up fancy new stuff, despite taking huge risks.
Well, Western bought STEC, which had bought Unity, which was developing its version of memristor. HP started it all, IIRC. There's a bunch of candidates to replace NAND/NOR flash. Have been for more than a decade. Stacking may just make those replacements replaceable.
For posteriority, it looks like a year since these have gone into "mass production" and now these are available for what looks like competitive pricing (to grab the early adopters) before literally lowering the boom on HDD.
We’ve updated our terms. By continuing to use the site and/or by logging into your account, you agree to the Site’s updated Terms of Use and Privacy Policy.
33 Comments
Back to Article
lmcd - Tuesday, August 6, 2013 - link
Hopefully this ends up desirably priced. It'd be great to see SSD storage overtake Disk for mainstream and value desktop, laptop storage by three years from now.Vigneshj - Tuesday, August 6, 2013 - link
Having entered into mass production mode, doesn't that mean significant portion of Validation & testing is completed already? (forgive me, noob question!)coder543 - Tuesday, August 6, 2013 - link
Exactly. Mass production is a different scale than what the article describes, at least to me. If this isn't on the market in six months, someone is using a weird definition for mass production.lilmoe - Tuesday, August 6, 2013 - link
I agree. Since these chips should be cheaper to produce (smaller process), while holding a significantly greater amount of data, and having at least 2x more reliability, the only reason to delay release of these chips would probably be to get rid of the current stock of SSDs.If the info holds true, I believe current SSDs will render obsolete in price, performance, and capacity. This should also be really bad news for conventional spinning drives.
All depends on pricing though. I don't believe Samsung is stupid enough to sell those for cheap, at least not till they make a good margin on them if they get market lead.
JarredWalton - Tuesday, August 6, 2013 - link
Just because you're producing the NAND (or V-NAND) doesn't mean it's going into fully functional products, let along SSDs. Once the NAND is being mass produced, someone needs to integrate that into a useful product. We might see simple flash drives, SD/Micro SD type devices, etc. at first followed by the integration of the product into a fully functional SSD. That last one is a big step, though, as it needs controller logic, testing, validation, debugging, etc. which is why I think this could be more like 1-2 years out.Also, 10nm is very cutting edge, unless Samsung is somehow using a different definition than others. IMFT is on 19nm I think, and Intel's CPU stuff is still at 22nm with 14nm coming next year followed by 10nm probably in three years. NAND is easier to manufacture (well, test and validate at least) is my understanding, which is why most first wafers off a new assembly line are NAND wafers. When did the first 1xnm NAND start mass production vs. when did it show up in the first consumer SSDs? I'd say a year at least passed.
Anyway, we'll see what happens. I for one would love to pay twice the cost of HDD storage to get an SSD of similar capacity, where right now it's more like 10x the cost for the same storage capacity. I'm looking forward to the day where I can run a 1TB SSD and not have it cost as much as the CPU, GPU, and motherboard combined.
p1esk - Tuesday, August 6, 2013 - link
Samsung has not disclosed the process size. Where did 10nm info came from?http://www.samsung.com/global/business/semiconduct...
JarredWalton - Tuesday, August 6, 2013 - link
From Ars Technica...looks like they misinterpreted something in the document. Editing...garadante - Tuesday, August 6, 2013 - link
I think it's just a common misinterpretation for people to read 10 nm class devices as 10 nm devices. 10 nm class is anything from 10-19 nm, and when they say 10 nm class devices, they typically mean the process is actually 19 nm and they're just trying to make it seem more incredible and ahead of the competition than it really is. Woohoo, marketing lingo.Mondozai - Wednesday, August 7, 2013 - link
"From Ars Technica...looks like they misinterpreted something in the document. Editing..."The danger of blindly re-writing somene else's news post.
JarredWalton - Wednesday, August 7, 2013 - link
Thanks for that...that was highly useful. There was no "blind rewriting" involved, thanks. It was a minor error that has been corrected, based on their use of "10nm-class" vs. "10nm".Jaybus - Thursday, August 8, 2013 - link
Samsung has several technologies in the development phase. 3D will help, but there are also some more disruptive technologies in the pipeline. For one thing, Samsung has a history of licensing RAM technology from smaller companies such as Symetrix, from where they got the FeRAM technology they are implementing in small very low power MCUs. Symetric also has a memristor type nonvolatile called CeRAM "Correlated Electron RAM" that looks very promising. Unlike other phase-change efforts, CeRAM does not involve inducing transition metal filaments between electrodes, so does not require electroforming and uses a homogenous material. It will be cheap and easy to produce. The advantage is that endurance is orders of magnitude greater than flash, on current is lower, and Set/Reset is far faster (close to DRAM rates). The phase change itself takes only femtoseconds. I think first testing runs are 64M chips at the 22 nm node and already being made.lilmoe - Wednesday, August 14, 2013 - link
Guess they were fully functional when announced....http://www.engadget.com/2013/08/13/samsung-unveils...
mmonnin03 - Wednesday, August 7, 2013 - link
Only a fool believes new process nodes are actually cheaper to produce than the last. To process shrink you need more Photolithography tools and newer/more expensive ones at that. Each process shrink requires and and more layers printed on immersion (193 wet) and less on 193 dry, 248 and i-line tools. New NXT 193 immersion tools go for like $50mil+ a pop and a fab needs several to run latest gen products.Maybe at a per Gb level prices will drop but per wafer absolutely not. Looking at per Gb cost doesn't do the cost to end consumers justice as chip density goes up but the chip's Gb in SSDs also go up. So the sq/mm of a wafer used in an SSD will stay much more steady (8 die in a chip and 8 chips on the circuit board) all the while per wafer cost isn't dropping.
Just like Samsung likes to market '10nm class' hoping some clueless people think it's half pitch is actually 10nm Samsung also does the same thing with 'mass production.' News releases like this are well, well ahead of the chips making it into products, lease of all SSDs. SSDs probably have some of the higher stringent qualifications and qualification is many months after wafer starts begin in a fab.
Initial measurements show this being like 20-30% die size reduction. If Samsung wants to market these chips closer to SLC due to the CFT technology the end client SSDs may not see any price reduction at all.
1Angelreloaded - Thursday, August 15, 2013 - link
Samsung however is on a progressive kick "SSDs for everyone" so pricing could be very very competitive in an effort to sell mass units at lower costs, rather than fewer at a premium price point, tbh they are the only ones who have the numbers that can either justify or damn certain mainstream marketing technics due to the success of the 830/840 lines.gnx - Tuesday, August 6, 2013 - link
No wonder Samsung stepped out of the hdd market awhile ago. With the added storage and longevity, looks like they're seeking to make sdds dominate mainstream quicker than most expected ... hmm ...MDme - Tuesday, August 6, 2013 - link
Meanwhile...another company "innovates" by patenting a shape of a rectangle phone.flyingpants1 - Tuesday, August 6, 2013 - link
Excuse me, only rectangles with rounded corners are patented. The rectangle with 90 degree angles is still fair game.flyingpants1 - Tuesday, August 6, 2013 - link
And they will price-fix the hell out of them, just like RAM and SSDs and probably everything else.klmccaughey - Wednesday, August 7, 2013 - link
Absolutely. No way are we getting a decent priced product until other manufacturers have it and they are on compete.LeftSide - Tuesday, August 6, 2013 - link
We could see phones and tablets with 1tb of storage in the next few years.Mondozai - Wednesday, August 7, 2013 - link
Tablets, possibly. Phones? Heh.Mondozai - Wednesday, August 7, 2013 - link
By the way, does this, in any significant way, affect thickness of a slim smartphone or is the stacking so neglible that it won't matter? Enlightenment, please.JarredWalton - Wednesday, August 7, 2013 - link
My understanding is that the stacking only adds a very small amount to the z-height of one chip, and it should be easy enough to accommodate. I believe some of the upcoming CPUs (Broadwell or Skylake) are expected to do something similar with stacked DRAM vs. stacked NAND. If each layer of NAND adds even 1000nm to the total height of the chip (which would be pretty extreme I would think -- 50-100nm would seem a more likely target), 24 layers would increase the height of the chip by 24um, or 0.024mm.Jaybus - Thursday, August 8, 2013 - link
Negligible and won't affect it.beginner99 - Wednesday, August 7, 2013 - link
If true IMHO this could be one of the most disruptive inventions in the tech sector lately. The only problem I see is a lot higher price compared to normal NAND and hence less of an actual benefit. And then my media files are perfectly fine on a slow ass green drive. So if this new SSD would only cost double, well it's still a lot more given one does not need the performance.althaz - Wednesday, August 7, 2013 - link
I wouldn't bother for my media drives at double the price, but I sure as hell would for my games drive!dali71 - Wednesday, August 7, 2013 - link
RRAM will laugh at NAND's puny capacity and speed:http://www.engadget.com/2013/08/06/crossbar-rram/
klmccaughey - Wednesday, August 7, 2013 - link
Been hearing about this for years now and it's still pie in the sky - when I see a finished product at a reasonable price from RRAM I will believe it.MrSpadge - Wednesday, August 7, 2013 - link
To be fair, Crossbar showed much more than HP ever did - a taped out chip produced at 25 nm and being significantly smaller than NAND at this node. That's something companies can actually use and test - and if it's as good as they someone will start building devices out of it. At least OCZ, they're always up fancy new stuff, despite taking huge risks.klmccaughey - Wednesday, August 7, 2013 - link
Also, I remember at the time it had terrible endurance. Hopefully this has been solved.glugglug - Wednesday, August 7, 2013 - link
How long before this gets obsoleted by memristors?FunBunny2 - Wednesday, August 7, 2013 - link
Well, Western bought STEC, which had bought Unity, which was developing its version of memristor. HP started it all, IIRC. There's a bunch of candidates to replace NAND/NOR flash. Have been for more than a decade. Stacking may just make those replacements replaceable.bellasys - Tuesday, September 9, 2014 - link
For posteriority, it looks like a year since these have gone into "mass production" and now these are available for what looks like competitive pricing (to grab the early adopters) before literally lowering the boom on HDD.