I am not smart enough to understand all this but can i condlude 1)they will only license the technology not make it themself 2)It scales less well in size than Intel Optane but is as fast RAM and is persistence?
1: It would appear so. Not a bad thing either, foundry business is extremely expensive.
2: Quite the contrary, should scale really well due to the way each cell is constructed. Past that it's just a matter of implementing the front-end (which is no different than doing so for DRAM for example).
Remember that CNT's are discreet elements, and as such, have a fixed size (though small). These are likely MW tubes so are not nearly as small as SW tubes. Also, I imagine that to create and open current paths (plural is important) involves multiple tubes and paths. As such, statistics will play a role in the resistance of the on state, so with scaling, one might assume that fewer paths will be available and enabled, and as such, more variability in the on (and possibly off) state resistance. This will not only impact sensing of the on/off state at lower supply voltages, but also close the door on the possibility of multi-bit storage in a single cell. I'm not convinced that scaling is a slam dunk at all.
In my experience, flashy PPT themes are not taken seriously at technical conferences. People assume it's either marketing material with no technical depth or it's there to distract you from some other lack of content.
Is the "300C limit" temperature? I think that's beyond automotive grade, but they seem impressive.
I'd assume nanotube production would be an issue, but probably not. Scaling in size vs. Optane isn't all that much of an issue: expect the road to <5nm to be *slow*.
I'm curious why persistence is in such: I'd assume that [data corrupting] software and even hardware issues should be vastly more common than power issues. And it shouldn't be that hard to make a motherboard that can boot without clearing ram for the software issues.
But I want RAM cheaper than DRAM. Anybody know what the odds of this company being bought by Intel and this tech being quietly smothered? Cheap memory (i.e. DRAM, one of Intel's biggest inventions) helped microcomputers kill the minicomputer and mainframe, and even cheaper DRAM threatens Wintel (and Lintel) as once you need "real" amounts of memory, you can't justify putting a lower performance chip than high end x86 (ok, sometimes that means AMD, but you know what I mean).
If the technology is as good as it sounds I don't think there is any real chance of Intel being able to buy it. Natero being the first with this tech has the potential of taking over a large size of the memory market. And being a fabless company receiving royalties from every other memory or chip manufacturer It would basically be tens of billions in 5 years and probably more than Intel itself in 10 years.
If memory serves (pun intended), this company was founded 17 years go. I have to believe that they have pitched this to (and been evaluated by) many companies both large and small.
If production could scale to the point where the chip cost is properly cheap is there any reason why you couldnt load a system up with NRAM and run the system in a ram disk? for the business machines I roll out (8GB/256GB) if a system had 4 64GB NRAM modules and in UEFI I reserved 8GB for system and the rest for storage that would make for a very small for factor, same principle with laptops...
I hope it isnt suppressed or purposely ignored, we wait and see who creates a fab and starts producing it at scale, perhaps the Chinese will spin up a couple of fabs for internal use they can afford to and if the benefits are to be believed would reap the benefits immediately. Western markets are hyper competitive battlegrounds I dont see NRAM breaking into that market certainly not in anything other than niche areas, definitely not on the desktop :( Europe would be the big unknown, I dont follow nm scale fabrication to any great degree but i remember reading about AMD having a fab or two in continental europe presumably there are other companies who mine chips there as well?
NRAM is based on the positions of carbon nanotubes -- touching or not. How sharp an external force would change these positions?
Imagined downside: could NRAM be corrupted by an earthquake? (so not permanent for servers in the ring-of-fire? But it survived rocket lift-off?) A dropped phone? A vehicle crash? (so not for 'black-box' recorders in computer controlled vehicles?) Imagined upside: is it reliably erased by impact? (so maybe dud smart munitions are rendered dumb?)
This looks even better than MRAM in terms of density and 3D scaling. And you can build HBM out of this, potentially with higher stacks, as the energy density will be lower, no refresh. I wonder if they can mix with logic as well as MRAM to give almost NV-logic with caches or even register files built out of CNT? Can't wait to see this happen, too bad all my current stuff is DDR3 (DRAM recycling ;-)
One big question is the cost of the CNT layer (and there are multiple layers based in the presentation). In any commodity semiconductor segment, cost is the driver. You might have the best widget it town, but if it's expensive, then it better offer some amazing benefits over the existing technology, and even then, it might end up being a niche play since the current product base is heavily biased towards the existing technologies (NAND, DRAM, SRAM).
This thing is far too good to be true, surely something is missing? Even if it made DRAM persistent would have been a huge breakthrough, but it is talking about increase in capacity to something ( I assume ) like Die stacked SLC NAND, lower power, with ECC while being faster.
A quick look on WIkipedia shows they've been trying to sell these for well over a decade. "2-3 weeks ago Fujitsu announced they were taking CNT to mass market." Actually, the deal with Fujitsu goes back to 2016. It is strange that in the presentation, Nantero revealed that a 16 Gb device hasn't been built yet. Back in 2016, the plan was to be manufacturing 256 Gb devices this year. https://www.theregister.co.uk/2016/08/31/nram_dev_... I wish them luck. Surely, the skyrocketing price of DRAM last year bolstered NRAM's value proposition.
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.
19 Comments
Back to Article
plopke - Tuesday, August 21, 2018 - link
I am not smart enough to understand all this but can i condlude1)they will only license the technology not make it themself
2)It scales less well in size than Intel Optane but is as fast RAM and is persistence?
Still i loved this article xD!
Arnulf - Wednesday, August 22, 2018 - link
1: It would appear so. Not a bad thing either, foundry business is extremely expensive.2: Quite the contrary, should scale really well due to the way each cell is constructed. Past that it's just a matter of implementing the front-end (which is no different than doing so for DRAM for example).
zapshere01 - Monday, December 31, 2018 - link
Remember that CNT's are discreet elements, and as such, have a fixed size (though small). These are likely MW tubes so are not nearly as small as SW tubes. Also, I imagine that to create and open current paths (plural is important) involves multiple tubes and paths. As such, statistics will play a role in the resistance of the on state, so with scaling, one might assume that fewer paths will be available and enabled, and as such, more variability in the on (and possibly off) state resistance. This will not only impact sensing of the on/off state at lower supply voltages, but also close the door on the possibility of multi-bit storage in a single cell. I'm not convinced that scaling is a slam dunk at all.Qwertilot - Tuesday, August 21, 2018 - link
Fun :) Potentially a lot of it if it can manufacture at any sort of speed.Can’t help but think they musn’t be trying hard enough when they claim ‘unlimited’ endurance!
Gunbuster - Tuesday, August 21, 2018 - link
Is it 2001? Those people need to drop some coin on a professional website and PowerPoint deck make over.RandomUsername3245 - Tuesday, August 21, 2018 - link
In my experience, flashy PPT themes are not taken seriously at technical conferences. People assume it's either marketing material with no technical depth or it's there to distract you from some other lack of content.p.s. very cool technology; hope to see it soon
Ian Cutress - Tuesday, August 21, 2018 - link
Correct. Simple slides are the best. It's for engineers, not marketing.wumpus - Tuesday, August 21, 2018 - link
Is the "300C limit" temperature? I think that's beyond automotive grade, but they seem impressive.I'd assume nanotube production would be an issue, but probably not. Scaling in size vs. Optane isn't all that much of an issue: expect the road to <5nm to be *slow*.
I'm curious why persistence is in such: I'd assume that [data corrupting] software and even hardware issues should be vastly more common than power issues. And it shouldn't be that hard to make a motherboard that can boot without clearing ram for the software issues.
But I want RAM cheaper than DRAM. Anybody know what the odds of this company being bought by Intel and this tech being quietly smothered? Cheap memory (i.e. DRAM, one of Intel's biggest inventions) helped microcomputers kill the minicomputer and mainframe, and even cheaper DRAM threatens Wintel (and Lintel) as once you need "real" amounts of memory, you can't justify putting a lower performance chip than high end x86 (ok, sometimes that means AMD, but you know what I mean).
Ian Cutress - Tuesday, August 21, 2018 - link
300C limit means at 300 Celsius, data retention goes down from 12000 years to 300 years.Dr. Swag - Wednesday, August 22, 2018 - link
Looks like optane might have some solid competition! :Dsgeocla - Friday, August 24, 2018 - link
If the technology is as good as it sounds I don't think there is any real chance of Intel being able to buy it. Natero being the first with this tech has the potential of taking over a large size of the memory market. And being a fabless company receiving royalties from every other memory or chip manufacturer It would basically be tens of billions in 5 years and probably more than Intel itself in 10 years.zapshere01 - Monday, December 31, 2018 - link
If memory serves (pun intended), this company was founded 17 years go. I have to believe that they have pitched this to (and been evaluated by) many companies both large and small.WatcherCK - Tuesday, August 21, 2018 - link
If production could scale to the point where the chip cost is properly cheap is there any reason why you couldnt load a system up with NRAM and run the system in a ram disk? for the business machines I roll out (8GB/256GB) if a system had 4 64GB NRAM modules and in UEFI I reserved 8GB for system and the rest for storage that would make for a very small for factor, same principle with laptops...I hope it isnt suppressed or purposely ignored, we wait and see who creates a fab and starts producing it at scale, perhaps the Chinese will spin up a couple of fabs for internal use they can afford to and if the benefits are to be believed would reap the benefits immediately. Western markets are hyper competitive battlegrounds I dont see NRAM breaking into that market certainly not in anything other than niche areas, definitely not on the desktop :( Europe would be the big unknown, I dont follow nm scale fabrication to any great degree but i remember reading about AMD having a fab or two in continental europe presumably there are other companies who mine chips there as well?
Gc - Thursday, August 23, 2018 - link
NRAM is based on the positions of carbon nanotubes -- touching or not. How sharp an external force would change these positions?Imagined downside: could NRAM be corrupted by an earthquake? (so not permanent for servers in the ring-of-fire? But it survived rocket lift-off?) A dropped phone? A vehicle crash? (so not for 'black-box' recorders in computer controlled vehicles?) Imagined upside: is it reliably erased by impact? (so maybe dud smart munitions are rendered dumb?)
abufrejoval - Friday, August 24, 2018 - link
This looks even better than MRAM in terms of density and 3D scaling.And you can build HBM out of this, potentially with higher stacks, as the energy density will be lower, no refresh.
I wonder if they can mix with logic as well as MRAM to give almost NV-logic with caches or even register files built out of CNT?
Can't wait to see this happen, too bad all my current stuff is DDR3 (DRAM recycling ;-)
zapshere01 - Monday, December 31, 2018 - link
One big question is the cost of the CNT layer (and there are multiple layers based in the presentation). In any commodity semiconductor segment, cost is the driver. You might have the best widget it town, but if it's expensive, then it better offer some amazing benefits over the existing technology, and even then, it might end up being a niche play since the current product base is heavily biased towards the existing technologies (NAND, DRAM, SRAM).iwod - Friday, August 24, 2018 - link
This thing is far too good to be true, surely something is missing? Even if it made DRAM persistent would have been a huge breakthrough, but it is talking about increase in capacity to something ( I assume ) like Die stacked SLC NAND, lower power, with ECC while being faster.alex_d - Monday, August 27, 2018 - link
A quick look on WIkipedia shows they've been trying to sell these for well over a decade. "2-3 weeks ago Fujitsu announced they were taking CNT to mass market." Actually, the deal with Fujitsu goes back to 2016. It is strange that in the presentation, Nantero revealed that a 16 Gb device hasn't been built yet. Back in 2016, the plan was to be manufacturing 256 Gb devices this year. https://www.theregister.co.uk/2016/08/31/nram_dev_... I wish them luck. Surely, the skyrocketing price of DRAM last year bolstered NRAM's value proposition.Anymoore - Friday, September 7, 2018 - link
I see a big problem that their SET and RESET voltages are so close.