The less the voltage states, naturally the lower the latency. But not on the level of Optane which was built from the ground up for low latency. Optane is more like static DRAM whereas this and Samsung Z-NAND are highly optimized NAND.
sure there is. what is labeled SRAM is, in fact, refresh required DRAM, just more complicated and faster. true SRAM is made with true flip-flops and doesn't require refresh. I don't even know whether anyone even makes it any longer.
You are confused. SRAM and DRAM are completely different technologies. SRAM uses transistors which can hold state as long as power is supplied to the chip. SRAM is present in every modern CPU in the form of caches. DRAM uses capacitors which leak charge and must be refreshed periodically. There's no such thing as "static DRAM", because DRAM by definition is "dynamic", as in "not static". I suggest Wikipedia for more info.
There is no escaping the similarities in architecture between X-Point and RAM, not withstanding the fundamental difference in architecture allowing X-Point to hold a cell state without power not unlike NAND.
But alas, I should have put static in quotations or just said alternatives to MRAM, TRAM, etc. Everyone is taking their own approach to scaling up NVRAM.
Yeah, they're similar in that they both use small things to store data... AFAIK optane is pretty different and seems to be a sort of rram, which is quite different from dram
It should be possible with firmware to operate the standard TLC NAND as MLC or SLC. This would result in 3X increase in cost per bit at SLC and have improved latency performance. But for really great performance, we would like a custom NAND with smaller page size, perhaps 1-2KB, versus the standard 16KB
This is not a new standard, it's an new technology. No company shares technology, that's what they make money with. They will use the same NVMe Standard as other Flashdrive manufacturers. Don't see your problems.
I don’t think that NOR write times are spectacular. It’s not going to get you to the holy grail of non-volatile memory that has the speed of DRAM and the cost and scaling of NAND. Optane is the only thing remotely close. MRAM, ReRAM are still niche and don’t seem to be scaling economically.
Yeah NOR writes are still in blocks like NAND. Reads are much much faster and simpler though. You read more often than you write too. This isn't necessarily about beating optane in speed, but about offering better performance per dollar at similar storage sizes.
I just feel like with what WDC just announced here with a higher performance, higher cost NAND, that they could have done better on both the speed and cost by just doing NOR.
Doing NOR probably is quite a bit of redesign work though. You are playing around with your most fundamental storage structures. If you mess it up and all your drives have an issue that is a lot of product recalls. WDC seems to have taken a safer route.
I was under the impression that Martin Fink pretty much burned his hope on the Memristor while he was still working for HP: Actually I'd love to hear why that isn't moving forward when they were so convinced that they'd have product in 2016.
Today he seems to argue that the technology doesn't matter as long as it works and the price point can be made to match the performance you need. And that IT purchasers show worry more about their architecture and software debt than the base technology.
While he may be right, I still long to satisfy my curiosity.
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.
Facebook
Twitter
RSS
22 Comments
Back to Article
Bigos - Tuesday, March 12, 2019 - link
"The memory will cost 10 times less than DRAM, but 20 times more than 3D NAND"That's 3 times more than 3D NAND, not 20 times.
FreckledTrout - Tuesday, March 12, 2019 - link
Agree. The 10x cost of LLF Nand multiplied by 3 = 30x of 3d Nand.PeachNCream - Tuesday, March 12, 2019 - link
"...using 1 bit-per-cell and 2 bit-per-cell architectures."That sounds suspiciously like SLC and MLC. Hmmmm....
Samus - Tuesday, March 12, 2019 - link
The less the voltage states, naturally the lower the latency. But not on the level of Optane which was built from the ground up for low latency. Optane is more like static DRAM whereas this and Samsung Z-NAND are highly optimized NAND.p1esk - Tuesday, March 12, 2019 - link
There's no such thing as "static DRAM" :)FunBunny2 - Tuesday, March 12, 2019 - link
" There's no such thing as "static DRAM" :) "sure there is. what is labeled SRAM is, in fact, refresh required DRAM, just more complicated and faster. true SRAM is made with true flip-flops and doesn't require refresh. I don't even know whether anyone even makes it any longer.
p1esk - Tuesday, March 12, 2019 - link
You are confused. SRAM and DRAM are completely different technologies. SRAM uses transistors which can hold state as long as power is supplied to the chip. SRAM is present in every modern CPU in the form of caches. DRAM uses capacitors which leak charge and must be refreshed periodically. There's no such thing as "static DRAM", because DRAM by definition is "dynamic", as in "not static". I suggest Wikipedia for more info.Samus - Tuesday, March 12, 2019 - link
There is "static" DRAM: X-Point :PThere is no escaping the similarities in architecture between X-Point and RAM, not withstanding the fundamental difference in architecture allowing X-Point to hold a cell state without power not unlike NAND.
But alas, I should have put static in quotations or just said alternatives to MRAM, TRAM, etc. Everyone is taking their own approach to scaling up NVRAM.
Dr. Swag - Thursday, March 14, 2019 - link
Yeah, they're similar in that they both use small things to store data... AFAIK optane is pretty different and seems to be a sort of rram, which is quite different from dramshabby - Tuesday, March 12, 2019 - link
Wd is going back... back to the future!PeachNCream - Thursday, March 14, 2019 - link
Do they need to hit 88 mph to do that?jchang6 - Tuesday, March 12, 2019 - link
It should be possible with firmware to operate the standard TLC NAND as MLC or SLC. This would result in 3X increase in cost per bit at SLC and have improved latency performance. But for really great performance, we would like a custom NAND with smaller page size, perhaps 1-2KB, versus the standard 16KBMrSpadge - Wednesday, March 13, 2019 - link
You are describing Samsungs Z-NAND ;)coburn_c - Tuesday, March 12, 2019 - link
Great so... another proprietary tech that won't see proper [driver] support or adoption and will never see competitive pricing due to ... propriety.coburn_c - Tuesday, March 12, 2019 - link
To further my indifference: xkcd has a comic about standards.... to which to WDC contributes: There are 14 competing standards, let's make our own.KarlKastor - Thursday, March 14, 2019 - link
This is not a new standard, it's an new technology. No company shares technology, that's what they make money with.They will use the same NVMe Standard as other Flashdrive manufacturers.
Don't see your problems.
MrSpadge - Wednesday, March 13, 2019 - link
It won't need driver support per se, just a suitable controller. Software wise anything built for Optane, Z-NAND etc. should equally work with this.Khenglish - Tuesday, March 12, 2019 - link
Why don't flash makers just make NOR memory if they want more speed? Or is that what this actually is?NOR memory should be several times faster than NAND while having roughly half the density, at least according to my quick mental math.
flgt - Tuesday, March 12, 2019 - link
I don’t think that NOR write times are spectacular. It’s not going to get you to the holy grail of non-volatile memory that has the speed of DRAM and the cost and scaling of NAND. Optane is the only thing remotely close. MRAM, ReRAM are still niche and don’t seem to be scaling economically.Khenglish - Wednesday, March 13, 2019 - link
Yeah NOR writes are still in blocks like NAND. Reads are much much faster and simpler though. You read more often than you write too. This isn't necessarily about beating optane in speed, but about offering better performance per dollar at similar storage sizes.I just feel like with what WDC just announced here with a higher performance, higher cost NAND, that they could have done better on both the speed and cost by just doing NOR.
Doing NOR probably is quite a bit of redesign work though. You are playing around with your most fundamental storage structures. If you mess it up and all your drives have an issue that is a lot of product recalls. WDC seems to have taken a safer route.
abufrejoval - Thursday, March 14, 2019 - link
I was under the impression that Martin Fink pretty much burned his hope on the Memristor while he was still working for HP: Actually I'd love to hear why that isn't moving forward when they were so convinced that they'd have product in 2016.Today he seems to argue that the technology doesn't matter as long as it works and the price point can be made to match the performance you need. And that IT purchasers show worry more about their architecture and software debt than the base technology.
While he may be right, I still long to satisfy my curiosity.
gfkBill - Thursday, March 14, 2019 - link
"The manufacturer admits (LLF) is very fast because it is tailored for performance"Wow, that's quite the admission.