The article appears to be a rehash of this Nantero press release:[1] That says "Signaling a new era in memory, Nantero’s NRAM has already been installed in multiple production fabs..."
This technology has been around for a while. Lockheed Martin built a 4MB NRAM back in 2008, and sent it up with a memory tester on a Space Shuttle flight in 2009.[2] That was done in conjunction with Nantero; they had an arrangement for government products. The technology is inherently rad-hard, so it's useful for space operations.
This technology has been two years from volume production since 2008.[3] It's not clear why it hasn't reached production yet. It clearly works. It may just cost too much.
FWIW, this predates the memristor. There are several types of RAM all trying to replace NAND and DRAM. It seems likely that one will win, with a couple others for niche uses (IIRC MRAM is already favorable for some aerospace applications due to its resistance to switching from cosmic rays).
Memristors are turning into yet another falso hope from HP. They never created a product that is as good as flash, nevermind this newfangled nanotube memory.
It would be better phrased "will not suffer write cycle limitations", but I think it communicates decently to a general audience what's going on.
If you asked me what the write cycle limitation of the registers in my haswell chip was, I'd shrug and say effectively unlimited. Which isn't far off what they said.
Well, to be fair, if that's true, that's an improvement over 10^2 - 10^3 write cycles is what current NAND-flashes can manage. 10^12 is enough for it not to ever matter.
No, that's a low enough number that wear-leveling logic will be needed, as with flash memory. This is a fast device, approaching DRAM speeds. A program could easily bang on a cell 10^12 times.
You write to your RAM, or even more pertinently, your SSD, on every CPU clock cycle?!
How many times have you written to the same cell on your SSD? Current NAND chips have a write cycle limitation of somewhere around 10^6. Apparently a Taiwanese firm got it to 10^9 in 2012, but I'm guessing that hasn't arrived in production yet. So they have raised the write cycle limit by somewhere between 1000 and 1,0000,000. If you're wearing out your SSD once a year (I'll assume you are using it for swap or something), they've increased it to 1000 yrs. That's pretty much the minimum. So "almost infinite" seems reasonable.
But if you're writing to your RAM at a 10 microsecond interval (say you're doing something like a firewall or HFT) and you keep re-using the same few megs of RAM then instead of 32000 years it's more like 4 months.
So it's fantastic for "durable" storage, but definitely not as good as SRAM or DRAM in general. The lifetime of SRAM and DRAM are both going to be based on the lifetime of the silicon which is roughly related to the temperature of the chip and the thermal migration of dopants. Hotter means more energy for dopants to move, but generally measured in many years or decades.
So, something akin to magnetic relays at the nanometer level?
I have a hard time believing the durability presented (especially with remarks on how it might as well be an infinite number) for a process that involves movement and contact points.
So, if it's been tested continually for 10 years, why are the fabrication technologies behind? This sounds like marketing for vapor ware with as much hype shoved between the specs.
This technology has been around for a while. Lockheed Martin built a 4MB NRAM back in 2008, and sent it up with a memory tester on a Space Shuttle flight in 2009.[2] That was done in conjunction with Nantero; they had an arrangement for government products. The technology is inherently rad-hard, so it's useful for space operations.
This technology has been two years from volume production since 2008.[3] It's not clear why it hasn't reached production yet. It clearly works. It may just cost too much.
[1] http://nantero.com/nantero-closes-30m-series-e-round-its-nex... [2] http://www.nsti.org/events/NNI/sld/pdf/65.pdf [3] http://spectrum.ieee.org/tech-talk/semiconductors/devices/na...