Nantero Exits Stealth: Using Carbon Nanotubes for Non-Volatile Memory(anandtech.com)
anandtech.com
Nantero Exits Stealth: Using Carbon Nanotubes for Non-Volatile Memory
http://www.anandtech.com/show/9314/nantero-exits-stealth-using-carbon-nanotubes-for-nonvolatile-memory-with-dram-performance-unlimited-endurance
7 comments
I wouldn't be that concerned even if the endurance isn't any better than the lower bound. A large, fast DIMM today would have 8GB of storage and 25 GB/s of bandwidth. If you write to the whole thing at full speed that'll get you a millennium of writing. If you're only ever writing to the smallest bit that won't be masked by a 8MB L3 then you've still got a year of continuous writing even in the absolute worst case.
> First, 10^11 P/E cycle and "practically infinite" aren't really two concepts I would associate with each other when discussing RAM.
I think their point of comparison for this statistic is NAND flash. 10^11 is much better than current NAND P/E cycles ( < 10^9 I believe).
I think their point of comparison for this statistic is NAND flash. 10^11 is much better than current NAND P/E cycles ( < 10^9 I believe).
It's roughly 3000 cycles. Yes, three thousand. See: http://www.anandtech.com/show/7237/samsungs-vnand-hitting-th...
With the endurance numbers they're publishing, I can't wait to hear about the applications for reconfigurable computing.
Who wouldn't love a glob of 7nm computronium that reconfigures itself on a per-process basis?
Who wouldn't love a glob of 7nm computronium that reconfigures itself on a per-process basis?
Even small amounts (megabytes) of NRAM could be pretty sweet as a non-volatile write & read cache and book-keeping for SSDs and eMMCs.
Everyone knows there's a direct relationship between stealth and FOMO. Maximize FOMO by remaining stealth as long as possible.
When I first heard about their techniques I was worried about their ability to precisely guide nanotubes onto the chip pads, something that has deviled people attempting to build nanotube transistor computers. Luckily their diagrams make it look like they expect a number of nanotubes to cross each connection making the whole thing far more robust to process variation.
What "stealth"? Even the article says Nantero has been working on this for over a decade. They and Lockheed had a NVRAM back in 2008. It was sent into space in 2009, along with a memory tester. It's radiation-hard, so it has space applications.
The technology seems to work fine. Manufacturing cost is apparently the problem. Nantero has been trying to fab the things in something close to a standard fab. They may eventually succeed at getting the cost down, but they've been struggling financially.
The technology seems to work fine. Manufacturing cost is apparently the problem. Nantero has been trying to fab the things in something close to a standard fab. They may eventually succeed at getting the cost down, but they've been struggling financially.
The write/erase endurance is practically infinite as independent university study has shown Nantero’s NRAM technology to have over 10^11 P/E cycles,(for your information, 10^11 translates to 100 billion).
First, 10^11 P/E cycle and "practically infinite" aren't really two concepts I would associate with each other when discussing RAM.
And second, an Anandtech article explaining what 10^11 means? If I have trouble with that concept, about 98%+ of their articles would be completely over my head.