Nanotube memory
Followers of this blog may remember some previous posts I'd made about new non-volatile memory that looked like it could also fulfill the same role that RAM does now. That is, it's reasonably dense and fast and so could be used as the working memory of your computer but also doesn't lose the information it holds when you turn off the power.
Well, yet another type of memory with these properties has been in the news recently. The stuff is called NRAM after "nanotube." The nanotubes in the name are the same carbon nanotubes that people talk about making a space elevator with if they can be made long enough and in sufficient bulk. People have been trying to make transistors out of nanotubes for a while and they work but there's a big problem with manufacturing them at scale. You make a big silicon chip with conventional techniques with a bunch of pads on it which you want the nanotubes to stick to. You wash it with a solution containing the tubes and some bonding agent but you'll be lucky to get a tube at 90% of the places you want them. That's enough to do experiments but it's still way too high to build any sort of computer out of. When I first read the EE Times article on this I thought that NRAM would work in a similar way and that they might have manufacturing difficulties. But then a new article with a bunch of presentation slides came out at Anandtech showing that each connection is made up of several nanotubes so you don't have to have every junction have the same configuration and some variation shouldn't prevent it from functioning.
Then I managed to find a link to an actual PowerPoint presentation form a Real World Tech forum thread. It has a lot of details, including the actual current defect rate of .00003 which ought to be fine.
So how does this stack up against the other potential non-volatile memories people have been working on? Well it does seem to be pretty tough. There are actual molecules being shifted into different configurations so it's resistant to electrical damage which is why NASA was interested enough to fly some up to test it in space. Cosmic rays flipping bits is a big problem for computers in space and even if NRAM doesn't have anything else that might be a niche for it.
The write endurance also seems to be pretty good, just like with RRAM. With at least 10e11 writes you could be writing across typically sized RAM card at typical max speeds and it would take you a millennium to exhaust the writes. Even if you write to the smallest area that won't be absorbed by the cache system on a modern CPU you've got a year of writing before you burn a hole there. For typical usage I'd guess it would last something like 20 years which ought to be good enough.
Which of this and MRAM and RRAM will win out if any does? I've got no idea but I'm finding it interesting.
Well, yet another type of memory with these properties has been in the news recently. The stuff is called NRAM after "nanotube." The nanotubes in the name are the same carbon nanotubes that people talk about making a space elevator with if they can be made long enough and in sufficient bulk. People have been trying to make transistors out of nanotubes for a while and they work but there's a big problem with manufacturing them at scale. You make a big silicon chip with conventional techniques with a bunch of pads on it which you want the nanotubes to stick to. You wash it with a solution containing the tubes and some bonding agent but you'll be lucky to get a tube at 90% of the places you want them. That's enough to do experiments but it's still way too high to build any sort of computer out of. When I first read the EE Times article on this I thought that NRAM would work in a similar way and that they might have manufacturing difficulties. But then a new article with a bunch of presentation slides came out at Anandtech showing that each connection is made up of several nanotubes so you don't have to have every junction have the same configuration and some variation shouldn't prevent it from functioning.
Then I managed to find a link to an actual PowerPoint presentation form a Real World Tech forum thread. It has a lot of details, including the actual current defect rate of .00003 which ought to be fine.
So how does this stack up against the other potential non-volatile memories people have been working on? Well it does seem to be pretty tough. There are actual molecules being shifted into different configurations so it's resistant to electrical damage which is why NASA was interested enough to fly some up to test it in space. Cosmic rays flipping bits is a big problem for computers in space and even if NRAM doesn't have anything else that might be a niche for it.
The write endurance also seems to be pretty good, just like with RRAM. With at least 10e11 writes you could be writing across typically sized RAM card at typical max speeds and it would take you a millennium to exhaust the writes. Even if you write to the smallest area that won't be absorbed by the cache system on a modern CPU you've got a year of writing before you burn a hole there. For typical usage I'd guess it would last something like 20 years which ought to be good enough.
Which of this and MRAM and RRAM will win out if any does? I've got no idea but I'm finding it interesting.
Comments
Post a Comment