> By simply swapping carbon nanotube transistors for conventional ones in a
> simulated IBM microprocessor, they were able to increase speeds by a factor
> of seven or, alternatively, achieve power savings almost as significant, said
> Wilfried Haensch, an IBM physicist who is a member of the research group.
In technology, one often comes across the use of the word "simply" to refer to something that's actually quite complicate (see the Druid docs[0]) but this is quite a doozy.
It's crazy to think that a major breakthrough like this with a factor of seven increase in transistor density is little more than four years of progress as measured by Moore's Law.
> It's crazy to think that a major breakthrough like this with a factor of seven increase in transistor density is little more than four years of progress as measured by Moore's Law.
In a single step, by switching to an experimental new process. It seems likely that, if that process pans out, it'll then get years of extensive refinement to improve both yield and density.
I don't know about that. They were talking about line widths of 40 atoms. Another four years (factor of seven improvement) takes you to 6 atoms; another four years (factor of 7 improvement) takes you to one atom. At that point you're not talking about a "carbon nanotube" any longer - it's just a row of carbon atoms, and you still can't improve it any further.
I mean, yes, there will be process improvements. The article talked about 28 atoms as the next step. But there are fundamental limits, and they aren't very far away...
(I could be overstating the case here, in that transistor density may scale as the square of the line width, rather than linearly with it. The point about fundamental limits still stands, however.)
There will always be a hard limit, but one thing I could see us looking back on and finding weird is that we have a single layer of transistors. If we could build more complex 3D structures I'd expect we could see vast improvements.
Not part of Moore's Law but there are also
1. Significant storage improvements, and bringing the storage closer to the processing (or the other way around) which massively improves the overall speed.
2. Better chip design. Not knocking the current designers, but it'd be foolish to assume we have the theoretical optimum chip designs.
3. Cost. Sure, you're not going to improve every use case but what if my motherboard was just a big layer of cpus with non-volatile huge caches because it cost pennies? Particularly when we hit some larger limits, we'd be likely to see that current best design being produced by lots of people and getting cheaper and cheaper.
Along with 3. comes more custom designed chips. If we can get the cost of design and fabrication down, we can speed things up by having a bunch of chips for specific functions.
In technology, one often comes across the use of the word "simply" to refer to something that's actually quite complicate (see the Druid docs[0]) but this is quite a doozy.
[0] http://cse.google.com/cse?cx=004325268462231820898:e4htgtatb...