Of note: current CPU transistors are actually reasonably near this limit already. Consider Intel's recently announced 22 nanometer process
, where an SRAM cell consists of 6 transistors, and is 0.1 micrometers square. Plus, the size of SRAM includes connections between those transistors, so figure maybe 3 atoms per transistor in "wiring" gives a size of an SRAM cell of about 24 atoms. Therefore, assuming the atomic radius of a single silicon atom is ~100 picometers: (sqrt(0.1) micrometers / 24) / 100pm or about 131.761569. A significant gain, to be sure, but not really "enormous" (to give an idea, that would be about 7 "die shrinks" from now, assuming you could mass produce such transistors. The 2000s decade had about 5 die shrinks, to give an idea. Moreover, this ignores that even if you can make a single atom sized transistor, there are still other parts of the chip that would have to be dealt with, which would (presumably) be greater than the size of an atom, and therefore the total size of the entire apparatus doesn't shrink that much.
Is it a step forward? Yes, yes it is. But it's not an unbelievable advancement. I think it's likely we'll see other technologies hitting limits before we see transistors pushed to this kind of size. (e.g. we're already having problems scaling CMOS lithography at the current (22nm) process node; creating transistors one by one with a SEM isn't really practical for something like a CPU, which has (in Ivy Bridge) some 2.27 billion transistors)
The more amazing thing is that *current* transistors are as small as they are and still mass produced.
Edited by Billy O'Neal, 20 February 2012 - 08:54 PM.