Archive for February 21st, 2012

Is one atom enough to seal the fate of Moore’s law?

Tuesday, February 21st, 2012 by Roberto Saracco

Every now and then there is someone voicing that the end of the applicability of Moore’s law is in sight. This is again the case now as a paper on Nature Nanotechnology is reporting the success of a team of researchers to build a transistor mane by a single atom…

A rendering of the so called one atom transistor: the one atom that is used as a gate is the one in the middle of the image

Clearly, one cannot imagine anything smaller than  that. As a matter of fact, if you really look at the paper, and study the physics of computation, you see that one the one hand the transistor created by this team is made up of more atoms than your whole laptop has, and on the other hand the physics of computation extends to sub atomic particles, hence even if a one atom transistor would have been made we would still be quite afar from the ultimate limits of computation.

First of all, it is not the first time that scientists manage to use a single atom to control the flow of electrons, that is to make a transistor. The first news of this, that I am aware of, goes back to 2009 (ages ago in technology evolution terms…).

What is new, in the article published by Purdue and University of Melbourne, Australia, researchers, is the process for controlling the atom of phosphorus.

Having an atom doing the job of controlling the flow of electrons is not exactly the same as having a transistor. As it can be seen in the figure, the single atom is in the middle of the sort of gate through which electrons flow but you need plenty of other atoms to create the gate. In this particular case, you also need plenty of machinery to keep the transistor at minus 196 degree Celsius.

We have now reached  2.3 billion transistor in a chip (the Intel Sandy Bridge Chip), each one taking about 32 nanometre. A single atom takes about 0.1 nanometre so in principle you get a 300 fold higher density. But that is just in principle since, as I have noticed, the current device consists of many more atoms. Secondly, the Moore’s law is about the number of transistor on a chip, and you can get, up to a certain point, more transistor in the chip by shrining their size or by increasing the size of the chip!
The limiting factor, beyond technology, is “cost”. Increasing the chip surface means decreasing the yield per wafer, thus increasing the cost of the single chip. That is why decreasing the transistor (or the etching) size is so important: it keeps the cost stable.

I guess the press got very excited with this news since it seems so easy to understand: we got down to one atom, nothing smaller can exist! Great result, end of the line. As I said it is not true, and it is not true in many ways.

Of the several “ways” it is not true I just want to highlight one (in addition to the two already mentioned): the Moore’s law has subtly changed over the past ten years. We no longer measure the processing power by the number of transistor on a chip, nor by the “clock” of the chip. Practical limitations like power dissipation and cost, have pushed the researchers to create new forms of computation architectures, where the words pervasive and disseminated are taking the upper hand. The overall computation capability is now becoming a property of a system, no longer the one of a chip. And it also goes beyond the multicore structures since these have shown several bottleneck. Probably, the future computation architectures will resemble much more our brain and body where processing is happening every where and although the single processing unit is pretty slow (the neuron and the sensory cells), the whole is capable of amazing processing, not of amazing processing speed! Even if you multiply the number of neurons in your brain (100 billion) for the speed of the single neuron (in the range of 0.1-1 Hz) you get very little processing power (100 GFLOPS … if you can ever measure in GFLOPS…). That is so much less than the chip inside the laptop I am using right now!

The truth is that as we move to pervasive processing (and the associated networking) the definition of computation change dramatically. And we are now moving in that direction.
No end of the line for the conceptual Moore’s law in sight.

Anyhow, take a look a the clip on the one atom transistor, but I hope you will look at it with different “eyes” now.