How many electrons in each processor cycle?

January 8, 2008 5:04 AM

In a modern CPU, how many individual electrons, approximately, are "let through" in each clock-cycle of, say, an individual transistor or logic gate?

Are we close to managing the paths of individual electrons? Or are we still shuffling around millions of the little blighters in each transistor in each tick of a 2ghz processor?

posted by Jimbob to technology (7 comments total)

We can use math to make an estimate.

This 3.2 GHz CPU uses 125W at 1.4V = 90 amps. At 3.2 GHz that's 174 Billion electrons per cycle (6.2e18 electrons per amp). The Windsor has about 150 Million transistors, so on average each transistor is letting through about a thousand electrons.

This completely disregards the fact that not all the transistors are actually doing something each tick, or that the processor isn't 100% efficient, but it's enough to show that we're definitely not operating on the individual electron level.
posted by 0xFCAF at 5:46 AM on January 8, 2008

Let's look at the individual transistor level. The newest 45 nm transistors are reported to run a current of about 1 mA/um. Assuming a width of 45 nm, that's about 50 uA. Dividing this by one billion clock cycles, we have 5x10^-14 C per clock cycle. A coulomb is about 10¹⁹ electrons, so we're still talking about hundreds of thousands of electrons per transistor per clock cycle.

This estimate is in agreement with 0xFCAF's if we assume that only 1% of the transistors are switching per clock cycle.
posted by Mapes at 5:59 AM on January 8, 2008

I agree with the above, but you might be interested in knowing that single electron transistors are currently being researched. See here for example.
posted by PercussivePaul at 9:39 AM on January 8, 2008

0xFCAF's estimate also doesn't take into account the fact that, in CMOS, a lot of electrons get reused by gates wired in series.
posted by blenderfish at 9:40 AM on January 8, 2008

To first order, a CMOS logic gate only needs to conduct enough current to charge up the input capacitance of the logic gates that it is driving. This value is on the order of fF to tens of fF, which corresponds to tens to hundreds of thousands of electrons.

(A note: in CMOS scaling, the nanometer value generally refers to the nominal gate length, not width, which is larger.)
posted by Krrrlson at 9:56 AM on January 8, 2008

we're not close, in industry at least. There is a lot of experimental research being performed on single electron transistors and that is just a catchphrase, some stuff with quantum dots deal with single electrons too, but it's all barely operable in laboratories. Researchers make thousands of devices, and if 4 or 5 work, that makes a paper.
posted by Large Marge at 3:48 PM on January 8, 2008

My Honours thesis dealt with single electron transistor fabrication using metal deposition. Single electron transistors don't really work much like their semiclassical cousins - quantum effects mean that instead of exponential increases in current flow with gate voltage, you get oscillations. Also, the SETs I was working on had to be cooled to liquid helium temperatures.

This was 5 years ago, so possibly fabrication techniques are improving to get down to the thousands of electrons a mentioned in answers above.
posted by claudius at 12:46 AM on January 9, 2008