Why is lithography so important and why is smaller better? What is the thing in the CPU/GPU that is that size? I don't get it
Propagation of information => light speed is constant for a particular medium. Time Delay = Derivative of Phase Delay, for very trivial linear phase velocity, Time Delay = velocity of light x distance. Larger distance results in larger time delay. Period (Duty Cycle) is inversely proportional to the clock frequency, more GHZ = lower Period. Once the Period is within an order of magnitude of the Propagation Delay, can have phase distortion effects which affects edge detection which must be equalized. Result = cannot make chips too large by putting too many transistors in them. Solution = shrink Transistors.
Fabrication of Transistors also is an issue. Larger Gates require more power and yield larger switching latency (think of a gate as a capacitor). Shrinking the Transistor helps on this front.
Issue is we will one day hit the Quantum Wall where Quantum Tunneling and other Quantum effects will come into play. There is a limit to how small we can ultimately make Transistors to be usable in a sense that Engineers are familiar with.
7nm is well beyond the start of the quantum wall, although plenty of things are done to mitigate it. I think somewhere around 90nm is when electrons started leaking through the gate, which is roughly when "xxnm" ceased to have any real meaning (before that the number was the gate width, and you could compare the number between fabs. Note that there were still differences for things like high/low power, but the lower number was almost always "better").
Two things traditionally limit the size of a chip: the first being just how large a chip you can get that is perfectly flawless (of course you can always disable things like cores, memory lines, and other redundant parts but these have to be sold cheaper and there will always be flaws that simply kill the chip. The second is the reticle size of the fab (typically 400-700mm2 in size, nvidia Titans tend to hit this limit): this is the amount of silicon that a single mask can lay down a common circuit. You can "stitch together" separate chips, but that involves trying to match exactly where each mask is located and is typically not cost effective.
I've heard rumblings of the "waferscale" (basically a round super chip with a 1 foot radius): huge chips also have the issue of trying to get all that power in as electricity and out again as heat, as well as the issue of connecting all the input/output.
Name of the process as determined by the manufacturer.
In the past it used to be the size of the structures able to be made (there is way more to a architecture then the transistor), but that transitioned to the minimum adjustments that could be made to a structure being formed, to now they are referred to as "Class" i.e. 14nm class, 10nm class, 7nm class processes.
Unfortunately every manufacturer defines what their "NM" means differently so there is no longer a fixed meaning to it.
I have not looked at it in some time, but it is easier to think of it as developing film, and then plating the image you shot onto the 'paper' you put the image on. So they design a huge image of transistors like a huge circuit board (a cpu) and then shrink it way down and 'print' it out micro tiny in a sense. And note it has multiple layers too. Then you get into how thick is the layer, how wide are the traces, how many circuits did you fit on there. Its like the 'billion customers served' sign at McDonalds, they used to have a number (in 1994) now they don't bother it is so huge.
That’s the individual size of the transistor, by making CPUs better we just shrink them and add more essentially.
The "7nm" you see has nothing to do with actual transistor sizes
It hasn't in generations and is more of an advertising point for large generational improvements
It also takes less electricity to run an individual transistor when its smaller which is why the power requirements stay around the same even though there's WAY more transistors now. This in turn also helps with heat not building up as quick which helps in getting to higher clock speeds.
Thats why you can see performance gains from the same design when they put it on a smaller process/lithography.