[qsmi]

> node names are entirely unrelated to any physical dimensions

I agree they're not necessarily related to any particular transistor dimension but are you sure it's entirely unrelated to any physical dimensions? If so, here is a question I have. 28nm, 22nm, 20nm, 14nm, 10nm, 7nm, 5nm, 3nm. What progression is that? Why skip 6nm and 4nm?

[dragontamer]

A "full node" step is 1/2 the area of the previous. Since area is dimension-squared, 28 x 28 == ~800 area, while 20 x 20 == ~400 area.

So 28 -> 20 is a "full node" decrease (ie: twice the transistors).

---------

We can see the full-node step from 14nm -> 10nm as well: 14 x 14 == ~200, 10x10 == 100. So its a full node step. The next full-node is 7 x 7 == ~50, which is half of 100. After 7 comes 5, because 25 is half of 50.

The next full node is sqrt(12.5) or 3.5, smack dab between 3 and 4, so not really easy to round.

---------

The steps in between are "half-node" decrements, where you didn't quite achieve a full 1/2 area reduction. An incremental technology that represents something in between the full node step progression.

[qsmi]

That's an interesting explanation and somehow I never heard it explained that way before. It makes sense if one thinks of the transistor count doubling relative to where that company previously was, and not where the industry is. In the transistor density plot each curve is roughly x^2, but with different initial conditions.

https://www.techcenturion.com/7nm-10nm-14nm-fabrication

[dragontamer]

2^x, not x^2.

x^2 is 1, 4, 9, 16, 25.

2^x is 1, 2, 4, 8, 16, 32