• Guest, The rules for the P & N subforum have been updated to prohibit "ad hominem" or personal attacks against other posters. See the full details in the post "Politics and News Rules & Guidelines."

How do Global Foundries/Samsung 14 nm, TSMC 16 nm, and Intel 14 nm all compare?

norseamd

Lifer
Dec 13, 2013
13,990
179
106
There are at least 3 majors foundries in the computer business, and that is not even counting countless other more minor foundries. There are 3 current major lines from these 3 major foundries in the 14/16 nm node process, but these are more marketing terms and broad category alignment than anything else, as there are many details of multiple components of various process sizes of any "node process" that ultimately make up the true physical size of these process nodes.

So I was wondering how these 3 major process node lines all compare to each other, and not just in the physical/technical characteristics, but also the performance, ease and quality of production, value, etc.

Mods I have no idea where a thread like this should go, so feel free to move it if you feel it is necessary.
 

witeken

Diamond Member
Dec 25, 2013
3,876
154
106
There are at least 3 majors foundries in the computer business, and that is not even counting countless other more minor foundries. There are 3 current major lines from these 3 major foundries in the 14/16 nm node process, but these are more marketing terms and broad category alignment than anything else, as there are many details of multiple components of various process sizes of any "node process" that ultimately make up the true physical size of these process nodes.
Agreed.

So I was wondering how these 3 major process node lines all compare to each other, and not just in the physical/technical characteristics, but also the performance, ease and quality of production, value, etc.
For the layman, there are two things that matter: density (transistors per area) and transistor characteristics. There other minor things, but those are usually not big difference (like every foundry has multiple libraries), and not important for the layman. The designer will chose how they will use the features of the proces node.

1) So first up is density. You can compare SRAM sizes. Every design uses quite a bit of SRAM. SRAM is also an actual circuit that is about the most dense you will find.

TSMC 16nm = 0.07µm² (high density)
Samsung 14nm = 0.080µm² / 0.064µm² (high density)
Intel 14nm = 0.0588µm² / 0.0500µm²

So Intel wins by some margin.

The second method is to calculate the theoretically minimum area of a transistor by multiplying its height and width, namely the gate length and the interconnect length.

TSMC 16nm = 5.120 nm²
Samsung 14nm = 4.992 nm²
Intel 14nm = 3.640 nm²

Again, Intel is smaller transistor area by at least 1.37x.

2) Second, you have transistor characteristics. There are measures of performance and power, but I don't have those, at least not for the foundries. So the next best proxy is to use other transistor characterisrics that tell something about it.

Channel width: All are 8nm
Channel length: 20nm for Intel, I don't know the others.
Fin heigth (higher is better): Intel 42nm, Samsung 38nm.

So here there isn't a whole lot either, although Intel's higher fin should deliver higher performance.

The third best proxy is to note that while Intel has done 2 major finFET nodes, TSMC and Samsung only have their 1st generation finFET. That's also just a proxy because in reality Samsung/TSMC 14/16nm will be somewhere between Intel 14/22, because fin heigth is higher than Intel 22nm, and Intel's 22nm fin was pretty triangular, which is worse than a rectangular.
 

ASK THE COMMUNITY