Array
(
    [content] => 
    [params] => Array
        (
            [0] => /forum/threads/smic-n-3-confirmed-kirin-9030-analysis-reveals-how-close-smic-is-to-5nm.24183/
        )

    [addOns] => Array
        (
            [DL6/MLTP] => 13
            [Hampel/TimeZoneDebug] => 1000070
            [SV/ChangePostDate] => 2010200
            [SemiWiki/Newsletter] => 1000010
            [SemiWiki/WPMenu] => 1000010
            [SemiWiki/XPressExtend] => 1000010
            [ThemeHouse/XLink] => 1000970
            [ThemeHouse/XPress] => 1010570
            [XF] => 2030770
            [XFI] => 1060170
        )

    [wordpress] => /var/www/html
)

SMIC N+3 Confirmed: Kirin 9030 Analysis Reveals How Close SMIC Is to 5nm

Fred Chen

Moderator
TechInsights has completed an exploratory teardown and process analysis of the Huawei Kirin 9030 application processor used in the new Mate 80 Pro Max. Our structural and dimensional analysis confirms that the chip is manufactured using SMIC’s N+3 process, a scaled evolution of its 7nm-class technology, and a key indicator of how close SMIC is to achieving a true 5nm-equivalent node without EUV lithography.

 
While SMIC’s N+3 process shows meaningful density improvements, our comparative measurements confirm it remains significantly less scaled than leading commercial 5nm nodes offered by TSMC and Samsung. Full competitive metrics—including gate density comparisons
 
While SMIC’s N+3 process shows meaningful density improvements, our comparative measurements confirm it remains significantly less scaled than leading commercial 5nm nodes offered by TSMC and Samsung. Full competitive metrics—including gate density comparisons
I think this characterization by TechInsights is off. The author provided me some key details. I posted only a rough version on X:


Maybe a few % is already significant in this context?

They had to go beyond the double patterning of N+2. The metal pitches are close to TSMC N5.
 
Last edited:
TSMC 5nm is 137 and Samsung 5nm is 126. So you mean 115~125 mtr?

You are calculated by cell height x Gate pitch which provided by Techinsight ?

Or M2 pitch x6=cell height and x gate pitch?
 
AFAIK Samsung's 5nm node is half node of their 7nm EUV process, and TSMC's is full node from their N7. So if SMIC's density is similar to Samsung's then it's more like 7nm-equivalent process.
I'm just going after the "significantly less scaled" than Samsung "5nm" part.
 
Yes I see that. Anyway I had to delete this comment because the number I knew from TSMC was wrong.
I went back to check the numbers, what I have is:

Samsung and TSMC 7nm and 5nm.png


I am not sure if SMIC N+2 density was ever released or calculated?
 
N+2 is 93 mtr.

It has 63nm gate pitch and 252nm cell height which revealed by Techinsight.

118 mtr mean ~27% density improve, very huge.
Yeah, I remembered the gate pitch had been "abnormally" large, but actually it's just that SMIC density < Samsung density < TSMC density for a given process family or cluster.
 
I went back to check the numbers, what I have is:

View attachment 3966

I am not sure if SMIC N+2 density was ever released or calculated?
Just a spot check, the Semiwiki article on TSMC N5, has:

171 MTr/mm2, but agrees with 51 and 57 gate pitches for N5 and N7 respectively.

 
Just a spot check, the Semiwiki article on TSMC N5, has:

171 MTr/mm2, but agrees with 51 and 57 gate pitches for N5 and N7 respectively.

This article explained that some discrepancies may arise, depending on the layout assumptions, such as the ratio of NAND to Flip-Flop: https://www.angstronomics.com/p/the-truth-of-tsmc-5nm

I used the formula in that article, but actually I applied it wrongly for the 7nm generation nodes since they were double diffusion break. Here is the correction:

Samsung and TSMC 7nm and 5nm.png


Assuming SMIC has the same double vs. single diffusion break difference, N+3 should be more readily associated with the 5nm crowd.

To get away from these layout assumptions, I'd rather use minimum metal half-pitch as the defining feature, as that sets the main patterning difficulty.

Or maybe cells/um2 (1/gate pitch/cell height)?
 
Last edited:
Appreciate all the discussion and accurate numbers on minimum pitches and raw maximum cell densities, but without any discussion of routing rules and litho effects on routing rules, the raw density is a bit meaningless. I still remember the whole fracas back at 14nm/16nm, where Intel was claiming a raw library density around 37–45 MTr/mm² but only delivering a density in the mid-teens on Atom, while TSMC was delivering 20 MTr/mm² in the Apple A9, even though they had a far lower raw density. Some of that Intel “loss” is related to chip composition and methodology, but they also had routability issues.
 
Last edited:
It seems like N+3 is truly G57H228 ( come from Techinsight decap data).

Hard to judge whether this is 5nm or not.

Techinsight think N+3 is 7nm scaled, not 5nm due to density is lower than 5LPE and N5: 120 vs 134 vs 145.

(And Techinsight list N6 at 114 mtr, so N+3 slightly higher than N6 but far inferior to TSMC N5)

But to me, N+3 M0 pitch is so impressive, which can be classified as 5nm node.

But M2 pitch only slightly reduce from 7nm node of 40nm to 38nm, not as intensive as M0 pitch scaled. Wonder why?
 
Techinsight think N+3 is 7nm scaled, not 5nm due to density is lower than 5LPE and N5: 120 vs 134 vs 145.

(And Techinsight list N6 at 114 mtr, so N+3 slightly higher than N6 but far inferior to TSMC N5)
TSMC N6 changed diffusion break from double to single, that in itself is enough to put it above 100 MTr/mm2. So it is a noticeable jump in density from earlier 7nm. The design rules were (supposedly) the same as N7 though I think gate pitch shrunk to 54 nm to get the number below.

TSMC N6 vs N5 and N7.png


I had checked with the author, they may not be using the same ratio of FF:NAND. So absolute numbers may vary from source to source. But within a given author's consistency, the density ratios would be the same and still give the same plot appearance.
 
Last edited:
Back
Top