Wednesday, February 23rd 2022

Qualcomm Said to be Moving to TSMC for 3 nm Chips

Although nothing has been officially confirmed by Qualcomm, it looks like the company will be moving away from Samsung for its 3 nm based chips, in favour of TSMC. The Elec also mentions that Qualcomm has moved some of its Snapdragon 8 Gen 1 production to TSMC, something that has already been hitting the rumour mill. The first batch of 4 nm Snapdragon 8 Gen 1 chips are said to already have entered the early stages of production. The main reason for the move is said to be poor yields by Samsung Foundry on its 4 nm node.

The yield rates are said to be a measly 35 percent for the Snapdragon 8 Gen 1, with Samsung's Exynos 2200 having even lower yields. This also helps explain why Samsung's mobile division has decided to limit the availability of its Exynos 2200 based phones to only a few regions. Apparently Qualcomm had to send staff over to Korea to help get the yields up to their current rate, but it's not hard to see why the company is shifting back to TSMC, as a 35 percent yield rate is simply not acceptable. Samsung is said to be auditing Samsung Foundry to find out what has gone wrong, as anything below 80-90 percent in terms of yield rate is simply not acceptable for mass production. Qualcomm will apparently continue to use Samsung Foundry for its 7 nm RF chips, where the yields must be within industry norms.
Source: The Elec
Add your own comment

14 Comments on Qualcomm Said to be Moving to TSMC for 3 nm Chips

#1
TheoneandonlyMrK
What so while n3 isn't suitable for AMD yesterday it is today for Qualcomm, odd.
Posted on Reply
#2
TheLostSwede
News Editor
TheoneandonlyMrKWhat so while n3 isn't suitable for AMD yesterday it is today for Qualcomm, odd.
As with every node, there are several different types of optimisations that are done.
Keep in mind that TSMC are pretty much building their upcoming nodes to suit Apple's needs, so the focus is on making the nodes really good at making low-power chips.
High-power chips have different requirements and takes time to get to after the low-power chips have reached good yields. This can sometimes take up towards a year.
This is not news, as it has been the case since at least 28 nm from TSMC.
There's a reason why TSMC has all these sub-nodes and variants.



I'd suggest reading this for a better understanding.
fuse.wikichip.org/news/3398/tsmc-details-5-nm/
Posted on Reply
#3
_Flare
Chronologically:
N7 - no EUV
N7+ - introduction of EUV
N7P - no EUV, compatible with N7 designrules
N5 - 2nd node using EUV
N6 - 3rd node using EUV compatible with N7 and N7P designrules, but easier to use

N5P and N4 are compatible with N5 designrules, N4 will be easier to use, like N6
Posted on Reply
#4
TheLostSwede
News Editor
_FlareChronologically:
N7 - no EUV
N7+ - introduction of EUV
N7P - no EUV, compatible with N7 designrules
N5 - 2nd node using EUV
N6 - 3rd node using EUV compatible with N7 and N7P designrules, but easier to use

N5P and N4 are compatible with N5 designrules, N4 will be easier to use, like N6
That was mostly meant as an illustration to show that things aren't as simple as one node being suitable for one company, but not another.
The diagram doesn't even show the various design rules, nor the LP/HP versions.
I guess this is the most up to date diagram from Wikichipm, which further complicates things.


fuse.wikichip.org/news/6439/tsmc-extends-its-5nm-family-with-a-new-enhanced-performance-n4p-node/
Posted on Reply
#5
defaultluser
No surprise, after Qualcomm went to Samsung by-default, the contract was theirs to lose!

I'm sure that the piss-poor efficiency also helped make the move back?
Posted on Reply
#6
zlobby
Samsung made a big, big poopoo this time.
Posted on Reply
#7
TheLostSwede
News Editor
zlobbySamsung made a big, big poopoo this time.
It's getting harder and harder to make chips on the cutting edge.
Posted on Reply
#8
DeathtoGnomes
TheLostSwedeIt's getting harder and harder to make chips on the cutting edge.
My magic 8-ball thinks that there will soon™ be a time that chip manufacturing will be behind the cutting edge until production can meet current demands without issue, meaning more fabs/forges will be needed even as currently planned ones are being built. Planning farther ahead than then just the next process.
Posted on Reply
#9
zlobby
DeathtoGnomesMy magic 8-ball thinks that there will soon™ be a time that chip manufacturing will be behind the cutting edge until production can meet current demands without issue, meaning more fabs/forges will be needed even as currently planned ones are being built. Planning farther ahead than then just the next process.
Question is what is beyond 3nm? Beyond 1nm?
Posted on Reply
#10
Wirko
We also don't know whether QC and Samsung can sell/use some partly defective chips as lower-tier products or not.
Posted on Reply
#11
seth1911
I dont have such fast SoC in my phones, in my passport is a Krait 400 and in my second smartphone is a Snapdragon 425 :laugh:

Yeah i can not game on it but discord etc. runs verry well,
even on my second one with the SD 425 and 2GB RAM in kiwi browser are sometimes 90 TABS open and dont have a problem (G Services etc. disabled):D


Did u know that google playservices runs in background and take from smartphones with 1GB RAM about 180MB for those services while u dont use it?
Posted on Reply
#12
Prima.Vera
zlobbyQuestion is what is beyond 3nm? Beyond 1nm?
pm? they will start with 900pm, 800pm...and so on. But probably Silicon needs to be replaced with something else.
Not sure why the supercomputers are not doing active R&D on this. Maybe they already are...
Posted on Reply
#13
DeathtoGnomes
Prima.VeraBut probably Silicon needs to be replaced with something else.
This was talked about before but I cant find where I read it. I want to say there is ongoing research for that, but dont quote me here.
Posted on Reply
#14
zlobby
Prima.Verapm? they will start with 900pm, 800pm...and so on. But probably Silicon needs to be replaced with something else.
Not sure why the supercomputers are not doing active R&D on this. Maybe they already are...
We all know the scale. Problem is the Si. It just can't go any smaller.

There are alternatives of course, but which one will be mature and cheap enough for mass production?
Posted on Reply
Add your own comment
Dec 21st, 2024 21:25 EST change timezone

New Forum Posts

Popular Reviews

Controversial News Posts