[Digitimes] Foundries 10nm yield rates lower-than-expected

[witeken]

Foundries 10nm yield rates lower-than-expected

Both Samsung and TSMC have claimed they have 10nm in "high" volume production at the moment. Yeah right.

TSMC is scheduled to start making Apple's A10X chips for the next-generation iPad series slated for launch in March 2017. Unsatisfactory yields for the foundry's 10nm process could disrupt the schedule, the sources indicated.

Meanwhile, yield rates for Samsung's 10nm process technology have been low prompting Qualcomm to turn cautious about its product roadmap for 2017, the sources said. Qualcomm originally planned for the Snapdragon 835 and other chips including the 660 (codenamed 8976 Plus) built using Samsung's 10nm process, but has revised its roadmap by having only the 835-series made using the newer node technology.

That bodes well for TSMC's 7nm which is supposed to go "HVM" in H1'18, certainly if you know their 7nm will make use of quadruple patterning, which is literally twice as hard as double patterning .

How low are TSMC and Samsung willing to drop their yields to meet mobile phones' customer demands? Well, I guess as long as their customers are willing to pay per wafer (lower yields = more wafers = higher cost for customers), until they move to Intel.

Edit: Poor yields apparently mean below 50%. Impressive. http://www.fool.com/investing/2016/12/17/taiwan-semiconductor-mfg-co-ltd-to-start-building.aspx

 

[dark zero]

I feel that Intel will realize that 10 nm will be another 20 nm all over again.... they had a hard time with 22 nm.

No wonder why GloFo is skipping it going to 7 nm.

 

[witeken]

I feel that Intel will realize that 10 nm will be another 20 nm all over again.... they had a hard time with 22 nm.

They had a hard a time with 14nm. 22nm is their highest yielding process ever.

No wonder why GloFo is skipping it going to 7 nm.

Why do you think 7nm will be easier than 10nm? BTW, do you realize that GLoFo's 7nm will be very similar to Intel's 10nm?

 

[itsmydamnation]

because EUV should hopefully finally be ready by then.

 

[dark zero]

because EUV should hopefully finally be ready by then.

That's what I am expecting. EUV is the true refresh of the process. FinFet wasn't as effective after all with the current materials.

 

[Andrei.]

I heard TSMC's 10nm is a bit of a mess.

 

[krumme]

I heard TSMC's 10nm is a bit of a mess.

As in unexpected mess?

 

[witeken]

I heard TSMC's 10nm is a bit of a mess.

How about Intel 10nm? There's something behind a paywall.

http://semiaccurate.com/2016/12/22/intel-10nm-cannon-lake-back/

 

[dark zero]

They had a hard a time with 14nm. 22nm is their highest yielding process ever.

Wasn't 32 nm the best process of Intel's history? And even more, isn't that process the best of the WHOLE processor history?

 

[turtile]

They had a hard a time with 14nm. 22nm is their highest yielding process ever.


Why do you think 7nm will be easier than 10nm? BTW, do you realize that GLoFo's 7nm will be very similar to Intel's 10nm?

Also consider that GF can use all of it's resources working on 7nm instead of split between 7 and 10.

 

[dark zero]

Why do you think 7nm will be easier than 10nm? BTW, do you realize that GLoFo's 7nm will be very similar to Intel's 10nm?

You mean the refresh of 10 nm? Because GF 7 nm targets that instead of being a new unmature process.

 

[witeken]

Wasn't 32 nm the best process of Intel's history? And even more, isn't that process the best of the WHOLE processor history?

No.

 

[Arachnotronic]

I heard TSMC's 10nm is a bit of a mess.

DigiTimes and BlueFin Research both say 10nm yields at TSMC aren't great. DigiTimes also says 10nm at Samsung is in bad shape, too.

The foundries are pushing too fast to meet customer schedules.

 

[ninaholic37]

Wasn't 32 nm the best process of Intel's history? And even more, isn't that process the best of the WHOLE processor history?

Sandy Bridge and "real" Atom (before they started calling it Celeron/Pentium). Good times.

 

[witeken]

DigiTimes and BlueFin Research both say 10nm yields at TSMC aren't great. DigiTimes also says 10nm at Samsung is in bad shape, too.

The foundries are pushing too fast to meet customer schedules.

So no 10nm GPUs in '17?

 

[jpiniero]

There won't be 10 nm GPUs period. Both AMD and nVidia will skip it.

 

[dark zero]

There won't be 10 nm GPUs period. Both AMD and nVidia will skip it.

There will be... but not the desktop GPU, only mobile like Intel HD, Adreno, Mali and Power VR

 

[oak8292]

TSMC has responded to the rumors.

"TSMC’s 10nm process is “totally on track” and will contribute to sales revenue for the first time in the first quarter of 2017, according to Elizabeth Sun, senior director, TSMC's Corporate Communications, refuting a report by Taiwan’s Digitimes websitethat was picked up by other news media."

http://www.eetimes.com/document.asp?doc_id=1331059

HVM and poor yields are not well defined terms. Intel defined 'volume' as anything over 1 million processors in a quarter which can be less than one FOUP (cartridge of 25 wafers) per day depending on die size and yield. The yield ramp also depends on expectation. A yield of less than 50% at this point in the production may exceed internal expectations?

 

[witeken]

"The company predicted that high-end smartphones, the main driver in the semiconductor business, will move to 10nm from 16nm during 2017."

Phones? So iPhone in September '17. That's only weeks before Intel 10nm.

" The company’s 10nm process will account for less than one percent of TSMC’s overall revenue during the first quarter of 2017, she said."

So just the iPad at most.

Edit: Also Huawei http://www.gizmochina.com/2016/12/2...k-10nm-tsmc-process-cat-12-lte-octa-core-cpu/.

Does MediaTek go with SS or TSMC?

 

[Lodix]

"The company predicted that high-end smartphones, the main driver in the semiconductor business, will move to 10nm from 16nm during 2017."

Phones? So iPhone in September '17. That's only weeks before Intel 10nm.

" The company’s 10nm process will account for less than one percent of TSMC’s overall revenue during the first quarter of 2017, she said."

So just the iPad at most.

Edit: Also Huawei http://www.gizmochina.com/2016/12/2...k-10nm-tsmc-process-cat-12-lte-octa-core-cpu/.

Does MediaTek go with SS or TSMC?

Mediatek uses TSMC along with a lot of not so loud companies, and they all move huge volumes of chips.

 

[oak8292]

At one point the Mediatek Helio X30 was rumored to be a 'pipe cleaner' processor for the 10nm process.

https://www.xda-developers.com/mediatek-officially-unveils-the-10-nm-helio-x30-and-16-nm-helio-p25/

The iPad will be one of at least three. Who is on point to bring up the 10nm process is still an open question? Apple has the deepest pockets but it may fall to another company to pick up the costs.

 

[Headfoot]

That's what I am expecting. EUV is the true refresh of the process. FinFet wasn't as effective after all with the current materials.

Not true. FinFET has panned out to be extremely useful, necessary even for 14nm. 20nm was a total turd because it was still planar. There's a pretty direct line to draw from FinFET to lower vcores to lower power consumption.

 

[Nothingness]

There won't be 10 nm GPUs period. Both AMD and nVidia will skip it.

My understanding is that 10 nm is an interim process for foundries. Everyone is waiting for 7 nm. Well of course the same was said of 20 nm.

Let's see how Intel do for LG with their 10 nm. I wouldn't bet they are doing well atm Of course Intel has to learn what being a foundry really means...

 

[witeken]

My understanding is that 10 nm is an interim process for foundries. Everyone is waiting for 7 nm. Well of course the same was said of 20 nm.

Not interim process for Intel and Samsung, which are half of the leading edge companies. So that statement is not at all true.