[WCCF] AMD Contracts TSMC To Produce Zen At 16nm Amidst Concerns Over 14nm

[gdansk]

Samsung isn't shipping anything high-performance on 14nm.

iPhone's A9 are on Samsung's 14nm:
http://www.chipworks.com/about-chipworks/overview/blog/inside-the-iphone-6s

High performance given the IPC. 14LPP is what Zen would end up on anyhow.

 

[Idontcare]

iPhone's A9 are on Samsung's 14nm:
http://www.chipworks.com/about-chipworks/overview/blog/inside-the-iphone-6s

High performance given the IPC. 14LPP is what Zen would end up on anyhow.

The A9s are produced by both foundries, hopefully we'll get to see a teardown comparison in the coming days since the chips themselves won't be exactly identical.

 

[gdansk]

The A9s are produced by both foundries, hopefully we'll get to see a teardown comparison in the coming days since the chips themselves won't be exactly identical.

Yeah, I meant some A9. Either way, it renders suspect the "Samsung isn't shipping anything high-performance on 14nm."

 

[Arachnotronic]

Yeah, I meant some A9. Either way, it renders suspect the "Samsung isn't shipping anything high-performance on 14nm."

It's high performance for a mobile chip

 

[hawtdawg]

Yeah, I meant some A9. Either way, it renders suspect the "Samsung isn't shipping anything high-performance on 14nm."

In what universe is 5W TDP CPU considered high performance?

 

[ViRGE]

The A9s are produced by both foundries, hopefully we'll get to see a teardown comparison in the coming days since the chips themselves won't be exactly identical.

Are you sure about that, IDC? That seems really questionable. As you note the chips won't be identical, which would be a bad thing for Apple.

 

[IntelUser2000]

It's high performance for a mobile chip

Yes it is. At this rate we'll be able to truly call it high performance if Intel keeps on making 3-5% generational gains and Apple keeps up their big ones.

4770K-level chip on a late 2017 iPad with 10nm A11?

Also I doubt the TDP on the Apple chips are 5W. The whole device gets nearly 3 hours when gaming on a 9WHr battery. The chip is probably at 3.5W at max.

The way Intel is struggling with their 14nm(Broadwell/Skylake availability) means there's big opportunity even for AMD at 10nm.

 

[Arachnotronic]

The way Intel is struggling with their 14nm(Broadwell/Skylake availability) means there's big opportunity even for AMD at 10nm.

Lol

There's a world of difference between rich and powerful Apple (which I hear poaches top semiconductor talent by literally offering to double people's salaries) and AMD, which can barely keep the lights on.

 

[videogames101]

Are you sure about that, IDC? That seems really questionable. As you note the chips won't be identical, which would be a bad thing for Apple.

There are two potential problems with "non-identical" chips. It's bad for Apple if the power/perf is not (mostly) equivalent, and it's bad if doing the physical design twice impacts the overall project schedule. Apple easily has the resources to do physical designs for both processes at once, they may even have two design teams if they are targeting two foundries. Then if we assume the two processes can both meet the chip's target PPA, two sources actually has tangible benefits for Apple. One source can have problems; see Intel Skylake availability. When you're riding on a single source foundry, you're at the mercy of their yields and schedule. Potential benefits also include silicon pricing (because neither foundry has Apple "by the balls" as it were) and supply chain reliability.

Yes it is. At this rate we'll be able to truly call it high performance if Intel keeps on making 3-5% generational gains and Apple keeps up their big ones.

4770K-level chip on a late 2017 iPad with 10nm A11?

Also I doubt the TDP on the Apple chips are 5W. The whole device gets nearly 3 hours when gaming on a 9WHr battery. The chip is probably at 3.5W at max.

The way Intel is struggling with their 14nm(Broadwell/Skylake availability) means there's big opportunity even for AMD at 10nm.

Modern commercial CPU architecture is limited by the state of academic architecture research. Most of the companies doing CPU cores (not very many left) continue to adopt the same structures which have first been validated in academia. Although they target different frequencies and use different foundries, the overall structures that each company uses inside their CPUs are quite similar. Because Apple has no magic bullet to advance this research more quickly, I believe their IPC gains will level out. That's not to say they can't do better than other companies, the designs themselves are obviously enormously complex and some companies will get the details down better than others. But in my opinion continued large generational improvements in IPC (while maintaining comparable frequency) just aren't possible.

As for Intel's 14nm struggles, I see no reason why you would expect the foundries (meaning TSMC and Samsung/GF) to get 10nm off the ground any easier than Intel will. Of course it's possible, but I don't know why you would predict it. I would understand if you were using 14nm as a data point, but I am not aware of any public data which directly compares Samsung and Intel's 14nm yields (much less data that also accounts for process parameter/performance differences!)

 

[zir_blazer]

Can anyone explain me how GloFo is such a disaster? I mean, they used to be AMD fabs. Aren't they staffed by mostly the same people? I would have expected that AMD and GloFo had more synergy. Instead, is a type of ball and chain.

 

[Idontcare]

Are you sure about that, IDC? That seems really questionable. As you note the chips won't be identical, which would be a bad thing for Apple.

Back when TI had 80% marketshare in mobile phones, we fabbed the same chip at three different foundries plus internally manufactured in our own fabs. We did this for multiple nodes over many years without any problem.

You can do it provided you drive the requirement of "electrical equivalence". Leakage specs, operating voltages, etc. were all required to be the same, which was possible provided the design factored in this requirement.

To my knowledge the A9 contract is setup so that 50% comes from TSMC and 50% from Samsung. But I have no idea how you would determine which foundry fabbed the specific A9 that is in your iPhone 6S without tearing apart your phone and destructively investigating the A9. And to further complicate the situation, supposedly because of the yield difference between TSMC and Samsung, the majority of A9's (more than the 50% contractual stipulation) are coming from TSMC until whatever time in the future that Samsung gets their yields up.

But there is only one company that knows for sure, and that is Apple. I have no contacts or interaction with Apple. So its just a guess as to what game they are really playing at any given point in time with the foundries.

 

[DrMrLordX]

Can anyone explain me how GloFo is such a disaster? I mean, they used to be AMD fabs. Aren't they staffed by mostly the same people? I would have expected that AMD and GloFo had more synergy. Instead, is a type of ball and chain.

No, not really. I don't know that anyone can without insider knowledge. All I can add is that AMD, prior to divestiture of their fabs, licensed most/all of their process tech from others. I think?

 

[zir_blazer]

Come to think about it, it may be part of a more complex strategy. As GloFo seems to be a big responsible in making that what is left of AMD is sinking, you can say that they're directly making AMD fail. Makes for a much cheaper takeover if you can sabotage them that way, don't you think so?

 

[ShintaiDK]

Come to think about it, it may be part of a more complex strategy. As GloFo seems to be a big responsible in making that what is left of AMD is sinking, you can say that they're directly making AMD fail. Makes for a much cheaper takeover if you can sabotage them that way, don't you think so?

And takeover what? A dead patient on life support without hope of recovery?

No...

 

[zir_blazer]

AMD Intellectual Property is still there. That is not minor.

 

[ShintaiDK]

If it was worth the trouble, someone would have bought it now.

Remember, smaller AMD=Less Glofo revenue too.

 

[Fjodor2001]

http://www.kitguru.net/components/a...cts-using-second-gen-14nm-process-technology/

September 26th, 2015
GlobalFoundries: We started to tape-out chips using second-gen 14nm process technology

GlobalFoundries on Friday confirmed that the first products to be made using the company’s advanced 14nm LPP [low-power plus] manufacturing technology had been taped out. The contract maker of chips did not reveal any details, but indicated that prototype chips had demonstrated “excellent” performance and yields. One of the customers, who will use the 14LPP is Advanced Micro Devices.
[...]
“The performance-enhanced version of the technology (14LPP) is set for qualification in the second half of 2015, with the volume ramp beginning in early 2016,” said Jason Gorss, senior manager of corporate and technology communications at GlobalFoundries.

According to Mr. Gorss, the company started to tape-out products, which will be manufactured using the 14LPP a while ago
[...]
“Prototyping on test vehicles has demonstrated excellent logic and SRAM yields and performance at near 100% of target,” explained the official for GlobalFoundries.
[...]
Earlier this year Advanced Micro Devices confirmed that it had taped-out its first two products to be made using FinFET manufacturing technologies. According to unofficial information, AMD will use GlobalFoundries’ 14LPP fabrication process to produce its code-named “Summit Ridge” central processing unit with up to eight “Zen” cores, which will be marketed under “FX” and “Opteron” brands late next year.

http://news.softpedia.com/news/glob...ut-14nm-process-technology-chips-492865.shtml

Recent rumors that leaked in the press reveal a possible AMD foundry switch from GlobalFoundries to TSMC, since sources in the industry speak of troubles in building the 14nm wafers at GlobalFoundries, while TSMC still offers reliable 16nm FinFET wafers used by NVIDIA and Apple.

Probably to counter this wild rumor, the UAE-owned company confirmed that it managed to tape-out the first chips using the company’s 14nm LPP (low power plus) manufacturing technology.

To prove these wild rumors of AMD leaving GlobalFoundries aren't true, the Taiwanese foundry company insisted that the chips produced demonstrated “excellent” performance and yields, and said there were no issues that may prompt AMD to abruptly switch to TSMC. As we mentioned before, the main chips GlobalFoundries is referring to are more or less the AMD’s “Zen” processors

 

[Arachnotronic]

http://www.kitguru.net/components/a...cts-using-second-gen-14nm-process-technology/

http://news.softpedia.com/news/glob...ut-14nm-process-technology-chips-492865.shtml

To be perfectly blunt: Global Foundries is full of it.

 

[Phynaz]

Just read between the lines. They just admitted yield issues. Also, notice they have only taped out prototypes, no mention of customer tape outs.

We now have further confirmation of a possible reason for Zen's delay.

 

[ShintaiDK]

Quite clear they had no hope of ever producing their A9 allocation. And unless AMD moves Zen away from GloFo, that chip will be far far away.

 

[DrMrLordX]

So are we to take the word of WCCFTech over Kitguru and softpedia? Personally I think it all should be taken with a grain of salt. Show us the engineering samples or something, or the commercial products. Then we'll know more.

 

[ThatBuzzkiller]

It's strange how anyone has yet to find any Apple A9 samples that are produced from TSMC ...

Some people are so quick to count out GF without actually examining any further details ...

Note how they said that they were taping out "second-gen" (LPP) 14nm process and not "first-gen" (LPE) 14nm process, well so much for that parade ...

 

[NTMBK]

Come to think about it, it may be part of a more complex strategy. As GloFo seems to be a big responsible in making that what is left of AMD is sinking, you can say that they're directly making AMD fail. Makes for a much cheaper takeover if you can sabotage them that way, don't you think so?

Bleeding the life out of your main customer is not exactly great publicity. After seeing what has happened to AMD, who would want to be dependent on GloFo's ability to deliver?

 

[NTMBK]

To be perfectly blunt: Global Foundries is full of it.

Yup. Remember their press releases about 20nm tapeout? That worked out well...

 

[mrmt]

Bleeding the life out of your main customer is not exactly great publicity. After seeing what has happened to AMD, who would want to be dependent on GloFo's ability to deliver?

IBM. They even paid GLF for the honor.