GLOBALFOUNDRIES Introduces New 12nm FinFET Technology for High-Performance Applications

[turtile]

I bet this won't be in any normal AMD product. If the rumors are correct about AMD and Tesla making an automotive chip, this would be the process they would use....

 

[Ajay]

I bet this won't be in any normal AMD product. If the rumors are correct about AMD and Tesla making an automotive chip, this would be the process they would use....

You think AMD is going stand still on 14LPP until 7LP is ready?

 

[NostaSeronx]

22FDX/12FDX still better....

22FDX has been steadily improving, it is at the point of being around 10nm FinFET performance. The thing is the 12FDX node because of it has to go even further... it is better than 7nm FinFETs currently. The extra performance comes with the SOI cost, other than that 22FDX/12FDX is no more expensive than 28nm Bulk(22FDX) and 20nm Bulk(12FDX).

The fact that no one has taken this cheap, and excellent process. Means that there are a lot more mentally challenged folk than I thought.

GlobalFoundries hasn't even left Gate-First Metal-inserted Poly-si for: Gate First Full metal gate or Gate last full metal replacement gate. They haven't even touched stressed insulators, dual insulation, etc. The switch of gate tech increases mobility and improves electrostatics. The improved insulation tech would allow for increased body bias and faster switching of body biasing voltages. Insane stuff, to bad though.

 

[turtile]

You think AMD is going stand still on 14LPP until 7LP is ready?

Vega will likely move to 12LP but I think it's unlikely that AMD will release an improved CPU unless 7LP is delayed (unless14nm+ is the same as 12LP)

 

[Ajay]

Vega will likely move to 12LP but I think it's unlikely that AMD will release an improved CPU unless 7LP is delayed (unless14nm+ is the same as 12LP)

I have a hard time finding a reason for Pinnacle Ridge without the 10% performance gain from GF's 12LP process. I don’t think there is a 14LPP+, 12LP is it.

 

[Jan Olšan]

http://www.tomshardware.com/news/amd-ryzen-vega-12nm-lp-2018,35502.html

Tom's HW seems to have been present.

At the Global Foundries Technology Conference, AMD’s CTO Mark Papermaster announced that the company will be transitioning “graphics and client products” from the Global Foundries 14nm LPP FinFET process it uses today to the new 12nm LP process in 2018. Global Foundries also announced that 12LP will begin production in 1Q18.

Edit: Bolding mine. AMD is GloFo's prime customer so there is some chance that they were working with them early and managed to get a head start on the technology. However, they will still need some time to get some decent inventory before a CPU launch and there are logistics still. So we should definitely be careful and don't expect launches very early. Maybe GPUs could be brought out more agressively. Maybe.

We followed up with Papermaster in person and confirmed directly that the company will transition both Vega GPUs and the Ryzen line of processors to the 12nm LP process.

 

[krumme]

Its a "difficult" time for the usual news market when their usual business supporter is under pressure.
Look at this crap from thg
"However, this is nothing really new, as Intel has been ahead in process technology for quite some time, and if anything, third-party foundry companies such as Samsung/Global Foundries and TSMC have closed the gap a little once they adopted the FinFET structure for their transistors. "
Do they take us for morrons. Thg is nothing but a whore these days. But what should we complain as customers that dont even pay.

When 7nm high perf ibm process hits it wil be a huge change kissing feets for the new masters. Thg notebookcheck and heck even AT at a smaller degree will change their melody. It would be nice if we could get something just a bit less colored. There is a huge business opportunity for those that solve that dilemma.

 

[NTMBK]

22FDX/12FDX still better....

22FDX has been steadily improving, it is at the point of being around 10nm FinFET performance. The thing is the 12FDX node because of it has to go even further... it is better than 7nm FinFETs currently. The extra performance comes with the SOI cost, other than that 22FDX/12FDX is no more expensive than 28nm Bulk(22FDX) and 20nm Bulk(12FDX).

The fact that no one has taken this cheap, and excellent process. Means that there are a lot more mentally challenged folk than I thought.

Yes, I'm sure you know better than the hundreds engineers with decades of experience at AMD, Qualcomm, NVidia, Apple and Mediatek.

 

[ao_ika_red]

"However, this is nothing really new, as Intel has been ahead in process technology for quite some time, and if anything, third-party foundry companies such as Samsung/Global Foundries and TSMC have closed the gap a little once they adopted the FinFET structure for their transistors. "

That ticks me as well. But, let them be. As long as they able to keep their website free to the readers, I don't mind if they do "sponsored" news occasionally.

 

[DrMrLordX]

I think its highly likely that 12LP was developed by the IBM semiconductor team which was acquired by GF in 2015.

I'm sure they made significant contributions to the effort.

14LPP was the last officially licensed Samsung process by GF. I think 12LP and 7LP are developed by GF with the resources gained from the IBM acquisition.

7nm LP? I thought the process AMD was using on 7nm was HP . . .

Good speculation in this thread, but the 12" LP node is meant to capture the lucrative vinyl head market. As well as well as unsuspecting DJs and turntablists.

two turntables and a microphone

 

[raghu78]

I'm sure they made significant contributions to the effort.



7nm LP? I thought the process AMD was using on 7nm was HP . . .



two turntables and a microphone

GF 7LP stands for 7nm Leading Performance.

Sent from my SM-G935V using Tapatalk

 

[ajc9988]

That ticks me as well. But, let them be. As long as they able to keep their website free to the readers, I don't mind if they do "sponsored" news occasionally.

Move the quotations marks from sponsored to news and this reads better.

 

[Ajay]

Edit: Bolding mine. AMD is GloFo's prime customer so there is some chance that they were working with them early and managed to get a head start on the technology. However, they will still need some time to get some decent inventory before a CPU launch and there are logistics still. So we should definitely be careful and don't expect launches very early. Maybe GPUs could be brought out more agressively. Maybe.

That make some sense. AMD was probably in on discussion of the new process node earlier than anyone. Clearly, GF designed 12LP to require a minimum of changes to existing 14LPP designs. Risk production could apply to new customers, whereas AMD likely already has designs complete for 12LP and is already ramping up. IF these points are accurate, I could see a Q2 release for Pinnacle Ridge - which would be impressive. The only issue at this point is I don't recall any tapeout notification from AMD on 12LP, which may simply mean they are holding their cards very close to their chest.

 

[inf64]

That make some sense. AMD was probably in on discussion of the new process node earlier than anyone. Clearly, GF designed 12LP to require a minimum of changes to existing 14LPP designs. Risk production could apply to new customers, whereas AMD likely already has designs complete for 12LP and is already ramping up. IF these points are accurate, I could see a Q2 release for Pinnacle Ridge - which would be impressive. The only issue at this point is I don't recall any tapeout notification from AMD on 12LP, which may simply mean they are holding their cards very close to their chest.

Pinnacle Ridge on 12LP in Q2 '18 would be very impressive if it happens. Just about in time to get AMD needed MHZ push and fine tune Zen1 before next big update arrives in 2019.

 

[Ajay]

Pinnacle Ridge on 12LP in Q2 '18 would be very impressive if it happens. Just about in time to get AMD needed MHZ push and fine tune Zen1 before next big update arrives in 2019.

Yes. If Zen on 12LP were here now, I'd probably be buying that vs Coffee Lake. I overclock my CPUs and back them off of their max - that would probably mean 3.8 GHZ for Ryzen and likely much higher for Coffee Lake. The ~25% higher ST performance of Coffee is worth it to me (unless I change my mind at the last minute based on price!).

 

[Jan Olšan]

Risk production could apply to new customers, whereas AMD likely already has designs complete for 12LP and is already ramping up. IF these points are accurate, I could see a Q2 release for Pinnacle Ridge - which would be impressive. The only issue at this point is I don't recall any tapeout notification from AMD on 12LP, which may simply mean they are holding their cards very close to their chest.

I'm not so sure. There are several parts of bringing a chip out, some can be helped by being in the bed with the fab, some likely not.
1) tape-out: this probably yes, when you get advanced access and are active in the pre-production development, you can probably advance this stage quite a bit
2) mass production: here the process needs to mature and the advance in tape-out might be of limited usefulness. If the yield is not there or the clocks are low, you can't launch, even if you have the design ready. Like Intel with 10nm (☝ﾟ∀ﾟ)☝
3) having enough volume to be able to launch. This is more of the same as 2) but additionally also depends on the product, likely. For Ryzen, you would need a lot of chips so this adds weeks if not months of time buffer to release date, even excluding the logistics (shipping from factory to factory to package/test/finalise, then ferrying to the channel). With low-volume GPU, you need relativelly small inventory, but there is again time needed for the partners to make cards et cetera.

(note that I'm completely armchair-talking here, but these things are probably just common sense)

 

[Ajay]

I'm not so sure. There are several parts of bringing a chip out, some can be helped by being in the bed with the fab, some likely not.
1) tape-out: this probably yes, when you get advanced access and are active in the pre-production development, you can probably advance this stage quite a bit
2) mass production: here the process needs to mature and the advance in tape-out might be of limited usefulness. If the yield is not there or the clocks are low, you can't launch, even if you have the design ready. Like Intel with 10nm (☝ﾟ∀ﾟ)☝
3) having enough volume to be able to launch. This is more of the same as 2) but additionally also depends on the product, likely. For Ryzen, you would need a lot of chips so this adds weeks if not months of time buffer to release date, even excluding the logistics (shipping from factory to factory to package/test/finalise, then ferrying to the channel). With low-volume GPU, you need relativelly small inventory, but there is again time needed for the partners to make cards et cetera.

(note that I'm completely armchair-talking here, but these things are probably just common sense)

No, good point. I think a CPU takes 3 months to go from a wafer going into the fab to retail - so more like Q3 to get volume - assuming all goes well. It's based on an existing good node, 14LPP, so that should move things along quickly. AMD said no uarch changes for Pinnacle Ridge, so it's just a port to a very similar process with a minor shrink plus more speed. Mobos will need just a BIOS update for the new chips - it should go really well - on paper.

 

[Octoploid]

I think a CPU takes 3 months to go from a wafer going into the fab to retail.

No, that is way too short. Currently it is approaching half a year.

 

[raghu78]

I'm not so sure. There are several parts of bringing a chip out, some can be helped by being in the bed with the fab, some likely not.
1) tape-out: this probably yes, when you get advanced access and are active in the pre-production development, you can probably advance this stage quite a bit
2) mass production: here the process needs to mature and the advance in tape-out might be of limited usefulness. If the yield is not there or the clocks are low, you can't launch, even if you have the design ready. Like Intel with 10nm (☝ﾟ∀ﾟ)☝
3) having enough volume to be able to launch. This is more of the same as 2) but additionally also depends on the product, likely. For Ryzen, you would need a lot of chips so this adds weeks if not months of time buffer to release date, even excluding the logistics (shipping from factory to factory to package/test/finalise, then ferrying to the channel). With low-volume GPU, you need relativelly small inventory, but there is again time needed for the partners to make cards et cetera.

(note that I'm completely armchair-talking here, but these things are probably just common sense)

Since 12LP is based on a mature 14LPP the timeframe to move from risk to volume production might be much faster than what it takes on a brand new node which is usually 12 months. I am just guessing here without any technical knowledge as to the specifics of 12LP. I think 12LP has been designed around AMD's product requirements I think GF should be able ramp the process into HVM quickly . I think late Q2 or early Q3 2018 is a realistic timeline for Pinnacle Ridge.

No, good point. I think a CPU takes 3 months to go from a wafer going into the fab to retail - so more like Q3 to get volume - assuming all goes well. It's based on an existing good node, 14LPP, so that should move things along quickly. AMD said no uarch changes for Pinnacle Ridge, so it's just a port to a very similar process with a minor shrink plus more speed. Mobos will need just a BIOS update for the new chips - it should go really well - on paper.

Yes. The fact that 12LP is based on a mature 14LPP process should definitely help in the speed of the ramp from risk to high volume. 12LP is basically a process designed for AMD's needs which will benefit other GF customers too. imo TSMC designs processes for Apple and Nvidia and GF designs processes for AMD and IBM. The move to Pinnacle Ridge for consumers should be simple. Current AM4 motherboards will probably need BIOS updates to support the new processors. But thats about it.

 

[Ajay]

No, that is way too short. Currently it is approaching half a year.

I've always read 3-4 months. Has something changed. I am being optimistic here, but 6 months?!

 

[Octoploid]

I've always read 3-4 months. Has something changed. I am being optimistic here, but 6 months?!

On 7nm without EUV it will take ~5-6 months. If EUV could be inserted
(a big if), to get rid of the otherwise necessary multi-patterning on
some layers, then it will take a month less.

 

[Arachnotronic]

I've always read 3-4 months. Has something changed. I am being optimistic here, but 6 months?!

The more multi-patterning steps required, the longer the cycle times. Double patterning for the critical metal layers increased cycle time during the 20nm/16nm/14nm generation, and now the 10nm generation adds triple/quadruple patterning in some areas, further elongating cycle time.

 

[Phynaz]

That make some sense. AMD was probably in on discussion of the new process node earlier than anyone. Clearly, GF designed 12LP to require a minimum of changes to existing 14LPP designs. Risk production could apply to new customers, whereas AMD likely already has designs complete for 12LP and is already ramping up. IF these points are accurate, I could see a Q2 release for Pinnacle Ridge - which would be impressive. The only issue at this point is I don't recall any tapeout notification from AMD on 12LP, which may simply mean they are holding their cards very close to their chest.

I agree with everything you wrote except the schedule. I really doubt that something as large as Zen could be used for risk production, or that AMD could afford it. Remember risk production means that customer designs have not been manufactured yet. What happens if the design rules change? How about if a mask property has to be changed? Mask sets at 12nm are going to cost tens of millions of dollars, and that's capital, not expense.

 

[Arachnotronic]

I agree with everything you wrote except the schedule. I really doubt that something as large as Zen could be used for risk production, or that AMD could afford it. Remember risk production means that customer designs have not been manufactured yet. What happens if the design rules change? How about if a mask property has to be changed? Mask sets at 12nm are going to cost tens of millions of dollars, and that's capital, not expense.

TSMC's definition of risk production is when a technology reaches a certain level of reliability and small quantity production begins. There is usually a pretty significant amount of time, anywhere from 6 months to more than a year, between when a company begins risk production and when you start seeing products rolling off the lines and sold to customers.

I would be willing to bet that we don't see 12LP products in the market until 2H 2018.

 

[Ajay]

Well, wasn't there a slide saying 2H18 - so I guess that's it. It's ready when it's ready. Don't know why I was so inclined to be optimistic, guess I want things to go well for AMD.