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

[maddie]

7LP might have a slower ramp than normal as EUV won't be used at first. This should significantly reduce early wafer throughput. Can we then expect a lower volume die to be the first, using 7LP? Namely a GPU especially if designed using the same philosophy as Zen with very small die.

 

[tamz_msc]

I don't think this 12nm "half-node" will take 6-12 months to ramp up like 7nm. I wouldn't be surprised if we get a Zen refresh in March-April next year, exactly a year later Summit Ridge.

 

[ajc9988]

7LP might have a slower ramp than normal as EUV won't be used at first. This should significantly reduce early wafer throughput. Can we then expect a lower volume die to be the first, using 7LP? Namely a GPU especially if designed using the same philosophy as Zen with very small die.

7LP is taping out Q4 2017. This was said by Su in the past couple months. So that puts it in second half to Q1 2019.

 

[raghu78]

7LP is taping out Q4 2017. This was said by Su in the past couple months. So that puts it in second half to Q1 2019.

Yes. Lisa Su said they expect to tapeout 7nm chips before end of 2017. But there was no indication about which chips were being taped out and whether it was at TSMC or GF. AMD is using both TSMC and GF at 7nm. My guess is the discrete GPUs are made at TSMC and are likely to tapeout first. I think the Zen 2 CPU is made at GF and will tapeout in Q2 2018.

https://m.eet.com/content/images/eetimes/5 PDK roadmap x 800_1505974269.jpg

https://www.cadence.com/content/cad...d-full-flow-digital-and-signoff-tools-en.html

AMD is involved from the very early stages of the 7LP process creation and is likely to have started designing the Zen 2 CPU using the v0.5 PDK which was released a few months back. But i think tapeout will happen with v1.0 PDK in Q2 2018 (which I think is the procedure followed at the foundries). I do not think its possible for any volume production on 7LP in 2018. imo its a H1 2019 HVM ramp for GF 7LP. These processes are very complex and very hard to ramp to volume production at reasonable yields. I would say a 12 month timeframe for risk to volume production is the least (if not more).

 

[ajc9988]

Yes. Lisa Su said they expect to tapeout 7nm chips before end of 2017. But there was no indication about which chips were being taped out and whether it was at TSMC or GF. AMD is using both TSMC and GF at 7nm. My guess is the discrete GPUs are made at TSMC and are likely to tapeout first. I think the Zen 2 CPU is made at GF and will tapeout in Q2 2018.

https://m.eet.com/content/images/eetimes/5 PDK roadmap x 800_1505974269.jpg

https://www.cadence.com/content/cad...d-full-flow-digital-and-signoff-tools-en.html

AMD is involved from the very early stages of the 7LP process creation and is likely to have started designing the Zen 2 CPU using the v0.5 PDK which was released a few months back. But i think tapeout will happen with v1.0 PDK in Q2 2018 (which I think is the procedure followed at the foundries). I do not think its possible for any volume production on 7LP in 2018. imo its a H1 2019 HVM ramp for GF 7LP. These processes are very complex and very hard to ramp to volume production at reasonable yields. I would say a 12 month timeframe for risk to volume production is the least (if not more).

That's fine except that GF previously announced risk for H1 and volume for H2 2018.
https://www.electronicsweekly.com/news/business/7nm-customer-socs-globalfoundries-h1-2018-2017-06/

 

[Ajay]

Here's a thought.

The top overclocks on Zeppelin reach 4.2GHz (Threadripper reserved dies). Assuming that 10% boost increases Fmax and not just F at ISO power, that would mean the top dies would do 4.6GHz. Now we can't know what variation would look like on 12nm, but it is POSSIBLE those top dies would become more common, allowing them to be put into a Ryzen SKU like 2800X.

So on a best case scenario making plenty of assumptions, we're talking 4.6GHz XFR for a 2800X type SKU.

Main problematic assumptions with what I said:
- Fmax increases by 10%, and not just F at ISO power
- 12nm variation smaller than 14nm, and the curve trends towards the higher end rather than lower end
+ AMD made no tweaks over Zeppelin to lift frequency ever so slightly

Well said. Even though AMD said the is no uarch change for Pinnacle Ridge, I'm sure they had done a fair bit of bug fixes and small improvements. I wouldn't be surprised, for instance, to find that faster memory speeds work with Pinnacle Ridge (assuming the issue isn't some major architectural design limitation). In terms of variation in top clocks, that will depend heavily on process maturity and the yield curve as a function of clock speed. I expect AMD to make some nice improvements - and they will have to in order to compete with Intel's 10nm+ product line (multi-core 14nm++ products as well, given recent developments).

 

[raghu78]

That's fine except that GF previously announced risk for H1 and volume for H2 2018.
https://www.electronicsweekly.com/news/business/7nm-customer-socs-globalfoundries-h1-2018-2017-06/

The slides from Sep 20, 2017 at Globalfoundries Technology Conference 2017 are the latest and they clearly state H1 2018 for risk production (which is generally Q2 2018 if not they would clearly state Q1 2018). The time to go from risk to volume production is 12 months at least so volume production in Q2 2019 at the earliest. You have to filter the marketing talk from the real tech stuff.

 

[ajc9988]

The slides from Sep 20, 2017 at Globalfoundries Technology Conference 2017 are the latest and they clearly state H1 2018 for risk production (which is generally Q2 2018 if not they would clearly state Q1 2018). The time to go from risk to volume production is 12 months at least so volume production in Q2 2019 at the earliest. You have to filter the marketing talk from the real tech stuff.

So you make an assumption, not on company statements, when the current slide agrees with statements to investors given about 2-3 months ago which included volume production in H2 2018. So you see the problem here?

Stated another way, the most recent slide agrees with what they said 3 months ago in June. You then ignore the second part of the statement from June and substitute your own timeline in its place when the company has not yet given a reason to change the reliance on the prior statement. That I find problematic.

 

[raghu78]

So you make an assumption, not on company statements, when the current slide agrees with statements to investors given about 2-3 months ago which included volume production in H2 2018. So you see the problem here?

Stated another way, the most recent slide agrees with what they said 3 months ago in June. You then ignore the second part of the statement from June and substitute your own timeline in its place when the company has not yet given a reason to change the reliance on the prior statement. That I find problematic.

I am not making assumptions. I am being realistic about what to expect from Globalfoundries. The fact that we will see a 12LP Zen in 2018 means there is no 7nm Zen 2 in 2018. TSMC 7nm entered risk production in Q2 2017 and is expected to enter high volume production in Q2 2018. GF is not exempt from the laws of physics or the risks and difficulties involved in ramping a leading edge node from risk to volume production. Moreover GF has never been able to bring a new node to volume production on time with good yields, let alone speeden up things. 32nm SOI was delayed and had serious yield problems Their in house 14XM node was scrapped and they licensed 14LPP from Samsung and still they had yield issues. Even though GF acquired IBM in 2015 and have developed a in-house node they are yet to prove they can deliver on time and with competitive yields. The cycle times for 7nm DUV are longer and it will take 4-5 months from wafers in to actual finished chips. I don't expect a 7nm Zen 2 in retail before mid to Q3 2019. Thats my opinion. Others can disagree.

 

[Ajay]

The slides from Sep 20, 2017 at Globalfoundries Technology Conference 2017 are the latest and they clearly state H1 2018 for risk production (which is generally Q2 2018 if not they would clearly state Q1 2018). The time to go from risk to volume production is 12 months at least so volume production in Q2 2019 at the earliest. You have to filter the marketing talk from the real tech stuff.

It's really hard to tell this far out. I'm sure Pinnacle Ridge will be being produced during the 'risk production' period. AMD wouldn't be so foolish as to miss an opportunity to get data on the new chip as soon as possible. The question, and only one we care about, is when the 12LP process starts producing in sufficient quality (Fmax) and quantity (yield) for AMD to begin it's ramp to release. So it's a problem of both process and product maturity - neither of which we cannot know in advance. We can reasonably hope for sooner rather than later, but that it - mostly science and a bit of art will determine the real outcome. Some may reasonably assume a more pessimistic ramp and, that to is subject to the same unknown reality.

Personally, I hope AMD gets Pinnacle Ridge out in 3Q18 - for the sake of the company's success and no other reason. It's nice to see an underdog come back from near death.

 

[ajc9988]

I am not making assumptions. I am being realistic about what to expect from Globalfoundries. The fact that we will see a 12LP Zen in 2018 means there is no 7nm Zen 2 in 2018. TSMC 7nm entered risk production in Q2 2017 and is expected to enter high volume production in Q2 2018. GF is not exempt from the laws of physics or the risks and difficulties involved in ramping a leading edge node from risk to volume production. Moreover GF has never been able to bring a new node to volume production on time with good yields, let alone speeden up things. 32nm SOI was delayed and had serious yield problems Their in house 14XM node was scrapped and they licensed 14LPP from Samsung and still they had yield issues. Even though GF acquired IBM in 2015 and have developed a in-house node they are yet to prove they can deliver on time and with competitive yields. The cycle times for 7nm DUV are longer and it will take 4-5 months from wafers in to actual finished chips. I don't expect a 7nm Zen 2 in retail before mid to Q3 2019. Thats my opinion. Others can disagree.

I disagree, but only because the tapeout for 7nm was announced months before the TSMC announcement. Further, the agreement allows use of TSMC if they cannot fulfill the order on the timeline. Because of this and 9-15 months from tape out, I put Q4-Q1 2019, not Q3. Also, there is a chance they plan to have a certified risk design during tape out to then try to accelerate to market. Because of this and trends, they likely off loaded GPU chips to TSMC to make it more likely to meet demand of AMD by the timeline. As such, I think you are being overly pessimistic, especially considering what it will take to deliver the later EUV designs by late 2019 and into 2020.

 

[mikk]

The fact that we will see a 12LP Zen in 2018 means there is no 7nm Zen 2 in 2018.

Are you sure about 12LP for Pinnacle Ridge? If it comes in H1 2018 I would expect 14nm.

 

[ajc9988]

Are you sure about 12LP for Pinnacle Ridge? If it comes in H1 2018 I would expect 14nm.

On this, I think it is a low production run as a refresh. People mostly already bought the main chip. Those are not potential purchasers, for the most part. Because of that, seeing one at the beginning of the year and one in the last month or so, or first quarter of the following year, makes sense.

 

[raghu78]

It's really hard to tell this far out. I'm sure Pinnacle Ridge will be being produced during the 'risk production' period. AMD wouldn't be so foolish as to miss an opportunity to get data on the new chip as soon as possible. The question, and only one we care about, is when the 12LP process starts producing in sufficient quality (Fmax) and quantity (yield) for AMD to begin it's ramp to release. So it's a problem of both process and product maturity - neither of which we cannot know in advance. We can reasonably hope for sooner rather than later, but that it - mostly science and a bit of art will determine the real outcome. Some may reasonably assume a more pessimistic ramp and, that to is subject to the same unknown reality.

Personally, I hope AMD gets Pinnacle Ridge out in 3Q18 - for the sake of the company's success and no other reason. It's nice to see an underdog come back from near death.

Oh i definitely agree. I made the statement about Q2 2018 for start of risk production and Q2 2019 for start of volume production for 7LP. For 12LP the risk production is likely to be early Q1 2018 and the ramp to volume production should be much quicker as it builds on a mature 14LPP process. I expect Zen on 12LP to arrive in Q3 2018 and Zen 2 on 7LP to arrive in Q3 2019.

 

[ajc9988]

https://www.electronicsweekly.com/news/business/semiconductor-golden-age-coming-says-jha-2017-07/
July - CEO re-iterated the 2H 2018 volume production. Now, if they don't make that marker, they have to show a reason why due to securities law, based on false statements or omissions of material fact, which so include scienter (hard to prove here) and reliance. Certain forward looking statements are protected, but there can be an embedded facts component to the statement. As such, missing into Q1 could potentially be OK, but being off by a year may bring a shareholder lawsuit.

THIS IS NOT LEGAL ADVICE. THIS IS A DISCUSSION OF GENERAL LEGAL PRINCIPLES. THIS IS NOT AN INTERPRETATION OF LEGAL RIGHTS NOR IS MEANT TO DISCUSS ANY SPECIFIC RIGHTS THAT ANYONE MAY OR MAY NOT POSSESS. IF LEGAL ADVICE IS NEEDED, PLEASE SEEK THE ADVICE OF COUNSEL IN YOUR JURISDICTION FOR AN EVALUATION OF YOUR RIGHTS AND ADVICE RELATED TO SUCH RIGHTS, IF ANY.

 

[Abwx]

Are you sure about 12LP for Pinnacle Ridge? If it comes in H1 2018 I would expect 14nm.

From the OP :

"We are pleased to extend our longstanding relationship with GLOBALFOUNDRIES as a lead customer for their new 12LP technology,” said Mark Papermaster, CTO and senior vice president of technology and engineering, AMD. “Our deep collaboration with GF has helped AMD bring a set of leadership high-performance products to market in 2017 using 14nm FinFET technology. We plan to introduce new client and graphics products based on GF’s 12nm process technology in 2018 as a part of our focus on accelerating our product and technology momentum.”

 

[Phynaz]

Global Foundries' PDK roadmap

 

[Ajay]

Global Foundries' PDK roadmap

More confusion. Slide has no identifying items on it. Is 14LPP-XL now 12LP or something else all together.

 

[ajc9988]

More confusion. Slide has no identifying items on it. Is 14LPP-XL now 12LP or something else all together.

Pretty sure I recognize 7LP and know that AMD is one of the first in line for it...

 

[Ajay]

Pretty sure I recognize 7LP and know that AMD is one of the first in line for it...

Yes, but that has nothing to do with the thread topic.

 

[ajc9988]

Yes, but that has nothing to do with the thread topic.

The image likely was made before the announcement of 12LP. Because of that, the image itself would also be OT, as well as half the discussion in this thread on 7 and 14nm, even though ancillary in trying to figure out how the refresh fits in, something that has been of debate since the announcement of the refresh. So what is your point?

 

[Arachnotronic]

DIGITIMES reporting that AMD is going to launch Pinnacle Ridge in February 2018. Impressive schedule.

 

[Phynaz]

The image likely was made before the announcement of 12LP. Because of that, the image itself would also be OT, as well as half the discussion in this thread on 7 and 14nm, even though ancillary in trying to figure out how the refresh fits in, something that has been of debate since the announcement of the refresh. So what is your point?

You need to chill with your hostility about what's on topic. The slide is from the same event 12LP was announced at.

 

[ajc9988]

You need to chill with your hostility about what's on topic. The slide is from the same event 12LP was announced at.

Then, with this slide and context, the developer kit for 14LPP-XL would most likely apply to the 12LP process, considering here they are listed together, whereas the others fill out different areas on the diagram.

 

[Phynaz]

DIGITIMES reporting that AMD is going to launch Pinnacle Ridge in February 2018. Impressive schedule.

At the announcement Sanjay Jha said risk prduction of 12nm FinFET begins in the first half of 2018. Rick Merritt said "This makes sense because 12FinFET is a minor tweak of 14 and is needed to give current volume 14 users like AMD a 2018-2019 refresh."

I'm doubtful of Digitimes reporting. Seems like they are trying to get wccf's audience lately.