EETimes: TSMC starts FinFETs in 2013, tries EUV at 10 nm

[podspi]

I liked that analyst near in the end, who asked about AMD's process nodes in 2014 when intel plans to have 14nm and AMD will be behind for 2 nodes.

The answer: We will focus on "user experience" ... lol .. you could hear Lisa's anger in her voice ^^

The guy even then asked again for clarification how AMD will be better than ...
Answer: Face recognition and low power ... yeah ... low power with 28m against 14nm ...

The bad thing: Lisa really didnt had an answer .. seems AMD really has no clue what to do in 2014 - besides 28nm.

Scary but what can they do? They are at the mercy of their 'partners'.

 

[SocketF]

Scary but what can they do? They are at the mercy of their 'partners'.

Yes true, I guess they simply don't plan a lot, because they dont trust GF's roadmap...
Concerning FD-SOI, as IDC said, adoption time is already too late for 28nm FDSOI, it would probably be ideal for 20nm FDSOI, GF already has some 20nm wafer starts, so they can see (the probably horrible ;-)) yields, but .. there is no 20nm FD-SOI node Next node is 14nm ... probably due to after 14XM. Too far away imo.

AMD didn't acknowledge revenues from console chips this quarter, but they did acknowledge royalties revenues, meaning that your hypothesis of AMD not handling manufacturing, just the design is rather probable.

Are you referring to the "Milestone-Payment"? That was from Nintendo, the CFO said that around the 23:00 minute mark.

 

[mrmt]

Are you referring to the "Milestone-Payment"? That was from Nintendo, the CFO said that around the 23:00 minute mark.

Thanks, that explains it, and makes my assumption wrong.

I'll have time to read AMD's results and Q&A by the weekend. Could you summarize the best items in the Q&A? That one about beyond 28nm was a hit.

 

[Ajay]

I liked that analyst near in the end, who asked about AMD's process nodes in 2014 when intel plans to have 14nm and AMD will be behind for 2 nodes.

The answer: We will focus on "user experience" ... lol .. you could hear Lisa's anger in her voice ^^

The guy even then asked again for clarification how AMD will be better than ...
Answer: Face recognition and low power ... yeah ... low power with 28m against 14nm ...

The bad thing: Lisa really didnt had an answer .. seems AMD really has no clue what to do in 2014 - besides 28nm.

Well if they are talking low power @ 28nm, then they are doing FD-SOI. Or she was referring to Kabini

Is the con-call available online to the public?

 

[SocketF]

Thanks, that explains it, and makes my assumption wrong.

I'll have time to read AMD's results and Q&A by the weekend. Could you summarize the best items in the Q&A? That one about beyond 28nm was a hit.

The full transcripts will be online at SeekingAlpha soon. So no need to do it on my own. Just be a bit patient.

Well if they are talking low power @ 28nm, then they are doing FD-SOI. Or she was referring to Kabini

Is the con-call available online to the public?

Yes it is available here:
http://ir.amd.com/phoenix.zhtml?c=74093&p=irol-eventDetails&EventId=4901035

You have to type a name and email, but this is not checked.

She used "low-power" without any deeper meaning, anything at AMD is low power nowadays, even without FD-SOI ...

 

[Ajay]

The full transcripts will be online at SeekingAlpha soon. So no need to do it on my own. Just be a bit patient.

Yes it is available here:
http://ir.amd.com/phoenix.zhtml?c=74093&p=irol-eventDetails&EventId=4901035

You have to type a name and email, but this is not checked.

She used "low-power" without any deeper meaning, anything at AMD is low power nowadays, even without FD-SOI ...

TY! Geez, no elaboration, sounds like she was totally unprepared for the question. I guess in their quest to reduce costs, AMD execs have opted to scratch doing dry runs before the actual call.

 

[SocketF]

I'll have time to read AMD's results and Q&A by the weekend. Could you summarize the best items in the Q&A? That one about beyond 28nm was a hit.

It's online now:
http://seekingalpha.com/article/135...-2013-results-earnings-call-transcript?page=2

 

[mrmt]

It's online now:
http://seekingalpha.com/article/135...-2013-results-earnings-call-transcript?page=2

Thanks. I didn't go too deep in the numbers or in the conference, but it seems that they have pretty much stabilized the company, which means that they might have time to wait for their efforts to bear fruits. What remains to be seen is the Haswell impact on their Q3 numbers. As AMD won't have a true answer to Haswell, just half-baked upgrade, the impact might be something interesting.

We can already see a change in AMD choice of words. While in the previous quarter they were keen to show that they still had a hand on the PC market, today they are showing that they are probing other markets and those other markets will be able to sustain the company. For the PC market their only forecast is that it will remain important on AMD revenues, but didn't tell much more. I finished the transcript with the feeling that they are slowly retreating from the PC market, and whatever last stand they plan to do it will be with Kabini and Temash.

BTW, Lisa put Kabini against Pentium, Celeron and even i3. I don't think it will be able to compete against i3, but this is not the most important point here. What Lisa said is that Kabini will compete against Trinity too, and this could spell more WSA trouble, if they are not to manufacture console chips in GLF.

Debt is also weighting. Today debt expenses are 10% of their OPEX, not a good prospect for a company bleeding cash.

The lack of plans for their post-28nm offers, or at least lack of plan they could talk about, was stunning. They simply said that they don't have anything to compete against Intel and that they will have to resort to price cuts. Whatever plan they have to embedded, they must succeed in implementing not before their cash runs out, but also before Intel start to mass production of 14nm chips. Once they do, AMD is done for in the market they are competing now.

Break even at Q313 also vanished from their financial statements. What we have now is cash flow positive in H213, and they are stubbornly refusing to give guidance, yet another bad prospect.

All in all, I'm expecting Q213 to be uneventful, the next important milestone will be Q313. From there we'll see how Haswell is impacting their revenues, how strong Kabini and Temash are in generating cash and how much money they will make from consoles. I wouldn't call their situation good, but it's under control. If they can execute they will still bleed cash but at least have a chance in pivoting to embedded.

 

[AtenRa]

What remains to be seen is the Haswell impact on their Q3 numbers.

First hasewell CPUs will be Core i5 and Core i7, those are less than 5-10% of market share and could only impact the FX line.

As AMD won't have a true answer to Haswell, just half-baked upgrade, the impact might be something interesting.

Kaveri is scheduled for H2 2013 release, thats the answer to Haswell APUs.

The lack of plans for their post-28nm offers, or at least lack of plan they could talk about, was stunning.

They (AMD) may not want to disclose what they are planning after Kaveri/Steamroller right now. They are focusing on Kabini/Temash and Kaveri as of now.

 

[AtenRa]

The full transcripts will be online at SeekingAlpha soon. So no need to do it on my own. Just be a bit patient.

Yes it is available here:
http://ir.amd.com/phoenix.zhtml?c=74093&p=irol-eventDetails&EventId=4901035

You have to type a name and email, but this is not checked.

She used "low-power" without any deeper meaning, anything at AMD is low power nowadays, even without FD-SOI ...

She didnt say that,

Lisa Su

Yeah, we believe that Kaveri and the heterogeneous systems architecture will allow us to do things like natural user interface processing much more efficiently, especially in low-power environments, so things like facial recognition, speech recognition, those types of sort of graphics intensive types of applications we will be able to do at a lower power with higher performance.

 

[SocketF]

All in all, I'm expecting Q213 to be uneventful, the next important milestone will be Q313. From there we'll see how Haswell is impacting their revenues, how strong Kabini and Temash are in generating cash and how much money they will make from consoles. I wouldn't call their situation good, but it's under control. If they can execute they will still bleed cash but at least have a chance in pivoting to embedded.

Yes I agree, but Q2 will be interesting regards the number of design wins for Kabini. If there will be a lot of designs from top OEMs, then everything should be fine .. if not .. well ..

They (AMD) may not want to disclose what they are planning after Kaveri/Steamroller right now. They are focusing on Kabini/Temash and Kaveri as of now.

Then she should have said, that they dont disclose it yet. Focusing now at Kabini and Kaveri is a very risky plan, as IDC pointed out once, you need a 3y lead time to get a processor out of the fab. The longer AMD waits the more they will lag behind intel.
What it means is that they don't care so much about intel, because now they are bling bling low-power but intel is intel aka the 600 pound gorilla, they will have 14nm soon and their GPU-parts are getting better, too. At least she could said something about planned shrinks or Beema.

She didnt say that,

Speaking about "low-power environments" in the context of a 28nm chip, compared to a 14nm one is in my opinion meaningless.

 

[AtenRa]

Speaking about "low-power environments" in the context of a 28nm chip, compared to a 14nm one is in my opinion meaningless.

35W TDP Trinity is 32nm and has better performance in Graphics than 45W TDP 22nm Intel FinFet Core i7. Also, Kabini at 28nm will beat 32nm and 22nm ATOM and Celeron/Pentium, not to mention Kaveri at 28nm and GCN/HSA.

One more thing, I don’t see 14nm products before H2 2014 the earliest (I will not be surprised if they will launch at Q1 2015)

 

[mrmt]

Then she should have said, that they dont disclose it yet. Focusing now at Kabini and Kaveri is a very risky plan, as IDC pointed out once, you need a 3y lead time to get a processor out of the fab. The longer AMD waits the more they will lag behind intel.

And Intel isn't the only game in town in the new AMD markets. ARM will be there, and most likely with better manufacturing capabitilies than AMD by 2014. I'm rather curious to see what happens once you have subpar ARM64 cores competing against 20nm cores and subpar 28nm x86 core competing against ARM 20nm cores.

Their only chance is embedded, where reliability and costs are far more important than performance.

 

[R0H1T]

And Intel isn't the only game in town in the new AMD markets. ARM will be there, and most likely with better manufacturing capabitilies than AMD by 2014. I'm rather curious to see what happens once you have subpar ARM64 cores competing against 20nm cores and subpar 28nm x86 core competing against ARM 20nm cores.

Their only chance is embedded, where reliability and costs are far more important than performance.

Oh so you're saying that AMD/ARM both will be subpar in 2014, what a surprise ! But guess what, in this day & age over 90% of all chips sold are subpar(according to your defintion) & no one cares if "can run it crysis 3" or not ! The only major chipmaker getting squeezed harder going forward is Intel cause they don't make cheap stuff, which btw does 90% of your avg Joe's work !

 

[mikk]

Kaveri is scheduled for H2 2013 release, thats the answer to Haswell APUs.

It isn't. Only shipments will start late in Q4. Richland shipment started Q4 2012 and there is no Richland notebook/APU available yet.

 

[mrmt]

Oh so you're saying that AMD/ARM both will be subpar in 2014, what a surprise ! But guess what, in this day & age over 90% of all chips sold are subpar(according to your defintion) & no one cares if "can run it crysis 3" or not ! The only major chipmaker getting squeezed harder going forward is Intel cause they don't make cheap stuff, which btw does 90% of your avg Joe's work !

I'd suggest you to tone down a bit and actually read what I'm writting, because I never said what you wrote in the quoted paragraph.

What I said is that AMD will have subpar offers in both markets, in x86 for obvious reasons, and in ARM because they'll be stuck in 28nm with GLF and a vanilla core, while most of the 1st tier players will have 20nm offers with custom cores.

Now you can argue or disagree with this point like an adult and we'll do fine and learn from each other, or you can proceed with the diatribe and I'll ignore you from now on, just like I do with some famous trolls here.

 

[Phynaz]

35W TDP Trinity is 32nm and has better performance in Graphics than 45W TDP 22nm Intel FinFet Core i7.

You should just put this in your sig.

 

[AtenRa]

You should just put this in your sig.

No need, i can give you the AT review link

http://www.anandtech.com/show/5831/amd-trinity-review-a10-4600m-a-new-hope

 

[R0H1T]

I'd suggest you to tone down a bit and actually read what I'm writting, because I never said what you wrote in the quoted paragraph.

What I said is that AMD will have subpar offers in both markets, in x86 for obvious reasons, and in ARM because they'll be stuck in 28nm with GLF and a vanilla core, while most of the 1st tier players will have 20nm offers with custom cores.

Now you can argue or disagree with this point like an adult and we'll do fine and learn from each other, or you can proceed with the diatribe and I'll ignore you from now on, just like I do with some famous trolls here.

I dunno why you continue with this idiom "subpar" for AMD as if they haven't done much on 32nm FDSOI ? I've already said that going forward price will be the biggest factor outside of the high end enthusiast/server market which isn't that big as compared to the overall consumer electronics market & is shrinking btw, the latter also has the biggest potential for growth.

For Intel, if they wish to be a major player in this arena, will have to lower their prices or alternatively ramp up performance to 2~3x ARM/AMD low power chips if they are to maintain their margins & overall revenues, however I can't see them follow both of these routes at the same time ! If they are to increase revenue they will have to lower margins, unless they plan to fab chips for others, thereby going forward we won't see another node shrink on schedule for their tick/tock by which time ARM/AMD will make further inroads into the traditional Intel strongholds.

 

[SocketF]

I have a rather fundamental question now:

How "short" can you run a semiconductor company "on sight"? The whole story with Lisa's evasion tactics about the manufacturing processes, empty roadmaps, and the sudden pop-up of Richland, then the bitter experiences with GF's 28nm process in 2012 and the cancellations of Krishna let me believe that the AMD bosses are trying to steer the ship "on sight" only, without a detailed long term plan.

Also Raja Koduri who just came back from apple told anandtech:

Today, AMD is a much smaller and more agile company

http://www.anandtech.com/show/6907/the-king-is-back-raja-koduri-leaves-apple-returns-to-amd

Which further consolidates my pov in my opinion.

But ... I wonder how "short" you can do that business. IDC previously mentioned a 3-year period that is needed for designing and producing a chip. Now the question to me is, how much of that 3 years can you cut off, e.g. by shipping A-revisions of your chips (which AMD is doing now with their APUs) or by using synthesized designs, or by using I don't know what ...

Maybe an example question:
If AMD would decided to make an FD-SOI @28nm version of Kabini right now, could they have that chip next year? (Assuming that GF wont be a problem).

 

[Idontcare]

I have a rather fundamental question now:

How "short" can you run a semiconductor company "on sight"? The whole story with Lisa's evasion tactics about the manufacturing processes, empty roadmaps, and the sudden pop-up of Richland, then the bitter experiences with GF's 28nm process in 2012 and the cancellations of Krishna let me believe that the AMD bosses are trying to steer the ship "on sight" only, without a detailed long term plan.

Also Raja Koduri who just came back from apple told anandtech:
http://www.anandtech.com/show/6907/the-king-is-back-raja-koduri-leaves-apple-returns-to-amd

Which further consolidates my pov in my opinion.

But ... I wonder how "short" you can do that business. IDC previously mentioned a 3-year period that is needed for designing and producing a chip. Now the question to me is, how much of that 3 years can you cut off, e.g. by shipping A-revisions of your chips (which AMD is doing now with their APUs) or by using synthesized designs, or by using I don't know what ...

Maybe an example question:
If AMD would decided to make an FD-SOI @28nm version of Kabini right now, could they have that chip next year? (Assuming that GF wont be a problem).

In their current business, designing ICs, they absolutely must know today (internally) what they plan to be bringing to market next year, the year after and so on for the next 4yrs because if they want to bring them to market in their respective launch year for the next 4yrs then they need to already be at specific project milestones on this day.

As for the 28nm FDSOI Kabini...it all comes down to just how process-transparent the FDSOI vs Bulk situation is. STM says it is dead-easy to move from 28nm bulk to their 28nm FDSOI.

If they are telling the truth then yes, it would be 1yr for AMD's engineers to vet the validation and qualification cycle for the chip on the new process tech.

If STM is not telling the whole truth then you can start to add another year, or two, to that timeline as more and more of the front-end (xtor stuff) has to be re-laid out.

 

[SocketF]

In their current business, designing ICs, they absolutely must know today (internally) what they plan to be bringing to market next year, the year after and so on for the next 4yrs because if they want to bring them to market in their respective launch year for the next 4yrs then they need to already be at specific project milestones on this day.

Ok, but do they already have to know which process they are going to use?
Cant they just work on the general Excavator module design and then in later stages
a) Decide on specific SoC configurations (How many cores / L3 / GPU-Shaders / PCIe / RAM-Channels / whatever )
b) implement that to a specific process?

As for the 28nm FDSOI Kabini...it all comes down to just how process-transparent the FDSOI vs Bulk situation is. STM says it is dead-easy to move from 28nm bulk to their 28nm FDSOI.

Well .. I guess STM's statement was only valid for other bulk designs based on GF's Low-Power libraries. Kabini however is based on TSMC High-Performance libs... no clue what effects that has, but that probably means more than 1 year then.

Now I just remember the CEO's statement about using "standard processes" and an easier exchange of designs between different fab companies.

I guess his plan could really be to use TSMC as the "bleeding edge" partner and then being able to move the design easily over to GF's fab a year later, or whenever they are ready ...

 

[mrmt]

I dunno why you continue with this idiom "subpar" for AMD as if they haven't done much on 32nm FDSOI ?

When you have to resort to sell 246mm^2 chips cheaper than your competitor sell 110mm^2 chips, when you have to resort to breach your own already scandalous 125W specifications in order to be competitive with your competitor's 77W solution, when your market share in a given market goes to less than 2% because nobody wants your products, it is obvious that you fell behind in the technology race and is using a group of subpar solutions, and this is no offense, it's just an obvious fact.

But Globalfoundries fell so behind in the technology race that what you see with Intel will happen with their ARM offers too, because GLF cannot deliver a node of the same efficiency of TSMC in the same time frame. They are a subpar foundry too.

I've already said that going forward price will be the biggest factor outside of the high end enthusiast/server market which isn't that big as compared to the overall consumer electronics market & is shrinking btw, the latter also has the biggest potential for growth.

This.

This is essentially what AMD marketing has been saying, but on the other side we have Intel, Nvidia, Qualcomm, Samsung and others actually developing bleeding edge solutions that they think will yield more money than a vanilla low-cost solution that AMD will give us. And between putting my money in any of those player and put my money in AMD, I guess the choice is clear.

 

[mrmt]

But ... I wonder how "short" you can do that business. IDC previously mentioned a 3-year period that is needed for designing and producing a chip. Now the question to me is, how much of that 3 years can you cut off, e.g. by shipping A-revisions of your chips (which AMD is doing now with their APUs) or by using synthesized designs, or by using I don't know what

I posted some two or three Rory speeches and in these was very keen about "reusable" IP, meaning that he wants to avoid development of all-custom chips, and instead just put pieces together and then validate only the overall product, not just the specific part. In other words, sacrifice the scope in order to keep schedule and costs under control.

This is more or less what they did with the PS4 and Xbox. They took a GPU they had, they took a core they had, they already had the interconnect between the two parts, they just had to bolt whatever MSFT and Sony wanted and they were set. They didn't need 4 years to actually develop the PS4 chip, probably 2-3 years. These two chips won't be as bleeding edge as the PS3 chip for example, but costs should be lower and developed in a smaller time frame.

So this is where I think the "agile" comment fits.

 

[FwFred]

35W TDP Trinity is 32nm and has better performance in Graphics than 45W TDP 22nm Intel FinFet Core i7.

What is the max power usage (w/ Turbo) for the GPU in 45W Ivy? How about for GPU in 35W Trinity? How many mm2 was reserved for the GPU in either?

Do you think the Haswell GPU power budget may go up with GT3 or GT3e?