TSMC Shows Path to 16nm, Beyond

[kevinsbane]

Not at all. As I understand it by "final market value" TSMC means what their customers (e.g. Qualcomm) in turn sell their chips for to their customers (e.g. Samsung or Sony).

Saying that this compares to the sales of all products that have an Intel CPU in them is just ridiculous, because then you are adding the sales value of all other components in those products (e.g. SSD, motherboard, RAM, etc).

And even if you were right about the second point, what exactly is the relevance of the first? It's not as if the two numbers are comparable anyways - I mean, it's not as if anyone has the numbers for all of the money people put into upgrades of their processor (such as paying $200 to go from an i5 -> i7) on prebuilts either. It's a useless metric to use.

If you want to use a number to impress people, use one which actually makes sense. To compare "final market value" for one company to "total revenue" for another is not only useless, but actively misleading.

Here's a number which is actually useful.
TSMC Q2 Revenue: $5.1B
Intel Q2 Revenue: $12.8B

Edit:

And believing the article when it says that "TSMC is the largest integrated circuit manufacturer" is like believing Samsung when it says that their Note 3 is the fastest smartphone on the planet. Perhaps true, but still full of shenanigans and fuzzy math.

 

[dahorns]

And what is bread and butter for Intel really then?

Is this a serious question?

Look, what I'm saying isn't controversial at all. CAPEX as an isolated statistic is relatively useless in determining whether one competitor is gaining technologically on another. TSMC could be building another 200nm fab, they'd still take a CAPEX hit, but it certainly doesn't indicate they are catching up with Intel's cutting edge process.

TSMC might be catching up, but looking at CAPEX alone gives you a very narrow and inaccurate picture.

 

[Fjodor2001]

Here's a number which is actually useful.
TSMC Q2 Revenue: $5.1B
Intel Q2 Revenue: $12.8B

Yes, I agree that those are useful numbers too. But the problem is that Intel sells lots of different products apart from just chips, which also are included in those revenue figures. I'm talking about motherboards, SSDs, Ethernet controller cards, SW, etc. So unfortunately you cannot compare the value of the chip sales alone like that.

 

[blackened23]

So is this the thread where the fans of a certain firm pot shots at intel non stop? Another one? Cool.

 

[kevinsbane]

Yes, I agree that those are useful numbers too. But the problem is that Intel sells lots of different products apart from just chips, which also are included in those revenue figures. I'm talking about motherboards, SSDs, Ethernet controller cards, SW, etc. So unfortunately you cannot compare the value of the chip sales alone like that.

Then let us consider only the PC Client revenue - $8.1B.

PC Client Group: Delivering platforms designed for the notebook (including Ultrabook™ systems and convertible) and desktop (including high-end
enthusiast PCs) market segments; wireless and wired connectivity products

If you wish to make the case that wireless and wired connectivity products provide $3B revenue, you are free to do so. However, i think you will find it hard to make that case. I submit that in no way is that true, and that Intel is still the largest integrated circuit manufacturer in the world by any reasonable measure.

 

[6PPC]

What I do not understand is why people say Intel with all their advanced technology and so say years in front TSMC are unable to produce a SoC with Baytrail in 22nm that is suitable for cell phones, it is restricted to tablets.

No doubt they are years ahead with pc processors, but the technology used with SoC's, which is a different, Intel are scrabbling like mad to catch up.

I expect TSMC to be in full production with 20nm SoC chips at the same time as Intel with their 22nm SoC chips in 1Q14

 

[Arachnotronic]

What I do not understand is why people say Intel with all their advanced technology and so say years in front TSMC are unable to produce a SoC with Baytrail in 22nm that is suitable for cell phones, it is restricted to tablets.

No doubt they are years ahead with pc processors, but the technology used with SoC's, which is a different, Intel are scrabbling like mad to catch up.

I expect TSMC to be in full production with 20nm SoC chips at the same time as Intel with their 22nm SoC chips in 1Q14

There's a lot more that goes into making a suitable smartphone platform than SoC power. There are power-guzzling A15's that fit into smartphones, and even Krait 400 at full load burns more power than a Silvermont at full load.

Intel is rolling out smartphone-oriented "Merrifield" in Q4 2013 (for devices in Q1 2014) and then Moorefield (likely quad core) in 1H 2014.

 

[ShintaiDK]

I dont expect Intel to do a full push for smartphones until Airmont. But Silvermont really made a large impact in the tablet segment. Seems its a oneway battle, Intel just taking from ARM.

 

[krumme]

So is this the thread where the fans of a certain firm pot shots at intel non stop? Another one? Cool.

No its not. You can safely start at page one.

 

[blackened23]

What I do not understand is why people say Intel with all their advanced technology and so say years in front TSMC are unable to produce a SoC with Baytrail in 22nm that is suitable for cell phones, it is restricted to tablets.

No doubt they are years ahead with pc processors, but the technology used with SoC's, which is a different, Intel are scrabbling like mad to catch up.

I expect TSMC to be in full production with 20nm SoC chips at the same time as Intel with their 22nm SoC chips in 1Q14

As mentioned there are tons of ARM SOCs with very much worse power characteristics than Bay Trail being used for phones; Intel is making Merrifield as their "phone" chip, so to speak. The key metric for phones right now is integration, and by that metric Qualcomm is currently the leader. They are the only ones to my knowledge with integrated LTE, while everyone else is playing catch up to them. I expect this to change next year, and Merrifield is reported to include integrated LTE. We'll see what happens, I suppose.

As well, TSMC's 20nm is still worse than intel's 22nm with FinFETs, and intel has already mass produced Bay Trail on 22nm. So I really have no idea where your "1Q14" figure comes from, Bay Trail is being manufactured now. Next year, the successor on 14nm will be in production, as compared to TSMC's 20nm which will still be worse than intel's 22nm. TSMC as yet does not have FinFETs, Intel does.

 

[krumme]

As mentioned there are tons of ARM SOCs with very much worse power characteristics than Bay Trail being used for phones; Intel is making Merrifield as their "phone" chip, so to speak. The key metric for phones right now is integration, and by that metric Qualcomm is currently the leader. They are the only ones to my knowledge with integrated LTE, while everyone else is playing catch up to them. I expect this to change next year, and Merrifield is reported to include integrated LTE. We'll see what happens, I suppose.

As well, TSMC's 20nm is still worse than intel's 22nm with FinFETs, and intel has already mass produced Bay Trail on 22nm. So I really have no idea where your "1Q14" figure comes from, Bay Trail is being manufactured now. Next year, the successor on 14nm will be in production, as compared to TSMC's 20nm which will still be worse than intel's 22nm. TSMC as yet does not have FinFETs, Intel does.

You might add along this glorious story of Intel technological triumpf that TSMC will make better and better profit while Intel can get in a situation to cut capex yet more.

 

[blackened23]

Pfft, yeah, okay if you say so.

 

[krumme]

Besides the good assesments in the eetimes for the story what is perhaps interesting is the chairmans prerecorded talk to the group of engis; that there were still opportunities. But that they were getting smaller and smaller. That future better cooperation between customers and tsmc was needed to get a profit in a future market that would become more and more compettitive.

In my ears it tells small margins. And even better synergy needed to get that profit. I think its a message that the times that were is ending and they should be prepared to leaner times that demands new habits and openness to cooperation.

Its an important message imho right now when there is this sky high demand for tsmc products. Its a message it can not continue.

 

[jdubs03]

As mentioned there are tons of ARM SOCs with very much worse power characteristics than Bay Trail being used for phones; Intel is making Merrifield as their "phone" chip, so to speak. The key metric for phones right now is integration, and by that metric Qualcomm is currently the leader. They are the only ones to my knowledge with integrated LTE, while everyone else is playing catch up to them. I expect this to change next year, and Merrifield is reported to include integrated LTE. We'll see what happens, I suppose.

As well, TSMC's 20nm is still worse than intel's 22nm with FinFETs, and intel has already mass produced Bay Trail on 22nm. So I really have no idea where your "1Q14" figure comes from, Bay Trail is being manufactured now. Next year, the successor on 14nm will be in production, as compared to TSMC's 20nm which will still be worse than intel's 22nm. TSMC as yet does not have FinFETs, Intel does.

I expect Intel to have a fully capable integrated LTE solution (XMM 7160/7260), along with Nvidia in their Tegra 5.

When it comes to process matters, going along the same line of thought as yours, even when TSMC gets to 16nm FinFet its not a "real" 16nm (like one of you said). And this is at least 2H 2015. By that time Intel will have already been a year through 14nm and will be sorting through 10nm issues (new materials = III-V?).

But when it comes to performance of low-power SOC's I do not see much of an increase for Intel. Airmont (Cherry Trail) will be a significant overhaul, but I think the ARM vendors will also experience a solid bump especially going from 28nm to 20nm along with the A57 cores. The performance advantage Intel has will likely be similar, but since Bay Trail was their first push, Cherry Trail could shake up the dynamic even more.

Intel at 10nm vs the rest at 14/16nm FinFet is another story.

 

[hawtdawg]

I still don't see why Intel doesn't use this advantage to enter the GPU market. This gives them the ability to be competitive even if their architecture isn't as good as Nvidia or AMD.

 

[ShintaiDK]

I still don't see why Intel doesn't use this advantage to enter the GPU market. This gives them the ability to be competitive even if their architecture isn't as good as Nvidia or AMD.

Because the dGPU segment uses large dies, low margins and is without a future.

 

[hawtdawg]

Because the dGPU segment uses large dies, low margins and is without a future.

Considering that Nvidia is now a legit competitor to them in the server market place, I don't see how you can say this.

 

[jdubs03]

I still don't see why Intel doesn't use this advantage to enter the GPU market. This gives them the ability to be competitive even if their architecture isn't as good as Nvidia or AMD.

MIC/Xeon Phi is Intel's approach to get market share in the highly parallel HPC space, to compete against Nvidia (GK110/Titan), and AMD (Firepro). While Xeon Phi is a coprocessor, I think eventually Intel will correct the mistakes of Larrabee.

Intel will likely start integrating their MIC type features once Skylake comes out?, or around there. They are learning with Xeon Phi, eventually they are going to incorporate their exploits there into their PC Client CPUs, and maybe mobility later on.

 

[ShintaiDK]

Considering that Nvidia is now a legit competitor to them in the server market place, I don't see how you can say this.

Why do you think nVidia is trying all they can to get into something else than dGPU segment? And when Skylake and the next Xeon Phi shares AVX3.2. It will be pretty hard to even consider a Tesla or Firepro card. Plus the node difference is increasing. Intel will ship 14nm while nVidia and AMD still fiddles around on 28nm.

Not to mention 2Tflops GT4 Broadwell parts.

 

[Blitzvogel]

I still don't see why Intel doesn't use this advantage to enter the GPU market. This gives them the ability to be competitive even if their architecture isn't as good as Nvidia or AMD.

They already dominate Nvidia and AMD with the number of supplied GPUs. They might not have top end GPUs, but Intel HD graphics have come a long way. Beginning with HD 3000, Intel has been on a warpath.

 

[Idontcare]

I think anybody who follows this industry and TSMC's previous timelines will know not to fall for their "roadmaps" ;-)

Hey, IDC or anybody else, what in the hell does "risk production" mean? As far as I can tell, at guys like Intel, "risk production" means running wafers before a design is qualified for sale. What does it mean in TSMC parlance?

Risk production is something you call your production run when the process flow itself is still not functioning within the spec'ed electrical parametrics but you intend to tweak and fiddle with the process flow in parallel to your production run.

Everybody does it. It is all about managing time-to-market risk.

So let's say for example that your have a 20nm process flow which is spec'ed on paper as delivering 1.2mA/um drive current at 100pA/um off current. But in reality the process flow is delivering wafers with xtors that are both too cold (low drive current, say 0.9mA/um) and too leaky (say 200 pA/um), oh and the yield is a paltry 8%.

Obviously your R&D team is still hustling like mad to get the process flow tweaked so the electrical parametrics come in on target, and in parallel they are hustling like mad to get the yield to 80% instead of 8%.

But in the meantime there will always be a customer or two waiting in the wings who just cannot get their hands on the 20nm early production stuff soon enough. They'll pay a price-premium for it, on top of knowing full well that the yields are very bad and the electrical properties are not yet up to spec.

That is risk production. And if you are an IDM like Intel then your customer is an internal customer but a customer all the same. Someone will always want your buggy unpolished unfinished process node if for no other reason than to start getting through there own stages of design validation or early sales of limited run products.

 

[jpiniero]

Intel will ship 14nm while nVidia and AMD still fiddles around on 28nm.

And yet Tesla will be faster and better perf/watt. It's the same thing with mobile - they can win, but they'd have to dump x86 for something better. But they won't, and it will cost them.

 

[ShintaiDK]

And yet Tesla will be faster and better perf/watt. It's the same thing with mobile - they can win, but they'd have to dump x86 for something better. But they won't, and it will cost them.

A 14nm Xeon Phi would easily beat a 28nm Telsa in any metric. Xeon Phi already got the FirePro beaten in performance/watt. And the Xeon Phi is a lot more flexible and can handle more tasks than Tesla can.

 

[Arachnotronic]

And yet Tesla will be faster and better perf/watt. It's the same thing with mobile - they can win, but they'd have to dump x86 for something better. But they won't, and it will cost them.

That's probably the least informed statement I've seen on this board...today. You do realize ~2% of Xeon Phi's logic is x86, right? Most of the work is done by a fat vector unit.

 

[Arachnotronic]

Risk production is something you call your production run when the process flow itself is still not functioning within the spec'ed electrical parametrics but you intend to tweak and fiddle with the process flow in parallel to your production run.

Everybody does it. It is all about managing time-to-market risk.

So let's say for example that your have a 20nm process flow which is spec'ed on paper as delivering 1.2mA/um drive current at 100pA/um off current. But in reality the process flow is delivering wafers with xtors that are both too cold (low drive current, say 0.9mA/um) and too leaky (say 200 pA/um), oh and the yield is a paltry 8%.

Obviously your R&D team is still hustling like mad to get the process flow tweaked so the electrical parametrics come in on target, and in parallel they are hustling like mad to get the yield to 80% instead of 8%.

But in the meantime there will always be a customer or two waiting in the wings who just cannot get their hands on the 20nm early production stuff soon enough. They'll pay a price-premium for it, on top of knowing full well that the yields are very bad and the electrical properties are not yet up to spec.

That is risk production. And if you are an IDM like Intel then your customer is an internal customer but a customer all the same. Someone will always want your buggy unpolished unfinished process node if for no other reason than to start getting through there own stages of design validation or early sales of limited run products.

Thanks, Idontcare. You're the best!