TSMC signs chip deal with Apple

[NTMBK]

Rumours say the Xbox is fabbed at GF.

Seems likely, given that GloFo is on IBM's common platform- rumours say that IBM had a hand in the ESRAM for the XBone.

 

[SiliconWars]

Globalfoundries has yet to get a bleeding edge customer for them, AMD was doing everything it could to get away from their foundry "partner", they lost 600 million dollars last year... You know, silly facts outside GLF's PR-land that shows that things are not good for Globalfoundries. Not bad for a very "successful business", isn't it?

I don't think you have any PR statement to counter this "FUD", do you?

GF doesn't need a bleeding-edge customer, they can't produce enough wafers to meet demand already!

What do you expect them to do, just magic up another 5 foundries overnight so they are on the same level as TSMC?

 

[ShintaiDK]

GF doesn't need a bleeding-edge customer, they can't produce enough wafers to meet demand already!

What do you expect them to do, just magic up another 5 foundries overnight so they are on the same level as TSMC?

Why are they running at a loss if their fabs are fully utilized? And why reduce investment then?

 

[mrmt]

I'll consider that you do not have anything to show regarding the "FUD" I brought here. That's bad, no? How can you call out someone like that and don't have anything to show? But let's dig into your pie-in-the-sky assumption here:

GF doesn't need a bleeding-edge customer, they can't produce enough wafers to meet demand already!

What do you expect them to do, just magic up another 5 foundries overnight so they are on the same level as TSMC?

They don't need a bleeding edge customer? Even losing 600 million dollars a year? How do you expect them to pay for that funny half-baked finfet node of them? With TSMC leftovers?

And they can't produce wafers to meet demand? Where are the great 28nm orders that are eating all GLF capacity? You know, 28nm, a node that is operational in TSMC since 2011, that the equivalent is fully deprecated at Intel, and last, but not least the node that AMD paid to not have to use it.

What we are seeing here is a foundry with bleeding edge costs and lagging edge revenues, and more important, a foundry that nobody with a bleeding edge business worth something is betting on it being a reliable foundry partner. That's far from the successful business you are trying to portray here.

 

[SiliconWars]

I'll consider that you do not have anything to show regarding the "FUD" I brought here. That's bad, no? How can you call out someone like that and don't have anything to show? But let's dig into your pie-in-the-sky assumption here:

Let me point out your FUD, so you understand exactly what I mean.

there is plenty of capacity within Samsung Foundry Division to spare, and more important, capacity could be on the cheap.

- The same can be said about Globalfoundries.

FUD. As I have already proven, there is no "plenty of capacity spare at GF". If there was they wouldn't be in a constant stage of expansion and be the fastest growing semi in the world.

In Globalfoundries' case, we know the reason nobody goes to them:

Nobody goes to GF? How come they have 150 customers? FUD.

Their PR slides haven't anything to do with reality, so unless you are looking for something far behind the bleeding edge, Globalfoundries is not for you

FUD. 28nm Rockchip SoC in production

So let's see..according to you GF has tons of spare, cheap capacity, no customers and "far from bleeding edge" technology. You're a MIRACLE, you got nothing right.

And they can't produce wafers to meet demand? Where are the great 28nm orders that are eating all GLF capacity? You know, 28nm, a node that is operational in TSMC since 2011, that the equivalent is fully deprecated at Intel, and last, but not least the node that AMD paid to not have to use it.

See above, your information on GF appears to be oh...3 years out of date? Do you even bother reading about other semi's elsewhere or do you just get your outdated information from this forum?

What we are seeing here is a foundry with bleeding edge costs and lagging edge revenues, and more important, a foundry that nobody with a bleeding edge business worth something is betting on it being a reliable foundry partner. That's far from the successful business you are trying to portray here.

What we're seeing is a 3 year old business with increasing revenues making them the worlds fastest growing semi and your ludicrous attempt to make it look as if they are failing.

How many foundry customers does Intel have? What happened to Apple, wasn't that supposed to be a given? How come Intel lost Apple to TSMC? While I'm at it, wasn't it Intel who scythed production recently? If anyone is sitting with empty fabs, it's them.

 

[mrmt]

FUD. As I have already proven, there is no "plenty of capacity spare at GF". If there was they wouldn't be in a constant stage of expansion and be the fastest growing semi in the world.

I wasn't talking about capacity in general but capacity at the bleeding edge, because you obviously know that Apple uses bleeding edge nodes and this is the capacity vacated by Apple at Samsung, right?

And where's GLF growing, in the bleeding edge or in the lagging edge? The numbers you brought here just show that revenues are increasing, not that revenues are increasing in the bleeding edge. Same with your customer numbers. 150 nobodys isn't anything great. LOL at Rock Chip, they are not bleeding edge, sorry.

Oh, FYI, it's easy to grow when you have a cash cushion to support losses. That's what you call buying market share in business parlance. Right now, the worlds fastest growing semi is also one of the top money losers of the world, together with... AMD.

 

[SiliconWars]

Apple uses bleeding edge nodes does it? That's a new one to me.

Let's see about GF - currently fabbing Rockchip's SoC (28nm), Kaveri (28nm) starting very soon, Xbox One (28nm) probably already in production...what is Apple doing that is more on the "bleeding edge"?

 

[mrmt]

Apple uses bleeding edge nodes does it? That's a new one to me.

No, they don't. 20nm isn't bleeding edge at TSMC. 32nm wasn't at Samsung.

 

[SiliconWars]

No, they don't. 20nm isn't bleeding edge at TSMC. 32nm wasn't at Samsung.

So which one is it, last post you were saying Apple uses bleeding edge nodes? Do they or don't they? What does Apple need that GF couldn't provide, had they had the spare capacity?

Oh btw, GF is expanding at Singapore and NY - $2.3 billion to expand Fab 8 (28nm and below) http://www.bizjournals.com/albany/blog/2012/07/globalfoundries-moves-forward-on-23b.html so there goes your theory that they are only selling old tech.

"Most leading-edge R&D will be done here," he said, focusing on development of future computer chips.
“Today we see increasingly strong demand from our customers,” Choh said.

 

[Exophase]

Exophase, question for you:

- After Apple departure there is plenty of capacity within Samsung Foundry Division to spare, and more important, capacity could be on the cheap.

- The same can be said about Globalfoundries.

- In Globalfoundries' case, we know the reason nobody goes to them: Their PR slides haven't anything to do with reality, so unless you are looking for something far behind the bleeding edge, Globalfoundries is not for you.

- We didn't hear anything about Samsung having any kind of problems with their node. Things seem to be ok.

Given the above, why aren't bleeding edge customers flocking on Samsung's door to get deals on the cheap instead of having to fight (and pay dearly) for capacity on TSMC?

Samsung released their first 32nm product long after GF did (over a year). They released their first 28nm product shortly after GF did. Apple hasn't released a 28nm Samsung product yet. Does that answer your question at all?

If Samsung has yield problems or not we wouldn't necessarily even hear about it because almost no one external is using them to begin with and there's no point for Apple to complain publicly if they're having them. Especially when they're already in the process of transitioning to TSMC.

At any rate, the Apple departure hasn't actually happened yet and transitions to new foundries take a long time, so it's way too early to conclude that no one is interested in using them - it's not like any such interest is guaranteed to be publicly documented well in advance.

 

[mrmt]

Oh btw, GF is expanding at Singapore and NY - $2.3 billion to expand Fab 8 (28nm and below) http://www.bizjournals.com/albany/blog/2012/07/globalfoundries-moves-forward-on-23b.html so there goes your theory that they are only selling old tech.

Isn't it? GLF is expanding their 28nm capacity when Intel is building 14nm capacity and TSMC is building 20nm capacity. They will pick up TSMC leftovers and the guys that are in 65/40nm now.

 

[Exophase]

Seems likely, given that GloFo is on IBM's common platform- rumours say that IBM had a hand in the ESRAM for the XBone.

Those rumors must be coming from people who don't understand that eSRAM is trivial to implement. They must have thought "well, IBM manufactured XBox 360 and thar had eDRAM, and eSRAM is more or less the same thing so they probably got IBM's help." Of course IIRC not all XBox 360s even used IBM manufactured eDRAM dies, there were some NEC ones in there.

I wouldn't have expected the console chips to be fabbed at GF instead of TSMC but I also didn't expect XBox One's chip to be nearly as huge as it is so I don't really know. I'm sure AMD would be happy to fulfill some of their obligations to GF this way.

 

[Exophase]

How many foundry customers does Intel have? What happened to Apple, wasn't that supposed to be a given? How come Intel lost Apple to TSMC? While I'm at it, wasn't it Intel who scythed production recently? If anyone is sitting with empty fabs, it's them.

IMO that whole Apple using Intel rumor was pretty silly to begin with, Intel has been very clear that they're not interested in opening up their fabs to enable competitors and Apple would very much be a competitor. I think the logic behind this rumor was that Apple was supposed to in turn use Intel SoCs in iPads.. basically both companies doing something totally against their business model which I guess was supposed to cancel each other out.

 

[Idontcare]

If Samsung has yield problems or not we wouldn't necessarily even hear about it because almost no one external is using them to begin with and there's no point for Apple to complain publicly if they're having them. Especially when they're already in the process of transitioning to TSMC.

Samsung has no yield problems. In fact it is the opposite. Samsung's yields are higher than Fishkill's yields running in parallel on the exact same development node.

It has caused a huge ruckus amongst the fab-club because IBM management won't allow the Samsung yield enhancement ideas to be back-fed into the fab-club so that everyone's yields can be elevated.

Why you might ask? Because IBM wants to retain sole control and ownership of the licensable IP that is captured by the fab-club nodes. Samsung's engineers upstaging IBM's engineers is too much for the entrenched NIH syndrome to cope with.

 

[mrmt]

Samsung released their first 32nm product long after GF did (over a year). They released their first 28nm product shortly after GF did. Apple hasn't released a 28nm Samsung product yet. Does that answer your question at all?

No, it doesn't. If anyone plans to release something in Samsung's bleeding edge node (20nm) by H214/H115 they must have an agreement *now*, and so far we haven't heard anythig.

At any rate, the Apple departure hasn't actually happened yet and transitions to new foundries take a long time, so it's way too early to conclude that no one is interested in using them - it's not like any such interest is guaranteed to be publicly documented well in advance.

It might not have happened from a contractual POV, but from an engineering POV, yes, it did. Apple design teams aren't working Samsung guys to develop their future chips there, all the engineering guys are with TSMC design rules books optimizing Apple designs for it.

 

[Exophase]

Samsung has no yield problems. In fact it is the opposite. Samsung's yields are higher than Fishkill's yields running in parallel on the exact same development node.

It has caused a huge ruckus amongst the fab-club because IBM management won't allow the Samsung yield enhancement ideas to be back-fed into the fab-club so that everyone's yields can be elevated.

Why you might ask? Because IBM wants to retain sole control and ownership of the licensable IP that is captured by the fab-club nodes. Samsung's engineers upstaging IBM's engineers is too much for the entrenched NIH syndrome to cope with.

That's interesting information, is there any public source on that or is it pretty much insider knowledge?

No, it doesn't. If anyone plans to release something in Samsung's bleeding edge node (20nm) by H214/H115 they must have an agreement *now*, and so far we haven't heard anythig.

Would we necessarily hear anything?

The rumor mill won't always be as interested in some lesser SoC makers as they are in Apple.. I mean, we're talking anyone, right?

Anyway, I thought (and tell me if I was wrong on this) that you were arguing that no one uses GF because they're really late to market, and I'm saying in terms of products released Samsung isn't better.

Another factor is I don't think Samsung really offers the same density at the same node designations. Here's one example: look at the Cortex-A9s used in TI's OMAP4430 (TSMC 45nm). Die size is 70mm^2 (8.2mm x 8.5mm). Here's a die shot:

http://www.techinsights.com/uploade...Teardown_-_Blackberry_Playbook/OMAP4430-3.jpg

CPU area not counting L2 cache is about 4.9mm^2.

Now look at Apple's A5 on Samsung's 45nm. Die size is 122mm^2 (12.1mm x 10.1mm). Here's a die shot:

http://www.chipworks.com/media/wpmu...files/2011/03/APL0498_APL0498E01_Backside.jpg

CPU are not counting L2 cache is about 7.77mm^2.

That's a pretty huge difference. It's possible some of it can be accounted for in different design goals but I would think at least some of it has to do with the process.

You can also compare blocks from Brazos's VLIW5 (TSMC 40nm) and Llano's (GF 32nm), but I'll leave that one to you if you're interested.

What I'm trying to say is that there appear to be plenty of good technical reasons to choose TSMC over Samsung.

It might not have happened from a contractual POV, but from an engineering POV, yes, it did. Apple design teams aren't working Samsung guys to develop their future chips there, all the engineering guys are with TSMC design rules books optimizing Apple designs for it.

That's true, but the fact remains that the fabs are still occupied with Apple devices and we won't necessarily know that other fabs want Samsung until we see those devices. For as long as it's taking Apple to transition to TSMC it'll take even longer for others to transition to using their available capacity.

We don't even know if Samsung is advertising the space (or if they even want it instead of filling it with Samsung SoCs or just closing it down altogether)

 

[Ajay]

I wouldn't have expected the console chips to be fabbed at GF instead of TSMC but I also didn't expect XBox One's chip to be nearly as huge as it is so I don't really know. I'm sure AMD would be happy to fulfill some of their obligations to GF this way.

I would have expected AMD management to push for at least one console APU to be produced at GFL in exchange for a second source agreement (to fulfill WSA). To do otherwise would have been fiduciary incompetence.

 

[NTMBK]

Another factor is I don't think Samsung really offers the same density at the same node designations. Here's one example: look at the Cortex-A9s used in TI's OMAP4430 (TSMC 45nm). Die size is 70mm^2 (8.2mm x 8.5mm). Here's a die shot:

http://www.techinsights.com/uploade...Teardown_-_Blackberry_Playbook/OMAP4430-3.jpg

CPU area not counting L2 cache is about 4.9mm^2.

Now look at Apple's A5 on Samsung's 45nm. Die size is 122mm^2 (12.1mm x 10.1mm). Here's a die shot:

http://www.chipworks.com/media/wpmu...files/2011/03/APL0498_APL0498E01_Backside.jpg

CPU are not counting L2 cache is about 7.77mm^2.

That's a pretty huge difference. It's possible some of it can be accounted for in different design goals but I would think at least some of it has to do with the process.

Apple's Swift core is a much more powerful core than the A9- I'm not surprised it takes up more die space.

 

[Arachnotronic]

Apple's Swift core is a much more powerful core than the A9- I'm not surprised it takes up more die space.

A5 isn't based on Swift.

 

[Exophase]

A5 isn't based on Swift.

Just in case that doesn't fully clarify it, A5 uses the same dual-Cortex-A9 setup as OMAP4430, hence why I made the comparison.

Incidentally, this ratio in density is similar to what you get between the estimated die size for Cortex-A15 on Tegra 4 on TSMC 28nm and what Samsung publishes for their Cortex-A15 on Exynos 5410 on Samsung 28nm. So maybe there's more to that than the Tegra 4 die shots simply being totally bogus.

 

[bryanW1995]

You seem to place a lot of trust in corporations, which isn't necessarily a wise thing to do. A lot of people seem to hedge their bets on who the "good guys" and "bad guys" are among corporations but they're all pretty much the same. Businesses and corporations definitely have an ugly / cut-throat side across the board when it comes down to it.

I'm certain that Samsung used Apple's business to their advantage (as they compete in many of the same areas), and i'm also sure there is not a shred of concrete evidence anywhere. Where there is proof, you can also be sure there is plausible deniability.

Some guys are more "good" than others, though, and that is what IDC is talking about. Samsung just has a bad reputation in the industry. Quite likely they earned it, but it is at least theoretically possible that they were themselves the victims of corporate espionage, people jumping to conclusions regarding their products, they have a couple of rogue employees looking for advancement, etc etc etc.

Regardless, it seems dumb to allow a major competitor to have access to your IP. In the immortal words of jhh: “Globalfoundries is an AMD fab, right? Globalfoundries is AMD's fab. Our strategy is TSMC,”

 

[bryanW1995]

And yet I can't think of a single way in which Samsung Mobile moved products which put them closer to Apple earlier than their other competitors.

Personally, I don't see a difference in magnitude between Samsung Semiconductor leaking logic details to the SoC makers (Samsung LSI) or the mobile device makers (Samsung Mobile), nor the former being more egregious than the latter. Both are not automatically a given and some evidence actually would apply. I also think Samsung's divisions are more walled off than some people realize, word has been going around that SamMobile came really close to completely dumping Exynos 5 Octa from any Galaxy S4s (and even if that's not true, they've so far made it into far fewer than expected)

This particular concern also doesn't make sense in the context in which people have been arguing - that manufacturing with Samsung is a risk for other SoC makers. Being as none of those other SoC makers are really making phones and tablets.

You are putting forth a lot of time/effort to dispute somebody's opinion. Normally, I would be inclined to favor your "fact-based" line of reasoning, but you're just trying too hard. And anybody with common sense recognizes the absurdity of giving your competitor your strategic plans. Plus, IDC has a long-established reputation for objectivity. In the rare instances such as now that he doesn't provide lots of evidence, I'm inclined to accept his judgement.

ATIC and Mubadala dont agree with you.

http://www.xbitlabs.com/news/other/...es_Unlikely_to_Become_Profitable_Shortly.html

And their investment is dropping. Just like the Abu Dhabi facility was cancelled.

Your data is from April 2012, while SiliconWar's links were from Nov 2012 and June 2013. I think that his data is a bit more relevant.

 

[NTMBK]

A5 isn't based on Swift.

Whoops Guess who screwed up their codenames!

 

[Idontcare]

You are putting forth a lot of time/effort to dispute somebody's opinion. Normally, I would be inclined to favor your "fact-based" line of reasoning, but you're just trying too hard. And anybody with common sense recognizes the absurdity of giving your competitor your strategic plans. Plus, IDC has a long-established reputation for objectivity. In the rare instances such as now that he doesn't provide lots of evidence, I'm inclined to accept his judgement.

I wouldn't make the statements if they weren't based on objective information which I have at my disposal...but not all objective information is available in the public domain (one is a subset of the other).

That is where "reading between the lines" is required, and not just with my posts. There is quite a large group of active members here in these forums who for various reasons must restrict their contributions to posts which are intentionally crafted to enable the reader to read between the lines if the reader is willing to do so...

Comedy generally ensues when a member doesn't realize this and they begin to debate the "insider" member over the technical details of their post, "lol sighs" are dropped with reckless abandon

That said, reading between the lines requires a leap of faith (and imagination as one connects the dots in their own time), and I would not ask anyone to just take it on faith that I know what I'm talking about.

But for those who are ok with reading between the lines, and feel the dots I try to leave in my posts for connecting by the reader are acceptable for connecting on their own time, then I suspect those individuals have a rather darn good insight into how things are happening in the industry in ways that will simply never become impressed on the mainstream tech journalist.

I could refuse to accept that the moon landings ever happened until I am provided incontrovertible evidence that it had...but if Buzz Aldrin was in my living room telling me about his experience being the second person to walk on the moon I don't think I'd yell "shens!" to his face and demand he prove to me he actually walked on the moon :whiste: I'd probably be inclined to take him at his word at that point

That's interesting information, is there any public source on that or is it pretty much insider knowledge?

If you wait for this kind of info to become public sourced then you'll be waiting for at least 2 or 3 more years. For now it is just stuff I happen to know for fact is true, but the facts themselves are not in a form that lend itself to easy communication or verification.

That said, its not like it is a secret either. I don't talk about stuff in these forums that isn't already pretty much a widely known fact with the relevant industry circles for which the information applies.

Take for example when I said around a year ago (IIRC) that Apple had already signed contracts with TSMC and was moving all future production there. That wasn't secret information, literally thousands of people knew it.

Anyone who had been onsite at TSMC's Fab15 saw it with their own eyes, and if you asked just about anyone onsite they would tell you why the pace of construction was not only record-setting for TSMC but was more than double (2x!) that of any prior capacity expansions.

Now it was easy enough for me to communicate what I saw, read, and knew from other contractual contacts (the truth is always the easiest thing to write about)...but not so easy to prove it with supporting third-party links at the time. So you either had to accept it on faith, read between the lines, pm me, or otherwise find any individual who had eyes and ears in Taiwan and you'd know all there was to know a full year in advance of it becoming "news" in all the tech sites.

Samsung is a snoozer of a topic though, last year's discussion material, the real juicy stuff right now is about what is happening with the fab club, GloFo, and that Malta fab :whiste: (yes that was a silly ridiculous teaser, but also completely true )

 

[Silverforce11]

TSMC as well. The loser is AMD/nVidia. 20nm GPUs are not coming anytime soon I guess. Apple/Qualcomm bidding war.

I think we long knew this would happen. The mobile wars are in full swing, they simply need to be on the cutting edge and they all know this. With the margins these guys operate with as well as the volume, they can outbid the "niche" PC GPU market easily.

20nm GPUs are going to be late and very very expensive. Sad panda.