TSMC signs chip deal with Apple

[Arachnotronic]

Well, that further tidbit does raise another point of interest - it's likely that the 20nm A8 that the article is talking about won't be in actual products 'til basically Q4 of 2014. (Refresh later this year should be using A7 if they continue their current numbering scheme.) So either TSMC isn't going to have 20nm products shipping until after Intel likely releases 14nm Broadwell or Apple is still giving themselves a bit of buffer between planned process availability for mass production and when they actually need it to work.

I can almost guarantee you that I'll be able to buy a Broadwell before a 20nm TSMC product.

 

[Subyman]

Why doesn't TSMC scale up? They obviously have the business.

 

[Exophase]

I can almost guarantee you that I'll be able to buy a Broadwell before a 20nm TSMC product.

So you're confident that Broadwell will be out early Q2 or Q1, or what?

Ivy Bridge came out about 15 months after Sandy Bridge. Looking back, Penryn took about 15 months and Westmere took about 14 months, so this is nothing new for tick duration.

14-15 months places a Broadwell release in August or September of 2014, firmly in Q3. I see no reason to believe Broadwell will be a faster tick. I see it taking longer than usual as being more likely.

Why doesn't TSMC scale up? They obviously have the business.

Building new fabs is very expensive and takes a long time, so it can take foundries a while to react to big changes in volume demand.

 

[blackened23]

I think it's more related to the fact that 22nm was in development at intel for the greater part of 8 years. I really don't expect TSMC and company to switch on a whim, as there is a lot of R+D and expenditures involved. It isn't like you can start a new process like this with a blink of an eye.

I'm pretty certain Broadwell will be on market prior to any TSMC 20nm product.

 

[Hitman928]

Why doesn't TSMC scale up? They obviously have the business.

Building new fabs is very expensive and takes a long time, so it can take foundries a while to react to big changes in volume demand.

Yep, even a company like TSMC has to be able to justify a new fab. At $5 billion + (they are getting more and more expensive), you better be sure you can look a few years into the future and say, yes, when this fab comes online we can recoup the expenses within a reasonable time frame. This is not just an easy, hey, we have more demand, let's build another fab!

So you're confident that Broadwell will be out early Q2 or Q1, or what?

Ivy Bridge came out about 15 months after Sandy Bridge. Looking back, Penryn took about 15 months and Westmere took about 14 months, so this is nothing new for tick duration.

14-15 months places a Broadwell release in August or September of 2014, firmly in Q3. I see no reason to believe Broadwell will be a faster tick. I see it taking longer than usual as being more likely.

Agreed. Most people blasted that AMD guy recently when he said that Moore's law was at an end (as a constant predictor) because technology advances were slowing down. What those people didn't realize is that the people within most semi companies agree with him, they just don't say it. It's not a technological/science issue (yet), it's a return on investment issue.

 

[Exophase]

Well, that further tidbit does raise another point of interest - it's likely that the 20nm A8 that the article is talking about won't be in actual products 'til basically Q4 of 2014. (Refresh later this year should be using A7 if they continue their current numbering scheme.) So either TSMC isn't going to have 20nm products shipping until after Intel likely releases 14nm Broadwell or Apple is still giving themselves a bit of buffer between planned process availability for mass production and when they actually need it to work.

But the article itself says A8 will be released in an iPhone early 2014. That'd pull in the release well ahead of the normal time, but that's not that surprising given that the last iPad release was pulled in well ahead too.

If they release the next iPad before the next iPhone the naming's already going to be broken anyway, unless they call the iPad SoC A6XX..

 

[Hitman928]

If they release the next iPad before the next iPhone the naming's already going to be broken anyway, unless they call the iPad SoC A6XX..

I gave up on Apple's naming conventions after the "new ipad" and its successor, the new "ipad with retina display even though the last one had the same exact display"

 

[Exophase]

I think it's more related to the fact that 22nm was in development at intel for the greater part of 8 years. I really don't expect TSMC and company to switch on a whim, as there is a lot of R+D and expenditures involved. It isn't like you can start a new process like this with a blink of an eye.

You think what's related to that, IB taking 15 months after SB? But like I showed, the previous two ticks took about as long, that's clearly not a 22nm problem.

I'm pretty certain Broadwell will be on market prior to any TSMC 20nm product.

Maybe we should start taking bets

 

[blackened23]

You think what's related to that, IB taking 15 months after SB? But like I showed, the previous two ticks took about as long, that's clearly not a 22nm problem.

It was in response to intel17. If you think TSMC can switch to a new process on a whim, you are mistaken, which is why i'm nearly certain that 20nm TSMC will not happen this year or next. Intel had 22nm in development for an extremely long and costly period of time, longer than the 15 months between SB and IB - it was in development for nearly 8 years. It is not a trivial matter, the R+D process takes much longer than many realize. To see anyone suggesting otherwise, all I have to say is.....good luck with that to TSMC.

 

[Hitman928]

It was in response to intel17. If you think TSMC can switch to a new process on a whim, you are mistaken, which is why i'm nearly certain that 20nm TSMC will not happen this year or next. Intel had 22nm in development for an extremely long and costly period of time, longer than the 15 months between SB and IB - it was in development for nearly 8 years. It is not a trivial matter, the R+D process takes much longer than many realize. To see anyone suggesting otherwise, all I have to say is.....good luck with that to TSMC.

Just because I've seen this argument several times now in different threads, do you know how long TSMC has been working on 20nm? I'm not trying to single you out on this or anything because I don't know the answer. It's just that every time I see people talk about how long intel worked in finFET and 22nm and how far ahead they are, etc., they make it seem like TSMC didn't start working on 20nm until 28nm was finished and shipping. Perhaps 20nm has been in R&D for 8 years already? I'm not saying intel isn't ahead, they certainly are, but I don't think we can say for sure how far until we get actual working silicon from TSMC.

 

[Exophase]

It was in response to intel17. If you think TSMC can switch to a new process on a whim, you are mistaken, which is why i'm nearly certain that 20nm TSMC will not happen this year or next. Intel had 22nm in development for an extremely long and costly period of time, longer than the 15 months between SB and IB - it was in development for nearly 8 years. It is not a trivial matter, the R+D process takes much longer than many realize. To see anyone suggesting otherwise, all I have to say is.....good luck with that to TSMC.

No I don't think TSMC can switch to a new process on a whim. I don't know why you're putting it that way.

What I do think TSMC won't necessarily slip substantially on current projections (including financial ones) for release pending for a little over half a year away. And that the current 20nm projections are in line with the previous node schedule. And that the projection for 28nm known ~6 months before release didn't end up grossly off.

You keep bringing up Intel's 22nm when talking about TSMC's 20nm. Maybe you don't realize that TSMC's 20nm doesn't use FinFETs. It doesn't matter how long Intel was working on FinFETs, Intel clearly has stuck fairly closely to a release schedule for their tock to tick transitions. Why is it so impossible that TSMC can do the same?

 

[Wreckem]

Broadwell will ship in Q1/Q2 ONLY IF Broadwell = Haswell refresh. Thats TBD and many seem to think its very unlikely, while TSMC is very likely to ship their 20nm in volume in 1H 2014.

 

[galego]

From 3.5B to 4.5B, yes thats fantastic until you look on the competition. TSMC grew from 14.6B to 17B.

Therefore for you a 29% increase is fantastic until you see a 16% elsewhere?

 

[blackened23]

Broadwell will ship in Q1/Q2 ONLY IF Broadwell = Haswell refresh. Thats TBD and many seem to think its very unlikely.

Broadwell will initially be for mobile only and released next year. Likely not released for the desktop. At least, that's what everything points to.

There is no reason to bring broadwell to desktop outside of an E type processor because all of the improvements with it are related to efficiency, not desktop IPC.

 

[Exophase]

Broadwell will ship in Q1/Q2 ONLY IF Broadwell = Haswell refresh. Thats TBD and many seem to think its very unlikely.

Indeed we know there's something Intel has referred to (in many slides now) as "Haswell refresh" for desktops.. now why would that also be Broadwell? It just means Broadwell is probably not coming out for desktops at all, or at least not in socketed form. Not at first anyway.

These codenames go back several years but they're really only for internal use and early promotion anyway, the actual products don't bear the name.. so renaming Haswell CPUs to Broadwell in any context at all would be quite bizarre.

I've seen some people who have taken this to mean Broadwell isn't happening in 2014 at all (almost definitely not the case either), but not that they renamed a Haswell update to Broadwell.

 

[Arachnotronic]

Indeed we know there's something Intel has referred to (in many slides now) as "Haswell refresh" for desktops.. now why would that also be Broadwell? It just means Broadwell is probably not coming out for desktops at all, or at least not in socketed form. Not at first anyway.

These codenames go back several years but they're really only for internal use and early promotion anyway, the actual products don't bear the name.. so renaming Haswell CPUs to Broadwell in any context at all would be quite bizarre.

I've seen some people who have taken this to mean Broadwell isn't happening in 2014 at all (almost definitely not the case either), but not that they renamed a Haswell update to Broadwell.

 

[blackened23]

Well that answers the Broadwell question.

 

[AtenRa]

Where is that slide from??

 

[Exophase]

One of Ashraf Eassa's Seeking Alpha articles More seriously, he got it from a leak from CeBIT although he doesn't come right out and say it. You can see it here. ftp://62.153.231.170/Praesentationen/CeBIT_Kickoff_2013/TERRA_CeBIT_KickOff_2013_Intel.pptx

I wouldn't be totally surprised if Intel fudged it a little bit and pushed a July or August release onto 1H 2014 for that slide, although I guess it wouldn't be a huge stretch for it to really launch in 1H. They do say that the other mobile versions are coming 2H, which means Broadwell could be releasing early but in very limited (and expensive) quantity since it's only targeting premium ultrabooks. This would be enough of a break in precedent to facilitate a quicker turn-around than usual.

But before guaranteeing that this will be before TSMC's first 20nm products hit because they project small revenue in Q2 2014 remember that one of the first products will be FPGAs, which are very low volume.

 

[AtenRa]

@Exophase

Thx a lot for the link

 

[Idontcare]

Agreed. Most people blasted that AMD guy recently when he said that Moore's law was at an end (as a constant predictor) because technology advances were slowing down. What those people didn't realize is that the people within most semi companies agree with him, they just don't say it. It's not a technological/science issue (yet), it's a return on investment issue.

Its not an "issue", it is the sole reason why the pace was ever the pace it was. At every point along the curve, that point existed on the curve because it was by definition the point necessary for maximal ROI.

Moore's law was always about economics, never about performance.

 

[Hitman928]

Its not an "issue", it is the sole reason why the pace was ever the pace it was. At every point along the curve, that point existed on the curve because it was by definition the point necessary for maximal ROI.
*snip images*
Moore's law was always about economics, never about performance.

Good point. Moore's law is horribly misunderstood today. I blame the severe lack of technical understanding in the media (in general) as well as the severe lack of business understanding amongst most "techies"

 

[carop]

One of Ashraf Eassa's Seeking Alpha articles More seriously, he got it from a leak from CeBIT although he doesn't come right out and say it. You can see it here. ftp://62.153.231.170/Praesentationen/CeBIT_Kickoff_2013/TERRA_CeBIT_KickOff_2013_Intel.pptx

Thanks for the link.

I do not love roadmaps, but admit that the confidence a roadmap instills can be priceless.

Curiously, the following spy shot contradicts the contents of this document.

At any rate, Altera on 14 nm node at Intel and Xilinx on 16 nm node at TSMC have inevitably turned into an Intel and TSMC competition. If there is any way to prove out a new process, it is with memory or programmable logic, mostly uniform structures that can be replicated easily.

 

[blackened23]

Thanks for the link.

I do not love roadmaps, but admit that the confidence a roadmap instills can be priceless.

Curiously, the following spy shot contradicts the contents of this document.

A desktop roadmap doesn't contradict a mobile roadmap. These are different products. It's no secret that desktop isn't intel's priority.

 

[Arachnotronic]

A desktop roadmap doesn't contradict a mobile roadmap. These are different products. It's no secret that desktop isn't intel's priority.

Sure they are...if you're an AIO using a mobile grade chip