Samsung and GLOBALFOUNDRIES Forge Strategic Collaboration to Deliver 14nm FinFET

[AtenRa]

Even if you want to ignore the volatility associated with the customer base you're still going to be left with the fact that the total foundry revenue across TSMC, Samsung, and GF doesn't match Intel.

Wafer volume of TSMC alone is higher than Intel, revenue is irrelevant since Intel sell SKUs when TSMC only wafers.
If you want to compere Revenues, then take all the revenue of NVIDIA, AMD, Qualcomm and the rest that buy wafers from TSMC, combine them and compare them to Intel.

Also, Intel is not immune to the higher cost. 14nm products are one year late because they had to subsidize the cost of 22nm in 3 years.

One more think, everyone forgets that IC volume across x86 and ARM is rising it is not static. That means more wafer volume that translates to eventually lower cost for everyone.

 

[Khato]

Wafer volume of TSMC alone is higher than Intel, revenue is irrelevant since Intel sell SKUs when TSMC only wafers.
If you want to compere Revenues, then take all the revenue of NVIDIA, AMD, Qualcomm and the rest that buy wafers from TSMC, combine them and compare them to Intel.

Here in the real world only the individual company's revenue is applicable when talking about their available resources to devote to R&D. Last I checked TSMC gets paid by their customers for wafers, they don't receive payments to subsidize their R&D, so how is the revenue of their customers relevant in the least? Unless you want to claim that TSMC can take a greater chunk of those revenues by upping their pricing to subsidize the greater costs of continuing node shrinks... But that's basically the point in question here - can TSMC actually do that and still obtain enough volume going forward? Sure they claim that they can, but you'd have to be crazy to believe that they'd say anything to the contrary publicly. They're still obtaining what, almost two thirds of their revenue from wafers on the 40nm and larger nodes? Yeah, having more wafer volume than Intel means so much when over two thirds of it is on legacy process nodes whereas Intel is almost all leading edge.

Also, Intel is not immune to the higher cost. 14nm products are one year late because they had to subsidize the cost of 22nm in 3 years.

You're free to your opinion, but that doesn't make it correct. Far from it in fact. Intel isn't making more money due to the 14nm delay, they're losing it. Now clearly they've done quite well at mitigating it, but that doesn't change the fact that Q2 would've been even better if 14nm hadn't had issues and Broadwell shipped on time. (They're losing money both to the longer term 14nm fab capital investments and the design costs of Broadwell on a shorter product life cycle.)

 

[witeken]

Also, Intel is not immune to the higher cost. 14nm products are one year late because they had to subsidize the cost of 22nm in 3 years.

At the recent conference call, Brian Krzanich literally said 14nm is 6 months too late, not 1 year. And it isn't because of it took 22nm to get ROI of 22nm, but because 14nm was a difficult node for them, apparently. If it really took 3 years to subsidize 22nm, 10nm would also have been delayed, but it hasn't.

 

[AtenRa]

Here in the real world only the individual company's revenue is applicable when talking about their available resources to devote to R&D. Last I checked TSMC gets paid by their customers for wafers, they don't receive payments to subsidize their R&D, so how is the revenue of their customers relevant in the least? Unless you want to claim that TSMC can take a greater chunk of those revenues by upping their pricing to subsidize the greater costs of continuing node shrinks... But that's basically the point in question here - can TSMC actually do that and still obtain enough volume going forward? Sure they claim that they can, but you'd have to be crazy to believe that they'd say anything to the contrary publicly. They're still obtaining what, almost two thirds of their revenue from wafers on the 40nm and larger nodes? Yeah, having more wafer volume than Intel means so much when over two thirds of it is on legacy process nodes whereas Intel is almost all leading edge.

Intels Revenue combine the Wafer sale + the SKU sale.
Intels cost combine the R&D for the process + the IC design cost

TSMC Revenue comes only from the Wafer sale
TSMC cost is from the R&D process only.

So, at the end you cannot compare Intels total Revenue against TSMC because they are not the same.

If you want to compare to Intels total Revenue then you get TSMC Wafer revenue + (AMD, NV, Qualcomm etc) sales.

Also, TSMC has announced that 20nm will be 20% of its total revenue buy the end of 2014.

You're free to your opinion, but that doesn't make it correct. Far from it in fact. Intel isn't making more money due to the 14nm delay, they're losing it. Now clearly they've done quite well at mitigating it, but that doesn't change the fact that Q2 would've been even better if 14nm hadn't had issues and Broadwell shipped on time. (They're losing money both to the longer term 14nm fab capital investments and the design costs of Broadwell on a shorter product life cycle.)

Did i say they are making more money ?? I have said they need 3 years to subsidize their 22nm instead of 2.

 

[AtenRa]

At the recent conference call, Brian Krzanich literally said 14nm is 6 months too late, not 1 year. And it isn't because of it took 22nm to get ROI of 22nm, but because 14nm was a difficult node for them, apparently. If it really took 3 years to subsidize 22nm, 10nm would also have been delayed, but it hasn't.

He only said there is no delay on 10nm because of the 14nm delays. That is soooo PR and vague but thats his job

 

[witeken]

He only said there is no delay on 10nm because of the 14nm delays. That is soooo PR and vague but thats his job

Wrong.

"We have done no changes or shift to our 10-nanometer schedule but we won’t really talk about 10-nanometer schedules until next year."

http://seekingalpha.com/article/231...-results-earnings-call-transcript?part=single

Cannonlake should be coming somewhere in 2016, which is nowhere near your 3 year number. Even if it does cost 3 years, that it still no problem since those fabs don't magically disappear after 2 years. But I think you randomly invented this 3 year number without any source or reason.

 

[raghu78]

Here in the real world only the individual company's revenue is applicable when talking about their available resources to devote to R&D. Last I checked TSMC gets paid by their customers for wafers, they don't receive payments to subsidize their R&D, so how is the revenue of their customers relevant in the least? Unless you want to claim that TSMC can take a greater chunk of those revenues by upping their pricing to subsidize the greater costs of continuing node shrinks... But that's basically the point in question here - can TSMC actually do that and still obtain enough volume going forward? Sure they claim that they can, but you'd have to be crazy to believe that they'd say anything to the contrary publicly. They're still obtaining what, almost two thirds of their revenue from wafers on the 40nm and larger nodes? Yeah, having more wafer volume than Intel means so much when over two thirds of it is on legacy process nodes whereas Intel is almost all leading edge.

You're free to your opinion, but that doesn't make it correct. Far from it in fact. Intel isn't making more money due to the 14nm delay, they're losing it. Now clearly they've done quite well at mitigating it, but that doesn't change the fact that Q2 would've been even better if 14nm hadn't had issues and Broadwell shipped on time. (They're losing money both to the longer term 14nm fab capital investments and the design costs of Broadwell on a shorter product life cycle.)

Intel is an IDM which has more cost overhead as it spends billions on CPU architecture, GPU architecture, baseband R&D on top of the process R&D and capital expenditure. TSMC is a pureplay foundry which only has to focus on process R&D and capital expenditure

http://intc.client.shareholder.com/intel-annual-report/2013/Intel_ARand10K_13.pdf

(page 34)

Intel 2013 revenue - USD 52.7 billion
Net Income - USD 9.6 billion

http://www.tsmc.com/download/ir/annualReports/2013/english/annual2013e.pdf

(page 64)

TSMC 2013 revenue - USD 19.8 billion
Net income - USD 6.2 billion (1 TWD = 0.033 USD)

TSMC is more profitable than Intel when you consider profit as a % of revenue.

btw TSMC spent USD 9.7 billion in capex in 2013 and is expected to spend the same amount in 2014. Intel spent USD 10.6 billion in 2013 and is expected to spend USD 11 billion in 2014.

http://www.icinsights.com/news/bull...ccount-For-52-Of-Semiconductor-CapEx-In-2014/

btw what does it matter what node the revenue comes from. What matters is -

1. Is TSMC growing revenue rapidly - Yes they are increasing at 20% Y-o-Y
2. Is TSMC investing enough in capex to keep transitioning to the leading 20nm/16FF/16FF+ - Definitely yes
3. Is TSMC profitable - More than Intel from a % of revenue point of view.

But definitely TSMC faces challenges from Samsung at 16/14 FINFET and going forward. For that matter Intel faces a different kind of threat - losing market share to ARM ecosystem in the server market. Intel owns literally 100% of the x86 server market (97+%) and any loss to ARM is going to hurt Intel as they come with hefty margins.

The next 5 years are going to be very interesting. We will see how well ARM does higher up the stack (servers,desktops and notebooks) and how well Intel does lower down the stack (smartphones and tablets).

 

[AtenRa]

Wrong.

"We have done no changes or shift to our 10-nanometer schedule but we won’t really talk about 10-nanometer schedules until next year."

http://seekingalpha.com/article/231...-results-earnings-call-transcript?part=single

"We have done no changes"

Yeap, that is vague and a PR answer.

But I think you randomly invented this 3 year number without any source or reason.

22nm Ivy 2012
22nm Haswell 2013
22nm Haswell Refreash 2014

That is 3 years, 14nm was supposed to be 2013 according to Intel roadmap.

 

[AtenRa]

Intel is an IDM which has more cost overhead as it spends billions on CPU architecture, GPU architecture, baseband R&D on top of the process R&D and capital expenditure. TSMC is a pureplay foundry which only has to focus on process R&D and capital expenditure

http://intc.client.shareholder.com/intel-annual-report/2013/Intel_ARand10K_13.pdf

(page 34)

Intel 2013 revenue - USD 52.7 billion
Net Income - USD 9.6 billion

http://www.tsmc.com/download/ir/annualReports/2013/english/annual2013e.pdf

(page 64)

TSMC 2013 revenue - USD 19.8 billion
Net income - USD 6.2 billion (1 TWD = 0.033 USD)

TSMC is more profitable than Intel when you consider profit as a % of revenue.

btw TSMC spent USD 9.7 billion in capex in 2013 and is expected to spend the same amount in 2014. Intel spent USD 10.6 billion in 2013 and is expected to spend USD 11 billion in 2014.

http://www.icinsights.com/news/bull...ccount-For-52-Of-Semiconductor-CapEx-In-2014/

btw what does it matter what node the revenue comes from. What matters is -

1. Is TSMC growing revenue rapidly - Yes they are increasing at 20% Y-o-Y
2. Is TSMC investing enough in capex to keep transitioning to the leading 20nm/16FF/16FF+ - Definitely yes
3. Is TSMC profitable - More than Intel from a % of revenue point of view.

But definitely TSMC faces challenges from Samsung at 16/14 FINFET and going forward. For that matter Intel faces a different kind of threat - losing market share to ARM ecosystem in the server market. Intel owns literally 100% of the x86 server market (97+%) and any loss to ARM is going to hurt Intel as they come with hefty margins.

The next 5 years are going to be very interesting. We will see how well ARM does higher up the stack (servers,desktops and notebooks) and how well Intel does lower down the stack (smartphones and tablets).

Nicely explained thx

 

[Fjodor2001]

Wrong.

"We have done no changes or shift to our 10-nanometer schedule but we won’t really talk about 10-nanometer schedules until next year."

http://seekingalpha.com/article/231...-results-earnings-call-transcript?part=single

Cannonlake should be coming somewhere in 2016, which is nowhere near your 3 year number. Even if it does cost 3 years, that it still no problem since those fabs don't magically disappear after 2 years. But I think you randomly invented this 3 year number without any source or reason.

If Cannonlake is released in 2016, that gives Intel a ~1 year sales window to get ROI from their 14 nm process tech R&D investments (since 14 nm is about 1 year late, and most Intel 14 nm chips will hit the market in 2015Q1-Q3). Do you really think it is realistic for Intel to get satisfactory ROI from such a short sales window?

 

[Khato]

Intels Revenue combine the Wafer sale + the SKU sale.
Intels cost combine the R&D for the process + the IC design cost

TSMC Revenue comes only from the Wafer sale
TSMC cost is from the R&D process only.

So, at the end you cannot compare Intels total Revenue against TSMC because they are not the same.

If you want to compare to Intels total Revenue then you get TSMC Wafer revenue + (AMD, NV, Qualcomm etc) sales.

Also, TSMC has announced that 20nm will be 20% of its total revenue buy the end of 2014.

Quite correct, when comparing revenue of Intel against TSMC you have to account for the fact that Intel's cost includes both manufacturing and design costs... but that's it. Once you do that it's a perfectly valid comparison in the context of the company's available cash flow. Because in terms of being able to continue the necessary capital expenditures for future process nodes that's what matters - Intel can divert the profit it derives from the design side to the manufacturing side if necessary, TSMC can't.

Taking the full year 2013 results as an example, TSMC's net revenue was $20.11 billion with a gross profit of $9.46 billion and $1.62 billion spent on R&D. Just for a fun comparison, just on Q4 of 2013 Intel had net revenue of $13.83 billion with a gross profit of $8.57 billion and $2.82 billion spent on R&D. Yes, Intel spends more on 'R&D' in a quarter than TSMC does for the entire year, and far more than a quarter of that R&D expenditure is going to the manufacturing side of the business. Going to the full year results for Intel it's a net revenue of $52.7 billion, gross profit of $31.5 billion, and $10.6 billion spent on R&D. There's no escaping the fact that Intel has over double the cash flow at its disposal compared to TSMC and the majority of that revenue is coming from the leading edge process node.

As for TSMC's claims of massive 20nm income this year - maybe so. Here's to hoping for their sakes that other customers will be able to use up all of that capacity if Apple is indeed dropping TSMC for Samsung on the '16/14nm' node next year. Of course that's assuming that either of the foundry FinFET processes are actually able to reach mass production on schedule.

Did i say they are making more money ?? I have said they need 3 years to subsidize their 22nm instead of 2.

Yes, you did, and you just did again. By definition if they 'need 3 years to subsidize their 22nm instead of 2' then moving on to 14nm after 2 years would result in lower profits due to making 14nm capital investments before 22nm was paid off. But that's obviously not the case as Intel has already invested quite a bit of capital for 14nm production - not as much as they'd planned to due to the production ramp issues, but more than enough for it to be quite clear that the delay was not planned and hence not due to needing more time to subsidize 22nm... You don't build a fab and buy equipment just to have it all sit idle intentionally.

 

[witeken]

"We have done no changes"

Yeap, that is vague and a PR answer.

I don't think so. It's a very clear statement.

22nm Ivy 2012
22nm Haswell 2013
22nm Haswell Refreash 2014

That is 3 years, 14nm was supposed to be 2013 according to Intel roadmap.

14nm Broadwell 2014

14nm was never supposed to launch in 2013. It was supposed to start being manufactured.

 

[witeken]

If Cannonlake is released in 2016, that gives Intel a ~1 year sales window to get ROI from their 14 nm process tech R&D investments (since 14 nm is about 1 year late, and most Intel 14 nm chips will hit the market in 2015Q1-Q3). Do you really think it is realistic for Intel to get satisfactory ROI from such a short sales window?

Yes. Intel spends 10B$ on capital expenditures and 10B$ on R&D every year, with a net income of 10B$, which allows a 2 year tick-tock cycle. They have to spend the R&D and capex anyway, so when something happens to one node which doesn't influence the health of another node (aka yield), then surely they could give 14nm a shorter life to not delay future nodes. Not delaying future nodes always makes sense when you can pay it (which Intel can) because it is good for revenue (people have a reason to upgrade and you have a competitive advantage to gain more market share or go into new markets) and margins.

But it isn't really 1 year. 14nm availability starts in Q4'14, while Cannonlake will be Q3'16, give or take 1 quarter. They could release desktop 10nm mid-2016, while 10nm for the data center follows much later. I'm not saying they will do this, but theoretically they could, and it seems they will, if BK isn't lying.

What I'm basically saying is that they have the money for a 2 year cycle and they had to spend the money anyway, so it isn't strictly necessary to delay 10nm.

 

[Fjodor2001]

=> TSMC/Samsung 14nm ~ Intel 22nm

By what metrics? Transistor density, price per transistor, power consumption, ... ?

And how does TSMC/Samsung 14 nm vs Intel 14 nm compare by those metrics?

 

[Khato]

Intel is an IDM which has more cost overhead as it spends billions on CPU architecture, GPU architecture, baseband R&D on top of the process R&D and capital expenditure. TSMC is a pureplay foundry which only has to focus on process R&D and capital expenditure

http://intc.client.shareholder.com/intel-annual-report/2013/Intel_ARand10K_13.pdf

(page 34)

Intel 2013 revenue - USD 52.7 billion
Net Income - USD 9.6 billion

http://www.tsmc.com/download/ir/annualReports/2013/english/annual2013e.pdf

(page 64)

TSMC 2013 revenue - USD 19.8 billion
Net income - USD 6.2 billion (1 TWD = 0.033 USD)

TSMC is more profitable than Intel when you consider profit as a % of revenue.

Yes, which is due to a number of factors. As an IDM Intel spends a considerable portion of its revenue on advertising compared to practically nil for TSMC. Also, Intel's R&D budget is over 4x that of TSMC, and while the majority of that does go to the design side that still leaves plenty for their process R&D to be well ahead of TSMC's capabilities. Even if you want to claim that Intel doesn't maintain much of a process advantage in terms of node size there's no way to deny that Intel leads the industry in terms of putting transistor innovation into production.

btw TSMC spent USD 9.7 billion in capex in 2013 and is expected to spend the same amount in 2014. Intel spent USD 10.6 billion in 2013 and is expected to spend USD 11 billion in 2014.

And there are many reasons for that. Regardless, CapEX is primarily a function of equipment for the leading edge process and fab space to increase capacity. TSMC has been on a spending spree in order to do both en mass in order to gain Apple as a 20nm customer. And again I do wonder what their CapEX for 2015 is going to look like if the hint at their recent conference call does mean that they've lost Apple on the next node.

btw what does it matter what node the revenue comes from. What matters is -

1. Is TSMC growing revenue rapidly - Yes they are increasing at 20% Y-o-Y
2. Is TSMC investing enough in capex to keep transitioning to the leading 20nm/16FF/16FF+ - Definitely yes
3. Is TSMC profitable - More than Intel from a % of revenue point of view.

But definitely TSMC faces challenges from Samsung at 16/14 FINFET and going forward. For that matter Intel faces a different kind of threat - losing market share to ARM ecosystem in the server market. Intel owns literally 100% of the x86 server market (97+%) and any loss to ARM is going to hurt Intel as they come with hefty margins.

1. Yup, for the last two years they've been increasing their revenue quite nicely - 17.8% increase in 2013 and an 18.5% increase in 2012. But go back to 2011 and it drops like a rock to all of a 1.8% increase. I expect that they'll tick up to that 20% figure or possibly even a bit beyond it this year, but next year?
2. Assuming that their CapEX maintains constant sure. But right now they're just investing enough for 20nm. (By the way, anyone else find it odd that they didn't mention '16 FinFET+' once in their recent conference call? Despite comparing '16 nanometer' against '10 nanometer'.)
3. Yup, and they'll continue to be. The question is whether or not they continue to be by becoming the next in the long line of foundries that throws in the towel. (By the way, that's the reason why the node source of TSMC's revenue matters, especially going forward.)

And there's no question that Intel faces challenges... But ARM in the server market isn't one of them Intel's doing quite well at guarding all angles of that cash cow. It helps that the software ecosystem is also doing its part in playing to Intel's strengths - that's not x86 by the way, but rather performance per core/socket. No the challenge for Intel is to actually gain ground in mobile rather than continuing to come in second place to Qualcomm as that's their primary growth opportunity right now. And it's a prospect that's worrying to all of the competition...

 

[ShintaiDK]

By what metrics? Transistor density, price per transistor, power consumption, ... ?

And how does TSMC/Samsung 14 nm vs Intel 14 nm compare by those metrics?

Price per transistor depends on IC design. So unless you are willing to lay down a little billion $ there. You get higher transistor price vs 28nm.

Both TSMC and Samsungs 16 and 14nm uses 20nm design tools. TSMC claims a 15% area reduction compared to 20nm. Samsung 14% compared to 20nm. That alone should make your alarm bells go off that marketing have been creative.

I assume its because TSMC and Samsung is quickly running out of customers actually able to pay the IC design cost for lower transistor cost. Its hard to see more large scale customers besides Apple and Qualcomm for that node. So TSMC and Samsung is busy in a PR war with one another. GloFo did the exact same before they gave up. Their 14XM was the exact same PR nonsense by fiddling with the numbers.

Not to mention how far behind both of them are in implementing transistor technology.

It seems at 10nm that Intel may be 4-5 years ahead. Assuming that 10nm ever materialize for the foundries.

 

[Khato]

By what metrics? Transistor density, price per transistor, power consumption, ... ?

And how does TSMC/Samsung 14 nm vs Intel 14 nm compare by those metrics?

Really, we only have a partial picture to compare even against Intel's 22nm. Sure we've been given SRAM size and some drive currents, but we don't know other process characteristics which are typically trade-offs with those. In terms of both SRAM size and drive currents TSMC's "16 FinFET" has a slight advantage over Intel's 22nm - again though, that really doesn't give us the full picture of how they compare.

On both Samsung "14nm" and Intel's 14nm though I'm not aware of any information having been released? Other than the chart from Intel's investor meeting last year which implied a 14nm SRAM cell size below 0.05 um^2. It certainly is odd for Intel to have not released any real information regarding 14nm characteristics yet. Whereas I don't recall Samsung ever saying much? (Equally plausible that I've just never paid much attention to them though, haha.)

 

[witeken]

By what metrics? Transistor density, price per transistor, power consumption, ... ?

And how does TSMC/Samsung 14 nm vs Intel 14 nm compare by those metrics?

I'm talking about the average.

Intel's 14nm compared to TSMC's 16nm has substantially higher performance, density and price per transistor, lower power and faster TTM.

@Khato: According to Intel17, Intel will give information about 14nm at IDF.

 

[Fjodor2001]

I'm talking about the average.

I'd like you to be a bit more precise than that.

According to Khato, not much details are known about neither Intel 14 nm, nor Samsung/GF 14 nm. So it's a quite bold statement for you to say that Samsung/GF 14 nm should be equal to Intel 22 nm. In that case I expect you to be able to back that statement up with some hard facts, like the metrics I mentioned. So do you have such information?

Intel's 14nm compared to TSMC's 16nm has substantially higher performance, density and price per transistor, lower power and faster TTM.

I said Samsung/GF 14 nm, not TSMC 16 nm. I.e. I was requesting a comparison of Samsung/GF 14 nm vs Intel 14 nm (and Samsung/GF 14 nm vs Intel 22 nm since you mentioned that too).

 

[ShintaiDK]

I said Samsung/GF 14 nm, not TSMC 16 nm. I.e. I was requesting a comparison of Samsung/GF 14 nm vs Intel 14 nm (and Samsung/GF 14 nm vs Intel 22 nm since you mentioned that too).

Samsungs 14nm looks to be alittle worse than TSMC 16nm.

 

[raghu78]

Yes, which is due to a number of factors. As an IDM Intel spends a considerable portion of its revenue on advertising compared to practically nil for TSMC.

these are all disadvantages for Intel as an IDM and advantages for TSMC as a pureplay foundry. The low cost structure of a pureplay foundry is well understood.

Also, Intel's R&D budget is over 4x that of TSMC, and while the majority of that does go to the design side that still leaves plenty for their process R&D to be well ahead of TSMC's capabilities.

Do you know what Intel's process R&D portion is of their overall R&D ? I agree Intel has tremendous resources but TSMC is no pushover.

Even if you want to claim that Intel doesn't maintain much of a process advantage in terms of node size there's no way to deny that Intel leads the industry in terms of putting transistor innovation into production.

Agree completely. Intel will be the world's No.1 semiconductor company well into the future.

And there are many reasons for that. Regardless, CapEX is primarily a function of equipment for the leading edge process and fab space to increase capacity. TSMC has been on a spending spree in order to do both en mass in order to gain Apple as a 20nm customer. And again I do wonder what their CapEX for 2015 is going to look like if the hint at their recent conference call does mean that they've lost Apple on the next node.

Losing Apple will affect the perception of others who view TSMC as the world's No.1 foundry. But still TSMC will dominate 20nm which will be the major process node for 2015 and 2016. Today TSMC 28nm is 34% of TSMC production and in its 3rd year of production. It took more than a year for 28nm to catch up with 40nm.

http://www.xbitlabs.com/news/other/..._Fourth_Quarter_Totaled_Nearly_1_Billion.html

TSMC has literally no competition at 20nm. Also TSMC's yield learning at 20nm helps with 16FF and 16FF+ as they share the same BEOL.

Samsung may have got the deal for A9 but that does not mean TSMC cannot get back the deal for A10. TSMC remains by far the best foundry to ramp a leading edge process at high yields. Only Intel has better yields in the industry at any stage of the leading edge process node ramp.

1. Yup, for the last two years they've been increasing their revenue quite nicely - 17.8% increase in 2013 and an 18.5% increase in 2012. But go back to 2011 and it drops like a rock to all of a 1.8% increase. I expect that they'll tick up to that 20% figure or possibly even a bit beyond it this year, but next year?
2. Assuming that their CapEX maintains constant sure. But right now they're just investing enough for 20nm. (By the way, anyone else find it odd that they didn't mention '16 FinFET+' once in their recent conference call? Despite comparing '16 nanometer' against '10 nanometer'.)
3. Yup, and they'll continue to be. The question is whether or not they continue to be by becoming the next in the long line of foundries that throws in the towel. (By the way, that's the reason why the node source of TSMC's revenue matters, especially going forward.)

TSMC is the world's largest and most profitable foundry. They have the resources and the knowhow to lead the foundry business. A slight lead of 6 months to Samsung in getting to FINFET first won't matter in the long run. TSMC will invest even higher amounts at 16FF/16FF+ and in the long run will be a dominant player.

And there's no question that Intel faces challenges... But ARM in the server market isn't one of them Intel's doing quite well at guarding all angles of that cash cow. It helps that the software ecosystem is also doing its part in playing to Intel's strengths - that's not x86 by the way, but rather performance per core/socket. No the challenge for Intel is to actually gain ground in mobile rather than continuing to come in second place to Qualcomm as that's their primary growth opportunity right now. And it's a prospect that's worrying to all of the competition...

ARM getting into the server market is the biggest threat Intel faces. Given the inherent efficiency advantage of the ARMV8 ISA over x86-64 the fact is you can design a high performance ARMv8 core which has IPC like Intel's big core Haswell and at similar or better power efficiency (btw with a slightly inferior foundry process like TSMC 16FF/16FF+ against Intel 14nm). Apple already has shown the direction with Cyclone as to what a big ARMv8 core looks like. AMD with K12 and others ARM licensees are going to go straight up against Intel's big cores. The Linux and Open Source server software vendors are embracing ARMv8 in a big way and that is the key to disruption.Microsoft too will embrace ARMv8 sometime in the future when it sees the writing on the wall.

Intel in mobile is a competitor who will keep Apple, Qualcomm and others honest and driving the performance aggressively. I never take extreme views. Intel can enter mobile and ARM can enter server meaningfully in the long term. Long term I see ARM as a bigger participant in the computing spectrum, not just mobile and embedded as it is today.

 

[raghu78]

Samsungs 14nm looks to be alittle worse than TSMC 16nm.

do you have any technical information presented by Samsung on their 14nm (14LPE/14LPP) finfet process at events like IEDM and can compare it to TSMC 16FF and 16FF+.

 

[witeken]

these are all disadvantages for Intel as an IDM and advantages for TSMC as a pureplay foundry. The low cost structure of a pureplay foundry is well understood.

What? IDM is better than a foundry. Ask Nvidia.

Do you know what Intel's process R&D portion is of their overall R&D ? I agree Intel has tremendous resources but TSMC is no pushover.

Yes.

Which is good enough to gain a 2-4 year process lead.

ARM getting into the server market is the biggest threat Intel faces. Given the inherent efficiency advantage of the ARMV8 ISA over x86-64 the fact is you can design a high performance ARMv8 core which has IPC like Intel's big core Haswell and at similar or better power efficiency (btw with a slightly inferior foundry process like TSMC 16FF/16FF+ against Intel 14nm).

There is no inherent efficiency advantage with ARM. If you want to design a good CPU with similar IPC as Intel, you'll have to spend billions per year, and then you have to put in on an inferior process.

 

[raghu78]

I'm talking about the average.

Intel's 14nm compared to TSMC's 16nm has substantially higher performance, density and price per transistor, lower power and faster TTM.

@Khato: According to Intel17, Intel will give information about 14nm at IDF.

Intel can claim a lot of things and so can TSMC. TSMC claims 16FF+ has a 15% perf increase and 15% density increase over TSMC 16FF and a 38% perf increase over 20 SOC and a close to 60% perf increase over 28HPM.
(In terms of perf at same power 28HPM = 1x , 20 SOC = 1.15x, 16FF - 1.38x , 16FF+ - 1.6x )

As for density TSMC claims Intel 14nm has a marginal lead of 10 - 15% over 16FF+. But the key to all of this discussion is how do actual products fare. So that would mean Intel 14nm Broxton against Qualcomm and other ARMv8 licensees products at TSMC 16FF+. Intel does have a 12 - 18 month Time to market lead at 14 FINFET compared to the others when you consider all products. For Intel's first baseband integrated 14nm SOCs the lead is not much. 14nm Broxton will go up against 16/14 FINFET Qualcomm ARMv8 SOCs and Apple A9.

 

[witeken]

I'd like you to be a bit more precise than that.

According to Khato, not much details are known about neither Intel 14 nm, nor Samsung/GF 14 nm. So it's a quite bold statement for you to say that Samsung/GF 14 nm should be equal to Intel 22 nm. In that case I expect you to be able to back that statement up with some hard facts, like the metrics I mentioned. So do you have such information?

Density: look at Intel's slide.
Power consumption: Intel cut it in half with 22nm, TSMC claims something similar with their FinFETs.