Is Intel's "Process Lead" Somewhat a Sham?

[SlimFan]

It's not "transistor scaling". It says "mm2/transistor" scaling.

If you include the less scaling area's (like the I/O) in the "area-scaling", then
the "mm2/transistor" scaling becomes well over 3 and not less.

Don't forget metal. An IP might very well scale at better than mm2/transistor depending on the metal / library / perf characteristics, etc.

For example TSMC is claiming that they get area benefit at "16nm" even though the fundamental geometries of the process are the same.

 

[Fjodor2001]

Much lower transistor cost, 67% lower power, 40% higher peformance (take those 2 claims with a pinch of salt until we see independent benchmarks, though), continuation of Dennard scaling. Also see this informative article: Intel’s 14nm milkshake: It’s better than yours.

14nm seems like an excellent node, vastly superior compared to 20nm, 20nm FinFET and probably competitive/similar to the competition's 10nm. So I wouldn't worry about Intel process lead. It certainly isn't shrinking, that's for sure.

Edit: And I'd better listen to Intel's claims. As we've seen with TSMC's rebuttal, it doesn't really contain any facts.

So what frequency increase can we derive from this?

 

[Idontcare]

So what frequency increase can we derive from this?

Depends on 14nm TJmax at this point. The density is scaling faster than the reduction in power consumption, so thermal density is increasing which means so too are the operating temperatures.

Recall that 32nm->22nm saw considerably theoretical improvement in power consumption normalized clockspeeds, but in practice both 32nm and 22nm saw themselves thermally limited in clockspeed at their respective TJmax envelopes when employing comparable cooling, etc.

At this point I expect 14nm to offer no real improvement in Fmax at TJmax, instead it will offer considerable improvement in power at any given point on the shmoo plot (and smaller die size which will help with the production cost economics).

If you are operating at the cusp of a thermally limited threshold, which Intel has been for both 32nm and 22nm, then making your chips ever smaller will do you no good in terms of clockspeed and so forth unless you also simultaneously scale the power consumption at an equivalent or better rate.

Every indication for Intel 14nm is that this is not happening. That is bad news if you want high(er) clocks with fat(ter) cores.

 

[blackened23]

Depends on 14nm TJmax at this point. The density is scaling faster than the reduction in power consumption, so thermal density is increasing which means so too are the operating temperatures.

Recall that 32nm->22nm saw considerably theoretical improvement in power consumption normalized clockspeeds, but in practice both 32nm and 22nm saw themselves thermally limited in clockspeed at their respective TJmax envelopes when employing comparable cooling, etc.

At this point I expect 14nm to offer no real improvement in Fmax at TJmax, instead it will offer considerable improvement in power at any given point on the shmoo plot (and smaller die size which will help with the production cost economics).

If you are operating at the cusp of a thermally limited threshold, which Intel has been for both 32nm and 22nm, then making your chips ever smaller will do you no good in terms of clockspeed and so forth unless you also simultaneously scale the power consumption at an equivalent or better rate.

Every indication for Intel 14nm is that this is not happening. That is bad news if you want high(er) clocks with fat(ter) cores.

Great info IDC. I firmly believe intel has delivered closer to their promises far more than TSMC, but that's aside from the point. I am curious as to your last sentence. Do you think that this is an intentional trade-off? Necessary trade-off? If you cant have your cake and eat it so to speak, i'm almost inclined to say that this is a necessary evil in the pursuit of performance per watt, and intel will be focusing more on many core CPUs for the high end desktop since all out clockspeed with insane voltages don't seem to jive with that design. And since mobile has exceeded desktop sales for a long time now, it seems that performance per watt would be the focus to have extreme low voltage SKUs for ultrabooks, tablets, so on and so forth.

Maybe this is why HEDT platforms will have 8 cores as a base. Intel may possibly be slowly increasing core counts for desktop platforms, since the design architecture must be designed for PPW to be viable in the overall picture. Now this COULD be a good thing. Instead of software and games relying on 1-2 threads/cores, perhaps this will spur developers to take increased advantage of many core processors?

Would you agree with that? I dunno. I can see how some desktop purists (I love mobile, but I'm also a desktop guy at heart) would prefer insane voltages with high clockspeeds, but it doesn't seem to be compatible with the new metric that everyone is pursuing - performance per watt. Would you agree/disagree? Or am I way off here?

 

[Abwx]

What's interesting is that, on top of that fact that the article claims that Intel's slide is accurate, another indication of the correctness is that Intel's slide claims a 1.7x area scaling of 32->22nm (1/0.585). This is spot on if we look at the data that is available: Sandy Bridge has 1.16B layout transistors and an area of 216mm², while Haswell has 1.6B layout transistor in 177mm², resulting in a scaling of 1.68.

This would be spot on if these were straight die shrinks but it s not, the smaller the node the bigger the part, the GPU, that allow high density so real scaling is likely significantly less than 1.68.

 

[Idontcare]

Great info IDC. I firmly believe intel has delivered closer to their promises far more than TSMC, but that's aside from the point. I am curious as to your last sentence. Do you think that this is an intentional trade-off? Necessary trade-off? If you cant have your cake and eat it so to speak, i'm almost inclined to say that this is a necessary evil in the pursuit of performance per watt, and intel will be focusing more on many core CPUs for the high end desktop since all out clockspeed with insane voltages don't seem to jive with that design. And since mobile has exceeded desktop sales for a long time now, it seems that performance per watt would be the focus to have extreme low voltage SKUs for ultrabooks, tablets, so on and so forth.

Maybe this is why HEDT platforms will have 8 cores as a base. Intel may possibly be slowly increasing core counts for desktop platforms, since the design architecture must be designed for PPW to be viable in the overall picture. Now this COULD be a good thing. Instead of software and games relying on 1-2 threads/cores, perhaps this will spur developers to take increased advantage of many core processors?

Would you agree with that? I dunno. I can see how some desktop purists (I love mobile, but I'm also a desktop guy at heart) would prefer insane voltages with high clockspeeds, but it doesn't seem to be compatible with the new metric that everyone is pursuing - performance per watt. Would you agree/disagree? Or am I way off here?

It is a necessary evil in the pursuit of lowering the cost on a transistor/area normalized basis.

Adding more cores, albeit at comparable Fmax, is an R&D savings decision at the tradeoff of higher per-unit production cost (because the die is larger than it might otherwise have been on a performance normalized basis - see Piledriver versus Sandy Bridge).

Intel's focus on continued area scaling is arguably the single-most "game changer" coming down the pipe in terms of potentially shaking up the industry.

It is the last thing fabless giants like Qualcomm, Broadcom, Apple want to see happen. It leaves them incredibly vulnerable to Intel rolling out a market-share killer.

And they know how quickly that can happen, it wasn't all that long ago (2006) that TI held some 80% of the mobile phone market (in terms of underlying silicon, not brands), and it took a mere 7 years for that number to go to 0%.

But that is why I stated earlier that the wrong question is being posed. It isn't a question of the lead, it is a question of the ability of Intel to execute and leverage it into something tangible in terms of products, market share, and financials in any area outside their core x86 business units.

Intel has failed to leverage their process advantage in any appreciable manner (ex-core business) for many nodes now (45nm, 32nm, and 22nm). If they don't make considerable mobile inroads with 14nm while competing with hamstrung area-scaled 16/14nm nodes then they never will.

 

[witeken]

Depends on 14nm TJmax at this point. The density is scaling faster than the reduction in power consumption, so thermal density is increasing which means so too are the operating temperatures.

Recall that 32nm->22nm saw considerably theoretical improvement in power consumption normalized clockspeeds, but in practice both 32nm and 22nm saw themselves thermally limited in clockspeed at their respective TJmax envelopes when employing comparable cooling, etc.

If Intel claims 67% lower power, and FinFET supposedly continues Dennard scaling for gate length, isn't that an indication that it might be better than 22nm? According to Homeles, the trade-off at higher clock speeds was something that would happen only 1 time from going to FinFET, so that also gives a sign that 14nm could be better.

At this point I expect 14nm to offer no real improvement in Fmax at TJmax, instead it will offer considerable improvement in power at any given point on the shmoo plot (and smaller die size which will help with the production cost economics).

If you are operating at the cusp of a thermally limited threshold, which Intel has been for both 32nm and 22nm, then making your chips ever smaller will do you no good in terms of clockspeed and so forth unless you also simultaneously scale the power consumption at an equivalent or better rate.

Every indication for Intel 14nm is that this is not happening. That is bad news if you want high(er) clocks with fat(ter) cores.

I do not have enough knowledge about this, but I'm more optimistic than your "every indication... not happening," although probably nothing too substantial.

 

[witeken]

But that is why I stated earlier that the wrong question is being posed. It isn't a question of the lead, it is a question of the ability of Intel to execute and leverage it into something tangible in terms of products, market share, and financials in any area outside their core x86 business units.

Intel has failed to leverage their process advantage in any appreciable manner (ex-core business) for many nodes now (45nm, 32nm, and 22nm). If they don't make considerable mobile inroads with 14nm while competing with hamstrung area-scaled 16/14nm nodes then they never will.

I think Brian Brzanich certainly has the ability to execute. In fact, going into mobile was 1 of his main goals from day 1: "I look forward to working with our leadership team and employees worldwide to continue our proud legacy, while moving even faster into ultra-mobility, to lead Intel into the next era."

The decision to put SoFIA first on TSMC's silicon to bring it faster to market is a great example of that (even though it wasn't his idea, but he put in on the roadmap nonetheless). Also, Broxton is specifically designed to allow faster execution and it combines the tablet and smartphone roadmaps into 1 product.

Also, it's only since Silvermont and actually since Airmont that Atom will enjoy N-0 silicon from day 1.

 

[R0H1T]

Intel has failed to leverage their process advantage in any appreciable manner (ex-core business) for many nodes now (45nm, 32nm, and 22nm). If they don't make considerable mobile inroads with 14nm while competing with hamstrung area-scaled 16/14nm nodes then they never will.

So how do you think their R&D expenses, especially their node shrinks, play a role in all of this? They lost close to a billion dollars last quarter on their mobile division, some (or most) of it was seemingly towards R&D while the rest being contra revenue, & they're set to loose over a billion in each of the quarters following this last one!

So what I'm trying to say is that if it takes a fraction of the cost, for Intel, to make these chips, as compared to say Qualcomm or even Apple, then why don't they price these SoC accordingly as in cheaper than their desktop or server chips? I understand you might not be the right person to answer this or you may not be in a position to do so but do you, like many others, think that Intel can (ever) command a premium in the mobile arena especially when their (real) competition consists of Apple & Samsung atm D:

 

[Homeles]

Serious question for debate. With TSMC claiming that its 16nm FinFET+ will have similar performance to Intel's 14nm transistors, and with Samsung claiming 14nm high volume production this year, just how much of a process lead does Intel actually have? Why does Intel keep claiming that it's *extending* its lead here?

Seems to me that there is one but that the gap is shrinking...at least if these PRs from TSMC/GloFo are to be believed (and that's the real debate here...whether they are to be believed).

Discuss.

TSMC and the rest are getting 2x density moving from 28nm to 20nm. Intel already has a density advantage with 22nm, and they're getting >2x the density increase with 14nm.

It's that simple, really.

 

[blackened23]

So how do you think their R&D expenses, especially their node shrinks, play a role in all of this? They lost close to a billion dollars last quarter on their mobile division, some (or most) of it was seemingly towards R&D while the rest being contra revenue, & they're set to loose over a billion in each of the quarters following this last one!

So what I'm trying to say is that if it takes a fraction of the cost, for Intel, to make these chips, as compared to say Qualcomm or even Apple, then why don't they price these SoC accordingly as in cheaper than their desktop or server chips? I understand you might not be the right person to answer this or you may not be in a position to do so but do you, like many others, think that Intel can (ever) command a premium in the mobile arena especially when their (real) competition consists of Apple & Samsung atm D:

It's a pretty simple answer. Intel has enough cash in the war chest and enough profits to do it. Admittedly intel was late to mobile and are playing catch up, and that requires R+D and money expenditures. Guess what? Intel isn't AMD. Intel has money in the bank, AMD doesn't. Intel can do it, AMD can't. Making money in the silicon business requires spending money. Always has always will. Period. Spending money on R+D doesn't guarantee success, but NOT spending money on R+D guarantees failure in the long term. Think 5 years down the road. Intel wants to be in a good position 4-5 years down the road. That's why they're spending all of this money on R+D. Is it ideal? No. But can they do it thanks to high profit levels? Yes, they can, and it doesn't hurt them due to the cash they have on hand and their net income levels. I mean. Come on. 900million spent and they still had overall 2 billion in net income. Those are great financial results while intel is hedging their bets for 4-5 years from now.

Intel does it because they can and their profit levels allow them to; 900 million is literally a drop in the bucket for them at this point and time especially when they still had 2 billion in net income. You can say what you will about their R+D spending but they own the desktop, corporate, and server markets and still make a huge chunk of change there. And for premium level portable PCs, they still own the vast majority of that market (specifically i'm referring to x86 level macbook pro and core i5/i7 based ultrabooks). All of this profit allows them to do what they do. And as mentioned, they're spending that much on R+D and still making 2 billion in profit. Doesn't sound like they're hurting to me. Rest assured AMD would do the same if they could, because their mobile products are simply not good in terms of performance per watt. But they can't because they don't have the profit or war chest. But you would see them do the same if they could because breaking into mobile is the only path towards long term sustainability. I don't think AMD likes their prospects there, but we'll see.

 

[Abwx]

It's a pretty simple answer. Intel has enough cash in the war chest and enough profits to do it. Admittedly intel was late to mobile and are playing catch up, and that requires R+D and money expenditures. Guess what? Intel isn't AMD. Intel has money in the bank, AMD doesn't. Intel can do it, AMD can't. Making money in the silicon business requires spending money. Always has always will. Period. Spending money on R+D doesn't guarantee success, but NOT spending money on R+D guarantees failure in the long term. Think 5 years down the road. Intel wants to be in a good position 4-5 years down the road. That's why they're spending all of this money on R+D. Is it ideal? No. But can they do it thanks to high profit levels? Yes, they can, and it doesn't hurt them due to the cash they have on hand and their net income levels. I mean. Come on. 900million spent and they still had overall 2 billion in net income. Those are great financial results while intel is hedging their bets for 4-5 years from now.

Intel does it because they can and their profit levels allow them to; 900 million is literally a drop in the bucket for them at this point and time especially when they still had 2 billion in net income. You can say what you will about their R+D spending but they own the desktop, corporate, and server markets and still make a huge chunk of change there. And for premium level portable PCs, they still own the vast majority of that market (specifically i'm referring to x86 level macbook pro and core i5/i7 based ultrabooks). All of this profit allows them to do what they do. And as mentioned, they're spending that much on R+D and still making 2 billion in profit. Doesn't sound like they're hurting to me. Rest assured AMD would do the same if they could, because their mobile products are simply not good in terms of performance per watt. But they can't because they don't have the profit or war chest. But you would see them do the same if they could because breaking into mobile is the only path towards long term sustainability. I don't think AMD likes their prospects there, but we'll see.

You seem obssesed by AMD since the post you re answering to doesnt mention AMD at all and speak about a completely different matter but i guess that some habits lead to some kind of pavlovian behaviour..

 

[blackened23]

Well, it's the die hard AMD fans that seem so concerned about intel's financials and no one else really caring. So I thought i'd kindly point out the results of NOT spending money on R+D, which can be seen in AMD. Don't get me wrong, the situation for AMD sucks. I would like to see them be a better competetor in mobile. But the fact of the matter is, their financial situation is their own doing due to the lavish overspending of Hector Ruiz and a culture of hubris during the Athlon era. They didn't spend money on R+D during that time and now it's biting them in the rear. As a consumer, it sucks. Having AMD strong in that sector, which they're not right now, would benefit consumers. But it is what it is. Their own doing.

So like I said what I see are die hard AMD fans pointing out flaws in intel's plan for mobile. So I thought i'd point that out. People who live in glass houses shouldn't throw stones. Right? Anyway, intel's plan isn't fool proof. But they are playing catch up, and breaking into a new sector requires spending money on R+D. It's not ideal. But a necessary evil.

 

[Wall Street]

Another thing that I keep in mind when comparing processes is that TSMC's highest density processes are good to about 2 ghz while Intel makes 4 ghz chips. There are RAM chips more dense than either, but we discount those because they are too slow for logic. I think Intel should get credit for its power and frequency characteristics in addition to density when considering process advantage.

 

[raghu78]

Well, it's the die hard AMD fans that seem so concerned about intel's financials and no one else really caring. So I thought i'd kindly point out the results of NOT spending money on R+D, which can be seen in AMD. Don't get me wrong, the situation for AMD sucks. I would like to see them be a better competetor in mobile. But the fact of the matter is, their financial situation is their own doing due to the lavish overspending of Hector Ruiz and a culture of hubris during the Athlon era. They didn't spend money on R+D during that time and now it's biting them in the rear. As a consumer, it sucks. Having AMD strong in that sector, which they're not right now, would benefit consumers. But it is what it is. Their own doing.

So like I said what I see are die hard AMD fans pointing out flaws in intel's plan for mobile. So I thought i'd point that out. People who live in glass houses shouldn't throw stones. Right? Anyway, intel's plan isn't fool proof. But they are playing catch up, and breaking into a new sector requires spending money on R+D. It's not ideal. But a necessary evil.

Intel's process lead with 14 FINFET just got cut down to a year. TSMC 16FF / Samsung 14 FINFET will be in volume production by late Q4 2014 / early Q1 2015. You can expect the Apple A9 manufactured at one of these two mega foundries to launch for holiday 2015. Guess what is the biggest threat to Intel. Its Apple. The Cyclone core is of a completely different class than any mobile core.

http://www.anandtech.com/show/7910/apples-cyclone-microarchitecture-detailed

Cyclone thrashes Silvermont/Jaguar/A15/Krait in single threaded performance. Two of the beastly cores at 1.4 Ghz match four of the other mobile cores running at 2+ Ghz. This is a fact.

Now can you imagine the impact of an ipad Pro or a low cost Macbook Air which uses the A8/A9 on the PC notebook market. Better imagine both products attacking the USD 500 - USD 750 market. Apple has shown to be not averse to platform changes. They moved from PowerPC to x86. So a USD 700 Macbook Air running A8/A9 and OS X will cause some serious headaches at Intel. The price of Intel's high end chips is USD 225+ . The A7 fabbed at Samsung is USD 19.

http://macdailynews.com/2014/02/19/apple-prepping-a8-powered-macbook-air-for-2014-release/

http://seekingalpha.com/article/2137723-apples-laptop-opportunity

This is the single biggest threat Intel and AMD face. AMD finally has some respite as it will have access to TSMC 16FF/16FF+ and Samsung 14 FINFET. btw did you know that the ARM competitors especially Apple and Qualcomm are at 28 nm planar and still competing with Intel. Apple and Qualcomm at TSMC/Sammy FINFETs are a serious headache for Intel in the smartphone and tablet business and for both Intel/AMD in the notebook business. By late 2015 Apple and Qualcomm will have FINFET SOCs ramping and thats an even playing ground with no process lead for Intel. Its not looking any easier for Intel than it was 3 years ago.

 

[videogames101]

Intel's process lead with 14 FINFET just got cut down to a year. TSMC 16FF / Samsung 14 FINFET will be in volume production by late Q4 2014 / early Q1 2015. You can expect the Apple A9 manufactured at one of these two mega foundries to launch for holiday 2015. Guess what is the biggest threat to Intel. Its Apple. The Cyclone core is of a completely different class than any mobile core.

http://www.anandtech.com/show/7910/apples-cyclone-microarchitecture-detailed

Cyclone thrashes Silvermont/Jaguar/A15/Krait in single threaded performance. Two of the beastly cores at 1.4 Ghz match four of the other mobile cores running at 2+ Ghz. This is a fact.

Now can you imagine the impact of an ipad Pro or a low cost Macbook Air which uses the A8/A9 on the PC notebook market. Better imagine both products attacking the USD 500 - USD 750 market. Apple has shown to be not averse to platform changes. They moved from PowerPC to x86. So a USD 700 Macbook Air running A8/A9 and OS X will cause some serious headaches at Intel. The price of Intel's high end chips is USD 225+ . The A7 fabbed at Samsung is USD 19.

http://macdailynews.com/2014/02/19/apple-prepping-a8-powered-macbook-air-for-2014-release/

http://seekingalpha.com/article/2137723-apples-laptop-opportunity

This is the single biggest threat Intel and AMD face. AMD finally has some respite as it will have access to TSMC 16FF/16FF+ and Samsung 14 FINFET. btw did you know that the ARM competitors especially Apple and Qualcomm are at 28 nm planar and still competing with Intel. Apple and Qualcomm at TSMC/Sammy FINFETs are a serious headache for Intel in the smartphone and tablet business and for both Intel/AMD in the notebook business. By late 2015 Apple and Qualcomm will have FINFET SOCs ramping and thats an even playing ground with no process lead for Intel. Its not looking any easier for Intel than it was 3 years ago.

samsung 14nm =/= intel 14nm

 

[Idontcare]

If Intel claims 67% lower power...

If true then of course that changes everything...but I had read 14nm was 67% "of the power", not 67% "lower" power.

If 14nm is 0.67x the power of 22nm then power is only 33% lower, not 67% lower.

So the question is - is 14nm 0.67x that of 22nm, or is it 0.33x that of 22nm in terms of power consumption at comparable clockspeed and operating temperature?

22nm was only around 67% the power of 32nm, 0.67x under best conditions, so I'm not expecting miracles with 14nm where static leakage only continues to be more of an issue than it was at 22nm and 32nm.

So how do you think their R&D expenses, especially their node shrinks, play a role in all of this? They lost close to a billion dollars last quarter on their mobile division, some (or most) of it was seemingly towards R&D while the rest being contra revenue, & they're set to loose over a billion in each of the quarters following this last one!

So what I'm trying to say is that if it takes a fraction of the cost, for Intel, to make these chips, as compared to say Qualcomm or even Apple, then why don't they price these SoC accordingly as in cheaper than their desktop or server chips? I understand you might not be the right person to answer this or you may not be in a position to do so but do you, like many others, think that Intel can (ever) command a premium in the mobile arena especially when their (real) competition consists of Apple & Samsung atm D:

The contra-revenue strategy is a new tool in Intel's toolbox that is only possible because they are operating in a market segment with this business unit for which Intel is not an effective monopoly. That's not philosophical contemplation, that is straight from folks at Intel who deal with this situation.

The regulatory environment for the mobile group is delineated in a specific manner such that it can operate with contra-revenue.

The reason Intel doesn't simply discount the price of the chips is one of standard business practice - you never discount the price of something if you have any aspirations of going back to the higher price someday. Your customers simply won't stand for the price hike down the road.

So you keep the price where you want your customers to pay, it sets their expectations and yours in appropriate fashion, and then you subsidize the product development aspects on their side.

This lowers their risk of losses should the higher priced (to cover Intel's chip price) Intel-enabled product be a flop in the market. If it is a flop then its now loss to the OEM, they don't have an un-amortized R&D bill on their books (its on Intel's instead).

But should the product take-off, then presumably it does so while being positioned at the right price-point so as to be able cover the elevated chip costs associated with using Intel's chips while still leaving money on the table for the OEM to be OK with the new Intel-enabled product cannibalizing sales of their other product line-ups.

The BOM is up because it has Intel expensive chips inside, but it commands a price premium for the features and performance. So it would all work out.

It is an interesting strategy, one that Intel can afford financially and one that they can do in terms of the regulations because the products are not competing in a business environment where Intel is a duopoly or effective monopoly.

Another thing that I keep in mind when comparing processes is that TSMC's highest density processes are good to about 2 ghz while Intel makes 4 ghz chips. There are RAM chips more dense than either, but we discount those because they are too slow for logic. I think Intel should get credit for its power and frequency characteristics in addition to density when considering process advantage.

Idrive is the difference. Having a small numbered node label doesn't mean much if the underlying electrical parametrics are inferior.

 

[IntelUser2000]

If Intel claims 67% lower power,

They are spinning, just as competitors are.

http://forums.anandtech.com/showpost.php?p=36276848&postcount=34

Look at the "67% reduction" graph. It's going from about 75%"to about 0%. Now that really makes no sense, unless you assume 0% means 1x, and 75% means 1.75x.

"67%" reduction literally means x% - 67. So if we assume x as 75%, the result is 8%. 1.08/1.75 = 61%, or 39% lower.

At 45/40nm, Intel had a lead of about 1,5 years. At 32/28nm and 22/20nm, Intel had a lead of 2 years. Intel's 14nm node, however, will not correspond to the competition's 20nm FinFET node ("14/16nm"), but rather to their 10nm node. So because 20nm will be here for over 3 years (50% longer), Intel's lead will increase to a staggering 3+ years.

Again, Intel is spinning. But so are others. So its really the same.

20nm is denser than Intel's 22nm, and partway between Intel's 14nm and 22nm. Transistor performance is said to be nearly equal, if anything maybe few % better on Intel.

14nm is less dense than Intel's 14nm, and about a year behind(that's true for both Samsung and TSMC, 2H 2015, at most few months off). 14nm is better than Intel's 22nm in transistor performance, but worse than Intel's 14nm. Honestly, I put Intel's lead at 2 years(cumulative gains of TTM, Idrive, power use). Anything else is a spin.

Samsung hitting 14nm this year...probably. For their NAND FLASH A lot of guys are pretty far down for flash. The requirements on that stuff isn't nearly as stringent as that for CPUs/GPUs/RAM.

That's a spin too. They'll get 20nm first, which will be end of the year, and 14nm a year after that. And their "14nm" is a half-node just like TSMC and GF. Countless times I've seen such claims of "acceleration" and countless times they've never done it. Intel spins, Samsung spins, TSMC spins. I believe no spins and count on my own.

 

[Nothingness]

If true then of course that changes everything...but I had read 14nm was 67% "of the power", not 67% "lower" power.

If 14nm is 0.67x the power of 22nm then power is only 33% lower, not 67% lower.

So the question is - is 14nm 0.67x that of 22nm, or is it 0.33x that of 22nm in terms of power consumption at comparable clockspeed and operating temperature?

My guess is that it's 0.67x, since Intel showed this Broadwell vs Haswell power graph.

BTW excellent post with information and analysis not directly coming from marketing :thumbsup:

 

[raghu78]

samsung 14nm =/= intel 14nm

Intel does not have a significant transistor density lead at 14nm over the foundries TSMC 16FF/ Samsung 14FINFET. The lead is marginal if any. All the 3 companies use a M1 metal pitch of 64 nm. All of them use dual pattern immersion lithography. According to TSMC official statements Intel's 14 nm is matched in transistor performance by TSMC 16FF+.

We are yet to see official data on the latest FINFET process of these 3 companies at technical seminars like IEDM, ISSCC where they talk about the extremely technical details like SRAM cell size, Idsat etc. But its fair to say that Intel does not have a full node like lead by any stretch of imagination in both transistor density and performance. Intel has a 1 year lead though, which is down from the traditional 2+ year/ full node lead.

 

[Idontcare]

Intel does not have a significant transistor density lead at 14nm over the foundries TSMC 16FF/ Samsung 14FINFET. The lead is marginal if any. All the 3 companies use a M1 metal pitch of 64 nm. All of them use dual pattern immersion lithography. According to TSMC official statements Intel's 14 nm is matched in transistor performance by TSMC 16FF+.

We are yet to see official data on the latest FINFET process of these 3 companies at technical seminars like IEDM, ISSCC where they talk about the extremely technical details like SRAM cell size, Idsat etc. But its fair to say that Intel does not have a full node like lead by any stretch of imagination in both transistor density and performance. Intel has a 1 year lead though, which is down from the traditional 2+ year/ full node lead.

Going back 2+ decades, TSMC has never matched the xtor performance of Intel at *any* node.

Do you really believe TSMC's statements claiming they have erased a 20yr gap with their first stab at productized finfets compared to Intel's second-gen finfets?

TSMC doesn't have the R&D budget to pull off such miracles, let alone any kind of precedence with their R&D team to suggest such a coup is even remotely plausible.

But it doesn't matter where Intel is at in terms of the process space relative to TSMC so long as Intel isn't using its process technology to fab products that compete with any of TSMC's customers.

Which so far has pretty much been the reality of the situation. One that Intel would love to change, but that change has yet to materialize outside of projections and market goals.

 

[Arachnotronic]

Going back 2+ decades, TSMC has never matched the xtor performance of Intel at *any* node.

Do you really believe TSMC's statements claiming they have erased a 20yr gap with their first stab at productized finfets compared to Intel's second-gen finfets?

TSMC doesn't have the R&D budget to pull off such miracles, let alone any kind of precedence with their R&D team to suggest such a coup is even remotely plausible.

But it doesn't matter where Intel is at in terms of the process space relative to TSMC so long as Intel isn't using its process technology to fab products that compete with any of TSMC's customers.

Which so far has pretty much been the reality of the situation. One that Intel would love to change, but that change has yet to materialize outside of projections and market goals.

IDC,

Thanks for your great posts. I guess the question is, how does TSMC get away with making statements such as the following on its earnings call:

Then I will cover the updates on 16-FinFET plus and I’ve two general offers for customers, 16-FinFET and 16-FinFET plus. 16-FinFET plus offers 15% speed improvement the same total power compared to 16-FinFET . More importantly 16-FinFET plus offers 30% total power reduction at the same speed compared to 16-FinFET. Our 16-FinFET plus matches the highest performance of both 60 nanometer and 40 nanometer technologies in the market today. Compared to our own 20 SoC, 16 16-FinFET plus offers 40% speed improvement.

More importantly, if this is true, how would they have pulled it off and left Intel - which had FinFETs at 22nm - without a process node advantage?

 

[witeken]

Intel's process lead with 14 FINFET just got cut down to a year. TSMC 16FF / Samsung 14 FINFET will be in volume production by late Q4 2014 / early Q1 2015.

You're a 'victim' of TSMC's/Samsung's/GlobalFoundries' marketing lies. 14nm FinFET is nowhere near Intel's 14nm process. For starters, it's made on the same 20nm process as the foundries' 20nm process, which is 1.5x less dense and much costlier. Even if we take their FinFET+, but we don't know when that will be released, it might be immediately or 1 year after FinFET, Intel's process is still 1.3x denser. Secondly, Intel's 14nm will be their 2nd generation Tri-Gate, compared to TSMC's 1st gen.

You can expect the Apple A9 manufactured at one of these two mega foundries to launch for holiday 2015. Guess what is the biggest threat to Intel. Its Apple. The Cyclone core is of a completely different class than any mobile core.

http://www.anandtech.com/show/7910/apples-cyclone-microarchitecture-detailed

Yeah, and it's also extremely power hungry.

Cyclone thrashes Silvermont/Jaguar/A15/Krait in single threaded performance. Two of the beastly cores at 1.4 Ghz match four of the other mobile cores running at 2+ Ghz. This is a fact.

Cyclone is 1.4x faster than 2.4GHz Silvermont, but it consumes ~3x as much energy per core, which means the performance/watt is awful.

And you forget that Intel also has Core, which in its turn demolishes Cyclone. So Apple could also ask Intel for a dualcore Broadwell or Skylake.

Now can you imagine the impact of an ipad Pro or a low cost Macbook Air which uses the A8/A9 on the PC notebook market. Better imagine both products attacking the USD 500 - USD 750 market. Apple has shown to be not averse to platform changes. They moved from PowerPC to x86. So a USD 700 Macbook Air running A8/A9 and OS X will cause some serious headaches at Intel. The price of Intel's high end chips is USD 225+ . The A7 fabbed at Samsung is USD 19.

http://macdailynews.com/2014/02/19/apple-prepping-a8-powered-macbook-air-for-2014-release/

http://seekingalpha.com/article/2137723-apples-laptop-opportunity

Your 20$ estimate is way too optimistic. And comparing the price of a high-end chip too a slow ARM one doesn't make much sense either.

This is the single biggest threat Intel and AMD face. AMD finally has some respite as it will have access to TSMC 16FF/16FF+ and Samsung 14 FINFET. btw did you know that the ARM competitors especially Apple and Qualcomm are at 28 nm planar and still competing with Intel. Apple and Qualcomm at TSMC/Sammy FINFETs are a serious headache for Intel in the smartphone and tablet business and for both Intel/AMD in the notebook business. By late 2015 Apple and Qualcomm will have FINFET SOCs ramping and thats an even playing ground with no process lead for Intel. Its not looking any easier for Intel than it was 3 years ago.

I don't think so. Intel's at a comfortable 2 year lead, which will increase to 3 year at 14nm (TSMC's 10nm is the equivalent of Intel's 14nm) and further expand at 10nm.

 

[witeken]

They are spinning, just as competitors are.

http://forums.anandtech.com/showpost.php?p=36276848&postcount=34

Look at the "67% reduction" graph. It's going from about 75%"to about 0%. Now that really makes no sense, unless you assume 0% means 1x, and 75% means 1.75x.

"67%" reduction literally means x% - 67. So if we assume x as 75%, the result is 8%. 1.08/1.75 = 61%, or 39% lower.

That makes sense, thanks.

 

[witeken]

IDC,

Thanks for your great posts. I guess the question is, how does TSMC get away with making statements such as the following on its earnings call:

More importantly, if this is true, how would they have pulled it off and left Intel - which had FinFETs at 22nm - without a process node advantage?

Interesting, I though FinFET+ was just a density improvement. Then I think FinFET Plus will likely be released 1 year after FinFET, around the start of 2017, which is more than 2 years behind Intel's 14nm (and still 1.3x less dense).