(Semiwiki) Intel 14nm Delayed Again?

[Homeles]

A question regarding the rumors about the delayed 450 mm EUV to 2023, if this is indeed the case then Moore's law is completely thrown out of the window unless we find a new material and quickly, right? I mean, Moore's law is already slowing but can you even hit 7 nm before the end of the decade without the 450 mm EUV?

I think you're confusing things. EUV and 450mm wafer processing are two separate technologies. EUV allows for smaller patterning; 450mm wafers are a cost saving technology.

EUV isn't necessary at 10nm. Quad patterning works, although it is less optimal from a cost standpoint. As far as 7nm goes... maybe EUV will actually be ready by then. It's a shame we didn't get EUV at 32nm like it was initially projected... chips would be dirt cheap.

There are still other technologies that can reduce costs, like Multiple Electron Beam.

Moore's Law is actually accelerating for Intel. Not significantly so, but they are outpacing the $/transistor curve for the next 2-3 nodes.

 

[bullzz]

http://semiaccurate.com/2014/03/06/intel-altera-tsmc-sad-sad-state-tech-reporting/

charlies article on altera moving to tsmc. hilarious (esp linking to McDs website)

 

[SiliconWars]

Moore's Law is actually accelerating for Intel. Not significantly so, but they are outpacing the $/transistor curve for the next 2-3 nodes.

Link?

 

[witeken]

Link?

 

[jpiniero]

Intel's actions don't exactly support that chart.

 

[witeken]

What do you mean?

 

[Homeles]

Intel's actions don't exactly support that chart.

 

[videogames101]

Intel's actions don't exactly support that chart.

now whether people buy the transistors is another story, that curve simply speaks to production costs

 

[Piroko]

Last I checked Moore's law was minimum cost over timespan. That graph will only correlate as long as there are no delays...

 

[jpiniero]

What do you mean?

Closing the two 14 nm fabs, Haswell Refresh, cutting back the Broadwell launch then delaying because yield isn't perfect, etc... It is a marketing slide after all, probably represents an ideal state that Intel might never reach.

It's not the end of the world for Intel since their competition is going through the same issues. TSMC's 20 nm isn't going to be any cheaper per transistor than their 28 nm in 2014.

 

[mrmt]

It's not the end of the world for Intel since their competition is going through the same issues. TSMC's 20 nm isn't going to be any cheaper per transistor than their 28 nm in 2014.

Cheaper to who? TSMC or TSMC's customers?

 

[Homeles]

Closing the two 14 nm fabs, Haswell Refresh, cutting back the Broadwell launch then delaying because yield isn't perfect, etc... It is a marketing slide after all, probably represents an ideal state that Intel might never reach.

None of that has to do with cost/tansistor, though.

 

[AtenRa]

cost/transistor alone means nothing. You have to factor in the higher R&D/capex cost, lower fab utilization, smaller or bigger die size, yields etc etc.
At the end of the day, 14nm wafers today could be 2x or more than 22nm.

Personally i believe that is the reason for the 14nm delays. 4 year old plans made for the 22nm were not taking in to account a declining market in 2012-2013. It takes more than 2 years to depreciate each node and Intel needs to keep selling 22nm products longer than planed.

 

[witeken]

Why would they do that if they could release 14nm products with higher margins? I mean, the fabs are/were ready, but yields simply weren't high enough. If there weren't yield issues, Intel would be volume producing their 14nm products since Q4 for release in Q2/Q3.

Yields are the only reason for the delay.

 

[AtenRa]

Why would they do that if they could release 14nm products with higher margins? I mean, the fabs are/were ready, but yields simply weren't high enough. If there weren't yield issues, Intel would be volume producing their 14nm products since Q4 for release in Q2/Q3.

Yields are the only reason for the delay.

Because nobody would buy higher priced products. Every OEM feels fine with low cost Ivy and Haswell. Why bother with higher cost Broadwell at this time ??? the market is going for cheaper Laptops and Tablets, not high end Laptops. It is the reason UltraBooks never got the sales Intel wanted.

 

[Mondozei]

Because nobody would buy higher priced products. Every OEM feels fine with low cost Ivy and Haswell. Why bother with higher cost Broadwell at this time ??? the market is going for cheaper Laptops and Tablets, not high end Laptops. It is the reason UltraBooks never got the sales Intel wanted.

Exactly. The raw processing power has been "good enough" for years now, and even in the budget segment has this been true. The big issue has been battery life but Haswell made huge improvements there.

Another issue is that Intel is a lot stronger in the tablet space than in the smartphone space, where battery life is even more crucial.
Nevertheless, the options of not going for 14 nm this year isn't there. Intel can't assume that TSMC will do badly, and Samsung is rapidly increasing their own capex costs to bring themselves faster into the ~20 nm node. And unlike Intel, Samsung's business is a lot more diversified and their tech business has growing revenue rates which far surpasses the growth Intel can manage. Intel can't afford to wait too much longer, but it can't afford to jump too fast either. It's not a good situation for them. This is where they have traditionally bled out their opponents but now they are being bled out.

 

[Homeles]

Because nobody would buy higher priced products. Every OEM feels fine with low cost Ivy and Haswell.

I would love to see your source for that one. And why would Broadwell be higher cost? Intel doesn't lower its prices meaningfully over time.

 

[R0H1T]

I would love to see your source for that one. And why would Broadwell be higher cost? Intel doesn't lower its prices meaningfully over time.

How about you prove otherwise ? Intel is heavily subsidizing Atom, through the sale of their high end/server chips, & there's still no real market for it yet ! The mobile/tablet line of chips will have to be subsidized for (at least) another year or two that'll further pull their margins down, so they'll have to increase the prices for desktop/notebook parts or the server/HPC chips, since no one is going to buy an overpriced Atom SoC even if it gives 2x the performance of its nearest competitor

 

[AtenRa]

I would love to see your source for that one. And why would Broadwell be higher cost? Intel doesn't lower its prices meaningfully over time.

Do you believe that Intel will lower its Margins for Broadwell ?? Because if they would sell low volume of Broadwell CPUs now at the same price as Haswell, then they would have to significantly lower there Margins to achieve that.

14nm is currently more expensive with lower yields than 22nm. Broadwell Die size is smaller than Haswell but not that much. Broadwell today cost more than Haswell, how Intel will sell at the same price and keep the high margins they have ???
They would/could do that if the market volumes were increasing. The market currently is stagnant at best. Intel Fab utilization is decreasing not increasing. They already mentioned they will close a fab.

But even if they would sell Broadwell at the same prices as Haswell, OEMs would not jump like crazy for High End High cost Laptop CPUs. As i have said before, the market is transitioning to cheaper Lower end laptops and Tablets.

 

[Sweepr]

Broadwell Die size is smaller than Haswell but not that much.

Citation needed.

 

[bullzz]

wow one audacious claim followed by the other
@atenra
"Because nobody would buy higher priced products" - tell that to intel since they just made $30+ billion by selling more high end products compared to low end. clue: their ASP went up when pc sales are going down

"Every OEM feels fine with low cost Ivy and Haswell. Why bother with higher cost Broadwell at this time" - so are u claiming apple/ms/lenovo are going to release only haswell products and not broadwel ones? the "every OEM" ur talking abt dont care abt processors and what to sell it $10 less than their competitor

@rohit
"How about you prove otherwise ?" - well he didnt make any claim. not sure what he has to prove

"Intel is heavily subsidizing Atom, through the sale of their high end/server chips, & there's still no real market for it yet" - I think you are getting confused here. intel is subsidizing atom for tablet market, not for micro server market. atleast we dont have info on that
yes intel is losing money on atom. but they compensate that by not jacking up prices. heard of market growth? their server division is expected to grow. and that is how they are planning to compensate

 

[R0H1T]

@rohit
"How about you prove otherwise ?" - well he didnt make any claim. not sure what he has to prove

So you think device makers, server/PC vendors & phone manufacturers, are willing to pay more for the chips than what they're currently doing ? I think common sense says that the trend is for cheaper devices & only lifestyle products with a brand name(like Apple) can command a premium & certainly Samsung's fall from grace will attest to that, their growth rate is also down drastically as compared to last year, cause a bunch of Mediatek powered devices are proving more than a handful for them, due to their lower prices.

"Intel is heavily subsidizing Atom, through the sale of their high end/server chips, & there's still no real market for it yet" - I think you are getting confused here. intel is subsidizing atom for tablet market, not for micro server market. atleast we dont have info on that
yes intel is losing money on atom. but they compensate that by not jacking up prices. heard of market growth? their server division is expected to grow. and that is how they are planning to compensate

In a growing market, where you aren't making much money, the cost of subsidies is ballooning & the PC division is down(YoY) with the servers barely making it, in terms of positive growth rate. Oh I forgot to mention that the fab utilization is going down & R&D costs are through the roof now !

So what we're looking at is single digit growth for servers, barely able to negate the decline in traditional PC sales, increased R&D & other related expenditures(plus inflation & employee wages) with no bottom in place for their Atom line. Their only saving grace is that the spending for 14nm node is behind them & they own the fabs themselves, therefore if the Atom line doesn't take off(you can quote me on that) then expect the next node shrink to come one half to a full year behind schedule, that'll make the Moore's law pretty much obsolete, after factoring in the delay this year.

 

[witeken]

Do you believe that Intel will lower its Margins for Broadwell ?? Because if they would sell low volume of Broadwell CPUs now at the same price as Haswell, then they would have to significantly lower there Margins to achieve that.

14nm is currently more expensive with lower yields than 22nm. Broadwell Die size is smaller than Haswell but not that much. Broadwell today cost more than Haswell, how Intel will sell at the same price and keep the high margins they have ???
They would/could do that if the market volumes were increasing. The market currently is stagnant at best. Intel Fab utilization is decreasing not increasing. They already mentioned they will close a fab.

But even if they would sell Broadwell at the same prices as Haswell, OEMs would not jump like crazy for High End High cost Laptop CPUs. As i have said before, the market is transitioning to cheaper Lower end laptops and Tablets.

Your whole point is flawed, in my opinion. It makes always sense to go to a lower node. If not, then Intel wouldn't do to it or wait longer, which they aren't going to do in the near future.

By the way, your claim that Broadwell isn't that much smaller, is incorrect: Broadwell ULT is about 63% the size of Haswell ULT.

 

[witeken]

So what we're looking at is single digit growth for servers, barely able to negate the decline in traditional PC sales, increased R&D & other related expenditures(plus inflation & employee wages) with no bottom in place for their Atom line. Their only saving grace is that the spending for 14nm node is behind them & they own the fabs themselves, therefore if the Atom line doesn't take off(you can quote me on that) then expect the next node shrink to come one half to a full year behind schedule, that'll make the Moore's law pretty much obsolete, after factoring in the delay this year.

It doesn't seem that your doom scenario is going to come true: Intel already stated that volume production of 10nm will start in 2015. TSMC's volume production of 10nm (= Intel's 14nm) will start at least 1 year later. With 14nm they will likely be able to take an enormous amount of market share in tablets and smartphones, which they will strengthen with 10nm. At that time (2016 and further), TSMC will be more than 3 years behind Intel, so no one is going to want TSMC's silicon.

 

[Homeles]

How about you prove otherwise ?

Go learn about "burden of proof," and then come back to me. You just flunked out of Writing Arguments 101.

Do you believe that Intel will lower its Margins for Broadwell ?? Because if they would sell low volume of Broadwell CPUs now at the same price as Haswell, then they would have to significantly lower there Margins to achieve that.

14nm is currently more expensive with lower yields than 22nm. Broadwell Die size is smaller than Haswell but not that much. Broadwell today cost more than Haswell, how Intel will sell at the same price and keep the high margins they have ???
They would/could do that if the market volumes were increasing. The market currently is stagnant at best. Intel Fab utilization is decreasing not increasing. They already mentioned they will close a fab.

But even if they would sell Broadwell at the same prices as Haswell, OEMs would not jump like crazy for High End High cost Laptop CPUs. As i have said before, the market is transitioning to cheaper Lower end laptops and Tablets.

You're basing all of this off an erroneous belief that Broadwell is more expensive than Haswell. Broadwell is a good 60% of the size of Haswell (GTx for GTx, e.g. BRW DC-GT2 vs HSW DC-GT2), IIRC, and the density of 14nm should be more than double that of 22nm. Yields should be mostly fixed by now.

If it is going to be more expensive, it won't be by much. Regardless, Broadwell will be cheaper for the same level of performance, offer better battery life, and other good stuff.