Intel investing $1B to meet 14nm demand

[Jimzz]

https://www.anandtech.com/show/1341...4nm-demand-prioritizing-highend-core-and-xeon


Not looking good to spend that much on a old tech if 10nm is trully that close to being used for mass-production. This is not a single site upgrade either, its for 4 sites (2 outside the US). I am guessing even in 2019 10nm will still be limited for intel at best for them.

Could open the door for AMD if TSMC can produce enough at 7nm with good quality.

 

[Dayman1225]

Interestingly they mention investing in Israel for this problem, but their Israel fab is 22/10nm not 14nm?

 

[LTC8K6]

Sounds good to me. I'm glad Intel is working on the 14nm supply and packaging problems. It's been an incredible node, and it has more life left at the top, imo.

Plus, the competition is great.

 

[Nothingness]

There's a typo:

We’re prioritizing the production of Intel® Xeon® and Intel® Core™ processors so that collectively we can serve the high-performance segments of the market.

Should read:

We’re prioritizing the production of Intel® Xeon® and Intel® Core™ processors so that collectively we can maximize benefits.

 

[Campy]

Good, the last thing we need is supply problems to create another long-lasting price increase.

 

[LTC8K6]

Good, the last thing we need is supply problems to create another long-lasting price increase.

May just reduce it from a really long delay, to just a long delay.

Lots of doom and gloom forecasts out there about Intel.

 

[Topweasel]

May just reduce it from a really long delay, to just a long delay.

Lots of doom and gloom forecasts out there about Intel.

I doubt it and I doubt this will really affect stock level before it shouldn't matter. I mean Intel is big enough to waste a billion dollars just to tell stockholders that they are trying. But this is a move for Late 2019 or early 2020 at best. If they need increased 14nm production at that time then they are screwed in more ways than 1.

By the time Intel gets this up in running they need to have 10nm ready for prime time. Whether it's to compete, on clocks or core count, power usage, or size. If 7nm AMD goes well they will it will be do or die time for Intel to get 10nm out the door. If they don't the supply problem goes away anyways, because they will start to lose OEM orders.

 

[jpiniero]

I doubt it and I doubt this will really affect stock level before it shouldn't matter. I mean Intel is big enough to waste a billion dollars just to tell stockholders that they are trying. But this is a move for Late 2019 or early 2020 at best. If they need increased 14nm production at that time then they are screwed in more ways than 1.

Thing is, Cooper Lake is literally two 28 core dies fused together. That's going to eat up a ton of wafers, and while Rome should take a decent amount of share there's only so much share that would be realistic. And despite what Intel says, I don't think Icelake Server will ever be in any volume.

 

[Topweasel]

Thing is, Cooper Lake is literally two 28 core dies fused together. That's going to eat up a ton of wafers, and while Rome should take a decent amount of share there's only so much share that would be realistic. And despite what Intel says, I don't think Icelake Server will ever be in any volume.

A ton of wafers with questionable Yields. Yikes. Still if that is the late 2019 solution they are going to be hurting. I can see how supply might be an issue if they keep up this demand. But I will start to wonder if there is even going to be as much demand at that point.

 

[maddie]

A ton of wafers with questionable Yields. Yikes. Still if that is the late 2019 solution they are going to be hurting. I can see how supply might be an issue if they keep up this demand. But I will start to wonder if there is even going to be as much demand at that point.

How does the saying go? When only lemons are available, learn to make lemonade.

The amusing thing is that some are only now praising the taste of lemonade

 

[Mopetar]

A ton of wafers with questionable Yields. Yikes.

Yields for Intel on 14 nm should be as good as they're going to get (and reasonably good) by this point in time. I think that problem is that they had planned for a good chunk of their production to be on 10 nm already, but since that didn't pan out they're tight on wafers, especially when they need to make enough massive dies to hit their bin targets.

 

[Topweasel]

Yields for Intel on 14 nm should be as good as they're going to get (and reasonably good) by this point in time. I think that problem is that they had planned for a good chunk of their production to be on 10 nm already, but since that didn't pan out they're tight on wafers, especially when they need to make enough massive dies to hit their bin targets.

Yeah I am just saying Yields on a 14nm 28c part are going to be bad. First you have the fact that you are taking a big hit on just how many dies you can fit on a wafer. Then each die is going to be so big that there is a much larger chance of multiple defects. Then on top of that now it's going to take two of them for one CPU. That's harsh.

 

[jpiniero]

I used a die size calculator and got 93% yield (for fully enabled) if the defect rate was 0.01/sqcm. Which doesn't sound so bad until you realize you only get 64 dies per wafer and Cooper Lake would need two per chip.

What 14 nm's yield is anyone's guess.

 

[Topweasel]

I used a die size calculator and got 93% yield (for fully enabled) if the defect rate was 0.01/sqcm. Which doesn't sound so bad until you realize you only get 64 dies per wafer and Cooper Lake would need two per chip.

What 14 nm's yield is anyone's guess.

Yeah that's what I am getting at. AMD's Yields on full featured dies for SR were like 83% and that was supposed to be really great. Binning and lasering probably pushed it up to 90% usable. Intel will be able to do that to that as well. But how much before it starts to affect its usability. I saw a figure that had the SL-SP at something like 40% full featured. Which means for their top sku's they will only get 20% of the wafer potential that they could have compared to 40% before.

But I mean even if everything is perfect. That will be a whole lineup of CPU's at only 32 CPU's per wafer. That sounds bad. Just in silicon cost at what 10k per wafer. It's like $310 a CPU and again that's at 100%. 90% $360 80% $400. Now Intel more than any company outside Oil and Pharma companies could deal with a significant loss in margin. But it's not great. If it is costing you $400 per CPU and your competitor is matching it on $100-$150 dollars.

 

[jpiniero]

Intel's 14 nm wafers actually shouldn't be that much, although if they have to depreciate now that extra $1B of equipment, that adds some cost.

OTOH I wouldn't be surprised that yes TSMC is charging over 10 grand for a 7 nm wafer.

 

[Mopetar]

Yeah I am just saying Yields on a 14nm 28c part are going to be bad. First you have the fact that you are taking a big hit on just how many dies you can fit on a wafer. Then each die is going to be so big that there is a much larger chance of multiple defects. Then on top of that now it's going to take two of them for one CPU. That's harsh.

The bad dies can still be used so it’s hardly a case of complete waste.

For what Intel charges for their top chips than could afford to only have a few fully functioning dies per wafer.

The yields aren’t a problem. The issue is that with chips this big, even with a perfect yield you get a small fraction of chips compared to the mainstream parts. There aren’t enough wafers to go around.

 

[Topweasel]

The bad dies can still be used so it’s hardly a case of complete waste.

For what Intel charges for their top chips than could afford to only have a few fully functioning dies per wafer.

The yields aren’t a problem. The issue is that with chips this big, even with a perfect yield you get a small fraction of chips compared to the mainstream parts. There aren’t enough wafers to go around.

Yeah but you can only take so many defects in so many area's before the CPU becomes actually unusable or just unsellable. Look at Ryzen So many configurations for one realitively small chip that would have a severely less likely chance of having multiple systems impacted by defects and still ~90% usable yield.

So with this. What's the yeild like at top level. 2-4 28c dies? Now one or two of the top CPU per Wafer? That's got to suck. Best case scenario I think is probably ~80% usable dies. 20-30% Fully functional dies. Double that up and now they are effectively 40% and 10-15% (because they lose half the working dies to make one CPU) compared to SL-SP.

 

[Mopetar]

It’s probably better than that. A lot of times companies intentionally bin dies below their full capabilities because they can only sell a set number of dies.

It used to be quite popular to buy phenom chips and unlock them. AMD had chips that weren’t defective, but they artificially locked them down because they had too many full chips and there wasn’t enough demand. I wouldn’t be surprised if Intel were hobbling the bottom part of their bin either under normal circumstances.

If you have 90% of chips yield good you don’t keep them that way. You raise the bar on what those chips need to hit in terms of performance until you reach a point where enough fall out of that bin based on what sales are projected to be. Even if Intel could get 100% yield on their top chip, they can’t sell them all at the price they want to charge. The same goes for AMD.

 

[maddie]

When we get the argument that Intel can afford lower margins as their costs are still well below the selling price, we often forget that it's a public company, not a private one. Yes the company will survive, but will the board, upper management and even some lower level employees also survive the inevitable reduction in share price caused by falling margins. I suggest this is on the mind of many there now.

 

[scannall]

All this really says is that 10nm is still stuck in the mud, and will likely get pushed back again.

 

[HurleyBird]

With the lag time on something like this, it could really backfire if Zen 2 "Conroes" them and AMD is able to get good allocation from TSMC.

 

[DrMrLordX]

On the one hand, I see this as a positive for Intel, since they are still able to push product despite Spectre/Meltdown, the 10nm fiasco, and pressure from multiple angles.

On the other, I see it as a negative since they are sinking a lot of money into expanding production on a fully-matured node. We are going to see 14nm from them for awhile.

 

[LTC8K6]

We did hear that 14nm++ was better than the intial 10nm node in every way.

Presumably 14nm+++ increases that difference.

So perhaps 14nm won't be as bad as we think against 10/7nm?

 

[Dayman1225]

On the one hand, I see this as a positive for Intel, since they are still able to push product despite Spectre/Meltdown, the 10nm fiasco, and pressure from multiple angles.

On the other, I see it as a negative since they are sinking a lot of money into expanding production on a fully-matured node. We are going to see 14nm from them for awhile.

Most 14nm equipment is usable for 10nm.

 

[frozentundra123456]

When we get the argument that Intel can afford lower margins as their costs are still well below the selling price, we often forget that it's a public company, not a private one. Yes the company will survive, but will the board, upper management and even some lower level employees also survive the inevitable reduction in share price caused by falling margins. I suggest this is on the mind of many there now.

Their margins are still far higher than AMD, and nobody seems to be calling for the resignation of their top management.