[TechEye] GlobalFoundries to buy IBM Semi

[krumme]

You forgetting 32nm SOI HKMG, GloFo's 28nm HKMG Bulk is half-Node of the 32nm process. GloFo is not 1-2 years behind TSMC, their 20/14nm are on par with TSMC.

We can hope that. But try to go 3 years back and look at the gf press material eg the writeups here at AT. It was just so far from reality its incredible. Its 2 years back i stopped reading their fairytales. Now i beliewe it when there is actual products on the market using the new nodes.

 

[mrmt]

This. It is a huge mess. What is the purpose of gf if its 1-2 years behind tsmc?

To grab whatever leftovers TSMC couldn't or didn't want to get. Globalfoundries isn't competition to tsmc, let alome Intel, but for UMC, SMIC and others.

 

[lefty2]

There are rumours that GlobalFoundries will skip 20nm altogether:
https://markets.jpmorgan.com/research/email/prtvmqli/GPS-1343864-0

That makes sense because 20nm only exists as an accident in the first place.
This came out in the TMSC conference call:
"We got committed, and then suddenly, we realized that there was a need for FinFET or a faster FinFET. Since then, one of the guerillas in the business announced that there was a FinFET business. Now if we hadn't committed us already to 20, we might have skipped 20. I think we would have skipped 20. But we were committed already."

 

[mrmt]

There are rumours that GlobalFoundries will skip 20nm altogether:
https://markets.jpmorgan.com/research/email/prtvmqli/GPS-1343864-0

That makes sense because 20nm only exists as an accident in the first place.
This came out in the TMSC conference call:
"We got committed, and then suddenly, we realized that there was a need for FinFET or a faster FinFET. Since then, one of the guerillas in the business announced that there was a FinFET business. Now if we hadn't committed us already to 20, we might have skipped 20. I think we would have skipped 20. But we were committed already."

Funny interpretation of accident. What you see as an accident I see as a huge miscalculation from TSMC's part.

 

[lefty2]

Funny interpretation of accident. What you see as an accident I see as a huge miscalculation from TSMC's part.

Maybe not, because they would have lost the customer if they had not gone 20nm. But the question is why is Globalfoundries going 20nm? Monkey see, monkey do?

 

[mrmt]

Maybe not, because they would have lost the customer if they had not gone 20nm. But the question is why is Globalfoundries going 20nm? Monkey see, monkey do?

This looks like more of risk mitigation from both parties. Instead of developing a new node AND finfets at the same time, they develop a node first, get things stabilized and then focus on develop and deploy finfets.

 

[Phynaz]

Im sure you have technical data to provide ???

Edit: Not to mention that 28nm SLP (Gate First) got 40% lower power, 30% higher performance and 2x times the density of 40nm LP (Gate Last). Not only that, being a Half Node of 32nm means they can save huge amount of resources, time and money both GloFo and AMD.
Also people always forget that Gate First is 10-20% cheaper than the same process on Gate Last.

Common sense man.

If Gate First is better how come in a year nobody will be using it?
http://www.eejournal.com/archives/articles/20111114-gate/

 

[AtenRa]

Common sense man.

If Gate First is better how come in a year nobody will be using it?
http://www.eejournal.com/archives/articles/20111114-gate/

I didnt say it is better, i have asked him to provide technical data that TSMCs 28nm is "overall superior".

 

[DigDog]

.

The future IBM only does in service.

ftfy

 

[MisterMac]

The more interesting question is - GF buys IBM Tech and engineers\physicists.

...can they keep them and integrate them?


IBM has been known for research - but not actual engineering production feats.
Something GF also lacks.
(Well lack both, if your being bull).

How would this in actual non-theory even work out for GF\Mubadala?

 

[witeken]

There are rumours that GlobalFoundries will skip 20nm altogether:
https://markets.jpmorgan.com/research/email/prtvmqli/GPS-1343864-0

That makes sense because 20nm only exists as an accident in the first place.
This came out in the TMSC conference call:
"We got committed, and then suddenly, we realized that there was a need for FinFET or a faster FinFET. Since then, one of the guerillas in the business announced that there was a FinFET business. Now if we hadn't committed us already to 20, we might have skipped 20. I think we would have skipped 20. But we were committed already."

20nm is just the logical successor of 28nm. Unfortunately for TSMC and others, implementing important technologies (here: FinFET) 1 node later than Intel (HKMH: 28&20 instead of Intel's 45&32) isn't a good thing because leakage get worse every node.

 

[Phynaz]

I didnt say it is better, i have asked him to provide technical data that TSMCs 28nm is "overall superior".

Sure you did. Anyway there's your evidence.

 

[NTMBK]

Does anybody know how the costs of IBM's processes compare to the foundries? For instance, how much does a transistor on their 22nm SOI process cost compared to a 28nm GloFo wafer? (I honestly have no idea.)

 

[AtenRa]

Does anybody know how the costs of IBM's processes compare to the foundries? For instance, how much does a transistor on their 22nm SOI process cost compared to a 28nm GloFo wafer? (I honestly have no idea.)

From a purely wafer cost, the SOI wafer price is higher.

 

[AtenRa]

Sure you did. Anyway there's your evidence.

where is the evidence ??

 

[krumme]

To grab whatever leftovers TSMC couldn't or didn't want to get. Globalfoundries isn't competition to tsmc, let alome Intel, but for UMC, SMIC and others.

That can never be the purpose - (but reality yes) - because of the enourmous investments.
Huge investments -> bleeding edge

 

[NTMBK]

From a purely wafer cost, the SOI wafer price is higher.

But then you also need to consider the cost of process steps. For instance STMicro were claiming that FD-SOI worked out cheaper than bulk overall, because you saved money on process steps. (I sincerely doubt that IBM's SOI is cheaper though, because IBM.) Just wondering if anyone had seen any ballpark figures or comparisons.

 

[krumme]

Funny interpretation of accident. What you see as an accident I see as a huge miscalculation from TSMC's part.

Perhaps. Lets see when actual fin fets appear - and what the performance is. And lets see what impact 20nm have on the market for the end user.

Take eg qq 810 as an intermediate soc on 20nm to the own successor. How would qq do without 20nm? - even for less than a year?

Interpretation on what is effective technologically is just as much about the market reaction. Lets wait and see. I think the 810 series is a nessesary step for qq until own baked next gen arives because they would otherwise risk loosing to others. You dont want that. It means lack of control and more future risk. You always want to protect your monopoly like situation.

 

[AtenRa]

But then you also need to consider the cost of process steps. For instance STMicro were claiming that FD-SOI worked out cheaper than bulk overall, because you saved money on process steps. (I sincerely doubt that IBM's SOI is cheaper though, because IBM.) Just wondering if anyone had seen any ballpark figures or comparisons.

That STmicro claim was for low volume production, at higher volumes the FD-SOI was starting to get more expensive than bulk.
Comparing GloFos 28nm Bulk to IBMs 22nm SOI from manufacturing only perspective at the low volumes IBM have i will say it may be very close between the two.

 

[krumme]

The more interesting question is - GF buys IBM Tech and engineers\physicists.

...can they keep them and integrate them?


IBM has been known for research - but not actual engineering production feats.
Something GF also lacks.
(Well lack both, if your being bull).

How would this in actual non-theory even work out for GF\Mubadala?

Interesting perspective.

Well if we look at the results you can say they are not the experts on actual engineering productions feat. But man it contradicts all normal asumptions. I think they must be there now competence wise, because what was going wrong was the gate first decision giving problems for years. Even with a imcompetent owner good engineering will prevail and can hardly be destroyed 🙂

I think the problem is rather - what is IBM research bringing to the table?

And as competences is often tied to specific persons. How are the competences actually physically geographically located? - ibm vs gf personel?

 

[mrmt]

Perhaps. Lets see when actual fin fets appear - and what the performance is. And lets see what impact 20nm have on the market for the end user.

Either way there is miscalculation. If 20nm flops then it is obvious that the initial scope of their 20nm R&D effort was not effective enough from the get go, and they couldn't correct it until it was too late to incorporate finfets on the node development. If 20nm is good enough, they still are sacrificing one year to incorporate finfets in their next node and delaying the learning curve of a solution they will need for 10nm.

 

[mrmt]

But then you also need to consider the cost of process steps. For instance STMicro were claiming that FD-SOI worked out cheaper than bulk overall, because you saved money on process steps. (I sincerely doubt that IBM's SOI is cheaper though, because IBM.) Just wondering if anyone had seen any ballpark figures or comparisons.

That is STM/GLF theoretical scenario, where (surprise) SOI comes cheaper.

But what is more probable:

- TSMC and Intel, the players that actually make all the money in this business, have committed a huge mistake and did not adopt the best solution?

- STM/GLF theoretical business case isn't what most customers will find in practice.

 

[NTMBK]

That is STM/GLF theoretical scenario, where (surprise) SOI comes cheaper.

But what is more probable:

- TSMC and Intel, the players that actually make all the money in this business, have committed a huge mistake and did not adopt the best solution?

- STM/GLF theoretical business case isn't what most customers will find in practice.

That seems plausible, but that's just a guess. 🙂 I was hoping there would be some actual analysis that has made estimates of costs for the processes. We can argue in circles all day about the relative merits of FinFETs or SOI, but I for one don't have enough technical knowledge to say for sure. But putting some numbers into the problem would be interesting.

 

[AtenRa]

I will try to find the FD-SOI analysis later on and post back with numbers.

Edit: Hmmm, that was FinFet SOI vs FinFet Bulk.

Here is the link if you want to have a look.
http://www.soiconsortium.org/pdf/Comparison study of FinFETs - SOI versus Bulk.pdf

 

[mrmt]

That seems plausible, but that's just a guess. 🙂 I was hoping there would be some actual analysis that has made estimates of costs for the processes. We can argue in circles all day about the relative merits of FinFETs or SOI, but I for one don't have enough technical knowledge to say for sure. But putting some numbers into the problem would be interesting.

The only ones who are able to answer this question are Intel, TSMC and IBM, because those were the ones actually RESEARCHING nodes and making the design decisions. Anyone else outside these three will provide you a biased/incomplete picture of costs/TTM/yields/etc.

Given that the only two foundries making money on the market are shunning the solution, I'm amazed that SOI is still being discussed at all.