Speculation: Intel will become fabless

[moinmoin]

Alder Lake S is also expected very late this year into early next year.

I see...

Rumors are Sapphire Rapids will only be for the high end and that Ice Lake Server is going to stay on the market for awhile.

So just covering the top die, with the rest of the line up just being refreshes of the previous gen?

Years ago Sapphire Rapids was introduced to me as the new architecture that would finally overhaul the whole Core design.

Already having low expectations I'm surprised at myself still ending up being disappointed of Intel. This is not the competition I want!

 

[jpiniero]

So just covering the top die, with the rest of the line up just being refreshes of the previous gen?

Based upon the leaked picture of it, it might be that way because it's like Zen 1 Epyc where you need all 4 chiplets. Maybe you will also get an HEDT with 2 chiplets. The chiplets are huge too so it's a lot of silicon.

The "revolutionary" core (Ocean Cove) was cancelled a long time ago. Never got leaked as to why although one theory I have is that Smeltdown made it impractical.

 

[DrMrLordX]

Rumors are Sapphire Rapids will only be for the high end and that Ice Lake Server is going to stay on the market for awhile.

That doesn't make any sense. Intel is barely shipping Ice Lake-SP at this point. Unless they're going to seriously ramp up production of 10nm+ or port Ice Lake-SP to 10SF, something else has got to give.

 

[jpiniero]

port Ice Lake-SP to 10SF

I was thinking that they might do that.

 

[LightningZ71]

Why stop at SF? Why not continue to 10esf?

 

[moinmoin]

Why stop at SF? Why not continue to 10esf?

I propose going with 10SESF right away.

 

[DrMrLordX]

Why stop at SF? Why not continue to 10esf?

Intel hasn't shipped anything using 10SFE yet so . . . there's a lot we don't know. All that we know at this point is that Intel has struggled to bring any meaningful number of Ice Lake-SP CPUs to market on 10nm+. If they need to up yields and get product out the door to cover for delays to Sapphire Rapids, then a "proven" node which has produced millions of 4c Tiger Lake-U dice might be the safest bet they have. We also don't know how freely Intel can move fab lines from 10nm+ to 10SF to 10SFE, or if it can be done at all without major retooling. Using 10SFE might cut into Alder Lake production (and vice versa).

 

[IntelUser2000]

Why stop at SF? Why not continue to 10esf?

Actually, that is what they are doing. Sapphire Rapids will replace all.

 

[Vattila]

Hah!

WSJ News Exclusive | Intel Is in Talks to Buy GlobalFoundries for About $30 Billion

A deal for chip maker GlobalFoundries would be Intel’s largest acquisition. It comes as Intel is launching a major push to become a chip manufacturer for other companies.

www.wsj.com

Intel in talks to buy GloFo | AnandTech Forums: Technology, Hardware, Software, and Deals

It will be very interesting to see if anything materialises out of this rumour. As I expressed in my OP, I think a merger or some kind of joint venture makes sense. GlobalFoundries needs a competitive process roadmap — which Intel can, or at least will attempt to, provide —while GlobalFoundries is an established foundry with a solid share in mainstream and speciality nodes (various SOI processes, in particular). And both are based in USA with a lot of manufacturing history at long established sites around the country.

 

[NTMBK]

Hah!

WSJ News Exclusive | Intel Is in Talks to Buy GlobalFoundries for About $30 Billion

A deal for chip maker GlobalFoundries would be Intel’s largest acquisition. It comes as Intel is launching a major push to become a chip manufacturer for other companies.

www.wsj.com

Intel in talks to buy GloFo | AnandTech Forums: Technology, Hardware, Software, and Deals

It will be very interesting to see if anything materialises out of this rumour. As I expressed in my OP, I think a merger or some kind of joint venture makes sense. GlobalFoundries needs a competitive process roadmap — which Intel can, or at least will attempt to, provide —while GlobalFoundries is an established foundry with a solid share in mainstream and speciality nodes (various SOI processes, in particular). And both are based in USA with a lot of manufacturing history at long established sites around the country.

Putting all the eggs of American chip manufacturing in one basket makes me nervous... But I guess they need to compete with the Taiwanese and Korean giants.

GlobalFloundering can at least bring along experience of working with specialist nodes, and of genuinely being a third party foundry. But I still think Intel Foundry won't succeed until they split off from Intel Design.

 

[DisEnchantment]

Putting all the eggs of American chip manufacturing in one basket makes me nervous... But I guess they need to compete with the Taiwanese and Korean giants.

GlobalFloundering can at least bring along experience of working with specialist nodes, and of genuinely being a third party foundry. But I still think Intel Foundry won't succeed until they split off from Intel Design.

Two stellar Fabs with very competent leadership, excellent corporate culture and a history of consistent execution coming together, what could go wrong. /s

 

[coercitiv]

Putting all the eggs of American chip manufacturing in one basket makes me nervous... But I guess they need to compete with the Taiwanese and Korean giants.

It's a genius move, they get a foundry for free as long as Intel is the only American manufacturer able to receive those $50+ Billion in subsidies. It's almost like the US likes to finance monopolies with taxpayer money. /s

 

[A///]

Who's the Korean giant other than Samsung which is reportedly flailing. SK Hynix?

 

[NTMBK]

Who's the Korean giant other than Samsung which is reportedly flailing. SK Hynix?

Samsung is who I was referring to. They're struggling, but they've done a lot better than GloFo did.

 

[moinmoin]

Samsung is who I was referring to. They're struggling, but they've done a lot better than GloFo did.

And even when not being at the cutting edge of the process tech Samsung Semiconductor is in a way better position to persevere in the process race than both GloFo and Intel since it can rely on Samsung's huge in-house memory and smartphone business in any case. On the other hand GloFo has nothing and Intel is losing market share fast.

 

[A///]

I've not heard good things about Samsung's DDR5 production. Quite the opposite actually. Not sure how verifiable it is given that not much in the way of DDR5 has hit the markets yet.

 

[aigomorla]

I think this thread can be laid to rest since Intel seems to be doing reverse..... buying more FABS

 

[Doug S]

I think this thread can be laid to rest since Intel seems to be doing reverse..... buying more FABS

Intel buying GF might make it MORE likely they become fabless - the same way AMD did by spinning off their fabs into a separate company. Buying a foundry might be necessary for Intel to be able to do so, as their fabs have very little experience working with companies other than Intel, and only a few recent nodes rather than the wide selection of nodes that proper foundries have.

 

[Failnaught]

Intel buying GF might make it MORE likely they become fabless -

Yes, completely agree! But it means the new Intel manufacturing company will be in the game for leading edge nodes (in my opinion), even though they will probably need to become independent of Intel chip design, because otherwise nobody would trust them as a neutral partner.

Of course it's debatable whether the new Intel fab company will pursue leading edge nodes. I think they must, if they want their fab to grow. All of the highest margin parts are done at the bleeding edge. The old nodes will just pick up the scraps.

 

[Vattila]

I think this thread can be laid to rest since Intel seems to be doing reverse..... buying more FABS

As a consequence of their struggles with manufacture, since the creation of this discussion thread back in 2018, Intel has undergone an incredible restructuring already, with major changes in management, organisation and design methodology. I think it will be interesting to follow Intel's fabrication journey in the time ahead — and this discussion thread should be useful in that regard.

Right now it looks like they are adopting Samsung's model, with Intel Foundry Services gradually taking over all the manufacturing operations, as a fully independent subsidiary, akin to Samsung Foundry.

 

[Doug S]

Yes, completely agree! But it means the new Intel manufacturing company will be in the game for leading edge nodes (in my opinion), even though they will probably need to become independent of Intel chip design, because otherwise nobody would trust them as a neutral partner.

Of course it's debatable whether the new Intel fab company will pursue leading edge nodes. I think they must, if they want their fab to grow. All of the highest margin parts are done at the bleeding edge. The old nodes will just pick up the scraps.

TSMC makes half their revenue on 16/20 nm and older nodes. While they charge less for those wafers, everything is pretty much fully depreciated at that point so the margins are huge. Definitely not "scraps".

Intel didn't keep the old nodes around because they didn't have anything to make with them. Perhaps they would have been smarter to have a foundry subsidiary they spun off older nodes to. They would have generated additional revenue and already have the foundry business figured out, so they wouldn't have to spend $30 billion to get there. I imagine if anyone had ever suggested such a plan, their beancounters vetoed it because it would reduce that juicy gross margin Wall Street is in love with.

 

[Mopetar]

TSMC makes half their revenue on 16/20 nm and older nodes. While they charge less for those wafers, everything is pretty much fully depreciated at that point so the margins are huge. Definitely not "scraps".

Revenue has nothing to do with costs. I think the reality is that with these newer nodes there just isn't the same amount of capacity as with the older nodes. I think people tend to forget there are a huge amount of chips that aren't CPUs that go into computers, phones, and other electronics. Most of those don't need to be on the bleeding edge node or even the second best node.

 

[LightningZ71]

Maybe not the bleeding edge, or even the current node, but many of them can't actually be more than two or three full node generations behind. They ALL consume power, and trailing nodes aren't as power efficient. That power produces heat, which must be dealt with, and the power adds to the expense of larger power supplies. They also yield far fewer chips per wafer, so, even if the equipment is long paid for, the actual raw materials cost per die isn't as good if you go too far back.

 

[MasterofZen]

My claim is simply: on relevant, leading edge performance nodes (which, as we see in benchmarks, Intel's 14nm node is one), TSMC has substantially less capacity than Intel. TSMC has N7, it's various refinements on the same lines, and N5, in only a couple of fabs. Intel has 14nm, and 10nm in multiple variants, in nearly a dozen high volume fabs.

Does TSMC have more total, all-in capacity than Intel? Yes, I believe so. But I'm specifically focusing on leading edge processes. This excludes TSMC'S large capacity in their less than competitive 10nm node and larger. This is where AMD and Intel compete. AMD is hopelessly capacity constrained due to their situation. They can't push for more capacity without tanking profits because it results in a bidding war against MUCH deeper pockets.

Your claim is totally wrong! Even in leading edge, TSMC have much more capacity than Intel.
According to a research by Mizuho securities. As of 2020, the capacity of TSMC and Intel's advanced node are as follows
Advanced node capacity(12''wpm)
16/14nm 7nm/10nm 5nm/7nm
TSMC 120k 136k 80k
Intel 70k 35k 0k

Be Careful when you make a bold claim.

 

[LightningZ71]

I conceed to your more recent numbers. The charts that I was looking at ended 1q 2020 and showed different numbers than yours for 2019.