News Intel to fab chips for Qualcomm

[NTMBK]

AnandTech Forums: Technology, Hardware, Software, and Deals

Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

www.anandtech.com

Starting at the 20A process in a few years. I wonder what parts they will build at Intel? If it's the laptop chips meant to compete with Intel, that's going to be weird.

 

[Nothingness]

What an interesting news! A friend of mine working at Intel was telling me about 10 years ago that Intel considered Qualcomm as their main competitor. How things change

 

[DrMrLordX]

That's weird. What kind of volume does Qualcomm expect out of what is probably the IBM 2nm node?

 

[moinmoin]

So depending on when exactly these 20A Qualcomm chips will be actually available nearly 20 years will have gone from Intel rejecting to build Apple's phone chips Intel to building phone chips for a competitor.

 

[jpiniero]

That's weird. What kind of volume does Qualcomm expect out of what is probably the IBM 2nm node?

I'll say the modems.

You know what would be really interesting, if Samsung ends up licensing IBM's node too.

 

[DrMrLordX]

I'll say the modems.

Hmm! Interesting thought. You may be right. Intel did once have a modem business of their own, after all.

You know what would be really interesting, if Samsung ends up licensing IBM's node too.

Yyyyup. And Globalfoundries! Hyuck.

 

[Doug S]

This Qualcomm announcement smacks as a pay for play deal to help Intel market its foundry. They know their foundry service has a terrible reputation that has severely impacted some companies that stupidly believed their roadmap, so having a big name customer announce "we're going to Intel!" is just what the marketers to want see.

If I had to bet, I'd say Intel is getting that in exchange for funding Qualcomm's full cost of taping out one or more designs in Intel's process, and an option on a guaranteed wafer allocation.

That makes it a no-risk proposition for Qualcomm to say they're using Intel, while they continue with their former plans of designing for TSMC and/or Samsung. If Intel's promises come to fruition, great, if not no harm done to Qualcomm and Intel will have leveraged Qualcomm's blessing to secure other suckers customers for their foundry.

I mean, this is a marketing organization that paid OEMs to use Intel SoCs in mobile devices - not just a discount, they actually got the SoCs for free and got money on top of it. Doing something like this would be small potatoes, they might be offering Qualcomm free or discount wafers as well!

 

[Kenmitch]

Starting at the 20A process in a few years.

These years are Qualcomm or Intel years? Intel years tend to stretch a lot at times.

 

[JasonLD]

That's weird. What kind of volume does Qualcomm expect out of what is probably the IBM 2nm node?

Is Intel 20A based on IBM 2nm? IBM 2nm had 3 stacks GAA, while Intel's version looks like 4 stacks.

 

[RTX]

When are they producing for Amazon?

 

[DrMrLordX]

When are they producing for Amazon?

The ad specifies packaging capabilities. So. They may be utilizing one of Intel's packaging/test facilities for future silicon.

Is Intel 20A based on IBM 2nm? IBM 2nm had 3 stacks GAA, while Intel's version looks like 4 stacks.

Hmm, good catch. Might be a derivative process.

 

[Thibsie]

I mean, this is a marketing organization that paid OEMs to use Intel SoCs in mobile devices - not just a discount, they actually got the SoCs for free and got money on top of it. Doing something like this would be small potatoes, they might be offering Qualcomm free or discount wafers as well!

And I still don't get how Intel could get away with this ...

 

[NTMBK]

And I still don't get how Intel could get away with this ...

"Contra-revenue". It's well documented, look it up. Their mobile chip division was basically financial antimatter. They didn't just have running costs higher than their revenue, they had negative revenue.

 

[coercitiv]

Just as @Doug S described, this new deal between Qualcomm and Intel looks a lot like a (very expensive) publicity stunt. Not only does Intel benefit from contracting a big name customer, but they're also showing the US government (and others) they can play nice with the competition. What better proof to offer than bringing one of their biggest competitors onboard to demonstrate their commitment of becoming a world leader in semiconductor manufacturing? Please insert subsidies in 3, 2, 1...

"Contra-revenue". It's well documented, look it up. Their mobile chip division was basically financial antimatter. They didn't just have running costs higher than their revenue, they had negative revenue.

For anyone looking for a decent write-up of what went wrong, here's one postmortem worth the read:
How Intel Lost the Mobile Market, Part 2: The Rise and Neglect of Atom

There's more to the story though, as the original article from 2016 was updated at the end of 2020 with some very interesting hindsight:

Update (11/25/2020): The article below may have been written in 2016, but it still stands up as a postmortem of what went wrong with Intel’s mobile efforts — with one very important omission. Back in 2016, we didn’t know Qualcomm had been ruthlessly enforcing licensing and purchasing terms that made it effectively impossible for manufacturers to offer Intel-based mobile devices. I remember wondering why Intel couldn’t find a single US company to produce a phone around its hardware platform for love or money when the original Xolo X900 compared well enough against a then-current iPhone.

Intel still made a number of mistakes with Atom, as this article discusses, but the fact that Qualcomm had a stranglehold on the market behind the scenes obviously had an impact on what kind of success Intel was ever going to achieve.
I genuinely liked the Xolo X900 device I tested all those years ago, and the Bay Trail tablets I had circa 2013 were great devices. Atom’s mobile efforts will always remain an enticing might-have-been. The US 3rd Circuit Court of Appeals has since thrown out the antitrust findings against Qualcomm. It should be noted that the FTC disagreed with that decision. Intel continues to manufacture 5G base stations, but the scope of its 5G business has shrunk significantly after selling its 5G modem to Apple.

 

[A///]

Yes, agree with @Doug S's post, but on the other hand, should Intel commit and pull their promises off, it would be incredible to have another player on the market. I really do hope IFS works out since Samsung is obviously a poor choice for the time being and TSMC will be forced to rethink their ever increasing prices despite node jumps. Though I do wonder if this is Qualcomm also hedging their bets?

 

[bullzz]

This Qualcomm announcement smacks as a pay for play deal to help Intel market its foundry. They know their foundry service has a terrible reputation that has severely impacted some companies that stupidly believed their roadmap, so having a big name customer announce "we're going to Intel!" is just what the marketers to want see.

If I had to bet, I'd say Intel is getting that in exchange for funding Qualcomm's full cost of taping out one or more designs in Intel's process, and an option on a guaranteed wafer allocation.

That makes it a no-risk proposition for Qualcomm to say they're using Intel, while they continue with their former plans of designing for TSMC and/or Samsung. If Intel's promises come to fruition, great, if not no harm done to Qualcomm and Intel will have leveraged Qualcomm's blessing to secure other suckers customers for their foundry.

I mean, this is a marketing organization that paid OEMs to use Intel SoCs in mobile devices - not just a discount, they actually got the SoCs for free and got money on top of it. Doing something like this would be small potatoes, they might be offering Qualcomm free or discount wafers as well!

Or.. Samsung has consistently fallen behind TSMC and the only reason qualcomm is sticking with them is to be in their Galaxy phones. and TSMC is sucking blood out of everyone. So QCOM wants a third viable player to stick it to others. Not everything has to be a conspiracy. Even JHH wants Intel to be a foundry option according to some reporting.

I do agree that this is a no risk option for QCOM. They can always use this as a negotiating tactic. That's capitalism for ya.

 

[Roland00Address]

Is Intel 20A based on IBM 2nm? IBM 2nm had 3 stacks GAA, while Intel's version looks like 4 stacks.

We do not know, it is likely a mixture of Intel and IBM tech, and possibly other research IP, patents, etc.

We will not know till later (and hard even then) due to trade secrets and you will not want to disclose this information to competitors until the final product is implemented.

 

[Doug S]

"Contra-revenue". It's well documented, look it up. Their mobile chip division was basically financial antimatter. They didn't just have running costs higher than their revenue, they had negative revenue.

"financial antimatter"

Love the phrase, I would remember it to use in the future but I'm only aware of the one example so I'll probably never run across it again...

 

[beginner99]

For anyone looking for a decent write-up of what went wrong, here's one postmortem worth the read:
How Intel Lost the Mobile Market, Part 2: The Rise and Neglect of Atom

This quote from the 2016 article is hilarious:

Intel’s 14nm problems delayed its next-generation tablet processors from 2014 to 2015. Its 10nm node, once expected to secure enormous economies of scale over TSMC, has been pushed to 2017 as well.

lol. pushed to 2017. Here we are mid 2021 and large part of the chips is still on 14nm. Just a reminder how far ahead intel was and how much they blundered.

EDIT: Albeit now with hindsight 20/20 and intel still having all fabs full on an outdated process their decision might unintentionally not have been that bad. Every smartphone and IoT device ultimately relies on a server. The more smartphones, the more servers are needed and hence more high margin server cpu sales for intel.

 

[coercitiv]

lol. pushed to 2017. Here we are mid 2021 and large part of the chips is still on 14nm. Just a reminder how far ahead intel was and how much they blundered.

Yeah, there's plenty of these things around. Just yesterday I posted one from March 2020:

In the 7 nm production at the end of 2021, Intel wants to reach the level of parity with the competition and be significantly stronger than with 10 nm

Three months later they announced that 7nm would be delayed.

 

[moinmoin]

Unnecessarily publicly fueling these unrealistic expectations over a long period likely created immense pressure on the involved engineers. Can only image many of the engineers that weren't laid off wanting to get out due to such.

While Gelsinger does have clearer messaging now, it's still plenty ambitious, so I can only hope for Intel that the internal structure to support these efforts is there already.

 

[Roland00Address]

This quote from the 2016 article is hilarious:


lol. pushed to 2017. Here we are mid 2021 and large part of the chips is still on 14nm. Just a reminder how far ahead intel was and how much they blundered.

EDIT: Albeit now with hindsight 20/20 and intel still having all fabs full on an outdated process their decision might unintentionally not have been that bad. Every smartphone and IoT device ultimately relies on a server. The more smartphones, the more servers are needed and hence more high margin server cpu sales for intel.

The time travel mistake was thinking you can do density of Intel 10nm or TSMC 7nm (roughly 90 to 100 MTr/mm2 ) without EUV. TSMC bought the EUV machines in 2017, as much as ASML was able to make and supply (for there were multiple customers.) Intel was confident they would not need EUV for that milestone and the next milestone.

And Intel 7nm which is being renamed to Intel 4 is Intel accepting they actually do need EUV. The reason why Intel thought they could not do EUV was ASML was not shipping EUV machines till 2017 and they wanted to push forward and it bit them in the "regrets." Intel did actually make a working 90 MTr/mm2 without EUV it took over 4+ years to do so, and during that time their competitors were behind them is now several years ahead. Regrets / Hubris it is all the same thing, just different place in the timeline.

 

[beginner99]

Intel did actually make a working 90 MTr/mm2 without EUV

With quadruple patterning and all it's great downsides (poor yield). Right? That os what you are hinting at?

But then all their actual chips usually have half or less of that density...

 

[Roland00Address]

And I still don't get how Intel could get away with this ...

They treated it like "marketing money" and so on, certain businesses inside of a big business are allowed to be loss leaders in order to forge OEM relationships with phone makers and so on. This is very similar to the 90s and 00s Intel with "rebates" (similar not quite same), and it is not just hardware makers, drug makers do similar type of thing with rebates with hospitals and doctors and so on.

With quadruple patterning and all it's great downsides (poor yield). Right? That os what you are hinting at?

But then all their actual chips usually have half or less of that density...

Yep. It is more complicated than that, more complicated than anyone can put in 200 words, but yes.

There were trade offs with not using EUV and it "worked" but is it worth the tradeoffs. I say no, but this is hindsight speaking.

 

[Doug S]

And Intel 7nm which is being renamed to Intel 4 is Intel accepting they actually do need EUV. The reason why Intel thought they could not do EUV was ASML was not shipping EUV machines till 2017 and they wanted to push forward and it bit them in the "regrets." Intel did actually make a working 90 MTr/mm2 without EUV it took over 4+ years to do so, and during that time their competitors were behind them is now several years ahead. Regrets / Hubris it is all the same thing, just different place in the timeline.

Intel always planned their 7nm was going to be made with EUV. Do you have a source showing otherwise?