Speculation: Intel will become fabless

[Spartak]

Apple has never delivered a chip that has outperformed x86. I wouldn’t get your hopes up that they will do it now. What they will deliver is reasonable performance with decent battery life. However, this thread isn’t about Apple, it is about Intel, or more specifically, their fabs.

If I were Intel, I would consider opening things up to others. It would require additional investment, but it would help them increase capacity and decrease the dependence on their CPUs for volume and profitability.

I guess there is still someone around here not aware of the incredible gap in IPC and performance/power between Apple and the rest of the industry.

 

[chrisjames61]

I think Apple was always going to switch to ARM. Intel's woes just sped that process way up.

I think so. 68K, PPC to Intel and pretty much back full circle.

 

[A///]

If I had to guess when Apple made the decision, I'd say sometime between 2014 and 2016.

 

[Vattila]

EETimes is in no doubt Intel's future is fabless:

Intel Outside … Just Like all the Others

"“Our 7-nanometer products will be delayed … I’m not pleased with our 7 nanometer-process performance,” Swan added. “We’re going to be pragmatic about if, and when, we should be making stuff inside or outside and making sure that we have optionality to build internally, mix and match, inside and outside or go outside in its entirety.”"

"Similar statements have been issued before by Western semiconductor executives, but the outcome has always been this: eventual wholesale outsourcing of manufacturing. Intel will undoubtedly head in this direction."

https://www.eetimes.com/intel-outside-just-like-all-the-others

 

[biostud]

I thought AMD was dead before they launched Ryzen. If AMD can come back from Bulldozer, so can Intel come back from 14+++, if they take the right decisions. They are still in a much better state than AMD was, but some pride will have to be swallowed if they split production and design into different businesses.

 

[Asterox]

EETimes is in no doubt Intel's future is fabless:

Intel Outside … Just Like all the Others

"“Our 7-nanometer products will be delayed … I’m not pleased with our 7 nanometer-process performance,” Swan added. “We’re going to be pragmatic about if, and when, we should be making stuff inside or outside and making sure that we have optionality to build internally, mix and match, inside and outside or go outside in its entirety.”"

"Similar statements have been issued before by Western semiconductor executives, but the outcome has always been this: eventual wholesale outsourcing of manufacturing. Intel will undoubtedly head in this direction."

https://www.eetimes.com/intel-outside-just-like-all-the-others

There is other solutions, but that would be a big slap to Intel face.

For example, Intel can(for future products) licensed and use TSMC nm nodes or tehnologies.

Globaloundries 14nm(first Ryzen 1000 series), it is nm node licenced from Samsung.

As we now Globalfundries didnt die of shame, so can Intel do the same thing and not die of shame?

 

[soresu]

Samsung's 3nm is probably close to TSMC 5nm than 3nm in terms of density but theoretically, performance might be better with GAA, so I really hope they can get it right though I am not sure if they can deliver it on 2022.

MBCFET Nanosheet device also has lower density for a given pitch than finFET - something that its successor the Forksheet device will fix and achieve both superior density to finFET, plus superior power to the Nanosheet device.

 

[soresu]

I thought AMD was dead before they launched Ryzen. If AMD can come back from Bulldozer, so can Intel come back from 14+++

It's worse for Intel, because unlike AMD who divested most financial ties to their foundry, Intel still has a serious amount of capital tied up in theirs - so every misstep with process nodes has serious ramifications for their bottom line.

Couple that with their process lead essentially having been eroded by TSMC, and AMD nipping at their CPU heels they have more problems than AMD had in 2017 prior to Ryzen's release.

If TSMC has some misfortune of their own it will help Intel, something that does make me wonder exactly how much of the recent US govmt spat with TSMC had to do with China, and how much had to do with a certain company lobbying against them, given that we already know that this current administration has been open to certain industrial parties pushing govmt sponsored actions behind the scenes.

 

[scannall]

My take on it is that Intel will eventually sort out their issues and keep their fabs. They need to do a lot of soul searching on how the company is managed to start with, and they seem to be at least starting that. They will in the long run be successful and competitive. But there's a rough patch they need to get through first. They can do it, just won't be fun.

But they will never again have the massive lead on everything that they used to. Overall though, I think that's a great thing for the company, and for us consumers.

 

[soresu]

I think Apple was always going to switch to ARM. Intel's woes just sped that process way up.

Not always - this began with Intel snubbing Apple on developing a low power x86 chip for iPhone, something they backtracked on far too late in developing Atom to compete in mobile markets.

This in turn led to Apple's successful iPhone/Pad line being ARM based, and their subsequent investment in a high performing custom ARM core (by buying PA Semi to design it) to enable further differentiation from the Android based competitors on the market.

In short, hubris kicked Intel in the teeth twice - first in snubbing the Apple iPhone contract to begin with, and then overreaching on 10nm node complexity, which subsequently amplified their problems moving to new cores which were designed to run on their future 10nm node.

Intel snubbing them actually did Apple a favour as ARM gives them far more room for custom, and likely any terms for an x86 license would have been significantly more costly to them than an ARM ISA license.

 

[Markfw]

Living so close to the new Intel fab, and watching it grow everyday, this place is HUGE and must be in the billions in cost. No way they are throwing this away. There is constant construction, literally square miles of buildings. Their parking lot is bigger than Disneyland !

 

[soresu]

As we now Globalfundries didnt die of shame, so can Intel do the same thing and not die of shame?

Define die.

Seriously, their current market position is less than nothing in the state of the art sector of the market.

Of course they still have plenty of older nodes likely getting orders from all corners, so they won't go out of business, but I don't see them getting back into the state of the art sector any time soon - unless through a more lateral route, like a new material that no one is willing to invest in.

 

[soresu]

Living so close to the new Intel fab, and watching it grow everyday, this place is HUGE and must be in the billions in cost. No way they are throwing this away. There is constant construction, literally square miles of buildings. Their parking lot is bigger than Disneyland !

I'd be careful living so close to any industry monster, they aren't known for being especially careful of their residential neighbours safety regs wise historically speaking.

 

[Markfw]

I'd be careful living so close to any industry monster, they aren't known for being especially careful of their residential neighbours safety regs wise historically speaking.

Its actually about 5 miles away.... Nothing but industrial or business anywhere close.

 

[DrMrLordX]

For example, Intel can(for future products) licensed and use TSMC nm nodes or tehnologies.

While I agree, do remember that GF licensed Samsung's node essentially to allow GF to serve as a second-source for a production contract. If Intel goes to TSMC, hat-in-hand, asking to license their node tech for internal adaptation (Intel won't just copy/paste) so they can convert all their 14nm fabs to something useful for the future:

a). it'll take a ton of work
b). it'll take a lot of money and other bribes to get TSMC to play ball

Best-case scenario, Intel builds fabs for TSMC state-side to help them expand capacity, and Intel produces some hybrid nodes for their own products to get back on-track.

 

[maddie]

It's worse for Intel, because unlike AMD who divested most financial ties to their foundry, Intel still has a serious amount of capital tied up in theirs - so every misstep with process nodes has serious ramifications for their bottom line.

Couple that with their process lead essentially having been eroded by TSMC, and AMD nipping at their CPU heels they have more problems than AMD had in 2017 prior to Ryzen's release.

If TSMC has some misfortune of their own it will help Intel, something that does make me wonder exactly how much of the recent US govmt spat with TSMC had to do with China, and how much had to do with a certain company lobbying against them, given that we already know that this current administration has been open to certain industrial parties pushing govmt sponsored actions behind the scenes.

Was wondering about how they can begin to divest the fabs. Sell them? I don't think a Mubadala like buyer will be available. Split the company into 2 and assign shares to present stockholders? The good thing is that they have plenty of cash reserves, so some flexibility.

 

[Doug S]

Living so close to the new Intel fab, and watching it grow everyday, this place is HUGE and must be in the billions in cost. No way they are throwing this away. There is constant construction, literally square miles of buildings. Their parking lot is bigger than Disneyland !

If they decided to go fabless (something I'm very much not convinced of, I think it is too early to throw in the towel) they wouldn't just abandon it and let the tumbleweeds take over. They'd sell it, and their other fabs that have new enough technology, with TSMC and Samsung the most likely buyers.

You can't just quit making your own chips and start buying someone else's, as many have pointed out TSMC doesn't have the capacity to absorb Intel's business. But if Intel turns over their fabs, with Intel staff operating them as an Intel fab until TSMC was ready one by one to upgrade them to do TSMC technology and take over operations. Presumably most of the employees would stay on, it would just be the upper management that TSMC would need to replace.

Samsung might be a more likely partner, just because TSMC is already so big and unlike TSMC Samsung has a lot of excess capacity currently. They are having trouble keeping up with TSMC, but they wouldn't if they had big customer like Intel paying the bills to develop new processes. Just make sure they check resumes carefully so they don't accidentally hire any former Intel managers!

 

[Ajay]

If they decided to go fabless (something I'm very much not convinced of, I think it is too early to throw in the towel) they wouldn't just abandon it and let the tumbleweeds take over. They'd sell it, and their other fabs that have new enough technology, with TSMC and Samsung the most likely buyers.

You can't just quit making your own chips and start buying someone else's, as many have pointed out TSMC doesn't have the capacity to absorb Intel's business. But if Intel turns over their fabs, with Intel staff operating them as an Intel fab until TSMC was ready one by one to upgrade them to do TSMC technology and take over operations. Presumably most of the employees would stay on, it would just be the upper management that TSMC would need to replace.

Samsung might be a more likely partner, just because TSMC is already so big and unlike TSMC Samsung has a lot of excess capacity currently. They are having trouble keeping up with TSMC, but they wouldn't if they had big customer like Intel paying the bills to develop new processes. Just make sure they check resumes carefully so they don't accidentally hire any former Intel managers!

Yeah, I don't see Intel 'selling' it fabs. Seems like a corporate partnership would be best. I'm sure the Feds (and states w/Fabs) will bend over backwards to make this happen if necessary. Pooling R&D with Samsung would be a win-win, if Samsung got all Intel's ASIC business (GPUs, etc). TSMC is a head in process development, so Intel would really have to pay the piper to get involved in any process-node sharing. It's going to be an interesting ~5 years while Intel figures out what to do (unless 7nm ultimately outperforms current internal expectations). I think Intel has shown it can manage being 1 node behind - 2 nodes behind is too much though.

 

[Vattila]

There is other solutions

Bolaji Ojo, writing for EETimes, has another opinion piece on the Intel fab crisis:

Intel is a Potentially Great Foundry

"Intel Corp. dropped the ball. It should today be the world’s biggest semiconductor foundry; not TSMC. This can still change but it will require a radically different vision and strategy completely at odds with what CEO Bob Swan outlined last week. The board of directors will also need to intervene."

https://www.eetimes.com/intel-is-a-potentially-great-foundry

As previously stated, I think a joint venture with GlobalFoundries makes sense. They operate in the USA with facilities and thousands of employees in upstate New York, and they need a future process roadmap beyond 12LP FinFET and 12FDX FD-SOI. Former GlobalFoundries CTO Gary Patton went to Intel last year, taking a key role as the Corporate Vice President and General Manager of Design Enablement.

 

[Doug S]

They burned a lot of bridges in their failed attempt to become a foundry a few years ago. It would be very difficult to get someone to trust them now, especially since they are not only not the process leader they were when they tried the first time but are behind. There's also the problem that Intel's fabs are pretty highly utilized, because they didn't plan to still be making 14nm chips in 2020. Heck, five years ago they thought they'd be phasing out 10nm production around now.

 

[DrMrLordX]

As previously stated, I think a joint venture with GlobalFoundries makes sense. They operate in the USA with facilities and thousands of employees in upstate New York, and they need a future process roadmap beyond 12LP FinFET and 12FDX FD-SOI. Former GlobalFoundries CTO Gary Patton went to Intel last year, taking a key role as the Corporate Vice President and General Manager of Design Enablement.

GF doesn't have the IP to get Intel back on track.

 

[beginner99]

1. Apple moving to ARM means more professional software will support ARM, which allows Windows ARM to grow.
2. Apple successfully moving to ARM means PC makers will have to consider ARM options in order to compete.

Like Doug I agree with 1. but not with 2. But 2. isn't that important as 1. because the main success of x86 was because client-hardware was x86 meaning software was being produced on x86 for x86. And it's hard to build a server infrastructure if you don't have any client devices to build software for said servers. This will change with Apple going ARM.

However the main difference to x86 and Intel is that Apple is more and more a closed platform. Software made on Macs will on be for other Apple devices hence not really affecting anyone outside the apple platform at all meaning it doesn't matter what ISA they use.

Yep. Source: I'm a web app developer. Chrome OS has demonstrated that most people can use mostly web apps fine.

Which I find a very scary prospect if the future of client devices are dumbed-down devices with limited functionality. If this becomes mainstream it means more complex machines will slowly cease to exists and get very expensive. Eg. the good-enough aspect of performance is a scary thing.

 

[teejee]

I think it is almost impossible for Intel to succed as foundry. They need to create a seperate independent foundry company that takes all the plants and production R&D to get the trust needed.
Intel is a competitor with too many of the normal foundry customers.

 

[Doug S]

I think it is almost impossible for Intel to succed as foundry. They need to create a seperate independent foundry company that takes all the plants and production R&D to get the trust needed.
Intel is a competitor with too many of the normal foundry customers.

I think this is probably why they only made a 'half hearted' attempt at becoming a foundry. The x86 side would be putting too many restrictions on the customers that the foundry business could seek - they wouldn't want anyone fabbing CPUs on Intel's then-superior process because that would take away one of the advantages (the other being x86 as the defacto requirement for Windows) that has been responsible for Intel's success in the CPU market.

If you narrow your market down to custom ASICs, FPGA, network switch chips and the like and avoid anything that has computation as its primary role (thus no GPUs either) then you really limit your potential customer base - and a lot of that potential base cares more about cost than they care about top performance. i.e. a chip in an ethernet, fibre channel etc. device gains no benefit from going faster than line speed so a leading edge node may be a waste of money for it.

Now that they are behind TSMC it is too late for them to become a foundry, while foundries behind the leading edge can obviously succeed they have to compete on price or have some specialty (like GF and their fully depleted SOI which is ideal for RF/analog, or SMIC which has the home field advantage for Chinese customers)

 

[Spartak]

GF doesn't have the IP to get Intel back on track.

But GF has the knowledge how to operate as a foundry and this I think could be very valuable to Intel.

If 7nm gets any further delay Intel will be forced to fab their first gen 7nm CPUs elsewhere (TSMC or Samsung). This, I think will ultimately force a spin-off of their fabs as you either want to fab inhouse or outsource. Doing both is a recipe for losses and persisting political and technical conflict between your design and manufacturing departments. GF isn't needed for it's cutting edge node technology but for it's foundry expertise and experience how to transition from an in-house to an independent fab. Running decent FD-SOI based nodes could further diversify a combined Intel/GF foundry as this is where a large chunk of the market is where Intel has no presence whatsoever.

I think they might be a perfect match.