News Report claims that Intel will build Core i3s at TSMC

[NTMBK]

Reporting from Tom's Hardware here, and the actual report from Trendforce here.

Intel has outsourced the production of about 15-20% of its non-CPU chips, with most of the wafer starts for these products assigned to TSMC and UMC, according to TrendForce’s latest investigations. While the company is planning to kick off mass production of Core i3 CPUs at TSMC’s 5nm node in 2H21, Intel’s mid-range and high-end CPUs are projected to enter mass production using TSMC’s 3nm node in 2H22.

...

TrendForce believes that increased outsourcing of its product lines will allow Intel to not only continue its existence as a major IDM, but also maintain in-house production lines for chips with high margins, while more effectively spending CAPEX on advanced R&D. In addition, TSMC offers a diverse range of solutions that Intel can use during product development (e.g., chiplets, CoWoS, InFO, and SoIC). All in all, Intel will be more flexible in its planning and have access to various value-added opportunities by employing TSMC’s production lines. At the same time, Intel now has a chance to be on the same level as AMD with respect to manufacturing CPUs with advanced process technologies.

I have no idea how credible Trendforce are, or what their track record is, but they seem pretty confident. Sounds like Intel might be pivoting their internal fabs to be "IBM-like"- lower volume, higher margin products, tuned to absolute max performance, and don't need to yield high enough to manufacture cheap PC chips.

 

[jpiniero]

Clock speed is actually an argument for doing i3 at TSMC... would be a lot less pressure to hit high frequencies.

 

[mikk]

None. Just that Intel is the one having to prove being able to deliver on time.


With Intel's tight coupling of process and silicon design one can't tell either way. We don't even know the yield, only that Intel so far has been incapable of launching dies beyond a specific die size and number of cores, thus keeping them all in the mobile market.

Just a reminder to you, not long ago people like you didn't believe Intel could fix the awful 10nm clock speeds of CNL and Icelake and many didn't believe in 10nm for desktop either. You are trying to downplay the process improvements which speaks for itself. Tigerlake-U is growing really fast in the market which is a good sign. Furthermore Icelake-SP is based on the awful 10+, it will always struggle because of that. It's hard to believe for you Intel is really improving their 10nm.

 

[Hulk]

Do we know the transistor density of 10SF vs TMSC's 7nm that is fabbing Zen 3? With all of the different process structure measurement it' seems the best way to gauge process size would be transistor density.

 

[LikeLinus]

Couple of question. I've seen many people saying that they didn't expect or understand the i3 being built by TSMC. Question is, what is the actual percentage of i3 sales across all markets vs i5 or i7? Volume.

Also, I saw people complain about 4 core. What are corporate, enterprise (for employees) and low end customers purchasing? I know everyone here wants 16c/32T (I have a 5950X and understand), but realistically, what does the market demand and what is the biggest seller? High core/thread CPUs or 4C CPUs that are more than enough to run Office or a web browser? Thats a huge market segment.

AMD (or TSMC) currently can't produce enough CPUs, that much has been proven through the 5000 Series, Xbox and PS5. But they probably can easily produce non-high performance lower cores easily and it would offset the demand of Intel and allow them to focus on high-core and high-performance products.

Just a though....

 

[LikeLinus]

Maybe, why AMD didn't snatch that up is anyone's guess.

I think most 3rd party manufactures would not want to sell all their new capacity to one company. Diversity is in the best interest of any corporation. Additionally, you're assuming that AMD can afford that. While they are on a rise and producing great products, that doesn't mean they have the cash reserves of some of the companies they also produce for. AMD doesn't want to spend all of their cash reserve not knowing what could happen in the next year or two. Believe me, I've seen it happen with companies during the market fallout in 2001, 2008/9, and 2020 You have to be fiscally responsible. Anyone who disagrees with this, simply hasn't worked for a Fortune 50 company during a down-turn.

 

[Hulk]

Couple of question. I've seen many people saying that they didn't expect or understand the i3 being built by TSMC. Question is, what is the actual percentage of i3 sales across all markets vs i5 or i7? Volume.

Also, I saw people complain about 4 core. What are corporate, enterprise (for employees) and low end customers purchasing? I know everyone here wants 16c/32T (I have a 5950X and understand), but realistically, what does the market demand and what is the biggest seller? High core/thread CPUs or 4C CPUs that are more than enough to run Office or a web browser? Thats a huge market segment.

AMD (or TSMC) currently can't produce enough CPUs, that much has been proven through the 5000 Series, Xbox and PS5. But they probably can easily produce non-high performance lower cores easily and it would offset the demand of Intel and allow them to focus on high-core and high-performance products.

Just a though....

I am sitting here typing this on my main rig powered by a 4770k from 2013. I don't game on this, true. But I do edit video, record/edit multitrack audio, work with Photoshop, etc.. Honestly with a properly set up computer like mine even a lowly 4770k can churn through many workloads quite effectively.

Computer processors for the masses may have already reached the point where any 4 core part can do what most people require. Remember when you had to spec out a computer for a family member or friend if they wanted to do a little video editing. "Make sure you get a fast hard drive and CPU...." Not necessary anymore.

With that in mind imagine the volume of CPU's Intel has to reliably supply to all of their vendors. I'm sure the most important thing is just to get them the darn processors! So what if another one is 5 or 10% faster. 95% of the people will never notice it. But the vendors will notice is Intel doesn't get them their parts.

So I'm sure Intel looked at they produce in the most economical fashion and what TMSC could help their supply chain with most efficiently. We would think they would go to TMSC for some low yielding upper part of the stack part, not the i3. But who knows? Maybe because TMSC can do these parts with really high yields Intel can maximize the value with this relationship.

I would bet all 10SF capability is going to Tiger Lake. Maybe another big chunk of their fabs are producing Rocket Lake and that's why they are going to TMSC for I3 supply?

 

[jpiniero]

Couple of question. I've seen many people saying that they didn't expect or understand the i3 being built by TSMC. Question is, what is the actual percentage of i3 sales across all markets vs i5 or i7? Volume.

i5 is the best seller by a wide margin but i3 is still a lot. More than i7.

 

[VirtualLarry]

You do what you have to survive. If you have to go into the heating business as a side job you do it.

Space heaters, with "Intel Inside", coming soon to Home Depot...

That being said, I am kind of surprised that Intel is going with an i3 with TSMC fabbing it, but I'm pretty sure that's just a "test vehicle", and that faster / higher-end CPUs may follow, should it be successful. That would pit Intel directly against AMD, using TSMC to fab their CPUs.

 

[LikeLinus]

i5 is the best seller by a wide margin but i3 is still a lot. More than i7.

I wasn't sure because I know a lot of corporations have lower spec laptops/desktops that they give employees. But it doesn't surprise me that the i5, middle of the road, CPU is the best seller. Would be nice to see the sales numbers though. Wonder how huge the margin is?

As VirtualLarry mentioned, it may just be a lower volume run to see how thing goes and then they can ramp up. The thing is, while AMD has better high performance chips, Intel is still significantly larger and can purchase manufacturing at a much higher volume, if they decided to do so. At least Intel has realized outsourcing may be the way to go and focus on R&D. It'll be interesting to see how this i3 run plays out.

 

[LikeLinus]

That would pit Intel directly against AMD, using TSMC to fab their CPUs.

Intel needs to keep their fabs for the various other products they manufacture. But, what would stop Intel from selling/rent their fab facilities to TSMC (long term vision) to allow them to become the manufacturing partner of Intel? I realize it would take several years to ramp up, but a partnership with TSMC could be very beneficial in the long term.

 

[DrMrLordX]

Intel needs to keep their fabs for the various other products they manufacture. But, what would stop Intel from selling/rent their fab facilities to TSMC (long term vision) to allow them to become the manufacturing partner of Intel? I realize it would take several years to ramp up, but a partnership with TSMC could be very beneficial in the long term.

You aren't the only one to think that way. Personally I see a move like that (or a similar move with Samsung) to be in Intel's best interest long-term, even if it means finally admitting that TSMC is now the market leader in cutting-edge silicon fabrication. Intel would be forced to make some serious concessions, but at least they could sort-of continue to survive as an IDM.

 

[beginner99]

Clock speed is actually an argument for doing i3 at TSMC... would be a lot less pressure to hit high frequencies.

Same can be said for server CPUs. And intels real issue with 10nm seems to be yield (only quad-core parts so far).

One could imagine Intel outsources the volume parts so that they can use up all of their 10nm for server parts because they will need the full capacity due to the terrible yields. No idea about the actual numbers ($) but I doubt this route is more profitable?

 

[Insert_Nickname]

Also, I saw people complain about 4 core. What are corporate, enterprise (for employees) and low end customers purchasing? I know everyone here wants 16c/32T (I have a 5950X and understand), but realistically, what does the market demand and what is the biggest seller? High core/thread CPUs or 4C CPUs that are more than enough to run Office or a web browser? Thats a huge market segment.

I don't know if they're the biggest sellers currently, but 4C/8T seems the ideal budget option right now. Both AMD and Intel have very reasonable priced options in that category. In many cases they provide similar performance to older i7's, and are even 65W TDP rated.

For the really budget end AMDs Athlons provide excellent value. Or at least used to before prices skyrocketed. For basic usage my Athlon 200GE is more then sufficient.

 

[Zucker2k]

Same can be said for server CPUs. And intels real issue with 10nm seems to be yield (only quad-core parts so far).

One could imagine Intel outsources the volume parts so that they can use up all of their 10nm for server parts because they will need the full capacity due to the terrible yields. No idea about the actual numbers ($) but I doubt this route is more profitable?

If yields is the problem with 10nm right now, the last thing Intel should be outsourcing are the i3 chips. I think capacity is the issue. Even with all the gains made in node density, the new server chips from both AMD and Intel are humongous compared to a few years ago. Those things are going to eat up a hefty shipment of silicon, and with Covid-19 inspired upsurge in PC demand, Intel's bets seem to have been placed.

 

[moinmoin]

Just a reminder to you, not long ago people like you didn't believe Intel could fix the awful 10nm clock speeds of CNL and Icelake and many didn't believe in 10nm for desktop either. You are trying to downplay the process improvements which speaks for itself. Tigerlake-U is growing really fast in the market which is a good sign. Furthermore Icelake-SP is based on the awful 10+, it will always struggle because of that. It's hard to believe for you Intel is really improving their 10nm.

Like me? Reminder? Source? If I ever were to think Intel couldn't fix its stuff I'd suggest Intel to sell its foundries, which I don't. What I do think is Intel is taking way too long to fix its stuff thus successfully turning its competitive IDM advantage into a disadvantage.

For that matter the fixation on frequencies is seriously overblown. What's important is performance and yield, and the safest way to performance is increasing IPC whereas raising the frequency is a way out if you are stuck with a given IPC level. Raising frequency is possible by decreasing density (which can be done both through silicon design and on a process node level) as well as by binning (the higher the level the lower the yield). The smaller the dies the better the yield. Intel's trouble with 10nm and 7nm by all accounts is yield.

I have no trouble whatsoever believing Intel is really improving its 10nm, it's the dates it sets that are the issue. It promised Ice Lake SP to launch last year (originally around mid year) and now says volume production only starts this year. Intel's track record of slipping dates is really awful by now. What gives you the confidence that Asper Alder Lake S will be different?

 

[jpiniero]

If you remember, 7 nm was supposed to be ready and ramping right now. The Arizona fab which was intended for 7 nm probally has the 10 nm equipment that was taken down when they decided to go back mostly to 14 nm.

 

[Gideon]

If you remember, 7 nm was supposed to be ready and ramping right now. The Arizona fab which was intended for 7 nm probally has the 10 nm equipment that was taken down when they decided to go back mostly to 14 nm.

And let's also not forget that TSMC started Volume Production of 5nm in April 2020 with products out in Q4. That 3-phase plan will be finished this year with N5P node also out (7% perf or 15% power improvement)

TSMC 5-nanometer node will be ramping at Fab 18, a new 12-inch EUV GigaFab being constructed in three phases. Phase one finished in early 2018 which is where 5-nanometer is ramping. Phase 2 started a little later and is expected to enter volume production in 2020 as well. The final phase, Phase 3, started in 2019 and is planned for volume production in 2021. Fab 18 will also be the future home of their 3-nanometer process which is planned for 2022.

While Intel's 7nm could theoretically be a better node than TSMC 5nm (far from sold on that, but let's give them some benefit of a doubt). Even if that's the case, there is no way it will be better when initially released. History just doesn't make that a probable outcome. And I'm not even talking about Ice Lake (or ahem, Cannon Lake), but even Broadwell wasn't all that smooth sailing as a first 14nm product in 2014.

And 7nm won't be ramping in April 2021. Based on Intel's claims in July 2020 it will be ramping up in Q2 2022 at the earliest (with some limited products out in Q4). That's at least 2 years behind.


Now onto another metric:

• In 2018
  • TSMC had a capex of $10.5 billion
  • Intel ended up with $15.5 billion
• In 2019
  • TSMC expected a capex of ($10 billion to $11 billion) but ended up with $14.9 billion
  • Intel had $16.2 billion
• In 2020
  • TSMC expected a capex of $15 - $16 billion ended up with $17.2 billion
  • Intel's capex budget was $14.2 to $14.5 billion
• In 2021
  • TSMC is expecting a capex of $25 billion to $28 billion vs actual ...

This is why I have an issue with suggesting Intel to produce significant volume at TSMC while they fix their processes. Just think how that would change the numbers above (and thus R&D money).

The only way producing significant volume in TSMC makes sense, is if Intel intends to either become fabless or licence TSMC tech in the future. I just don't see a natural partnership between them otherwise.
It would be much smarter from TSMCs perspective to produce for Apple, AMD, Altera, etc and take Intel's market with partners who have no intention to compete with them in the foundry market in the future.

 

[Doug S]

Intel needs to keep their fabs for the various other products they manufacture. But, what would stop Intel from selling/rent their fab facilities to TSMC (long term vision) to allow them to become the manufacturing partner of Intel? I realize it would take several years to ramp up, but a partnership with TSMC could be very beneficial in the long term.

It might be beneficial to Intel, but I'm struggling to understand where TSMC benefits from this. If Intel is already coming to them with orders it seems it would make more sense for them if Intel wound down its operations as they slowly phased off using their own fabs. If they wanted to own that level of capacity in the US they would have been building a full sized fab in Arizona instead of the smaller version they are building.

 

[Exist50]

This particular rumor is false. I'm saying that as a statement of fact, not a prediction.

If Intel were to use TSMC for any of its high margin products, it would only do so to take advantage of TSMC's superior PPA and better ecosystem IP availability. Any other reason would be foolhardy. It's anything but trivial to port across processes, especially when Intel's one of them.

 

[jpiniero]

It's anything but trivial to port across processes, especially when Intel's one of them.

Intel did say they were moving their designs to be more portable. And if you are going to validate it, you gotta start somewhere.

 

[Exist50]

Intel did say they were moving their designs to be more portable. And if you are going to validate it, you gotta start somewhere.

Let me translate. The Core team (capital-C) refused to update their design methodology to the 21st century, despite everything else having gone in that direction. Supposedly they have recently been convinced to do so, but from what I hear, that's a ways out in the future. This is at least the limiting factor in "making designs more portable". Well, that, and presumably whatever disaggregation/chiplet/"client 2.0" system they come out with.

That aside, there are some IPs that are inherently tied to a particular process. PHYs (e.g. DDR and PCIe) and power delivery (e.g. FIVR) are the biggest culprits. You basically either need to redesign them for the new process, or buy them from someone who already has. And even IPs that are perfectly synthesizable need some optimization. The effort and cost involved makes it untenable for dual sourcing except as an emergency measure for a high priority product.

 

[dr1337]

The effort and cost involved makes it untenable for dual sourcing except as an emergency measure for a high priority product.

Well with the m1 being a thing and apple guaranteed to be increasing core count, and AMD is already at a comfortable 8, not getting absolutely owned in their largest market these next few years might be a high priority.

I agree though, its not likely i3s. Seems like the worst possible idea to split up your flagship mobile production. Would be so much cheaper just to scale down and stay on the same node.

Also wasn't it reported a while ago that Xe was being made at tsmc? That would make 10x more sense to me than low end mobile/desktop processors. edit: I guess this leak is saying both huh? idk I still don't buy it.

 

[John Carmack]

Just a reminder to you, not long ago people like you didn't believe Intel could fix the awful 10nm clock speeds of CNL and Icelake and many didn't believe in 10nm for desktop either. You are trying to downplay the process improvements which speaks for itself. Tigerlake-U is growing really fast in the market which is a good sign. Furthermore Icelake-SP is based on the awful 10+, it will always struggle because of that. It's hard to believe for you Intel is really improving their 10nm.

"Not long ago" was over two and a half years ago which is how much time elapsed between CNL and TGL and it might be a full three years before consumers see anything beyond a small 28W quad core because evidently not all of the process issues have been fixed yet. Three years is in fact an eternity in the process business.

 

[Gideon]

"Not long ago" was over two and a half years ago which is how much time elapsed between CNL and TGL and it might be a full three years before consumers see anything beyond a small 28W quad core because evidently not all of the process issues have been fixed yet. Three years is in fact an eternity in the process business.

For reference 3 years ago Apple's 10nm A11 had only been out for 4 months, and first 7nm chips were still ~8 months away. We are in the same place now for 5nm chips (being out ~ 4 months out). Effectively TSMC has managed 2 (almost) full-node shrinks since then

 

[LikeLinus]

It might be beneficial to Intel, but I'm struggling to understand where TSMC benefits from this. If Intel is already coming to them with orders it seems it would make more sense for them if Intel wound down its operations as they slowly phased off using their own fabs. If they wanted to own that level of capacity in the US they would have been building a full sized fab in Arizona instead of the smaller version they are building.

Intel has deep pockets and huge market share. TSMC is struggling to even keep up with current demand.

Intel provides the facilities, which would save TSMC time and billions of dollars. Why just phase out using your own fabs and then do what with them? Rent it out to TSMC or allow them to use the fabs at a reduced cost for them to manufacture Intel products.

You're taking a rumor as something that has already happened. Even if they did come to TSMC, there is still a good reason to utilize what you already own and can use as a barging chip or to create a mutually beneficial partnership.

This is all theoretical, of course. But doing so would meet shareholders' demands and help solve their manufacturing issues.

For TSMC, it would allow them to sell more product, faster and larger overall production without having to build out the infrastructure. Plus, it allows them to keep increasing fab space and manufacturing for AMD, Apple and whoever else at the same rate they are already doing. TSMC has loyalty to no one company. This would be an expansion play to acquire the manufacturing partnership with a huge company like Intel.