Intel Could Manufacture Apple's Next-Generation A20 SoC for iPhone

[StimpsonJCat]

They might use Intel foundries for low-end chips, but not for the cutting-edge chips that actually matter.

 

[Wirko]

Intel 7 is similar to n7, just exactly what suggest the density of both nodes:
N7 - 91.2
Intel 7 - 100.76
Intel 3 is also slightly behind N5:
Intel 3 (same density as 4) - 123.4
N5 - 138.2
But A18 is build on N3E - 216 and intel's 18A definitely won't be any close to this even if it was existing node today, let alone after 2 years when A20 will be on N2 or N2P.
Their 20A node has 15% higher perf/watt than Intel 3 so this node is waaaaay behind even the current TSMC nodes

This also means that TSMC's fabs in Arizona alone are going to be more advanced than anything from Intel, right?

 

[londiste]

This also means that TSMC's fabs in Arizona alone are going to be more advanced than anything from Intel, right?

No, currently.
The first TSMC Arizona fab is there, N4 and should start production first half next year. Second fab is 2-3 years after that, on N3 and N2. Third more advanced fab is planned but later.
Intel has Intel 4 in production. Whether they have or if/when will have 18A ready is debatable (18A should be more of a match for N2 rather than N3).

Edit:
The problem with transistor density comparisons is that you need to look at what variation of the process is being mentioned. Intel should still largely have one main variation that is tuned for performance - frequencies mostly. TSMC has 2-3 variations, high performance and high density being the canonical ones. Rule of thumb - when looking at the declared TSMC density numbers, 2/3rds of that is what a high performance variation gets (by some accounts even less with latest nodes).

This is of course simplified. Intel is starting to offer options. TSMC has more than 2 variants and offers a certain amount of mix and match these days even on the same die. Etc. but for general comparison, Intel's processes should be compared to TSMC ones using HP libraries.

Intel 3 is also slightly behind N5:
Intel 3 (same density as 4) - 123.4
N5 - 138.2
But A18 is build on N3E - 216 and intel's 18A definitely won't be any close to this even if it was existing node today, let alone after 2 years when A20 will be on N2 or N2P.
Their 20A node has 15% higher perf/watt than Intel 3 so this node is waaaaay behind even the current TSMC nodes

Disputable. To dissect some of these MTr/mm^2 numbers more:
Intel has given 123.4 number for 4, yes. For Intel 3 they said 10% more density, so 135-136.
TSMC N5 HP reportedly gets somewhere between 90-95. Not sure if that is accurate but Wikichip estimated N3E HP density to 124 which would put Intel 4 and Intel 3 squarely against TSMC N3 group.

For the future TSMC has not claimed very high improvements for N2 over N3. Neither has Intel as you mentioned for 18A over 3. It does seem that Intel 18A will go head to head with TSMC N2. The question isn't about the size/density or performance of the node - the question is if and when Intel will be able to manufacture on given process with sufficient yields.

Edit2:
From a year ago but not much has changed since then:

TSMC N3, And Challenges Ahead

A Look At TSMC N3 Process

fuse.wikichip.org

 

[Chane]

Unless it's much, much, cheaper for Apple to do so, this won't happen. Apple moved away from using Intel CPUs because they stopped being able to deliver on their node promises. Intel still to this day can't get their foundry in order and has outsourced Arrow Lake production to TSMC because of that. This story is a nothing burger.

 

[chrcoluk]

Only way this happens or even makes sense is if they actually bought Intel first.

Its a thought, but TSMC could also be getting greedy, and Apple throwing something Intel's way could be a way for them to remind TSMC they not the only option.

 

[usiname]

Disputable. To dissect some of these MTr/mm^2 numbers more:
Intel has given 123.4 number for 4, yes. For Intel 3 they said 10% more density, so 135-136.
TSMC N5 HP reportedly gets somewhere between 90-95. Not sure if that is accurate but Wikichip estimated N3E HP density to 124 which would put Intel 4 and Intel 3 squarely against TSMC N3 group.

For the future TSMC has not claimed very high improvements for N2 over N3. Neither has Intel as you mentioned for 18A over 3. It does seem that Intel 18A will go head to head with TSMC N2. The question isn't about the size/density or performance of the node - the question is if and when Intel will be able to manufacture on given process with sufficient yields.

Edit2:
From a year ago but not much has changed since then:

TSMC N3, And Challenges Ahead

A Look At TSMC N3 Process

fuse.wikichip.org

Just learn to read

 

[londiste]

Just learn to read

Did you happen to read anything I wrote?

 

[lexluthermiester]

When the hell did THIS happen? Well, it seems Apple will be doing Intel again. Fabs are Fabs after all, they can make any IC as long as the stencils work right.

 

[Flanker]

Not sure if this is possible for chip manufacturing. But Apple does have a track record of making sure they have more than one viable supplier for each component, and they would help out the worse ones to be able to compete. So that the dominant supplier wouldn't have a stranglehold on Apple.

 

[kondamin]

Not sure if this is possible for chip manufacturing. But Apple does have a track record of making sure they have more than one viable supplier for each component, and they would help out the worse ones to be able to compete. So that the dominate supplier wouldn't have a stranglehold on Apple.

“Help” like giving them other companies designs to make their own version and then Dropping them like a burning piece of coal the moment they get caugt.

 

[user556]

It might be a case of Intel packaged TSMC dies. Isn't that basically how Intel's newest CPUs are put together now?

 

[Wirko]

It might be a case of Intel packaged TSMC dies. Isn't that basically how Intel's newest CPUs are put together now?

My speculation is wickeder than yours: Intel could buy TSMC's stuff by the wafer, cut the wafers into chips and sell them under their own brand!

Or chips to run the bidet toilets at Apple HQ.

"Our transistorized water heater is 386% faster than the competition, see the slide deck here!"

 

[Ferrum Master]

Apple never puts all his eggs in one basket...

 

[Dr. Dro]

Provided their 18A node doesn't flop, I see Apple using Intel's foundry services. If not on the very latest iPhone, at least for the more budget-friendly models.

 

[mrnagant]

Maybe Intel is offering production at cost or something. Or Intel eats the cost of poor yields. To even have Apple entertain the idea of this.

But it could also be just for diversification, since China hasn't been too friendly with Tawain. But then, they do have that new plant in Arizona. Having that as an option could potentially depend on when US TSMC will be allowed to manufacture 2nm chips.

 

[lexluthermiester]

Maybe Intel is offering production at cost or something. Or Intel eats the cost of poor yields. To even have Apple entertain the idea of this.

No.

But it could also be just for diversification, since China hasn't been too friendly with Tawain.

This is far more likely.