TSMC capacity glut?

[jav6454]

that's not how it works. When you design a chip, it's for a specific node, say 5nm and you ue the PDK (Process Development kit) that comes with that process. when done with design, you create a mask set to run actually wafers in the fab, those masksets are for 5nm. If you were "switching to 3nm, you basically have to start from scratch. now using the PDK for 3nm ... masksets for 5nm are rumored to be in the $10M range, 3nm masksets are rumored to be closer to $100M. So you're dealing with some heavy expenditures here and don't change plans midway thru ...
"Testing" occurs on actual wafers/chips, during the design phase you "simulate", and your simulation depends on the actual node designed to ...

When you design systems and manufacturing process, you design for one possible outcome but with the flexibility to move to something else should the opportunity arise. Apple is no fool and likely tested 3nm and went as far as even build sample MBPs with 3nm chips. After all TSMC wasn't mass production ready this year. That still doesn't mean sampling and engineer validation didn't occur.

Changing plans on what? An inexistent product line right now? In fact, right now is the best moment as nothing is built.

 

[jz0309]

When you design systems and manufacturing process, you design for one possible outcome but with the flexibility to move to something else should the opportunity arise. Apple is no fool and likely tested 3nm and went as far as even build sample MBPs with 3nm chips. After all TSMC wasn't mass production ready this year. That still doesn't mean sampling and engineer validation didn't occur.

Changing plans on what? An inexistent product line right now? In fact, right now is the best moment as nothing is built.

no, not semiconductor, you design for a specific fab process, you do not have "flexibility" as you describe.
And yes, Apple and TSMC are working hand in hand developing 3nm. A fab cannot create a process without having a product, so yes, Apple has been designing 3nm "products" for easily 2 years, but those are not necessarily real products, but sure they get characterized in real world computers.
"Mass" production in H2'22 means that the process is "frozen" now, Apple I'm sure is in the process of "qualifying" their 3nm products now, maybe even done.
Also keep in mind that it takes 3-4 months to manufacture a chip (fab + assembly)

 

[jav6454]

no, not semiconductor, you design for a specific fab process, you do not have "flexibility" as you describe.
And yes, Apple and TSMC are working hand in hand developing 3nm. A fab cannot create a process without having a product, so yes, Apple has been designing 3nm "products" for easily 2 years, but those are not necessarily real products, but sure they get characterized in real world computers.
"Mass" production in H2'22 means that the process is "frozen" now, Apple I'm sure is in the process of "qualifying" their 3nm products now, maybe even done.
Also keep in mind that it takes 3-4 months to manufacture a chip (fab + assembly)

Agreed on the time to manufacture. In another thread, we discussed that possibility where Apple could well be doing 3nm and delay the M2 Pro/Max to March 2023.

 

[jz0309]

Agreed on the time to manufacture. In another thread, we discussed that possibility where Apple could well be doing 3nm and delay the M2 Pro/Max to March 2023.

Could be, or, the rumor that the A15 will only go into the iPhone Pro models could mean that A15 is actually on 3nm but because of ramping only lower volume available … only Apple really knows at this point…

 

[jav6454]

Could be, or, the rumor that the A15 will only go into the iPhone Pro models could mean that A15 is actually on 3nm but because of ramping only lower volume available … only Apple really knows at this point…

That rumors to me means two things:

1. Apple likely was left out of the initial 3nm capacity by Intel so they prioritized the Pro line.
2. Apple wanted to make the Pro and regular line become more differentiated.

 

[jz0309]

That rumors to me means two things:

1. Apple likely was left out of the initial 3nm capacity by Intel so they prioritized the Pro line.
2. Apple wanted to make the Pro and regular line become more differentiated.

That I find the most unbelievable rumor out there:
. Apple is TSMC's #1 customer, they're not going to "piss" them off.
. Intel has declared last year that they are aiming to replace TSMC as the #1 foundry - what is the incentive for TSMC to enable a competitor?
. Intel has been claiming that their node 4 (forget the name right now) is ahead of schedule, so why do they need TSMC capacity?
. x86 is going to see falling sales in notebook and server as well. Intel needs to keep its fabs full, otherwise they're losing money, a lot of money (which will translate into significant loss of profit).

So we can speculate as much as we want, no-one here "knows"

 

[senttoschool]

Good point regarding the A15 and 5nm, but on the other hand: I really don’t think making M2 Max on 5nm will be enough of a jump to be worthwhile for anything when Apple can make that on 3nm in the start of 2023.

In any event, I can’t wait to see a Mac Studio Ultra when the chip is done on 3nm. There is a decent chance that will already be out during 2023, like May 2023? A M2 Ultra on 5nm would be very meh in my book regarding to what is already out now.

M2 is on 5nm and the jump is good. M2 Max will have an equivalent jump over the M1 Max. Apple has already accepted the M2 jump over M1. Why wouldn't they accept the M2 Max jump over the M1 Max?

I think M3 on 3nm in Spring 2023 or Summer 2023 makes more sense. Making a huge chip on a brand new process probably isn't a good idea anyways.

On a related note, I assume that the M3 will be based on the A16, which is designed for the 4nm process. So there's actually no guarantee that M3 will be on 3nm either. It'd mean that Apple designed the A16 cores for both 4nm and 3nm.

 

[jav6454]

That I find the most unbelievable rumor out there:
. Apple is TSMC's #1 customer, they're not going to "piss" them off.
. Intel has declared last year that they are aiming to replace TSMC as the #1 foundry - what is the incentive for TSMC to enable a competitor?
. Intel has been claiming that their node 4 (forget the name right now) is ahead of schedule, so why do they need TSMC capacity?
. x86 is going to see falling sales in notebook and server as well. Intel needs to keep its fabs full, otherwise they're losing money, a lot of money (which will translate into significant loss of profit).

So we can speculate as much as we want, no-one here "knows"

• TSMC likely contractually allocated 3nm capacity to Intel early due to Intel's negotiations. However, as we know, that failed to fall thru.
• Intel has historically only made Intel designed chips. I haven't seen any of their intentions to become a foundry as GloFo or TSMC. I could be very well wrong. However, I highly doubt Intel will ever enter such competitive market as they have failed time and time again to even get their own designs smaller.
• Likely their node 4 (or whatever the call it) is once again delayed like all previous nodes.
• x86 is not dead yet, but Intel is taking a beating from AMD on the desktop front and on the laptop side, Apple has come out with a superb competitor. I am unsure how AMD's Zen architecture is holding on the mobile/laptop space. But I do reckon their APUs are better than what Intel offers.

 

[senttoschool]

• Intel has historically only made Intel designed chips. I haven't seen any of their intentions to become a foundry as GloFo or TSMC. I could be very well wrong. However, I highly doubt Intel will ever enter such competitive market as they have failed time and time again to even get their own designs smaller.

Then you're living under a rock. Intel has made it clear that they intend to manufacture chips for other companies. Look up "IDM 2.0".

 

[jav6454]

Then you're living under a rock. Intel has made it clear that they intend to manufacture chips for other companies. Look up "IDM 2.0".

I haven't been following Intel ever since they have failed to reduce their nodes.

 

[senttoschool]

I haven't been following Intel ever since they have failed to reduce their nodes.

If you haven't, then why do you write as if you're very confident in these node predictions?

I'd suggest that you read and understand Intel's side first.

 

[jav6454]

If you haven't, then why do you write as if you're very confident in these node predictions?

I'd suggest that you read and understand Intel's side first.

Those are all my opinions. Furthermore, I clearly stated “I highly doubt” which means opinion.

I suggest you don’t try to make everything written here as a fact unless explicitly stated.

 

[bobmans]

M2 is on 5nm and the jump is good. M2 Max will have an equivalent jump over the M1 Max. Apple has already accepted the M2 jump over M1. Why wouldn't they accept the M2 Max jump over the M1 Max?

I think M3 on 3nm in Spring 2023 or Summer 2023 makes more sense. Making a huge chip on a brand new process probably isn't a good idea anyways.

On a related note, I assume that the M3 will be based on the A16, which is designed for the 4nm process. So there's actually no guarantee that M3 will be on 3nm either. It'd mean that Apple designed the A16 cores for both 4nm and 3nm.

Why would A16 be on 4nm when 5NP is better than 4nm? Apple, and probably almost everyone, is not going to be bring out a single chip on TSMC 4nm.