Intel indicating Arrow Lake will not have a 20A version

[Xebec]

Continued Momentum for Intel 18A

Progress on lead product designs and process readiness is enabling us to bridge from Intel 20A earlier than we’d planned.

www.intel.com

One of the benefits of our early success on Intel 18A is that it enables us to shift engineering resources from Intel 20A earlier than expected as we near completion of our five-nodes-in-four-years plan. With this decision, the Arrow Lake processor family will be built primarily using external partners and packaged by Intel Foundry.

(I was personally really hoping to see proof of yield with Arrow Lake 20A…)

 

[Yunus]

Question: How hard is it to port a design from 20A to 18A ? my understanding is that 18A is the full version of 20A which is lacking some libraries.

 

[siliconbruh999]

I guess same as I4 -> I3

 

[MKWVentures]

Now that Intel has Finally admitted 20A will not be used (it was decided a while ago), some things to look for:

1) Intel will have less than one full fab running all sub 7nm processes. 4,3,20, 18A in 2025 is less than 1 full fab. This will save a ton in capital spending.
2) Intel will start with announcements on the success of 18A, then in January will start to talk about the success of 14A, they will downplay 18A.
3) 20A and 18A didnt have and still dont have a financially responsible way to ramp. Intels plans are to "bring the wafers home" but that is spending tons of money with no real return. This is part of the problem with Pats vision. It is financially impossible unless some obviously unrealistic ramp happens. Porting to Intel 3, not 18A is the plan today.
4) what could work and what will probably happen: Intel will launch products from D1. all will be in lower volume until at least 2027.

Now that Intel has finally announced this, we have a note to discuss why this happened, why it was predictable, and what the next steps will be.

This decision plus the lack of foundry volume in 2025 and 2026 has major implications.

 

[Erik F]

Now that Intel has Finally admitted 20A will not be used (it was decided a while ago), some things to look for:

1) Intel will have less than one full fab running all sub 7nm processes. 4,3,20, 18A in 2025 is less than 1 full fab. This will save a ton in capital spending.
2) Intel will start with announcements on the success of 18A, then in January will start to talk about the success of 14A, they will downplay 18A.
3) 20A and 18A didnt have and still dont have a financially responsible way to ramp. Intels plans are to "bring the wafers home" but that is spending tons of money with no real return. This is part of the problem with Pats vision. It is financially impossible unless some obviously unrealistic ramp happens. Porting to Intel 3, not 18A is the plan today.
4) what could work and what will probably happen: Intel will launch products from D1. all will be in lower volume until at least 2027.

Now that Intel has finally announced this, we have a note to discuss why this happened, why it was predictable, and what the next steps will be.

This decision plus the lack of foundry volume in 2025 and 2026 has major implications.

Isn't it possible to bring a partner (like a real partner that comes in to actually produce chips) to the foundry business to bear the 2 years of low volumes, maybe even broaden the list of possible clients?

 

[jorgequinonez]

Now that Intel has Finally admitted 20A will not be used (it was decided a while ago), some things to look for:

1) Intel will have less than one full fab running all sub 7nm processes. 4,3,20, 18A in 2025 is less than 1 full fab. This will save a ton in capital spending.
2) Intel will start with announcements on the success of 18A, then in January will start to talk about the success of 14A, they will downplay 18A.
3) 20A and 18A didnt have and still dont have a financially responsible way to ramp. Intels plans are to "bring the wafers home" but that is spending tons of money with no real return. This is part of the problem with Pats vision. It is financially impossible unless some obviously unrealistic ramp happens. Porting to Intel 3, not 18A is the plan today.
4) what could work and what will probably happen: Intel will launch products from D1. all will be in lower volume until at least 2027.

Now that Intel has finally announced this, we have a note to discuss why this happened, why it was predictable, and what the next steps will be.

This decision plus the lack of foundry volume in 2025 and 2026 has major implications.

They only have or only will have one fab making all three nodes ("4, 3... 18A in 2025...")?

 

[MKWVentures]

They only have or only will have one fab making all three nodes ("4, 3... 18A in 2025...")?

One total fabs worth of loadings on all of those nodes combined in 2025. its doesnt matter if you create node if you dont ramp it. In fact it, creating new technologies and not ramping them is a disaster as Intel is seeing now.... Wafers from these nodes will run in 34 and in D1.

 

[siliconbruh999]

What about Arizona they recived a prepay for it on 18A

 

[kevin01]

One total fabs worth of loadings on all of those nodes combined in 2025. its doesnt matter if you create node if you dont ramp it. In fact it, creating new technologies and not ramping them is a disaster as Intel is seeing now.... Wafers from these nodes will run in 34 and in D1.

In this case, it is actually more understandable that they don't want to spin up 20A manufacturing. Why waste limited euv capacity on low end ARL chips?

 

[MKWVentures]

I cant wait for Pat to claim 20A was a success and that Intel delivered 20A on time. It is very familiar (IBM delivered 2nm GAA 2 years ago).

 

[Xebec]

3) 20A and 18A didnt have and still dont have a financially responsible way to ramp. Intels plans are to "bring the wafers home" but that is spending tons of money with no real return. This is part of the problem with Pats vision. It is financially impossible unless some obviously unrealistic ramp happens. Porting to Intel 3, not 18A is the plan today.

I'd like to play this out a little bit. How far in advance / how many years of wafer volume do you think Intel committed to on TSMC N3?

I'm assuming at least 2024 and 2025 are set for handling Intel's desktop and mobile platforms. However, when we get to 2026, in theory Intel would want something along the lines of TSMC N2 (and definitely by 2027) to stay ahead of competition in Mobile and Desktop. That would necessitate a full volume ramp of 18A around the end of 2025 to sustain the mobile/desktop market in 2026.

While that is "bring the wafers home" it's also an automatic self-fulfilling prophecy if Intel didn't prebook N3 capacity after a certain date. I don't see any way Intel could back off 18A without a critical shortage of sellable products in FY2026. The market is softening in the near term for these types of products (Zen 5 is already on sale days after launch, and Intel's desktop failure nightmare).

Thoughts?

(Also I'm assuming Battlemage is a TSMC N4 product; meaning not wafers they could shift to mobile/desktop unless they're "using" N3 Celestial planned capacity for a potential "original plan" of late 2025 or early 2026...)

 

[MKWVentures]

TSMC N3 was not done because the capacity was allocated. Intel has increased the wafers starts from plan dramatically. TSMC is easily able to accept them.

There is not enough demand to ramp 20A/18A in a new fab. There only was enough demand if foundry started to ramp in 2025 on 18A which is not happening

Can't get costs down without ramping a new fab, Can't afford a new fab if costs dont come down....
If you don't build foundries for others, you can't get costs down. If you don't have costs down, you cant build foundries for others.

... wow, it is almost like there is a reason for using a foundry business and not trying to be a foundry.

Some CEO had the idea in 2020 for Intel to use foundry because the finances don't ever work out for Intel being an IDM and it takes decades to be successful in foundry. hmmm

 

[kevin01]

I'd like to play this out a little bit. How far in advance / how many years of wafer volume do you think Intel committed to on TSMC N3?

I'm assuming at least 2024 and 2025 are set for handling Intel's desktop and mobile platforms. However, when we get to 2026, in theory Intel would want something along the lines of TSMC N2 (and definitely by 2027) to stay ahead of competition in Mobile and Desktop. That would necessitate a full volume ramp of 18A around the end of 2025 to sustain the mobile/desktop market in 2026.

While that is "bring the wafers home" it's also an automatic self-fulfilling prophecy if Intel didn't prebook N3 capacity after a certain date. I don't see any way Intel could back off 18A without a critical shortage of sellable products in FY2026. The market is softening in the near term for these types of products (Zen 5 is already on sale days after launch, and Intel's desktop failure nightmare).

Thoughts?

(Also I'm assuming Battlemage is a TSMC N4 product; meaning not wafers they could shift to mobile/desktop unless they're "using" N3 Celestial planned capacity for a potential "original plan" of late 2025 or early 2026...)

As Intel exeuctives communicated in April,

So it's on the order of 30-ish percent of our wafers today that we bring in externally. We'll be in-sourcing some level. As I said, a couple of fab modules, we expect over this period of time. So we expect that to moderate to down below 20% over the period even as we continue to use.

It seems to me that Intel has booked TSMC leading edge capacity till at least 2026. I would not be surprised that Diamond Rapids is actually made on N2.

 

[kevin01]

TSMC N3 was not done because the capacity was allocated. Intel has increased the wafers starts from plan dramatically. TSMC is easily able to accept them.

There is not enough demand to ramp 20A/18A in a new fab. There only was enough demand if foundry started to ramp in 2025 on 18A which is not happening

Can't get costs down without ramping a new fab, Can't afford a new fab if costs dont come down....
If you don't build foundries for others, you can't get costs down. If you don't have costs down, you cant build foundries for others.

... wow, it is almost like there is a reason for using a foundry business and not trying to be a foundry.

Some CEO had the idea in 2020 for Intel to use foundry because the finances don't ever work out for Intel being an IDM and it takes decades to be successful in foundry. hmmm

I am actually not concerned with the lack of external 18A demand. To me, external demand and internal demand are the same, financially speaking. Meaning, both can provide funding to the fab.

However, if Intel does not dare to ramp up 18A with their own products because they are not confident with 18A performance and yield, that is serisoully worrying. Eat your own dog food first.

That said, at least for now, seems that 18A is progressing reasonably well. Let's wait for PTL ES performance leaks,

 

[Daniel Nenni]

TSMC N3 was not done because the capacity was allocated. Intel has increased the wafers starts from plan dramatically. TSMC is easily able to accept them.

There is not enough demand to ramp 20A/18A in a new fab. There only was enough demand if foundry started to ramp in 2025 on 18A which is not happening

Can't get costs down without ramping a new fab, Can't afford a new fab if costs dont come down....
If you don't build foundries for others, you can't get costs down. If you don't have costs down, you cant build foundries for others.

... wow, it is almost like there is a reason for using a foundry business and not trying to be a foundry.

Some CEO had the idea in 2020 for Intel to use foundry because the finances don't ever work out for Intel being an IDM and it takes decades to be successful in foundry. hmmm

The whole 5 processes in 4 years was flawed from day one. Intel 4/3 should be one process and Intel 20/18A should be one. That is what is happening now. There is not enough volume to support all of these separate fabs. Pat set expectations too high and now he is going to pay for it, possibly with his job. I said this from day one, Intel's foundry expectations are unreachable.

Intel booked a ton of TSMC N3 fab space with a massive pre-pay and a very lucrative wafer agreement. Intel has not signed a TSMC N2 agreement yet which is a good sign. Remember, Intel is moving to the chiplet architectures and chiplets are much easier to manufacture than complex AI chips or SoCs.

It will be interesting to see what happens after the Intel board meeting. My guess is there will be another pivot. Still IDM 2.0 but scaled back quite a bit.

What odds do you give Pat Gelsinger remaining CEO? I would say 50/50 at best.

 

[Xebec]

What odds do you give Pat Gelsinger remaining CEO? I would say 50/50 at best.

Is there any former engineer in the potential running for CEO? If they choose an MBA, stick a fork in Intel..

 

[hist78]

The whole 5 processes in 4 years was flawed from day one. Intel 4/3 should be one process and Intel 20/18A should be one. That is what is happening now. There is not enough volume to support all of these separate fabs. Pat set expectations too high and now he is going to pay for it, possibly with his job. I said this from day one, Intel's foundry expectations are unreachable.

Intel booked a ton of TSMC N3 fab space with a massive pre-pay and a very lucrative wafer agreement. Intel has not signed a TSMC N2 agreement yet which is a good sign. Remember, Intel is moving to the chiplet architectures and chiplets are much easier to manufacture than complex AI chips or SoCs.

It will be interesting to see what happens after the Intel board meeting. My guess is there will be another pivot. Still IDM 2.0 but scaled back quite a bit.

What odds do you give Pat Gelsinger remaining CEO? I would say 50/50 at best.

In a more typical Intel's style, they will call it IDM 3.0.

4 years 5 nodes put too much burden on Intel's finance and engineering resources while there are several Intel's Board of Directors are finance experts. Isn't it strange?

 

[delong.height]

The whole 5 processes in 4 years was flawed from day one. Intel 4/3 should be one process and Intel 20/18A should be one. That is what is happening now. There is not enough volume to support all of these separate fabs. Pat set expectations too high and now he is going to pay for it, possibly with his job. I said this from day one, Intel's foundry expectations are unreachable.

Intel booked a ton of TSMC N3 fab space with a massive pre-pay and a very lucrative wafer agreement. Intel has not signed a TSMC N2 agreement yet which is a good sign. Remember, Intel is moving to the chiplet architectures and chiplets are much easier to manufacture than complex AI chips or SoCs.

It will be interesting to see what happens after the Intel board meeting. My guess is there will be another pivot. Still IDM 2.0 but scaled back quite a bit.

What odds do you give Pat Gelsinger remaining CEO? I would say 50/50 at best.

I don't think booking too much capacity and planning even more for 5 nodes in 4 years was the biggest mistake to be pointed Pat Gelsinger.

A lot of these were done to de-risk the potential drawback the IFS business would potentially have to encounter. Back to 2020 and 2021, the saying was Intel design were hold back by its manufacturing arm, hence the need to outsource. They just don't know how much was actually needed. I think they do it for the worst possible outcome. And here we are, the result (5N4Y) aren't worst, it's just great. Having 0.4 in defect speak 18A is indeed very healthy.

Also, when looking the balance sheet Intel had back in 2021, the margin were health, the revenue were strong. Everything looks so good, and they executed the plan. Then not long after that, the inventory accumulate because the demand has peaked during COVID. Nobody need a new PC. The semiconductor were in depression. The memory chips has to be sold at discount. Also AMD were eating Intel's lunch alone the way, Intel doesn't have good margin at data center, they were losing money. Then suddenly AI came, overthrowing Intel. And Nvidia ate all the data center revenue because their GPU are scarce.

I mean who would thought this was possible back in beginning of 2021. All of these bad things were just miracle to me, and I think to Pat G as well.

I said this from day one, Intel's foundry expectations are unreachable.

It would not be possible if intel still has 2021's balance sheet and income statement , like when Pat G was taking over. I think the worst he had predicted was low-mid 60B in terms of revenue and mid/high 40 gross margin at worst.

 

[delong.height]

It will be interesting to see what happens after the Intel board meeting. My guess is there will be another pivot. Still IDM 2.0 but scaled back quite a bit.

They will need a partner, a partner who they can offload this burden (panther lake, clearwater forest, diamond rapid, falcon shore, with their dependence on 18A). The financial burden to invest ahead of time versus the wafer required are a positive linear equation. The more wafer/product that depend on 18A, the more financial burden they will they to go through.

So they need a partner, Samsung is the only one out there who also need some help. https://semiwiki.com/forum/index.ph...split-of-intel-and-to-intel-management.20914/

 

[delong.height]

What worry me the most, is the financial burden of 18A, unless they outsource 18A to samsung, I don't think they will survive post-2025. Yes you hear me right. If they are delivering all those products that are at leading edge at late 2025-early 2026. They have the biggest possibility of running out of cash and bankupted before the dawn.