Intel CEO implies Intel 4 already being skipped with Granite Rapids (Intel 3) tapeout

[Fred Chen]

Intel (INTC) Q1 2022 Earnings Call Transcript | The Motley Fool

INTC earnings call for the period ending March 31, 2022.

www.fool.com

Intel Q1 2022 Earnings Call
Apr 28, 2022, 5:00 p.m. ET

Intel 3, we'll see the test wafers on that with our leadership products with Sierra Forest. In fact, just today, we taped out our first Granite Rapids compute die as well. We'll have the test wafers on 20a and 18a, which we expect to be a big foundry node as well. So I'll say, overall, the technology pipeline is doing tremendously well, and really proud of our teams there.

 

[semiman]

From what I remember, this was a plan from the start. Intel's plan was to achieve 'lead product test wafer in 2H of 2022'. Intel 4 can't be a high volume node.

Anyway, Intel 7(former Intel 10nm) will have a tough time fighting TSMC N5 Genoa. Granite Rapids arrives in 2024, and in 2023 what's left for Intel is refreshed Intel 7 node.

 

[Fred Chen]

In the opening, Pat Gelsinger outlined the planned milestones, indeed making it sound like that they had planned it to be like this:

Intel 7 is ramping extremely well with Alder Lake and on Intel 4, Meteor Lake has now successfully booted Windows 12 and Linux. The speed at which the team was able to achieve this milestone is a significant sign of the health of both Meteor Lake and our Intel 4 process technology. We plan to deliver several additional milestones in 2022, demonstrating our process technology development remains on track. This includes early Sierra Forest preproduction wafers on Intel 3, IP test wafers on Intel 28 (sic), and foundry customer test chips and initial IP shovels (sic) on Intel 18a.

It looks like development of 4 nodes at the same time, possibly pulling the ribbonFET into this year. If they succeed in doing that, it would be an impressive coup indeed, taking the lead from TSMC.

 

[Fred Chen]

I've always been suspicious they got some hints from TSMC's N3 PDK. It's an insidious form of customer-foundry competition.

 

[benb]

I've always been suspicious they got some hints from TSMC's N3 PDK. It's an insidious form of customer-foundry competition.

This is a weakness of the late-stage Foundry business model, where there are so few customers, and some are customer/competitor. It erodes the advantage of being pure-play foundry; that purity can be spun into a weakness, since TSMC cannot retaliate to theft and repay in-kind. Unless TSMC started making it's own products and building them at Intel Foundry.

 

[Daniel Nenni]

This is a weakness of the late-stage Foundry business model, where there are so few customers, and some are customer/competitor. It erodes the advantage of being pure-play foundry; that purity can be spun into a weakness, since TSMC cannot retaliate to theft and repay in-kind. Unless TSMC started making it's own products and building them at Intel Foundry.

Very true. A big TSMC advantage is building to scale (good yields at very high volumes with a wide range of customers). In fact they are the only foundry that can do that today.

Intel and Samsung can claim small foundry victories but unless they can build fabs to scale they will never overtake TSMC.

 

[blueone]

Very true. A big TSMC advantage is building to scale (good yields at very high volumes with a wide range of customers). In fact they are the only foundry that can do that today.

Intel and Samsung can claim small foundry victories but unless they can build fabs to scale they will never overtake TSMC.

It's not clear to me that Intel is going after the same foundry model as TSMC. TSMC has, what, thousands of customers, including start-ups? I think Intel has a vision of targeting big customers going after diversification for their supply chain. All of the big chip designers know that any foundry can slip dates and miss objectives, and the probability of misses is getting higher as there's a major technology transition coming up (GAA). A lot of companies grow up with TSMC; Intel doesn't appear to be anywhere near ready for that.

I would feel better about Intel's chances if PG was an experienced visionary in being a foundry customer or in running a foundry. He has done neither one. Thakur, Intel's most senior foundry guy, apparently hasn't either, from reading his bio. It looks like the closest senior person they have to being in a foundry business is Hong Hao, who they hired away from Samsung's US foundry group. I suspect Intel's foundry group is going to be discovering this business as they go.

 

[Xebec]

It's not clear to me that Intel is going after the same foundry model as TSMC. TSMC has, what, thousands of customers, including start-ups? I think Intel has a vision of targeting big customers going after diversification for their supply chain. All of the big chip designers know that any foundry can slip dates and miss objectives, and the probability of misses is getting higher as there's a major technology transition coming up (GAA). A lot of companies grow up with TSMC; Intel doesn't appear to be anywhere near ready for that.

I would feel better about Intel's chances if PG was an experienced visionary in being a foundry customer or in running a foundry. He has done neither one. Thakur, Intel's most senior foundry guy, apparently hasn't either, from reading his bio. It looks like the closest senior person they have to being in a foundry business is Hong Hao, who they hired away from Samsung's US foundry group. I suspect Intel's foundry group is going to be discovering this business as they go.

FWIW - this is one of the reasons they bought Tower. From an interview elsewhere - PG has indicated that he wants Intel to adopt Tower's culture (not the other way around) in Foundry 2.0.

 

[blueone]

FWIW - this is one of the reasons they bought Tower. From an interview elsewhere - PG has indicated that he wants Intel to adopt Tower's culture (not the other way around) in Foundry 2.0.

I see what you're saying, but Tower is in a completely different semi market. Do you think they really know anything about the cutting edge foundry business for the likes of Qualcomm and Apple? I honestly don't know, but it does not strike as an intuitive conclusion to come to, that they do.

 

[tooLongInEDA]

In the opening, Pat Gelsinger outlined the planned milestones, indeed making it sound like that they had planned it to be like this:

Intel 7 is ramping extremely well with Alder Lake and on Intel 4, Meteor Lake has now successfully booted Windows 12 and Linux. The speed at which the team was able to achieve this milestone is a significant sign of the health of both Meteor Lake and our Intel 4 process technology. We plan to deliver several additional milestones in 2022, demonstrating our process technology development remains on track. This includes early Sierra Forest preproduction wafers on Intel 3, IP test wafers on Intel 28 (sic), and foundry customer test chips and initial IP shovels on Intel 18a.

It looks like development of 4 nodes at the same time, possibly pulling the ribbonFET into this year. If they succeed in doing that, it would be an impressive coup indeed, taking the lead from TSMC.

Can we here at least just call a spade a spade ?

Sometimes it's "4 nodes in 5 years", sometimes 4 nodes.

Either way, looking at the density changes, two of these look like half nodes (20% area shrink rather than 40-50%).

So it's not really 4 full nodes, is it ?

I do hate this "chaos marketing".

 

[Daniel Nenni]

I would feel better about Intel's chances if PG was an experienced visionary in being a foundry customer or in running a foundry. He has done neither one. Thakur, Intel's most senior foundry guy, apparently hasn't either, from reading his bio. It looks like the closest senior person they have to being in a foundry business is Hong Hao, who they hired away from Samsung's US foundry group. I suspect Intel's foundry group is going to be discovering this business as they go.

This was a big issue for me but the purchase of Tower Semiconductor solved that one. I have a huge amount of respect for what Tower has accomplished over the years. There is a lot of talk about new nodes but I think the Tower people can fill the 22nm and 14nm Intel fabs up without a problem. Tower has been stuck at 45nm so there is probably pent up demand for FinFET chips at Tower customers. I also think the Tower folks can unlock the value of Intel IP quite quickly, much better that Intel themselves. Tower is also very strong in the automotive market which is high profile.

 

[Portland]

This is the reason for the tower "purchase."

Everyone is ignoring mems when they shouldn't.

 

[Fred Chen]

Can we here at least just call a spade a spade ?

Sometimes it's "4 nodes in 5 years", sometimes 4 nodes.

Either way, looking at the density changes, two of these look like half nodes (20% area shrink rather than 40-50%).

So it's not really 4 full nodes, is it ?

I do hate this "chaos marketing".

I feel so too, more like two nodes. But the milestones are crammed so it looks like they're trying to rush the RibbonFET endgame. In a way, this follows Samsung's precedent.

 

[blueone]

This was a big issue for me but the purchase of Tower Semiconductor solved that one. I have a huge amount of respect for what Tower has accomplished over the years. There is a lot of talk about new nodes but I think the Tower people can fill the 22nm and 14nm Intel fabs up without a problem. Tower has been stuck at 45nm so there is probably pent up demand for FinFET chips at Tower customers. I also think the Tower folks can unlock the value of Intel IP quite quickly, much better that Intel themselves. Tower is also very strong in the automotive market which is high profile.

I'm glad you feel that way, and now that you explain it, I agree, Tower could be a significant market-maker for Intel's older fabs. I have so many friends who work there, I want nothing but success for Intel.

 

[semiman]

It's not clear to me that Intel is going after the same foundry model as TSMC. TSMC has, what, thousands of customers, including start-ups? I think Intel has a vision of targeting big customers going after diversification for their supply chain. All of the big chip designers know that any foundry can slip dates and miss objectives, and the probability of misses is getting higher as there's a major technology transition coming up (GAA). A lot of companies grow up with TSMC; Intel doesn't appear to be anywhere near ready for that.

I would feel better about Intel's chances if PG was an experienced visionary in being a foundry customer or in running a foundry. He has done neither one. Thakur, Intel's most senior foundry guy, apparently hasn't either, from reading his bio. It looks like the closest senior person they have to being in a foundry business is Hong Hao, who they hired away from Samsung's US foundry group. I suspect Intel's foundry group is going to be discovering this business as they go.

PG is not from either Foundry or Fabless. He's a kinda manufacturing guy. But he was the one who introduced IDE (Intel Developer Forum). This indicates that he's the one who understands the surrounding ecosystem of a certain industry. At least he knew developers are the ones who make Intel CPUs so valuable. CPU and developer's relation resembles that of TSMC's market dominance + IP alliance members. Maybe he can learn from that.

Will PG succeed? maybe not. But I think he's the best bet Intel can find. Intel has a long history and has its own unique culture with a massive number of people, so they need to find 'Intel way' of doing foundry things anyway. The road's kinda bumpy. Intel has a massive IP portfolio which can scare big fabless customers out, but there are some chances in the market. For example, companies like Microsoft has an economy of scale to start designing their own chips but have weak IP lineups. Intel x86 IPs might attract them.

 

[blueone]

PG is not from either Foundry or Fabless. He's a kinda manufacturing guy. But he was the one who introduced IDE (Intel Developer Forum). This indicates that he's the one who understands the surrounding ecosystem of a certain industry. At least he knew developers are the ones who make Intel CPUs so valuable. CPU and developer's relation resembles that of TSMC's market dominance + IP alliance members. Maybe he can learn from that.

Will PG succeed? maybe not. But I think he's the best bet Intel can find. Intel has a long history and has its own unique culture with a massive number of people, so they need to find 'Intel way' of doing foundry things anyway. The road's kinda bumpy. Intel has a massive IP portfolio which can scare big fabless customers out, but there are some chances in the market. For example, companies like Microsoft has an economy of scale to start designing their own chips but have weak IP lineups. Intel x86 IPs might attract them.

What do you mean by PG being a "kinda manufacturing guy"? That his strategy is to maintain and extend Intel's fabrication capability, or that he has manufacturing experience? To my knowledge, he does not have any experience in manufacturing, though when he was at Intel previously he was a direct customer of internal manufacturing, which at least gives him a strong basis for understanding what chip developers want from foundries.

 

[VCT]

If Intel Foundry Spin off in the future , they won't have enough profit to build next generation Fab.