Intel shakes up manufacturing leadership as key Oregon executive sets retirement

[fansink]

That was my point, though. What goals Intel wants to set are arbitrary. Let's say, hypothetically, that back in 2012 Intel said their goal for Intel 10nm was to go into HVM before 2020. Because Intel launching ICL in Sept 2019, 10nm would have been perfectly on time. Now, anyone hearing that goal in 2012 would think that was a sad and pathetic goal. But as long as it was met, Intel by definition didn't underdeliver. So if Intel's roadmap had 3 derivatives on it, that is their business so long as they deliver on it. Rightfully, nobody would say that TSMC had undelivered on N2 because the PPA uplift versus N3P is minor or because the BEOL process is mostly reused with N3. Assuming all continues to go well, TSMC will have delivered what they said they would.

That is just universally true. Until commercially available products are available, everything is academic.

Yes. But that is clearly not going to happen since it is outside the laptop refresh windows and when Intel normally launches new client products. By the letter of the law 5N4Y was not delivered since no 20A and 18A not delivered by July 26th. You won't hear me argue with anyone on that point, as it is objectively true. But what is also objectively true, is that if we want to go down that rabbit hole, we are being pedantic. To say that Ann's plan to pull ahead of TSMC (which was later given the 5N4Y moniker) was a flop is a disingenuous strawman. The point of 5N4Y wasn't a goal in of itself. It was never about putting checkmarks next to a pretty powerpoint slide and saying "Yay team. We did it!". The point was achieving Intel's aspiration of regaining technology leadership. All that really matters long term is the end goal of tech leadership in 2025. Intel 7, 4/3, or 20A are not Intel's future, nor were they ever going to make Intel's external foundry business a $15B per year company by 2030. Those hopes and dreams mostly come down to 18A and its successors.

By that metric, TSMC also only did 1N4Yc.

This is my point. The arbitrary milestones taken to get to the goal are arbitrary. As long as the final destination is reached, that is what really matters (for Intel's business anyway).

I'm anticipating their success to be a pleasant surprise.

 

[siliconbruh999]

I'm anticipating their success to be a pleasant surprise.

yeah it would be surprise considering 18A PPA is on Par with N2 for according to @Daniel Nenni and if Panther Lake launched H2 25.
It would be a big success definitely but the guy who planned it won't be at Intel

 

[Artificer60]

yeah it would be surprise considering 18A PPA is on Par with N2 for according to @Daniel Nenni and if Panther Lake launched H2 25.
It would be a big success definitely but the guy who planned it won't be at Intel

I really believe that is due to the economics more than the plan. Gelsinger seemed to be unwilling to adapt the plan to match the economic realities that Intel is facing. Hence, he is gone. And yet the general vision he laid out still seems to be largely intact from what I can see.

 

[siliconbruh999]

I really believe that is due to the economics more than the plan. Gelsinger seemed to be unwilling to adapt the plan to match the economic realities that Intel is facing. Hence, he is gone. And yet the general vision he laid out still seems to be largely intact from what I can see.

Yeah I agree there was no issue with the strategy but the financing of his plan

 

[Daniel Nenni]

You say that as if it wasn't her idea and her plan... Said plan that she wanted to do before Pat, but she couldn't because Bob was too busy not investing in R&D for her to actually have the resources to execute said catchup plan. And considering the results, I can hardly call 5N4Y an overpromise considering it looks like Intel will have not just have "unquestioned power-performance leadership" and "unquestioned process technology leadership" in 2025; but also unquestioned PPA leadership until first N2 products break onto the market 3-4Q after the first 18A products start getting sold to Intel product's customers. Unless intel doesn't start selling 18A chips this year, then I don't see how one could say that Intel has even partially failed to reach their publicly stated end goal.

No, I clearly said Pat's Schtick. Pat's 5N4Y song and dance. It was an over promise.

Process leadership is debatable. Once Intel 18A has HVM yield we can talk about that.

Let's circle back after the Intel and TSMC events to see what is officially said. Lip-Bu, like TSMC, does not over promise.

You should also look at the foundry side of the business since TSMC is a foundry, not an IDM. Do an Apples-to-Apples comparison.

 

[siliconbruh999]

There is no Apple at Intel only at TSMC so can't do Apple to Apple comparison

 

[nghanayem]

No, I clearly said Pat's Schtick. Pat's 5N4Y song and dance.

Everybody knows it was Ann's vision not Pat's. Pat just gave her the money and acted as a cheerleader. I don't know why Ann would lie about 5N4Y not being her idea after Pat asked her what she could do with a blank check during prior interviews she has done.

It was an over promise.

What deliverable in particular are you talking about that Intel had promised and completely failed to deliver on?

Process leadership is debatable.

I can think of three different vectors to measure that and by every single one Intel is back in a leadership position.

1) From a technology perspective: Tell me one product that is planned to launch in 2025 (or even 2026 for that matter) that has GAA and BSPD.

2) From a PPA perspective:
CC Wei himself admittied that his team believed that 18A is better than any in production or on roadmap N3 or N3E derivative (from a PPA perspective) and that he thought from a PP perspective that it was something of a toss up with N2 but that from a total PPA perspective he expected N2 to lead. And let us not forget that TSMC'S best process with products launching this year is N3P not N2. iPhone HVM would have needed to have started months ago for that to be the case.

3) From a competitive position perspective:
Would you consider TSMC to be a technology leader? If yes then regardless of where 18A sits between N3 and N2, Intel by being in the same ballpark would be extension also be a leader.

Once Intel 18A has HVM yield we can talk about that.

Ok... So, right now since products are launching this year. And according to Intel 18A DD is supposedly tracking ahead of Intel 4 iso time and that was indisputably a very healthy process.

Let's circle back after the Intel and TSMC events to see what is officially said. Lip-Bu, like TSMC, does not over promise.

Unless you expect 18A to suddenly be a finFET no BSPDN process, N2 to get pulled up a year despite prior consistent communication that this isn't happening, N3P suddenly being a 10-15% uplift rather than a 5% or the usual 2-4%, or TSMC suddenly announcing an N2.5 launching products now there is nothing that Intel or TSMC can announce that will change the current competitive positioning assuming TSMC and Intel continue to execute on their respective roadmaps.

You should also look at the foundry side of the business since TSMC is a foundry, not an IDM. Do an Apples-to-Apples comparison.

We've talked about this ad nauseam before. Pat wasn't even CEO when N2 development work started. If there was even an Intel 3 chip available before N2 that would mean Intel was WAY better at foundry service and ecosystem building than TSMC. Just because the ecosystem is less mature doesn't mean the process isn't good or ready for HVM. And if we are talking purely about technical capability it makes no sense to talk about only technologies that are available to fabless chip makers as much of the semiconductor process development is done by IDMs for IDMs. Unless Micron's DRAM and NAND technologies don't count as more advanced than TSMC'S embedded 28nm equivalents because Micron isn't letting fabless companies design their own memory chips l, and Sony's CIS technology also doesn't count because it is only for Sony's imaging group. Heck by that logic Intel is the last foundry to develop a HKMG, strain, and finFET process since their first external foundry tape out with strain, HKMG, and finFET was Altera in like 2014.

 

[Daniel Nenni]

We've talked about this ad nauseam before. Pat wasn't even CEO when N2 development work started. If there was even an Intel 3 chip available before N2 that would mean Intel was WAY better at foundry service and ecosystem building than TSMC. Just because the ecosystem is less mature doesn't mean the process isn't good or ready for HVM. And if we are talking purely about technical capability it makes no sense to talk about only technologies that are available to fabless chip makers as much of the semiconductor process development is done by IDMs for IDMs. Unless Micron's DRAM and NAND technologies don't count as more advanced than TSMC'S embedded 28nm equivalents because Micron isn't letting fabless companies design their own memory chips l, and Sony's CIS technology also doesn't count because it is only for Sony's imaging group. Heck by that logic Intel is the last foundry to develop a HKMG, strain, and finFET process since their first external foundry tape out with strain, HKMG, and finFET was Altera in like 2014.

I'm a foundry person, that is what I do for a living. If you want to compare Intel and TSMC foundry offerings I can do that. If you want to cross match Intel internal products/processes with TSMC I am not the person for that. When Intel Foundry is shipping wafers to customers information will come out. Until then it is just click bait speculation.

I do however feel that people who insist on comparing Intel to TSMC doom Intel Foundry to fail. Pat did this and he is no longer associated with Intel. You will see a much different approach with Lip-Bu Tan. He will never publicly bad mouth another company, especially TSMC or Intel or AMD or Nvidia etc...

I will be at the Intel and TSMC Foundry events. I will write them up in detail.

 

[Mad2]

Anne was a good manager, and a nice person who will be missed, but had nearly zero influence on current TD accomplishments and directions. It is however, very significant that her heir is now Naga, and not Navid, who really has zero experience at all with fab.

Naga has great potential to move intel technology into more cost competitive space with tsmc, important to many devices in consumer space, which must be balanced against their strong offering in high performance space that tsmc covets to retain customers like Nvidia and AMD. To significantly reduce costs, it will take aggressive changes to intel's integration to reduce their mask count, which will help with yield in many cases, but also negatively impact performance in others. TSMC is also adding masks and complexity to be more competitive in high performance space to counter intel with their fastest growing customers who are demanding different perf/power specs than Apple. It is critical intel does not give up this strong advantage in the fastest growing data center space in their efforts to break even on foundry profit margin, or they will give away their strongest opportunity to grow their foundry business. I hope Naga makes a careful analysis of intel's true overall business opportunity in logic space and doesn't assume the memory business is the right model for intel's success.

IMHO, Intel foundry best bet is to become THE high performance/power king. Intel will never win apple, nor qualcomm unless strongarmed by DJT - intel's strength is definitely not in high density, low cost, low power devices, and if intel foundry persists in pursuing this distraction, they will fail - it is tsmc's strong suit. They can win Nvidia, AMD and the CSP's ASICs by focusing on performance/watt for the workloads each of them care about most in high performance data center and network edge. The "system foundry" model with integrated, customized x86 cores with customer IP could also become a very strong proposition if they can make this work, and big reason to keep intel groups together (but firewalled).

Bon Chance, Intel!

 

[XYang2023]

PG discussed his tenure and view on Intel in the beginning of the program.

 

[nghanayem]

Anne was a good manager, and a nice person who will be missed, but had nearly zero influence on current TD accomplishments and directions.

She is a manager not a fellow. Leading and building an organization was her job not innovation. Empowering the right people to do a good job is what a what a good manager should do.

It is however, very significant that her heir is now Naga, and not Navid, who really has zero experience at all with fab.

If memory serves I think Navid used to have some fab experience many moons ago. But yes Naga is a much better choice. Navid made no sense when you have so many rising stars in Ann's new LTD. Heck I would even take Choon over Navid, and as far as I know he has only ever worked in assembly test.

Naga has great potential to move intel technology into more cost competitive space with tsmc, important to many devices in consumer space, which must be balanced against their strong offering in high performance space that tsmc covets to retain customers like Nvidia and AMD.

Naga isn't an individual contributor. He isn't going to be writing recipes, finding cost reduction opportunities, making tool selections, or designing integration schemes. Like Ann a job well done will be ensuring LTD and ATTD are well led, have organizational structures that promote innovation, and setting a vision for the team to accomplish.

To significantly reduce costs, it will take aggressive changes to intel's integration to reduce their mask count, which will help with yield in many cases,

I mean that has already been happening. Intel 4 was a large layer count reduction and power via is another layer count reduction. Now maybe 10A or 7A show a lot more opportunities for process simplification in areas where Intel was overally complex for little or no reason than what Intel found on 4, 18A or 14A. But if that does happen I don't know if I would 100% attribute that to Naga since this tend of finding simplification opportunities was ongoing for years.

but also negatively impact performance in others.

Eh not really how that works. The most common example of that happening is reducing metal layer count. But metal layer count is product dependent.

TSMC is also adding masks and complexity to be more competitive in high performance space to counter intel with their fastest growing customers who are demanding different perf/power specs than Apple.

Mask count increases would be happening even without performance enhancement. That is always how it goes. With that said, while it does seem like TSMC is giving more focus to HPC than the afterthought with it all being mobile mobile mobile, I don't think TSMC is doing that to drag race Intel. TSMC was working in that direction pre Pat, and TSMC is just responding to HPC growing rapidly while mobile stagnates. After all depending on when you look over recent times HPC is as big or mobile now.

It is critical intel does not give up this strong advantage in the fastest growing data center space in their efforts to break even on foundry profit margin, or they will give away their strongest opportunity to grow their foundry business. I hope Naga makes a careful analysis of intel's true overall business opportunity in logic space and doesn't assume the memory business is the right model for intel's success.

Why would Naga make Intel a memory maker? I can all but garuentee that this thought has never once been something he has thought of. It makes no sense.

IMHO, Intel foundry best bet is to become THE high performance/power king. Intel will never win apple, nor qualcomm

I wouldn't say never. Qualcomm was all Samsung for snapdragon for almost a decade across 5 process node families.

unless strongarmed by DJT - intel's strength is definitely not in high density,

Why not? density leadership isn't exactly something Intel has struggled with in the past. If TSMC continues with 3 year process gaps and Intel can maintain the post 5N4Y pace of 2 years I think it seems very likely that TSMC'S meager density lead will turn into Intel's meager density lead.

low cost, low power devices, and if intel foundry persists in pursuing this distraction, they will fail

No. Just no. I really don't think you know what you are talking about.

it is tsmc's strong suit.

Low power performance is an essential skill set especially for HPC. The only garuenteed path to failure is not continuing to enhance POWER-performance, shifting the VF curve down, and miniacally reducing capacitance. If you don't reduce power and cap chip performance will fall off a cliff.

They can win Nvidia, AMD and the CSP's ASICs by focusing on performance/watt for the workloads each of them care about most in high performance data center and network edge. The "system foundry" model with integrated, customized x86 cores with customer IP could also become a very strong proposition if they can make this work, and big reason to keep intel groups together (but firewalled).

People don't seem super interested in Intel's IPs or system foundry. Maybe once Intel is a good foundry and OSAT then system foundry becomes more attractive to smaller players. But that isn't seemingly today (at least assuming Intel's non merchant customers are just getting wafers or assembly/test and not any of the other goodies).

Bon Chance, Intel!

 

[Daniel Nenni]

PG discussed his tenure and view on Intel in the beginning of the program.

I watched the video. Not a good look for Pat G. The first question was very awkward and do you really think Christianity and AI will get along?

 

[XYang2023]

I watched the video. Not a good look for Pat G. The first question was very awkward and do you really think Christianity and AI will get along?

I also don’t fully understand his effort. Is it for profit or charity?

My mother told me that her church hire priests from other churches to conduct teaching sessions on a regular basis. I assume churches could subscribe to such a service to save money, especially since there are many churches worldwide.

 

[Daniel Nenni]

I also don’t fully understand his effort. Is it for profit or charity?

My mother told me that her church hire priests from other churches to conduct teaching sessions on a regular basis. I assume churches could subscribe to such a service to save money, especially since there are many churches worldwide.

Gloo raised more than $100M so it is definitely for profit.

Gloo serves over 70,000 churches and more than 1,000 resource partners. The recent funding aligns with plans to expand the Discover marketplace on Gloo, offering the faith community a comprehensive range of products and services tailored to the specific and ongoing needs of ministry leaders.

 

[Mad2]

OK, you selectively responded to the first part of my sentence here:

unless strongarmed by DJT - intel's strength is definitely not in high density,

Why not? density leadership isn't exactly something Intel has struggled with in the past. If TSMC continues with 3 year process gaps and Intel can maintain the post 5N4Y pace of 2 years I think it seems very likely that TSMC'S meager density lead will turn into Intel's meager density lead.

and the second part of the sentence here:

low cost, low power devices, and if intel foundry persists in pursuing this distraction, they will fail

No. Just no. I really don't think you know what you are talking about.

The "Low cost" part that you chopped is the most important qualifier of the first phrase!

The only reason that density matters in mobile is it's impact on cost (more chips/wafer). It is the A/C of PPAC, and critical for low cost chips in mobile consumer space in $5-20 ASP range. Intel matching density at much higher cost still makes them non-competitive in this space and they have an extremely long way to go to change this according to their own financial data. Consumer CPU's in the $100-500 range are much more forgiving business to be in, especially if you can hide your non-competitive fab costs in your product division. Still better, the fast-growing market in datacenter (albeit still relatively small in units) with $1,000-30,000 ASPs that leverages the high performance/power optimized process already developed for Xeon. If you're intel, it doesn't seem hard at all to me to prioritize what part of the business to go after.

People don't seem super interested in Intel's IPs or system foundry. Maybe once Intel is a good foundry and OSAT then system foundry becomes more attractive to smaller players. But that isn't seemingly today (at least assuming Intel's non merchant customers are just getting wafers or assembly/test and not any of the other goodies

The CSPs are all doing things with semi-custom Xeons already. Not a big stretch for me to see that relationship expanding to more specialized systems-in-package, if they can deliver the goods.

 

[Daniel Nenni]

IMHO, Intel foundry best bet is to become THE high performance/power king. Intel will never win apple, nor qualcomm unless strongarmed by DJT - intel's strength is definitely not in high density, low cost, low power devices, and if intel foundry persists in pursuing this distraction, they will fail - it is tsmc's strong suit. They can win Nvidia, AMD and the CSP's ASICs by focusing on performance/watt for the workloads each of them care about most in high performance data center and network edge. The "system foundry" model with integrated, customized x86 cores with customer IP could also become a very strong proposition if they can make this work, and big reason to keep intel groups together (but firewalled).

You cannot determine how your design will perform using conference papers, PowerPoint slides, or press releases. That is what the PDK is for. The people I know that have evaluated the PDKs have said that Intel 18A is comparable to TSMC N2 in PPA. They have also said that the TSMC N2 PDK is much better than 18A. I know the Intel 18A PDK recently had a new release so that may have changed but the big N2 decisions have already been made.

The biggest market today for Intel 18A today is the NOT TSMC business which was dominated by Samsung Foundry, up until the 5/4/3nm yield debacle. Samsung still has not recovered from that. There is pent up demand for NOT TSMC business and that will be the first 18A customers for sure. Not ones looking for better PPA because it is just not there yet and the risk was too high.

Trust is a big part of the foundry business and choosing a foundry in turmoil without a CEO is a tough one. Now that Lip-Bu is at the helm, in my opinion, trust is much less of an issue with Intel Foundry. Lip-Bu is great in front of customers and has a lot of sway with the C level. I wish he was hired immediately after Pat left but it is what it is.

Bon Chance, Intel!