Array
(
    [content] => 
    [params] => Array
        (
            [0] => /forum/index.php?threads/intel-believes-idm-model-is-better.16119/
        )

    [addOns] => Array
        (
            [DL6/MLTP] => 13
            [Hampel/TimeZoneDebug] => 1000070
            [SV/ChangePostDate] => 2010200
            [SemiWiki/Newsletter] => 1000010
            [SemiWiki/WPMenu] => 1000010
            [SemiWiki/XPressExtend] => 1000010
            [ThemeHouse/XLink] => 1000970
            [ThemeHouse/XPress] => 1010570
            [XF] => 2021370
            [XFI] => 1050270
        )

    [wordpress] => /var/www/html
)

Intel believes IDM model is better

hist78

Well-known member
“This initiative is a terrific example of how integrated manufacturing is foundational to Intel’s success. Our global factory network and supplier ecosystem directly enable a more adaptable and resilient supply of products. Over the past year, as substrates were constrained across the industry, our ability to leverage internal capacity created more than $2 billion in revenue upside for Intel, allowing us to respond with agility to meet dynamic customer demand.”

–Keyvan Esfarjani, Intel executive vice president and chief global operations officer

Source: https://www.intel.com/content/www/u...ctory-helps-overcome-substrate-shortages.html


My thoughts:

Gauging Intel's 2021 revenue growth against other major semiconductor companies, it doesn't show that Intel's IDM model is superior. Most major semiconductor companies can grab the opportunity to grow their revenue by double digits percentage in 2021. But in the same year Intel can only increase their revenue by 1% under their supposedly excellent IDM capabilities.

1653950672849.png


From the Intel Q1 2022 negative YoY revenue growth, I can't see the IDM model is helping Intel greatly either.

 
Last edited:
I don't think IDM model is to blame here. IDM model works as long as certain products hugely benefit by economies of scale. Vertically integrated products are pretty strong and give considerable benefits to users as well. Intel failed due to a lack of contingency plans. They didn't have usable half-nodes to fill a gap when a leading-edge node fails, nor had foundry-friendly design cultures to switch their CPUs.
I think this is how Intel tries to fix its IDM model. Keep vertically integrated products, learn and adapt to foundry businesses(in leading-edge) to get contingency plans and to get access to different microfabrication technology to design new products which are not suitable for IFS. Interesting days indeed.
 
I don't think IDM model is to blame here. IDM model works as long as certain products hugely benefit by economies of scale. Vertically integrated products are pretty strong and give considerable benefits to users as well. Intel failed due to a lack of contingency plans. They didn't have usable half-nodes to fill a gap when a leading-edge node fails, nor had foundry-friendly design cultures to switch their CPUs.

Intel did process improvements 14nm, 14+, 14++, 14+++ etc.... but did not market them as half nodes. We used to do shrinks between nodes but they were really only for the high volume chips. TSMC and Apple were really behind the half node marketing strategy. Apple came to TSMC at 20nm then 16nm which was 20nm with FinFETs using the same fabs. That was the first TSMC half node. Then N10, N7, N7+, N6, N5, N4, and N3. We call them half nodes but what they really are is yield learning process steps.

TSMC will be talking about their new N2 node at the Technical Symposium this month. That will be a new node for sure followed by many enhancements (half nodes). Semiconductors have never been more exciting, absolutely.
 
Intel did process improvements 14nm, 14+, 14++, 14+++ etc.... but did not market them as half nodes. We used to do shrinks between nodes but they were really only for the high volume chips. TSMC and Apple were really behind the half node marketing strategy. Apple came to TSMC at 20nm then 16nm which was 20nm with FinFETs using the same fabs. That was the first TSMC half node. Then N10, N7, N7+, N6, N5, N4, and N3. We call them half nodes but what they really are is yield learning process steps.

TSMC will be talking about their new N2 node at the Technical Symposium this month. That will be a new node for sure followed by many enhancements (half nodes). Semiconductors have never been more exciting, absolutely.
Daniel - I thought the "first TSMC half node" was much further back -- example: TSMC 55nm, inbetween 65nm and 45nm. TSMC had all three 'nodes', while other companies like Intel and AMD/Glofo jumped from 65nm to 45nm..? (I also remember foundry 80nm and 110nm 'half nodes')..

(I really like the "yield learning process steps" quote here).
 
“This initiative is a terrific example of how integrated manufacturing is foundational to Intel’s success. Our global factory network and supplier ecosystem directly enable a more adaptable and resilient supply of products. Over the past year, as substrates were constrained across the industry, our ability to leverage internal capacity created more than $2 billion in revenue upside for Intel, allowing us to respond with agility to meet dynamic customer demand.”

–Keyvan Esfarjani, Intel executive vice president and chief global operations officer

Source: https://www.intel.com/content/www/u...ctory-helps-overcome-substrate-shortages.html


My thoughts:

Gauging Intel's 2021 revenue growth against other major semiconductor companies, it doesn't show that Intel's IDM model is superior. Most major semiconductor companies can grab the opportunity to grow their revenue by double digits percentage in 2021. But in the same year Intel can only increase their revenue by 1% under their supposedly excellent IDM capabilities.

Not that I'm expecting this to prove Intel's case (I suspect it'll show the opposite) but I think margin might more accurately show whether IDM has the expected benefits. The main reasons (I think) Intel would insource would be to make more $ with what they already have, rather than try to focus on capacity only.

There are often reasons to eschew topline growth only, such as abandoning the lower end of the market for a time when it might cheapen the brand image or when you simply don't have enough capacity to service the entire market. AMD, for example, seems to be employing this strategy with their CPUs..
 
Daniel - I thought the "first TSMC half node" was much further back -- example: TSMC 55nm, inbetween 65nm and 45nm. TSMC had all three 'nodes', while other companies like Intel and AMD/Glofo jumped from 65nm to 45nm..? (I also remember foundry 80nm and 110nm 'half nodes')..

(I really like the "yield learning process steps" quote here).

55nm was a shrink (die shrink, optical shrink, process shrink). Half nodes in my experience involve more process improvements which means you have to do design modifications.
 
Not that I'm expecting this to prove Intel's case (I suspect it'll show the opposite) but I think margin might more accurately show whether IDM has the expected benefits. The main reasons (I think) Intel would insource would be to make more $ with what they already have, rather than try to focus on capacity only.

There are often reasons to eschew topline growth only, such as abandoning the lower end of the market for a time when it might cheapen the brand image or when you simply don't have enough capacity to service the entire market. AMD, for example, seems to be employing this strategy with their CPUs..

I remember when Intel first claimed IDM dominance:


10 years later it is a whole new ball game.
 
I remember when Intel first claimed IDM dominance:


10 years later it is a whole new ball game.
LOL - echoes of "Real Men Have Fabs"
 
Intel did process improvements 14nm, 14+, 14++, 14+++ etc.... but did not market them as half nodes. We used to do shrinks between nodes but they were really only for the high volume chips. TSMC and Apple were really behind the half node marketing strategy. Apple came to TSMC at 20nm then 16nm which was 20nm with FinFETs using the same fabs. That was the first TSMC half node. Then N10, N7, N7+, N6, N5, N4, and N3. We call them half nodes but what they really are is yield learning process steps.

TSMC will be talking about their new N2 node at the Technical Symposium this month. That will be a new node for sure followed by many enhancements (half nodes). Semiconductors have never been more exciting, absolutely.
Intel really messed up the marketing badly on the process shrink naming here (dug themselves into the "stuck at 14nm" hole). For supposed marketing wizards, they've done some pretty dumb stuff.
 
Intel's IDM mentality is killing themselves. Intel's product development and innovation are dictated and planned on Intel own inhouse manufacturing capabilities. Until the Arc Alchemist GPU manufacturerd by TSMC, there wasn't any major Intel products outsourced to outside foundries (Excluding those acquired companies).

The negative impacts can be seen clearly on the Intel's revenue growth as described in the table I posted at the beginning of this thread. When every major semiconductor company reached double digits growth in the boom time 2021, Intel struggled to reach only 1% YoY growth. Then again in Q1 2022, Intel retreated to -1% Non-GAAP or -7% GAAP YoY revenue growth while all other major semiconductor companies are showing double digits growth.

Intel's number one problem is not in their manufacturing capabilities. Instead it's in Intel's product offerings. Intel doesn't have the right products for the market.

Intel's IDM thinking has limited Intel to choose products towards those can be made inhouse instead what the market really wants (and it might need to be made by non Intel foundries)
 
The problem with IDM is the ingrown relationship. Like running a 3-legged race while everyone else sprints. When you have that closed world, crazy things get accepted as sensible. Things like taking 4 years (best case) from feature lock to product, because the feature waits for the process, and the process has to wait for everyone to get on board, and then it needs to be copy exact. God help you if you stumble. Oh, and 4 years from now your customers totally forgot what they asked you and had to go do something else, because real customers wanted something ages ago, and then they never get around to implementing support for half the new features because they are by then irrelevant. But, you will be doomed to backward support for all those obscure features, including the outright mistakes, because each feature is used by some 10% of your customers. Sound familiar? Oh, you've used Intel chips, huh? Good thing you spend, what, 20% of your engineering budget on ensuring compatibility ! That sure helps be competitive. And does no end of good for agility. Right.

Whereas in the rest of the industry a new process comes online every year. Fabs are all different and they all work, because the fabs figured out that every year they need something shiny and new and hey, once you get used to it you can build a fab that works by just putting in features that are proven and rescheduling everything not yet ready to the next fab. There is always a next fab. And a good thing too, because your top 10 customers will never want the same process so copy exact is a mirage. Flexibility becomes your culture.

I worry about Intel. So many signs they think IDM was great. 20 years ago it may have been the best idea. Now, it is a tired and fragile model that ignores the strengths that the independent fab model has forced others to grow. Even their own projects go to outside fabs, so one hopes they will figure it out. A first class fab needs a diverse set of customers to teach it to fire on all cylinders. Customers that do not think the fab defines the products. Fabs that skate where the customers want the puck to be.
 
While I generally agree with comments about IDM being a significant risk model in the modern fab era, Intels revenue changes are largely due to engineering products that can't compete well with AMD in the data center. Even if the latest. (Ice Lake, Sapphire Rapids) Xeons were fabbed at TSMC 7nm like AMDs, those products would still fall significantly behind in scalability and performance.

Also for Intel the high margin corporate laptop market is cooler lately after big business bought lots of PCs to enable increases WFH during the pandemic. That trend is now reversing.

Unless I'm missing something, this lack of revenue gain in 21 is likely not mostly due to IDM?
 
While I generally agree with comments about IDM being a significant risk model in the modern fab era, Intels revenue changes are largely due to engineering products that can't compete well with AMD in the data center. Even if the latest. (Ice Lake, Sapphire Rapids) Xeons were fabbed at TSMC 7nm like AMDs, those products would still fall significantly behind in scalability and performance.

Also for Intel the high margin corporate laptop market is cooler lately after big business bought lots of PCs to enable increases WFH during the pandemic. That trend is now reversing.

Unless I'm missing something, this lack of revenue gain in 21 is likely not mostly due to IDM?


With only few exceptions, IDM business model traps Intel to sell those products they can make in-house, instead of offering products the market needs (no matter they will be made in-house or by outside foundries).

So when Semiconductor industry is experiencing a historical boom time, Intel can't grab the golden opportunity except blaming supply chains and the cooling down PC demand.

Every major semiconductor company was/is doing very well in 2021 and Q1 2022, except Intel. The root cause is Intel IDM business model and its impact on Intel's product development and selection.

Intel and Pat Gelsinger are trying to use Intel Foundry Service (IFS) to salvage their IDM model. But it doesn't solve what Intel is lacking, the good and right products.

A semiconductor company makes money by selling the good and right products to people but do people care who really "manufactured" those products?

A successful semiconductor product company, with or without a fab, must let products and market to dictate the company's directions.

Every major semiconductor product company has figured it out, except Intel.
 
Last edited:
With only few exceptions, IDM business model traps Intel to sell those products they can make in-house, instead of offering products the market needs (no matter they will be made in-house or by outside foundries).

So when Semiconductor industry is experiencing a historical boom time, Intel can't grab the golden opportunity except blaming supply chains and the cooling down PC demand.

Every major semiconductor company was/is doing very well in 2021 and Q1 2022, except Intel. The root cause is Intel IDM business model and its impact on Intel's product development and selection.

Intel and Pat Gelsinger are trying to use Intel Foundry Service (IFS) to salvage their IDM model. But it doesn't solve what Intel is lacking, the good and right products.

A semiconductor company makes money by selling the good and right products to people but do people care who really "manufactured" those products?

A successful semiconductor product company, with or without a fab, must let products and market to dictate the company's directions.

Every major semiconductor product company has figured it out, except Intel.

Thought experiment -- if Intel had kept it's node guidance going and was still ahead of TSMC, would we still say that IDM has broken down and has failed Intel?

There's been a few interviews with Jim Keller in recent years that pointed out that (the company of Gordon Moore) morale was at an all time low as Intel engineers were convinced that scaling was dead.. Jim explained working to re-energize the team by challenging them to come up with methods for scaling 100x vs today, and the team got to 50x within a week after stopping to think about it deeply. (Anandtech and Lex Fridman each covered some of this content if you're curious). Intel's advanced nodes always allowed them some opportunity to salvage situations when they were significantly behind AMD, such as the K7-K8 eras.

re: Foundry Scaling -- Recently, AMD could have actually profited a lot more last year also if they had even more of TSMC's capacity for their chips. They went a whole year without really trying to sell any CPUs below $300 (Ryzen 5000 series) on the desktop for example. Foundries also take time to ramp up capacity for customers..

Intel did rode the cloud consolidation of the 2010s wave very well, with quarterly revenue going from $38-40B in early 2010 to $75-78B in early 2020. That's when the revenue model 'broke' because they fell behind on engineering new nodes and not keeping up with competition overall. If they were outsourcing fabrication that entire time, I'm not sure what the evidence is they would have done any better. By comparison, AMD quarterly revenue rose only 35% (vs ~ 90%) between 1H 2010 ($5.4-5.8B) and 1H 2020 ($7.2-$7.5B in 1H 2020), and this is with AMD having toes in two big markets (CPU and GPU) and multiple foundries to leverage. The failure of the last couple of years for Intel seems to be increased competition, and a corporate culture (thanks BK..) that didn't support innovation. IDM may make that a little worse on the down side, but IDM looked pretty good on the upswing last decade..

FWIW, I personally like Intel's current path of doing both IDM and outsource to reduce risk...

TL;DR - IDM has pros and cons but it was the company culture (not IDM specifically) that caused Intel to flatten revenue (and lose margins) over the past few years vs. competition.
 
Thought experiment -- if Intel had kept it's node guidance going and was still ahead of TSMC, would we still say that IDM has broken down and has failed Intel?

There's been a few interviews with Jim Keller in recent years that pointed out that (the company of Gordon Moore) morale was at an all time low as Intel engineers were convinced that scaling was dead.. Jim explained working to re-energize the team by challenging them to come up with methods for scaling 100x vs today, and the team got to 50x within a week after stopping to think about it deeply. (Anandtech and Lex Fridman each covered some of this content if you're curious). Intel's advanced nodes always allowed them some opportunity to salvage situations when they were significantly behind AMD, such as the K7-K8 eras.

re: Foundry Scaling -- Recently, AMD could have actually profited a lot more last year also if they had even more of TSMC's capacity for their chips. They went a whole year without really trying to sell any CPUs below $300 (Ryzen 5000 series) on the desktop for example. Foundries also take time to ramp up capacity for customers..

Intel did rode the cloud consolidation of the 2010s wave very well, with quarterly revenue going from $38-40B in early 2010 to $75-78B in early 2020. That's when the revenue model 'broke' because they fell behind on engineering new nodes and not keeping up with competition overall. If they were outsourcing fabrication that entire time, I'm not sure what the evidence is they would have done any better. By comparison, AMD quarterly revenue rose only 35% (vs ~ 90%) between 1H 2010 ($5.4-5.8B) and 1H 2020 ($7.2-$7.5B in 1H 2020), and this is with AMD having toes in two big markets (CPU and GPU) and multiple foundries to leverage. The failure of the last couple of years for Intel seems to be increased competition, and a corporate culture (thanks BK..) that didn't support innovation. IDM may make that a little worse on the down side, but IDM looked pretty good on the upswing last decade..

FWIW, I personally like Intel's current path of doing both IDM and outsource to reduce risk...

TL;DR - IDM has pros and cons but it was the company culture (not IDM specifically) that caused Intel to flatten revenue (and lose margins) over the past few years vs. competition.

”Thought experiment -- if Intel had kept it's node guidance going and was still ahead of TSMC, would we still say that IDM has broken down and has failed Intel?“

I used to think if Intel can keep the leadership in manufacturing then Intel will be the king of the semiconductor industry. Now I don't think this is the real world scenario anymore.

Looking at 2021 full year and Q1 2022 revenue growth from all major semiconductor companies, Intel is the only one in a miserable situation. Why so?

Those companies are in all kinds of technology categories and market segments. Some of them have fabs but many don't. Some of them have advanced products using advanced manufacturing process but many of them don't. Some of them are big companies but many of them are smaller than Intel.

They all share a common successful trick: to sell the right products the market needs at the right time with the right price. The double digits YoY growth became an industrial standard phenomenon for those semiconductor companies while Intel is struggling to avoid negative growth. All good reasons provided by Intel became meaningless once you look what the whole semiconductor industry is experiencing and doing.

I believe Intel's number one problem is that they don't have the right products that market needs. To break that, Intel must consider what market needs instead which "Intel fab" can make it. It's a big shift from the IDM culture.

Pat Gelsinger's stated goal is to regain Intel manufacturing leadership again around 2025. I'm wondering without taking care of Intel product offerings and meeting the market demand, does that manufacturing leadership really matter?
 
Last edited:
”Thought experiment -- if Intel had kept it's node guidance going and was still ahead of TSMC, would we still say that IDM has broken down and has failed Intel?“

I used to think if Intel can keep the leadership in manufacturing then Intel will be the king of the semiconductor industry. Now I don't think this is the real world scenario anymore.

Looking at 2021 full year and Q1 2022 revenue growth from all major semiconductor companies, Intel is the only one in a miserable situation. Why so?

Those companies are in all kinds of technology categories and market segments. Some of them have fabs but many don't. Some of them have advanced products using advanced manufacturing process but many of them don't. Some of them are big companies but many of them are smaller than Intel.

They all share a common successful trick: to sell the right products the market needs at the right time with the right price. The double digits YoY growth became an industrial standard phenomenon for those semiconductor companies while Intel is struggling to avoid negative growth. All good reasons provided by Intel became meaningless once you look what the whole semiconductor industry is experiencing and doing.

I believe Intel's number one problem is that they don't have the right products that market needs. To break that, Intel must consider what market needs instead which "Intel fab" can make it. It's a big shift from the IDM culture.

Pat Gelsinger's stated goal is to regain Intel manufacturing leadership again around 2025. I'm wondering without taking care of Intel product offerings and meeting the market demand, does that manufacturing leadership really matter?
I agree with all of your points. The lack of a leading edge fab process is one thing, but uncompetitive product designs are a worse problem. A less than competitive design in a leading edge process is not going to cut it anymore. Intel is behind AMD in x86 CPU design, behind Nvidia in GPU design, behind Nvidia and Broadcom in networking chip design, Optane has not met expectations, and they're even behind Ampere and Amazon for cloud CPU designs. Personally, I can't consider Intel investable until I see some product leadership thinking from them. The only place I can see them truly dominating is the invention of new I/O industry specifications. USB, PCIe, CXL, and UCIe were all created in-house at Intel. Now if they could only use that leadership to create leadership products. I'm not hopeful.
 
Last edited:
“This initiative is a terrific example of how integrated manufacturing is foundational to Intel’s success. Our global factory network and supplier ecosystem directly enable a more adaptable and resilient supply of products. Over the past year, as substrates were constrained across the industry, our ability to leverage internal capacity created more than $2 billion in revenue upside for Intel, allowing us to respond with agility to meet dynamic customer demand.”

–Keyvan Esfarjani, Intel executive vice president and chief global operations officer

Source: https://www.intel.com/content/www/u...ctory-helps-overcome-substrate-shortages.html
Here I think "integrated manufacturing" just means "we have fabs". I don't think he's defending IDM 2.0 or IDM generally.

"Resilient supply" This is a big advantage. TSMC has multiple fabs but not globally distributed (not yet, AZ/JP fabs will start to change this).

Geopolitics is important now. Your neighborhood is important now. This is a disadvantage for TSMC.

"Respond with agility" As hist78 points out 1% revenue growth paints a different picture than "agility".

Some basic comparisons might apply here. TSMC has 56,000 employees, Intel 121,100. TSMC 2021 revenue was $54B, Intel 2021 revenue was $79B. Dividing revenue by employee, TSMC is at 964,285 per employee, Intel is at 652,353. So, you can say TSMC is more agile by hundreds of thousands of dollars per employee. This seems like a big problem for Intel, and a big advantage for TSMC.
 
Here I think "integrated manufacturing" just means "we have fabs". I don't think he's defending IDM 2.0 or IDM generally.

"Resilient supply" This is a big advantage. TSMC has multiple fabs but not globally distributed (not yet, AZ/JP fabs will start to change this).

Geopolitics is important now. Your neighborhood is important now. This is a disadvantage for TSMC.

"Respond with agility" As hist78 points out 1% revenue growth paints a different picture than "agility".

Some basic comparisons might apply here. TSMC has 56,000 employees, Intel 121,100. TSMC 2021 revenue was $54B, Intel 2021 revenue was $79B. Dividing revenue by employee, TSMC is at 964,285 per employee, Intel is at 652,353. So, you can say TSMC is more agile by hundreds of thousands of dollars per employee. This seems like a big problem for Intel, and a big advantage for TSMC.
Is revenue per employee the correct metric for "agility" ?

I'm not sure you can directly compare revenue per employee for Intel and TSMC as Intel has a large proportion of it's employees in design and associated sales and marketing and is running a diverse set of business and not just one large one like TSMC.

My impression from things I've read is that the effective number of hours worked per week in TSMC (Taiwan) is significantly higher than at Intel (US). So we might also need to adjust for that.

$650K sales per employee at Intel doesn't sound like a "problem" to me. Especially when there is also a large profit margin.

You are forcing me into the unfamiliar position of defending Intel !
 
Back
Top