Morris Chang talks about Intel's reliance on TSMC

[fansink]

Is Intel destined to follow Samsung’s downward spiral with single digit yields?

Morris Chang, the founder of TSMC, predicts that Intel will continue to rely on TSMC for fabricating their advanced chips.

Morris Chang Asserts Intel Foundry Will Remain in TSMC's Shadow

Founder of TSMC does not see Intel as a major threat.

www.tomshardware.com

 

[samwilde]

This is not the first time Intel tried dipping their toes into the foundry business.
Their last attempt was an utter failure. The company is just too focused on making a process to make their own CPUs.

It is also debatable if companies like AMD or NVIDIA would make chips at a fab owned by someone who is perceived to be their competitor.

 

[hist78]

In Intel destined to follow Samsung’s downward spiral with single digit yields?

Morris Chang, the founder of TSMC, predicts that Intel will continue to rely on TSMC for fabricating their advanced chips.

Morris Chang Asserts Intel Foundry Will Remain in TSMC's Shadow

Founder of TSMC does not see Intel as a major threat.

www.tomshardware.com

This article was published in October 2023 and there are some serious translation errors. For example, Mr. Morris Chang originally said Intel's advancement in foundry business will give TSMC some challenges (like the shadow created by the passing cloud). Somehow that Tom's Hardware writer Anton Shilov mistakenly translated it to "Intel Foundry Will Remain in TSMC's Shadow"!! It's very funny and unprofessional.

 

[fansink]

This article was published in October 2023 and there are some serious translation errors. For example, Mr. Morris Chang originally said Intel's advancement in foundry business will give TSMC some challenges (like the shadow created by the passing cloud). Somehow that Tom's Hardware writer Anton Shilov mistakenly translated it to "Intel Foundry Will Remain in TSMC's Shadow"!! It's very funny and unprofessional.

Weird, this article was in my date sorted news results, stating it was 16 hours old.

However, my question still remains:

Is Intel destined to follow Samsung’s downward spiral with single digit yields?

 

[hist78]

Weird, this article was in my date sorted news results, stating it was 16 hours old.

However, my question still remains:

Is Intel destined to follow Samsung’s downward spiral with single digit yields?

According to Intel, Intel 3 has gone into high volume manufacturing so I assume its yield is acceptable at this stage. But I think its outdated business model and lacking attractive and high demand products are more problematic to Intel. The later is especially critical to Intel's revival and survival.

Unfortunately Intel's leadership and board of directors are still believing inhouse manufacturing is the most important thing that Intel needs to pursue. They ignore the fact that most of those Intel's strong competitors do not own a fab at all. Pat Gelsinger is betting $100 billion to prove the whole industry has been wrong for the past 30 years.

 

[fansink]

According to Intel, Intel 3 has gone into high volume manufacturing so I assume its yield is acceptable at this stage. But I think its outdated business model and lacking attractive and high demand products are more problematic to Intel. The later is especially critical to Intel's revival and survival.

Unfortunately Intel's leadership and boar of directors are still believing inhouse manufacturing is the most important thing that Intel needs to pursue. They ignore the fact that most of those Intel's strong competitors do not own a fab at all. Pat Gelsinger is betting $100 billion to prove the whole industry has been wrong for the past 30 years.

The undeniable truth is, success at the bleeding edge of this art, may ultimately be limited to the few, or the one.

All else, may be destined to remain in the realm of commodity.

I suspect 2nm will separate the mice from the men.

 

[Daniel Nenni]

However, my question still remains:
Is Intel destined to follow Samsung’s downward spiral with single digit yields?

No, definitely not. Very different cultures, yield is a priority for Intel. I do not see Intel getting better cost or better yield than TSMC but they will not be too far off. Even if Samsung gets better at yield who will believe them?

Intel may pass TSMC on paper for performance but TSMC will always be first with customers. From what I see inside the ecosystem, TSMC N2 will again dominate design starts next year in the same manner N3 did.

 

[Maximus]

According to Intel, Intel 3 has gone into high volume manufacturing so I assume its yield is acceptable at this stage. But I think its outdated business model and lacking attractive and high demand products are more problematic to Intel. The later is especially critical to Intel's revival and survival.

Unfortunately Intel's leadership and board of directors are still believing inhouse manufacturing is the most important thing that Intel needs to pursue. They ignore the fact that most of those Intel's strong competitors do not own a fab at all. Pat Gelsinger is betting $100 billion to prove the whole industry has been wrong for the past 30 years.

Intel CFO admitted they will not get big orders for 18A that 18A is for building customers' confidence in Intel.
The next major node for Intel is 14A: Targeting to come out in 2027. Usually customers engagement start from 2-3 years before a node comes out.
That said if Intel do not catch a whale in 2025 for 14A, then it seems tsmc will dominate again in A14.

 

[hist78]

Intel CFO admitted they will not get big orders for 18A that 18A is for building customers' confidence in Intel.
The next major node for Intel is 14A: Targeting to come out in 2027. Usually customers engagement start from 2-3 years before a node comes out.
That said if Intel do not catch a whale in 2025 for 14A, then it seems tsmc will dominate again in A14.

That's a very risky situation for Intel to wait for 14A. It will be too late and too little to save the Intel even if a "whale' is willing to adopt 14A in 2027.

IMO Intel should use TSMC as much as possible to offer competitive products because customers won't wait and Intel's competitors won't sit idle either. Intel Foundry is a nice thing to have but Intel products are more important, no matter using inhouse manufacturing capability or external foundries.

 

[Xebec]

That's a very risky situation for Intel to wait for 14A. It will be too late and too little to save the Intel even if a "whale' is willing to adopt 14A in 2027.

IMO Intel should use TSMC as much as possible to offer competitive products because customers won't wait and Intel's competitors won't sit idle either. Intel Foundry is a nice thing to have but Intel products are more important, no matter using inhouse manufacturing capability or external foundries.

I think they’re in the situation where they really have no choice but to wait for 14A to be successful (or not) with whales.

They had a few failed Foundry attempts before. The Tower option fell through which would have added some buzz of credibility. Intel 3 has higher transistor performance than TSMC N3, but the confidence to get it to yield on time would have been low because of recent Intel history (10nm). They need a few years of solid performance to show they’re back (Intel 3 should be doing that this year, and they’re betting on 18A in 2025-2026), and then they’ll get business with whatever is cutting edge next.

That said, Intel has a solid story of “we’re betting our own company on these products (18A, 14A)”, which shows how serious they are to potential customers. Intel also has some pretty impressive packaging tech, apparently offering a few tricks TSMC can’t (like packaging chips from other fabs, and some benefits of Foveros, not that TSMC doesn’t have some advantages too here).

Intel is running a lot of volume through TSMC right now that can be shifted back to Intel, and they have much more competitive products in the process of release now (Sierra Forest, Lunar Lake, Arrow Lake) which should help buy some time.so. I think 14A is their real chance given the increasing cost curve of new fab tech.

..

Intel and TSMC are competing but they’re also not/never will be the same. All things being equal, Intel doesn’t need to match TSMC’s capacity to make the same amount of money as Intel is also selling internal chips with margin additive on top of the fab margins. If they do it right they can capture the margin of “TSMC + AMD” on a CPU sold for example.

If Intel fails to sell enough chips to justify foundry, then we’re eventually going to end up with a monopoly on cutting edge tech.. which stinks for a lot of reasons.

 

[Daniel Nenni]

Intel CFO admitted they will not get big orders for 18A that 18A is for building customers' confidence in Intel.
The next major node for Intel is 14A: Targeting to come out in 2027. Usually customers engagement start from 2-3 years before a node comes out.
That said if Intel do not catch a whale in 2025 for 14A, then it seems tsmc will dominate again in A14.

Yes, now Intel admits that 18A will not get a whale customer but that is a different story from when Pat joined Intel, right? Why did big customers not use Intel 18A? And just how does 18A build confidence when big customers will not use it?

One of the challenges of being a leading edge foundry is delivering wafers on spec (PPA), on time, and with predictable yield. Intel and Samsung really need to work on this. TSMC has set the bar high for big customers.

The other thing you have to consider is the ecosystem. Intel 3 and 18A are false foundry starts. EDA and IP companies put a lot of effort into getting tools and IP ported over. Some were paid a lot of money (Synopsys/Arm/Cadence) but others were on their own dime. Intel Foundry is now in the same pool as Samsung where the customers will have to request IP porting, which is time consuming, and the cost will be built into the IP price. Meanwhile TSMC is the foundry where IP companies go first for development and silicon validation. So a customer must really NOT want to do business with TSMC to port over to Intel or Samsung, absolutely.

 

[MKWVentures]

Couple thoughts (I have detailed data on these):
1) Intel is a major customer of TSMC. Not a competitor. Intel is not a top 15 foundry as of today. Someday that might change. Today is not that day.
2) Intel production (going into real products) wafer starts on Intel 3,4,20A, 18A are incredibly low and dont ramp significantly until end of 2025
3) Intel will succeed in foundry when : they can get the process cost to below 150% of TSMC, get a couple low volume products from customers working, and then execute as well as TSMC or have lower price in order to get major products. THEN Intel will get major products and ramp.

CFO is very clear on these in his presentations. Pat is a more aspirational in his presentations (Which is fine)

 

[Daniel Nenni]

Remember, Morris Chang retired from TSMC in 2018. CC Wei did the Intel deal with Bob Swan.

 

[siliconbruh999]

Imo Intel 18A will tell whether they can deliver on What they say and if they can 14A can be a big winner so 18A is make or break point

 

[MKWVentures]

Imo Intel 18A will tell whether they can deliver on What they say and if they can 14A can be a big winner so 18A is make or break point

20A should give us earliest look. It was committed to Arrow Lake with a end of year launch. Lets see what actually happens. it might be the most political thing to happen in November LOL

 

[siliconbruh999]

20A should give us earliest look. It was committed to Arrow Lake with a end of year launch. Lets see what actually happens. it might be the most political thing to happen in November LOL

If we follow Intel's Release cadence ARL is in Q4 24 which is for K parts and most probably 20A is going to be 6+8 die rather than the 8+16 die N3B for top SKU and the lower non K sku are released in Jan Next year that's how they launch products

 

[MKWVentures]

If we follow Intel's Release cadence ARL is in Q4 24 which is for K parts and most probably 20A is going to be 6+8 die rather than the 8+16 die N3B for top SKU and the lower non K sku are released in Jan Next year that's how they launch products

100% agree. There may be fallout if the releases pan out like you stated. and we need to use the term launch which means an end product is available to end customers. Some people are predicting a paper launch and cannon lake like release (trivial volume) on 20A product. There are a lot of conflicting messages.

 

[siliconbruh999]

100% agree. There may be fallout if the releases pan out like you stated. and we need to use the term launch which means an end product is available to end customers. Some people are predicting a paper launch and cannon lake like release (trivial volume) on 20A product. There are a lot of conflicting messages.

I don't think there is a fallout for the usual schedule if 20A products are available in January for non-k sku it is their regular cadence so I think it will be a pass but if it's Feb yeah not good I am more interested in comparing 6+8 Intel 20A die vs 6+8 N3B die it will also give a hint of the PPA for the nodes

 

[Maximus]

Remember, Morris Chang retired from TSMC in 2018. CC Wei did the Intel deal with Bob Swan.

Morris did gave his opinions to Mark Liu about Intel's deal that he mentioned in a speech on Dec 06, 2021:
"Morris Chang said that Intel approached TSMC to make high-end products last year (2020). Chairman Mark Liu was excited about this, but he thought about it and asked Mark Liu not to get excited too early. Because technology and manufacturing are the soul of Intel, the case may not be successful. Sure enough, Intel disagreed and asked Intel's CEO at the time to leave and find Pat to come back and take over."

英特爾CEO講台積電壞話 張忠謀看衰反嗆 - 自由財經

〔記者溫于德／台北報導〕司法院今天下午舉辦「趨勢講堂」，台積電創辦人張忠謀應邀分享「珍惜台灣半導體晶圓製造的優勢」，過程中，張忠謀被大法官張森林問到英特爾（intel）CEO季辛格（Pat Ge

ec.ltn.com.tw

----
Intel still made a deal with tsmc on N3 but the volume shrink.

 

[siliconbruh999]

Morris did gave his opinions to Mark Liu about Intel's deal that he mentioned in a speech on Dec 06, 2021:
"Morris Chang said that Intel approached TSMC to make high-end products last year (2020). Chairman Mark Liu was excited about this, but he thought about it and asked Mark Liu not to get excited too early. Because technology and manufacturing are the soul of Intel, the case may not be successful. Sure enough, Intel disagreed and asked Intel's CEO at the time to leave and find Pat to come back and take over."

英特爾CEO講台積電壞話 張忠謀看衰反嗆 - 自由財經

〔記者溫于德／台北報導〕司法院今天下午舉辦「趨勢講堂」，台積電創辦人張忠謀應邀分享「珍惜台灣半導體晶圓製造的優勢」，過程中，張忠謀被大法官張森林問到英特爾（intel）CEO季辛格（Pat Ge

ec.ltn.com.tw

----
Intel still made a deal with tsmc on N3 but the volume shrink.

He is not wrong though they have been manufacturing focused since their inception they should have invested in Fabs years ago after the 14nm slip they paid for it dearly with 10nm