Morris Chang: Taiwan’s semiconductor advantage will not exist in 20 to 30 years

[tonyget]

【台積電運動會3】張忠謀：台灣半導體優勢 2、30年後不會有

台積電創辦人張忠謀認為，台灣的半導體產業優勢，在過2、30年後，也不會那麼有利，這跟經濟發展有關。

tw.news.yahoo.com

TSMC founder Morris Chang believes that Taiwan’s semiconductor industry advantages will not be so beneficial in 20 to 30 years. This is related to economic development.

TSMC founder Morris Chang said that the U.S. semiconductor manufacturing environment from 1955 to 1973 was the same as Taiwan today. However, Taiwan will also change, and Taiwan’s semiconductor environment will not be so favorable in 20 to 30 years. This is related to the country's economic development. Although the United States has lost most of its semiconductor manufacturing, it has upgraded to an IC design industry that does not require a lot of capital. This is a high-end industry. Companies such as nVidia, Apple, Google, and Microsoft are all developing. In the IC design industry, these companies make a lot of money.

Morris Chang said that TSMC’s advantage is that it uses machinery and equipment every minute, because these equipment are too expensive. Moreover, if the machinery and equipment breaks down at 12 o’clock in the middle of the night, call the engineers and the engineers will come to solve it immediately. There is still no problem in doing so now. , his wife doesn’t ask too much about him going out in the middle of the night and is used to it. However, he believes that in 20 or 30 years, Taiwanese engineers will no longer be on call 24 hours a day like this. As for who will replace Taiwan in 20 or 30 years? nation? Is it India, Vietnam or Indonesia? Morris Chang said: "Who Knows."

However, Morris Chang believes that Japan is an ideal country for the development of semiconductors because when he was working at Texas Instruments in the United States 50 years ago, he went to Japan to set up a factory and found that the local resources such as land, water and electricity were quite abundant, and the work culture was also very good. Although Singapore It is also quite ideal and very good for fabs, but the land is too small and there are not enough resources such as water, electricity, and people.

 

[Barnsley]

And?

Is this supposed to be some farsighted view that nobody else can see?

 

[blueone]

Morris would do his legacy a lot of good if he stopped making public statements.

 

[hist78]

And?

Is this supposed to be some farsighted view that nobody else can see?

Morris would do his legacy a lot of good if he stopped making public statements.

You can skip this news article altogether. This particular article, written in Chinese and translated into English, is a low quality piece with lots of omissions and mistranslation. It quoted the Q&A out of context in several places.

@Daniel has shared a better written article published by Nikkei.

Post in thread 'Morris Chang on TSMC Arizona, An Exercise in Futility' https://semiwiki.com/forum/index.ph...zona-an-exercise-in-futility.18957/post-64784

 

[Daniel Nenni]

What Morris fails to mention is that semiconductors are critical to national security. That alone justifies the onshoring money required for semiconductor design and manufacture, the whole ecosystem. As they say, plan for the worst, hope for the best. Given the violence around the globe that should be foremost on politician's minds, right? Other than just getting re elected.

 

[hist78]

What Morris fails to mention is that semiconductors are critical to national security. That alone justifies the onshoring money required for semiconductor design and manufacture, the whole ecosystem. As they say, plan for the worst, hope for the best. Given the violence around the globe that should be foremost on politician's minds, right? Other than just getting re elected.

He did. He spoke in serious tone about national security is now the top priority for counties, more important than globalization and free trade.

 

[Daniel Nenni]

He did. He spoke in serious tone about national security is now the top priority for counties, more important than globalization and free trade.

Yes, here is one:

TSMC founder concerned about Taiwan semiconductor future | Taiwan News | Oct. 14, 2023 13:47

Morris Chang names Kyushu, Singapore as potential future chip centers | Oct. 14, 2023 13:47

www.taiwannews.com.tw

But he is still focused on cost and work force.

 

[VCT]

Each person's definition of successful is different.
If IFS make small loss or small profit with huge revenue in the next 20 year it will be considered successful to US government.
But not considered successful to shareholder and Intel management.

For Morris Chang, his bar is high.

 

[blueone]

The only way IFS can be a success in the long run is to have high revenue and profit. However, the only example of that in the foundry industry is TSMC, and TSMC did not achieve high revenue and profit in a couple of years. Or even five years. The first ten years of TSMC operations did not result in fabulous growth either. I think it's unrealistic to expect IFS to sprint to tens of billions of dollars in high margin external foundry revenue in a few years. IFS success will likely be more of a marathon than a sprint.

 

[nghanayem]

The only way IFS can be a success in the long run is to have high revenue and profit.

I guess I think about it in a different way. Let's say intel fabrication group (or whatever the external name they give that BU is) can get to break even and with enough revenue to be nearly cashflow neutral. In my mind that would be the lowest hurdle to clear before IFS can be viewed as successful. Even a mere 1% op margin would mean that Intel corp is seeing cost and margin upside over being a pure fabless company. Combine this with the peace of mind that intel doesn't have to rely on the execution of a 3rd party, the benefits of being their fab's number one priority rather than playing second or third fiddle to Apple, and owning intel fabrication group would only be to the wider intel corp's benefit.

And I know, I know. 1% op margin with net neutral cashflows is not great for any standalone business. But if the intention of IDM2.0 is to make intel the IDM stronger, then I think that this is all that is needed for IFS to be a good success story for intel. In this scenario intel would be enjoying all of the benefits of being an IDM and if the external business snags enough revenue that they don't even need to invest very much cash from the design side of the business then they are also seeing none of the downsides of being an IDM.

However, the only example of that in the foundry industry is TSMC, and TSMC did not achieve high revenue and profit in a couple of years. Or even five years. The first ten years of TSMC operations did not result in fabulous growth either. I think it's unrealistic to expect IFS to sprint to tens of billions of dollars in high margin external foundry revenue in a few years. IFS success will likely be more of a marathon than a sprint.

Besides being an upstart and that nobody in the foundry world besides TSMC has high margins, there are other factors that will act as margin headwinds for intel fabrication group. The largest one long term is what they do vs what TSMC doesn't do. Namely being an OSAT. The OSAT business operates on razor thin margins. So even if the wafer fab side had equal revenue, COS, and COGS to TSMC, intel manufacturing would have substantially lower margins on account of the packaging and test services they provide. Another major issue that cannot really be solved in the short/medium term is node age. If one looks at TSMC revenues half of it comes from the vintage stuff being made on depreciated tooling. Given the new nodes compress TSMC's margins, I think it is safe to assume the majority of the profits come from the trailing edge side of the business. For example let's take the early phases of TSMC's fab18. The tools still won't be fully deprecated until the 2025/N2 timeframe. In the case of the later N3 phases they won't be fully depreciated until the 2028/29 timeframe when we are talking about N1.5 HVM or even N1 ramp. Intel on the other hand all the way back in 2021 mentioned that the 10nm ramp passed 14nm starts as those products were getting phased out of production. Heck on the CPU side I am pretty sure almost all of the 14nm products have already been EOLd. Basically the only 14nm parts you can find these days are in ultra budget celerons for chromebooks. In the foundry world and with the adoption of disaggregated design intel will be able to milk their old technology and tools for longer.

 

[blueone]

I guess I think about it in a different way. Let's say intel fabrication group (or whatever the external name they give that BU is) can get to break even and with enough revenue to be nearly cashflow neutral. In my mind that would be the lowest hurdle to clear before IFS can be viewed as successful. Even a mere 1% op margin would mean that Intel corp is seeing cost and margin upside over being a pure fabless company. Combine this with the peace of mind that intel doesn't have to rely on the execution of a 3rd party, the benefits of being their fab's number one priority rather than playing second or third fiddle to Apple, and owning intel fabrication group would only be to the wider intel corp's benefit.

And I know, I know. 1% op margin with net neutral cashflows is not great for any standalone business. But if the intention of IDM2.0 is to make intel the IDM stronger, then I think that this is all that is needed for IFS to be a good success story for intel. In this scenario intel would be enjoying all of the benefits of being an IDM and if the external business snags enough revenue that they don't even need to invest very much cash from the design side of the business then they are also seeing none of the downsides of being an IDM.

Being a publicly-traded company, Intel is subject to the same cost of capital expectations as any other company in that category, and low net margins on such a capital-intensive business such as fabrication is unsustainable. It'll lower Intel's share price, increase interest costs on Intel debt, and may even attract an activist shareholder to get on the BoD and propose a different direction. I'm not suggesting that IFS has to have similar net margins to TSMC anytime soon, but Intel will have show the financial community a financial roadmap to getting competitively profitable, or Gelsinger and the board will get a lot of pressure improve performance or get replaced.

Besides being an upstart and that nobody in the foundry world besides TSMC has high margins, there are other factors that will act as margin headwinds for intel fabrication group. The largest one long term is what they do vs what TSMC doesn't do. Namely being an OSAT. The OSAT business operates on razor thin margins. So even if the wafer fab side had equal revenue, COS, and COGS to TSMC, intel manufacturing would have substantially lower margins on account of the packaging and test services they provide. Another major issue that cannot really be solved in the short/medium term is node age. If one looks at TSMC revenues half of it comes from the vintage stuff being made on depreciated tooling. Given the new nodes compress TSMC's margins, I think it is safe to assume the majority of the profits come from the trailing edge side of the business. For example let's take the early phases of TSMC's fab18. The tools still won't be fully deprecated until the 2025/N2 timeframe. In the case of the later N3 phases they won't be fully depreciated until the 2028/29 timeframe when we are talking about N1.5 HVM or even N1 ramp. Intel on the other hand all the way back in 2021 mentioned that the 10nm ramp passed 14nm starts as those products were getting phased out of production. Heck on the CPU side I am pretty sure almost all of the 14nm products have already been EOLd. Basically the only 14nm parts you can find these days are in ultra budget celerons for chromebooks. In the foundry world and with the adoption of disaggregated design intel will be able to milk their old technology and tools for longer.

I don't have enough industry expertise to comment on this.

 

[VCT]

And I know, I know. 1% op margin with net neutral cashflows is not great for any standalone business.

Agreed. 1% op margin with net neutral cashflows is not great for investors/shareholders but it will be quite an achievement.
People outside of this business don't know how difficult it is.

 

[Tanj]

Heck on the CPU side I am pretty sure almost all of the 14nm products have already been EOLd. Basically the only 14nm parts you can find these days are in ultra budget celerons for chromebooks.

Intel has a track record of producing support chips on the older processes. IO hubs, NICs, FPGAs, etc.

As it moves to an IFS model it may need to plan for customers who want long life assurance. A lot of government contracts require that, more than the 4 to 5 years Intel has traditionally offered for long life embedded or enterprise SKUs. But it may be able to manage that by excess production and holding inventory reserves, possibly cheaper than keeping old lines in use.

 

[Tanj]

If one looks at TSMC revenues half of it comes from the vintage stuff being made on depreciated tooling. Given the new nodes compress TSMC's margins, I think it is safe to assume the majority of the profits come from the trailing edge side of the business.

That much profit seems unlikely from old processes. Other foundries offer cheap alternatives, and the products themselves often only survive by being put into cheap products. A wafer at $3000 is not going to contribute more than the N-1 products in HVM with excellent yields at $15k per wafer. The $3000 wafers are also a distraction in OPEX like personnel, maintenance, power, consumables, even if the CAPEX is depreciated.

An Intel IFS will not have those distractions and may even go shopping for customers least likely to need long legacy support (other than excess inventory reserves). It does not need the same business model as TSMC, indeed better not.

 

[raikk6]

Morris has standard old Chinese people’s mindset. I am surprised that he has that since he got his education and started career in the States. Perhaps he worked in Taiwan for too long.
Truth is if the pay is high enough TSMC will get enough workforce willing to sacrifice work life balance in Taiwan. Problem is semiconductor manufacturing tech is slowing in advancement. Might be difficult to keep high profitability as China will catch up in the long run. So 20 years from now, TSMC might not be competitive enough staying in Taiwan.

 

[hist78]

Yes, here is one:

TSMC founder concerned about Taiwan semiconductor future | Taiwan News | Oct. 14, 2023 13:47

Morris Chang names Kyushu, Singapore as potential future chip centers | Oct. 14, 2023 13:47

www.taiwannews.com.tw

But he is still focused on cost and work force.

Morris has standard old Chinese people’s mindset. I am surprised that he has that since he got his education and started career in the States. Perhaps he worked in Taiwan for too long.
Truth is if the pay is high enough TSMC will get enough workforce willing to sacrifice work life balance in Taiwan. Problem is semiconductor manufacturing tech is slowing in advancement. Might be difficult to keep high profitability as China will catch up in the long run. So 20 years from now, TSMC might not be competitive enough staying in Taiwan.

Again, this is not what Morris said or what he tried to express during this informal Q&A with reporters. If you can understand Mandarin Chinese, here is a complete and unedited video recording:

Because it's in the quiet period before the quarterly earning conference, Morris declined to answer several questions he or TSMC CFO Wendell Huang feel that may have problems with the regulations or TSMC policy. This 30+ minutes Q&A session was held during the annual TSMC Sport Day, a companywide team building, fun games, and family festival day (with US$500 Sports Day bonus for each employee). With such backdrop Morris went through several questions in a more interesting, funny or sometimes sarcastic tone.

For example:

1. If he has any suggestions for TSMC.
Morris: I won't tell you (the reporter). (reporters laugh...)

2. If he has any suggestions for Intel.
Morris: No, I won't tell Intel. (reporters laugh and laugh...)

3. When will he publish his second half of his autobiography after releasing the first half 25 years ago.
Morris: The first half of the autobiography covered my career before starting TSMC and it was mush easier to write. I hope the second half will be completed in the Spring 2024.

4. Applying "Experience Curve" theory in semiconductor manufacturing by Morris (at TI) and Joseph Bannister (a consultant from Boston Consulting to work at TI and later the founder of Bann Consulting) works well and helps TSMC to improve quality, increase output, and reduce cost. A simple reason is that it's much easier to improve the semiconductor manufacturing quality and efficiency when you have a large production volume. But execution and R&D are very important to gain the true benefit of "Experience Curve".

5. Upon the request from Morris, Sophia Chang (Morris' wife) described how TSMC Charity Foundation helps disadvantage young adults to start careers in various companies.

6. Morris will give a speech at MIT in October regarding the semiconductor industry. Consequently Morris said he probably won't be able to meet Nvidia's Jensen Huang in Taiwan next week.

7. He felt so happy about the TSMC Sports Day after the three-year Covid-19 interruption. He pointed out the 6,000-meter relay helps people to recognize that every person in the team is critical and essential.

8. TSMC is much better prepared to build and operate Arizona fabs successfully than starting the WaferTech 27 years ago. It's because TSMC today is in a much stronger position in personnel, financial, and technology.

9. TSMC OIP program and Grand Alliance partnership are TSMC's powerful advantages and harder for competitors to duplicate.

10. The current war in Israel will have limited impact on Intel and global semiconductor supply chain.

11. First for US then for Taiwan, the semiconductor manufacturing has evolved along with the progress of economic devolvement. For US it has moved up to the less capital intensive yet extremely profitable fabless model (like Nvidia, Apple). Morris said in 20 to 30 years, Taiwan's superior position in semiconductor manufacturing environment may not be as strong as today. But 20 to 30 years to the future which country will be the new leaders/winners in providing the best semiconductor manufacturing environment?

Morris: It may be India, Vietnam, or Indonesia. Who knows?! (IMO, it's more sarcastic than some people thought).


There are several more Q&A not listed in the above summary.

 

[nghanayem]

Being a publicly-traded company, Intel is subject to the same cost of capital expectations as any other company in that category, and low net margins on such a capital-intensive business such as fabrication is unsustainable. It'll lower Intel's share price, increase interest costs on Intel debt, and may even attract an activist shareholder to get on the BoD and propose a different direction. I'm not suggesting that IFS has to have similar net margins to TSMC anytime soon, but Intel will have show the financial community a financial roadmap to getting competitively profitable, or Gelsinger and the board will get a lot of pressure improve performance or get replaced.

I think this is a case of both could be right. In the situation I presented for the minimum we need to see for IDM2.0 to be considered a success, I think it is certainly possible investors demand that either improvements get made or manufacturing gets spin out. However as I stated in that proposed situation there is no way in which that wouldn't hurt the wider intel group. As you also said I think we should all expect more than 1% op margin long term. Even if that and IFS customers helping fund the CAPEX needed to keep intel's scale is all that is needed to ensure the IDM business model stays a successful model for intel. If margins were only about break even" after the end of this build-up phase, after getting significant external wafer volumes to supplement internal volumes, and having a variety of depreciated fabs chugging along then that would be a failure in the spirit the goal even if it isn't really a failure from what the objective goal is.

Intel has a track record of producing support chips on the older processes. IO hubs, NICs, FPGAs, etc.

As it moves to an IFS model it may need to plan for customers who want long life assurance. A lot of government contracts require that, more than the 4 to 5 years Intel has traditionally offered for long life embedded or enterprise SKUs. But it may be able to manage that by excess production and holding inventory reserves, possibly cheaper than keeping old lines in use.

I don't disagree. My main point was that very clearly the main volume runners have moved past 14nm at this point, and that the current 14nm capacity definitely has to be lower than what it was in 1H 2021 given the lack of new fabs until just recently.

That much profit seems unlikely from old processes.

You don't have to take my word for it, it is in TSMC's earnings reports. 0% N3 revenue in Q2 and 40-something percent was N7 and N5 if memory serves.

Other foundries offer cheap alternatives, and the products themselves often only survive by being put into cheap products.

Yes but you must also consider the TSMC ecosystem. Unless you want something specialized that TSMC doesn't do someone like UMC or GF needs to offer better pricing to make up for the extra design legwork or IP porting costs. TSMC's larger scale at most nodes also gives them a wafer cost advantage over the smaller guys (all else being equal). Now does TSMC pass this cost onto customers, well you can just look at their margins compared to UMC and it is a pretty obvious N-O.

A wafer at $3000 is not going to contribute more than the N-1 products in HVM with excellent yields at $15k per wafer. The $3000 wafers are also a distraction in OPEX like personnel, maintenance, power, consumables, even if the CAPEX is depreciated.

I have no doubt that TSMC makes a killing on N7 and N5. Their foundry competition is very weak at those nodes, and the depreciation expenses get smaller and smaller every year along the dep curve (assuming they are smart and using DDB, MACRS, or something like that).

An Intel IFS will not have those distractions and may even go shopping for customers least likely to need long legacy support (other than excess inventory reserves). It does not need the same business model as TSMC, indeed better not.

Intel has said multiple times that they want to run their fabs as never ramp down like TSMC does. Moving to an 8 year depreciation model also only makes sense if you planned to do keep fabs running for well over a decade. Meanwhile the old pre BS 4 year depreciation curve only made sense when fabs only kept the nodes for a few years.

 

[Xebec]

I'm picturing Chang's advice coming from someone who has lived a very long time to see macro trends that aren't so obvious.

I think what he's saying is in 20-30 years, there's a high chance that the Taiwanese work culture won't be so 'hardcore', and other factors (cost of labor, etc.) may reduce Taiwan's special competitiveness that it has today.

This could be a lot of reasons: a population that is more aged than it is today (smaller working pool), the engineers seeing how it's "easier" to work elsewhere over time and expecting more, and the country itself maturing workforce wise. The latter part - this is not an opinion of support or not, but EU labor laws are certainly more lax than the US. Some of that is due to the work culture getting more 'relaxed' towards work/life balance. I don't see why that can't happen with Taiwan relative to younger / "more hungry" countries.

This is just a guess but there's probably some logic like this behind his statements. Business Founders of major corporations like TSMC are often on a different level of thinking..

 

[Tanj]

You don't have to take my word for it, it is in TSMC's earnings reports. 0% N3 revenue in Q2 and 40-something percent was N7 and N5 if memory serves.

Revenue is not profit. Did you mean profit (aka. earnings)?