Chinese semiconductor industry

gelgoog said:

Of course it will. US workers don't work double shifts. And US salaries are higher.

Click to expand...

I fully support US salaries in the semiconductor industry becoming even higher. I don't care if it comes at the cost of the rest of the US economy. (in fact, thats a plus) May the wealth disparity between the rich and poor become even greater in the US.

gelgoog said:

Of course it will. US workers don't work double shifts. And US salaries are higher.

Click to expand...

And lower productivity

t2contra said:

Welp, Morris Chang had said the cost of making chips in the US will be higher than making in Taiwan. Something like 50% more, IIRC.

Click to expand...

Has anyone do the math on this, cuz tsmc will still make a lot of money, just less than they would if they had the choice to pen the fab in Taiwan or maybe expand Nanjing.

1. most of the cost of a greenfield fab is from the fab and equipments, not from salaries or opex.

2. Revenue: $3.6B annually
- 20K wpm, >$15,000/5nm wfr

3. Cost of Revenue:
A. AZ Equipment depreciation: $2.4B for first five years (original $12B CAPEx)
B. Salaries+OPEX: ~$0.5B
- assuming avg $150K/TWN employee in 2021 (I got this number by estimating the 2021 depreciation by taking into account of CAPEX spent from 2017-2021. Subtract this from 2021 cost of revenue. The difference is assumed to be salary+opec. Divided by 55,000 employees)
- let’s assume US salary and OpEx is 100% higher than TWN —> $300K/US employee
- announced 1,600 employees planned for AZ


for year 1-5:
Revenue: $3.6B/yr
Profit=3.6-2.9=$0.7/yr
OpMargin: ~19%

year 6 and out:
Revenue: $3.6B/yr
Profit=$3.1B/yr
OpMargin: ~86%

this is just a conservative estimate with assumption US salary+OPEX doubles that if Taiwan. From this, you can see Morris is just complains about making less money than tsmc could. He hinted AZ fab will be profitable, but did not clarify it will make so much. I think he is just being a good businessman laying the foundation so they can:
1. justify asking his customers to pay even more for wafers or to justify not discounting wafer price,
2. ask their supplier for bigger discount.
3. Ask US government for even more subsidy or tax break
4. Justify not having to invest any further than existing commitment.

somebody feel free to check my math, but it sure look like even if AZ costs more to operate tsmc will still make lots of money.

tsmc’s overall gross margin is 50%, but a fully depreciated fab will have much higher margin. Even with high salary, cost in AZ, this is just a small portion of the revenue.
Opening a high wafer price, high margin 5nm fab in the US would still make financial sense. But investing in say a 28nm fab wouldn’t make sense (only ~$720M per annum so this type of fab will barely break even in the US).

The TSMC experience is that if you have the same people working two shifts you get better productivity and quicker turnaround in solving problems. So your fab's yields will be better, and they will get better quicker than your opponent's. You will ramp up production faster.
So it is not just about salaries. It is a problem of communication and knowledge being too spread out among production teams.
Add to that Samsung's experience at Texas where they have issues loading up their fab with enough orders and it is a recipe for inefficiency.

hvpc said:

Has anyone do the math on this, cuz tsmc will still make a lot of money, just less than they would if they had the choice to pen the fab in Taiwan or maybe expand Nanjing.

1. most of the cost of a greenfield fab is from the fab and equipments, not from salaries or opex.

2. Revenue: $3.6B annually
- 20K wpm, >$15,000/5nm wfr

3. Cost of Revenue:
A. AZ Equipment depreciation: $2.4B for first five years (original $12B CAPEx)
B. Salaries+OPEX: ~$0.5B
- assuming avg $150K/TWN employee in 2021 (I got this number by estimating the 2021 depreciation by taking into account of CAPEX spent from 2017-2021. Subtract this from 2021 cost of revenue. The difference is assumed to be salary+opec. Divided by 55,000 employees)
- let’s assume US salary and OpEx is 100% higher than TWN —> $300K/US employee
- announced 1,600 employees planned for AZ


for year 1-5:
Revenue: $3.6B/yr
Profit=3.6-2.9=$0.7/yr
OpMargin: ~19%

year 6 and out:
Revenue: $3.6B/yr
Profit=$3.1B/yr
OpMargin: ~86%

this is just a conservative estimate with assumption US salary+OPEX doubles that if Taiwan. From this, you can see Morris is just complains about making less money than tsmc could. He hinted AZ fab will be profitable, but did not clarify it will make so much. I think he is just being a good businessman laying the foundation so they can:
1. justify asking his customers to pay even more for wafers or to justify not discounting wafer price,
2. ask their supplier for bigger discount.
3. Ask US government for even more subsidy or tax break
4. Justify not having to invest any further than existing commitment.

somebody feel free to check my math, but it sure look like even if AZ costs more to operate tsmc will still make lots of money.

tsmc’s overall gross margin is 50%, but a fully depreciated fab will have much higher margin. Even with high salary, cost in AZ, this is just a small portion of the revenue.
Opening a high wafer price, high margin 5nm fab in the US would still make financial sense. But investing in say a 28nm fab wouldn’t make sense (only ~$720M per annum so this type of fab will barely break even in the US).

Click to expand...

@hvpc bro TSMC policy is to maintain that Profit Margin of 50%, since it's a US request anything lower then the American should shoulder the cost with subsidies. TSMC is trying to leverage its advantages and it's a smart move, they're there for the profit.

I don't want to politicized this thread BUT I think TSMC executive knows full well that when the Arizona FAB is completed there is a possibility of war in the Taiwan Strait. The US gets what it wanted and to solve the manpower issue those 50,000 Taiwanese technician will be the first one to be evacuated and issued a green card ASAP. Killing 2 birds with one stone, creating a market for the Vanity Arizona FAB (Taiwan FAB is affected) and solving the manpower issue. Typical American regime solution...lol

gelgoog said:

The TSMC experience is that if you have the same people working two shifts you get better productivity and quicker turnaround in solving problems. So your fab's yields will be better, and they will get better quicker than your opponent's. You will ramp up production faster.
So it is not just about salaries. It is a problem of communication and knowledge being too spread out among production teams.
Add to that Samsung's experience at Texas where they have issues loading up their fab with enough orders and it is a recipe for inefficiency.

Click to expand...

Samsung and tsmc is Apple and oranges, can’t draw a parallel.
Being a tsmc alum, I can tell you the experience/knowledge get thin out every year when we continued to build new fabs. So what you described is burning new.
The AZ fab will be more challenging, but given what tsmc go through on a almost a yearly basis (spreading out the talent and hiring new inexperienced people for new fabs/backfill people that left e siting fab to support new fabs), I don’t think it’s as catastrophic as people make it out to be.

this is also why tsmc had originally planned 5Q for AZ for trial-run before 2024. With recent construction delay, the bigger is just 3Q. But this is 3Q more than any other new fabs here in Taiwan. 3Q should allow tsmc to work out most of the kinks.

ansy1968 said:

@hvpc bro TSMC policy is to maintain that Profit Margin of 50%, since it's a US request anything lower then the American should shoulder the cost with subsidies. TSMC is trying to leverage its advantages and it's a smart move, they're there for the profit.

I don't want to politicized this thread BUT I think TSMC executive knows full well that when the Arizona FAB is completed there is a possibility of war in the Taiwan Strait. The US gets what it wanted and to solve the manpower issue those 50,000 Taiwanese technician will be the first one to be evacuated and issued a green card ASAP. Killing 2 birds with one stone, creating a market for the Vanity Arizona FAB (Taiwan FAB is affected) and solving the manpower issue. Typical American regime solution...lol

Click to expand...

Haha. Tsmc should hire you to lobby Uncle Sam to subsidize them for the “lost” profit. And you are absolutely right that tsmc is in it for the profit and should definitely lobby to US to shoulder some of the “extra” expenses.

your may be right about US’s evil plan to poach the 50k tsmc employees in the event of a war. But they will not be able to build up enough Gigafabs to replace tsmc…it would take at least a decade. I’m sure many idiot politicians may be drafting up such unrealistic plan without understanding that Rome is not (could be) built in “a few years”

any how, people are commenting on the extra salary and cost so I just thought I provide another perspective through actual numbers. Key point is, even if the AZ situation is not ideal, there’s enough of a profit potential/buffer that tsmc will not lose their shirt/pants I’m this investment. It’s not a doomsday situation for tsmc that some seems to insinuate.

hvpc said:

I just checked out this Hybrid Bonding tool's spec. This still need improvement to be used for advanced packaging. The alignment accuracy of 200nm & throughput of 12-18wph is not competitive for manufacturing. But it is a good foundation to build on.

For integration of very advanced chips with dense bonding pads of small pitch (<1um), the leading Hybrid bonding tool is spec'ed at <50nm & >50wph and still improving per mandate from the likes of tsmc and Intel.

Click to expand...

According to their 2020 report looks like they are trying get into the EVG hybrid bond market. They say they are working on a product to catch up with more advanced tools but they have yet to released it, maybe they are in verification phase (who knows). Looks they are getting like 20% of their sales from hybrid bonding, some people are interested in these kind of machines.

tokenanalyst said:

According to their 2020 report looks like they are trying get into the EVG hybrid bond market. They say they are working on a product to catch up with more advanced tools but they have yet to released it, maybe they are in verification phase (who knows). Looks they are getting like 20% of their sales from hybrid bonding, some people are interested in these kind of machines.

View attachment 87609

Click to expand...

Ah, looks like U-precision is benchmarking against EVG.

tsmc advanced packaging fab is in fact working with EVG to help define what will be required to meet their mass production requirements. From my understanding this small Austrian company seems to have a leg up on everyone on heterogeneous integration, at least at tsmc & Intel. The last time I talked to the EVG marketing person, she said they are still struggling with bot inter and intra-die overlay fingerprint correction and that it's difficult because the wafer warpage induced error is not consistent. It's not possible to correct random errors. I heard they are working with scanner companies to form a holistic solution to reduce the wafer-to-wafer bonding overlay fingerprint.

The leading edge tool is actually the Gemini FB XT.


I think it's important for U-precision to be working with someone domestically to refine their hybrid bonding tool just like EVG is with tsmc. There're subtle nuances that makes a tool production worthy that only the end-user can educate the suppliers. I hope U-precision hopefully is wise enough to already working with SMEE and leverage their new 2.5/3D packaging iline stepper if they want to take a share of the advanced packaging.

The HBS300 system's spec may be okay for niche application where the resolution and bonding density may not be so critical. I'm thing maybe for lowered CIS, MEMS, or perhaps compound semiconductor power devices.

Fedupwithlies said:

I fully support US salaries in the semiconductor industry becoming even higher. I don't care if it comes at the cost of the rest of the US economy. (in fact, thats a plus) May the wealth disparity between the rich and poor become even greater in the US.

Click to expand...

the problem is that semiconductor industry does not pay enough in America. The smartest people go for the highest salary jobs in Silicon Valley or Wall Street (or crypto now). Whether or not the policy makers like it, American job market is going further away from Industrial Age to digital age. Why get paid $150k a year at a fab plant when you can make 500k working remotely for a crypto company? That's the problem Intel faces.

hvpc said:

B. Salaries+OPEX: ~$0.5B
- assuming avg $150K/TWN employee in 2021 (I got this number by estimating the 2021 depreciation by taking into account of CAPEX spent from 2017-2021. Subtract this from 2021 cost of revenue. The difference is assumed to be salary+opec. Divided by 55,000 employees)
- let’s assume US salary and OpEx is 100% higher than TWN —> $300K/US employee
- announced 1,600 employees planned for AZ

Click to expand...

The issue here is that you don't seem to understand the labour market here in America. Talented engineers do not want to work in Arizona or in semiconductor industry. There is very high turnover for employers that don't give the kind of lifestyle that they would get from other tech firms. Among low skilled labour, the cost is ever increasing. I can guarantee you that the productivity rate in Arizona would be a fraction of what it is in Taiwan due to these reasons. So while they may announce that 1600 employees are planned, they are going to be paying a lot more people than that due to the ongoing new employee training, early termination and transitioning they are going to be dealing with. Just wait until the environmentalists start to converge in their Arizona plant a few years from now. A lot of money will need to be set aside from lobbyists and lawyers.

Just to give you an idea of turnover rate at large tech firms. In my first year out of college at Bloomberg, they told me 27% of employees there were hired in the past year. The first 2 or 3 months I was there was purely training. I really wasn't productive until at least 6 months in. Now, just consider what it would be like in Arizona where techies don't want to go.

None of this even factors in the higher corporate tax rate, payroll taxes, 401K and the outrageous health care costs we have here in America.

The other issue is that you assume that they would be able to get the same output at Arizona plant as they would from other plants. You also assume they can get the same revenue for each 5 nm chips 5 years from now. What's the point of investing in new tools and equipments if the old chips continue to bring in the same revenues? Let's never pay for upgrades or buy new machines if we can sell the current generation chips at the same price 20 years from now.

Where is the raw material costs in your calculation?