Chinese semiconductor industry

[hvpc]

If they can get SN1/SN2 up to 70k wpm by 2024, I think most people here should be happy. Heard earlier that getting equipment for advanced nodes is hard. Maybe that will be easier now that domestic suppliers are further along. I don't know how much more they can expand SMSC location. Maybe Lingang will also do advanced node.

As for other stuff, my understanding is the Shenzhen fab is coming online this year. The Beijing one might start production next year but probably won't hit close to 100k until 2025. And then there is a second phase of 100k wpm. Lingang started construction in January, so probably a year behind Beijing and start production in 2024. Tianjin just started construction this week, so probably another 2024 production start. All of this will take a while. They will need to continue to buy Western equipments, especially ASML machines. It's important to not underestimate the difficulties of scaling up production of domestic equipment.

Your expectations is more realistic, but I think still unlikely SMIC will have 70K wpm of 14nm/7nm by 2024. With constrained WFE supply, the expected SMIC CAPEx the next few years, and timing of SN2 fab readiness, it's more likely 2026 is when SMIC could output 70K wpm. 2024/2025 for tool-move in, take some time to qualify tools, so E of 2025 or some time in 2026 to be able to turn actual capacity into actual wafer output.

Now, let's take a look at the math that I was talking about. Here's information based on SMIC's announcement:

SMIC SN1	35K	14nm/7nm	$12B
SMIC SN2	35K	14nm/7nm	$12B
SMIC Beijing	100K	28nm	$7.7B
SMIC Linggang	100K	28nm	$8.9B
SMIC Tianjing	100K	28nm	$7.5B
SMIC Shenzen	40K	28nm	$2.35B
Total	410K		$50.45B

All of these 300mm expansions summed up to $50B in CAPEx. As you had mentioned, Shenzen started to move-in tools this year, in 2022. How fast and how much capacity come online by 2024 will depend on how much they spend in 2023/2024.

As you mentioned, Shenzen started to move-in tools in 2022. SN1 actually moved-in tools as well (I think they had 15K wpm last year), so let's assume $6B worth still need to be spent from beginning of 2022.
- SN1/SN2 to be full, still need to spend $18B ($6B+$12B).
- let's assume Shenzen reach max capacity by 2024 as well: $2.35B

With just these three fabs, SMIC needs to spend $20.35B. Subtract the $5B spent in 2022, SMIC needs to spend another $15B in 2023/2024. ==>$7.5/year in the next two years. that's 50% more worth of equipment than 2022. Even if SMIC actually spend this type of money, it's not like ASML and other WFE could allocate 50% more tools to SMIC over 2022.

So, I think it's more likely SN1/SN2 get filled with tools by 2025 & wafer production 2026. If we assume the $20B is spent in 2023/2024/2025, then it's about $5B/year (which is probably what the WFE suppliers could support).

The above assumes zero investment in Beijing, Tianjing, and Linggang. It's likely SMIC would have to divvy up the wafer fab equipment allocated to them with these 28nm fabs, in that case timing for SN1/SN2 to reach 70K get pushed back even further.

Well, of course the above is assuming ASML/AMAT/KLA/LAM continue to allocated same amount of tools to SMIC in 2023/2024/2025. Any increase or decrease in tool allocation will influence the actual outcome. With ASML telling investors that they will not be able to increase their capacity until 2025/2026, SMIC's expansion plan, like everyone around the world, has to be tapered.

This is my quick and dirty ad hoc analysis for your reference.

 

[hvpc]

Makes me wonder how far along is Intels new high end fab?
Is there a change it will come online in next few years?

inbe4 TSMC goes into money problems so Intel with government money can buy those EUV machines for pennies on the dollars.

Intel just broke ground for their fab in Ohio earlier this month:

Please, Log in or Register to view URLs content!

You guys are all reading too much into media reports and jumping to conclude this is tsmc's demise. Apple objecting a price hike to 3nm/4nm wafers doesn't mean they will not consume any 3nm wafers at the current price point. So, how would tsmc be hurt bad enough that they need to sell their EUVL for pennies on the dollar?

Even if tsmc could not get Apple to pay more, tsmc's current wafer price is, and historically has been, at up to 20% premium over SMIC/UMC/Global Foundries. Since 2020, tsmc increased their wafer price and others followed. If tsmc has to drop their wafer price, the other would not be spared of the same fate and would have to adjust their price downward as well. This is how the foundry market works. Everything get trickled down from tsmc. So, if tsmc suffers, that would mean SMIC would, too.

Looking at the tsmc/Apple situation, if wafer price stay as is, tsmc will continue to make in excess of 50% GM. I don't really buy-in to tsmc's story that the raw material cost is going up so they have to increase wafer price. Because in the last two years, tsmc's Gross Margin continues to climb, so I think their wafer price increase is much higher than needed to cover inflationary cost.

Now, you may argue the domestic demand is high enough that SMIC will not be impacted. So you all think, but this is not what Zhao Haijun said to the investors last month. Of all the foundries, the outlook sentiment provided by SMIC is the most negative. SMIC has a much larger exposure to China, coupled with slowdown in Chinese market, the slowing economy impacts SMIC the most. Segments that exhibits slow-down (mid to low end mobile, consumer electronics, and CMOS Imaging Sensors) are where SMIC mostly play in.

"We are still closely monitoring the pandemic situation surrounding us in China. ... We have more interactions with the Chinese market so we see more impact," said Zhao Haijun.​

The High Performance Computing and high-end mobile inventory is relatively more healthy. These are basically 7nm/5nm/3nm where tsmc derives over half of their revenue from. With their bread and butter segments doing well, this is why tsmc maintain the most positive outlook in their last quarterly report and still expects 30% Y/Y growth for 2022. The media has been hyping the industry down turn will impact tsmc by reporting all kinds of new since March, and yet tsmc continue record monthly revenue month-after-month. Even if tsmc have a mediocre September, they will have hit another record quarter for Q3. tsmc acknowledge inventory correction by customers until maybe first quarter of 2023, but despite these, their order book is healthy and they are not worried.

If we don't believe tsmc's optimistic outlook, then how about considering Zhao Haijun's comment that the downturn may reach a turning point at the end of this year. If so, everybody continue making money, SMIC and tsmc. If the economy gets worse, no one would bet money that SMIC will come out of this better off than tsmc.

NEV segment is growing. But this segments currently account for less than 10% of IC market. It's also <10% of SMIC's revenue contribution. Strength in this segment could offset some but not all weakness in other segments. If the economy doesn't turn around per Zhao Haijun's assessment, NEV segment strength will not be able to save SMIC, either.

So, unlike you guys, I hope the economy will not continue the downward spiral and it won't be doom and gloom for tsmc, because this would mean SMIC will get hurt as well. I don't want SMIC to get hurt, too.

 

[tokenanalyst]

Interesting, looks like they are few companies making this kind of equipment,yet to see how advance they are or how independent they are to geopolitical trouble. Looks like is an area that still need some work to do.

Please, Log in or Register to view URLs content!

Please, Log in or Register to view URLs content!

View attachment 98429
View attachment 98430

Please, Log in or Register to view URLs content!

 

[PopularScience]

Your expectations is more realistic, but I think still unlikely SMIC will have 70K wpm of 14nm/7nm by 2024. With constrained WFE supply, the expected SMIC CAPEx the next few years, and timing of SN2 fab readiness, it's more likely 2026 is when SMIC could output 70K wpm. 2024/2025 for tool-move in, take some time to qualify tools, so E of 2025 or some time in 2026 to be able to turn actual capacity into actual wafer output.

Now, let's take a look at the math that I was talking about. Here's information based on SMIC's announcement:

SMIC SN1	35K	14nm/7nm	$12B
SMIC SN2	35K	14nm/7nm	$12B
SMIC Beijing	100K	28nm	$7.7B
SMIC Linggang	100K	28nm	$8.9B
SMIC Tianjing	100K	28nm	$7.5B
SMIC Shenzen	40K	28nm	$2.35B
Total	410K		$50.45B

All of these 300mm expansions summed up to $50B in CAPEx. As you had mentioned, Shenzen started to move-in tools this year, in 2022. How fast and how much capacity come online by 2024 will depend on how much they spend in 2023/2024.

As you mentioned, Shenzen started to move-in tools in 2022. SN1 actually moved-in tools as well (I think they had 15K wpm last year), so let's assume $6B worth still need to be spent from beginning of 2022.
- SN1/SN2 to be full, still need to spend $18B ($6B+$12B).
- let's assume Shenzen reach max capacity by 2024 as well: $2.35B

With just these three fabs, SMIC needs to spend $20.35B. Subtract the $5B spent in 2022, SMIC needs to spend another $15B in 2023/2024. ==>$7.5/year in the next two years. that's 50% more worth of equipment than 2022. Even if SMIC actually spend this type of money, it's not like ASML and other WFE could allocate 50% more tools to SMIC over 2022.

So, I think it's more likely SN1/SN2 get filled with tools by 2025 & wafer production 2026. If we assume the $20B is spent in 2023/2024/2025, then it's about $5B/year (which is probably what the WFE suppliers could support).

The above assumes zero investment in Beijing, Tianjing, and Linggang. It's likely SMIC would have to divvy up the wafer fab equipment allocated to them with these 28nm fabs, in that case timing for SN1/SN2 to reach 70K get pushed back even further.

Well, of course the above is assuming ASML/AMAT/KLA/LAM continue to allocated same amount of tools to SMIC in 2023/2024/2025. Any increase or decrease in tool allocation will influence the actual outcome. With ASML telling investors that they will not be able to increase their capacity until 2025/2026, SMIC's expansion plan, like everyone around the world, has to be tapered.

This is my quick and dirty ad hoc analysis for your reference.

All these new fabs are joint venture with the local governments, including SN1/2.

 

[hvpc]

@ansy1968, are you suggesting all fabs you listed will reach capacity and fully operational by 2024? Any idea how many total scanner systems this will require? How many total scanners do you foresee SMEE capable of delivery by 2024?

The math doesn't seem to add up.

Bro, since I'm in an analyzing/sharing mood, let me continue.

Let's look at another key bottleneck of expansion speed: availability of scanners . Fab expansion starts with scanners. No one goes out to buy more process equipment unless they could secure enough scanners. So let's look at how many scanners would be required.

If I remember correctly from when I'm in charge of scanners at a foundry, 28nm node flow typically takes about 50 litho steps, and 7nm about 75 litho steps (or # of reticles). Let's append this info to the table I shared previously:

SMIC SN1	35K	14nm/7nm	$12B	~25 ASML scanners
SMIC SN2	35K	14nm/7nm	$12B	~25 ASML scanners
SMIC Beijing	100K	28nm	$7.7B	~45 ASML scanners
SMIC Linggang	100K	28nm	$8.9B	~45 ASML scanners
SMIC Tianjing	100K	28nm	$7.5B	~45 ASML scanners
SMIC Shenzen	40K	28nm	$2.35B	~18 ASML scanners
Total	410K		$50.45B	~203 scanners

So, ~203 scanners is what SMIC needs.

- ASML ships 23systems/Q ==> 92systems/year
- assumes SMIC gets 1/3 of all systems (leaving 2/3 of systems to be divvy up between CXMT, YMTC, Huahong, and others): ~30systems for SMIC per year.
- This then means it'll take SMIC ~6 and half years to fill up fabs. This takes us to middle of 2027, which means 2028 to see all 410K wpm wafer output realized.

Assumming 6.5years to ramp:
- 2022: $5B CAPEx
- 2023 through 2026: $7B
- 2027: $3.5B CAPEx

Upon looking at this result, $7B/year which is 40% increase over 2022 CAPEx would seem plausible. However, this would also mean a 40% increase in scanners. Given ASML's announced supply constraint, I'm less confident if ASML could increase number of scanners they sell to SMIC by 40%.

And of course, this would mean, previous assumption of 30scanners/year is unrealistic. If $7B = ~30scanners. Then $5B in 2022 would mean SMIC purchased ~24 scanners in 2022.

Anyhow, analysis like this is some times iterative; many combinations. So let's adjust our original assumptions to come up with two possible scenarios:

1. If you are bullish on SMIC's willingness to spend and ASML capable to supply 30systems/year, then let's just call it 6.5years (from top of 2022). This means all SMIC 300mm fabs fully ramped before end of 2027.

2. If you are less bullish on ASML's capability to ship tools, let's call it 24systems/year @ $5B CAPEx spent by SMIC, then it'll be 8years to fill the fabs ==> SMIC 300mm fabs fully filled by end of 2029.

I'll stop right here. Y'all can decide for yourself if SMIC could pull off the entire capacity expansion faster than 6.5 years or slower than 8years. But, be mindful of the implication with the annual CAPEx and #scanners available that comes with whatever scenarios you chose.

 

[hvpc]

All these new fabs are joint venture with the local governments, including SN1/2.

Even with JV, all CAPEx are reported on SMIC's financial reports.

Funding from JV are tabulated in the 'Other Comprehensive Income' section of the Income Statement.

The final EBIT are then split and listed as separate line items at the end of the Income Statement as:

Total comprehensive income for the year attributable to: ​

Owners of the Company $xxx,xxx​

Non-controlling interests $xxx,xxx​

 

[PopularScience]

Even with JV, all CAPEx are reported on SMIC's financial reports.

Funding from JV are tabulated in the 'Other Comprehensive Income' section of the Income Statement.

The final EBIT are then split and listed as separate line items at the end of the Income Statement as:

Total comprehensive income for the year attributable to: ​

Owners of the Company $xxx,xxx​

Non-controlling interests $xxx,xxx​

You don’t worry about capex. Government will make sure the fabs have enough money.

 

[hvpc]

You don’t worry about capex. Government will make sure the fabs have enough money.

I think you misunderstood my point.

You only get as many tools as you spend. SMIC only spending $5B in 2022 and did not indicate they will increase it by much in subsequent years...hence the analysis that it will take time until they fill all their fabs.

And in terms of accounting, they are a publicly traded company and have to follow standard international accounting rules. Any funding from the government are logged in a very specific way in the financial reports.

We know SMIC gets plenty of support from government otherwise they would to be able to spend more than their profit these past few years.

Again, the focus here is that SMIC, either because the WFE suppliers could not allocate them enough tools or they are investing at a particular pace, are not spending $15-20 billion a year to support the theory that all fabs will be filled by 2024.

Remember, my other point, even if SMIC has money to spend, there's not enough equipment to buy.

I recommend you read through SMIC's annual reports if you haven't already. You'll find lots of concrete data point to use in your assessment/extrapolation, which is much more accurate than making qualitative speculations.

Please, Log in or Register to view URLs content!

 

[PopularScience]

Bro, since I'm in an analyzing/sharing mood, let me continue.

Let's look at another key bottleneck of expansion speed: availability of scanners . Fab expansion starts with scanners. No one goes out to buy more process equipment unless they could secure enough scanners. So let's look at how many scanners would be required.

If I remember correctly from when I'm in charge of scanners at a foundry, 28nm node flow typically takes about 50 litho steps, and 7nm about 75 litho steps (or # of reticles). Let's append this info to the table I shared previously:

SMIC SN1	35K	14nm/7nm	$12B	~25 ASML scanners
SMIC SN2	35K	14nm/7nm	$12B	~25 ASML scanners
SMIC Beijing	100K	28nm	$7.7B	~45 ASML scanners
SMIC Linggang	100K	28nm	$8.9B	~45 ASML scanners
SMIC Tianjing	100K	28nm	$7.5B	~45 ASML scanners
SMIC Shenzen	40K	28nm	$2.35B	~18 ASML scanners
Total	410K		$50.45B	~203 scanners

So, ~203 scanners is what SMIC needs.

- ASML ships 23systems/Q ==> 92systems/year
- assumes SMIC gets 1/3 of all systems (leaving 2/3 of systems to be divvy up between CXMT, YMTC, Huahong, and others): ~30systems for SMIC per year.
- This then means it'll take SMIC ~6 and half years to fill up fabs. This takes us to middle of 2027, which means 2028 to see all 410K wpm wafer output realized.

Assumming 6.5years to ramp:
- 2022: $5B CAPEx
- 2023 through 2026: $7B
- 2027: $3.5B CAPEx

Upon looking at this result, $7B/year which is 40% increase over 2022 CAPEx would seem plausible. However, this would also mean a 40% increase in scanners. Given ASML's announced supply constraint, I'm less confident if ASML could increase number of scanners they sell to SMIC by 40%.

And of course, this would mean, previous assumption of 30scanners/year is unrealistic. If $7B = ~30scanners. Then $5B in 2022 would mean SMIC purchased ~24 scanners in 2022.

Anyhow, analysis like this is some times iterative; many combinations. So let's adjust our original assumptions to come up with two possible scenarios:

1. If you are bullish on SMIC's willingness to spend and ASML capable to supply 30systems/year, then let's just call it 6.5years (from top of 2022). This means all SMIC 300mm fabs fully ramped before end of 2027.

2. If you are less bullish on ASML's capability to ship tools, let's call it 24systems/year @ $5B CAPEx spent by SMIC, then it'll be 8years to fill the fabs ==> SMIC 300mm fabs fully filled by end of 2029.

I'll stop right here. Y'all can decide for yourself if SMIC could pull off the entire capacity expansion faster than 6.5 years or slower than 8years. But, be mindful of the implication with the annual CAPEx and #scanners available that comes with whatever scenarios you chose.

SMEE planned to build 20 DUVs this year. I think this have been delayed.

 

[tokenanalyst]

Intel just broke ground for their fab in Ohio earlier this month:

Please, Log in or Register to view URLs content!

You guys are all reading too much into media reports and jumping to conclude this is tsmc's demise. Apple objecting a price hike to 3nm/4nm wafers doesn't mean they will not consume any 3nm wafers at the current price point. So, how would tsmc be hurt bad enough that they need to sell their EUVL for pennies on the dollar?

Even if tsmc could not get Apple to pay more, tsmc's current wafer price is, and historically has been, at up to 20% premium over SMIC/UMC/Global Foundries. Since 2020, tsmc increased their wafer price and others followed. If tsmc has to drop their wafer price, the other would not be spared of the same fate and would have to adjust their price downward as well. This is how the foundry market works. Everything get trickled down from tsmc. So, if tsmc suffers, that would mean SMIC would, too.

Looking at the tsmc/Apple situation, if wafer price stay as is, tsmc will continue to make in excess of 50% GM. I don't really buy-in to tsmc's story that the raw material cost is going up so they have to increase wafer price. Because in the last two years, tsmc's Gross Margin continues to climb, so I think their wafer price increase is much higher than needed to cover inflationary cost.

Now, you may argue the domestic demand is high enough that SMIC will not be impacted. So you all think, but this is not what Zhao Haijun said to the investors last month. Of all the foundries, the outlook sentiment provided by SMIC is the most negative. SMIC has a much larger exposure to China, coupled with slowdown in Chinese market, the slowing economy impacts SMIC the most. Segments that exhibits slow-down (mid to low end mobile, consumer electronics, and CMOS Imaging Sensors) are where SMIC mostly play in.

"We are still closely monitoring the pandemic situation surrounding us in China. ... We have more interactions with the Chinese market so we see more impact," said Zhao Haijun.​

The High Performance Computing and high-end mobile inventory is relatively more healthy. These are basically 7nm/5nm/3nm where tsmc derives over half of their revenue from. With their bread and butter segments doing well, this is why tsmc maintain the most positive outlook in their last quarterly report and still expects 30% Y/Y growth for 2022. The media has been hyping the industry down turn will impact tsmc by reporting all kinds of new since March, and yet tsmc continue record monthly revenue month-after-month. Even if tsmc have a mediocre September, they will have hit another record quarter for Q3. tsmc acknowledge inventory correction by customers until maybe first quarter of 2023, but despite these, their order book is healthy and they are not worried.

If we don't believe tsmc's optimistic outlook, then how about considering Zhao Haijun's comment that the downturn may reach a turning point at the end of this year. If so, everybody continue making money, SMIC and tsmc. If the economy gets worse, no one would bet money that SMIC will come out of this better off than tsmc.

NEV segment is growing. But this segments currently account for less than 10% of IC market. It's also <10% of SMIC's revenue contribution. Strength in this segment could offset some but not all weakness in other segments. If the economy doesn't turn around per Zhao Haijun's assessment, NEV segment strength will not be able to save SMIC, either.

A lot of the increase in demand of SMIC is due the desire of Chinese companies to localize their chip supply chain, which is mostly for geopolitical reasons not just demand, every single rumor of more restriction will likely push for more localization.
For 28nm+ SMIC could allocate tools from domestic players like NAURA, AMEC, Piotech and so on. Apart from international suppliers.

So, unlike you guys, I hope the economy will not continue the downward spiral and it won't be doom and gloom for tsmc, because this would mean SMIC will get hurt as well. I don't want SMIC to get hurt, too.

Sadly we don't have a say on that, but for central banks is either uncontrolled inflation or economic growth, the reality is that inflation eat through the middle class like hot knife in butter. If central bankers would have acted sooner to control the market bubble that themselves created they won't be now between a rock and hard place.