Array
(
    [content] => 
    [params] => Array
        (
            [0] => /forum/index.php?threads/results-from-semiconductor-supply-chain-request-for-information.15359/
        )

    [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
)

Results from Semiconductor Supply Chain Request for Information

hist78

Well-known member
"The specific kinds of products we identified as having significant semiconductor supply and demand mismatches are used by critical industries, including medical devices, broadband, and autos. They include:

~ Microcontrollers that are primarily made of legacy logic chips, including, for example, at 40, 90, 150, 180, and 250 nm nodes

~ Analog chips including, for example, at 40, 130, 160, 180, and 800 nm nodes; and

~ Optoelectronics chips including, for example, at 65, 110, and 180 nm nodes."



My thoughts about this official study of semiconductor shortage:

The report did not even include 28nm (TSMC started mass production in 2011) in the supply shortage list. Looking at those nodes in short supply, 40, 65, 110, 130, 150, 160, 180, 250, and 800nm, I'm thinking how lucky those automobile manufacturers are. Now in 2022 there are fabs who are still willing to make chips based on those old processes with cheap price. But on the other hand, is it reasonable to expect fabs to keep investing a lot more capital and resources into those old process nodes? For many of those companies crying for chip supply, do they ask for a lot but pay too little? Here I quote some text from Wikipedia about 800nm, a node adopted by the industry 30 years ago:

"The 800 nm process refers to the level of MOSFET semiconductor fabrication process technology that was reached around the 1987–1990 timeframe, by leading semiconductor companies like NTT, NEC, Toshiba, IBM, Hitachi, Matsushita, Mitsubishi Electric and Intel."


The reality is no matter how much government and private money will invest in the new equipment and new fabs, I doubt they are willing to spend too much on those legacy nodes mentioned. There is no quick solution for auto makers or any companies need those chips based on old process technology.
 
Last edited:
Thanks for the update; I had heard the report was coming out sometime soon but didn't know when.

The reality is no matter how much government and private money will invest in the new equipment and new fabs, I doubt they are willing to spend too much on those legacy nodes mentioned. There is no quick solution for auto makers or any companies need those chips based on old process technology.

The manufacturers in this market space would beg to differ. (well, except maybe the part about the "quick solution")

TI just had their 2021Q4 earnings call today, and mentioned this on the subject:

Vivek Arya: Yes. Thank you, Dave. So the other question is now on the supply side. There is an
investor concern that the semiconductor industry is over investing at a time when
demand might be peaking. And I know you guys have made it clear that you invest for
the longer-term. But how are you thinking about your current acceleration on the
investment side? When does that translate into actual useful capacity? And what are
you doing to make sure that you don't over invest, right, at least in the next couple of
quarters?

Rafael Lizardi: Yes. Vivek, I'll go ahead and take that. As you alluded to at the beginning of the
question, we think of the long-term when we make this decision. So this is not about
2021, 2022 or even 2023. This is over the long-term. And the secular trends in our
industry, we're confident of where those are pointing and specifically, in our products,
analog and embedded, and the end markets that -- where we put a strategic priority
industrial automotive. So on the manufacturing investments that you alluded to, we're
very excited about those.

As I mentioned during the prepared remarks, they're going to strengthen our competitive
advantage on manufacturing and technology. First, we're going to have significant 300-
millimeter capacity coming online with RFAB2 and Lehi. That's going to happen actually
this year and then going to next year with Lehi.

Second, with the announcement of the Sherman Complex, we're going to have a road
map that's going to support us out to 2035. And then finally, customers are very excited
about our investments specifically in 45- to 130-nanometer process technologies that
are optimized for Analog and Embedded and will support the customers' growth for
decades ahead.

So at the very least, TI is making a big bet on 300mm "old" technology.

I think the idea of what is "old" and "new" is somewhat relative; 45nm is "old" from the standpoint of leading-edge digital logic (TSM announced 45nm support in April 2007), but it's fairly recent in the embedded market.
 
Thanks for the update; I had heard the report was coming out sometime soon but didn't know when.



The manufacturers in this market space would beg to differ. (well, except maybe the part about the "quick solution")

TI just had their 2021Q4 earnings call today, and mentioned this on the subject:



So at the very least, TI is making a big bet on 300mm "old" technology.

I think the idea of what is "old" and "new" is somewhat relative; 45nm is "old" from the standpoint of leading-edge digital logic (TSM announced 45nm support in April 2007), but it's fairly recent in the embedded market.

Definitely some incumbent fabs will invest more in old process nodes because they have captive customers. But my question is how far can this go in the long run? Can automobile companies expect they will have cheap and reliable supply of 180, 250, 800nm products 10 years from now?

If the answer is "no", then they have to move on to newer (not newest) process nodes. If they don't manage the risk of technology obsolescence, no one is obligated to save them.
 
Definitely some incumbent fabs will invest more in old process nodes because they have captive customers. But my question is how far can this go in the long run? Can automobile companies expect they will have cheap and reliable supply of 180, 250, 800nm products 10 years from now?

If the answer is "no", then they have to move on to newer (not newest) process nodes. If they don't manage the risk of technology obsolescence, no one is obligated to save them.
I don't disagree with you completely, but it's a weird situation where supply and demand seem to be going in different directions.... TSMC at the leading edge and building a token 22/28nm fab in Japan for the trailing edge, while there's still plenty of demand for 40nm+ nodes.

Like a big party where TSMC is at the bar selling high-end drinks and sushi, but they've run out of water and orange juice, and aren't interested in getting more because the profit margins aren't good enough. But a bunch of the partygoers are starting to get really thirsty for water or at least something without alcohol, and hungry for something a little less exotic for their stomachs, and they're whining and moaning about it but no one feels like getting up to go out and fetch some water and sandwiches, until someone does and maybe they'll make some money off the deal... perhaps not as much money as the bar is making, but there's a demand for it.
 
In the past TSMC was slow to qualify processes for automotive grade chips. Now TSMC quickly qualifies new nodes (N5 is qualified) for automotive applications. Unfortunately car company chip suppliers did not get the memo. Now car companies are starting to design their own chips and you can bet they are not on the legacy nodes.

Over the last three years SemiWiki content and traffic for automotive grade EDA tools and IP has jumped. I can also see the domains and they now include all of the big car companies and literally hundreds of domaiins I have never heard of, especially from China.
 
  • Like
Reactions: VCT
I don't disagree with you completely, but it's a weird situation where supply and demand seem to be going in different directions.... TSMC at the leading edge and building a token 22/28nm fab in Japan for the trailing edge, while there's still plenty of demand for 40nm+ nodes.

Like a big party where TSMC is at the bar selling high-end drinks and sushi, but they've run out of water and orange juice, and aren't interested in getting more because the profit margins aren't good enough. But a bunch of the partygoers are starting to get really thirsty for water or at least something without alcohol, and hungry for something a little less exotic for their stomachs, and they're whining and moaning about it but no one feels like getting up to go out and fetch some water and sandwiches, until someone does and maybe they'll make some money off the deal... perhaps not as much money as the bar is making, but there's a demand for it.
“Like a big party where TSMC is at the bar selling high-end drinks and sushi, but they've run out of water and orange juice, and aren't interested in getting more because the profit margins aren't good enough”

TSMC’s margin are actually higher on the older nodes, they are the cash cows that pay for the new nodes. Every time TSMC introduces a new node it drags down their margins for the first couple of years. When a node becomes fully depreciated TSMC’s cost are cut in half but they only pass some of that on to the customers and their margin goes up. Of course if they buy new equipment for the older nodes the margins would come down.
 
“Like a big party where TSMC is at the bar selling high-end drinks and sushi, but they've run out of water and orange juice, and aren't interested in getting more because the profit margins aren't good enough”

TSMC’s margin are actually higher on the older nodes, they are the cash cows that pay for the new nodes. Every time TSMC introduces a new node it drags down their margins for the first couple of years. When a node becomes fully depreciated TSMC’s cost are cut in half but they only pass some of that on to the customers and their margin goes up. Of course if they buy new equipment for the older nodes the margins would come down.
I'm wondering on the Intel side, how do they utilize those legacy nodes capacity? If Intel only does advanced nodes and is building 5 new leading edge fabs in the next three years, how do they do with those old and not so old fabs?

How do you think the supply shortage on those legacy nodes (40, 65, 110, 130, 150, 160, 180, 250, and 800nm) pointed out by the Commerce Department's report?

Is it really the problem? Is there any meaningful solution?
 
In the past Intel upgraded their Fabs to new nodes, the oldest node Intel is currently running is 32nm. it may be part of why they want to get into the foundry business, they currently have no use for older nodes but foundries do.
 
TSMC’s margin are actually higher on the older nodes, they are the cash cows that pay for the new nodes. Every time TSMC introduces a new node it drags down their margins for the first couple of years. When a node becomes fully depreciated TSMC’s cost are cut in half but they only pass some of that on to the customers and their margin goes up. Of course if they buy new equipment for the older nodes the margins would come down.

Huh, that's interesting! I stand corrected; I figured at/near the leading node they were effectively a monopoly and could command a larger premium.

Since you brought up the trend of gross margin increasing for older nodes, is there anything quantitative you can throw our way to shed some light on the situation? Not an IC-Knowledge-investment-grade estimate for paying customers, but something a lot more coarse... say a "horseshoes-and-hand-grenades" grade estimate ("close enough") for how, say, their gross margin changes from the leading node to nodes that are more than 5 years old?

Just as an example, TSMC's quarterly reports document their profitability for the business as a whole; the latest earnings release for 2021 Q4 earlier this month mentions in the management report that the company shipped 3.725 million 300mm-equivalent wafers, gross margin was 52.7% and operating margin 41.7%, with these breakdowns per node for revenue:

1643259819344.png


Half the revenue in 2021 Q4 comes from 5nm/7nm nodes, another 24% from 16nm and 28nm, and the remaining 26% from 40nm and larger.

TSMC has been presenting their revenue in this approximate format for years. I picture the company surfing a kind of wave moving towards decreasing geometry (Morris Chang, C.C. Wei, and Mark Liu all hanging ten on surfboards, wearing sunglasses, if you prefer), with the leading edge at the crest of the wave and the trailing nodes behind them, decreasing gradually as a percentage of total revenue.

Rewind five years to 2016 Q4 and you see almost the same margins (52.3% gross, 41.9% operating) with this table:

1643260317028.png

16-28nm was leading edge at the time, and made up 57% of revenue, with the next two ranges (40-65nm) making 23% and the rest making up the remaining 20%.

Another five years back to 2011 Q4 and the margins are a little smaller (44.7% gross, 31% operating margin; commentary mentions these are down from a year earlier, "mainly due to lower capacity utilization and an unfavorable foreign exchange rate, partially balanced by cost improvements.") with 29% of revenues coming from 28-45nm, 38% from 65-90nm, and the rest making up the remaining 33%.
1643260583642.png



If I understand your point, let's say their gross margin is about 50% overall on average. Would I expect the situation to be something like:

a) 45% gross margin from leading nodes, 50% from nodes that have been around for 2-3 years, and 55% from older nodes,
b) 40% gross margin from leading nodes, 50% from nodes that have been around for 2-3 years, and 60% from older nodes,
c) 30% gross margin from leading nodes, 50% from nodes that have been around for 2-3 years, and 70% for older nodes...
d) something else; I'm not even close

?
 
...and regardless of the quantitative financials, there's an implied question here:

- If there's been so much sustained demand at 40nm+ (increased demand for nodes on 200mm wafers have been in the news since at least 2015)

- and these older nodes have got higher profit margin

- is there any reason TSMC hasn't been slowly adding a little capacity gradually to the older fabs to provide at least some further increase in long-term profits? Even 5-10% over five years. Not because they want to be charitable and help all their customers still using "mature nodes", but because they want to increase the income of their business.

If they have, it's been done very quietly; I haven't run across any mentioned of it in the earnings materials on their website. If they haven't, is there a non-obvious reason? (for example, I'm just guessing: you can't add 5% more equipment to a full fab building, you'd have to put that equipment in a new fab building and then you would only have 5% of a fab so it wouldn't provide the advantages of scale as a fully-equipped building and the net profit would be much lower, so not worth it to try to add 5%; you'd have to add a big chunk like 25-50% and the financial numbers just don't work to support that)
 
Huh, that's interesting! I stand corrected; I figured at/near the leading node they were effectively a monopoly and could command a larger premium.

Since you brought up the trend of gross margin increasing for older nodes, is there anything quantitative you can throw our way to shed some light on the situation? Not an IC-Knowledge-investment-grade estimate for paying customers, but something a lot more coarse... say a "horseshoes-and-hand-grenades" grade estimate ("close enough") for how, say, their gross margin changes from the leading node to nodes that are more than 5 years old?

Just as an example, TSMC's quarterly reports document their profitability for the business as a whole; the latest earnings release for 2021 Q4 earlier this month mentions in the management report that the company shipped 3.725 million 300mm-equivalent wafers, gross margin was 52.7% and operating margin 41.7%, with these breakdowns per node for revenue:

View attachment 629

Half the revenue in 2021 Q4 comes from 5nm/7nm nodes, another 24% from 16nm and 28nm, and the remaining 26% from 40nm and larger.

TSMC has been presenting their revenue in this approximate format for years. I picture the company surfing a kind of wave moving towards decreasing geometry (Morris Chang, C.C. Wei, and Mark Liu all hanging ten on surfboards, wearing sunglasses, if you prefer), with the leading edge at the crest of the wave and the trailing nodes behind them, decreasing gradually as a percentage of total revenue.

Rewind five years to 2016 Q4 and you see almost the same margins (52.3% gross, 41.9% operating) with this table:

View attachment 630
16-28nm was leading edge at the time, and made up 57% of revenue, with the next two ranges (40-65nm) making 23% and the rest making up the remaining 20%.

Another five years back to 2011 Q4 and the margins are a little smaller (44.7% gross, 31% operating margin; commentary mentions these are down from a year earlier, "mainly due to lower capacity utilization and an unfavorable foreign exchange rate, partially balanced by cost improvements.") with 29% of revenues coming from 28-45nm, 38% from 65-90nm, and the rest making up the remaining 33%.
View attachment 631


If I understand your point, let's say their gross margin is about 50% overall on average. Would I expect the situation to be something like:

a) 45% gross margin from leading nodes, 50% from nodes that have been around for 2-3 years, and 55% from older nodes,
b) 40% gross margin from leading nodes, 50% from nodes that have been around for 2-3 years, and 60% from older nodes,
c) 30% gross margin from leading nodes, 50% from nodes that have been around for 2-3 years, and 70% for older nodes...
d) something else; I'm not even close

?
TSMC reported gross margin is the weighted average of all their fabs running all their processes sold to all their customers.

- Large customers pay lower gross margins than small customers. Apple pays the lowest margins, a start up that needs a couple of engineering lots pays a much higher margin.
- Exchange rates effect margins, in the first half of 2021 TSMC margins were pulled down by exchange rate issues.
- Market forces effect margins, TSMC's margin increased in the second half of 2021 partly due to price increases and in 2022 announced price increases should drive up margins more.
- Margin is highly correlated with fab utilization, full fabs have lower costs and drive higher margins. For some older nodes TSMC had a period of excess capacity and lower margins trying to "buy" business.
- The node age isn't really what matters, it is the depreciation status of the equipment being used. This would be simple if TSMC built out a fab, fully equipped it and ran it full forever. Then margins would be relatively low for the first five years, then when the equipment became fully depreciated it would go up, and then again after 10 years when the building systems were fully depreciated and then again after 15 years when the building was fully depreciated. But if you look at 5nm there are three fabs running 5nm in Taiwan of differing ages and capacity ramp rates and another 5nm fab is due to come on-line in the US in several years. TSMC 5nm capacity will double from end of 2020 to end of 2021 and then double again by the end of 2023. So the overall depreciation status of 5nm is very complex.
 
...and regardless of the quantitative financials, there's an implied question here:

- If there's been so much sustained demand at 40nm+ (increased demand for nodes on 200mm wafers have been in the news since at least 2015)

- and these older nodes have got higher profit margin

- is there any reason TSMC hasn't been slowly adding a little capacity gradually to the older fabs to provide at least some further increase in long-term profits? Even 5-10% over five years. Not because they want to be charitable and help all their customers still using "mature nodes", but because they want to increase the income of their business.

If they have, it's been done very quietly; I haven't run across any mentioned of it in the earnings materials on their website. If they haven't, is there a non-obvious reason? (for example, I'm just guessing: you can't add 5% more equipment to a full fab building, you'd have to put that equipment in a new fab building and then you would only have 5% of a fab so it wouldn't provide the advantages of scale as a fully-equipped building and the net profit would be much lower, so not worth it to try to add 5%; you'd have to add a big chunk like 25-50% and the financial numbers just don't work to support that)
Generally TSMC builds oversized cleanrooms so they can add equipment as needed. Some capacity also gets redistributed/combined.

Until recently TSMC had enough capacity at older nodes to service demand and at 28nm they actually had too much for a couple of years.

Also keep in mind that if they buy new equipment they have to absorb the depreciation which drags the margin down. As I discuss above it is the depreciation status of the equipment, not the age of the node.
 
"The specific kinds of products we identified as having significant semiconductor supply and demand mismatches are used by critical industries, including medical devices, broadband, and autos. They include:

~ Microcontrollers that are primarily made of legacy logic chips, including, for example, at 40, 90, 150, 180, and 250 nm nodes

~ Analog chips including, for example, at 40, 130, 160, 180, and 800 nm nodes; and

~ Optoelectronics chips including, for example, at 65, 110, and 180 nm nodes."



My thoughts about this official study of semiconductor shortage:

The report did not even include 28nm (TSMC started mass production in 2011) in the supply shortage list. Looking at those nodes in short supply, 40, 65, 110, 130, 150, 160, 180, 250, and 800nm, I'm thinking how lucky those automobile manufacturers are. Now in 2022 there are fabs who are still willing to make chips based on those old processes with cheap price. But on the other hand, is it reasonable to expect fabs to keep investing a lot more capital and resources into those old process nodes? For many of those companies crying for chip supply, do they ask for a lot but pay too little? Here I quote some text from Wikipedia about 800nm, a node adopted by the industry 30 years ago:

"The 800 nm process refers to the level of MOSFET semiconductor fabrication process technology that was reached around the 1987–1990 timeframe, by leading semiconductor companies like NTT, NEC, Toshiba, IBM, Hitachi, Matsushita, Mitsubishi Electric and Intel."


The reality is no matter how much government and private money will invest in the new equipment and new fabs, I doubt they are willing to spend too much on those legacy nodes mentioned. There is no quick solution for auto makers or any companies need those chips based on old process technology.

There is a quick solution for car makers, and at German ones been on it for a year+.

Just replace those chips with something else. Thousands of embedded developers are being vacuumed from the world market by car companies

"A horse, a horse, my kingdom for a horse"


I guess they will no longer be that lazy from now on with throwing random micros on every small task imaginable. What you see a lot in cars is a lot of MCUs connected to CAN bus just to listen to a single button on the control panel, and blink its LED. In any Chinese engineering shop, an employee will be skinned alive for such engineering decisions.
 
Definitely some incumbent fabs will invest more in old process nodes because they have captive customers. But my question is how far can this go in the long run? Can automobile companies expect they will have cheap and reliable supply of 180, 250, 800nm products 10 years from now?

If the answer is "no", then they have to move on to newer (not newest) process nodes. If they don't manage the risk of technology obsolescence, no one is obligated to save them.


What to invest into to? All 2nd-hand 200mm equipment on the market been vacuumed up 6-5 years ago.
 
I guess they will no longer be that lazy from now on with throwing random micros on every small task imaginable. What you see a lot in cars is a lot of MCUs connected to CAN bus just to listen to a single button on the control panel, and blink its LED. In any Chinese engineering shop, an employee will be skinned alive for such engineering decisions.
Which decisions are you talking about?

MCU-on-all-the-things is either an economic or a marketing decision.

It's a marketing decision if it adds some whiz-bang feature that no one really needs, but has the potential to attract more customers. (I can't think of any particular example unfortunately...)

It's an economic decision if you reduce wiring harness costs by removing low-tech wires sent all around the car, and replace them with power + CAN bus and a microcontroller.

Neither are "lazy" decisions. They do open up supply chain risks.
 
- If there's been so much sustained demand at 40nm+ (increased demand for nodes on 200mm wafers have been in the news since at least 2015)
- is there any reason TSMC hasn't been slowly adding a little capacity gradually to the older fabs to provide at least some further increase in long-term profits?
(for example, I'm just guessing: you can't add 5% more equipment to a full fab building...)
IMO the issue is two-fold, fab space as you mentioned, and tool availability (for 200mm or smaller wafer equipment). It is my understanding that basically no one makes 6" or 8" equipment anymore, and a large majority of the equipment that was previously installed at various IDMs has been sold on the resale market to the Chinese foundry players through the mid-10's. So if you are TSMC (or UMC, or GF, or Samsung, or...) the costs to build new fab locations to equip for legacy nodes is a financial burden too large to cross, while for the Chinese players with a slush fund of capital to burn, standing up a new factory and buying used / depreciated equipment on the cheap was a good strategy to get going. Customers on these nodes, if they have not already multi-fab/sourced the designs, would require a re-qual of the products in the event a new fab location and associated ECN; their downstream customers then also need to absorb and re-qual, etc.
What to invest into to? All 2nd-hand 200mm equipment on the market been vacuumed up 6-5 years ago.
So again, my understanding was 200mm equipment is basically only resale market, nothing new being built. Am I mistaken here? SK Hynix acquired Key Foundry and is in the process of rehoming those 8" tools to Wuxi (https://english.etnews.com/20220114200002), so even in the case of a Foundry of entirely used, depreciated tools, it is still cheaper to move that all out of Korea (presumptively due to difference in labor prices)?
 
TSMC reported gross margin is the weighted average of all their fabs running all their processes sold to all their customers.

- Large customers pay lower gross margins than small customers. Apple pays the lowest margins, a start up that needs a couple of engineering lots pays a much higher margin.
- Exchange rates effect margins, in the first half of 2021 TSMC margins were pulled down by exchange rate issues.
- Market forces effect margins, TSMC's margin increased in the second half of 2021 partly due to price increases and in 2022 announced price increases should drive up margins more.
- Margin is highly correlated with fab utilization, full fabs have lower costs and drive higher margins. For some older nodes TSMC had a period of excess capacity and lower margins trying to "buy" business.
- The node age isn't really what matters, it is the depreciation status of the equipment being used. This would be simple if TSMC built out a fab, fully equipped it and ran it full forever. Then margins would be relatively low for the first five years, then when the equipment became fully depreciated it would go up, and then again after 10 years when the building systems were fully depreciated and then again after 15 years when the building was fully depreciated. But if you look at 5nm there are three fabs running 5nm in Taiwan of differing ages and capacity ramp rates and another 5nm fab is due to come on-line in the US in several years. TSMC 5nm capacity will double from end of 2020 to end of 2021 and then double again by the end of 2023. So the overall depreciation status of 5nm is very complex.

OK, so the answer to my question was

(e) something else; I'm not even close, and I'm asking the wrong question.

:)

Maybe the right question has something to do with pricing/profitability/capex strategy at TSMC (or other foundries; sometimes I just use "TSMC" when I really mean the major foundries including Samsung/UMC/GlobalFoundries/SMIC). As a corporation, TSMC is trying to maximize profit over the long term by meeting the needs of its customers and servicing as much demand as they can profitably, while the marginal benefit of serving additional demand is still positive. So in one sense, they need to choose supply to be as well-matched as they can with demand trends.

But they also create long-term demand with increased capacity... ("if you build it they will come")

And they don't seem to be adding capacity at 40nm and larger...

But 40nm and larger is still profitable (26% of their revenue in 2021 Q4)...

And their pricing structure can incentivize demand (for example if they want to nudge customers towards newer technology, they should be raising pricing a little bit more on older nodes; I suppose they did back in August, since WSJ reported 10% increases for newer and 20% for older nodes)...

Hmm.

Let's try this:

Aside from the Aug 2021 price increases, and exchange-rate fluctuation, and economy of scale (Apple gets better pricing than Little Fabless Inc), what does TSMC's pricing for a particular process look like over time? After the fab equipment gets depreciated, does it stay more-or-less constant? (profitability goes up as building systems/building depreciation completes) During the first few years of fab equipment depreciation, does it drop to pass along some of the savings to the customers? or stay more-or-less constant?

How much of TSMC's income is supported by 40nm-and-larger? (Given that it's 26% of revenue)
- something like 28-30% (a little more margin than leading-edge)
- something like 30-38% (moderately more margin)
- something like 38-50% (a lot more margin)
- above 50% (insane secret cash cow)
 
IMO the issue is two-fold, fab space as you mentioned, and tool availability (for 200mm or smaller wafer equipment). It is my understanding that basically no one makes 6" or 8" equipment anymore, and a large majority of the equipment that was previously installed at various IDMs has been sold on the resale market to the Chinese foundry players through the mid-10's. So if you are TSMC (or UMC, or GF, or Samsung, or...) the costs to build new fab locations to equip for legacy nodes is a financial burden too large to cross, while for the Chinese players with a slush fund of capital to burn, standing up a new factory and buying used / depreciated equipment on the cheap was a good strategy to get going. Customers on these nodes, if they have not already multi-fab/sourced the designs, would require a re-qual of the products in the event a new fab location and associated ECN; their downstream customers then also need to absorb and re-qual, etc.
And that gets back to scary thoughts. Like, convoy of cars speeding down the road full speed, knowing the highway ends at this time, but having faith that someone's going to extend it, but nobody's extending it, hmm, please will someone extend it, let's keep going and see.... That's not the automakers or Tier 1 suppliers, that's their suppliers, the semiconductor companies.

Because, historically, the demand is there... here's a graph of TSMC's revenue over the last years (in NT$, culled from their quarterly reports; I haven't tried to convert to US$ yet) by feature size; they've been fairly consistent with reporting it, so it falls nicely into certain buckets.

1643400486318.png

Some processes die quickly (10nm), and others decrease over time before settling into near-oblivion (90nm, 110/130nm), but otherwise revenue from processes is remarkably consistent over the long haul (150/180nm and 65nm and 40/45nm in particular). You can see it trending upwards in older process nodes in the last 3 years or so, starting even before COVID. In fact, there's no COVID decrease here, just a hit in 2009 from the 2008 financial crisis, and some wiggles here and there over the years since.
 
Back
Top