Array
(
    [content] => 
    [params] => Array
        (
            [0] => /forum/index.php?threads/tsmc-to-build-new-fab-in-kaohsiung-taiwan.14940/
        )

    [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
)

TSMC to build new fab in Kaohsiung Taiwan

hist78

Well-known member

From other Taiwanese news outlets it's reported that the new fab will be built on a 200-acre former oil refinery. The fab will be producing TSMC's N7 and N28 related products in 2024.

TSMC started their 28nm production in 2011 and it's interesting to see a new fab for this old node 13 years later in 2024.
 
Last edited:
From my post in another thread:

In the past high performance CPUs drive adoption of leading edge processes. As the high performance CPUs move to the leading edge, it frees up capacity on old nodes, which are adopted by companies with less stringent performance requirements. This model broke at 28nm and the move to FinFET. Design costs to move past 28nm became prohibitive and many chips ended up getting stuck here, creating a bottleneck at 28nm. These companies are not migrating causing demand for this node to go up while foundries were not building enough 28nm capacity assuming that companies would eventually migrate but they haven't and now >28nm is where shortages are concentrated.

Also, this is why some companies like Tesla, that have in house semiconductor design capabilities are not facing shortages to the same extent as other companies... Tesla has the ability to migrate it's designs while other companies may not.
 
7nm fab is good. why 28 nm fab?
Due to customers' demands, I guess. As long as TSMC is confidently to have enough orders to justify the cost, why not? 28nm is a very successful node for TSMC. The production cost should be low by now and the technical difficulty for them is very little, if any.
 
From my post in another thread:

In the past high performance CPUs drive adoption of leading edge processes. As the high performance CPUs move to the leading edge, it frees up capacity on old nodes, which are adopted by companies with less stringent performance requirements. This model broke at 28nm and the move to FinFET. Design costs to move past 28nm became prohibitive and many chips ended up getting stuck here, creating a bottleneck at 28nm. These companies are not migrating causing demand for this node to go up while foundries were not building enough 28nm capacity assuming that companies would eventually migrate but they haven't and now >28nm is where shortages are concentrated.

Also, this is why some companies like Tesla, that have in house semiconductor design capabilities are not facing shortages to the same extent as other companies... Tesla has the ability to migrate it's designs while other companies may not.

It also poses a challenge for IDM and any new and existing players in the foundry business. When TSMC's manufacturing technologies go smaller and more powerful every year, TSMC doesn't give up their existing customers and market.

TSMC today is still making 28nm products that they first started in 2011. If the new comers are thinking to replace or compete against TSMC in the mature nodes, the potential market for them is limited.
 
Last edited:
7nm fab is good. why 28 nm fab?

Not to let any mainstream competitors to raise a head. They even hinted at new 200mm capacity... That would be quite hard to pull though. All 200mm equipment is being vacuumed off the market.
 
Due to customers' demands, I guess. As long as TSMC is confidently to have enough orders to justify the cost, why not? 28nm is a very successful node for TSMC. The production cost should be low by now and the technical difficulty for them is very little, if any.

The problem with 28nm is that UMC, GF, and SMIC have compatible processes. Meaning that you can take a TSMC 28nm design to the other fabs without modification. Clearly TSMC can make a better and cheaper 28nm process with all of their past experience behind them but why go there? Other than to CRUSH the competition...
 
The problem with 28nm is that UMC, GF, and SMIC have compatible processes. Meaning that you can take a TSMC 28nm design to the other fabs without modification. Clearly TSMC can make a better and cheaper 28nm process with all of their past experience behind them but why go there? Other than to CRUSH the competition...
isn't that assumed TSMC can make cheaper or better 28nm than others? since 28nm is a mature tech...what's TSMC's edge?
 
The problem with 28nm is that UMC, GF, and SMIC have compatible processes. Meaning that you can take a TSMC 28nm design to the other fabs without modification. Clearly TSMC can make a better and cheaper 28nm process with all of their past experience behind them but why go there? Other than to CRUSH the competition...

It's an interesting situation. If TSMC builds more capacity for the mature nodes like 28nm, it will allow TSMC to satisfy increased demands from existing and new customers. Although potentially it will squeeze out other foundries' expansion based on those matured nodes.

But if TSMC doesn't increase their mature nodes capacity, they will be accused as careless and ignorant.
 
isn't that assumed TSMC can make cheaper or better 28nm than others? since 28nm is a mature tech...what's TSMC's edge?
It's a fact. Economies of scale, ecosystem, and the broad range of customers that TSMC serves. Boutique foundries can chip away at the edges of TSMC's business but they cannot get to the core which is the high volume chips amongst all of the markets.
 
I wonder how depreciation of TSMC's new equipment at an older node will impact its wafer costs... won't they be competing against others with fully depreciated toolsets?
 
I wonder how depreciation of TSMC's new equipment at an older node will impact its wafer costs... won't they be competing against others with fully depreciated toolsets?
Considering TSMC started 28nm in 2011and have been doing well on it since 2011, unless it's a new equipment bought in recent years, TSMC's 28nm should be very competitive and profitable.

"Depreciation is recognized so as to write off the cost of the assets less their residual values over their useful lives, and it is computed using the straight-line method over the following estimated useful lives:

buildings - 10 to 20 years; machinery and equipment - 2 to 5 years; and office equipment - 3 to 5 years. The estimated useful lives, residual values and depreciation method are reviewed at the end of each reporting period, with the effect of any changes in estimates accounted for on a prospective basis. Land is not depreciated."


Source: https://investor.tsmc.com/sites/ir/financial-report/2018/2018Q4-E.pdf
 

From other Taiwanese news outlets it's reported that the new fab will be built on a 200-acre former oil refinery. The fab will be producing TSMC's N7 and N28 related products in 2024.

TSMC started their 28nm production in 2011 and it's interesting to see a new fab for this old node 13 years later in 2024.
I have been looking into this, there are couple of reports in the press in Taiwan but nothing from TSMC. There are press releases from TSMC about the Fab in Japan but I can't find anything about this fab. Also, 7nm and 28nm in the same fab makes no sense to me. TSMC's 7nm business is all migrating to their EUV based 7+/6nm process and I wouldn't think it would make sense to combine that with non EUV 28nm. Not sure I believe this is real.
 
I have been looking into this, there are couple of reports in the press in Taiwan but nothing from TSMC. There are press releases from TSMC about the Fab in Japan but I can't find anything about this fab. Also, 7nm and 28nm in the same fab makes no sense to me. TSMC's 7nm business is all migrating to their EUV based 7+/6nm process and I wouldn't think it would make sense to combine that with non EUV 28nm. Not sure I believe this is real.
TSMC board of directors approved the Kaohsiung fab project on November 9, 2021, the same day TSMC announced their Sony/TSMC Japan fab. According to news report, Mayor of Kaohsiung stated in a City Council meeting last month that the fab construction will start in June 2022.


 
The problem with 28nm is that UMC, GF, and SMIC have compatible processes. Meaning that you can take a TSMC 28nm design to the other fabs without modification. Clearly TSMC can make a better and cheaper 28nm process with all of their past experience behind them but why go there? Other than to CRUSH the competition...
(my emphasis)

Is this really true? I get that digital logic would be fairly straightforward to migrate on the right process, but what about analog? Would other fabs be able to match the characteristics exactly?
 
TSMC board of directors approved the Kaohsiung fab project on November 9, 2021, the same day TSMC announced their Sony/TSMC Japan fab. According to news report, Mayor of Kaohsiung stated in a City Council meeting last month that the fab construction will start in June 2022.


The TSMC board resolutions for the November 9th meeting mention the Japan Fab but don’t say anything about Kaohsiung.

i have seen the news reports you mention but the lack of an official TSMC statement makes me doubt them.
 
Last edited:
The recent story of TSMC 28nm (or legacy nodes) is actually quite entertaining. Prior to covid, TSMC charges about 10-20% higher than its peers. For a long time, 28nm’s utilization rate was not impressive. As I recall, the TSMC chair once said the world had enough 28 nm Fabs. Then chip shortage hit, other foundries raised the wafer price much more than TSMC. As a result, TSMC’s old customers are back in droves thanks to the easily portable CMOS design. And here comes an unexpected and powerful customer-Intel.

Intel CPU division is massive, and so does non-CPU divisions. On Pat’s last year’s trip to Taiwan, all back-end companies were in his Hsinchu hotel’s lobby waiting to be summoned. It was quite a scene. At the end of the day, everyone had a wonderful Christmas gift from him. He has done something dramatic to his non-CPU divisions in case nobody notices.

TSMC has added more 28 nm capacity in Nanjing (60KWPM), Kumamoto (estimated > 50KWPM), and Kaohsiung (estimated > 50KWPM), compared to a capacity of 180KWPM pre-covid. As I read it somewhere, “It is a serious crime if you failed to meet your customers’ demands”.
 
The recent story of TSMC 28nm (or legacy nodes) is actually quite entertaining. Prior to covid, TSMC charges about 10-20% higher than its peers. For a long time, 28nm’s utilization rate was not impressive. As I recall, the TSMC chair once said the world had enough 28 nm Fabs. Then chip shortage hit, other foundries raised the wafer price much more than TSMC. As a result, TSMC’s old customers are back in droves thanks to the easily portable CMOS design. And here comes an unexpected and powerful customer-Intel.

Intel CPU division is massive, and so does non-CPU divisions. On Pat’s last year’s trip to Taiwan, all back-end companies were in his Hsinchu hotel’s lobby waiting to be summoned. It was quite a scene. At the end of the day, everyone had a wonderful Christmas gift from him. He has done something dramatic to his non-CPU divisions in case nobody notices.

TSMC has added more 28 nm capacity in Nanjing (60KWPM), Kumamoto (estimated > 50KWPM), and Kaohsiung (estimated > 50KWPM), compared to a capacity of 180KWPM pre-covid. As I read it somewhere, “It is a serious crime if you failed to meet your customers’ demands”.
Your numbers and locations are all wrong. According to TSMC's web site they have Fab 12A/12B in Hsinchu, Fab 14 and 18 in Tainan, Fab 16 in Nanjing and Fab 15 in Taichung.

From my research, at one time they made some 28nm in Hsinchu and may make some there still, say 20 to 30k wpm. Their main 28nm manufacturing is Fab 15 phases 1 through 4 in Taichung with around 160k wpm capacity between the 4.

Nanjing is used to make 16nm/12nm, not 28nm.

Kumamoto is in Japan and that fab hasn't even started construction yet.

Kaohsiung is a rumored site for a future fab but TSMC hasn't even announced it yet.
 
Your numbers and locations are all wrong. According to TSMC's web site they have Fab 12A/12B in Hsinchu, Fab 14 and 18 in Tainan, Fab 16 in Nanjing and Fab 15 in Taichung.

From my research, at one time they made some 28nm in Hsinchu and may make some there still, say 20 to 30k wpm. Their main 28nm manufacturing is Fab 15 phases 1 through 4 in Taichung with around 160k wpm capacity between the 4.

Nanjing is used to make 16nm/12nm, not 28nm.

Kumamoto is in Japan and that fab hasn't even started construction yet.

Kaohsiung is a rumored site for a future fab but TSMC hasn't even announced it yet.
TSMC is adding 28nm 40K monthly wafer capacity to their Nanjing fab. I believe it can go into production in late 2022 and reach full capacity around 2023 because it's using existing Nanjing fab's building.
 
Last edited:
Back
Top