Array
(
    [content] => 
    [params] => Array
        (
            [0] => /forum/index.php?threads/historical-examples-of-poor-decisions-to-build-a-new-fab.15104/
        )

    [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
)

Historical examples of poor decisions to build a new fab?

jms_embedded

Active member
Are there any "classic" examples of companies that have made overwhelmingly poor decisions to build a new fab? (e.g. borrow money, build fab, go out of business because of debt)

I know hindsight is 20-20, but I'm wondering if there are any cases where it just didn't make sense at all, and you have to scratch your head either to understand the reasoning, or to believe the bad luck. I'm going to assume that if there are, they are probably from the 1960s - 1990s, as the fabs nowadays are expensive enough that companies should be doing their due diligence before investing in significant CapEx projects.
 
Are there any "classic" examples of companies that have made overwhelmingly poor decisions to build a new fab? (e.g. borrow money, build fab, go out of business because of debt)

I know hindsight is 20-20, but I'm wondering if there are any cases where it just didn't make sense at all, and you have to scratch your head either to understand the reasoning, or to believe the bad luck. I'm going to assume that if there are, they are probably from the 1960s - 1990s, as the fabs nowadays are expensive enough that companies should be doing their due diligence before investing in significant CapEx projects.
I have a recollection from my time at TI that they used to sometimes help companies who wanted a fab get started. I don't remember if that was the case for Kobe Steel in Japan, but there was a short period where there was a huge DRAM price boom and all sorts of companies from unrelated industries with some spare cash thought "let's diversify into something more profitable". Like DRAM ! Needless to say, the over-capacity very quickly set in. I do wonder if we're heading that way again.

Exactly the same thing happened in the late 1990s tech boom. Suddenly everyone wanted to build next gen high capacity links for the booming internet traffic. There were only 3 companies worldwide capable of designing and laying the submarine cables - Alcatel was one (I was doing some consulting there at the time for a brief period). Their order book was off the scale - all sorts of weird consortia wanted to get in on the action. Again, over-capacity and the era of dark fibre. But brilliant news for consumers as we put in a decade of infrastructure in a few years and the investors picked up the tab. A bit like the mid 1850s railway boom.

[This was the era when Nortel had 90% of the new super-fast telecom switching market and were spending $250m a quarter on travel expenses. 10 years later they filed for bankruptcy. Before the Chinese cleaned out the Western telecoms switching equipment market].

And yes, TI helped Ericsson (a major customer) build a vanity wafer fab in Sweden. I head the Ericsson CEO got a Swedish knighthood for getting a leading edge wafer fab on Swedish soil.

Around the same time there were several new DRAM fabs built in the UK with heavy government support. Siemens on Tyneside (North East), LG in South Wales (never finished).

I'm sure this is just scraping the surface. Just the cases I remember.
 
Are there any "classic" examples of companies that have made overwhelmingly poor decisions to build a new fab? (e.g. borrow money, build fab, go out of business because of debt)

I know hindsight is 20-20, but I'm wondering if there are any cases where it just didn't make sense at all, and you have to scratch your head either to understand the reasoning, or to believe the bad luck. I'm going to assume that if there are, they are probably from the 1960s - 1990s, as the fabs nowadays are expensive enough that companies should be doing their due diligence before investing in significant CapEx projects.

The cases like you described are not rare. I think Wuhan Hongxin Semiconductor Manufacturing Co. (HSMC) may be one of the example.


Although HSMC crashed even before they can finish their "fab".

https://semiwiki.com/forum/index.php?threads/the-ruins-of-wuhan-hongxin-semiconductor.14927/
 
Last edited:
I have a recollection from my time at TI that they used to sometimes help companies who wanted a fab get started. I don't remember if that was the case for Kobe Steel in Japan, but there was a short period where there was a huge DRAM price boom and all sorts of companies from unrelated industries with some spare cash thought "let's diversify into something more profitable". Like DRAM ! Needless to say, the over-capacity very quickly set in. I do wonder if we're heading that way again.

Exactly the same thing happened in the late 1990s tech boom. Suddenly everyone wanted to build next gen high capacity links for the booming internet traffic. There were only 3 companies worldwide capable of designing and laying the submarine cables - Alcatel was one (I was doing some consulting there at the time for a brief period). Their order book was off the scale - all sorts of weird consortia wanted to get in on the action. Again, over-capacity and the era of dark fibre. But brilliant news for consumers as we put in a decade of infrastructure in a few years and the investors picked up the tab. A bit like the mid 1850s railway boom.

[This was the era when Nortel had 90% of the new super-fast telecom switching market and were spending $250m a quarter on travel expenses. 10 years later they filed for bankruptcy. Before the Chinese cleaned out the Western telecoms switching equipment market].

And yes, TI helped Ericsson (a major customer) build a vanity wafer fab in Sweden. I head the Ericsson CEO got a Swedish knighthood for getting a leading edge wafer fab on Swedish soil.

Around the same time there were several new DRAM fabs built in the UK with heavy government support. Siemens on Tyneside (North East), LG in South Wales (never finished).

I'm sure this is just scraping the surface. Just the cases I remember.

 
The cases like you described are not rare. I think Wuhan Hongxin Semiconductor Manufacturing Co. (HSMC) may be one of the example.


Although HSMC crashed even before they can finish their "fab".

https://semiwiki.com/forum/index.php?threads/the-ruins-of-wuhan-hongxin-semiconductor.14927/
Yes I'd quite forgotten the carnage going on/in the pipeline in China at the moment. If even a fraction of what is reported about some of these ventures is true there's a lot more to come.
 
Thanks both of you for helping point me at some names and places!

And yes, TI helped Ericsson (a major customer) build a vanity wafer fab in Sweden. I head the Ericsson CEO got a Swedish knighthood for getting a leading edge wafer fab on Swedish soil.

(Nov 1992)
DALLAS -- Texas Instruments Inc. and Sweden's L.M. Ericsson announced Thursday a cross-licensing agreement to develop semiconductor components for telecommunications applications.

TI said it will share with Ericsson its knowledge, technical support and rights to manufacturing of sub-micron integrated circuits for Ericsson's telecom systems using TI's .5-micron and .35-micron process technologies.

A team of Ericsson process and equipment engineers is enrolled in a training program at TI's Dallas headquarters. The curriculum includes process and prototype developments, design rules and semiconductor manufacturing steps.

Ericsson plans to build a prototype semiconductor wafer fabrication facility at Kista, Sweden. The sub-micron facility will have a planned capacity of about 1,000 wafers per month and is expected to begin pilot production in 1994.

TI said it plans to meet Ericsson's volume production requirements through one or more of its advanced sub-micron facilities.

[article goes on]
OK so a fab in Kista, Sweden; let's see where that history leads....

(Oct 1995)
Ericsson adopted a rather different approach when it planned a submicron fab at its Kista site in Sweden. Wafer manufacturing volumes were less important than time factors, since the principal role of the fab is to reduce the design and manufacturing lead times for the complex ASICs needed in the company's telecommunications products and systems. In this paper, the author describes Ericsson's rather unusual strategy for microelectronics sourcing.

(Sep 2003)

LUND, Sweden The Swedish government is scrambling to keep open at least one of the nation's last two operating fabs, both now owned by Infineon Technologies AG and both scheduled to be boarded up.

As the government searches for a prospective buyer, one fab's closing is weeks away while the second, a submicron facility, is scheduled to shut down within the next year. Both facilities had been owned by Ericsson Microelectronics in Kista, near Stockholm, and were acquired when Infineon bought Ericsson's internal semiconductor operations last year. When the deal was announced in June 2002, the purchase price was put at about $378 million.

The first shutdown is scheduled at a high-voltage bipolar and low-density MOS fab, internally called Fab 51, while the later closing is at Fab 6, which is capable of running a 0.18-micron process.

Infineon took over Ericsson Microelectronics' business to expand its wireless infrastructure and mobile phone portfolio. The acquisition included 400 Ericsson employees in Kista and 200 in Morgan Hill, Calif., most of them R&D engineers. In negotiating the acquisition, the companies agreed to close both Ericsson manufacturing facilities, a spokeswoman at Infineon Technologies Wireless Solutions AB said.

...

Details of the submicron fab's production capacity, the types of equipment already installed, its work force and other specifics have not been made public. Swedish government agencies now trying to structure a deal have not gotten Ericsson's approval for such disclosure. The manufacturing equipment in the submicron fab and its work force are leased to Infineon, but they are still Ericsson's property and employees, sources said. The fab today is completing orders already placed by Ericsson and others.

The fab reportedly can turn out about 450,000 wafers per year, making it “too small for mass production and too big for R&D prototyping, in my personal opinion,” said Engelmark. Whatever company takes over the facility must figure out the right products high-value added as well as relatively low volume.

Despite the government's enthusiasm, Ericsson is no longer interested in running a submicron fab. “The fab was once an important part of our operation, when we were trying to control everything at Ericsson,” said Pia Gideon, vice president at Ericsson. But as the Swedish telecom equipment giant has gone through a host of restructurings and become heavily dependent on outsourcing, Ericsson has determined that it is no longer economical to keep its fab.

Sandeep Chennakeshu, president of Ericsson Mobile Platforms, agreed. His unit, responsible for system engineering for mobile handsets, not only develops software for Ericsson mobile platforms but also does silicon design. But Chennakeshu said that the fabrication of such chips is best left to chip vendors already invested heavily in the advanced process technologies for 90-nanometer or smaller geometry, he said. “The key is to establish good partnerships with a few chip companies.”

...

That actually doesn't sound awful. Maybe not the best from a profit/loss on financial grounds only, but I guess they were trying to learn quickly. Expensive in dollars but yielding some intangibles.
 
Last edited:
I have a recollection from my time at TI that they used to sometimes help companies who wanted a fab get started. I don't remember if that was the case for Kobe Steel in Japan, but there was a short period where there was a huge DRAM price boom and all sorts of companies from unrelated industries with some spare cash thought "let's diversify into something more profitable".
bizarre...

(Nov 2001)
KYOTO, Japan –Kobe Steel Ltd. and Dainippon Screen Mfg. Co. Ltd. today announced a partnership to jointly develop next-generation semiconductor fabrication systems using supercritical fluid technology. Supercritical fluids have densities approaching liquids while possessing viscosities and diffusivity properties of gases, making them ideally suited for wafer cleaning and drying applications, said the two companies.

Dainippon Screen, based in Kyoto, is one of the world' top10 suppliers of chip-manufacturing gear and a leading producer of photolithography resist coat-and-develop equipment as well as a major producer of wafer-cleaning tools. Tokyo-based Kobe Steel supplies supercritical fluid process technology and high-pressure, contamination-free equipment.

The two Japanese companies said they plan to combine their expertise to co-develop and market new wafer fab systems using supercritical fluid technology in three areas:

  • “Super Critical C02 ” (known as SCC02 ) as a post-develop drying agent for sub-100-nm lithography;
  • Wafer cleaning using SCC02 enhanced with co-solvents;
  • and,
  • SCC02 in post-clean drying applications.
    Dainippon Screen and Kobe Steel said these types of supercritical fluid-based cleaning and drying systems will be needed within three years for mass production of ICs, based on device feature sizes less than 100 nm (0.10 micron).
Huh, that's more of a process research fab.

This article from Cadence has lots of interesting history....

(May 2009)
 
The only reason IBM ever built a site in Vermont was that the Watson heir loved to ski. The site commercialized the research coming out of the Hudson Valley by training farm boys to build electronics and later to design and manufacture DRAM ("memory capital of the world" at one point). But really it was a dream enabled by the monopoly, and when the monopoly was broken Vermont could not compete with NY tax breaks and energy prices. The 300mm fab went to Fishkill under the assumption that their experienced employees would migrate south. They did not, and IBM suffered generations of terrible yields that drove many nails into their semiconductor coffin today.
 
This is only a part of IBM Microelectronics demise. The Fishkill 300mm fab (one of the first 300mm built) came at the same time IBM decided to stop manufacturing things. They quickly went on a binge selling off printers, hard drives, desktop and laptop computers, only leaving mid/large servers (which much was subcontracted subassemblies) and microelectronics (to provide processors and such for those servers). The Vermont fab, then eventually Fishkill, moved to a foundry model. But while they had the technical chops to develop value products, neither had the manufacturing scale to be cost competitive, or even to produce the number of units customers required.
 
But while they had the technical chops to develop value products, neither had the manufacturing scale to be cost competitive, or even to produce the number of units customers required.

Makes sense although I'm not sure it absolves them of getting played like chumps by cynical ex-IBM managers at GF collaborating with McKinsey, although I could be Monday morning quarterbacking given the supply chain crunch and CHIPS. It's telling that the guy who did Fishkill would go on to do Watson, which did a fine job propping up brand value in the eyes of the 60+ Jeopardy demographic and nothing else.
 
I have a recollection from my time at TI that they used to sometimes help companies who wanted a fab get started. I don't remember if that was the case for Kobe Steel in Japan, but there was a short period where there was a huge DRAM price boom and all sorts of companies from unrelated industries with some spare cash thought "let's diversify into something more profitable". Like DRAM ! Needless to say, the over-capacity very quickly set in. I do wonder if we're heading that way again.

Exactly the same thing happened in the late 1990s tech boom. Suddenly everyone wanted to build next gen high capacity links for the booming internet traffic. There were only 3 companies worldwide capable of designing and laying the submarine cables - Alcatel was one (I was doing some consulting there at the time for a brief period). Their order book was off the scale - all sorts of weird consortia wanted to get in on the action. Again, over-capacity and the era of dark fibre. But brilliant news for consumers as we put in a decade of infrastructure in a few years and the investors picked up the tab. A bit like the mid 1850s railway boom.

[This was the era when Nortel had 90% of the new super-fast telecom switching market and were spending $250m a quarter on travel expenses. 10 years later they filed for bankruptcy. Before the Chinese cleaned out the Western telecoms switching equipment market].

And yes, TI helped Ericsson (a major customer) build a vanity wafer fab in Sweden. I head the Ericsson CEO got a Swedish knighthood for getting a leading edge wafer fab on Swedish soil.

Around the same time there were several new DRAM fabs built in the UK with heavy government support. Siemens on Tyneside (North East), LG in South Wales (never finished).

I'm sure this is just scraping the surface. Just the cases I remember.

I was in South Wales when LG came calling.
They were going to build a FAB and TV manufacturing plant, interviewed for their training program.
Though the folks who built the factory shells got employed , very little else was done as the FAB never started and the TV facility got crushed as we moved away from CRT monitors n TVs

Sent a couple of photomasks to Tyneside when it was under Atmel , but that didnt last long either.
 
IDK if White Oak / Infineon / Qimonda's memory fabs in Virginia were ever profitable, so maybe that qualifies as far as something in the USA.

- https://www.vedp.org/press-release/1997-10/white-oak-semiconductor-construction
- 2009 bankruptcy: https://www.eetimes.com/judge-approves-tis-purchase-of-qimonda-assets/
- https://www.pressreader.com/usa/richmond-times-dispatch/20160627/page/98/textview

QIMONDA (INFINEON TECHNOLOGIES AND WHITE OAK SEMICONDUCTOR​

  • Richmond Times-Dispatch
  • 27 Jun 2016
White Oak opened a memory chip-manufacturing plant in eastern Henrico in 1998 as a joint venture between Motorola and Siemens AG. Siemens separated its semiconductor business as Infineon Technologies in 2000. Infineon created Qimonda as a separate company in 2006. The local plant closed in February 2009, laying off thousands of workers, as the parent company filed for bankruptcy protection. Quality Technology Services acquired the 1.3 million-squarefoot former plant in 2010 and retrofitted it as a data center.
 
Back
Top