Taiwans Dram Failure

TITLE: Taiwan’s DRAM Failure CHANNEL: Asianometry DATE: ---TRANSCRIPT--- We often associate Taiwan with chips. Taiwanese chips. It’s their thing right? But Taiwan’s strength is actually only in logic chips. In the industry’s other big sector,
memory and DRAM memory in particular, Taiwan is second-tier. Hardly a player. It’s not that the Taiwanese haven’t tried to break into DRAM before. In fact,
they spent billions trying for two decades. They just keep losing at it over and over again. In this video,
we look at Vanguard, TI-Acer, Taiwan Memory Corporation and Taiwan’s DRAM failure.

Beginnings

In prior videos, I discussed Taiwan’s early successes in semiconductors. In the 1970s, a Taiwanese government-funded research institute called ITRI paid a few
million dollars to acquire a 7.5 micron CMOS process from RCA. In May 1976, Taiwan sent 35-40 individuals to RCA’s plant for 6-12 months to learn this
process node. Then ITRI transferred its acquired technology and facilities to the
private sector, founding two companies that became incredible success stories. The first, UMC, pioneered Taiwan’s semiconductor industry and later spun off its leading
fabless designer, MediaTek. They remain the world’s fourth largest foundry by revenue. The second company is of course TSMC. They need no further preamble. Both well-complemented Taiwan’s electronics and contract OEM industries, providing custom logic
chips or ASICs to accelerate their growth. With these twin successes under their belt,
the Taiwanese government then turned to the question of DRAM.

Going Into DRAM

Now, DRAM is a famously ferocious industry, with economic dynamics nothing like logic. Used in large quantities for computers, PCs, and consumer devices, DRAM is a commodity infamous
for its volatility. In the early days, you often had four good years and 1 bad year. The space was also crowded with global players. You had the Japanese,
the South Korean memory giants, and of course the Americans with IBM and Micron Technology. Success in the field required incredible R&D executed and flawlessly scaled every
two or three years. I have before covered the extensive work done to squeeze ever more
capacitors into the same real estate. Failure means falling out of the race. Was it wise to compete in such a crowded, R&D-intensive,
and high-stakes field? ASICs and other logic chips were less competitive and
better fit Taiwan’s business landscape of small to medium sized enterprises. Taiwan had even recently suffered its own painful failures in DRAM. In 1984,
the Taiwanese government raised $16 million of funding from banks and
investors including the PC giant Acer Group to start Quasel Taiwan Company (國善電子) in Hsinchu. Quasel had been founded by a Chinese-American entrepreneur with experience at both Burroughs
and Xerox. They sought to do R&D from the ground up. But as early as 1985,
the company started to suffer leadership issues. In-house fabrication began in January 1986, only to run head long into a brutal DRAM downturn,
with prices falling from $3-4 to just $1. Quasel ended up running out of money by mid-1986. Two other memory firms - Mosel and Vitelic

• also failed to make ground in the industry
  either. They eventually merged and will appear again later in this video. So anyway,
  the DRAM sector just seemed like cursed land for the Taiwanese.

A DRAM-atic Shortage

But then in 1988, a year after TSMC’s founding, there was a severe DRAM shortage. This global shortage had several causes. On the demand side,
there was a major surge in sales of PC clones. PCs were shifting
from older 8-bit CPUs to newer 16 and 32-bit CPUs, which needed more memory. And there was weirdness on the supply side. Several American producers had
exited the market after the aforementioned brutal downturn of 1986. The Japanese were
also transitioning their fabs from 256K to 1 megabit DRAMs, and had not yet fully ramped. The US press called it the “Great DRAM Shortage”. Prices in the third quarter
of 1988 surged to previously unseen heights. Computer, PC, and consumer device manufacturers
in both Taiwan and the US found it hard to get enough memory to ship product. The shortage underlined memory’s importance to Taiwan’s PC companies
and electronics ecosystem. They consumed increasingly more memory,
but depended on unreliable allocations given to them by the Japanese or Americans. During the shortage, Jim Dykes, TSMC’s first CEO,
recalled being pressured by Taiwan’s PC makers to produce DRAMs as well as ASICs.
Dykes refused, fully believing in the approach of staying only in logic chips. Though later on, TSMC did produce a few DRAMs for customers. They also offered
embedded memories like eDRAMs - special memories that sit right alongside the logic transistors.
They still offer these embedded memories today, though different ones like STT-MRAM.

Acer and ITRI

The first out with a DRAM venture was Stan Shih of Acer Group - then Taiwan’s leading PC company. Some time in late 1988, Shih formulated the bold idea to start an internal DRAM memory
fab inside Acer Group - gambling that future internal demand would be immense
and any leftover supply can also be sold to other Taiwanese companies. To shortcut time to market, Shih would license a DRAM process node from Texas Instruments. Shih
recalled in an oral history for the Computer History Museum that he talked to Sharp, NEC,
and others too but only TI, a former customer, was interested. The 1988 DRAM shortage also led ITRI’s vice president Dr. Shih Chin-tay to lobby for
ITRI to do DRAM too. Dr. Shih is a legend in Taiwanese semiconductor history - one of the
original engineers sent to the RCA factory in the 1970s. No relation to Stan Shih. In a 2011 oral history for the Computer History Museum, he noted his thinking about this path:

It happened in 1988 and 1989, we were short of memories [globally]. Therefore,
enterprises called ITRI for help … this could not be done in a single day,
it requires a long-term R&D [investment] Because our computer industry had been growing, people came to realize that
components had begun to influence the future. That’s when Stan Shih began his investment in
ASM. It was probably in the 90’s. ITRI had started thinking about this in the 80’s

The Submicron Project and TI-Acer

In 1989, ITRI kicked off the Sub-micron Memory Production
Technology Development Project, or just the Submicron Project. At about the same time, Acer cemented its collaboration with TI,
investing $185 million to start a DRAM joint venture called TI-Acer. It is important to know that this decision was extensively debated even as the Taiwanese were
making it. Yes, the DRAM market was competitive, but proponents had points to make too. The PC market was both growing and using more memory. In 1990, Microsoft released
the Windows 3.0 operating system, which featured a visual GUI with clickable applications. A big
commercial success, Windows 3.0 forced all PC vendors to upgrade their memory specs. And regarding the competition issue, the aforementioned devastating 1986
DRAM downturn washed out several big players, particularly in the United States. Like Intel! And with the 1985 Plaza Accord re-valuing the Japanese Yen upwards,
the Taiwanese and Koreans enjoyed a currency advantage. I should also mention that Samsung, Hyundai, and the Koreans had not yet
become the giants they were going to be. It seemed possible then to take them on. Even so, it was still going to be hard. Morris Chang recounted in his memoirs that the Submicron
Project had a staggeringly ambitious goal: To research and develop DRAM chips that
can compete at the leading edge against the Japanese, Koreans, and Americans in five years. And then there was the issue of who will design the memory. TSMC customers brought
their own designs, which lets TSMC entirely focus on the manufacturing. But memory-makers must design their memory, and it is a critical part of their success.
Neither ITRI nor TI-Acer owned memory design IP or had experience in memory design. ITRI
would be starting from scratch and TI-Acer was essentially borrowing TI’s technology via license. Both this went against Morris’ principle of owning the core
technology. He mused in his memoirs whether TSMC’s success as a spinoff
had been “too easy” - adding that the “tragic outcome was almost inevitable”. TI-Acer’s build greatly strained Acer Group’s financials in the early 1990s. The PC market
had crashed, causing the company to turn a $22 million loss in 1991 and layoffs.
But it finally started production in July 1991, with the two partners sharing the DRAMs output.

Vanguard

The Submicron Project leaned heavily on the efforts and knowledge of two
Chinese returnees from the United States. They were Dr. Chih-Yuan Lu, who had worked on memory processes at Bell Labs,
and his younger brother Chao-chun “Nicky” Lu, who had worked at IBM. The elder Lu agreed to join the project, but his younger brother
wanted to start a company - which is today the publicly traded Etron Technology. So
Etron participated as a contractor, contributing some critical designs. Despite challenges in procuring talented Taiwanese designers,
the Submicron Project set up an 8-inch wafer fab in Hsinchu and
per ITRI’s own milestone webpage, produced an advanced 16-megabit DRAM product in 1993. Though Morris Chang in his memoirs noted that the actually salable product was
a 4-megabit DRAM produced by Etron, the aforementioned Nicky Lu’s outfit. In 1994,
4-megabit was behind the leading edge but still good enough to sell for some profit. That same year, TSMC went public on the Taiwan Stock Exchange,
the stock price soaring 10 times its offering price and making many OG TSMC
project members very rich. The news motivated the Submicron Project team to push for a spinoff. This spinoff privatization was done differently than how it
was done with TSMC or UMC. UMC had been done as a straight transfer.
TSMC was set up as a joint venture with a foreign company, Philips. The Submicron Project on the other hand was privatized via an auction.
Despite his misgivings about the project and memory in general,
Morris Chang put together a group with his right-hand man FC Tseng and raised
investor money to win that auction and start Vanguard International Semiconductor (世界先進). Vanguard began in late 1994 with 330 employees transferred to the
company from ITRI, along with two product designs: The aforementioned
4 megabit DRAM product ready for sale and the 16-megabit design still in development.

Private Entry: Startups

Vanguard and TI-Acer led the way but they were not alone. A bunch
of private DRAM startups followed too.

They were backed with funding from Taiwan’s
diversified business groups - which were then flush with cash - and licensed IP from the
Japanese memory-makers - who were fading and saw licensing as an easy way to make a few dollars. The first of these arrangements was Nanya Technology Corporation (南亞科), established
in March 1995. Backed with money from the Formosa Plastics conglomerate and a
technology licensed from the Japanese firm Oki Electric, pilot DRAM production began in 1996. The second was Powerchip Semiconductor (力晶積成電子製造). Founded in Hsinchu in 1994 by Frank Huang and
others, they licensed a DRAM process node from Mitsubishi and started production in 1996. The third was ProMOS. They kicked off 1996 as a joint venture between the aforementioned
Taiwanese legacy firm Mosel-Vitelic and Siemens Semiconductor in Germany,
which contributed the IP for a 64megabit DRAM.

Private Entry: The “Me Too’s”

This DRAM gold rush was augmented by established Taiwanese companies also going into the field. The largest of these “me-too”, fast-follower DRAM-makers was Winbond. Winbond Semiconductor was founded back in 1987. They have an interesting founding story. Their
founder and assets came from ITRI, but their money came from a Taiwanese wire,
cable and metal conglomerate called Walsin Lihwa. Winbond started off providing logic chips for wired telephones, consumer electronics,
and computers before diversifying into NOR flash non-volatile memory
in 1995. That is and remains a pretty good business for them. In late December 1995, they announced a partnership with Toshiba to go make DRAMs.
Their 16-megabit DRAM fab - located outside of Taipei - cost about $1.2
billion and would produce 30,000 wafers a month. In addition to Winbond, there is Macronix,
a Taiwanese company founded in 1988 by former ITRI researchers. Originally, they also started off in non-volatile memories - mask ROMs and EPROMs. Then they too
got into DRAM, but only embedded types. The stuff that sits alongside logic transistors. We have to mention the vintage OGs like UMC and Mosel-Vitelic - yes,
separate from ProMOS - also producing some amount of DRAMs for the love of the game. Even Tatung, the ancient electronics company known more for rice cookers and fans,
declared plans to build a DRAM fab, though thankfully they never went
through with it. It was a gold rush, and that was the danger.

Initial Optimism

In a prior video, I discussed the Beer Game, and how it takes time for semiconductor end
user demand signals to filter down to producers like Samsung or TSMC. And then it takes time for those companies to go to their suppliers and ramp up production
in response. Unfortunately, the time it takes for these signals
to travel up or down the supply chain can cause a painful “bullwhip” effect. The DRAM cycle of the mid-1990s illustrates this well. In 1994,
the PC DRAM industry enjoyed a massive supercycle in anticipation of the
forthcoming Windows 95 operating system and its higher RAM requirements of 4-8 megabytes. Such times often snooker memory-makers into thinking that things are different
this time so they go build a bunch of expensive factories. And I mean,
really expensive. A memory fab in the 1990s cost something like a billion dollars. In 1994, Taiwan’s total DRAM production volume was worth about $600 million,
good for 2.5% market share of the $24 billion market. In 1995, new 8-inch wafer fabs brought online by Vanguard and
TI-Acer more than doubled that production to $1.25 billion. Morris Chang’s memoir related how Vanguard International began operations in January
1995 with optimism and energy. Their 4 megabit DRAM product still sold at a
good price of $12 to $15 per chip, with cash costs of about $5 or $6 per chip. The world DRAM market surged to $40 billion in 1995 - making $2 billion
of profits each month. In August 1995, the company broke even. Vanguard’s CEO
Bob Evans celebrated with employee bonuses of a full month’s pay. It was a good time. Unfortunately it would not last. In late December 1995,
Evans phoned Morris in alarm, telling him that 4-megabit DRAM prices were crashing.

Bullwhipped

DRAM-makers all thought that the Windows 95 and Internet boom had
changed things forever. So they built and built. They built too much. Nine new DRAM fabs in Taiwan came into production in 1995
and 1996 - including Powerchip and Nanya’s factories. Worldwide,
the number of fabs leapt from 73 in 1995 to 170 in 1996. The majority of a chip’s nominal accounting cost is depreciation,
i.e. sunk cost. The actual cash cost is quite low. If those are covered, memory-makers have
little incentive to cut back on production even if producing will cause prices to crash even more. All this new capacity coming online rolled down 4 megabit prices from $12 to $7.50 to $3.40. Then
when prices hit $2.30 in the third quarter 1996, Vanguard could not even cover its cash
costs. The only pathway out was to move up to 16-megabit, but that product was not yet ready. What is worse, new DRAM capacity was still coming online - a lagging
effect of decisions made years ago plus an unwillingness to yield market share. In 1997,
five more fabs entered the market - throwing off the demand-supply curve once again. By 1998, the value of the whole market had cratered to under $15 billion,
down over 60% from the sweet summer days of 1995. Classic bullwhip effect.
Great for buyers, but hard for memory-makers.

Washout

The Great DRAM Crash of 1996 smashed DRAM-makers all over. The Koreans suffered a double blow from the high debt inflicted upon them during
the go-go 1990s, i.e. the Asian Financial Crisis. After a very long and drawn out process, the Korean government would eventually
merge together Hyundai and Goldstar’s semiconductor divisions to create Hynix. Over in Taiwan. In 1997, TI-Acer lost $58 million. Powerchip suffered a pre-tax loss of $31 million.
Both very high numbers in corporate Taiwan. And with Samsung and Hynix refusing to give
up even an inch of market share, there was only so much money the Taiwanese can afford to burn. At Vanguard, Bob Evans left mid-way through 1996 and FC Tseng from TSMC took over.
Tseng fought fires at Vanguard for the next few years - switching production capacity
away from commodity DRAM to sell specialized high-speed DRAM and RAM for graphics card. The Taiwanese suffered additional damage because they did not own their technology. Licensing
someone else’s DRAM IP means paying them a percentage, cutting into precious margins. It also leaves them high and dry if the foreign partner pulls out or cannot deliver.
Nanya experienced this when Oki failed to move to the next DRAM generation. The backwardness
could have killed them had they not signed a technology sharing deal with IBM in 1998. Also in 1998, Texas Instruments announced that it would sell its money-losing DRAM business
to Micron, end its various memory joint ventures and cut 3,500 jobs. As part of that retreat, TI sold its stake in the TI-Acer joint venture
to Acer in March 1998. Acer took on the task of turning its DRAM
business around while also trying to find a new technology partner. In mid-1999, Acer sold a 33% stake in the money-losing TI-Acer division to TSMC. A
few months later, TSMC bought the whole thing and turned its factory into Fab 7. Can see today that Fab 7 looks quite distinctly different than
its peers. That’s because they didn’t build it. This purchase along that of another foundry challenger called Worldwide Semiconductor
helped boost TSMC’s wafer capacity ahead of rival UMC. One of their major turning points, really. A similar fate eventually befell Vanguard. After Tseng returned to TSMC,
Morris convinced his VP of Operations Rick Tsai to join Vanguard as its leader.
Tsai is an operations master, and managed a new technology transfer from Mitsubishi. But Vanguard still lost $155 million in 1998 - a corporate record for losses
in Taiwan - and was deploying technology behind the leading edge. Analysts started
asking questions about how long TSMC can keep subsidizing its floppy-haired little brother. Rick Tsai eventually privately told Chang that memory had no future and suggested
that Vanguard pivot to logic foundry. Morris agreed, and in January 2000 told the staff. For whatever reason, TSMC did not initially fully absorb Vanguard - probably due to the
financial losses and node differences. But they did establish a comprehensive
alliance - transferring process node technology and even referring business to them. Vanguard’s conversion into logic foundry was completed in 2004. And to this day,
the two companies remain separate but allied. Vanguard even recently
licensed TSMC’s power GaN technologies as the giant moves to free up more capacity for AI. Ditching DRAM can be argued to be the proper financial decision, but Dr. Chih-Yuan Lu and
others later reflected in the recent Chip Odyssey documentary that Vanguard should have
held on a little bit longer. Chang himself has said that the way it ended was “sad”. Vanguard and TI-Acer were the two major exits during this period of time. Taiwan’s
other DRAM makers - Nanya, Powerchip, ProMOS, Winbond, and Mosel-Vitelic - continued on.

The Two Trillion and Twin Stars

The end of the 1990s saw the Asian Financial Crisis,
the twin dotcom and telecom bubbles, and then 9-11. By 2001, the global economy remained sluggish. Weak demand
for high tech goods caused Taiwan’s economy to abruptly fall into what
was called its most severe economic recession since the first Oil Crisis. And then there was Mainland China. Both Taiwan and the Mainland ascended to the WTO
together but China’s cheaper labor costs and sheer size put pressure on Taiwan’s economic
model. Foreign and Taiwan-based investors far preferred to invest in the Mainland. This capital and labor flight was a concern. The Taiwanese
government - then led by Chen Shui-bian of the DPP party - sought policies to
expand Taiwan’s economic strengths, add new ones, and keep capital and people in Taiwan. Thus came the “Two Trillion and Twin Star” policy - a six-year initiative that was part
of a larger $2.65 trillion NTD “2008 challenge” to upscale Taiwan’s economy and reignite growth. The private-public policy disbursed government support in the form of tax incentives and special
budget funds to projects in two heavyweight technology fields: TFT-LCD Displays and DRAMs. The goal was for those two heavyweight sectors together generate 2 trillion NTD of value,
or about $60 billion USD, by 2006. The “two stars” part of the initiative’s name refers to two rising
”star” fields: Digital content and biotechnology. And some funds were put into that as well.

A New DRAM Boom

The Two Trillion Policy triggered a new wave of DRAM company formation and fab construction. In November 2003, Nanya announced that it would partner with Infineon - Siemens
Semiconductor’s spinoff company - for a new joint venture called Inotera. Inotera would build a cutting edge 12-inch wafer fab in the area of Taoyuan. The fab
cost 2.2 billion euro, and would churn out 20,000 to 50,000 wafers per month upon its
completion in 2006. To get the wafers, they licensed Infineon’s trench capacitor DRAMs. Three years later, Infineon spun out its DRAM division into a standalone
company called Qimonda. A financial move to get DRAM volatility off their books,
this cut off Qimonda from any future support from its siblings - making it vulnerable to a downturn. Other Taiwanese companies - ProMOS, Powerchip, and Winbond - all surged to build new fabs in
response to the policy - often cases taking out billions of NTD bank debt in order to do so. 2006 only threw gas onto the fire. It was yet another boom year for DRAM. Yes
the DRAM cycle cycling again. After years of working off inventory and
capacity - including another downturn in 2005 - things finally turned in 2006. In February 2006, Microsoft showed off their next flagship OS, the despicable
Windows Vista. The electronics industry expected this overstuffed,
graphics-heavy chonker to drive another DRAM supercycle just like with Windows 95 and 3.0. At the same time, PC demand in China unexpectedly soared as China became
the world’s second largest PC market - which in turn led to unexpected demand
in DRAMs that is said to have caught memory makers offguard. The memory industry was also absorbing the rise of flash memory and the switchover from DDR2.
Popular portable music players like the Apple iPod used lots of flash memory.
So Samsung switched over some of their DRAM capacity to fulfill that demand. Meanwhile, the switchover from older DDR2 to faster DDR3 technology did not go as well as
first thought. Demand for the older stuff surged just as the fabs started moving to the new. So anyway, the DRAM shortage whipped everything into a fervor. In December 2006,
Powerchip announced an ambitious joint venture with Japan’s Elpida Memory. Elpida was then Japan’s DRAM national champion, formed out of the merger of NEC,
Hitachi, and Mitsubishi’s DRAM divisions. This big Taiwanese-Japanese joint venture was named Rexchip. And it had the goal of building
a massive 450 billion NTD or $14 billion USD fab complex in Taichung. At the time,
it was called the single biggest FDI project in Taiwanese history. But while the Two Trillion policy did manage to stimulate companies to build
more big fab projects in Taiwan, none of those companies got close to matching Samsung or
even Hynix’s size. In total, Taiwanese firms produced less than 15% of the global market. And again, most of them simply struck deals with foreigners to import their
IP - perhaps believing that they can focus on manufacturing only. As I said earlier,
this limited their technology options in terms of product and thinned their margins. Yet more concerning was the fact that the Taiwan government explicitly identifying DRAM
as a nationally important technology field created an implicit government guarantee.
Companies moved ahead with their plans thinking
that the government would bail them out if they failed. A moral hazard.

The 2008 Crash

Anyway. The DRAM cycle turned again in 2007. Prices crashed some 30% and
all the DRAM vendors started taking heavy losses. That was BEFORE the Global Financial Crisis in 2008. The prices of certain
products like a standard DDR2 1Gb DRAM sank to as low as 75 cents in March 2009,
leading people to say that it was “cheaper than water”. Samsung reported its first financial loss since 2000. Hynix lost money for
five straight quarters. Elpida and Micron lost over 70 billion NTD in 2008 alone. Europe’s last serious DRAM maker, Qimonda, found itself in serious
crisis. After an attempted rescue by the German government fell through,
they filed for bankruptcy with their assets sold off piecemeal. But it were the Taiwanese who took an outsized proportion of the financial
losses. Jeho Lee in his thesis “The Chicken Game” notes the DRAM industry as a whole took
losses of $10 billion between first quarter 2007 and third quarter 2008. Taiwanese DRAM firms took 42% of those losses despite
having between 13-15% of world market share. And unlike Samsung Electronics - which can lean on profits from selling end
user products or cash inflows from its various chaebol siblings - the
Taiwanese funded their builds with short-term bank debt. A disastrous move. By early 2009, it seemed possible that the entire industry could
go under - putting thousands out of work and maybe even triggering a bank
crisis. Frank Huang - cofounder of Powerchip Semiconductor - said in a press conference:

How to help Taiwan’s DRAM industry survive is of major
importance to Taiwan’s computer sector and even the overall high-tech sector If Taiwan’s DRAM industry collapses, Huang warned,
then its PC makers would be at the mercy of Samsung and be next to fall. In December 2008, government officials met with leaders at Elpida and then a week later,
Micron. They then announced coordination efforts with
banks to defer loan payments and take some pressure off the DRAM companies. Then in March 2009, the Taiwanese government - now led by Ma Ying-jeou of
the KMT party - put forth its long-term proposal: Taiwan Memory Company, or TMC.

Taiwan Memory Company’s Failure

The TMC proposal would roll up Taiwan’s medium-sized DRAM
makers into a single government-backed company. The mooted merger candidates were Inotera, Nanya, Powerchip, ProMOS,
Rexchip, and Winbond. The company would be funded with a billion dollars from the government,
acquire Elpida’s core DRAM technology, and build enough capacity to take on Samsung or Hynix. On the surface, the plan seemed good, and several semiconductor analysts publicly
urged the companies to join in. But after the plan was announced,
it received criticism from both industry and the public. The public asked whether the Taiwanese government was throwing taxpayer money after bad money.
None of these Taiwanese DRAM companies looked capable of taking on Samsung or Hynix,
which combined had 60+% of the market. Will TMC be a zombie company that falls apart anyway? And most critically, Taiwan’s various DRAM companies did not
want to participate. It seems like they more expected handouts of government money (i.e,
bailouts) to help them tide over the current situation. In April 2009, the DRAM decline somewhat reversed itself. With Qimonda’s collapse, Elpida and
Powerchip taking supply off the market, and buyers filling inventories, prices rose. The market
recovery - precarious as it was - nevertheless took away much of the TMC project’s urgency. And underneath the hood, the Taiwanese DRAM firms were also quite technically different. Notably,
you had two major technology families. First those aligned with Elpida: Powerchip and Rexchip. And then those in the Micron camp: Nanya and Inotera. Inotera notably was then
switching over to Micron’s stacked capacitor technology after Qimonda’s
failure - which used trench capacitors - and they did not seem excited to switch again. This complex brew of alliances, deals, technology sharing contracts, and so on
made a fast consolidation impractical. Integration efforts were likely to take
years and be quite messy. Thusly, Nanya, Powerchip, and the like refused to join TMC. The government rejiggered the plan, turning Taiwan Memory Company into
Taiwan Innovation Memory Company. Armed with $150 million USD, it would acquire
and develop technologies so that Taiwanese firms can retain access to DRAM technology. But by then Taiwan’s Legislative Yuan had turned against the project
and voted it down - seeing it as a fruitless giveaway to a very risky,
very cyclical industry. With this, Taiwan’s DRAM industry was left to the free market.

Conclusion

In the wake of this monkey poop debacle, ProMOS, Rexchip, and Powerchip delisted their stocks. ProMOS and Powerchip shifted to new business
models. ProMOS sold their fab and became a fabless designer. Powerchip spun off its fabs to create PSMC,
a logic foundry. Their specialty is the display driver IC - a class of tricky
chips that turn digital signals into images on the screen. That basically saved their butts. Elpida Memory collapsed despite three years of support from the Japanese
government - filing for bankruptcy in 2012 and acquired by Micron for $2.5 billion. With that, Micron got Elpida’s stake in the Rexchip JV,
and later bought the rest from Powerchip. Micron then purchased Inotera in 2016, completing a massive memory consolidation. Today, Micron
is one of Taiwan’s biggest foreign employers with 15,000 employees. While Micron has fabs
in Singapore, US, Japan, China, and India, the majority of its DRAM is made in Taiwan. Winbond and Macronix both went back to making flash memory along with some logic
ICs and custom memory products. The two of them remain a dominant player in NOR flash.
Winbond still makes a little specialty DRAM. The only major Taiwanese player to survive while remaining in DRAM was Nanya Technology,
which only made it through thanks to money from the Formosa Plastics conglomerate. Nanya has continued churning along as a second-tier DRAM supplier behind the
big three and CXMT. They recently started construction on a brand
new 10-nanometer class fab in New Taipei City, Fab 5A. It even has an EUV machine! For them, that is fortunate timing, considering the current AI-fueled DRAM
shortage. Their 2026 revenues year to date are up something like 600% year-over-year. Once upon a time, the DRAM industry involved 20-30 companies. Today that
has consolidated to three to five, with only three of them really mattering.
Taiwan’s DRAM failure doesn’t get a lot of ink, because people tend to focus on the logic stuff.
But it is a big deal, and I think as this makes clear, a very very different market.