UMC: Taiwan’s Other Semiconductor Giant
If you want to watch the video, it is below
Dylan Patel of SemiAnalysis is coming to Taiwan later this month and we are thinking of having a small joint meetup. Would anyone be interested in joining? Let us know.
United Microelectronics Corporation (聯華電子) or UMC was Taiwan's first semiconductor company.
And like with many first-born sons, it was born with a golden spoon in its mouth.
But the company met its great early expectations and aggressively competed with its younger sibling TSMC for many years.
And even today it remains the third largest semiconductor foundry.
In this video, let's take a look at Taiwan's other semiconductor manufacturing giant.
I covered some of this story in a previous video about how Taiwan created TSMC. Feel free to watch that one too. The story diverges later on but I will try to keep it fresh and new.
In the 1970s, Taiwan faced new economic challenges. Its single biggest export category was textiles. But import quotas and international competition limited that industry’s future.
Taiwan's electronics industry was then its second largest export category. However, most of that industry still consisted of low value electronics assembly operations.
Various Taiwanese who spent time in the United States pushed for Taiwan's economy to pivot towards integrated circuits.
The government studied previous technology transfer situations and when they failed. They determined that the most common reasons for failure were insufficient R&D investment, lack of peripheral support after the transfer, and untrained personnel.
So the government set up the Industrial Technology Research Institute or ITRI to thoroughly acquire, introduce and transfer new technologies to the private industry. They would oversee the whole process like a project manager.
ITRI set up a committee to select several high potential technologies to bring to Taiwan. The committee selected Complementary Metal Oxide Semiconductors, or CMOS - an emerging technology also being adopted by the Japanese at the time.
They then also acquired a 7.5 micron CMOS technology from RCA. This was behind the 3 micron leading edge being used in the United States, but it was also reliable and quite mature.
In May 1976, Taiwan sent 35-40 individuals to RCA's plant for 6-12 months to learn this process node. RCA was in difficult financial straits at the time, so they were very willing to teach what they knew for the right price. In 1977, ITRI set up a small 3 inch wafer demonstration factory under the Institute of Electronics.
By December that year, yields had reached 81% - matching the original RCA factory's performance. That would have correlated at the time to a net profit rate of over 21% - then a very high number for a Taiwanese manufacturer.
From 1976 to 1980, ITRI spent $120 million USD to acquire IC design and manufacturing technologies from abroad. Soon thereafter, they set about transferring its acquisitions to the private industry. Thus came about UMC.
Setting Up UMC
In late 1979, ITRI set up a preparatory office for what would eventually be UMC - Taiwan's premier vertically-integrated electronics manufacturer. An Integrated Design Company, so to speak.
ITRI upgraded the aforementioned demonstration factory - worth $410 million NTD - from 7.5 microns to 3.5 microns. It also expanded the factory’s wafer sizes to 6 inches. They then transferred it and its employees to the company.
ITRI also invested millions to introduce computer simulation programs for IC design to the company. Spice II, CICAP, and other industry standard design tools.
ITRI even selected the products that UMC would sell. They transferred IC design IP for ten products, including for telephones, toys, timers, calculators, and watches. Mostly items that the Taiwan electronic industry itself exported.
The Taiwan government then reached out to several large Taiwanese companies to get them to invest into UMC. Many of Taiwan's biggest companies including China Development Financial or CDF, TECO, and the Taiwanese Bank of Communications contributed $12.5 million.
UMC was also one of the first companies to apply and receive space in the Hsinchu Science Park. Hsinchu would eventually become home to many of Taiwan's advanced microelectronics companies.
As you might guess, using the term "spin-off" seems to fall a bit short. This was a complete, soup-to-nuts transfer facilitated with all the resources of a nation-state. Due to the breadth and complexity of the transfer, it took over three years to complete.
Choosing a Leader
During its second year, the company needed to assign a deputy general manager to help run things after Eugene Du - the general manager - found himself stretched between too many projects.
The roster of talents to choose from was large - incredible talents who came back from the United States to contribute to Taiwan.
Hu Ding-hua (胡定華) - the Waishengren director of the Institute of Electronics - eventually chose his deputy director Robert Tsao (曹興誠). Sometimes goes by Bob.
Another Waishengren who grew up in Taichung, Tsao moved to Taipei as a student and lived in an informal house made from iron sheets - slum, better to say it - amidst cab drivers, cart pullers, and the like.
Tsao is a rare Taiwanese semiconductor pioneer who never went to the United States for school. He received a Bachelors from National Taiwan University and a Masters in Management Sciences from National Chiao Tung University in Hsinchu.
Hu chose Tsao because he felt that Tsao - at that time just in his 30s - was more suitable for managing a company. More entrepreneurial in his attitudes and abilities. And Hu was correct.
As leader of one of Taiwan's two major semiconductor makers, Bob Tsao would play a prominent role in Taiwan's rise to semiconductor dominance.
UMC's First Years
As the first born son, UMC had everything set up for it. And for that exact reason, the pressure was on for them to succeed.
As a first born son myself, I know the pressure of high expectations! If UMC failed, then that would snuff out the Taiwanese semiconductor initiative before it can reach its full potential.
UMC first began production in April 1982. And unfortunately it was a difficult time for the semiconductor industry. The Taiwanese economy was in a recession, and the company struggled to sell enough chips to customers.
As an IDM, UMC can only sell customers their own internally produced designs. Like I mentioned, ITRI had transferred IC designs for a few chip products but management wanted more.
At the close of 1982, UMC generated less than $5 million USD in revenue and turned a net loss. But the next year, things turned around.
Over in the United States, the AT&T Bell System monopoly was breaking up. This necessitated a large surge in ICs for new telephones, and the Japanese could not meet the sudden, massive demand.
UMC answered the call at the right time, and production surged 6 times over from 4 million pieces to 24 million pieces - turning a substantial profit. But this rapid export surge was only temporary, and the next year UMC had to find more business for itself.
TSMC and UMC have long been rivals. This dates back to Tsao's "accusation" that Morris Chang (張忠謀) stole the independent foundry idea, and then prevented UMC from adopting the model until after TSMC was founded.
In 1984, after the end of the telephone boom, Tsao submitted an expansion plan. In it, he said that the industry was moving away from Integrated Design Manufacturers. Every step was getting too complicated for any single company to master.
So he proposed an "OEM foundry" which would focus all its resources on semiconductor manufacturing. UMC would then invest in a variety of Taiwanese design houses to get them to bring their IC designs for UMC to manufacture - kind of like a loose semiconductor consortium.
Tsao and Chang
This idea - which Tsao later tweaked - sounds similar to though not quite entirely like the independent foundry idea that Morris Chang and TSMC pioneered.
The main difference was that Tsao's proposal would only serve design houses that UMC invested in - mostly favoring those in Taiwan. Chang's dedicated foundry would take on all comers.
According to Tsao, he showed the proposal to the Ministry of Economic Affairs. Dr. KT Li (李國鼎) - who had previously served as the Minister of Finance and is known as the father of Taiwan's modern economy - asked Tsao to bring the proposal to Morris Chang.
At this time, Dr. Chang had just become President of General Instrument - a semiconductor manufacturer in New York. Dr. Chang had already then gained widespread respect in the Chinese community for rising to the Number 3 spot in Texas Instruments.
Tsao sent the proposal to Chang and had dinner with him but nothing came out of it. Shortly thereafter, the Ministry of Economic Affairs dropped UMC’s expansion proposal. Tsao later claims that this was because Dr. Chang sent a letter to KT Li, saying that the proposal was unworkable.
Morris Chang for his part says that he has had his dedicated foundry idea since his days at Texas Instruments. He had proposed it to both Texas Instruments and Intel, who turned it down.
Hu Ding-hua, the director I mentioned earlier who first elevated Bob Tsao, adds some additional color. Behind the scenes, the government strongly considered UMC's proposal but opted to create a separate company, TSMC.
Ultimately, officials had two major concerns. First, they felt that it would be easier to set up a dedicated manufacturer from scratch than it would be to convert an IDM into one.
Their thinking on this turned out to be correct. UMC later discovered that becoming a foundry not only meant big cultural changes but also working with their former competitors. Which those competitors did not want to do.
And second, Taiwanese officials were concerned about possibly turning UMC into a state-backed monopoly like China Steel. They wanted competition in their growing semiconductor industry.
Who Invented the Idea?
Of course, arguing about who came up with what and when it came about is a bit of a silly matter.
The "idea" of an independent foundry has been around long before either UMC or TSMC.
Carver Mead - the Caltech professor famous for revolutionizing VLSI design - proposed the concept of a "fabrication facility" all the way back in 1979. He called it a "silicon press".
And then we have MOSIS, launched in 1981 as an e-commerce service provider for American university students. Students taking a VLSI course can submit their designs to MOSIS, who batched them together and fabbed them.
It is like a mobile app. Everyone can come up with an idea for an app to share pictures with filters. Fewer people can actually execute on it and make Instagram. Ideas are cheap. Execute on them.
UMC and TSMC
Regardless of how it happened, you can only imagine Tsao's frustration when TSMC opened its doors in 1987.
To add to the insult of losing the first mover's advantage, UMC felt that the Taiwanese government gave its second child a sweeter deal during the transfer and founding. This includes larger amount of subsidies and a better process node.
The sibling rivalry between the two companies bubbled up and intensified as the Taiwan semiconductor industry took off in the late 1980s and 1990s.
At the start, TSMC took orders from everyone except UMC. The reasoning apparently being that TSMC felt that doing wafer manufacturing for UMC would allow the latter to triple their capacity and take more share.
Early on, Morris Chang served as chairman of both UMC and TSMC. Then in 1991, Bob Tsao and the other directors of the UMC board asked Chang to resign in the name of "competition avoidance". Chang did and Tsao himself became chairman.
Then in 1995 UMC entered into a series of creative joint ventures with customers that allowed it to spin off their design firms into separate companies. This eventually turned them into a dedicated foundry like TSMC, putting the two into direct competition.
One of those spin-offs - MediaTek - would go on to become Taiwan's largest fabless company.
UMC chose to challenge TSMC for the semiconductor crown. But TSMC had an 8-year head start on its older sibling, grabbing many of the larger customers that had been available. Customers that the giant was not really excited about sharing.
Being the smaller Taiwanese semiconductor giant, however, meant that Tsao and his company could take more entrepreneurial risk and act more flexibly if it helps conserve their limited capital.
For instance, the creative joint venture structures in 1995 that eventually allowed the company to convert into an independent foundry.
UMC has also been far more willing than TSMC to make acquisitions and go overseas. A notable acquisition was the 1998 takeover of Nippon Steel's struggling semiconductor division, which they promptly turned around in a year. This gave UMC a factory in Japan long before TSMC.
Lacking the big, rich customers that TSMC has, UMC has had to resort to joint ventures with customers to raise the capital for their fabs.
Examples include ventures with Xilinx, Hitachi, and Infineon.
And finally, UMC was one of the first Taiwanese companies to provide their employees with a stock bonus plan in order to attract better talents. Stock and options are a common occurrence in Silicon Valley, but before UMC it was rather rare in Taiwan business. The idea has since been widely adopted.
It is good that UMC did all this. However, it is all for naught unless UMC can keep up with TSMC technically. If the company cannot maintain technology leadership, then it would be relegated to specialty fab status.
From about 1993 onwards, UMC managed to keep up with TSMC - even despite a devastating fire in 1997. The two companies spent large amounts on R&D to acquire the latest process nodes.
Then in 1997, a mistake. IBM announced their new copper interconnect technology to great fanfare. This replaces the old aluminium wires between circuits with copper to help those wires more easily transmit electrical signals as things got smaller.
IBM brought the concept to TSMC and UMC, seeking a technical cooperation. As I mentioned in the Liang Mong-song video, TSMC rejected this cooperation and successfully engineered their own implementation of the concept for the 130 nanometer node.
UMC took the deal, striking a cooperation agreement with IBM and Infineon to help create their 130 nanometer node.
It is here that I would like to make a correction to what I said in the Liang video. I said back then that IBM's technology never left the laboratory. That is incorrect. The technology did indeed enter commercial markets. What I meant to say was that IBM’s copper interconnects turned out to be commercially uncompetitive.
TSMC’s successful launch of their 130 nm node skyrocketed the company ahead of the competition, won major plaudits from the Taiwan government, and increased its revenue to $5 billion. And it was also UMC's first major technical stumble.
However it was not a fatal blow. And in fact, UMC managed to launch the 90 nanometer node first in 2003 while TSMC was still on its 130 nanometer node.
But TSMC was lagging behind because they were working on mastering the next step in lithography technology - immersion. They launched the next year in 2004, quickly followed that up with 65 nanometers in late 2005, and then marched on to 45 nanometers in 2006.
UMC struggled to keep up with this relentless march. They got to 90 nanometers first, but their own 45 nanometer process node would not come about until early 2008 - two years after TSMC.
TSMC stumbled on their 40 nanometer node which eventually - among other things - cost then-CEO Rick Tsai his job.
In June 2009, Morris Chang returned to lead TSMC in the midst of the Global Financial Crisis. Chang brought the company back on track - closely working on what would turn out to be an extremely impactful node - 28 nanometers.
The 28nm Gate Decision
28 nanometers is the most advanced process node still using a planar gate and is extremely difficult. To achieve it, fabs had to make a critical manufacturing decision: Gate-first, or Gate-last?
The deep technical details of this decision are quite arcane, but I am going to try to give a decent summary.
A transistor has three parts - a gate, source, and drain. Throughout semiconductor history, fabs produced the gate first before the source and the drain.
You then use that gate to help create the source and the drain. Then you go back to repair any damage that might have been done to the gate. This is Gate-first.
But starting with the 45 nanometer process node, the industry introduced a new type of gate - the High-K/metal gate. The abbreviation of this, HKMG, makes me think of Hong Kong. Anyway, these are better at preventing electrons from leaking as they pass through the gate from the source to the drain.
The drawback of this new gate is that your old Gate-first method basically causes it to fall apart - particularly during that repair stage.
So Intel, who first commercialized the HKMG, created a new process to deal with this - Gate last.
You make a sacrificial gate just for creating the source and drain. Then you destroy that sacrificial gate so that you can build the real gate after you finish making the source and drain.
It sounds conceptually quite similar to Gate first. And it is! But it radically changes the manufacturing process.
Furthermore, there are design implications. For instance, forcing all the gates to lie in the same direction. This was not a simple decision.
Anyway, the Gate first and Gate last manufacturing decision was the biggest talk of the town in those days. The industry split into two paths.
Intel and TSMC chose to make the switch to Gate last.
IBM and its technology partners - AMD, Chartered Semiconductor, Samsung and UMC included - chose to stay with Gate first.
In the end, Gate last was the right decision. Despite its benefits, Gate first did not yield, and the fabs who stuck with it suffered greatly.
Being the only independent foundry with a 28 nanometer process that actually worked, TSMC ate up the entire market and shut the door on UMC. Since then, UMC has essentially waved the white flag, leaving TSMC unchallenged on Taiwan.
One final thing. As part of their business strategy, UMC has been far more willing to enter new markets wherever it might be able to find some advantage. One of those markets was China.
In 2000, Richard Chang leaves TSMC and goes to Shanghai to set up SMIC. A year after that, UMC goes to the mainland as well, purchasing 15% of a company called Suzhou Hejian in the Jiangsu province.
Robert Tsao did this in secret and in violation of a 1991 Taiwanese law restricting such large investments in the Mainland called "Don't rush. Be patient". Later on, it was discovered that UMC had also transferred valuable 8-inch wafer production technology to Hejian.
This led to a large scandal. After DPP President Chen Shui-bian won re-election, his administration aggressively pursued Tsao. Tsao later resigned as UMC Chairman after 20 years and eventually was so fed up that he renounced his Taiwanese citizenship and moved to Singapore. He eventually returned a few years later.
Morris Chang and TSMC eventually got the law changed. I discussed the decision in a previous video of mine. After that, TSMC and UMC were able to legally establish subsidiaries on the Mainland.
There has since been one more significant incident. In 2018, UMC was indicted by the American government for transferring memory chip designs from its client Micron Technology to a Chinese semiconductor maker. In 2020 they paid a $60 million fine and is serving a 3-year probation period.
In 2021, UMC generated $7.7 billion in revenue - a 20% jump year over year.
In 2021, GlobalFoundries made $6.6 billion and SMIC, $5.44 billion. So UMC sits behind just TSMC and Samsung as the world’s third largest dedicated foundry.
The company's most advanced process in their 12 fabs is a 14 nm node.
However, most of their bread and butter comes from nodes in the 65 to 28/22 nanometer ranges. The sweet spot considering the nature of our recent chip shortages.
Amidst record setting profits and over 95% fab utilization rates, they have recently announced billion-dollar fab expansions to their fabs in both Singapore and Tainan.
Bob Tsao and Morris Chang have recently settled their differences, publicly shaking hands and making peace. The two companies that are their legacy remain the cornerstones of the global semiconductor manufacturing industry.