TSMC's renegade genius
The story of Liang Mong-song, semiconductor renegade, is a teaching tale about what is actually important in the industry. It is not about the million dollar EUV machines. It's the people.
Liang Mong Song (梁孟松) was one of TSMC’s founding geniuses.
Unquestionably brilliant but scathingly difficult to work with, he rose through TSMC’s ranks for nearly two decades.
Then he defected to South Korea to work for the company's biggest and fiercest competitor: Samsung.
In doing so, he near single-handedly pushed Samsung to overtake and overthrow the Taiwanese chip giant.
Most Western media seem to skip over TSMC's history after its founding. I think what happened between then and now is just as enthralling.
In this video, we are going to look at one of the company's most dramatic events: the defection of Liang Mong-song.
Beginnings
Liang was one of TSMC's brightest rising stars. After graduating from one of Taiwan's most prestigious engineering universities - NCKU in my family hometown of Tainan - he traveled to the United States to continue his studies.
There, he attended graduate school at the University of California, Berkeley under the supervision of Hu Chenming.
Professor Hu is a TSMC legend. He was one of the company's first technical hires, eventually serving as their Chief Technology Officer.
Later on in the mid-2010s, Hu would help bring the FinFET to market - helping TSMC make a transition that few other semiconductor foundries have been able to achieve. Liang - a specialist in this technology - would play a critical role in this move.
Liang received his PhD and then went to work at AMD for over a decade. During this period, he would be named in over 180 critical semiconductor patents and publish over 350 technical papers in both the US and Taiwan.
TSMC
In 1992 at the age of 40, Liang quit his job and returned to Taiwan to work for TSMC in their R&D department. There, he would join a group of incredibly brilliant semiconductor minds similarly returning from the United States to bring their talents to Taiwan.
Some of these other people include the aforementioned Professor Hu, the FinFET genius from Berkeley.
R&D team leader and future CTO Chiang Shang-yi, from Texas Instruments and Hewlett-Packard. You are going to hear his name a lot. No relation to Generalissimo Chiang Kai-shek.
Rick Tsai, from Cornell and also Hewlett-Packard. He would become TSMC's CEO for a brief time and is now the CEO of MediaTek.
Douglas Yu, from the Georgia Institute of Technology and Bell Labs. Today, a TSMC Distinguished Fellow.
And of course, the most famous of these American migrants: Morris Chang himself.
To leave the comfort and high salaries of United States - at that time one of the undisputed leaders in semiconductor technology - so to join this growing but peculiar Taiwanese company raised eyebrows.
Together, these brilliant executives would help TSMC navigate a number of technically challenging decisions time and time again. And in doing so, allowed the company to leapfrog its rivals.
The Rise of Copper
One of TSMC's most famous major technical moves came in 2003. The circuit elements on a chip have to be connected together with a structure called interconnects. Essentially, wires.
For decades, the industry used interconnects made up of aluminum. But it had long been known that copper interconnects offered tantalizingly better performance.
They offered 40% less resistance, 15% faster speed, and can be made much smaller than their aluminum counterparts. Chips equipped with such wires would run faster and more efficiently.
The catch, however, was that copper presented some tricky manufacturing challenges. For instance, copper atoms can diffuse into the silicon, degrading its properties. So a metal barrier has to be constructed around the interconnect. Which is hard.
In 1997, IBM invented, patented, and announced the industry's first copper interconnect implementation. The press hailed it as a magnificent leap forward in progress.
TSMC's Copper
But like with every new announcement of this or that battery chemistry, IBM's copper interconnect technology could not leave the laboratory. In the early 2000s they brought it to TSMC, asking whether or not they wanted to license it.
TSMC's R&D team believed that copper was the future. But they also believed that IBM's technology would not take them there. Shockingly, they rejected the American giant's offer and invested a great deal of resources to create their own implementation.
In about a year, a team of six elite semiconductor TSMC R&D scientists led the development of their own 130 nanometer process node with copper interconnects. In 2003, they started shipping wafers with it.
This was a critical event in TSMC's history, allowing the Taiwanese company to overtake the world-famous IBM. The Taiwan government awarded the 2003 Outstanding Scientific and Technological Worker Award to a team of three men:
Senior VP of R&D Chiang Shang-yi, Senior R&D manager Liang Mong-song, and Senior Director of R&D Burn Lin.
Burn Lin is also renowned for having led the industry's transition to 193 nanometer immersion lithography. ASML adopting immersion lithography allowed them to overtake the Japanese lithography makers - Canon and Nikon.
Other members of the team included the aforementioned Douglas Yu, Yang Guanglei, and Sun Yuan-cheng (孫元成), who goes by Jack.
Liang
There is no disputing that Liang is a technically brilliant man. He is named on nearly 500 TSMC patents. He has an obsessive dedication to technology and incredible knowledge of the entire advanced semiconductor manufacturing process.
TSMC rewarded him handsomely for his efforts. Commonwealth Magazine reported that his total stock and cash compensation throughout his 17 years at TSMC was over $21 million USD. Throughout the 2000s, he made over $1 million a year in Taiwan - at a time and place when the average Taiwanese made $15-22,000 a year.
Yet at the same time, his brilliance and stubborn nature have also led to frequent clashes with colleagues wherever he goes. A friend of his described him as being:
"Very strict on the technology, very rigorous ... he is like: 'This is the requirement. You need to meet this. There is no negotiation'."
His future colleagues in China would complain that he is inflexible, less than enthusiastic in anything other than technology, and aloof.
One said, "We are moved by his enthusiasm for work." Which I find to be an amusing understatement.
Taiwanese love to use literary and historical metaphors. And in this case, Liang is likened to Lu Bu - the mighty Chinese general who lived two thousand years ago during the Eastern Han Dynasty.
People today know Lu as being an incredible warrior, but also one with a ferocious temper. He is also known for freely betraying his allies - which will come into play later.
At this point, Liang Mong-Song had risen up through the ranks in an incredible way. But here, things took a nasty turn. The Game of Thrones inside TSMC had begun.
Game of Thrones
In 2005, Morris Chang retired as CEO of TSMC at the ripe old age of 74.
His chosen successor was Rick Tsai. A PhD with experience at Hewlett Packard, he had been deputy director of TSMC's early fabs and served at the company for fifteen years. The last four as COO.
Four years later in June 2009, in the midst of the Global Financial Crisis and a totally botched 40 nanometer ramp-up, Morris would overthrow Tsai and return to the Silicon Throne to guide the company through some of its most challenging years.
But for now, Tsai's ascension triggered a shuffle underneath him. The aforementioned Senior R&D VP Chiang Shang-yi also retired from TSMC at the age of 59, becoming advisor to the CEO.
Chiang's retirement opened up a spot in TSMC's sprawling R&D organization and it was decided that the position needed to be filled with two people.
The first position was filled by Dr. Lo Wei-jen (羅唯仁), a former Intel factory manager who today remains the Senior VP of R&D.
But the second position presented a bigger challenge. Liang thought that he was the best candidate for the job. But advisor Chiang ended up choosing one of Liang's R&D rivals for the job - the aforementioned Jack Sun.
Sun was as brilliant as Liang, but also possessed an even broader view of how everything within TSMC worked. This was due to his work in process integration - making sure all those process steps fit together right.
Furthermore, he was known to get along better with his colleagues. With regards to Liang, it was said that "He is very capable, but his personality sometimes ..." I guess that says everything.
Beyond Moore
Morris Chang and TSMC decided to transfer Liang to be in charge of a new initiative called "Beyond Moore" or "More Than Moore". The idea behind this would be to expand TSMC's trailing-edge foundry capabilities into new verticals like MEMs, automotive sensors, chiplet-style advanced packaging stuff and the like.
Today, this is a growing business for TSMC. But back then, it was small. Starting off with a four-person office and expanding into two small sized foundries.
Perhaps Morris Chang wanted to test out his prodigy scientist, wanting to see whether or not he can manage people and actually grow a business. Liang's successor in this department - C.C. Wei - would rapidly grow sales and later ascend to the CEO role, having passed the test.
If that was the intent, then Liang failed the test. Liang's technical obsession is for the leading edge. No matter where he goes, he wants to - as the Eagles song sings - take it to the limit.
So you can see why he would see this new position - with its trailing edge process nodes and 6-inch wafer factories - as an effective banishment to a far-off province. Liang's resentment bubbled and grew - feeling that company leadership had simply thrown him aside.
And like the fourth son of the Hongwu Emperor, founding emperor of the Ming Dynasty, he would start a rebellion against his brothers.
Leaving
In February 2009, several months before Morris Chang rejoined the company as CEO, Liang resigned his post after 17 years.
As part of his resignation agreement, TSMC put on a pair of "golden handcuffs", holding back a significant portion of his financial payout until the fulfillment of his two-year non-compete agreement.
Liang told the company that he was going to spend his two years taking care of his parents and teaching at National Tsing Hua University in Hsinchu.
But in October 2010, just six months into his time there, he took up a visiting scholar position at the Sungkyunkwan University in Korea. There, he taught a simple class for 3 hours a week.
For decades, Sungkyunkwan University has held a close association with the Korean chaebol - Samsung.
Samsung
Samsung Electronics is the biggest and most ambitious of the Korean chaebol. They led South Korea's insurgency against the Japanese chipmakers - eventually becoming the world's leading memory maker.
They then set their sights on displays and panels. Soon enough Samsung Display defeated Taiwanese panel-makers like AUO and Innolux. Now they sought to bring down TSMC itself. But they needed some help to do so.
Starting in 2009, about the time of Liang's departure from TSMC, Samsung started to rapidly make up ground in the process node race. First, 45 nanometers ... then 32 nanometers, and finally 28 nanometers. All in just two years. Incredibly rapid progress in a very short period of time.
As early as March 2010, TSMC company members were asking Liang about media reports of him working there, which would be a violation of his non-compete clause. Liang repeatedly denied this, swearing that he would never work at Samsung and that he had TSMC blood in his veins.
Then on July 2011, two months after his two-year non-compete expired and his golden handcuffs unlocked, Liang officially joined Samsung as both the executive Vice President of Samsung Foundry and technical director of Samsung LSI. He took with him a team of 10-20 TSMC engineers.
Company management put two and two together. And four months later, TSMC filed a lawsuit in Taiwan against Liang - alleging that he had broken the terms of his non-compete agreement. This lawsuit would drag out for over four years.
14 nanometers
I cannot say for sure, but it seems like Liang's defection would deeply affect TSMC's work. In 2009 they were preparing to bring out their 28 nanometer node, but despite their claims high volume production slipped from 2010 to mid-2011.
28 nanometers and its half-step sibling 20 nanometers would essentially represent the end of planar scaling as the industry had long known it.
After that, TSMC and Samsung needed to follow Intel's lead and bring their own FinFET gate implementations to the market - a monumental undertaking. TSMC named their node 16 nanometers, while Samsung called it 14 nanometers.
Pause here. I would like to remind viewers about the broad issues relating to process node numbering. It's all marketing. For additional information, I have another video that you can check out.
When Liang arrived, Samsung had been trying to move on to the 20 nanometer process. He promptly ended all that funny business and had them working on directly jumping from 28 nanometers to 14 nanometers, adopting FinFET at the same time.
This meant jumping 3 generations at the same time, a huge move that few companies would be comfortable trying. But Liang is known for taking it to the limit, and he was highly motivated to beat his former employer to the node.
Clouds
Liang ended up working miracles. The 14 nanometer node smoothly came online and in 2015, Samsung announced that they had successfully brought out a 14 nanometer process node - ahead of TSMC's 16 nanometer process.
It was a stunning move that broke TSMC's near-complete monopoly in the advanced semiconductor foundry business. And it reaped massive benefits for the Korean giant. Having the first FinFET based implementation allowed Samsung to lure a sizable chunk of TSMC's customers to defect.
First, Qualcomm announced a billion dollar deal where Samsung would produce their high-end mobile chips. Qualcomm had always used TSMC before for these chips so this was a massive blow.
And then, the biggest client of them all. Apple is TSMC's single most valuable customer - having first started with them in 2014 for its A8 chip.
That move came in part due to Apple and Samsung's long running lawsuit. But a year later for the Apple A9 in the iPhone 6s, Apple decided to reconcile and split their production run between TSMC and Samsung.
It was a warning to TSMC that Samsung was going to topple them like it had toppled AUO, Innolux and Taiwan's memory makers. At the shareholders meeting in January 2015, Morris Chang publicly admitted "Yes, we are a bit behind".
Trial
I am aware that many people on the internet would see Samsung's muscling in to the advanced foundry space as an absolute plus. More competition is a great thing. Save us from the TSMC monopoly - so that we may have cheaper GPUs.
I am also aware that non-compete clauses are not particularly popular in the West - who tend to believe that anybody should be able to work anywhere they want at any time. In my home state of California, noncompetes are prohibited.
I am sympathetic with those notions. But the semiconductor world - with its national security implications and government subsidies - is a twisted and ruthless one. None of its top companies got to where they are now without playing the game.
Just google the AMD versus Intel antitrust court battle.
Years before, TSMC sued China-based SMIC for patent infringement and won, essentially cutting the legs off a dangerous challenger. Now, Morris Chang personally pushed for a judicial pursuit to block out Samsung from progressing any further.
But Liang prevailed in the first trial. Defended by the famous DPP-aligned defense lawyer Wellington Koo (顧立雄) the judge ruled in his favor.
TSMC was undeterred and appealed the decision. So, a second trial commenced in front of a new judge, who was at first skeptical about its merits. But this time the company brought forth evidence showing that:
- Liang's “lectures” at Sungkyunkwan were to Samsung employees and took place in Samsung's campus;
- Technical analyses showing Samsung's 45 nanometer, 32 nanometer, 28 nanometer chips suddenly becoming very technically close to TSMC's after 2009. Where before, they more looked like IBM's, whose technology Samsung licensed;
- Liang had started using a Samsung email address as early as 2009 when he replied to a birthday email using that address.
Judge Hsiung Sung-mei - Berkeley-educated with a Korean husband - ruled in TSMC's favor, which was then upheld by the Taiwan Supreme Court. She mandated that Liang cannot work for Samsung until after the end of 2015.
Aftermath
Again, we cannot say for sure. Because teams - not people - ship chips. But the distraction of Liang's court battle as well as the court's final decision likely had repercussions for Samsung's manufacturing progress.
That year in 2015, Samsung encountered issues with their Apple A9 chips. Netizens found that Samsung's chips reached higher temperatures than their TSMC counterparts.
It also seemed to have had lower battery life, though Apple later put out a statement refuting this.
On the other side, Morris Chang lit a fire under TSMC and they mobilized a massive R&D effort to retake node leadership at the 10 nanometer, 7 nanometer, and 5 nanometer process nodes.
Apple’s next chip - the quad-core A10 - would be a TSMC exclusive and it has been as such ever since as of this writing. TSMC remains the leading semiconductor foundry, with Samsung taking second place.
SMIC
Our story now picks up again in Mainland China. In 2015, SMIC appointed Zhou Zixue, the former deputy director of the Ministry of the Electronics Industry, as their chairman.
After losing their patent lawsuit, SMIC stayed on trailing edge nodes, largely servicing customers wanting local manufacturing. Zhou wanted to bring SMIC back to the leading edge in direct competition with TSMC. He believed that Taiwanese talents could take him there.
First, he helped recruit TSMC's godfather of semiconductor R&D, Chiang Shang-yi, Liang's former manager, who joined in 2016. Then Zhou targeted Liang himself.
Liang said that he would only join if he got to handpick a team of 200 Taiwanese and South Korean engineers. SMIC complied, staying at a hotel in Hsinchu for over a year recruiting them. And this time, they respected the no-compete clause.
When the Samsung contract ended in 2017, Liang - the semiconductor Lu Bu - broke alliances and jumped to SMIC with his massive team. His reported salary was about $200,000.
Limit
Liang's entrance caused some tension. SMIC had just promoted Zhao Haijun to the CEO role. Liang won't join if it isn't as CEO, but parachuting someone in with a whole new team would be a great embarrassment for Zhao.
Zhou split the difference by making them co-CEOs. Zhao would handle the legacy trailing edge business, and Liang would handle the leading edge - which he always wanted. It worked, but rumors of feuding would continually waft out of the SMIC offices.
In February 2016, SMIC unveiled the 28 nanometer process node. And with Liang now on board a year later, the company was able to raise the yield from 60% to over 80%, which is very good.
However, the process was a business failure. That same year, the Taiwanese foundry UMC came out with their own 28 nanometer node and yield rates of 98%, drinking SMIC's milkshake.
Liang argued that SMIC wasted too much time working on 28 nanometers, and advocated another multi-generational jump to 14 and 12 nanometers. He led the company to do this, achieving a 95% yield in doing so. For this, SMIC raised his salary to $340,000.
He and Chiang attempted to leverage their connections to buy an EUV machine from ASML, but those efforts were blocked by the US Government. Thusly, some at SMIC felt it more prudent to focus on trailing edge nodes. But Liang is pushing ahead anyway to the 7 nanometer node - the last possible node achievable with older DUV technology.
In 2018, Chiang left to join the ill-fated Hongxin Semiconductor Company. Hongxin failed - that had been a real dumpster fire - and Chiang decided to return as Vice Chairman.
This effectively made Chiang Liang's manager again and it led to a hissy fit in which Liang attempted to quit. The story made the presses and forced SMIC to give him a $3.3 million mansion and raise his salary to $1.5 million.
Today, he remains co-CEO - taking it once more to the limit amidst Chinese complaints of his direction and management style. The 76-year old Chiang retired again in November 2021, likely to rejoin his family in the United States.
Conclusion
TSMC is a ruthless place full of brilliant, maniacally driven people trying to make the impossible mundane.
For instance, take Dr. Y. L. Wang (王英郎), Vice President of Operations and Fab Ops. Wang worked the night shift at TSMC while simultaneously getting his PhD in Electrical Engineering during the day. His bosses never knew.
Today, he has over 283 patents and his catchphrase is: "As long as it's interesting, you will never get tired". And that is just one guy I randomly picked.
So the company has a dilemma. Their younger people need and have to be able to come up into the senior ranks, but the elders already sitting there are too valuable to let go. So what to do?
The story of Liang Mong-song, semiconductor renegade, is a teaching tale about what is actually important in the industry. It is not about the million dollar EUV machines. It's the people.