When TSMC Sued Its Chinese Rival ...
Part 1 of 2
Author’s Note: If you want to watch the full video, it is below:
You will also want to watch the earlier video about SMIC, which stands for Semiconductor Manufacturing International. Today, it is China’s semiconductor crown jewel.
I was kind of surprised by the response that I got when I released this video. You can group together the responses into two camps:
How can TSMC bully this Chinese company? And on what grounds does an American court have over a Chinese company?
China has been doing this forever. Big deal. Meh.
We need to put the event and the company in context. This was a different time, and SMIC back then was not a Chinese national treasure.
SMIC at the time was essentially a foreign-run company that set up in China to take advantage of cheaper labor and favorable government subsidies. Their top management came from abroad - note the Italian Marco Mora below - and their technology was imported. A significant portion of their products were exported abroad, mostly to the US, which gave TSMC grounds to sue them there.
The success of the TSMC lawsuit combined with the Taiwanese company’s continued aggressive push would set SMIC back for a decade. Perhaps permanently.
It is 2003 and TSMC is grappling with the hot new thing in town: Semiconductor Manufacturing International, the rising startup based out of Shanghai.
Founded in 2000 by hard charging serial entrepreneur Richard Chang, SMIC was growing fast. Suspiciously fast. They and competitor Grace Semiconductor were stealing away some of TSMC's best people. What TSMC suspected was that SMIC was stealing more than *just* that.
Thus in 2003, the Taiwanese company launched a landmark legal battle against its mainland rival. The lawsuit would not be resolved for nearly six years, but it would end with a big TSMC victory. In this video, let us look at the time when TSMC sued SMIC ... and won.
The Semiconductor Process
I want to go over briefly the complexity of developing a new semiconductor making process. This is not a video about how to make semiconductors. So I am going to make a few generalizations in discussing the critical points.
In previous videos, I talked a lot about several specific methods and techniques like EUV and photolithography. It might leave you with the impression that that’s all there is to it. But fabricating a chip is more than just lithography. There are thousands of steps with complicated names like etching, ion implantation, and chemical mechanical polishing.
At a new process node, a company needs to figure out new technologies so that you can lay down the circuits while avoiding issues like cracks, current leakage, and particle contamination.
You start out with R&D that determines the specifications of the equipment. Then you select and integrate together that equipment. Then you got to determine the "recipes" and "design rules" behind the individual steps of the overall process flow. Recipes can mean like equipment settings - temperature, pressure, materials, and duration of the process. It really is like cooking.
You make changes and then test the product output for its quality - choosing the ones that make the best trade-offs. The knowledge gained from that trial-and-error process are often not patented but instead kept privately within the company as a trade secret. This is the valuable knowledge that SMIC targeted.
Richard Chang and his startup team had quite a lot of experience. No denying that. This was Chang's second rodeo. He knew what he was doing. But the speed with which SMIC was getting things up and running rang alarm bells.
Generally for a China-based foundry, you can expect to take about 21-30 months to go through this entire process and have a new multi-step node ready for customers. That’s how long Grace semiconductor took in launching their first foundry. SMIC was not taking that long.
SMIC was founded in April 2000. Four months later in August 2000, they broke ground on their first fab: SMIC Fab 1 in Shanghai. This 12,511 square meter building with all its respective cleanrooms was completed in June 2001.
It took SMIC just two months to install all of the necessary equipment. Then four months later, a 0.18 micrometer process had passed quality checks and was ready for customers with zero non-conformances. Volume production began January 2002 with around 500 200-millimeter sized wafers a month and scaled up by 2,000 each month with world-leading yields.
Six months from equipment installation to high volume production is kind of insane. Grace Semiconductor, SMIC's China nemesis, had an aggressive ramp-up schedule of their own for 0.18 micrometers. But they publicly announced that it would take them 18-21 months.
TSMC's director of brand management Charles Byers would say in a statement:
You begin asking questions when a company ramps a process in six months that took another company two years. That raises eyebrows.
Fab 1 single-handedly narrowed the technology gap between China and the rest of world from 4-5 generations to 1-2 generations. In just a year. And remember, they’ve never done this before.
SMIC, for their part, replied that they managed to do this so fast because of the global shift to outsourcing and competitive prices. And also, I suppose, the power of Friendship.
TSMC suspected that the real reason behind it was SMIC running an operation to obtain critical, proprietary information. Turns out, those suspicions were correct. But more on that later.
The First Lawsuit
In early 2002, TSMC filed a lawsuit in Taiwan trying to prevent SMIC from hiring away their senior managers. They won an injunction in a Taiwanese court, but they knew that this injunction had no power in the mainland.
So the company decided to file their December 2003 lawsuit in the United States. You might wonder then, how can a lawsuit in the United States affect a Chinese company?
The answer is that while SMIC's main operations are in Shanghai in the People's Republic, the actual corporate entity is a Cayman Islands corporation. This is a common structure for the purpose of allowing the company to be IPO'ed in the United States. Which it did in 2004.
Furthermore, a significant portion of the semiconductors that SMIC fabs make in China are being imported to the United States. The lawsuit sought as a remedy a permanent ban on all semiconductor imports from SMIC to the US. This would be immensely damaging.
After the legal discovery process was completed, TSMC argued for the existence of a coordinated, sustained effort to obtain proprietary process flow, fab layout and operations information.
The evidence begins with TSMC doing a forensic analysis of a Broadcom 0.18 micrometer process chip made by SMIC. They discovered that 90% of the logic process had been copied from TSMC. This is confirmed by sworn testimony by SMIC employees.
Willy Luo, a former TSMC senior process engineer who defected to SMIC, testified:
I was told that SMIC had a code name for TSMC’s process recipes called BKM1, which stood for ‘Best Known Method One'
Another SMIC employee testifies in a sworn statement:
This was confirmed by my co-workers at SMIC, many of whom had come from TSMC. To my knowledge, SMIC’s use of TSMC’s process information was both known by, and encouraged by, senior SMIC managers
One employee explicitly pointed out as having stolen TSMC information was C.Y. Shih. He was a Technical Manager of Technology Transfer Division, Diffusion and Thin Film Section before he resigned in Sept 2001. His job was to manage the records surrounding TSMC's trade secrets. The gatekeeper to the chicken coop, so to say.
The weekend before he left the company, Shih was observed by coworker Y.L. Wang photocopying private documents from TSMC's Fab 4. He had piles and piles of paper on his desk. Subtle.
"[He] repeatedly asked me and two of my colleagues ... to provide electronic and hard copies of all improvements made during the past quarter for thin-film processes"
Shih jumped ship a week later. Fab 4 in Hsinchu at the time made semiconductors using TSMC's 0.18 micrometer process - the same one that SMIC Shanghai Fab 1 would get going so fast and flawlessly.
More damningly, there are emails. Yes. Emails. Katy Liu, a high ranking TSMC employee, served as Program Manager of Quality and Reliability until Feb 2001. Unbeknownst to TSMC management, she had actually joined SMIC on July 2000 as a "Founding Project Manager". Between then and her termination, she basically acted as a corporate spy.
There, she solicited key employees to work for SMIC and sniffed out trade information to steal. She would offer those key employees something in the range of 50,000 to 80,000 shares of startup stock and stock options to leave TSMC for SMIC.
SMIC also directed her to sniff out specific critical trade secrets. In one email, SMIC COO Marco Mora told Katy:
I need a help from you to pull-out information from WSMC/TSMC. I think you can [unreadable] some help from your people ...
And then there is a long, specific list of exactly the TSMC technologies that SMIC wanted to copy. From the exact fabs. Marco closes the email with a kind apology:
Sorry for the long list, but we need a lot of material to set up the new operation.
How many times do we have to teach you this lesson, old man? Never put anything down in email that you don't want the world seeing!
SMIC's response to this in court was that there was no evidence that Katy ever got this information back to Mora. Katy Liu was convicted of corporate espionage on Taiwan but fled to the PRC and I guess permanently lives there now.
SMIC targets went beyond just those working at TSMC. Charles Chan worked at Broadcom as an account manager to TSMC. His position as a liaison between the two companies meant that he had access to private information like pricing, margin, and technical specifications. He left March 2002 to SMIC and used his knowledge to help SMIC land Broadcom as a client.
The fact that SMIC was copying TSMC could hardly be said to be a secret at the company. They literally marketed themselves as a replacement for TSMC. SMIC account managers told their prospective foundry customers that SMIC's fabrication processes were "TSMC compatible" and "just like TSMC's". One boasted that the two technologies were so similar that little design work would be needed on the customer's part so to make the switch.
SMIC for their part attempted to fight back. Their senior management strenuously denied the allegations. They issued a statement warning against "artificial or unfair restrictions" on fair competition.
They emphasized that SMIC has been "consistently investing a significant portion of its annual budget in both R&D activities as well as the acquisition of technologies from its partners in order to develop and build up its own technology and know-how". Their technology and processes are hard-won and their own!
For all their angry words, the lawsuit presented some pretty damning evidence. Then in 2004, TSMC expanded the lawsuit and added three more patents that SMIC had allegedly infringed on. In February 2005, it seemed clear that SMIC would not be able to get away from this and they settled.
The settlement agreement between SMIC and TSMC would have SMIC pay TSMC $175 million over six years. $30 million for each of the first five years and $25 million on the sixth. This is a significant chunk considering that the company made $120 million in profit for the 2004 year in total - they turned a $66 million loss in 2003.
In return, TSMC would allow a six-year cross license patent agreement to their respective portfolios. This is so long that SMIC return the stolen documents.
The Second Wave
This seemed like a satisfactory resolution, right? Wrong. A year later in August 2006, TSMC filed another lawsuit for another $130 million.
In it, they alleged that SMIC breached the settlement. They never returned the stolen documents and have continued copying manufacturing process information. As it turns out, SMIC had a lot more documents than at first revealed - 15,000 documents totaling 500,000 pages.
The complaint read:
SMIC has carried out massive corporate espionage directed by certain [of] SMIC’s top operating officers ... SMIC lavishly copied the information it stole from TSMC, word for word, line for line, diagram for diagram, and even typographical error for typographical error
SMIC this time decided to fight back. In November 2006, they filed a lawsuit in Beijing arguing that TSMC was essentially badmouthing their mainland rival.
The saga would last another three years. This was a time of difficulty for the semiconductor industry and as it turned out, TSMC had the resources to hold out. SMIC piled up increasing losses as the court battle and a bad economy distracted it.
In 2009, a California jury ruled that SMIC had indeed breached the agreement and stolen trade secrets. This would put SMIC on the hook for a billion dollars in damages. Furthermore, SMIC products would be permanently banned from the US.
SMIC finally caved a year later. The net losses and the jury decision made it clear that a new direction was needed. Chang resigned and the company settled. SMIC would pay TSMC $200 million and grant TSMC shares and options for a 10% stake in the business.
This new settlement would give TSMC a say in how its biggest mainland rival did things. It would also incentivize TSMC to keep the smaller company alive. Chang's resignation would also herald an era of calmer and friendlier relations.
Throughout the entire saga, TSMC wanted to protect its secrets. Trade secrets that cost billions of dollars to acquire. That was the foremost goal.
To the rest of the industry, the company showed that it could and would go to the mat in order to defend those secrets. And in the end, they likely took that stake in SMIC because they knew that if the company folded then its employees would scatter, taking those secrets with them.
After all of this, one can argue that SMIC’s shortcuts cost it dearly in the long run for two reasons. The lawsuit hamstrung the company's technological progress. I am reminded of how Microsoft’s monopoly battle in the 90s hamstrung it and prevented it from dominating the internet like it did in PCs. What customer would want to partner with a convicted plagiarist?
And finally, my big moral. Stealing someone else’s IP means that you didn’t make it yourself. And if you didn’t make it yourself then you have no idea how to make it better. You end up relying on someone else to do it for you.
SMIC has been a constant follower for a decade now. That's what happens when you don't own the technology to differentiate you from everyone else. It was only recently they’ve tried to do it themselves and despite having gotten quite far with it, that decade was indeed precious time lost.