In 2020, TSMC, the world’s largest contract maker of chips, announced it would invest US$12 billion to build a foundry in Arizona. That news completely changed the career path of onea 30-something TSMC engineer.

Taking advantage of the opportunity, he volunteered to be assigned to the factory-building team of “Fab 21” (the Arizona fab), signing a three-year contract to live in the United States. 

Young engineers pursue American dream

The charteredplane carried TSMC's engineers to the US (Source: Chien-Tong Wang)

On November 15, TSMC’s expeditionary force to the United States left Taiwan on six charter planes to Phoenix. CommonWealth witnessed the group of more than 300 TSMC employees and their families checking in for the flights, handing over large cardboard boxes and other luggage, and even bicycles, cats and dogs, all to be transported to the U.S.

Production at the Arizona plant is expected to begin in 2024, with a capacity of about 20,000 12-inch wafers in the project’s initial phase. On the surface, the new foundry will have a negligible impact on TSMC’s overall performance, accounting for less than two percent of its overall capacity and around three percent of total revenue.

But the three-year journey that this group of young TSMC employees have embarked on could affect not only TSMC’s future, but Taiwan’s as well.

A family of three took photos with friends and family before getting on the plane. (Source: Chien-Tong Wang)

Bloomberg News reported in mid-November that Apple CEO Tim Cook said in an internal company meeting in Germany in October that Apple had “already made a decision to be buying out of a plant in Arizona” that “starts up in ’24,” referring to Fab 21.

Cook was also cited by the report as saying that 60 percent of the world’s processor supply comes out of Taiwan and that “60 percent coming out of anywhere is probably not a strategic position.” 

Industry insiders had predicted that the Arizona plant would specialize in chips for sensitive military or other essential products that have to be made in the U.S. But Apple is TSMC’s largest customer, accounting for 25 percent of its revenues, and it may have decided to diversify risk by shifting production of some of the central processing units (CPUs) for its iPhones or MacBooks from Taiwan to Arizona.

This is simply history repeating itself. 

A retired TSMC executive recalled that when China began conducting missile tests in waters around Taiwan in 1995, sparking fears of a military confrontation, some customers asked TSMC to set up a factory in the U.S. 

That resulted in TSMC’s first U.S. foundry, called WaferTech, a joint venture set up in the Portland area in 1996 by TSMC, ADI, and Altera, in which TSMC had a 57 percent stake. It was similar to today’s Arizona project in that it was close to an existing semiconductor cluster on American soil.

The money-losing Portland experiment

The ambitious U.S. complex was planned as four foundries on 260 hectares of land. The general manager of the new company, who was hired from Micron, Ken Smith, told employees it would one day go public in the U.S. 

But in the fall of 1999, TSMC founder Morris Chang (張忠謀) flew to Portland to see how WaferTech was doing, only to go ballistic at what he observed, and decidinge on the spot to not go through with a factory expansion.

In April of this year, Chang talked about the WaferTech venture in a podcast produced by the Brookings Institution and recalled that “initially it was chaos.”

“It was just a series of ugly surprises,” said the retired, 91-year-old Chang, and because the costs of production there were 50 percent higher than in Taiwan, the facility was not competitive.

Many observers are struggling to understand why Chang on several occasions has bad-mouthed TSMC’s new foray into the U.S. and why he reportedly told U.S. House Speaker Nancy Pelosi during her visit to Taiwan in August that U.S. efforts to rebuild chip manufacturing at home were “doomed to fail.”   

But to the former TSMC executive, the explanation is simple: The WaferTech experience left Chang extremely downbeat about manufacturing in the U.S.

In 2001, when the dot-com bubble burst, WaferTech posted a loss of NT$1.2 billion (US$38.74 million) in a single quarter, threatening at one point to drag down the performance of the parent company.

Two mistakes that cannot repeated

A vice president of a large electronics manufacturer who once worked at WaferTech said TSMC made two critical mistakes when it launched its U.S. foundry.

First, it underestimated the difficulty of making semiconductors in the United States.

In his podcast with the Brookings Institute, Chang recalled that “when we first went in, we really expected the costs to be comparable to Taiwan. And that was extremely naïve.” 

Beyond the costs of land, employee salaries and taxes, there were many unexpected hidden costs.

Early on, for example, TSMC could not find local suppliers for certain chemicals, forcing the company to import them. It was only then that the company realized that strong acids and strong alkalis were not allowed on airplanes, and bringing them in by sea took longer than expected. 

In addition, it was hard to find talented people who were aligned with a manufacturing culture. Taiwanese employees were willing to be on stand-by around the clock, while American engineers shut off their phones after leaving work, and American line workers often challenged TSMC’s standard operating procedures.

The source felt, however, that TSMC is now much better prepared and has a higher chance of success than 25 years ago.

A key move was to ask members of its supply chain in Taiwan to set up shop with it in the U.S.

TSMC even recently requested that one of its equipment manufacturers install robotic arms to remove and replace protective pads during the wafer processing process.

“They know people will have trouble keeping up, so they want to automate as much as possible,” the vice president said.

TSMC’s second mistake was the composition of its management team, which the former WaferTech executive described as a “motley crew”. 

Chang embraced the philosophy of “Americans managing Americans”. The top American executives, led by CEO Smith, originated from Micron and Motorola, and were combined with a small Taiwanese team and ethnic-Chinese employees recruited in the U.S., such as Liu Xiaoqiang (劉曉強), who would later head TSMC’s Fab 18 in the Southern Taiwan Science Park. 

At the time, TSMC did not have nearly the cachet in the industry it has today, and American employees at WaferTech virtually ignored the TSMC model Taiwanese executives tried to introduce. After the situation got out of hand, Chang sent two highly trusted executives, senior vice-president Steven Tso (左大川) and Vice President Lung-chin Du (杜隆欽), to take over, and they eventually stabilized WaferTech’s operations after several large-scale personnel reshuffles.

Fast forward to today, and how the Taiwanese and American teams will work together remains a real challenge for the Arizona project.

This time, TSMC has appointed Rick Cassidy, senior vice president of the Corporate Strategy Office, as CEO and president of TSMC Arizona. 

That has not stopped Taiwanese and American employees from engaging in several skirmishes already. Last year, an American employee who was supposedly in Taiwan for training criticized TSMC online for what he called its military-style management, sparking an uproar. That was followed by a Taiwanese employee criticizing American workers as “big babies”.

Complying with Taiwan working model on American plants would be one of the hard tasks for TSMC. (Source: Chien-Tong Wang)

SOP translations: The more detail the better

With fewer than two years before the Arizona factory begins mass production, an urgent challenge is to replicate the 5-nanometer process, used in Fab 18 for two years and already a major source of TSMC revenue, in the United States.

Tony Chen (陳鏘澤), the technical director of TSMC’s Facility Division who has moved to Arizona, said the fab’s design is a clone of Fab 18 in Taiwan, and it involves far more than just the equipment.

A TSMC equipment engineer who will soon head to Arizona, said that while rotating shifts among different machines over the past year, he has been translating, with the help of his colleagues, 5-nm process SOPs and machine maintenance instructions to “teach the Americans how to operate the machines based on these documents.

The engineer and his crew have translated details related to the parts of every machine and operating instructions into English and even taken pictures of every part separately. The process was so thorough that even the materials used in the gloves of line workers were labeled.

“The more detail, the better. That way, if the Americans do poorly, they won’t have the Taiwanese to blame,” a supervisor said.

“America wants TSMC to go there, so it has to go. But there’s no way Americans will work in shifts around the clock like Taiwanese do. My colleagues and I don’t feel that the U.S. plant will succeed,” the engineer said.

US demanding the most advanced processes? 

TSMC’s Arizona fab is expected to begin volume production using the 5-nm process in the second half of 2024. But on November 21, after returning from the APEC Leaders Summit in Thailand, Chang revealed that TSMC would initiate a second investment phase to produce chips using its most advanced 3-nm process.

Though outsiders believed such a plan would take at least three years to materialize, Chang’s words sent shudders down Taiwanese spines. Taiwan is worried that if TSMC moves its most advanced technologies to the U.S., its special status as a semiconductor giant would be eroded, making it more likely for Washington to sit on the sidelines if China were to attack. 

In an analysis for the Center for a New American Security published in September, “Chip War” author Chris Miller wrote that even though TSMC is building fabs in Japan and Arizona, all of its advanced technology development remains in its home country.

With that in mind, he urged U.S. officials to adopt a policy of pressing TSMC to continue diversifying its fabrication base, “including by rolling out new technology nodes simultaneously in [its home country] and in the United States.” He recommended the same strategy for dealing with TSMC rival Samsung.

Though Miller’s recommendation was not a mainstream argument, a former TSMC executive who now lives in the United States said it was, in fact, the U.S. Department of Commerce’s stance, “but they are embarrassed to say it publicly,” 

Beyond political considerations, Apple’s view may count for even more because it has been the main driver of TSMC’s most advanced processes in recent years.

Tim Cook, CEO of Apple. (Source: AP)

An executive in the Apple supply chain said Apple has already begun the verification of products to be made at the Arizona facility, including 5-nm and 3-nm CPUs, as a backup to production from Taiwan. The former TSMC executive believed the process was accelerated by Pelosi’s August visit.

TSMC is likely not thrilled with the arrangement, reflected in the many questions an analyst for a foreign brokerage had, such as “Is it really a good idea to put your best technology on your rival’s [Intel’s] doorstep?” and “Is Apple willing to pay the bill [for the higher costs]?”

In addition, simultaneously churning out chips in Taiwan and overseas using the most advanced technology would be unprecedented for the company. Could TSMC pull it off?

“It would be very, very hard,” said an executive who works for a large TSMC client. If TSMC insisted on simultaneously proceeding with the newest technology node overseas, it could dissipate its R&D capacity and hold back TSMC’s overall R&D progress on advanced technologies, the executive suggested.

Intel vs. TSMC: Different approaches  

For Intel and Samsung, however, the dual approach is less of an issue. 

Intel’s 13th Gen Raptor Lake CPU that just hit the market uses the company’s most advanced Intel 7 process, and some of its Raptor Lake production is being handled by its Fab 28 in Israel. Its under-construction Fab 34 in Ireland will manufacture the next generation Intel 4 process technology.

How can Intel be this versatile while TSMC cannot? A former Intel executive attributed it to Intel’s famous “copy exactly” model that enables a process developed by the parent factory to be precisely replicated at all other foundries, down to the positioning of each machine, the angles of the electrical conduits, and the factory’s ambient temperature and humidity.

At the same time, Intel’s product development department carries far more weight than its manufacturing division, and the manufacturing parameters it sets during the R&D process cannot be changed once the product hits the factory floor.

This rigid process explains in part Intel’s high chip costs, but its consistency has enabled Intel to quickly pump out CPUs at its many factories around the world.

A former R&D executive at TSMC said TSMC takes a very different approach. When the development of a new technology by the R&D team reaches a certain point, process integration engineers test it on TSMC’s R&D test line in Hsinchu and give their feedback.

The technology is then brought to the main production line, and while the first run-through follows the script developed through the R&D line, similar to Intel’s “copy exactly” method, subsequent runs are used to raise yields and increase output. 

TSMC uses a “best known method” model to continually tweak its manufacturing formula and improve its process, in contrast to new Intel technologies, which cannot be modified after they enter volume production.

This is why many observers believe TSMC’s most advanced technologies cannot leave Taiwan, but also explains why its yields are so good. 

Another factor in its manufacturing prowess are its vendors, who help adjust and maintain the machines day and night.

“The CEO of GlobalFoundries wondered why TSMC’s output is 30 percent higher than his company’s using the same machines,” said an executive with a TSMC client.

Whenever a piece of equipment has a problem, regardless of the hour, a vendor will show up as soon as it gets a call for help. 

“TSMC doesn’t just have its own graveyard shift. Every vendor servicing the Hsinchu Science Park has their own graveyard shift. That won’t be the case in the United States,” the executive said. 

This “Taiwan model”, however, is also becoming more challenging in Taiwan itself. New-generation engineers often complain about late night shifts, and in the island’s crowded western half, water, electricity, land and talent are increasingly at a premium. 

As if Apple becomes TSMC's main client, moving TMSC's processor to the US to produce would be Mark Liu Chairman of TSMC's challenge. (Source: Ming-Tang Huang)

Can Chang’s successors succeed where he didn’t? 

TSMC may have reached the limits of growth in Taiwan because of those constraints. 

In Arizona, land and green energy are plentiful, and the state is home to a semiconductor cluster. If TSMC Chairman Mark Liu and CEO C.C. Wei can take TSMC out of its comfort zone and forge a workable U.S.-Taiwan operating model, it will lay the foundation for a new phase of sustainable growth for TSMC.

It was something Morris Chang could not achieve when he ventured into the U.S. decades ago, and it will be a real test of his two successors.

---

Have you read?

• TSMC alone cannot rebuild the US semiconductor supply chain - Taiwanese Hidden Champions need to follow 
• Supply Chain 3.0: the US ambition of Taiwan businesses

Translated by Luke Sabatier
Edited by TC Lin
Uploaded by Ian Huang