When Mino Morgese joined Taiwan Semiconductor Manufacturing Company in late 2021, his new employer arranged for a psychologist to prepare him and other newly hired US staff for the culture shock that might arise from working for a Taiwanese company.
Less than two years later, the Italian-born, US-trained engineer has embraced life at the world’s largest semiconductor manufacturer — so much so that he is managing the move-in of machinery into the new fabrication plant, or fab, in Arizona at TSMC’s first big manufacturing base in America.
“It doesn’t happen often that such a huge fab is being built and that it’s all cutting-edge technology,” says Morgese, who feels pride in being one of the pioneers helping to build the plant.
Sitting on a site the size of 640 football pitches in the desert on the northern outskirts of Phoenix, a cluster of hulking buildings and giant cranes around them are visible from highway bridges and hillsides 30km away. Working at TSMC, Morgese says, “there’s much more detail expected from us on every single level. [In previous jobs] we would not dig in as much as probably we should have.”
TSMC commands the most state-of-the-art chip production technology in operation in the world today, and controls 90 per cent of the global market for the most advanced made-to-order chips. But the Arizona project, where TSMC is building two fabs at a cost of $40bn, has given the company a culture shock of its own, and put both it and its employees on a steep learning curve.
So far, TSMC makes its most advanced chips exclusively in massive factories in Taiwan. Research and development staff nearby jump in to fix any problems, suppliers are clustered tightly around, and services providers who have been involved in building the company’s fabs for decades respond to its every whim. That well-oiled model has helped make TSMC more profitable than most other semiconductor manufacturers.
But growing anxiety among governments and industry in the US, Europe and Japan that a Chinese attack on Taiwan could leave the world without chip supplies is forcing the company to spread out its manufacturing footprint. Apart from Arizona, the company is also setting up a new plant in Japan and has committed to a €10bn joint-venture fab in Germany.
“Arizona is the first trial of our overseas mega-site development,” TSMC chair Mark Liu said in September. “It is a learning process of course.”
Whether that trial goes well could make or break the US’s quest to resurrect semiconductor production on its shores — an effort underpinned with subsidies worth $39bn for new fabs, which constitute part of the country’s largest public investment in industrial policy since the second world war. The Arizona project could also determine if TSMC can transform itself into a true multinational.
But the first two years have been rough, threatening the project’s timeline and the company’s ambitions. The Taiwanese company has struggled with the American approach to construction and labour, says Dylan Patel, chief analyst at SemiAnalysis, a California-based chip industry research and consulting firm. “TSMC didn’t quite get the culture.”
In July, Liu told investors that the date for starting production would be pushed back from next year to 2025.
How big the Arizona factory will become is also unclear. The two fabs the company has announced so far would take its monthly capacity there to 60,000 of the wafers on which circuits are printed and microchips cut. That would make Arizona a medium-size plant but far smaller than the “gigafabs” with more than 100,000 wafers a month that the company operates in Taiwan.
TSMC has acquired enough land for four more expansion phases which could bring Arizona to that scale, but it has not publicly committed to that. Several executives at TSMC suppliers now say they do not believe the company will build out all six phases.
“This is where there is a lot of uncertainty and tension,” says Caitlin Legacki, a former senior adviser to US commerce secretary Gina Raimondo who now runs a public affairs consultancy. “This is uncharted territory for TSMC, so it’s understandable why they are taking it a bit slowly.”
Construction challenges
Building the first stage of the plant has been rife with challenges.
The company’s way of awarding engineering contracts caused a lot of friction, leading to delays. In the US, chipmakers tend to choose three separate contractors for the office building, the central utility building which houses all electrical and plumbing infrastructure, and the cleanroom where the chip manufacturing itself happens. But TSMC typically subdivides the work in 20 or more packages.
“TSMC will slice those contracts very thinly for the sake of saving cost,” says Charles Lee, co-chief executive at Topco, a Taiwanese company which manages certain chemicals and parts supplies for TSMC. “They are used to handling everything in a top-down manner. In Taiwan, contractors are used to following those orders obediently and adapting to quickly changing instructions, but in the US there will be miscommunication.”
More recently, TSMC has struggled to find enough skilled workers for the installation of key advanced machinery. In June, the company sent 500 additional specialists from Taiwan to help in that phase.
Several of the new arrivals gathered for a barbecue late September at Aspire Pinnacle Peak, a gated apartment community 12km from the TSMC plant where they are staying.
Lin, an employee who asked to only be identified by his surname, says he had come to help set up extreme ultraviolet lithography, or EUV machines, the equipment made by Dutch company ASML that is needed to make the most cutting-edge chips.
Some industry experts say there is indeed less expertise in the US with regard to EUV. Apart from Intel’s technology development centre in Oregon, no US fabs have EUV installed yet. TSMC moved in its first EUV machine at its Arizona plant in May.
The delays in Arizona are also complicating things for TSMC’s supply chain.
The chipmaker relies on many suppliers at its home base. “They have a very strong ecosystem in Taiwan — spare parts and machining shops, some of the bulk chemicals, move-in and move-out shops. But many of these are small companies, and the question is how many of those folks will move?” says Peter Hanbury, a semiconductor expert at consulting firm Bain.
Several Taiwanese chemicals suppliers to TSMC, including Chang Chun Petrochemical, Kanto PPC and Topco, bought land in Casa Grande south of Phoenix with the intention of building plants and warehouses. For some of them, the fact they have to handle all planning work and permits for road connections and utilities themselves, as opposed to move-in ready plots in industrial parks back home, has come as a shock, together with drastically higher construction costs.
“TSMC’s construction cost is several times higher in the US than in Taiwan, and the same multiple applies to us,” says Vincent Liu, president of LCY, which makes cleaning chemicals used in TSMC’s chip production. “The accurate calculation should be five times.”
Fab capabilities
But the real challenge for TSMC, executives and analysts believe, will be to run the new fab as efficiently as its Taiwan-based ones.
“We believe it can be profitable, although not as profitable as our Taiwan fabs,” says one TSMC executive. The company could replicate fab operations in the US that are already in mass production in Taiwan, the executive says, but is not confident that it could build up production in Arizona using technology only recently developed in its global research and development centre.
The company’s success has been built on its ability to quickly raise yield — the ratio of chips without defects — at every new stage of process technology, which it has achieved by empowering engineers to experiment on the fab floor.
This contrasts with some other chipmakers’ practices. Intel, for example, has for many years followed an approach called Copy EXACTLY!, under which each new fab must operate under the exact parameters transferred from the company’s technology development centre in Oregon.
Any request to deviate from that has to be approved through a written process including experiments to demonstrate the merit of the proposed changes. The company needs this controlled process to ensure consistent products across its network of fabs, which operate in a number of US states and in countries from Ireland to Israel.
At TSMC, however, “fab workers have more autonomy to tweak dials to improve yield”, Patel says. “That is what makes them achieve such efficiency which Intel has never had.”
Patel sees TSMC’s model as the main reason why the company’s process technology has a far longer lifespan than that of its rivals.
But TSMC’s approach relies to a large extent on having the R&D team, which develops new process technology, only an hour-long high-speed rail ride away. With 20 hours of flight time between Arizona and headquarters, new procedures are needed.
Engineers with experience at other chipmakers say it is possible to start and run satellite fabs over vast distances. “I have seen this in practice when we ramp up production in Ireland or Israel,” says a senior engineer at Intel.
“The new fab will send a seed engineer to the technology development site in Oregon for anytime between a week or three months, so he can copy exactly what is going to be done. Then that person goes back and helps implement it on the other end.”
If you work at TSMC for a few years and can withstand the culture of drilling, you can join almost any other semiconductor company
As TSMC works towards getting its first Arizona fab ready, it has embraced aspects of the strict metrics mindset. Topco chief executive Lee says material supplies to Arizona are all exact copies of what is used in Fab 18, the site in the southern Taiwanese city of Tainan, where TSMC is running the same process technology that will be used in Arizona.
“They are even demanding that when we ship raw wafers to the Arizona fab, they must be from the same lot as those used in Fab 18,” he says.
Yet Morgese’s experience at TSMC indicates that the company, while taking a page out of Intel’s book, is still trying to preserve its more flexible approach.
Last year, Morgese spent six months at Fab 18. Hired as a manager for chemical-mechanical planarization, a certain kind of polishing process, he shadowed his counterpart in Fab 18 and then started solving related assignments on his own.
“When you transfer technology from Fab 18, your benchmark is how you match the yield from [that] fab,” says Morgese. He and his fellow engineers who are preparing for the start of production in Arizona first ensure that all processes and tool settings exactly match those at the fab in Taiwan.
But once they begin putting silicon wafers through the equipment, “it’s about making the small tweaks to see how we can optimise the process”.
Hiring locally
But industry experts caution that the flexibility TSMC grants its Taiwan fab workers will be hard to replicate with a US workforce.
“You can empower fab workers more as long as they are that highly skilled, but you can’t have that in the US,” Patel says. Technicians in Taiwan typically have completed four years of engineering school, he adds. But engineering graduates in the US “have job opportunities that pay more and are more inspiring, like developing new lenses for Apple or working for Meta, than in a fab.”
In Phoenix, that problem is clearly evident. TSMC has already hired more than 2,200 of the 4,500 staff it plans to employ there once the two fabs are in production. But almost half of those hired so far are assignees sent from Taiwan, according to two people familiar with the situation.
The company is trying to develop talent by supporting several local engineering schools, including at Arizona State University.
Engineering students there view TSMC as an interesting company because of its manufacturing technology leadership, but are wary that working there could land them in a factory job without opportunities for innovative work.
“I would consider working at TSMC after I finish my masters next year because the pay is really good and their technology is really good,” says Garima, a graduate student from India who would only give her first name.
“But from friends who work there I know that the hours are too long and it is very hard to get leave,” she says. She claims that TSMC wants their workers to be obedient “robots”, something she describes as a feature of Asian corporate culture also seen in her native India. “They need to go more towards Intel if they want to be attractive here,” Garima adds.
Many young engineers are starting to view TSMC as a good career start but plan to jump ship early. “If you work at TSMC for the first few years and can withstand the culture of drilling, it’s like you can join almost any other semiconductor company,” says Smitha Swain, who is doing his PhD at ASU and has friends who have worked at the company. “When people join TSMC, they spend a certain time in Taiwan to get trained on the tools, and when they come back, all the local companies try to pay them more and hire these trained people, so it’s an ongoing battle.”
Michael Kozicki, a professor at ASU’s school of electrical, computer and energy engineering, says TSMC faces a very different job fluidity environment in Arizona which the company is not used to in Taiwan, where it is among the most cherished employers.
Analysts say TSMC’s challenges will differ across its new manufacturing locations. Construction of its Japan fab is going much more smoothly than in the US. “Our cultures are closer to each other, and Japan has a similar manufacturing mindset and work ethic as we do,” says a TSMC engineer familiar with both projects.
As the fabs in both Japan and Germany will be using more mature technologies than the Arizona site, both construction and initial production should pose fewer challenges, analysts say. “But I suspect Germany will be more difficult in terms of labour issues,” says Hanbury, of Bain. “Consider the lessons from Arizona and magnify.”
TSMC is keenly aware of these deep-seated differences. Apart from the technical training, the company is putting foreign staff through language and culture training similar to the course Morgese underwent.
Over the past 12 months, about 6,000 TSMC employees have undergone intercultural training at TLI, a Taiwanese consultancy, according to people familiar with the situation.
After [TSMC] has developed a good reputation in the new location, it needs to develop a stable, sustainable local workforce
This includes training domestic staff to work more effectively abroad. TLI has seen a surge in demand from Taiwanese companies under pressure to globalise over the past five years as foreign customers and governments demand a diversification of tech supply chains away from China.
“Often they seek help just because Taiwanese management gives some orders and is surprised because there is a negative reaction from foreign staff,” says Yvonne Wu, deputy general manager at TLI. “But the much larger demand will be for training Taiwanese executives and middle managers to ready them for a globalised work environment.”
Consultants say TSMC and other Taiwanese companies expanding in the US or Europe initially sent large numbers of their existing staff because they struggle to hire and retain enough local staff to scale up business quickly.
“But after they have reached the scale and developed a good reputation in the new location, they need to systematically develop a stable, sustainable local workforce,” says Wu. “They cannot rely on running it with Taiwanese forever.”
Even if TSMC’s Arizona trial succeeds, the effect for diversifying global chip production away from Taiwan will remain limited due to the company’s massive scale.
Analysts and executives estimate that after the first two phases of the Phoenix plant reach full capacity and the company’s fabs in Japan and Germany are finished, total sales from production outside Taiwan are unlikely to account for more than 10 per cent of TSMC revenue.
But US observers say that is at least a first step towards rebuilding the country’s chip manufacturing capability. “Having some capacity,” says Legacki, the former Commerce Department adviser, “is better than zero.”
Cartography by Liz Faunce
Read the full article here