Capital is not the problem in building semiconductor fabs
Many governments today talk about semiconductor as a national security matter and are hoping to provide financial incentives in hope of TSMC building (and operating) fabs in their countries.
So far TSMC only committed to Arizona and expressed interest in building fabs in Kumamoto, Japan — at least in public. We have not heard about them even entering into negotiation with other governments about what could be the likely amount of subsidies.
This is because when it comes to building and operating fabs, capital is not the most critical issue.
Here I’ll talk about a few things that matter more than capital in modern semiconductor fabs.
Fab density & proximity
It doesn’t make sense to have a single fab operating by itself in a location. Fabs are best run in density and proximity.
If you visit HsinChu or Tainan Science Park in Taiwan, you will see lots of fabs built next to each other or at least in close proximity. HsinChu and Tainan are separated by 1h24mi of high-speed-rail transport, with a frequency of a train every 40mins. This allows the engineers of TSMC to move between fabs easily, if not instantly.
The reason the process engineers have to be able to move around fabs easily and frequently is because even with the exact same process recipe and fab equipments, each fab could still produce different results. The differences could be due to temperatures, humidity, micro earthquakes, etc. Or sometimes they just came out different and nobody knows why.
This is not a new phenomenon of the insanely fine modern nodes such as 7nm. Back when I was doing my own startup in 2004, we used UMC 0.18um process. UMC had many 0.18um fabs, but the simulation models of different fabs are slightly different. Because despite their best effort to align those fabs, the results still came out different (but consistently). It’s more pragmatic just to let the clients know about it and recommend them to beware of which fab they will eventually manufacture in.
At today’s 7nm node? I can only imagine it being worse.
This is also why most of the process engineers in Taiwan that I know are kind of superstitious. Despite having Ph.Ds from top schools in electrical engineering, chemical engineering, material engineering or physics, they would visit temples routinely praying for the fabs to run smoothly.
There’s also a well-known superstitious practice in putting a pack of Taiwanese snacks called “Kuai-Kuai” next to the machines. “Kuai-Kuai” means “behaves well” in Taiwanese Mandarin. By putting a pack of Kuai-Kuai next to the machines, the engineers hope that the machines will behave well and not give them headaches.
Because machines can behave “not well” and lead to yield issues, process engineers need to come in to “tune the parameters” trying to bring the results back in-line. This kind of tuning is almost black magic and depends heavily on experiences (on top of the doctoral degrees and the smartest brains). Really good and experienced process engineers that could bring machines back to behave well are few. They therefore have to move from fab to fab to firefight.
And since in a 24-hr non-stop operation any down time equals money flushed down the toilets, the quicker the process engineers could move to the fab that flagged an issue, the less the down time and the less the money lost.
This is why fabs in Taiwan are centralized in the two Science Parks. Though I’ve never visited Samsung’s fabs, I’m quite sure they’re also built in clusters with equally intelligent but superstitious process engineers shuttling back and forth among them like ants.
Oh, if you travel to Hillsboro, Oregon, you’ll see clusters of Intel fabs built next to each other as well.
Access to highly skilled labors/slaves
Of course any advanced country would have highly skill labors due to the excellent education systems. The problem is whether one can find lots of labors highly skilled in semiconductors, and ready to wake up, hop in a car to go to the fab upon receiving a text message at 3am on any day.
Today in US most STEM talents not only shun semiconductor in colleges, most just avoids hardware engineering completely. That’s understandable. Why suffer the headache of studying Maxwell Equations when your hero is Evan Spiegel?
Luckily US are an immigrant country. Lots of Taiwanese, Korean, Indian and Iranian talents still aspire to study and stay for work in US. For foreign talents that don’t speak native English, semiconductor is the best way to find a high-paid job.
Europe is a different situation. Even though many talents here still choose electrical, mechanical, chemical and material engineering as they major, they definitely won’t be willing to, upon receiving a text message at 3am on any day, wake up, hop in a car to go to the fab.
I also doubt if European labor laws allow for the (in-)famous 4-classes-2-shifts system that Taiwanese fabs employs. In 4-classes-2-shifts system, fab operators are divided into 4 classes. Each class will operate a 12-hr shift for 2 days, then take the next 2 days off. The operators also mostly live in company dormitories. They get picked up by buses shuttling between the fabs and the dorms to ensure high efficiency.
Can you imagine a French worker willing to live this kind of life?
Why Arizona or Kumamoto?
Now that we know the two other factors more critical than capital, let’s see why Arizona and Kumamoto works for TSMC.
While most people only know about its deserts and Grand Canyon, Arizona actually has a long history of semiconductors, with many companies still operating around Phoenix today. e.g. National Semiconductor, Dialog, NXP, On Semi, Cypress, etc. And since these are more likely the higher paid tech jobs in the area, local talents are interested in majoring in semiconductor-related engineerings in college. TSMC has a lot more flexibility here to manage its necessary talent pool. The first fab they’re building here is supposed to go into mass production in 2024. I won’t be surprised if they break ground for the 2nd one before that, and eventually build a cluster of fabs around Phoenix.
As for Kumamoto, Japan’s electronic makers historically built factories in the country side and built small towns alongside those factories so that the employees could live there. Most skilled semiconductor engineers in Japan actually prefer to live in this kind of environments. Life is simpler. Living cost is much lower. The children will be surrounded by children of like-minded STEM-background parents. Also in Japan lifetime employment used to be prevalent. Losing the job flexibility in metropolitan areas was (and still might be) not that critical to these skilled engineers.
Kumamoto has a few such satellite cities built around semiconductor fabs. The largest being the Sony’s fabs for camera sensor in Kikuchi-gun. They account for such a high percentage of Sony’s sensor production that when the 2016 Kumamoto Earthquake led to suspended production, it sent a supply-chain shockwave through all the camera makers around the world.
With the density of semiconductor fabs, suppliers and talents in Kumamoto, no wonder it’s of interest to TSMC to build a fab there.
What should countries that are not US or Japan do then?
To be honest I don’t know. I do not have an answer.
Semiconductor is difficult. Asking high-income countrymen to do difficult jobs while there are other options to make the same or higher incomes and have more fun at the same time is a tall order.
The more realistic way would be for these countries to build deeper trade relationships with Taiwan and South Korea, and be mindful of the fact that it takes a lot of sacrifices of the people of the two democracies for them to become the leaders in this highly competitive industry.
As for capital incentives in the form of subsidies? Yeh, that’s good to have but it’s not nearly as critical.
