Let's start with a picture for our dear readers...
On stage stood a lineup of tech giants, including U.S. President Biden, alongside NVIDIA's leather-jacketed Huang, AMD's Su Ma, Apple's Cook, and others.
But judging by their positions, it was evident that the real stars of the day were the three central figures from TSMC: founder Zhang Zhongmou, CEO Wei Zhejia, and chairman Liu Deyin.
What brought about this rare spectacle was the first machine installed at TSMC's U.S. wafer fabrication plant at the end of 2022.
And just like that, it got Biden all excited, saying, "Chip manufacturing is coming back."
Yes, you read that right—America's chip manufacturing, it turns out, has been lost...
All along, when people talked about America's strong chips, there was a qualifier: chip design. Apart from design, there's also wafer fabrication, circuit etching, testing, packaging, and other processes, all of which fall under chip manufacturing.
Let me put it this way: since the last century, their semiconductor manufacturing share has been like a slide, just going down and down.
The White House released the global semiconductor manufacturing share:
Apart from Intel, which is still holding onto its chip-making efforts, other companies like NVIDIA and AMD are practically begging Taiwan Semiconductor Manufacturing Company (TSMC), Samsung, and others for help.
The former semiconductor powerhouse has been losing its manufacturing business, and the reasons behind it have been analyzed to death over the years.
The mainstream view is basically this: during the Cold War between the US and the Soviet Union, the US supported chip technology in several East Asian countries and regions, only to be outpaced by Japan.
Realizing they couldn't compete, and amid deindustrialization, the US shifted its focus to the top of the value chain, prioritizing chip design. They then facilitated industrial relocation, believing they could maintain dominance by other means and control the chip hegemony.
But, did the US really not care about chip manufacturing? A recent revelation from Bloomberg may provide an answer. It turns out the US has been trying to revive its chip manufacturing industry all along.
And in the realm of EUV (extreme ultraviolet lithography) machines, the US almost turned the tables... but a single film caused them to miss a golden opportunity.
The lithography machine is likely familiar to many, as it's the device used to etch circuits onto wafers. EUV is currently the most advanced technology in this field. Before that, deep ultraviolet (DUV) lithography machines held the top spot.
However, the "strongest" silicon valley lithography group in the United States has been unable to surpass Japan's Nikon-Canon and the Netherlands' ASML in DUV lithography machine production. So, they're planning to take a different path to catch up, focusing on early development of EUV technology.
As early as 1997, the US Department of Energy teamed up with Intel to form the EUV LLC alliance. They not only enlisted the participation of three major national laboratories but also attracted scientists from companies like Motorola and AMD.
With everything in place, the US was caught off guard when it was discovered through comprehensive evaluation that EUV lithography machines were much harder to develop than DUV lithography machines. Moreover, their own Silicon Valley lithography group's technological prowess fell short.
Nevertheless, since the alliance had put in so much effort, they had to find a factory capable of producing the lithography machines. Besides Japan's Nikon, the only other option with such capabilities was ASML from the Netherlands.
Before America could shake off the shadow of Japan's household theft, ASML naturally seized the opportunity. However, not being a native son, ASML had to sign a series of "selling oneself" agreements before reaching cooperation, such as sourcing over half of the core components of its lithography machines from the United States, among other terms. After forging this kind of strategic partnership, Intel subsequently invested $4.1 billion in ASML, becoming its largest shareholder. It can be said that the great revival of the American semiconductor industry hinges on this move!
Of course, it turns out the U.S. investment in ASML wasn't in vain. In 2018, the EUV lithography machines went into mass production.
Now it's time to reap the benefits... As long as Intel says the word, ASML is ready to supply. However, the U.S. side seems to be dropping the ball again. Intel just can't seem to open its mouth, and as a result, they missed out on making the purchase.
When ASML initially began mass-producing EUV lithography machines, there was a critical technology hurdle: EUV pellicle. With Intel's technology at the time, there was no way to bypass this layer, and it didn't dare to bet on this technology being conquered in a short time.
But don't underestimate this technology just because of its name. It's a game-changer when it comes to the yield of chip manufacturing with lithography machines.
In operation, a lithography machine essentially shrinks the circuits on the mask onto the wafer.
But there's an issue in this "copying" process. The environment might contain dirt particles like dust, and if they land on the mask, the lithography machine will faithfully print them onto the wafer.
It's crucial to note that circuits on chips are incredibly tiny. Even a small defect means the printed chip is basically trash.
Moreover, according to media outlet Anandtech, since Intel had only one customer at the time, they typically used single-chip masks. This means that if any dust landed on one mask, the entire wafer was wasted.
This is where the EUV film comes into play. Positioned a couple of millimeters above the mask, it acts like a "guardian with a sword," keeping dust and such away from the mask's focal point. If it's not on the focus, it won't be imprinted onto the chip.
Comparison of chips produced without EUV film (left) and with EUV film (right)
But making this film is no easy feat.
First, EUV film has stringent requirements for extreme ultraviolet (EUV) light, which knocks down most materials. Even meeting this requirement, ensuring sufficient transparency requires making the film extremely thin.
Once thin, the film must withstand the ultra-high energy of EUV light, remaining undistorted within the temperature range of at least 600 to 1000 degrees Celsius.
In short, producing this film poses extraordinary challenges. That's why it took three years after the mass production of EUV lithography machines for EUV films to barely enter commercial production.
Since they couldn't wait for EUV lithography, why did TSMC boldly bring home EUV machines as soon as they went into production?
The reason is quite simple: they dared to gamble, and they had the capital to do so. When TSMC first bought EUV lithography machines, there were rumors circulating that they had developed a technology called laminar air knife, which had similar effects to EUV films.
Moreover, TSMC had orders from various manufacturers and for different types of chips. According to Anandtech, they mostly used multi-chip reticles. For instance, a reticle with 25 chip circuits, even without EUV films, could still maintain a yield rate of 96% in the presence of a single speck of dust.
In contrast, at that time, Intel didn't even have a decent hand when it came to chip manufacturing. They were really hesitant to gamble.
They couldn't even place their bets, so how could Intel even join the game at the table?
According to Bloomberg, it was precisely because Intel didn't adopt EUV lithography machines in time that the U.S. missed the opportunity to reclaim the peak of the chip manufacturing industry.
However, from a global perspective, it wasn't that Intel's failure to acquire EUV lithography machines led to the downfall of the US chip manufacturing industry; rather, it was already in decline, and the EUV lithography machines simply exposed this fact.
Of course, the US knew its chip manufacturing capabilities were slipping. Over the past few years, they have been openly and covertly bolstering their own semiconductor manufacturing industry.
As for Intel, in 2021, they made a significant move by announcing the IDM 2.0 plan. In summary, they're following in the footsteps of TSMC, expanding their production capacity to prepare for contract manufacturing. They even established a dedicated foundry services division for this purpose.
Even with the emergence of more advanced lithography machines, Intel hasn't hesitated. For instance, earlier this year, when ASML unveiled its new High-NA EUV lithography machine, Intel wasted no time in placing an order. Just a few days ago, they announced that assembly is already complete.
TSMC, on the other hand, seems to be taking a more cautious approach this time, opting to observe a bit longer...
The United States is also getting smarter, planning to put more eggs in multiple baskets. In recent years, they've been finding ways to entice TSMC to build factories in the US. Apart from chip subsidy bills, there's been subtle pressure from the authorities. As of now, TSMC has promised three factories to the US.
However, it's easy to lose the semiconductor manufacturing industry, but not so simple to regain it. TSMC started building a factory in 2021, but it's already stalled. The US can't even recruit the basic talent needed for construction, and the production timeline has been pushed from 2024 to 2025.
Looking at the stumbles the Americans have made in chip manufacturing, it seems there aren't many opportunities for overtaking on the inside track in this area. It's more about straight-line advancement with cooperation across the entire industry chain.
As for when the US can revitalize its semiconductor manufacturing industry, we don't know. At least in the short term, it seems hopeless.
Author: Squirrel Editor: Jiang Jiang Cover: Confucius
Image and data sources:
Reddit, the internet
Bloomberg, The US Had a Chance to Lead in Chipmaking Tech, and Missed It
Semiengineering, EUV Pellicles Finally Ready
Anandtech, EUV Pellicles Ready For Fabs, Expected to Boost Chip Yields and Sizes
AI Chip World, Depth: When EUV Pellicles Become the Key to "High-Precision" Chip Yield
As for why Intel's process has been leading all along but suddenly got stuck at 14nm for so long?
