For years, ASML built Ferraris. Now it has to make a Volkswagen – and that’s deceptively difficult.
The engineers currently working on a hybrid-bonding machine in Veldhoven must be having the time of their lives. For years, they were constrained by nanometer budgets – the limited margin they were given for their part of the system. Now, they suddenly have the freedom to pull off an entirely different trick.
As an avid tinkerer and amateur systems engineer, I enjoy putting myself in their shoes. What must that feel like? Spending years working on a Ferrari. Cost was almost no object – everything to stay ahead in the litho race.
Some have spent their entire careers working with the very best money could buy: the most powerful magnets, noble gases, titanium – you name it. They could choose freely from production techniques and tooling. Of course, it wasn’t truly unlimited. Their hands were tied: they had no room whatsoever beyond the agreed nanometer budget.
And then suddenly, a change of course – what would that feel like? Taking all that nanometer-level expertise into a completely different challenge: much more freedom in precision, but with new hard constraints. At least twice the speed of the competition, ruthless cost reduction in the supply chain and delivering a dramatic drop in cost per placement for customers.
The lucky ones now developing hybrid bonding in Veldhoven are entering an entirely different playing field. They’re tuning a Volkswagen. In that league, the rules are very different and there are many more players. Among the systems, you encounter clunkers, but also perfectly refined models. Most achieve alignment accuracy of around 200 nm, the best reach 100 nm. But they all share the same handicap: they’re painfully slow.
From what I hear in Korea and Taiwan, that’s exactly why major chipmakers have asked ASML to enter the race. It has to be faster, more reliable and so on. TSMC and Samsung know that their lithography supplier will push itself to the limit. Yes, I’m explicitly mentioning the Koreans – because I’ve heard from reliable sources that Samsung is also pulling on ASML.
You might think hybrid bonding would be easy for engineers used to working at sub-nanometer alignment precision. But outsiders tend to jump to conclusions too quickly – and, to be honest, I’m guilty of that myself. It was somewhat amusing to see the Korean publication The Elec suggest last week that maglev stages are increasingly being used in hybrid bonding. I have my doubts. First, because it can be done much more cheaply. Second, because it’s questionable whether the wafer stage is the critical factor in achieving alignment accuracy of, say, 10-20 nanometers at very high throughput.
If Veldhoven is indeed in catch-up mode, they’ll want to demonstrate results quickly. It wouldn’t be surprising if they’ve pulled an old stepper or scanner out of storage to build a test setup and show what’s possible. An old PAS 5500, a few fast motors, modern drives and control – and you’re already a long way there.
It reminded me of the very first stepper that Herman van Heek developed at Philips’ Natlab in 1973. To demonstrate the stepping principle, he used the base of an existing precision instrument (the Opthycograph) to build his first wafer stepper. The drawback was that during stepping, the entire lens column had to move (in the y-direction, while the x-movement came from the wafer stage). This meant the system had to settle after each step, reducing throughput. For Van Heek, that wasn’t an issue – he only needed to prove that he could pattern chips contactlessly in photoresist.
The positioning challenge resurfaces in hybrid bonding. A chip needs to be placed precisely onto a wafer. This isn’t comparable to the placement in pick-and-place machines used for PCB assembly. Placing a die in a hybrid bonding process requires a much more delicate motion and controlled touchdown. The placement head has to bend the chip slightly and, after initial contact, release it in a controlled way. The head has to allow the bond to propagate from that initial contact point (the center or a centerline) across the entire die – perfectly aligned with the corresponding copper contacts on the opposing circuit on the wafer. Die and wafer must merge in one continuous motion – somewhat like applying a glass screen protector to a smartphone. Needless to say, bubbles and particles in between the two are unacceptable.
From what I understood from a presentation by Jonathan Abdilla, technical marketing director at Besi, at Semi’s latest 3D & Systems Summit in Dresden, the placement head is the most critical component. ASML may want to reinvent the wheel entirely, but the Eindhoven region already has a great deal of expertise in pick-and-place technology. Just look at the (discontinued) development of placement machines for mini- and micro-LEDs at Kulicke & Soffa or at Itec (a Nexperia subsidiary) in Nijmegen.
To make things even more interesting, it’s not just hard physics at play; soft power matters too. ASML CEO Christophe Fouquet is currently [challenging his employees](https://www.linkedin.com/feed/update/urn:li:activity:7442179018994368512] to be more innovative. Hybrid bonding seems like the ultimate test case. It’s no small feat for a highly successful giant to deliver in an extremely cost-sensitive market.
Meanwhile, ASML is actively searching for suppliers across Asia. “ASML has finally woken up,” a front-end equipment supplier told me in Singapore. He was also surprised to see the litho giant setting up teams of dozens of people in countries like Malaysia. His short comment: “Totally oversized.”
And to what extent will a “build-it-yourself” mindset hold ASML engineers back? If speed is essential, it may be wiser to rely on the supply chain. After all, it was the network of suppliers around Eindhoven that enabled ASML to scale in the 1990s. But today, those same suppliers have become somewhat spoiled by ASML’s success.
Top image credit: ASML
