Discover more from The Asianometry Newsletter
A Deep Dive into Immersion Lithography Technology
Author’s note: If you want to watch the video, it is below
This newsletter is brought to you by my friend at Nightowl Sweets. Purveyor of fine baked goods in the Taipei area.
I know this might sound like a strange shoutout, but Hui Shan is a good friend of mine and she bakes great stuff. Go follow the Insta and if you live in Taipei, order some amazing cakes. I’d really appreciate it and you won’t regret it.
I get it. Everyone wants to talk about EUV. It's the sexiest lithography around with all the mirrors and the purple UV light.
But I think we shouldn't discount 193 nanometer immersion lithography. Coming about in the early 2000s, it has taken the industry further than anyone could have ever expected.
193i as it is sometimes called is still a workhorse and at the core of many leading edge process nodes. For this video, I want to look at how it works and the challenges overcome in developing it.
How Immersion Lithography Works
Let us first start with how this whole immersion lithography thing actually works. Previous lithography techniques had an air gap between the wafer and the machine.
Immersion lithography replaces that air gap with water. So now you are shooting ultraviolet light through water at the wafer. You have to literally stick the lens into water.
Why does this work?
A photolithography's attainable structure size is dependent on the resolution of the tool's optics. This resolution, sometimes also referred to as the critical dimension, is calculated using the famous Rayleigh formula.
I have discussed this formula in previous videos, but here it is again.
CD = k1 x λ / NA
Lambda λ stands for the light wavelength, which is the same. k1 is a process factor with a physical limit and we are already near it.
Immersion lithography improves lithography resolution by increasing the NA, or "numerical aperture".
It goes from a previous maximum of 0.93 to 1.35 or higher - collecting and focusing more light.
Immersion lithography derives from immersion microscopy. It is an old technique that dates back to the 1840s, when microscoper Giovanni Battista Amici first started using oil and after that water to improve the quality of his microscope images.
However, the first proposals to apply immersion principles to lithography emerged in the 1980s. A US patent filed in 1984 by several Hitachi employees talked about putting a liquid between the lens and the photoresist - the part that contains the IC design.
In 1985, Werner Tabarelli and Ernst Lobach, working for the American company PerkinElmer filed a patent that had the lens directly sitting in water.
Then in 1987, Burn Lin - at that time working for IBM - proposed using it to enhance the depth of focus for the then-emerging 249-nm lithography. 193 at the time was still in the conceptual stage.
In 1989, Hiroaki Kawata and his team at the University of Osaka demonstrated the technique for the first time at a 453-nm wavelength.
From then there on, the idea sort of receded into the background as the lithography industry continued on its development pathway.
Over the next decade, the semiconductor industry focused on the development of 193-nanometer lithography. This wavelength brought substantial challenges. For instance, most materials absorbed this type of UV light, requiring most new components to be re-engineered.
193 nanometers was eventually delivered, but what would come after it? The original idea had been to go to the 13.5 nanometer wavelength with Extreme Ultra Violet or EUV. However, delays in the technology made this unworkable for the upcoming process nodes.
So the industry pivoted towards 157 nanometer lithography - the last possible optical lithography technology.
I discussed 157 in a prior video. Progress on the technology was difficult, troubled by substantial challenges.
The Move to Immersion
In the summer of 2002, SEMATECH sponsored a 157 nanometer lithography workshop for the general industry. Dr. Burn Lin, now an R&D executive at TSMC, had been scheduled to do a speech there.
As originally planned, the speech was to discuss immersion lithography techniques and fluids to extend 157 nanometer lithography’s useful life. Several researchers had tested immersion techniques to extend 157 a few more years.
Having done experiments on immersion a decade ago while at IBM, Lin was qualified to discuss the technique.
Lin's remarks took a wild turn. In it, he finally spoke aloud what had been on everyone's minds. That the 157 nanometer emperor had no clothes and its technical challenges were too substantial to scale.
He instead proposed that people apply immersion techniques to the already-existing 193 nanometer lithography technology.
It would be first time the industry would transition from one wavelength to the same wavelength.
Lin's speech was the Ether of semiconductor lithography. About $2 billion dollars across the entire supply chain had already been invested into 157 development, and now Lin was dropping this sick diss track ... and at a 157 nanometer workshop no less.
But he was right. SEMATECH quickly organized another seminar in December, where everyone went over the 10 essential hurdles to 193i technology. Tests and experiments were run to see if those can be overcome. And they can.
For a while, the industry hemmed and hawed between 157nm and Immersion. Then in May 2003, Intel dropped 157 and the industry soon followed. Everyone pivoted to developing 193i technology fast enough to make the next major process node.
Looks like 193 nanometers was back on the menu, boys.
193i sounds like a simple spinoff of 193 nanometer technology. But what you are doing is literally throwing water on what is already a massive Jenga tower of precarious, delicate technologies.
New lithography technologies compete based on three market criteria: Resolution (which we mentioned earlier), accuracy, and productivity.
Resolution and accuracy make intuitive sense. A higher resolution machine can pattern increasingly smaller lines. A more accurate machine patterns without errors.
Productivity - sometimes also referred to as throughput - is not so often mentioned but is just as important. How fast can your machines make wafers?
Often measured as the number of wafers processed per hour, throughput is not to be underestimated.
One of the reasons that ASML TWINSCAN machines sold better than their Nikon and Canon counterparts is that they can process far more wafers.
A lithography machine needs to deliver on all three. If it cannot do that, then what you have is a laboratory experiment, not a shipping technology.
All three lithography makers chose ultrapure water as their first-generation immersion fluid. The right immersion fluid should not cause physical changes to the wafer, lens, or the applied coatings. It needs to flow freely, cannot absorb too much of the UV light and cannot be too susceptible to forming bubbles.
Ultrapure water met all of these criteria. Only one other liquid - Perfluoropolyether, an aerospace liquid lubricant - was seriously considered to start, but water made more sense. They were already using it in wafer fabrication.
The foundries would choose different fluids for later generations of immersion lithography.
Lithography makers experimented with a variety of ways to present the water in between the lens and the wafer. In order to keep the water fresh and prevent contamination on the wafer surface, the water has to be constantly moving.
First, they considered a simple circulating bath, technically called the wafer-based configuration. The wafer is immersed in a circulating tank that constantly brings water over and under the wafer. Dirty water is constantly being pumped out as new ultrapure water comes in.
It kind of reminds me of a jacuzzi.
The single biggest issue with this setup was that you had to fill and empty the whole jacuzzi - I mean, bath - with water every time before loading and unloading the wafer. This adds time to the process, slowing down throughput.
It also required a substantial redesign of the existing lithography machine's entire routing and setup, which seemed unnecessary.
They ended up choosing another set up - called the shower configuration.
Here, you pump water to flow only over the part of the wafer that is being exposed by the lens.
After the lens passes by, a suction picks up the water again.
This conceivably works better, and the foundries eventually chose this approach. The showers use less water overall. It can be integrated into existing 193 nanometer machines. And technicians do not have to wait for a bath to fill and empty first.
The challenges were how to keep the water flow consistent, how to suck up all the water so nothing gets left behind, and preventing watermarks especially on the surrounding areas of the lens.
The wafer's edges were particularly tricky. Yields at the edge are already half of that of the rest of the wafer. And now, when the water pump gets to the edge of the wafer, water can leak through the cracks. Then it can get trapped underneath the wafer, creating a vacuum.
If protection and edge inspection tools are not built to prevent water from getting underneath the wafer, then the water can suction the wafer to the tray and suck up contaminants - risking damage.
Another significant issue had to do with reducing the number of bubbles, stains and particles. During the lithography stage, there is a lot of movement.
The machine is moving the wafer around as the machine does its light exposures - about 500 millimeters each second. Furthermore, the water is being pumped very quickly through the system, on and then off the wafer.
When you shake water around like this, it tends to create bubbles. Early ASML demos uncovered this issue, finding bubbles in the 1 to 150 micrometer size - twice as big as a human hair.
Bubbles in the water scatter photons, which has two consequences. First, if the bubble is large enough - generally larger than the light wavelength in question - it creates flair imaging artifacts right on the chip;
And second, it weakens the DUV beam's power because photons are being carried away. A weaker beam means that the chip design image won't be as strongly imprinted onto the wafer.
They eventually learned that foundries would have to "de-gas" the water before it enters the immersion zone. De-gassing is now a regular part of the ultrapure water generation cycle.
Putting water directly on the wafer creates new contamination issues. SEMATECH particularly focused on water contamination and leaching issues relating to the resist.
Part of the lithography process involves the application of a photoresist chemical that reacts to the DUV light exposure. Water can carry contaminants from the wafer to the optics lens, or vice versa.
And if water somehow leaches away enough of the resist's chemical components, the resist might lose its ability to absorb DUV light.
In order to prevent the resist from washing off and contaminating the lens, foundries developed and applied a protective layer of what is called "topcoats" to the wafer.
This works, but it also adds more complexity to the semiconductor manufacturing process - creating more opportunities for errors and defects. The industry has generally asked not to do this.
The alternative would be to modify the photoresist chemical itself to make it hydrophobic or waterproof. But apparently only specifically during the water immersion stage - so engineers are working to make the photoresist's waterproof features come out only after a spin or bake step.
It is kind of amazing to consider the chemistries involved for these "self-segregating" photoresists as they call it, but that's where semiconductor manufacturing has gone now.
And finally, the optics. We cannot discuss an ASML lithography machine without also mentioning the Zeiss optics. Zeiss had to develop a brand new optics system for the immersion machine.
The old DUV machine had used an optics system made up of just refractive or dioptric lens.
But having a higher NA means adding even more complexity to the optics system. An NA of 1.2 would have increased the system's size 6 times over, which is unacceptable.
So Zeiss switched from a dioptric system to a catadioptric one - mixing lens and mirrors together to greatly cut down the system's size.
The problem with adding mirrors however, is that mirrors split beams into two - the original beam and its reflected twin - and you want to make sure that the two do not interfere with one another.
Several options were tried and rejected, due to vibration issues, polarization issues, and not meeting the required NA number.
The design they eventually chose is called an inline catadioptric system.
The light bounces back and forth between multiple mirrors carefully arranged so that the light pathways don't interfere with one another.
In 2003, ASML shipped the first prototype immersion lithography system to customers - the TWINSCAN XT:1150i.
They also shipped a pre-production version of their dry ArF tool - the XT:1250i.
The first immersion stepper to be shipped in volume was the XT:1700i, which was released in 2006.
ASML's Japanese rivals - Canon and Nikon - took longer to develop immersion. Canon never shipped one in the first place. Nikon's first prototype tool was completed in October 2004, nearly two years after ASML.
But the race was not over. After the first generation of immersion lithography tools, ASML and Nikon immediately moved to developing the next generations of immersion lithography.
These later generation systems raise the NA to as high as 1.92 so that foundries can pattern features down to 40 nanometers. But they also require new immersion fluids like hydrogen phosphate or aluminum chloride as well as brand new optics systems.
ASML handily won this race, gaining the market share dominance that they are today known for. The leading 193 nanometer immersion systems are ASML's now-mature XT:1900 series, introduced over a decade ago.
In 2010, ASML sold their 100th such immersion DUV tool, an amazing achievement.
Having been used in the West for nearly two decades now, some people might be tempted to label matured immersion lithography equipment as "old" or "legacy equipment".
But the reality is far from that. This is not a mechanical calculator, guys.
Modern Immersion lithographies can take you all the way to the N7 node. 193i is just one generation away from leading edge EUV, and there are indications that EUV might never become the workhorse 193i is right now.
ASML and Nikon are still refining their latest immersion lithography machines, regularly releasing new iterations packed with improvements. This is still state-of-the-art technology.