EUV And the State of Lithography

High-End Optical Lithography, what’s it all about

We live in interesting times, which means we are covered with news, some of which are tough to grasp. One of the topics is actually Lithography and particularly, EUV lithography.

With this tech-talk, I would like to cover this topic from a perspective I have not seen elsewhere, with the hope that it will illuminate the state of high-end lithography and touch upon the role of ASML.

So what is ASML doing exactly?

When ASML, TRUMPF, and Zeiss opened up the Extreme Ultraviolet Wavelength (EUV) spectrum, everyone was probably struck with awe, and if not, they should be. I know I was. To build at this wavelength is not only a matter of finding the light source. No doubt, the TRUMPF Laser Amplifier is an amazing product. That said, so is the optics.

Reaching the full performance potential of EUV isn’t just a matter of signing a contract and installing a new laser. It also requires ensuring that any photon reflected by the mask — when it finally reaches the wafer — must have traveled a path identical in length to every other path it could have taken through the optical system. Yes, that’s a bit of a mind-bender, but it’s essential: a good optical system must let each photon express its quantum nature, and in EUV lithography, this must happen with sub-nanometer precision.

It’s one thing to dip your finger in a pool of calm water and experience the spherical waves propagating away in all directions, it’s something completely different to make the clock go backwards and see those waves come in from the horizon to make a droplet of water jump at a single center.

To give Zeiss the kudos they deserve, imagine the Earth perfectly covered with a vast, undisturbed ocean stretching from the South Pole to the North Pole. You toss a small stone into the exact center of the South Pole. Then, quietly, move over to the North Pole and wait. Over time, the ripples travel across the planet, converging at a single point, releasing all their energy in one perfect, glistening droplet. In the big picture, they shape and polish the planet until it does exactly that.

So, Trumpf does this amazing laser, and Zeiss SMT do aspherical mirrors that seem impossible to do. Where does that leave ASML? I think that to get a bit more of a handle on that, we need to look back a little. And I know, this is a bit focused on optics. It’s so much more.

A Lithography primer

Before we can fully appreciate where lithography is today, we need to have a look at the basics and where it used to be.

Lithography systems are two overlapping 4-F projection systems. One that projects the source on the system pupil, and one that projects the mask onto something sensitive to light.

When we add structure to the mask, we will instead see something like this,

And had this been an I-line or KrF system where you could stick in a piece of paper without ruining everything, you might have seen this in the pupil,

Before we get too much ahead of ourselves, let’s just define k₁.

where H.P represents half of the pitch of a periodic structure, a binary grating, but in a loose sense, it represents our resolution limit. Each color represents one diffraction mode.

Starting from the left, if the pattern is not pushing resolution limit, we will have what we can call easy lithography. In the middle, we are approaching not so easy lithography. The center of the diffraction orders is just at the pupil’s edge, that’s nice, but we have lost some light, so our latent image is losing acuity. This still prints without too much pain. Moving further to the right, we are now in the painful region. This probably still prints. Maybe not so well with this particular illumination. Maybe we would have preferred an annulus. We would have regained back some of the slopes in the latent image. We will struggle with controlling line widths. Maybe we would like to add non-resolved assist lines to get back some focus tolerance. Maybe some additional serifs to make corners sharper, and so on, so on.

What is the point with all this. Well, the point is that it’s becoming hard to print. There is too much stuff close to the resolution limit. Or another way to put it, we are trying to do too much at once. We are still above k₁ > 1/4 but it is getting tough.

End of Primer and let’s have a look how “easy” lithography used to be.

A little history

The table below is an attempt to summarize how the lithography industry went from being pure optics to something else. By this I don’t mean that optics is not super important, it absolutely is, but for almost two decades the industry has been in the unprintable region, relying on anything it could get its hands on in order to print beyond the resolution limit.

The region of litho pain goes down to k₁ = 1/4. Strictly speaking, we cannot even print there, but it is possible to come surprisingly close. When we go below 1/4, we have to get creative. Scrutinize the axioms of optics and see which ones can be broken and we still can get away with it.

So where are we?

There was a time when CPU speed was all about the gate, and the gate was all about optics. That ship sailed long ago. Since then, we’ve kept shrinking nodes using optical lithography — but only by getting creative. Tricks like double (or triple) patterning, using interference with alternating phase-shift masks, vortex modes, assist lines, and more became part of the toolkit. Some of these ideas go back two decades, but they’ve become everyday essentials over time. Cell layouts had to adapt, instead of simply copying their intended shapes, they now twist themselves into pretzels just to look right after passing through the projection system.

There’s no chance I can paint the full picture — I honestly don’t have it. But it’s clear where ASML fits, they take care of everything else. Machines that move hundreds of wafers an hour, without rocking the planet (back to water-covered earth analogy), manage temperature, vacuum, source optimization, and the strange contortions of projected patterns that just want to look good in a curved mirror.

It actually reminds me a little of what I do through Senslogic, pushing the edge of what a crossover expert can be, arranging the small to paint a big picture. Like Gru arranging his Minions to spell out a giant “IT WORKS” — chaotic up close, perfect from a distance.

Today, lithography no longer decides the absolute width of a transistor gate — that baton was finally passed with the FinFET. The new Gate-All-Around structures are in the hands of process engineers and atomic layer deposition experts. Lithography’s focus has shifted: it’s now about how many billions of transistors we can squeeze onto a die, and how cleverly we can stack and connect them.

Take my AMD Ryzen 7 8700G. It packs 25.4 billion transistors across 178 mm². If you do the math, each transistor has about 84-by-84 nanometer of wafer real estate to live in. Imagine doing all that using optics with about 13-15 nanometer resolution.

It’s almost a leap of faith — fitting a source, gate, drain, local connections, and maybe a via or two — all into that tiny square and trying to print it, a million times. But as we’ve seen, we’ve long since crossed the line where pure optical resolution was the main constraint. Today, it’s about what works, how many mask layers you can afford, and how creatively you can bend the rules without breaking them. Making sure all that works is where ASML comes in, I believe.