How does 193 nm light print lines only tens of nanometers wide? Immersion lithography and multiple patterning

Deep ultraviolet light at 193 nm can print lines only tens of nanometers wide because wavelength is not the whole story. Four other things help: how steep an angle of light the lens can take in (its numerical aperture), a layer of water between the lens and the wafer (immersion lithography), splitting a layer that is too dense into two or three passes (multiple patterning), and reshaping the pattern on the photomask ahead of time. Immersion lithography is also called immersion exposure, and since its 193 nm light comes from an argon fluoride (ArF) excimer laser, ArF immersion. Multiple patterning is sometimes called multi-patterning, and its two-pass form double patterning.

The basic steps of lithography are in "What is photolithography?", and reshaping the mask is in "What is OPC?". This article covers the other three.

If the light is wider than the lines, how can it print them?

When light passes through a very narrow gap in the photomask, it spreads out to the sides. That's called diffraction, and the narrower the gap, the steeper the angle it spreads at. The scanner's lens has to collect that spread-out light and put the picture back together on the wafer, and the finest details of the picture are carried by the light that spread at the steepest angles.

So apart from the wavelength, what matters is how steep an angle of light the lens can take in. That's the numerical aperture, or NA: the bigger it is, the steeper the light it collects.

The textbook rule is that the finest feature you can print is roughly "a factor × wavelength ÷ numerical aperture". The factor reflects how good the process is (how the light is shone, how good the resist is). For a single exposure, it can't go below 0.25 in theory, and in production it is always a bit higher. "Finest" here means the width of one line when lines and spaces of equal width are packed side by side.

Why isn't air between the lens and the wafer good enough?

In a conventional "dry" scanner, there's air between the last lens and the wafer. When light goes from glass into air at a slant, past a certain angle it can't get out and is reflected straight back, the way the surface of a pool looks like a mirror when you look up at it at an angle from underwater. So in air the numerical aperture is always below 1, and the best dry scanners reached about 0.93.

Plug that into the rule: 0.25 × 193 ÷ 0.93 is about 52 nm, the theoretical limit for one dry exposure.

Why does a layer of water help?

You can think of the refractive index as how slowly light travels through a material. Air is 1; for 193 nm light, water is about 1.44. Fill the space between the lens and the wafer with water, and the steep light that used to be reflected back can pass into the water and reach the wafer. The numerical aperture can then go above 1, up to about 1.35.

Another way to see it: in water, the wavelength of light shrinks to 1 ÷ 1.44 of what it was, so 193 nm light becomes about 134 nm. Its "effective wavelength" is shorter.

With the same factor, going from an NA of 0.93 to 1.35 makes the finest line about 30% narrower: 0.25 × 193 ÷ 1.35 is about 36 nm. In production, one exposure gets to around 40 nm. Immersion has been in volume production since the mid-2000s.

What does immersion exposure have to take care of, step by step?

The water buys finer lines, and it also brings a few new things to look after.

Step 1: Protect the photoresist first

• What: After the photoresist is coated (see "What is photoresist?"), a very thin protective film, a topcoat, is often added on top, or a resist is used whose surface turns water-repellent by itself.
• Why: When resist sits in water, some of its ingredients leach out into the water, and water can seep into the resist.
• If not: The resist's sensitivity to light changes, the line width drifts, and defects can become more common.

Step 2: Keep the water only under the lens, feeding it in and sucking it out

• What: A ring-shaped hood under the lens keeps feeding ultrapure water into the narrow gap between the lens and the wafer, and sucks it back out at the outer edge.
• Why: During exposure the wafer stage races back and forth under the lens, printing one field at a time; keeping the water in a small patch under the lens stops it from being flung off.
• If not: Water ends up where it shouldn't, leaving droplets and dirt on the wafer and the tool.

Step 3: Keep the water very pure, very steady, and free of bubbles

• What: The ultrapure water has its dissolved gases removed (see "How pure is ultrapure water?"), and its temperature is held very steady.
• Why: Bubbles and particles scatter the light, like dust on a camera lens; and water's refractive index changes with temperature, so if the temperature drifts, the point where the light comes into focus drifts too.
• If not: The shadow of a bubble prints into the resist as a defect; and if the focus is off, the picture blurs.

Step 4: Leave no water behind after the lens passes

• What: The wafer's surface is made very water-repellent, and the water is fully collected wherever the hood has passed.
• Why: The faster the wafer stage moves, the more easily it drags little droplets behind it; the more water-repellent the surface, the better the water keeps up.
• If not: A droplet dries into a watermark, and the pattern there is ruined. How water-repellent the surface is also limits how fast the tool can run.

What if one exposure still isn't fine enough?

Denser than that, one exposure can no longer separate the lines. The answer is to split one layer into several passes, which is multiple patterning. There are two common ways.

Split it across two masks and print twice

Take a row of lines that is too dense and give every other line to a second mask. Each mask now has lines twice as far apart, which one exposure can print. Print the first mask and etch it, then print the second mask and etch it; together, the two sets of lines are half as far apart. This is often called LELE, for litho-etch-litho-etch.

The hard part is alignment. The second pass has to line up with the first, and if it's off even slightly, the spaces come out wide on one side and narrow on the other (see "What is overlay?").

Make sparse lines, then use the film on their sides as lines (self-aligned)

1. Make a row of sparse lines: twice the final spacing, so one exposure and etch can make them.
2. Grow an even film over the whole wafer: like giving every line a coat. However thick the film is, that's how wide the new lines will be, so it has to grow very evenly (deposition is in "What is thin film deposition?").
3. Etch straight down to remove the film on top and at the bottom: only the parts clinging to both sides of each line stay. Etch too little and the lines are still joined; too much and the side films get partly etched away too.
4. Remove the original lines: the side films stay behind, one line becomes two, and the spacing is halved. The new lines follow the original ones, so there's no second alignment to get right, which is why it's called self-aligned (self-aligned double patterning, SADP).
5. Print one more mask to cut the lines: what you get are rows of lines joined end to end, and the unwanted segments have to be cut. Skip it, and places that shouldn't connect stay connected.

Do the whole thing once more and the spacing drops to a quarter.

What's the price?

• More masks: the hardest layer needs two or more masks (why a set is so expensive is in "What is a photomask?").
• More steps: each extra print, etch or deposition sends the wafer through more steps, adds waiting time, and adds one more chance for something to go wrong.
• Tighter alignment: patterns from several passes have to combine into one layer, so the allowed offset is smaller than with one print.
• Less freedom in the design: two shapes that are too close can't go on the same mask, and the self-aligned method works best with rows of parallel lines.

What happened to the finest layers later?

The finest few layers later moved to extreme ultraviolet light with a wavelength of about 13.5 nm, and some of them can be printed in one exposure; see "What is extreme ultraviolet (EUV) light?". Many other layers still use 193 nm immersion: a chip has dozens of layers, and not every one of them is that fine.

In fab24hr

Tap "Clock in", choose "A day in the fab", then on the "Your role today" card tap "Production fab", swipe to the process engineer and tap "Litho" under the name: you're the litho process engineer. On the morning shift, "Overlay Out", lot P-3A08's overlay is out of spec, and you decide whether to rework it or run it as is.

Tap the lithography area on the left side of the cleanroom and press "Take it apart" on its card to get the scanner model. One of its parts, the "Immersion hood", says: "Between the lowest lens and the wafer there's only a very thin layer of ultrapure water." Press "See it work" to watch the mask and the wafer sweep in opposite directions, printing one field at a time. On the home screen, the "Learn" tab's "Scanner" opens the same tool. The game doesn't say what numerical aperture the scanner has, and it doesn't show multiple patterning. The tool model is a generic illustration; the companies, people, lot numbers and figures in the game are fictional, and immersion and multiple patterning are described from public, general sources. For other jargon, see Chip fab jargon in plain English.