Production optical lithography equipment accomplishes the seemingly impossible

Imagine a camera that takes photos containing half a trillion pixels. Such a camera exists, in the form of a semiconductor optical lithography tool called the Twinscan XT:1950i, which creates single-exposure images 22 x 33 mm in size with a 38 nm resolution in photoresist on a silicon wafer, for a total pixel count of 503 gigapixels. Just announced by ASML (Veldhoven, The Netherlands), the XT:1950i operates at the argon fluoride (ArF) excimer-laser wavelength of 193 nm and is built around an immersion catadioptric 4X reduction lens with a numerical aperture (NA) of 1.35. The tool is expected to begin shipping in the first quarter of 2009.

The previous leading-edge lithographic tool made by ASML, the XT:1900i, has a resolution of 40 nm. While the increase in resolution from 40 to 38 nm sounds small, it means that the same computer chip manufactured with the XT:1950i instead of the XT1900i is 10% smaller in area, making 10% more wafer area available.

Other improvements introduced in the XT:1950i include a 15% increase in wafer throughput, to 148 wafers per hour (the wafers are 300 mm in diameter). Combined with the 10% increase in wafer area available, this results in a 25% increase in performance, says ASML. In addition, the single-machine overlay (meaning the precision to which a wafer can be realigned to the same projected pattern in the lens field for the same machine) is 3.5 nm, according to ASML.

One reason I am personally interested in this announcement is that, long before becoming an editor at Laser Focus World, I was once an optical engineer developing this very sort of lithography tool, although I worked on far more primitive versions. The first version I worked on contained a projection lens with an NA of a mere 0.28; the lens operated at the mercury g-line, which is at 436 nm, while the tool handled wafers of about 100 mm in size at something like 60 wafers per hour. The last version I worked on had a lens with an NA of about 0.5 and used the krypton fluoride excimer wavelength of 248 nm; it handled wafers 150 mm in size.

Fabricating a lithographic lens that has an NA almost five times as large as the original system I worked on, and is designed to operate at less than half the wavelength, requires a stunning increase in precision of optical manufacturing. Handling wafers three times the diameter (nine times the area) at 2.4 times the speed, for a total increase in wafer-area throughput of 22 times, requires a stunning increase in everything from exposure speed to wafer-handler adeptness. Reducing single-machine overlay to 3.5 nm (I can't remember the overlay of the machines I worked on, but it was at least ten times larger) requires a stunning increase in precision of the optical alignment system and the laser interferometer that measures the stage position.

Back when I was an optical engineer, I never would have imagined that Moore's Law (which says that the number of transistors that can be packed onto a computer chip doubles about every two years) would extend this long. I still can't see how it can extend beyond a few more years from now--but it very well could.