Color-coded image for wafer inspection. The colors correspond to relative surface heights above or below the focal plane. (Carl Zeiss, Thornwood, NY)
Can you give us a brief history of microlithography?
If you remember, the integrated circuit was invented in 1958. Immediately following that there was some kind of lithography, whether you call it micro or not. People were doing 200-micron type of geometry for the first ICs using a simple method called contact printing, where a mask was put in contact with some kind of photosensitive material called a photoresist. The photoresist was coated on a wafer, a 5 to 8 inch single crystal silicon, 1 to 2 mm thick, where the integrated circuit is printed. They duplicated the mask to create a photoresist image (Figure 1).
Figure 1. Proximity printing to copy the mask image onto the photoresist.
That went on for quite a while. People kept improving contact printing by reducing defects and enhancing resolution. Actually, they changed to proximity printing because contact printing, when the wafer touches the mask-well, it is the photoresist on the wafer that first touches the mask-creates a lot of damage and defects. So the mask is separated from the wafer by 20 to 50 microns, depending on the geometry. People also scrambled the illumination, making it less coherent, to improve the resolution. Progressively, they were able to improve the linewidth from 200 µm to about 2 µm until about 1974. Any further shrinkage using this technique got into mask damage and defects unless the exposing wavelength was reduced from the vicinity of 400 nm to the 250 nm deep UV region and even to soft x rays in the 1 nm region. The wavelength reduction techniques did not impact manufacturing because of other new breakthroughs.
Around 1974, a great breakthrough occurred, namely, the so-called 1X all-reflective projection printing, using a reflective lens system between the mask and the wafer (Figure 2). This allowed the mask image to be focused on the photoresist, removing the requirement of proximity placement of the wafer. Because of the great popularity of proximity printing, researchers had to make the projection system do exactly what proximity printing could do.
Figure 2. 1X all-reflective projection system.
In addition to the wavelength independent nature of this projection system, they managed to image the entire mask onto the wafer by illuminating a slit at a time on the mask and then scanning, thereby breaking the small-field problem of other existing projection systems. They also created a mirror image so that the printed image was exactly the same as proximity printed images, thus making the mask usable for both types of printing. They had to add a three-face mirror block in the system to do that. The 1X all-reflective, full wafer system was a great breakthrough and people left proximity printing to get into projection printing. Of course, the system has gone through several stages of refinement. In proximity printing, the wavelength was reduced to the deep UV wavelength, below 250 nm. So, they did the same for this 1X all-reflective projection printing. Actually, the system has been used for extreme UV down to tens of nanometers. The wavelength reduction technique did not catch on in manufacturing also because of other new breakthroughs.
Practically, this kind of thing was phased out about 1985. People were trying to go from 2-µm linewidth to 1.5-µm linewidth in production, and it wasn't very successful. Another improvement came along in 1978. That was the reduction step-and-repeat system (Figure 3). People were no longer worried about compatibility with proximity printing, the full-wafer mask, and the requirement of a mirror image. They put a reduction refractive lens system between the mask and the wafer. The wafer is stepped repeatedly under the lens for exposure, therefore, the name step-and-repeat. The advantages are: (1) Large-numerical-aperture refractive lenses are easier to make. (2) Reduction facilitates better mask making. (3) Stepping makes better alignment between different masking levels.
Figure 3. Step-and-repeat projection system. The wafer is stepped to the size of the combination of chips within the lens field in the x and y direction. See Figure 5.
However, many masks have to be used for one wafer, if it contains more than just one repeating design. In addition, because of a narrow bandwidth requirement of refractive lens systems, multiple reflections in the resist started to become a problem.
So, the major industrial workhorse for microlithography is the reduction step-and-repeat system for 1.5-, 1.25-, 1.0-, 0.7-, 0.5-, 0.35-, and 0.25-µm geometries. A side-stepping breakthrough came along in 1980, when people combined step-and-repeat with 1X optics. They also combined reflective and refractive elements into a catadioptric imaging lens to come up with the 1X step-and-repeat (Figure 4). The system turned out to be, because of its simplicity, very economical. For a while, people used it for all the levels of the IC. Later on, when the geometry became minute, you'd use the reduction stepper for critical linewidths. The 1X stepper still survives, but its mission has changed from doing all levels to now just doing the noncritical levels.
Figure 4. Catadioptric projection lens used for step-and-repeat. This low-aberration system allows for a much broader band of wavelengths, is simple in design with very few elements, and is symmetrical.
Figure 5. Chips and lens field. The shaded area is not exposed.
Okay, what else?
Well, there were a lot of significant breakthroughs in 1989 when people realized that if you kept on doing step-and-repeat and kept on pushing the linewidths, pretty soon you wouldn't have enough field of view to cover the entire mask to satisfy the growing electronic requirement of the chip, because the chip gets bigger and bigger each generation even though the geometry gets shrunken down. Usually people like to put at least two chips, more if possible, into the field of view (Figure 5). If the chip size grows, only one chip, even less, will fit. That's not desirable. So people had to come up with something that had a larger field size. It wasn't easy. The breakthrough came along when researchers decided not to use the whole field of a step-and-repeat camera. Instead, they'd use only one long slot in the field (Figure 6) by scanning the slot over the chip. Actually, in 1989, what they proposed was a curved slit instead of a straight slot, due to the influence of 1X reflective optics. A straight slot, with many advantages associated with it, quickly replaced the slit. In that direction, there's no limitation to the field size, provided you build in the travel and the mask that allows for that kind of thing. That's called a step-and-scan system.
Figure 6. Step-and-scan principle. The circular wafer is covered by 18 lens fields in this example. A typical scan, step, scan occurs as follows. The light is turned on when the illuminating slot is scanned across line AB, turned off at CD. Then the slot is stepped down to the right of DE where it is scanned left with the light coming on at DE and turned off at FG. The arrows show the step and scan increments. The process is completed when the whole chip has been scanned and illuminated.
It's not the main workhorse yet, but sooner or later as chips grow and the geometries shrink, it will be. Some people are using it for 0.25 µm and it looks like you can push it to 0.18 µm or so and eventually to 0.13 µm. In 1989, the step-and-scan system didn't come out to compete with step-and-repeat on the other geometries. The step-and-scan system ranges from 1/2 µm to about 0.13 µm, even though it can do the larger geometries. But it would be economically unwise to use it.
So we're talking about improvements in the optical imaging tools. How about other things?
Well, as we know, people were initially using visible wavelengths to do the lithography. Then they found that that wasn't good enough, so they moved to ultraviolet light. The ultraviolet light at that time was still pretty long in wavelength, from 350 nm to 450 nm. Then people realized there was another part of the spectrum that could be used, and that was the deep UV, between 200 and 250 nm.
By going to shorter wavelengths, of course, you can improve the resolution in proximity and projection printing. That was in 1975. Then, it was mostly used for proximity printing because the imaging system was very simple. You didn't need a lens; all you needed was a deep UV illuminator. Even though 1/2 µm images were demonstrated and magnetic bubble circuits were fabricated this way, in order for deep UV lithography to become mainstream, you needed several things. First, you needed a very intense light source. Second, you needed a very robust deep UV photoresist, and one that's fast enough.
One breakthrough for the light source came in 1982, when people started using the excimer laser as a deep UV source, first in proximity printing, because of the freedom from imaging optics. In 1986, we started to see people putting in reduction step-and-repeat machines, marrying the excimer laser with a stepper. Right now, if you want to go to 1/4 µm production on a stepper, most likely you'll have to order yourself a deep UV reduction stepper.
Figure 7. Cross-sectional views of the positive and negative resist processes.
Now, let's talk about photoresists (Figure 7). A photoresist is a light sensitive material. When it is illuminated, it changes its dissolution characteristics. If it's a negative resist, when it's illuminated, it becomes insoluble and the unexposed part is soluble. So you develop an image and you remove the area that is unexposed and you retain the area that's exposed. The positive resist is the reverse. You remove the area that is exposed and you retain the area that is unexposed.
Initially, in the 1960s, there was a major resist that was called KTFR, which was a negative tone resist. It was a very fast resist, but its development processing characteristics were pretty lousy. It was gradually replaced, for the high resolution work, in the early 1970s by the so-called Novolak resist. The Novolak resist is still the main resist system up to 300 nm or so of light. So you still see it being consumed in large quantities for manufacturing all over the world.
After exposing the photoresist, the exposed or unexposed area is removed by a developing solution so that the wafer substrate is open in selected areas. You then etch out the uncovered parts of the wafer underneath the resist layer. That's how you transfer the image from a mask to a circuit pattern on the wafer (Figure 8).
Figure 8. The use of a photoresist as an etch mask. The device layer can be an insulator, semiconductor, or conductor. The etching is done by using an anisotropic reactive ion, which removes the exposed part of the device layer. What remains of the etched device layer can be used as gates, isolation fields, or as conducting pathways to carry the signal.
Now, back to the Novolak resist. It's desirable in many respects; it has good imaging and processing properties and reasonable sensitivity, except for deep UV. In that spectral region, it's not sensitive enough, it absorbs too much, and all those undesirable things. The breakthrough for the deep UV resist came in 1982. People realized that they could use a chemically amplified process in which a molecule of photogenerated acid reacts with the acid-labile (i.e., acid producing) group on the polymer to change the dissolution characteristics of the polymer and to generate a new molecule of acid to react with the unreacted acid-labile group. This process can go on for hundreds of times to greatly amplify the effect of each photogenerated acid. The photosensitivity is enhanced in terms of orders of magnitude. With the excimer laser and a chemically amplified resist, we now have a good, deep UV system that people could use in manufacturing.
What about x rays?
People have been intrigued with using x rays because the wavelength is so short. If we were to do proximity printing with such a short wavelength, it's almost like a ray, no wave property, and we should not be limited by diffraction. You could replicate anything you wanted, up to, down to any foreseeable geometry.
People have been toying with the so-called hard x ray (~0.1 nm), the x ray that you use for medical purposes and so forth, because sources and equipment are readily available. But a problem with hard x rays is that you cannot find a good absorber for it. You can't find anything to stop the x ray very effectively, which means you'd have a hard time making a mask of good contrast.
X rays weren't very successful until 1972 when people started using vacuum environments and longer wavelengths, since these longer wavelength x rays (around 1 nm) didn't penetrate through the atmosphere. By the willingness to go to vacuum, one can use these soft x rays, thus easing the mask requirement. You can now find materials that give you adequate contrast for the purpose of lithography.
Eventually, physicists realized that to get a really good x-ray source they would have to go to synchrotron radiation. The synchrotron storage ring type of x-ray source was first introduced in 1976. X-ray lithography has evolved mainly around these synchrotron storage x-ray sources. While the source and illuminator are very sophisticated and very expensive, the imaging system is simple. It's just a mask and a wafer, no more complex than that show in Figure 1, albeit in vacuum.
Then, there's electron-beam?
Right. People in the 1960s used to just take an SEM, scanning electron microscope, and convert it so that instead of scanning as a microscope, it would scan the whole chip area and turn the beam on and off while following the rasters (i.e., each line scan as in a TV).
Soon, they realized they could do better than just converting SEMs. They started to go in two major directions in 1974. One is the so-called vector scan system where you would save writing time by directing the beam to only the area you like. You don't have to scan the whole chip and turn the beam on and off. The other direction people took was to raster scan a small area not much larger than that of an SEM, so that the demand for the e-beam optics was not very high. They put a laser interferometric table underneath to position the wafer to a very high precision. The e-beam would write a small area on the wafer, which was much smaller than the chip. Then the table would move it to the next field very accurately so that the beam could repeat this process again to fill up the entire chip.
What kind of dimensions can you get with e-beam lithography?
I'm giving e-beam lithography about 0.10 µm, with a lot of qualifications. You can get an e-beam system that will generate nanometer size spots for you, but it's difficult to come up with such an e-beam system to write at a circuit level with linewidth control, placement control, throughput, and everything else. Currently, it's not doable.
I know that optical lithography is widespread in manufacturing right now, but what about the other technologies?
For e-beam lithography, manufacturers are using them in large quantities for mask making. For directly making circuits, it's just not economically viable. If there's an economically justifiable need to do 0.10-µm lithography, e-beam can write the supercritical levels. In an IC where you have 20 to 30 masking levels, of which about 10 levels are critical and 2 to 3 are supercritical, e-beam will write those supercritical ones.
X ray is not widely used in the industry, is it?
No, it is not. X ray is in reality just a proximity printing system. You have a separation between the wafer and the mask. That separation can be reduced to zero if one is not worried about defects and mask damage. Actually it's not zero, because you have the photoresist layer in there. So, let's say you make the mask-to-wafer distance 1 µm. Then you can make very small features, down to 0.05 µm, but that's not usable for manufacturing. For manufacturing, if one wants to use a mask-to-wafer gap of about 10 µm, then you cannot do better than 0.15 µm geometry, because despite the rosy speculation during conceptualization of x-ray proximity printing, there is actually a diffraction limitation to x rays just as there is in optical.
The major problem with x rays not being used is not because nobody is making 0.15 µm devices. People would be happy to use x rays for 0.25-µm devices, where there is no wafer-to-mask-to-wafer gap problem. The problem is that people can't make a mask that is economical enough, that is controlled in defect and linewidth variation, and in overlay placements. The e-beam machine is not there. People have to spend a lot of time and a lot of money to develop an e-beam mask making machine to support x ray. There is also the economic consideration in building a factory filled with storage rings and beam lines, along with the extra electric power and radiation protections you would need. One needs a tremendous volume of wafer exposures to justify such a venture.
What about the future?
Right now the limits are for optical, 0.13 µm; x ray, 0.15 µm; e-beam, 0.10 µm. There are other technologies we haven't talked about. One is ion-beam lithography; the other is extreme UV lithography. But they are not the right candidates for smaller linewidths. E-beam lithography started in the late '60s, so they have had about 30 to 40 years of development. X rays started in 1972, so it also had about a quarter century of development work. We have done very respectable work in all those areas. We've introduced many improvements in these technologies and know them very well. It's not foreseeable that any of them can bring us down to less than 0.10 µm, except possibly in very expensive demonstrations.
In terms of the commercial world, it will stop at about 0.13 to 0.15 µm. People will innovate elsewhere, perhaps go to 3D device structures, more effective use of chip areas, and more functions without making the geometry smaller because it's not easy to go below 0.10 µm.
I think microlithography will stay around 0.13 to 0.15 µm, dictated by optical lithography, with very specific supercritical levels using e-beam for 0.10 µm. That limit is going to stand for a long, long time unless there is a significant breakthrough that nobody knows about and, at this moment, nobody anticipates.
Burn J. Lin earned his PhD in electrical engineering from Ohio State University. He has been a research staff manager at IBM T. J. Watson Research Center; a department manager at IBM GTD, Burlington; staff member at IBM ASTC East Fishkill; assignee at SEMATECH; and currently is president of Linnovation, Inc. He is a senior member of IEEE and a member of SPIE. Accomplishments include: first deep-UV lithography article, portable conformable mask, first 3D partial-coherent projection imaging program, first article on E-D trees, first article on k1 reduction techniques, first article on attenuated phase shifting mask, first article on signamization. He has authored two book chapters, 50 articles, and holds 20 U.S. patents.
He was interviewed by Frederick Su.