Lithography: The road ahead
02/01/1999
Lithography: The road ahead
David A. Markle
It is generally agreed that optical lithography will run out of steam once 193-nm technology has reached its full potential. For some, 0.15-micron geometries may be the end of the road, and for others, with a much greater threshold for pain, the limit may be closer to 0.10 micron.Any discussion of lithography`s future requires perspective and an understanding of past lessons, particularly those revealing lithography`s inherent weaknesses.
Learning from the past
The Achilles heel of lithography is the mask or reticle. While it has become synonymous with high throughput, it is also a barrier to higher resolution and perfection in overlay and linewidth control. Much of our progress in lithography has been due to advances in reticle technology, but this technology`s limits may have a lot to do with lithography`s future. In lithography`s early days (the contact-printing era), the photo-emulsion mask was the primary yield limiter. The life of this mask was often just 10-20 wafers. Beyond that point, the mask would have accumulated so many defects from repeated contacts with the wafer that it was deemed useless. This short life made it economically unfeasible to perfect each mask, so inspection was cursory and repair was ignored. These "throwaway" masks were copies of a master or submaster and were inexpensive. As circuits became larger, the same number of defects took a larger proportional toll on yield, and photomasks became a barrier to further progress.
The introduction of the scanning projection printer (scanner) profoundly changed lithography and reticle technology. This optical imaging system, which provided a 9-in. separation between the mask and wafer, allowed the mask to be perfected by using better-quality substrates and hard chrome to carry the pattern. Thorough inspection and repair became standard procedures, and the practice of copying the mask from submasters or masters was eventually discontinued.
Another major breakthrough was the invention of the pellicle: a thin membrane stretched across a frame attached to the mask so that any particles are held well out of focus. This rendered the patterned surface of the mask immune to dust and dirt and eliminated the need for regular, periodic cleaning. However, full wafer 1? masks, required by scanners, were a fabrication challenge. The number of picture elements (pixels) required to cover a 125- or 150-mm wafer was massive, even at a relatively low resolution of 2 microns. Often the mask-limited yield was <100 percent because it was difficult to perfect such a large, patterned array.
The original GCA reduction stepper allowed maskmaking to move a step closer to perfection because it increased the minimum geometry on a reticle by a factor of five. At the same time, the patterned area on the wafer projected from the reticle was reduced about fiftyfold. From this point on, the reticle-limited yield was expected to be no less than 100 percent. Of course, this "mask-limited perfect yield" would have been next to impossible without the pellicle.
Back to the future
The technologies expected to replace optical lithography appear to fall into two categories: high throughput requiring a mask and maskless low throughput. The high-throughput technologies include x-ray, projection e-beam (SCALPEL), extreme ultraviolet (EUV), and ion beam. In all these cases, a pellicle is out of the question, although x-ray advocates have proposed a polyimide film between the patterned absorber and the wafer. This film would make it easier to find and remove particle contaminants; however, the x-ray mask would still be susceptible to particles on both sides. Even with 157-nm lithography, there is no hope of using a pellicle. Thus, all high-throughput, follow-on lithographic technologies are defect prone.
One other chilling characteristic of these follow-on technologies is that the mask or reticle is expected to be very expensive - from $50,000-$100,000. Imagine what that means to someone in a low-volume ASIC business requiring a 20-layer mask set.
Maskless technologies
Susceptibility to defects and high cost provides a strong incentive to find a maskless technology with reasonable throughput.
A number of promising possibilities are being examined. For example, multiple e-beams can be channeled through the same column by flooding a micromachined array of shutters, each of which is independently modulated. Multiple e-beams can also be generated from multiple points on a negative-affinity photocathode using independently modulated photodiodes. In both cases, throughput is likely to be eventually limited by the mutual repulsion between electrons, but not until throughput levels measured in wafers/hr are reached. Further improvement in throughput is possible with a 2-D array of microcolumns achieved using micromachining techniques. Similar maskless techniques have been proposed for x-ray and EUV technologies, and perhaps a proposal for multiple ion-beam columns is not far away.
The perceived need and the current level of interest virtually ensure that a maskless technology with reasonable throughput potential will eventually result. Most likely, the first place it will find application is in reducing the cost of $50,000-plus masks. That is only the beginning. Once out of the bottle, this genie could eliminate masks. Constant improvement in the throughput of multiple-beam systems opens the door to maskless wafer lithography, particularly for low-volume applications where mask cost may be prohibitive.
All multiple-beam, maskless systems are not the same. The number of pixels that span a minimum feature varies from 2 to 10, and the number of modulators involved in writing a pixel can vary from 1 to >100. For example, a minimum feature 0.1 micron square could be composed by turning on a square 3 ? 3 array of pixels, each 0.033 micron wide. If the pattern is printed with a single exposure, then the edge of the pattern can be positioned with a resolution equal to the pixel size (i.e., 0.033 micron). With multiple exposures, the pattern edges can be shifted between successive exposures, so that with 8 exposures the edge resolution becomes equal to 1/8 the minimum pixel size (4 nm). Thus, edge placement resolution can be much finer than pixel size.
In some maskless systems, a different set of pixels is involved in each successive exposure. Taking the above example, where 9 pixels are used 8 times, a total of 72 pixels is required to expose a single 0.1-micron minimum feature. Thus, an inoperative pixel (a defect) has only a 1.4 percent effect on determining the exposure of a minimum feature. Therefore, in some maskless systems, "voting lithography" can be used to achieve defect immunity.
Thus, in the long run - much more than five years - lithography could become both maskless and defect immune. However, in the short run, we will be faced with rapidly escalating mask costs because of continuously increasing pattern densities and the added complexities of optical proximity correction and phase shifting. Between these two periods, the transition to nonoptical masks will continue the cost escalation. Due to the loss of the pelllicle, lithographers will have to accustom themselves to lithography that is more susceptible to defects.
David A. Markle is VP and chief technical officer at Ultratech Stepper, 3050 Zanker Rd., San Jose, CA 95134; ph 800/222-1213, fax 408/325-6444.