Focused ion beam mask repair
John C. Morgan, Micrion Corp., Peabody, Massachusetts
In the late 1980s, the move to reduction lithography - instead of 1? lithography - gave mask equipment and process developers a vacation . Now that the linewidths on advanced photomasks are reaching and surpassing those used with 1? lithography, the vacation is definitely over. In addition, maskmakers must contend with phase shift techniques, optical proximity corrections (OPC) and shorter exposure wavelengths. While mask design, writing, and inspection have all become more complex, no task has increased in complexity more than mask defect repair.
When mask linewidths were on the order of 4-5 ?m, laser repair tools were the workhorses of the industry. They were fast and fairly easy to use, but limited in resolution and spot size, and had difficulty depositing material to repair clear defects. Removal of chrome defects is also becoming difficult because of quartz damage. Laser tools are not damaging the quartz any more than they used to do, but the small amount of damage they produce will now print as a defect on the wafer.
Focused ion beam (FIB) tools were typically used to repair clear defects by depositing an opaque carbon patch where the chrome was missing. FIB tools had limited ability to repair opaque defects. They created unacceptable quartz damage and implanted gallium into the underlying quartz, reducing transmission at the repair site. Some newer FIB mask repair tools use chrome gas assisted etching (GAE) to make high-quality repairs of opaque defects. In these systems, an etchant gas enhances the chrome etch rate, minimizing quartz damage and gallium implantation. Thus, the maskmaker can repair clear defects with carbon deposition and opaque defects with chrome GAE.
As lithographic wavelengths shrink to 248 nm and 193 nm, however, defect repairs become tremendously more difficult. Since implanted gallium absorbs light more strongly at these shorter wavelengths, quartz damage must be reduced, and edge accuracy improved. Whatever lithographic techniques lie beyond 193 nm, they will present a whole new series of challenges. This article explains the capabilities of today`s systems and discusses future requirements for focused ion beam repair.
Present mask technology
Chrome masks. The standard 5- or 6-in. chrome on quartz photomask is still the most commonly used. There are essentially two defect types, opaque and clear. Clear defects are areas where the chrome is missing and must be replaced. Opaque defects are supposed to be clear, but have chrome on them that must be removed. Critical repair issues for chrome photomasks include charge neutralization, edge placement accuracy, quartz damage, and transmission after repair.
Most focused ion beam systems image the substrate with a positively charged gallium ion beam, accelerated at approximately 20-50 kV and scanned over the substrate much like a scanning electron microscope (SEM). If the substrate is not conductive, it will charge up. Just a few thousand volts of substrate charging can deflect the ion beam from its intended target. Quartz is one of the worst substrates because it is a very poor conductor. Depositing a conductive material such as chrome on the quartz in random patterns makes a series of capacitors with varying charges. These capacitors make it very difficult to place an ion beam on the substrate to within 75 nm of the desired location, yet this is essentially the task required of a FIB mask repair system.
The charges that build up on the mask must be neutralized to obtain placement accuracy required for mask repair. One company uses an electron floodgun in conjuction with a neutralizing shield to reduce charge build-up. Conductive substrates such as x-ray masks can achieve much better accuracy.
When ions strike the substrate, it emits secondary electrons and secondary ions, which can be used to generate an image of the substrate. Unfortunately, the incoming gallium ions also penetrate the chrome and quartz. The gallium ion dose required to produce a good secondary ion image can implant enough gallium into the substrate to reduce the transmission at i-line to <70%. This problem becomes even more serious at shorter wavelengths. However, imaging with secondary electrons can reduce the implantation dose by more than 100?, since the beam and substrate interaction generates several orders of magnitude more electrons than ions. The reduced imaging dose helps keep gallium implantation to an acceptable level.
Quartz damage requirements have become tighter with higher numerical aperture steppers. Even slight damage to the quartz will print as a defect. Opaque defects must be removed without damaging the quartz under the defect area.
Phase shift masks. Chipmakers have developed strong cost incentives to try to extend stepper performance as steppers reach their resolution limits. One approach employs phase-shifting of the light in the exposure tool to cause destructive interference at the wafer plane, allowing the tools to print smaller features. Many types of phase shift masks have been proposed. The most popular type today seems to be the embedded shifter, or halftone, mask.
The most popular type of embedded shifter mask seems to be the MoSiON mask. This mask uses a thin layer of MoSiON as the absorbing material instead of chrome. MoSiON transmits a small percentage of the light and also shifts the phase of transmitted light by 180?. This phase-shift technique is not as strong as some others, but is the simplest to design, write, inspect, and repair.
The stronger alternating aperture, or Levenson, mask is seeing some limited use. Alternating aperture masks typically change the phase through the substrate. For example, in a chrome on quartz mask, alternate quartz spaces are etched to produce a 180? phase shift. This type of mask provides a good enhancement to the resolution, but is difficult to manufacture.
Repairing chrome masks
A carbon patch deposited over the quartz repairs clear defects in chrome masks. The carbon patch is opaque at the lithographic wavelength, is approximately the same thickness as the chrome, and stands up to commercial mask cleaning processes. DUV and i-line masks use these repairs today. This technique should work at 193 nm with some minor improvements .
Opaque defect repairs are more difficult for FIB systems. A good repair must completely remove the excess chrome with minimal damage to the underlying quartz. The repair must also be placed within 75 nm of any critical edges and have good transmission at the lithographic wavelength. Avoiding quartz damage is the greatest challenge. Milling a substrate with a focused ion beam sputters material away at a certain rate, which for the sake of this description we will call one, determined by factors such as the atomic masses of the ions and the substrate, and the accelerating voltage and angle of incidence of the ion beam. The sputtering rate of the substrate increases as the angle of incidence of the ion beam changes from perpendicular. As the angle gets very acute, the sputter rate can increase approximately sevenfold, creating serious problems at the chrome/quartz interface.
A focused ion beam sputters the chrome in the center of the defect at a sputter rate of one. The sputter rate increases tremendously on the perimeter of the defect because the angle of incidence of the ion beam is very acute. If nothing were done to remedy this problem, the resulting repair would look like Fig. 1. The chrome would be successfully removed in the center of the repair, but there would be an area of considerable quartz damage around the perimeter. This quartz damage, dubbed "the riverbed effect" because of its resemblance (in SEMs) to a dried-up riverbed, is sufficient to print on the wafer.
Figure 1. a) Top view of opaque repair - the riverbed is outlined in the perimeter of the repair; b) side view of an opaque defect repair representing the height profile along the dotted line in a).
Highly reactive gases are often used in conjunction with focused ion beams to deposit and selectively etch away materials. Examples from circuit modification include using tungsten hexacarbonyl gas to deposit tungsten lines on ICs or using halogens to enhance the etch rates of aluminum or silicon dioxide . Applying this technique to the selective removal of chrome from a quartz substrate is not nearly as easy, since chrome is not as reactive as the materials used to construct ICs. However, a recently developed gas process increases the etch rate of chrome approximately 2.5? and inhibits the etch rate of quartz by approximately 60%, increasing the selectivity of chrome removal by a factor of four. The operator can use less dose, thus reducing the riverbeds to acceptable levels. This process has other salubrious effects on the repair quality as well. Since less dose is required to remove the chrome, less gallium is implanted in the underlying quartz. The transmission of these repairs at i-line with no further processing is greater than 95%. Figure 2 shows the aerial image microscope simulation (AIMS) of print results of repairs.
Repairing embedded shifter masks
The repair issues for embedded shifter masks are still essentially the same as for chrome masks: there are still missing shifter defects and extra shifter defects. Missing material must be replaced and extra material must be removed without damaging the underlying quartz.
Depositing a thin layer of carbon to match the transmission of the MoSiON can repair missing shifter defects. This thin carbon deposition does not match the phase shift of the MoSiON, but in AIMS studies this did not seem to affect print quality adversely. In fact, it seems to make little difference whether the patch matches the transmission or is totally opaque.
Extra shifter defects can be repaired in a number of ways. One, called a framed repair, mills away the center of the defect and leaves a thin frame around the perimeter. This technique minimizes riverbedding due to milling near the edge. As edge effects allow rapid removal, a much lower dose than required for the rest of the repair then mills away the leftover frame. This technique implants unacceptable levels of gallium in the quartz in the repair area. A repair system equipped with XeF2 gas can remove the gallium by scanning the repair area lightly with the focused ion beam while flowing the gas. The ion beam and the gas etch away the thin layer of implanted quartz, restoring the transmission.
Work is continuing on true gas-assisted etching techniques for MoSiON masks. MoSiON has a high percentage of silicon and oxygen, so increasing the etch rate of MoSiON over that of the underlying quartz is challenging. Present repairs have a phase error <30? at i-line.
Repairing alternating aperture masks
Experiments have attempted to repair defect areas with too much shifter material (bumps) and with missing material (divots). Researchers repaired the divots, areas with 180? of phase error, by milling down an additional 180? and cleaning up the implanted gallium to restore transmission. The print results were very disappointing: this deep trench appears to create some sort of waveguide and greatly reduces the transmission in the repair area.
Figure 2. AIMS intensity profile at 365-nm illumination of a chrome bridge repaired using GAE; a) is reference pattern and b) is pattern with repaired bridge. Focus 6 is the in-focus condition. Foci 1 and 11 are +25 ?m and -25 ?m defocused relative to the mask surface. For a 5? stepper, this is equivalent to a ?1-?m focus range at the wafer plane. At focus 6, ignoring phase effects, the repair intensity is 98% of the reference intensity.
Figure 3. a) Phase bump before repair, and b) frame remaining after milling flooded area.
The framing technique mentioned previously can repair phase bumps. Imaging is very difficult on these masks as defective and nondefective areas are both quartz. Topographical contrast around the perimeter of the defect allows imaging of phase bumps, however. The perimeter produces more secondary particles and thus appears bright in a secondary electron FIB image. Once the outline of the defect is defined, a frame can be created. A predetermined dose mills away the inside of the defect and a smaller dose mills away the frame (Fig. 3). This technique requires phase bumps that are nearly full height and with nearly vertical sidewalls. Figure 4 shows an example of a repair using this technique.
Repairing OPC masks
OPC is used to improve print results by distorting or modifying mask features. Usually OPC adds serifs to the ends of line and in critical areas where adjacent lines come close to each other. These techniques do not create any major changes in the way a mask is repaired: OPC masks are just chrome on quartz masks with more small chrome features. The ability to reconstruct damaged features does become more important, though.
A line end with some serifs damaged must be reconstructed to look the way it was designed; for instance, by copying another line with the same features to the defect site. A bit-map image of the defective area, subtracted from the bit map of the good area, leaves a template describing which areas need to be etched away and which areas need carbon deposition. If this template is carefully aligned to the defective site, the features can be reconstructed to appear as they should have. Geometry reconstruction software can be used on any chrome mask, but seems to be especially useful when trying to recreate serifs and other such intricate features
Figure 4. Opaque defect a) before and b) after repair.
The next two generations of lithography seem well defined. Many people are presently using 248-nm lithography for production now. The following generation seems likely to be optical lithography using 193-nm illumination. The 193-nm generation will probably introduce 9-in. or 230-mm masks. After that, what will be used is anybody`s guess. The three most likely competitors seem to be x-ray proximity lithography, EUV and SCALPEL. These require very different mask types with different repair needs.
248 nm. A number of mask makers are producing DUV masks right now. The difficulties encountered in repairing these masks are basically the same ones that have already been discussed - edge accuracy, transmission after repair, and quartz damage. Most maskmakers want 50-nm repair accuracy or better on DUV masks, which exceeds the edge accuracy specifications of today`s equipment. Reaching this specification should be possible with careful calibration .
As the lithographic wavelength gets shorter, implanted gallium absorbs light more strongly. Repaired areas that have adequate transmission at i-line might not be acceptable at 248 nm. One solution uses in-situ etching to remove the gallium-implanted quartz at the repair site. A gas such as XeF2 can etch away approximately 30 nm of implanted quartz. In-situ etching typically restores the transmission to acceptable levels, but creates an unwanted phase shift in this area. To avoid the phase shift, the entire mask can be post-processed in an etchant solution, such as hydrofluoric acid. The acid etches off a small amount of quartz, including the implanted quartz, over the entire area of the mask.
193 nm. As the lithographic wavelength goes down to 193 nm, the specifications for edge accuracy, transmission, and quartz damage tighten even more. Preliminary work shows that carbon deposition repair will probably be acceptable at 193 nm. However, chrome GAE with a gallium ion may not work at this wavelength. An alternative ion source, such as a noble gas, may be needed to achieve the required transmission.
This generation of lithography will probably introduce the 9-in. mask as well. Since all FIB processing takes place inside a vacuum chamber, new systems will have to be designed to accommodate larger masks. Developing a new system with a 9-in. capacity and a gas field ion source would either require outside funding or a very large potential market.
Achieving the repair accuracy needed will probably require a conductive mask, such as a thin conductive layer that can be easily removed after repair. An aqueous coating of anti-static solution has demonstrated this capability.
X ray. FIB repair has been quite successful for x-ray masks. The edge accuracy specifications for the 0.13- or 0.10-?m generations will require some changes and development work. In addition, the recent switch to tantalum as the absorber will require further development of a tantalum deposition process.
X-ray masks have a number of advantages over regular photomasks. The most important is that the x-ray mask substrate is conductive. Charging is not a problem, and imaging and edge placement are much easier. Current x-ray mask repair tools achieve edge placement accuracy of ?25 nm (compared to ?75 nm for the photomask repair tool).
Second, the x-rays pass right through the substrate; implanted gallium at the repair site does not affect the print results. Likewise, a small amount of damage to the substrate does not shift the phase or the focus of the x-rays. Thus, repair of x-ray masks is easier than optical photomask repairs. X-ray masks are not reduction masks, however, and thus require small ion beams and very accurate placement. Future placement requirements will probably be around ?10 nm.
EUV. The final design of EUV masks is not well defined at this point. Apparently, they will be 4? reduction, reflective masks with approximately 80 thin layers of reflective material on them. The 80 reflective layers are impossible to repair back to the original specifications. The final construction will probably have a thin layer of oxide on top of the 80 reflective layers. A thin layer of metal on top of this will contain the mask pattern. Much like chrome masks, the metal layer would have clear and opaque defects. The extra metal defects would be milled away and the missing defects would require some sort of metal deposition. FIB techniques are available to etch various types of metal, including chrome, aluminum and, tungsten, and to deposit metals such as tungsten, gold, platinum, and tantalum.
Gallium implantation could be one area of concern. If the oxide layer is etched away after repair, it would probably take most of the gallium with it. If the layer is not etched away, a different ion source, such as a gas field ion source, would be necessary.
SCALPEL. Repair issues for SCALPEL masks are similar to x-ray masks. Since SCALPEL are 4? reduction masks, edge accuracy requirements will be easier to meet. The absorbing layer is made of tungsten and the substrate is silicon carbide. Extra absorber can be milled or etched away and missing absorber can be replaced by depositing tungsten. The tungsten deposition will probably need to be trimmed to meet the edge accuracy requirement. Gallium implantation could present a problem. Preliminary tests have demonstrated the feasibility of repairing these masks with gallium ions .
As masks get more difficult to manufacture and their cost rises, repair becomes more important. Advanced masks today can cost upwards of $50,000 and take 10 or more hours to write and inspect. These costs and manufacturing times are only going to increase.
All future lithographic generations will require better repair accuracy. Some approaches will promote better accuracy because the substrates are conductive. Other improvements could be made through ion column development. Manufacturers of FIB equipment are making constant improvements to ion beam resolution and placement accuracy, but nonconductive substrates negate many of these improvements. Some mask types may need conductive coatings to get the required placement accuracy.
Improved gas-assisted etching processes, smaller ion beams, and lower accelerating voltages can minimize quartz damage. These improvements come at a price, though. Smaller beams will reduce throughput and may hamper ion imaging resolution. Lower accelerating voltages likewise will produce ion beams with less resolution - the spot size is inversely proportional to the accelerating voltage.
Smaller beams and lower voltages can reduce gallium implantation, but at some point even a small amount of gallium may be unacceptable. Gas field ion sources may produce high-brightness ion beams with inert ions such as helium and neon. These are still under development and are probably years away from commercial introduction. It is still not certain if this technique will be useful in a production environment.
If mask sizes do progress to 9 in., new FIB systems capable of handling a substrate of this size will have to be developed. The cost to design and build such a system might be prohibitive, considering the relatively small market, and outside funding might be necessary.
Nonoptical lithography techniques will all require substantial R&D efforts in order for repair to work well. Repair of all of these mask types looks feasible, but pursuing all of them would be prohibitively expensive.
1. K. Levy, "The History and Future of Mask Making," Proc. 16th Annual Symposium on Photomask Technology and Management, SPIE, Vol. 2884, 2-5, 1996.
2. P. Yan, R. Remling, "Focused ion beam repair of 193 nm reticle at 0.18-?m design rules," Proc. 17th Annual Symposium on Photomask Technology and Management, SPIE, Vol. 3236.
3. J. Melngailis, "Focused ion beam technology and applications," J. Vac. Sci., pp. 469-495, March/April 1987.
4. M. Raphaelian, et al., "Study on edge placement accuracy of opaque and clear defect repairs using focused ion beam technology," Proc. 17th Annual Symposium on Photomask Technology and Management, SPIE, Vol. 3236.
5. L. Harriott, "SCALPEL: a projection electron-beam approach to sub-optical lithography," SEMATECH Conference on Lithography, Oct. 1997.
JOHN C. MORGAN is product marketing manager atr Micrion Corp., One Corporation Way, Peabody, MA 01960-7990; ph 978/538-6722, fax 978/531-9648, e-mail firstname.lastname@example.org.