While filters with 0.02-µm pores are advantageous for reducing these defects, small pore size alone does not eliminate them. In contrast, changes in the surface chemistry of the membrane without further pore-size reductions do result in the virtual elimination of defects in both the resist and antireflective coating layers. While some performance improvements depend on the filter surface and pore size only, other improvements depend on the volume of photochemical processed by the filter. In other words, once a prescribed volume of photoresist has been processed, defect levels return to near-prefiltration levels. This type of performance indicates that the filter surface relies on an adsorptive mechanism to remove contamination. In such cases, the membrane surface is active, enabling users to develop more-sophisticated algorithms to estimate filter lifetime.

This article discusses the nature of particulate and bubble defects in 193-nm photoresist processes and the interaction of the membrane surface with the photochemicals. In addition, it presents the results from laboratory and wafer-production tests performed by investigators at Infineon Technologies. Finally, the article offers recommendations for estimating filter lifetime.

Defect Types and the Role of Filtration in Reducing Defects

Increasingly fine linewidths require the use of new classes of photochemicals that are exposed using argon fluoride (ArF) excimer lasers. Known as ArF or 193-nm photochemicals, these resists rely on acid generators for chemical amplification. ArF photoresists generate microbridging defects, examples of which are shown in Figure 1.¹ To achieve sharp linewidths, 193-nm photochemicals also require the use of bottom antireflective coatings (BARCs) and top antireflective coatings (TARCs). BARCs and TARCs are acidic and have high concentrations of surfactant, resulting in very low surface tension.

Figure 1: Examples of microbridging defects associated with 193-nm ArF photoresists.

The physical properties of photoresists and antireflective coatings have prompted the semiconductor industry to modify its approach to the purification of point-of-use chemicals. To reduce photochemical defect levels, the industry is pursuing both an evolutionary approach (the development of filters with finer pore sizes) and a revolutionary approach (the modification of membrane surfaces).

The first filters used in photochemical purification were based on nylon or cellulosic microporous membranes. Initially, these membranes were developed for the removal of microbiological organisms from pharmaceutical fluids, but they were also able to reduce particle contamination in photochemicals, which contain many solvent bases.

Teflon PTFE membranes were introduced to the semiconductor manufacturing process because of their chemical compatibility. They were also used in photochemical filtration with some success. While Teflon filters were unaffected by the strong solvents used in photochemicals and their membrane structures had comparatively low organic and ionic extractables, the Teflon surface was much less wettable than the nylon and cellulosic filters they replaced. For many photoresists, high pressures are required to force the photochemical into the filters' pore structure. In particular, the use of cyclohexanone or N-methyl pyrrolidone–based photochemicals requires high pressure to wet PTFE filters.

Ultra-high-molecular-weight polyethylene (UPE) membranes were used quite successfully as photochemical filters. The UPE membrane and high-density polyethylene (HDPE) filter components were compatible with almost all photochemical solvents. The UPE membrane wetted spontaneously with all solvent-based photochemicals, eliminating the start-up difficulties of PTFE membranes. UPE membranes reduced photoresist waste and resulted in a significant reduction in wafer defects.²

While UPE membranes wetted spontaneously with ArF photoresist and antireflective coatings, there was empirical evidence that a membrane surface with an even higher critical surface energy results in lower wafer defects. A development program was initiated to understand the effects of that membrane surface on defects.

Improving Filter Performance

Increasing Particle Retention. While 0.05-µm membrane retention ratings became the standard for deep-ultraviolet (DUV) resists, the finer feature size of ArF photoresists required even finer filtration. Consequently, membranes with 0.03- and 0.02-µm retention ratings were developed.

These membranes' retention was tested using mono-dispersed 0.034-µm polystyrene latex beads and a modified Sematech test method.³ Retention was measured in terms of log reduction value (LRV), a very sensitive method for detecting the slight passage of particles. The LRV is a logarithm of the ratio of the number of particles in the feed to the number of particles in the filtrate. The maximum measurable LRV is controlled by the number of particles in the feed and the background counts of the particle counter. In the tests performed to determine the retention ratings of fine ArF photoresist filters, background system particle counts were subtracted from the filtrate values to improve sensitivity so that retention performance could be compared at high LRVs.

Although 0.05-µm UPE membranes were able to retain a high percentage of 0.034-µm PSL beads, retention degraded as particle loading increased, limiting the lifetime of filters in fluids containing high concentrations of contaminating particles. In contrast, the 0.03- and 0.02-µm rated filters demonstrated better retention rates under very high particle-loading conditions, as illustrated in Figure 2.

Figure 2: Retention rates of 0.05-, 0.03-, and 0.02-µm UPE membranes for 0.034-µm PSL beads.

Eliminating Bubbles. In photochemical purification applications, the elimination of bubbles is as important as particle removal. Bubbles in photoresist or antireflective coating layers result in wafer defects. Bubbles form in liquids when the solubility of dissolved gases decreases. In addition, pressure fluctuations, such as those created during fluid pumping, can cause bubbles to form. Three bubble-formation mechanisms have been proposed in the literature: homogeneous nucleation, heterogeneous nucleation, and cavitation.⁴

• Homogeneous nucleation results in the formation of microbubbles everywhere in a liquid when gas molecules form clusters and grow to a defined size. This phenomenon occurs when supersaturated dissolved gas in a liquid suddenly becomes insoluble—for example, when pressure is reduced.

• Heterogeneous nucleation is defined as bubble growth on hydrophobic surfaces. Hydrophobic surfaces or particles act as catalysts for bubble formation when gas solubility in a liquid is reduced.

• Cavitation is characterized by bubble formation at nucleation sites caused by a sudden pressure drop of a moving fluid.

A rare occurrence, homogeneous nucleation is not a likely mechanism for bubble formation in TARCs. In contrast, both heterogeneous nucleation and cavitation are the likely mechanisms for bubble formation in photochemicals. Both the surface energy and pressure drop of a membrane surface may play a role in microbubble formation, and the membrane's pore size plays a role in microbubble removal. Filter manufacturers must perform a balancing act by developing membrane surfaces that both remove bubbles from photochemical fluid and retard bubble formation downstream of the filter.

Optimal Filter Surface Properties. The optimal filter has a combination of physical properties that enables it to remove contamination from photochemicals:

• Its membrane is wetted spontaneously by the chemicals.

• It has high capillary forces to completely wet surfaces.

• It eliminates vapor from voids.

• It has a small pore size so that particles and bubbles can be removed.

• Because of its low-pressure-drop characteristic, it can have a small membrane area and reduce outgassing downstream of the membrane.

• Photochemical compatible, it does not have extractables and has a long lifetime.

While thermal-induced phase-separation UPE membranes meet most of these requirements, photochemicals with high surface tension (>35 dyn/cm) do not wet the membrane surface spontaneously. Hence, a modified UPE surface has been developed that accommodates chemicals with the surface tension of water (72 dyn/cm) and higher. The modified UPE surface has the following properties:

• A base membrane that is inert to photochemicals.

• A smooth structure with few crevices.

• A smooth surface with a cross-linked polymer matrix that increases the membrane's surface energy and maintains the bulk property of the UPE.

Figure 3 presents a schematic diagram of the modified UPE membrane.

In addition to the size-exclusion and wetting attributes of the membrane surface, the membrane may exhibit adsorptive effects. The surface removes trace contaminants as a result of hydrophobic binding or charge effects. Depending on the nature of the surface, these effects can be significant. While the membrane surface can be well characterized, its interactions with complicated photochemical chemistries can be difficult to predict. Nevertheless, these interactions can be measured empirically. Once they are recognized, a hypothesis describing their function can be developed.

Figure 3: Schematic diagram of the modified UPE membrane.

Optimizing Filter Design

Tests to optimize filter design commenced when process engineers at Infineon observed that unacceptably high defect levels resulted from using 0.05-µm filtration at the point of use to purify an ArF photoresist. While reducing the pore size of the UPE membrane to 0.02 µm reduced wafer defect levels, those levels remained high. Figure 4 shows that defect levels increased over time when 0.05- and 0.02-µm filters were used in the lithography process.

Figure 4: Wafer defect levels resulting from the use of 0.05- and 0.02-µm filters.

To address this problem, investigators from Infineon and Mykrolis (Billerica, MA) began to consider the use of nylon. Although they knew that the bulk properties of nylon limit the material's chemical compatibility, empirical evidence indicated that filters with a nylon surface can reduce wafer defect levels. Investigators hypothesized that a nylon surface has higher surface free energy than UPE alone—in other words, it is more hydrophilic or wets with high-surface-tension fluids. In addition, they observed what appeared to be an adsorptive interaction between the membrane surface and the resist (high-molecular-weight contaminants in the resist are adsorbed by the amide surface).

Consequently, Mykrolis membrane scientists developed a filter with a nylonlike surface membrane that maintains the bulk chemical compatibility of UPE. Figure 5 compares Fourier transform infrared data from four filters that were used to filter ArF photoresist. Initially, none of the membranes had an adsorption peak at 1720 cm–1. While filter modification 1 (a nonsieving UPE filter with a surface similar to nylon) adsorbed some contaminant material from the photoresist in the nylon-adsorption area, the base UPE membrane did not adsorb material. Since filter modification 2 had more amide functionality per mass than nylon, resulting in more contaminant adsorption, that modification became the basis for Mykrolis's PCM filter. The filter was tested at Infineon.

Figure 5: Adsorption of material from ArF photoresist on different membrane surfaces.

Prior to field testing, the chemical compatibility of the PCM membrane was compared with that of nylon to ensure that its physical properties did not degrade. UPE membrane samples with a PCM surface and nylon membranes were soaked in photoresist for one month. Then the strength and elongation of the membranes were evaluated using an instrument from Instron (Norwood, MA). Samples of the membrane were pulled at a controlled rate while force and elongation were measured. The samples' strength and elongation before and after the one-month soaking period are illustrated in Figures 6 and 7, respectively. While the nylon material's strength declined over that period and its elongation decreased by 90%, the UPE membrane's strength increased slightly and its elongation fell by only 50%.

Figure 6: Tensile strength of membranes before and after exposure to DUV resist.

Figure 7: Elongation of membranes before and after exposure to DUV resist.

To determine the effects of filter designs on wafer defect levels, tests were performed involving several filter samples with different surface types and pore sizes. The objective of the tests was to determine if pore size alone or surface alone could reduce defect levels. First, wafers were processed in an ACT 8 coater/developer i-line track tool from Tokyo Electron (Tokyo). Standard pump parameters and 193-nm acrylate-based photoresist were used. Finally, the wafers were inspected using a KLA 2351 from KLA-Tencor (San Jose). The filters tested are listed in Table I.

Filter Type Characteristics/Test Purpose Control 0.05-µm Impact LHVD (Mykrolis) Filter A 0.02-µm UPE (tight pore size) Filter B 0.05- and 0.02-µm double-layer UPE (improved gel particle retention) Filter C 0.05-µm UPE with modification 1 (check for adsorption) Filter D 0.02-µm UPE with modification 1 (check for adsorption and tight filtration) Filter E 0.02-µm nylon/0.02-µm UPE (check for adsorption and tight filtration) Filter F 0.02-µm UPE with modification 2 (PCM) (check for adsorption)

Table I: Filters tested to determine the effects of surface type and pore size on wafer defect levels.

Figure 8 summarizes the results of the tests. While both tighter membranes and improved surfaces reduced defect levels, the combination of a tighter membrane and the maximum amide surface concentration (Filter F) resulted in the lowest defect levels. Figure 9 shows the defect performance of a production PCM filter over an extended period of time.

Figure 8: Production wafer defect levels resulting from the use of different filters.

Figure 9: Defect performance of filter F (PCM filter) in a production setting over an extended period of time.

Filtration of Top Antireflective Coatings

Bubbles are the major source of defects in TARCs. Since TARCs have high levels of surfactant, resulting in low surface tension, they have a high propensity for outgassing. To determine bubble levels, investigators conducted laboratory experiments using a Mykrolis IntelliGen 2 dispense system and AZ Aquatar TARC from AZ Electronic Materials (Somerville, NJ). A LiQuilaz SO2 optical particle counter from Particle Measuring Systems (Boulder, CO) was installed on the dispense line. While optical particle counters are not designed to count bubbles, particle results can be used in a semiquantitative manner to determine differences in filter performance. All membranes tested were of the flat-sheet pleated variety, with the exception of a 0.1-µm hollow-fiber UPE membrane. The filters were primed with TARC and the dispense recipe was performed continually until particle counts leveled off.

Test results showing several different filters' retention efficiencies are shown in Figure 10. (Retention efficiency has the greatest effect on particle counts.) The lowest particle count (i.e., the best particle-retention efficiency) was achieved using a PCM 0.02-µm hydrophilic filter. Figure 11 shows the results of a flow-stoppage test. The use of the PCM filter resulted in the lowest particle counts when flow was reestablished. That filter resulted in a smaller particle-count spike than Mykrolis's LHVD 0.05-µm hydrophobic membrane and minimized bubble formation and bubble transmission.

Figure 10: Laboratory tests showing particle levels resulting from the use of different filters. Particle levels indicate the level of microbubbles in the dispense line.

Figure 11: Laboratory tests comparing wafer particle performance (bubble level) of a 0.02-µm hydrophilic filter and a 0.05-µm hydrophobic filter after flow had been stopped for 2 hours.

Following the laboratory tests, 0.02-µm Impact Plus PCM filters were evaluated at Infineon. The filters were installed in a Mykrolis Two-Stage Technology dispense system using AZ Aquatar TARC. The TARC-coated wafers were analyzed for defects using a KLA-Tencor AIT 2 patterned-wafer inspection tool. The tests revealed that the filter's pore size, retention efficiency, and surface energy had a dramatic effect on wafer-level defects. The PCM filter resulted in 57% fewer defects than a 0.04-µm nylon membrane, 85% fewer defects than a 0.1-µm nylon membrane, and 88% fewer defects than a 0.1-µm hollow-fiber UPE membrane.

Conclusion

As device geometries shrink, particulate contamination below 0.05 µm must be controlled. Both a filter's pore size and surface properties must be optimized to purify ArF photochemicals. The use of filter membranes with a fine (0.02-µm) pore structure reduces defects on the wafer surface. In addition, the membrane must be wettable in the photochemical to prevent the formation of gas bubbles and make filter start-up as easy as possible.

For many ArF resists, a surface with amide functionality may remove contamination because of its adsorptive nature. However, an adsorptive removal mechanism has a finite capacity. The number of sites on the wafer surface and the amount of contamination in the chemical control filter lifetime. In addition, flow rate, or residence time, can play a minor role in contamination removal. Membranes are particularly good adsorptive surfaces because they are in intimate contact with the chemical. While the removal kinetics of membranes are more favorable than those of resin beads, their capacity is lower than resins that are designed to remove particles using adsorption.

Since resist contamination levels are generally an uncontrolled variable, it is difficult to predict the lifetime of filters that use adsorption as a purification technique. Contamination levels follow a breakthrough curve, where defect levels start to increase gradually before significant degradation occurs. The higher the flow rate, the more apparent the degradation, since higher flow rates reduce the residence time. The only good monitor of resist contamination is wafer defect levels. When wafer defects increase, the filter should be replaced.

Active filter surfaces can be used to develop new photoresists. By testing each component of the resist formulation with the active filter, contamination sources can be segregated and controlled. The easiest point in the production cycle to remove contamination is during the resist manufacturing process, when sophisticated analytical techniques can be used. The active filter surface then becomes an insurance policy in the fab rather than the last line of defense, resulting in longer filter lifetime and a lower cost of ownership.

Acknowledgments

This article is an edited version of a presentation given at Semicon Korea, February 2–4, 2005, in Seoul. The authors would like to thank Infineon Technologies for providing the manufacturing-based data appearing in the article and for providing Figure 1. They also thank Joseph Zahka, Saksatha Ly, and Bipin Parekh for their technical contributions and for preparing the membranes evaluated in this study.

References

1. PB Sahoo et al., "Progress in Deep-UV Photoresists," Bulletin of Material Science 25, no. 6 (2002): 553–556.

2. L Mouche et al., "Photoresist Filtration Performance of UPE and PTFE Filters," in Proceedings of the IES Annual Technical Meeting (Mount Prospect, IL: Institute of Environmental Sciences, 1996), 259–267.

3. J-K Lee, BYH Liu, and KL Rubow, "Latex Sphere Retention by Microporous Membranes in Liquid Filtration," Journal of the IES 36, no. 1 (1993): 26–36.

4. J Duffner, "Defect Reduction in Top Antireflective Coating," Applications Note AN1019ENUS (Billerica, MA: Mykrolis, 2004).

---

Mutsuhiro Amari is manager of applications technology development at Mykrolis in Tokyo. He received a BS in industrial chemistry from Ibaraki University. (Amari can be reached at +81 3 54429713 or mutsuhiro_amari@mykrolis.com.)

Aiwen Wu, PhD, is an applications engineer at Mykrolis in Billerica, MA. Since joining the company in 2003, he has focused on wet etch and clean filtration, electrochemical-plating filtration, and photochemical filtration products. He received a PhD in chemical engineering from the University of New Hampshire in Durham. (Wu can be reached at 978/436-6820 or aiwen_wu@mykrolis.com.)

Hee Jun Yang is an Asia applications manager for Mykrolis. He has been with the company since 1996, working in application development for microelectronics, chemical manufacturing, and flat-panel manufacturing processes. Before joining Mykrolis, Yang was section chief of the process integration division of Hyundai Electronics. He received BS and MS degrees in materials sciences and engineering from Hanyang University, in Seoul. (Yang can be reached at +82 31 7385331 or hj_yang@mykrolis.com.)

Linda Chen, PhD, is a lithographic process engineer at Infineon Technologies in Richmond, VA. In 1998 she received a PhD in chemical engineering from the University of Rochester in Rochester, NY. (Chen can be reached at 804/952-8098 or Linda.chen@infineon.com.)

Thomas Bowling is a lithography equipment engineer at Infineon. Active in the semiconductor industry for 15 years, he received a degree in mechanical engineering in 1989 from the West Virginia Institute of Technology in Montgomery. (Bowling can be reached at 804/952-7607 or thomas.bowling@infineon.com.)

Michael Watt is a senior staff process engineer in the lithography department at Infineon. He received a degree in mechanical engineering from the University of Glasgow in Scotland in 1983. (Watt can be reached at 804/952-6131 or michael.watt@infineon.com.)