SURFACE CHEMISTRIES

Assessing the impact of megasonics on SC-1 bath life

Paul Ocansey and Mani Janakiram, ACT/Motorola

One of the challenges facing the semiconductor industry today is the ability to minimize production costs while maintaining overall product quality. This issue is particularly relevant to the front end of the line (FEOL), where control of metallic and particle contaminants in cleaning chemistries is critical. The adverse effects of contaminants in wet processes are well known and include a reduction of gate oxide integrity, an increase in junction leakage, the presence of mobile ions and silicide precipitates, a reduction in minority carrier lifetime, nucleation of stacking faults near the gate oxide—silicon interface, and microroughness.^1—7

During the past few years, efforts to ensure gate oxide integrity have been focused on several areas, such as the use of new cleaning mixtures, more dilute standard clean 1 (SC-1) solutions, optimization of the SC-1 process temperature, and megasonic wafer cleaning.^8—10 Among the advantages sought by implementing these practices are reduced Si/gate oxide surface roughness, more-efficient particle removal, reduced waste and environmental impact, and a decrease in chemical usage that in turn lowers process costs. Advanced Custom Technologies (ACT), which is based in Mesa, AZ, is one of Motorola's semiconductor wafer R&D fabs. At ACT, the primary goal is to develop new technologies for transfer to the company's manufacturing fabs. A recent project evaluated the impact of megasonic SC-1 cleaning (and the resulting SC-1 bath life) on production cost and equipment availability, throughput, and utilization. It was demonstrated through a process capability study and operational analysis that by integrating megasonics into the cleaning process the life of an SC-1 bath can be nearly doubled with no adverse effects on gate oxide integrity. During the study, data on throughput, work in progress (WIP), cycle time, and equipment availability and utilization were gathered at various ages of the SC-1 bath for purposes of comparison. These scenarios were also simulated for extended periods using ManSim software (Tyecin Systems, Palo Alto, CA) to see the impact on overall fab operations.

Experimental Details

Used as the final cleaning step in most FEOL processes to remove particles and, in some cases, residual organics from the wafer surface, the SC-1 chemistry consists of parts-per-billion-grade ammonium hydroxide and hydrogen peroxide, which are maintained at elevated temperature. During the study, this mixture was contained in a PCT quartz bath assembled in a SPEC hood (Semiconductor Process Equipment Corp., Valencia, CA). The mixture was megasonically agitated at all times, and recirculated and filtered to remove particles following each piranha clean. Throughout the life of the bath, several-milliliter aliquots of hydrogen peroxide were added just prior to processing a lot.

Oxide Removal Test. The effect of the age of the SC-1 mixture on oxide loss was monitored through thermal oxide etch studies. After approximately 6.5 kÅ of thermal oxide was deposited on a monitor wafer, it was immersed in the SC-1 mixture for 10 minutes, followed by a quick-dump rinse in an overflow bath. Final rinse and dry were performed in a spin-rinse dryer (Verteq, Santa Ana, CA). The oxide thickness was then remeasured and the amount of oxide lost was calculated. The experiment was performed at 1-hour intervals, starting from zero (when the SC-1 mixture was fresh) and continuing to 10 hours.

Particle Removal Test. A second test was used to monitor the particle removal efficiency of the megasonically integrated SC-1 mixture over time. By immersing the wafers in a nonfiltered 10:1 buffered oxide etch bath for 10 minutes, particles were deliberately deposited on monitors that had initial particle counts of <10 particles per wafer with sizes >0.1 µm. In this way, from 1200 to 1500 particles were deposited on each wafer. When the SC-1 mixture was fresh and at 1, 4, 6, 7, 8, 9, and 10 hours, two monitor wafers were immersed in the bath for 10 minutes. Both before immersion and after final rinse and dry the wafers were examined using a Surfscan 4500 particle counter (KLA-Tencor, San Jose). Particle removal efficiency was calculated using the equation

where P^_BEF is the number of particles on the wafer before the SC-1 clean and P^_AFT is the number of particles on the wafer after the SC-1 clean.

Metal Removal Test. Metallic contamination was monitored at predetermined intervals during the life of the SC-1 bath using eight bare silicon monitor wafers and one wafer designated as the control. The control wafer was not processed in the SC-1 bath; monitor wafers were immersed for 10 minutes in the fresh mixture and at 4, 6, 7, 8, 9, and 10 hours. Total reflection x-ray fluorescence (TXRF) and secondary ion mass spectroscopy (SIMS) were used to determine the presence of metal contaminants on each test wafer, and results were compared with the control wafer data. In order to simulate an actual production scenario, lots were run regularly during the first 4 hours of the SC-1 clean process.

Results and Discussion

Oxide Removal. Figure 1 shows thermal oxide loss from the monitor wafers following a 10-minute immersion as a function of SC-1 bath age, starting from zero, when the SC-1 was fresh. Each data point represents an average of 10 readings with across-the-wafer uniformity within 3% of the mean. It is apparent from this figure that as the SC-1 mixture ages, less oxide is etched away. The total oxide lost per immersion was insignificant, with measured oxide etch rates ranging from approximately 0.6 Å/min (registered at zero) down to 0.1 Å/min (at 10). The high etch rate at the beginning of the test is attributed to the initial high concentration of ammonium hydroxide; its concentration decreased with the age of the SC-1 bath as hydrogen peroxide was added prior to each process run. The observed range of etch rates differs from those reported elsewhere because of differences in process parameters.⁸

Figure 1: Thermal oxide loss as a function of the age of the SC-1 mixture. Test wafers were immersed for 10 minutes prior to measurements.

Particle Removal Efficiency. With integrated megasonic energy, the particle removal efficiency of the SC-1 clean remained above 95% throughout the 10-hour life of the SC-1 mixture. Figure 2, a histogram of two sets of experiments run on two separate days, clearly demonstrates that, with no other constraints, the SC-1 consistently removed surface particles. Earlier reports have theorized that the chemical reaction between ammonium hydroxide, hydrogen peroxide, and water undercuts surface particles while the vibrational energy released by the megasonic unit liberates the particles by overcoming the weak but attractive van der Waals forces holding them to the substrate.^8,9 We believe, however, that acoustic pressure buildup from the pulsed megasonic frequencies used—a result of random fluid motion or microstreaming—is more likely to be the cause of particle removal. The fact that particle removal efficiencies remained relatively constant at around 95% while the etch rate decreased with age of the SC-1 mixture suggests that etch rate is not a particle removal mechanism, as was reported elsewhere.⁸

Figure 2: Particle removal efficiency as a function of the age of the SC-1 mixture. Test wafers were immersed for 10 minutes prior to measurements.

During the entire study and beyond, particle count data were evaluated using statistical process control (SPC). Figure 3 compares the particle SPC charts for the SC-1 bath after 4, 6, and 8 hours. No significant variation was noticed and hence it is safe to assume that increasing the bath life does not compromise the quality of the SC-1 cleaning process.

Figure 3: Statistical control charts for particles at various times in the SC-1 bath life: (a) 4 hours, (b) 6 hours, and (c) 8 hours.

Metal Removal Efficiency. Figure 4 represents the quantitative results of the TXRF analysis, which reveal the effect of SC-1 mixture age on metallic contamination. For comparison purposes, data for the control wafer are also presented. In the entire 10-hour study with megasonic vibration, only iron (Fe) and zinc (Zn) were detected on the wafers. Their concentrations, however, were well within one order of magnitude of the detection limit of the TXRF detector until the SC-1 mixture was 9 hours old, at which time the metallic contaminants began to build up on the wafers. In a related experiment without megasonic energy these metals began to build up sooner, so it is assumed that agitation from the megasonic unit keeps the metals in solution for up to 9 hours. As reported elsewhere, metals build up because of a combination of several competing effects, which include physical and chemical interactions between the bath and the quartz material used to build the tank, the purity of the SC-1 constituent chemicals, and the combination of process steps.^1,7

Figure 4: TXRF data showing metal contamination as a function of SC-1 mixture age. Test wafers were immersed for 10 minutes prior to measurements; the control wafer was not processed in the SC-1 bath.

SIMS was also performed on all monitor wafers to confirm the presence of Zn and Fe as seen in the TXRF analyses, and these metals were observed only when the SC-1 mixture was 10 hours old. The SIMS profile at 10 hours also indicated there were trace amounts of calcium and potassium on the wafers, but the concentrations of these two elements were insignificant.

Model Predictions and Data Analysis

An increase in equipment uptime, or availability, can contribute significantly to cycle time improvement, particularly for constraint equipment. In general, mean time between failures (MTBF) and mean time to repair (MTTR) depend primarily on equipment reliability and will not be affected by changes in SC-1 bath life. However, mean time between assists (MTBA) is dependent on SC-1 bath life since the primary assist function is chemical changeover. Mean time to assist (MTTA) will remain the same, but MTBA will increase with an increase in bath life. In this study, MTBA for the SC-1 clean using various bath lives was compared to study its effects on operational metrics. The impact of the SC-1 clean MTBA on overall fab performance was modeled, and Figure 5 shows the cycle time—to—process ratio simulation data for 4-, 6-, and 8-hour baths. (The cycle time—to—process ratio depends on queue time, process time, and MTBA.) The results from the simulation scenarios were comparable with the actual data obtained and good mapping was observed.

Figure 5: Simulation data showing the impact of increased MTBA (resulting from increased bath life) on overall fab performance.

The actual lot cycle time, which is a sum of queue time and process time, was computed and compared with theoretical cycle time for all the recipes run on the SC-1 equipment.¹¹ Theoretical cycle time is the ideal time to process a wafer lot through the system, and does not include queue time and setup time. Multiples of theoretical cycle time (MTCT), a ratio of actual to theoretical cycle times, was also computed. As seen in Figure 6, which compares the MTCT when using various SC-1 bath lives, when the bath life is increased, MTCT decreases. The nature of the processes at ACT demand that most lots go through the SC-1 clean several times, and it had been observed that queue time at this step was often excessive. Increasing the bath life reduced the queue time, which in turn reduced the overall lot MTCT by 12%.

Figure 6: Multiples of theoretical cycle time (MTCT) as a function of SC-1 bath life.

The significant positive changes in equipment availability that occurred after bath life had been increased are shown in Figure 7. This additional capacity not only reduced the queue time, but permitted the processing of additional lots. Also, as a result of the increased bath life, less operator time is spent on chemical changeover and process qualification, and significant cost savings have been realized as a result of the reduction in chemical usage.

Figure 7: Equipment availability as a function of SC-1 bath life.

Conclusion

The study reported in this article indicated that by including megasonic energy, SC-1 bath life can be increased from the usual 4 hours to 8 hours without compromising gate oxide integrity. Oxide loss during a 10-minute immersion of monitor wafers was insignificant at all ages of the bath and decreased exponentially with the age of the SC-1 mixture. Particle removal was found to be very efficient even with a long bath life, while metal buildup in the SC-1 mixture was not apparent during the first 9 hours of the 10-hour test period. Based on these results, ACT increased its SC-1 bath life, which has resulted in higher equipment utilization, lower chemical costs, and lower cycle times.

Acknowledgments

The authors would like to thank Marque Holliness, Kathy Reavis, and Wendy Wilson for sample preparation; Dave Addie, Paul Dryer, and Tom Wood for useful technical discussions; Gari Harris and Kari Noehring for SIMS assistance; Erika Duda and Gary Mote for TXRF analysis; and the ACT cycle time team for its helpful input. Special thanks goes to Bruce Huling of MEMS-1 Motorola for providing useful feedback on the paper.

References

1. Talasek RT, Hunt BK, and Cooke DC, "Evaluating Reusable HDPE Containers for Delivery of High-Purity Hydrofluoric Acid," MICRO, 15(7):95—100, 1997.

2. Hall RM, Rosato JJ, and Lindquist PG, "Optimization of SC-1 Megasonic Cleaning Efficiency for 200-mm Wafer Processing," in Proceedings of the 1994 Microcontamination Conference, Santa Monica, CA, Canon Communications, pp 529—537, 1994.

3. Cohen SL, Rath D, and Lee G, "Studies of the Relationship between Megasonics Surface Etching and Particle Removal in SC-1 Solutions," in Materials Research Society Symposium Proceedings, vol 386, Pittsburgh, Materials Research Society, pp 13—19, 1995.

4. Henley WB, Jastrzbski L, and Haddad NF, "Monitoring Fe Contamination in Si by Surface Photovoltage and Correlation to Gate Oxide Integrity," in Materials Research Society Symposium Proceedings, vol 386, Pittsburgh, Materials Research Society, pp 299—312, 1995.

5. Helms CR, and Park HS, "Generalized Model of Metal Bonding and Cleaning from Wafer Surfaces," in Materials Research Society Symposium Proceedings, vol 315, Pittsburgh, Materials Research Society, pp 287—297, 1993.

6. Park HS, Helms CR, Ko DH, et al., "Effect of Surface Fe on Gate Oxide Integrity and Its Removal from Si Surfaces," in Materials Research Society Symposium Proceedings, vol 315, Pittsburgh, Materials Research Society, pp 353—358, 1993.

7. Francis T, "Not by Particle Count Alone: Getting to the Root Cause of Contamination," MICRO, 14(7):69—78, 1996.

8. Meuris SM, Mertens PW, Opdebeeck A, et al., "The IMEC Clean: A New Concept for Particle and Metal Removal on Si Surfaces," Solid State Technology, 38(7):109—113, 1995.

9. Singer P, "Wafer Cleaning: Making the Transition to Surface Engineering," Semiconductor International, 18(11):88—92, 1995.

10. Hall RM, Jarvis TD, Parry T, et al., "Investigating Particle, Metallic Deposition in Megasonic Wafer Cleaning," MICRO, 14(7):81—88, 1996.

11. Janakiram M, "Cycle Time Reduction at Motorola's ACT Fab," presented at the 7th Annual IEEE/SEMI Advanced Semiconductor Manufacturing Conference and Workshop, Cambridge, MA, November 12—14, 1996.

Paul Ocansey, PhD, is a process engineer at Motorola's Advanced Custom Technologies (ACT) R&D fab in Mesa, AZ, where his responsibilities include development of lead technologies in the areas of cleans and resist stripping. He is also involved in metals evaluation and deposition, including barrier materials, glue layers, polycides, and interconnects. Before joining Motorola, Ocansey was a postdoctoral research fellow in the department of materials science and engineering at the University of Arizona, Tucson, where he received a PhD in materials science and engineering. Ocansey can be reached at 602/655-5456 or rp4973@email.sps.mot.com.)

Mani Janakiram is a section manager at ACT, where he is active in manufacturing modeling, cycle time reduction, statistical analysis, production control, and manufacturing systems. He is a Motorola Six Sigma Black Belt, an ASQ-certified quality engineer, and an SME-certified manufacturing engineer. He holds a BS in mechanical engineering from University of Mysore, India, and an MS in industrial engineering from Arizona State University (Tempe). He is pursuing a PhD in industrial and management systems engineering at Arizona State University. (Janakiram can be reached at 602/655-5205 or rp5414@email.sps.mot.com.)

MicroHome | Search | Current Issue | MicroArchives
Buyers Guide | Media Kit