One method for reducing chemical and DI-water use involves the application of ultradilute wafer-cleaning chemistries. Actively used for various manufacturing processes, including prebonding, prediffusion/oxidation, preepitaxial deposition, post-laser-marking applications, and post–chemical-mechanical polishing (CMP) cleaning operations, ultradilute cleaning chemistries are highly efficient at removing submicron surface contamination and trace metallic impurities. Such chemistries reduce metal and particulate contamination from source chemicals, minimize surface haze and roughness, reduce decoration of crystal-originated particles, decrease particle redeposition resulting from low ionic strength, increase concentration stability, prolong bath life, shorten DI-water rinsing times, significantly lower chemical use, and have a lesser impact on the environment, safety, and health than other cleaning methods.

This article describes experiments involving a combined ultradilute cleaning chemistry/motionless drying (UD/MD) system that effectively removes both submicron particulate contamination and surface haze from silicon wafer substrates. The experiments were conducted at M/A-Com (Burlington, MA), which manufactures radio-frequency (RF) microwave devices using silicon and gallium arsenide processes.

The chemistry in question was ultradilute RCA-SC-1 followed by RCA-SC-2 solutions. All cleaning and DI-water rinsing steps were performed in the same processing vessel. Drying was performed in a motionless vessel. The cleaning, rinsing, and drying processes were carried out at ambient temperature. Following the cleaning, rinsing, and drying procedures, a laser-beam scanning system was employed to directly measure surface haze concentration and the density of light point defects (LPDs) at 0.5-µm or larger sizes.

Based on an analysis of the manufacturing data derived from the experiments, the article discusses the ability of the cleaning method to remove particles and trace metals efficiently. Electrical test data of total oxide surface charge, density of interface traps, and regeneration lifetime are also presented. Finally, data from a critical wafer-cleaning operation using an ozonated DI-water chemistry are discussed briefly.

Ultradilute RCA-SC-1/RCA-SC-2 Experiment

Cleaning Chemistry. Typically used to remove light organics from silicon wafer surfaces, alkaline RCA-SC-1 (pH = 9.0–10.0) is an aqueous solution of hydrogen peroxide (H₂O₂) and ammonium hydroxide (NH₄OH). The H₂O₂ grows a chemical oxide on a silicon surface and NH₄OH etches the oxide, thereby lifting particulate contamination off the wafer surface. The chemistries' particle-removal efficiency depends on the volumes of NH₄OH and H₂O₂ chemicals in DI water. Typically, a solution containing one part of NH₄OH to two parts of H₂O₂ is recommended to achieve maximum submicron particle removal.¹

Acidic RCA-SC-2 (pH = 1.6–3.0) is conventionally an aqueous solution of hydrochloric acid (HCl) and H₂O₂. Typically, this chemistry is used to remove trace metallic impurities.

In the experiments discussed here, the wafer surface was cleaned with the RCA-SC-1 solution followed by the RCA-SC-2 solution. The alkaline RCA-SC-1 chemistry was effective in providing a high negative zeta potential of silicon and oxide, thus aiding in the removal of submicron particles. In this study, the RCA-SC-2 chemistry did not contain H₂O₂. The acidic RCA-SC-2 chemistry had good metal oxide solubility, since the chloride ion acted as a complexing agent. Thus, trace metallic impurities, surface haze, and submicron LPDs were significantly reduced.

Controlling Boundary-Layer Thickness. One aspect of cleaning is to control the thickness of the boundary layer. The thickness of an acoustic boundary layer in water is 610 nm for RF megasonic baths operating at 850 kHz frequency, while the thickness of a hydrodynamic boundary layer is 4000 times higher, or 2.57 mm. The thickness of the boundary layer should be as small as possible to achieve effective particle removal. The higher the acoustic frequency, or streaming velocity, the lower the thickness of the boundary layer. Because a thinner boundary layer is better able to expose submicron and nanoscale particles on the surface to larger bulk-flow velocities than a thicker one, it increases the drag force and results in more-efficient particle removal.²

Chemical Cleaning System. Alkaline RCA-SC-1 and acidic RCA-SC-2 solutions were used in the same processing chamber. All chemical cleaning and DI-water rinsing steps were performed at ambient temperature in the same processing vessel.

A low-profile PFA Teflon wafer carrier was employed for all chemical cleaning and drying operations. The carrier's open-flow design was effective in removing particles from the hard-to-reach edges and corners of the silicon wafers. Furthermore, the carrier had a solid end wall, which acted to shield the tops of the wafers from particulate contamination.

Motionless Drying System. The drying vessel was motionless. In other words, the silicon wafers remained stationary during the entire drying operation. Less than 5 ml of gigabit-grade isopropanol (IPA) solvent were used at ambient temperature. Ultrapure, hot, low-pressure nitrogen acted as a carrier gas for the IPA solvent.

Surface-Contamination Measurement. A Surfscan wafer-inspection system from KLA-Tencor (San Jose) was employed to directly measure LPDs and surface haze on the silicon wafers. This system can measure LPDs down to 0.26 µm in size and surface haze in concentrations of <0.4 ppm. LPD counts and surface haze are generally regarded as true indicators of the performance efficiency of processing equipment, as well as of a unit's operation.

Surface-Charge Analyzer. A surface-charge analyzer (SCA) technique was used to characterize the electrically active properties of the thermal oxide film grown on the silicon wafers that had been chemically cleaned with the different systems. The SCA is a modified surface photovoltage instrument that provides immediate, real-time, nondestructive electrical measurements of dielectric films and dielectric/semiconductor interfaces. Since the probe tip of the SCA incorporates a thin Mylar dielectric film that forms the necessary capacitor, the instrument also can measure bare silicon surfaces having native oxide film.

The electrical parameters measured by the SCA—the density of interface traps at the midgap-inversion concentration (DitMG-inv), the total oxide charge-inversion region (Qox-inv), and surface generation lifetime values—are similar to those obtained by capacitance voltage (CV) techniques, but SCA values may not correlate with CV-determined values. Moreover, the SCA approach is simpler and faster than the CV approach.³

Assessing the Performance of the Ultradilute Cleaning System

The ultradilute cleaning system has been used in a manufacturing setting for more than 2 years to perform a variety of functions, including prebonding, prediffusion/oxidation, preepitaxial deposition, post-laser-marking, and post-CMP cleans. During that period, data have been gathered on submicron surface contamination, electrical properties, and costs, demonstrating that the system reduces chemical use and wastewater effluents.

Comparing the UD/MD System and a Wet Station/Spin-Rinse Dryer System. Systematic wafer-cleaning tests were performed to evaluate the combined UD/MD system and a system using a standard wet station followed by a centrifugal spin-rinse dryer (WS/SRD). Six 100-mm silicon wafers were cleaned in each system. Before cleaning, three of them, wafers 2, 4, and 5, had 1503–2272 LPD counts ranging in size from 0.5 to 1.0 µm and 697–1009 LPD counts >1.0 µm. In contrast, the other three wafers, 1, 3, and 6, had 80–152 LPD counts ranging in size from 0.5 to 1.0 µm and 35–70 LPD counts >1.0 µm.

Process Parameter	UD/MD Clean	WS/SRD Clean
Mode	Semiautomatic	Manual
Process temperature	Ambient (22°–25°C)	80°–85°C
RCA-SC-1	Ultradilute	Standard
RCA-SC-2	Ultradilute	Standard
Drying step	Motionless dryer	Centrifugal spin-rinse dryer

Table I: Process parameters for performing UD/MD and WS/SRD cleans.

The important process parameters for the UD/MD and WS/SRD cleans are presented in Table I. Comparative postclean LPD data for each system are shown in Figure 1, which indicates clearly that the particle counts on all wafers cleaned with the UD/MD system were several orders of magnitude lower than those cleaned with the WS/SRD system. The wafers cleaned with the UD/MD system had LPD counts of between 0 and 8 particles >0.5 µm in size and surface haze at or below 0.4 ppm. In contrast, postclean LPD counts of between 31 and 130 were observed on the wafers that had been cleaned using the WS/SRD system.

Figure 1: Data comparing postclean LPD results for the UD/MD and WS/SRD systems.

Comparing Voided Yield Data of Prebond Silicon Wafers Cleaned with the UD/MD and WS/SRD Systems. Since the presence of submicron particles can cause voids within the interface layers of bonded pairs, the UD/MD system was used to ensure that prebond silicon wafers had minimal surface contamination. After the chemical clean was performed, the wafers had <10 LPDs ranging in size from 0.5 to 1.0 µm. Surface haze was typically <0.4 ppm. More importantly, the wafers had no LPDs >1.0 µm. A single 1.0-µm particle can cause a void as large as a few millimeters in size, resulting in a failed pair. Thus, each wafer acted as a single die prior to the bonding step. About 4300 acceptable annealed pairs were manufactured during the 2-year manufacturing period. Figure 2, presenting itemized voided yield data for 25-wafer product lots, compares the cleaning performance of the UD/MD and WS/SRD systems for bonded pairs. The UD/MD system had a mean pass yield of >93%.

Figure 2: Itemized voided yield data for 25-wafer product lots comparing the cleaning performance for bonded pairs of the WS/SRD system (lots 7480023.1–7480110.1) and the UD/MD system (lot 7480111.1 and up). The UD/MD system had a mean pass yield of >93%.

Comparing Surface Charge Data for Oxidized Wafers Cleaned with the UD/MD and WS/SRD Systems. Comparative DitMG-inv and Qox-inv data for oxidized wafers are shown in Figure 3 for three different conditions: (a) no clean, (b) a UD/MD clean, and (c) a WS/SRD clean. Mean Qox-inv values were lower for wafers that had been cleaned using the UD/MD system than for those that had been cleaned using the WS/SRD system. Furthermore, mean DitMG-inv values were lowest for wafers that had been cleaned using the UD/MD system. No appreciable change in the surface lifetime of the wafers was noticed after they had been cleaned using the UD/MD system.

Figure 3: Data showing DitMG-inv and Qox-inv concentrations on oxidized wafers cleaned with the UD/MD and WS/SRD systems.

Determining Trace Metallic Impurities on Wafers Cleaned with the UD/MD System. The DitMG-inv value is regarded as an indirect indicator of the concentration of surface metallic impurities. To determine the level of trace metallic impurities, the point-of-use DI water supplied to the UD/MD system was microfiltered with a cartridge rated to remove particles down to 0.04 µm in size, and then processed through a metal-removal cartridge. The metal-removal step was necessary to effectively remove cations from the DI water. After chemical cleaning, metal concentrations on the cleaned silicon wafer surfaces were determined using a total x-ray fluorescence surface analysis technique. Typical concentrations of trace metallic impurities on the wafer surfaces are listed in Table II.

Metallic Impurity	Concentration (1X1010 atoms/cm2)
Calcium	<5.0
Cobalt	<0.3
Copper	<0.4
Chromium	<0.5
Iron	0.8
Manganese	<1.0
Nickel	<0.3
Titanium	< 0.9
Zinc	<0.3

Table II: Typical concentrations of trace metallic impurities on wafer surfaces.

Comparing UD/MD and WS/SRD Process Capacity, Chemical Costs, and Wastewater Loading. The comparative data for one 25-wafer product lot shown in Table III indicate the important advantages of the UD/MD system over the WS/SRD system. The use of the UD/MD system results in an approximately threefold reduction in chemical consumption, a threefold reduction in chemical costs, an eightfold reduction in wastewater loading, and a threefold increase in processing capacity. The system can enable fabs to cut process manpower costs by one-third.

Ozonated DI-Water (DIO₃) Cleaning Experiment

An experiment was conducted to test the performance of an ozonated DI-water (DIO₃) clean. Ozone gas is commonly employed to purify ground and surface waters in domestic and industrial settings. A highly toxic and oxidizing gas, ozone has a recommended threshold limit value of 100 ppb. However, its odor threshold limit of approximately 20 ppb is relatively low.

Process Parameter	UD/MD Clean	WS/SRD Clean
Process time	14 minutes	45 minutes
Ultrapure DI water	30 gallons ($3.00)	25 gallons ($2.50)
Chemicals used	Electronics-grade NH4OH (28–30%) Electronics-grade H2O2 (30%) Electronics-grade surfactant Gigabit-grade HCl (37%) Gigabit-grade IPA (99%)
Total chemical volume (cost)	556 ml ($1.43)	1840 ml ($4.25)
Wastewater loading	0.67%	0.08%

Table III: Data comparing UD/MD and WS/SRD chemical costs and wastewater loading from one 25-wafer product lot.

Ozone is an unstable gas that is produced at the point of use. Its half-life in ultrapure DI water is estimated to be 15 minutes at 20°C. A major benefit of DIO₃ processing is that the gas eliminates the need to perform an RCA-SC-1 clean, which involves the use of NH₄OH and H₂O₂ chemicals. Other environmental advantages of using ozonated DI water include reduced DI-water consumption and increased throughput. The employment of ozone reduces the consumption of DI water by 60–75% and decreases processing times by a factor of two.

All silicon wafer–cleaning trials were performed using a low concentration of ozone gas in DI water. The processing temperature was ambient. No RCA-SC-1 and RCA-SC-2 cleaning chemicals were employed in this experiment.

The ozonated DI-water process was evaluated for prebond cleaning—a critical semiconductor manufacturing operation. Three 25-wafer product lots were cleaned using the DIO₃ process. The experiment showed that the wafer surfaces cleaned with DIO₃ were as clean as those that had been cleaned with ultradilute RCA-SC-1 and RCA-SC-2 chemistries. Typically, <10 LPDs ranging in size from 0.5 to 1.0 µm were measured on each wafer, while surface haze was less than 0.4 ppm. About 70 acceptable annealed pairs were manufactured. For hydrophobic bonding, mean pass yield was >92%.

Conclusion

The UD/MD system is superior to the WS/SRD system because it can be used at ambient temperature, has shorter processing times, and is more efficient at removing submicron surface contamination and trace metallic impurities. The experiments discussed here indicate that the reduction in chemical and DI-water consumption achieved by shifting to the ultradilute cleaning system can generate net cost savings of $2.32 for each 25-wafer product lot. Consequently, the ultradilute cleaning method is both environmentally friendly and cost-effective. The additional use of ozonated DI water can enable fabs to dispense with RCA-SC-1 and RCA-SC-2 chemicals, further reducing the deleterious impact of IC manufacturing on the environment.

Acknowledgments

This article is a modified version of a poster paper presented at the 13th annual IEEE/SEMI Advanced Semiconductor Manufacturing Conference, Boston, April 30–May 2, 2002. The author would like to thank Bruce Cochran, Joel Goodrich, Manny Marcel, and Joseph Maniachi for their technical support.

1. H Maines, M Rathmell, and L Veldhuis, "Reduce Scrap: Control Oxide Loss in SC1," in Proceedings of the 13th Annual IEEE/SEMI Advanced Semiconductor Manufacturing Conference (Piscataway, NJ: IEEE, 2002), 184–186.