"Precision Cleaning - The Magazine of Critical Cleaning Technology"
Parts Cleaning

Ultrasonic and Megasonic Theory and Experimentation

by: Ahmed A. Busnaina, Glenn W. Gale, and Ismail I. Kashkoush
Pages: 13 - 19; April, 1994

Surface cleaning using ultrasound in liquids is a widespread practice which has been used to clean diverse parts and assemblies in the aerospace, optical, surgical, and electronics industries.1,2 Contaminants that can be removed include particles machining chips, grease, oils, rust, dyes, and pigments.3

The cleaning ability of ultrasonics has been attributed to various combinations of cavitation and acoustic streaming. While effective in cleaning many parts, ultrasound at less than about 100 MHz is also capable of inflicting surface damage under certain conditions.4,5

It has been shown3 that materials having high surface energy and high mechanical hardness - those most resistant to damage - are good candidates for ultrasonic cleaning.

Erosion results from cavitation implosion. For critical silicon surfaces used in the manufacture of ULSI circuits, even small-scale damage is unacceptable. To mitigate surface damage, ultrasound near 1 MHz (referred to as megasonics) is commonly used in semiconductor manufacturing. Such techniques have been studied recently.6-8


Ultrasonic Theory

A typical ultrasonic source is a plane surface oscillating at a single frequency, producing a longitudinal wave. Vibrational energy transmitted subsequently propagates through the fluid.

At each distance x along an axis perpendicular to the source, the displacement will be different than at other points, since a finite time is required for energy transmission. Pressure and velocity at a point are sinusoid functions of time and the angular frequency (equal to 2p f where f is the frequency in Hz).

The velocity of the wave, or speed of sound, depends on the fluid. At room temperature, the sound speed in water is about 1500 m/s. The wave period is T=1/f; the wavelength is l=cT. Loss of energy to the medium, called attenuation, is a function of a , the absorption coefficient.9



Cavitation is the formation and collapse of either gas or vapor bubbles in a liquid subjected to pressure changes. The formation of cavities in liquids is analogous to tensile failure in solids. When the "tensile strength" of a liquid is exceeded, cavities form.10

Actual values of these "strengths" are much lower than theoretical values, as a result of imperfections (gas pockets) in the liquid which serve as nuclei for cavitation.11 These nuclei grow, through net diffusion of dissolved gas from the liquid to the nuclei, to form cavitational bubbles.

When a high enough pressure amplitude (cavitation threshold) is reached, the nucleus becomes unstable and rapidly grows into a mostly vapor-filled bubble or transient cavity. The cavitation threshold has been studied for water as a function of various liquid properties.10

Cavitation threshold has been found to decrease with increasing dissolved gas tension and surface tension, increase with increasing hydrostatic pressure, decrease with increasing temperature, and decrease with increasing number of solid contaminants. It also generally increases with increasing dissolved ion concentration.10

In sonic cleaning, the important aspects of cavitation are its effect on particle removal and its infliction of surface damage. The conditions under which each of these occurs must be known and understood so that the former may be exploited and the latter avoided.

While the study of surface damage due to cavitation is relatively straightforward, the study of cavitation-induced particle removal is difficult, since acoustic steaming is also a removal mechanism. Whether removal is a result of cavitation or streaming or both) is not always clear. It is useful, to know at what frequencies and acoustic intensities cavitation will occur in a cleaning bath.


Sonic Frequencies

The existence of cavitation at low frequencies, up to 100 kHz, is well known. Shwartzman et. al.12 concluded from their pioneering work in megasonic cleaning that, in the 850 to 900 kHz range, there is insufficient time between pulses to allow the formation of cavitation bubbles.

Studies have confirmed that he cavitation threshold pressure increases considerably with frequency.13 Noltingk and Neppiras, in a review paper on cavitation, asserted that increasing the frequency can decrease the intensity of the cavitation to zero.14

Numerous experiments for determining cavitation thresholds in water have been reported.15,16-18

Esche18 studied cavitation over a range of frequencies and determined cavitation threshold pressure amplitudes for both aerated and degassed water. At 40 kHz, a typical ultrasonic cleaning frequency, the threshold according to Esche's curves is on the order of one atmosphere. At 850 kHz, at typical "megasonic" cleaning frequency, his data indicate a threshold in excess of 100 atmospheres.

In contrast, other data15 indicate a threshold of around 10 atmospheres at this frequency. The discrepancies result from extreme sensitivity to experimental conditions. Furthermore, different experimenters employ different criteria for determining the existence of cavitation, and have varying limits as to the smallest scale on which cavitation can be detected.

In fact, sound frequency does not affect cavitation threshold so much as it affects the maximum radius to which the cavities can grow.19 In some cases then, bubbles may exist but can be undetectably small.


Damage Control

Cavitation erosion of surfaces is likewise dependent on the conditions of sonication. Hile low gas content increases the cavitation threshold pressure, it will also increase damage since those cavities which do form collapse more violently in the absence of cushioning gas.20

Experiments have shown that damage can be eliminated, even at low frequencies, by increasing gas content in the liquid.4 Plesset21 suggested liquid jets emanating from collapsing bubbles as a potential cause of surface damage.

Indeed, jet velocities exceeding 100 m/sec have been theoretically calculated and experimentally measured.4 Experiments22 have shown that erosion is caused by these jets rather than from the extreme pressures and temperatures which also result from cavity collapse.

It has been reported that the amount of cavitation damage (expressed a weight loss from the surface) increases as the square of he radiated sound pressure amplitude.20

Noltingk and Neppiras14 suggested in their theoretical study that as acoustic pressure is increased, the violence of the cavity collapse will increase to a maximum value and the decrease. Their reasoning was that he maximum bubble radius will increase with pressure amplitude, also increasing the time (T) required for collapse.

Eventually, though, this collapse time would exceed one half of the sound field period (T/2). The collapse would be occurring in the half-period during which the pressure is increasing from its most negative value. However, if 2T/T is greater than unity, the sound pressure would become negative again before the collapse would be completed. The intensity of the collapse would thus be lessened.19 Experiments have supported this theory.20

Even very small-scale cavitation damage is unacceptable for silicon wafers. As a result, higher (megasonic) frequencies are used in order to mitigate cavitation. Another powerful cleaning mechanism is then responsible for particle removal


Acoustic Steaming

Despite the sinusoidal nature of sound waves, the particle velocities they induce are not strictly sinusoidal. Early researchers in sound theory noted the motion of dust particles in the air in the presence of sound.23 Patterns of circulation not time-dependent were observed.

When the Navier-Stokes equations governing the flow are solved,24 there indeed exists a time-independent component of velocity in addition to the oscillating component. The streaming component results from attentuation of he wave.25 Patterns are complex and geometry-dependent.

Different types of streaming exist, but what's most important in particle removal is the streaming activity close to a surface. So-called Schlichting streaming occurs in a viscous boundary layer in a sound field.26 This streaming produces vortices of a scale much smaller than the acoustic wave length. Velocity gradients are large, and transport is enhanced due to this streaming.

Increasing frequency significantly decrease the thickness of the acoustic boundary layer, faiclitating transport of particles away from the surface. For instance, increasing the frequency form 40 kHz to 900 kHz decreases the acoustic boundary layer thickness from 2.82 micron to 0.594 micron.

Also of significance is the microsteaming which occurs near small bubbles clinging to a surface (as they commonly do). The resulting eddies can be highly localized. Microstreaming has been shown to produce very high velocities and significantly accelerated reaction rates. Removal of surface contaminant films by microstreaming has been observed.26,27

Unlike cavitation, streaming is significant at all frequencies. Experiments in photographic development have shown that acceleration of that process is due to small-scale acoustic streaming and not cavitation.28 Streaming has been considered to the primary particle-removal mechanism in sonic cleaning.26 Drag forces imparted by the fluid motion can detach particle from the surface, after which they are swept away by the flow.



Testing the removal of submicron particles from silicon surfaces using ultrasonic and megasonic cleaning processes explored the efffcts of frequency, irradiation time, particle size, and particle compostiion. All experiments wee performed in the Class M2.5 (10) cleanroom of he Microcontamination Research Laboratory at Clarkson University (Potsdam, NY)

Particle removal efficiency was studied at frequencies of 40, 65, 80, and 850 kHz. The 850 kHz experiments wee performed using a commercially available megasonic cleaning system (nominal frequency 862 kHz an maximum power input 150 W), while commercial ultrasonic tanks and generators (40, 65, 80, and 100 kHz) wee use din the remaining experiments.

Silicon wafers used wee 125 mm p-type (100). Of 11 to 18 W -cm resistivity. The wafers were cleaned prior to deposition using an SC1 (1 NH4OH : 1 H2 O2 : 5 H2O) bath and scanned by a laser surface scanner (with 0.1 micron resolution) to establish a background particle count. This number of particle of unknown origin was subtracted from the total number of particles on the wafer before and after sonic cleaning.

Particles used were PSL spheres, SiO2 (silica) spheres, and non spherical Si3N4 (silicon nitride) particles. Mean diameters of 0.3, 0.4, 0.5, 0.6, 0.7, and 1.0 micron were employed. These particles, originally in a concentrated high-purity aqueous solution, were mixed with isopropanol to form a dilute solution, eliminating the problem of particle agglomeration.

The resulting suspension was atomized using a nebulizer with 0.1 micron filtered air, and deposited onto the wafers.

Approximately 150 to 300 particles were deposited on each wafer, resulting in relatively low pre-cleaning particle counts (1.1 to 2.4 particles/cm2).

Wafers wee tan loaded into a 25-wafer PFA teflon cassette, which was inserted vertically into the tank. After the required immersion time in deionized (DI) water, the cassette was removed from the tank. Wafers were rinsed in DI water, dried, the re-scanned to obtain the post-clean particle count .

A removal efficiency was computed as follows:

l (%)=Nbefore-Nafter x 100



In which Nbefore is the number of particles deposited on the water surface prior to sonic cleaning, and Nafter is the number of particles remaining on the surface after cleaning.

For any particular operating condition, 10 experiments were run, 10 removal efficiency values were measured, and their average was calculated.


Results and Discussion

Effect of Frequency

Table 1 shows removal efficiencies for various PSL sphere sizes in DI water at each frequency tested. Efficiency is expected to increase with frequency, as indeed it does with the exception of 80 and 100 kHz.

The reason fir the somewhat low efficiencies at these frequencies is that the experiments were performed using a transducer having a natural frequency of 65 kHz. Thus the transmission of energy was less efficient that it would have been had transducers of 80 and 100 kHz natural frequencies been used in the appropriate tanks.

Acoustic streaming velocity increases with increasing frequency. The vortex nature of streaming n the boundary layer results in high velocities close to the surface and efficient transfer from the surface to the free stream.

Also, the thickness of the acoustic boundary layer, given by (2v/w)1/2, decreases with increasing frequency. At. 40 kHz, the boundary layer thickness in water is about 2.82 micron, while at 0.9 MHz its only about 0.594 micron. The thinning of the boundary layer facilitates transport of contaminants into the free stream.


Effect of Irradiation Time

Removal efficiencies for PSL spheres in DI water as a function of time are shown in Figures 1 through 5. The data in Figures 1 through 4 were taken using ultrasonic tanks at 400 W, while Figure 5 data were taken in the megasonic tank at 150 W. Figures 1 through 4 show removal efficency at 40, 65, 80, and 100 kHz, respectively. Figure 5 shows removal efficiency at 862 kHz.

In all cases, removal increases with time until a maximum efficiency is reached, after which there is either no improvement or a slight degradation in cleaning. In the low-frequency case, the maximum occurs between 20 and 25 minutes, whereas he megasonic case shows no improvement after 15 minutes.

The improvement in cleaning with time is consistent with other reults,4,6,7,29-31 A drop in efficiency after 20 to 25 minutes may be due to particles redepositing on the surface.26


Effect of Particle Size

All data indicate that removal efficiency decreases with decreasing particle size. Figure 6 show this trend explicitly for megasonic removal of PSL spheres at different irradiation times. This result is expected due to the relationship between particle adhesion and removal forces as particle size becomes smaller

The dominant adhesion mechanism in liquid media is the long-range van der Waals force, the magnitude of which -- for a sphere adhering to a flat substrate - is given by

Fvdw=- Ar


where r is the sphere radius, z is the distance of separation between sphere and substrate (typically taken to be 4 Angstroms), and A is the Hamaker constant (a property of the materials involved). If two substances have Hamaker constants A11 and A22, the Hamaker constant between them is given by

A12= A11A22

If the two substances are immersed in a medium (material 3), then the Hamaker constant for the system is given by32


The constant c is about 1.5 to 1.6 for water. Hamaker constants and associated adhesion forces have been calculated for various particles adhering to silicon in water.33 The relationship between van der Waals force and particle diameter means that the adhesion force will decrease linearly with decreasing particle size.

In most particle removal techniques using forces such as hydrodynamic drag and/or centrifugal, however, removal forces are proportional to the second or third power of particle radius and therefore decrease at a high rate.34 Thus particles of smaller sizes are more difficult to remove.


Effect of Particle Composition

Figures 7 and 8 show the removal efficiencies in DI water of SiO2 and Si3N4 particles, respectively. Comparison of these figures with Figure 5, showing the removal efficiencies of PSL spheres under the same conditions, indicates the effect of particle composition. Since aplied removal forces are the same for each particle type, any differnces in removal efficiency can be attributed to different adhesion forces.

In the case of PSL spheres and silica spheres, the Hamaker constant is slightly lower removal efficiency for silica compared with PSL, the efficiencies are in fact comparable. The small discrepancy could be due to electrical double-layer effects (discussed below) or deformation of he PSl spheres, which has been shown to increase adhesion.33

The removal efficiency of Si3N4, however, is significantly lower - attributable to the nonspherical nature of the silicon nitride particles. Their shape facilitates greater numbers of contact points with the substrate, increasing the adhesion force.

Electrokinetics may also be a factor in the case of silicon nitride. Solid surfaces in liquid media form a layer of charge through absorption of ions of dissociation of surface groups. The tow layers are collectively called the "electrical double-layer," and the boundary between them is the shear plane.34 The electrical potential at the shear plane, defined as the zeta potential, determines whether a particle will be attracted or repelled by another charged surface in the fluid if their electrical double-layers overlap.

At the pH of water, SiO2 has a negative zeta potential (as does the wafer surface) while Si3N4 has a positive zeta potential.35 Thus the electrical double-layer repulsion whch exists for silica particles is not present with silicon nitride, and this may enhance net removal of SiO 2 relative to Si3N4.



Call Precision Cleaning at (908) 788-0343 for references.



This work was funded through the Center for Particulate Control in Process Equipment at Clarkson University. Funding for the Center is provided by Asyst Technologies, Clestra Cleanroom Technology, IBM Corp., Praxair, and Verteq. The use of equipment from Crest Ultrasonics and Verteq is also gratefully acknowledged.


About the Authors

Ahmed A. Busnaina , Ph.D., is an Associate Professor of Mechanical and Aeronautical Engineering, and director of the Microcontamination Laboratory at Clarkson University. He is the author of more than 130 publications and presentations on contamination control, computational fluid mechanics, environmental flows, and combustion.

Glenn W. Gale has been employed by IBM Corp. (Essex Junction VT) since 1988, with responsibilities in the areas of wet chemical equipment and process engineering . He received his M.S. and is currently pursuing a Ph.D. in mechanical engineering from Clarkson University. He has authored several publications concerning semiconductor wafer cleaning.

Ismail I. Kashkoush, Ph.D., process engineer for SubMicron Systems (Allentown, PA), is responsible for the firm's Class Mi.5 (1) applications laboratory. Specializing in microcontamination characterization, removal, and control, he has more than 25 publications and presentations to his credit in the area of wafer wet chemical cleaning, particularly using ultrasonics and megasonics.

Copyright 1999, Witter Publishing Corporation 84 Park Avenue, Flemington, NJ 08822 Phone: 908-788-0343 Fax: 908-788-3782
Email: PrecisionCleaning@WitterPublishing.com Please e-mail comments and questions to: mailto:webmaster@witterpublishing.com?subject=[www.PrecisionCleaningWeb.com]
Reposted with permission of Precision Cleaning.