Surface cleaning using energetic microcluster beams
07/01/1998
Surface cleaning using energetic microcluster beams
John F. Mahoney, Julius Perel, Carl Sujo, Phrasor Scientific Inc., Duarte, California
James C. Andersen, Applied Precision Inc., Issaquah, Washington
A new in situ vacuum surface-cleaning process uses microcluster beams formed by electrohydrodynamic emission from capillaries to clean surfaces contaminated by solid particulates and organic films. Proposed surface-cleaning mechanisms include a model based on transfer of impulsive forces during collisions between microclusters and solid particles. Microschocks induced in thin contaminant films during microcluster impact are also discussed.
More than 80% of the yield loss of volume-manufactured VLSIs has been attributed to particulate microcontamination [1]. As device geometries continue to shrink and wafer sizes increase, particulates and airborne and other trace contaminants will have an ever-increasing impact on device yields. Current cleaning technologies become less effective with the growing demand for removing submicron (=0.1 ?m) contaminants. New technologies will be required to clean wafer surfaces to meet national goals for producing 0.07-?m feature sizes by the year 2006, 0.05 ?m by 2009, and 0.035 ?m by 2012 [2].
Although great strides have been made in ex situ wafer cleaning technologies, contaminants generated within process cluster tool equipment can be a major source of yield loss. There is a critical need for in situ removal of contaminants arising from mechanical operations such as wafer handling, transport, shaft and disc rotations, pumping and venting, as well as chemical and physical reactions in process tools. Technology development for cleaning wafers in situ has not been the focus of concerted research efforts [3]. At present, wet chemistry dominates silicon wafer-cleaning technology [4]. Continued use of wet chemistry in the future will be tempered by conservation and environmental issues. With some wafer fabs using 1 billion gallons of water/year, the worldwide cost of ultrapure water for semiconductor production is projected to reach nearly $3 billion by the year 2000 [5]. For semiconductor device manufacturing to remain cost-effective, the need to conserve water and lower water costs is acute and new technologies for in situ cleaning of wafers are necessary. A successful cleaning technology should offer the following economic benefits: higher device yields; lower water costs; reduction of on-line contaminant diagnostic devices; and cost savings from a new generation of cleaner process tools.
What the industry needs is a fully automated "dry" system capable of being integrated into on-line fab device-processing steps [6]. To meet these requirements, a microcluster-beam-cleaning technology is under investigation for simultaneous in situ removal of particulates, organic films, and other contaminants [7]. An energetic beam of microclusters is directed at a target substrate. Upon impact, a large number of microcluster collisions with the surface remove both particulate and film contaminants. Depending on the type of contaminant removed in microcluster-surface interactions, the physical mechanisms responsible for cleaning may differ. Microclusters transmit impulsive forces to particulates sufficient to overcome their van der Waals adhesion forces. Film contaminants can also be removed via a shock wave mechanism.
Microcluster beam technology
Microcluster beams are generated by atomizing a conducting liquid exposed to high electric fields from the tips of capillary emitters (Fig. 1). A conducting liquid (typically a solvent mixture consisting of water and n-methyl-2-pyrrolidone) is fed at atmospheric pressure to a capillary tip mounted in vacuum. Atomization occurs when positive voltage is applied to the capillary, preferentially pulling positive charges in the solution to the surface of the liquid exposed to the high electric field, and attracting negative ions to the walls of the metal capillary where they transfer conducting electrons. When the electrostatic stress acting on the liquid-vacuum interface exceeds the solvent surface tension, the meniscus disrupts, forming a divergent beam of positively charged, energetic microclusters. To produce electrostatic atomization, the required fields are typically in excess of 105 V/cm. Figure 2 shows the qualitative dependence of microcluster properties (size, velocity, and impact energy) on process control variables (e.g., applied emitter voltage and solvent flow rates). For a given emitter voltage, microcluster sizes will increase if the solvent flow rate increases. If the solution flow rate remains unchanged, the size of microclusters will decrease by raising the applied emitter voltage. For surface cleaning, we prefer to generate microclusters with diameters ranging from about 0.01-1 ?m.
 |
Figure 1. Formation of a microcluster beam by electrostatic atomization.
 |
Figure 2. Microcluster properties as a function of process control variables.
Figure 3 shows a microcluster emitter source design used for surface-cleaning studies. This source performs adequately at vacuum levels ranging from about 1 mtorr down to 10-5 torr. A fused silica transfer line delivers a solution, stored in a reservoir, to the tip of a metal capillary assembly centered within a circular aperture machined in the extractor electrode. The inner diameter of the fused silica transfer line is 50 ?m and the outside diameter is about 375 ?m. The solution reservoir, exposed to constant atmospheric pressure, is located outside the vacuum housing enclosing the substrate to be cleaned. To control the solution flow rate, we selected the length and inside diameter of the fused silica line to provide the proper impedance to flow - usually in the range of 0.5-2 ?liters/min. A neutralizer assembly, consisting of a thermionic emitter and shield, can be installed on the microcluster beam source if necessary. The neutralizer injects low-energy electrons into the beam to compensate for charge build-up on insulating surfaces. Charge builds up from transporting positive charge or the ejection of secondary electrons from poorly conducting surfaces bombarded by microclusters.
 |
Figure 3. Microcluster beam emission source.
Mechanisms for contaminant removal
We propose a mechanism for particle removal whereby liftoff occurs when the impulsive force transmitted by an incident microcluster is sufficient to overcome van der Waals adhesion forces. We will examine this hypothesis by comparing the impinging forces to the van der Waals adhesion forces bonding spherical particles to a flat surface (Fig. 4). Consider the collision of a microcluster with a surface particle. The impulse (I) given to the particle is
 |
Figure 4. Particle removal by microcluster impact.
I = Dp = pi - pf = FI Dt (1)
where pi and pf are the initial and final momentum states, FI is the transmitted impulsive force, and Dt is the collision time. Neglecting momentum enhancement due to microcluster or particle material ejected backward ( i.e., pf = 0) simplifies the analysis. The impulsive force acting on the particle during a head-on collision (microcluster diameter less than the particle diameter) in time Dt is,
FI = mv/Dt(2)
where m and v are the mass and velocity of the incident microcluster. Assuming the collisional interaction time approximates the time it takes a cluster with velocity v to traverse a distance 2r (microcluster diameter), i.e., Dt ~2r/v, the impulsive force becomes,
FI = mv2 /2r (3)
or, since mv2 /2 = qVa, where q is the net charge carried by the microcluster and Va is the applied emitter voltage, Eqn. 3 can be expressed in the form
FI = qVa
(4)
which states that the impulsive force is equal to the microcluster impact energy divided by its radius. The microcluster charge q is related to its radius r by [8]
q = 4p (e0g)1/2 r3/2 (5)
where e0 is the permittivity of vacuum and g is the surface tension.
Substitution of Eqn. 5 into Eqn. 4 yields the following expression for the impulsive force,
FI = 4p (e0g)1/2 r1/2 Va (6)
Assuming clusters are sprayed from a solution with a surface tension (g) of 0.05 N/m and accelerated through a potential (Va) of 10 kV (typical values for some cleaning applications), the impulsive force given by Eqn. 6 reduces to
FI ~8.4 r1/2 (dynes) (7)
for r in ?m. This equation allows one to estimate the impulsive force exerted on surface particles as a function of the size of the incident cluster.
A theoretical analysis [9] for approximating van der Waals adhesive forces between spherical particles and a flat surface yields
Fv ~hr/8pz2 (8)
where h is the material dependent (particle and surface) Lifshitz-van der Waals constant and z is the adhesion distance assumed to be about 4 ?. Taking h = 8.0 eV (= 1.28 ? 1011 erg), Eqn. 8 reduces to
Fv ~0.032 r (dynes)(9)
for r in microns.
If the impulsive force generated by a single microcluster impact exceeds the van der Waals adhesive force ( i.e., FI >Fv ), a high probability exists for the force to dislodge a contaminant particle from the surface. The table compares Fv to FI for microclusters with contaminant particles of equivalent dimensions.
 |
Provided that contaminant particles have resided on surfaces for short times, microcluster impact forces can be significantly larger than the van der Waals adhesive forces. However, particles existing on surfaces for long periods tend to increase their adhesive surface area, leading to stronger van der Waals forces [9].
When microclusters are accelerated to hypervelocities, weak microshocks generated in the material can remove thin films (e.g., organic or oxide contaminants). The velocity of sound in SiO2 is 5.75 km/sec, while that for many polymeric materials lies between 1 and 3 km/sec. To prevent surface damage or "hazing" of a SiO2 wafer surface during particle or film removal, we project maximum microcluster velocities of ~5 km/sec. Shock waves propagated in thin films can lead to unloading of material after passage of decompression waves (Fig. 5). Local shock pressures generated by hypervelocity impacts can be roughly estimated by the Bernoulli pressure term:
 |
Figure 5. Film contaminant removed by shock unloading: a) hypervelocity impact and b) material ejected by shock unloading.
P = rv2/2 (10)
where r and v are the density and velocity of the incident microcluster [10]. Using this expression, shock pressures are estimated to exceed 1 GPa for impacts between microclusters and surfaces.
Particle removal feasibility investigations
Feasibility studies undertaken to demonstrate particle removal from a surface by energetic microcluster impacts used a stainless steel substrate seeded with monosized 1-?m polystyrene spheres. The substrate was mounted in a vacuum test chamber (3 ? 10-5 torr) and irradiated, using a single capillary emitter, with a 5-?A glycerol microcluster beam formed at +15 kV. Figures 6 and 7 show scanning electron microscope (SEM) photographs of the seeded substrate before and after exposure to the microcluster beam. Figure 7 shows that most of the seeded particles were removed from the substrate after exposure. Handling and transport of the substrate before and after the cleaning test were not conducted in a cleanroom environment. Although preliminary, these results demonstrate the surface-cleaning potential of microcluster impacts. Investigations are now underway to determine the optimal conditions for cleaning silicon wafer surfaces by studying multiple emitter array configurations, microcluster impact energy and size, substrate orientation, and target exposure times.
 |
Figure 6. SEM photograph of stainless steel substrate seeded with 1-?m polystyrene spheres before cleaning.
 |
Figure 7. SEM photograph of a stainless steel substrate after exposure to glycerol microcluster beam.
Comments on semiconductor wafer-cleaning applications
By expanding research and development efforts, eventual application of microcluster beam surface cleaning for semiconductor wafer conditioning has breakthrough technology potential. Compared to other beam technologies, including laser or cryogenic jet sprays, microcluster beam surface cleaning offers several advantages (see "Wafer probe card cleaning"on page 154). The process is compatible with in situ vacuum cleaning of silicon wafers or other surfaces. It contains no moving parts and consumes ultralow quantities of cleaning solution - potentially, tens of microliters/wafer. Since microclusters are formed by electrostatic atomization of water/solvent compositions, no hazardous chemicals are introduced into the surface-cleaning process. During operation, the microcluster-cleaning solution is not exposed to valves, fittings, seals, connectors, or extended lengths of plastic or metal piping, thus minimizing the introduction of secondary contaminants into the cleaning process. Most important, microcluster dimensions are of the same order as submicron particle contaminants and microelectronic device features. This latter property allows microclusters to remove contaminants trapped in crevices, e.g., organic residues trapped in vias. The capability of microcluster beam technology to remove metallic ion impurities has not been investigated, but should be examined in the future.
Wafer probe card cleaning
Semiconductor wafer cleaning is one of many potential uses for microcluster beam technology. Other opportunities include: optical surface cleaning; substrate conditioning prior to film deposition; glass slide preparation for gene chip research; and cleaning MEMS devices, computer disc drives, electrical pins and sockets, and microelectronic photomasks. Wafer probe card cleaning is one application being pursued by Applied Precision Inc. Wafer probe cards, circuit boards with traces connected to "probe needles," are used for testing semiconductor die in wafer form. Conventional probe needles are of the "cantilever beam" type constructed from drawn tungsten wire. Typical probe tip diameters range from 20-40 ?m , spaced as close as 50 ?m to each other. Low electrical resistance (1 W) and positional accuracy (?10 ?m) are critical to successful electrical testing of the die. Testing die in wafer form allows the manufacturer to determine the performance of a chip before going through considerable packaging expense.
The probe card is mounted to the prober and the wafer is aligned to the probe tips of the probe card. The wafer is then brought into contact with the probe tips, which scrub into the aluminum bond pads of the die. The tips break through the passivation layer and make electrical contact with the bond pads. This is repeated for each die and over time the tips accumulate aluminum from the bond pads. Oxides, acting as insulators, form at the probe needle tips, making electrical testing of the die inaccurate and unreliable. This problem not only contributes to prober/tester downtime but can reduce device yield as well. Prober or tester downtime can cost $0.25/sec ($900/hr) and a single wafer discarded due to faulty testing can represent a loss of $50,000. Increase in contact resistance, as a result of probing aluminum bond pads, can be significantly reduced by applying microcluster-beam-cleaning technology. The reduction in contact resistance is attributed to the microcluster removal of the oxide films.
The accepted method for reducing contact resistance uses abrasive sanding of the probe tips. For some probing techniques, abrasive sanding is not an option because the materials and designs cannot withstand the severe side-loading or deformation of the contact tip topography associated with sanding. Although effective for reducing contact resistance, this method can have several negative effects. Whenever probe tips are manipulated, there is risk of causing misalignment and damage. The abrasive media and the probe needle material removed by sanding can adhere to the probe needles and deposit on the wafer during probing. As the tapered shafts of the probe needles are sanded down, the tip diameters enlarge and the needle lengths decrease, effectively reducing the life of the probe card. Extending useful card life and minimizing tip damage are of obvious concern.
Research indicates that microcluster beam technology offers the positive results of abrasive cleaning without its negative effects. The figure illustrates that microcluster beam technology can reduce contact resistance (Rcontact) as effectively and consistently as conventional methods. In almost all cases, previously high probe needle Rcontact values were reduced below 1 W after microcluster beam cleaning.
 |
Contact resistance reduction via microcluster beam cleaning.
Microcluster beams are a noncontact, nondestructive method for reducing the contact resistance of probe tips. They have no effect on the planarity or alignment of probe needles, and original tip diameters and lengths are preserved. Materials and topographies are inconsequential when applying microcluster beam technology, offering a method for reducing probe contact resistance that would not be not possible using conventional methods.
As microelectronic features continue to shrink, the disadvantages of wet chemistry cleaning methods could become more apparent. For example, wet chemistry can be a source of submicron particle contaminants suspended in DI aqueous solutions. Although ppb purity levels are standard specifications now, ppt levels may be required in the future. As chip complexities increase, cleaning steps are predicted to increase also, adding to already high water usage levels. With submicron chip features, it will become increasingly difficult to get liquids in and out of finer geometries. The most limiting aspect of wet chemistry may be its inability to integrate in situ with cluster tools where vacuum or low-pressure processing is required.
References
1. T. Hattori, "Detection and Identification of Particles on Silicon Surfaces," in Particles on Surfaces, ed., K.L. Mittal, Marcel Dekker Inc., p. 201, New York, 1995.
2. National Technology Roadmap for Semiconductors, Semiconductor Industry Association, 1997.
3. R. Blewer, V. Menon, "Trends and Future Requirements in Contamination-free Manufacturing Research," in 1994 Microcontamination Conference Proc., Canon Communications Inc., p.10, 1994.
4. R. Iscoff, "Wafer Cleaning: Wet Methods Still Lead the Pack," Semiconductor International, p. 58, July 1993.
5. M. Lancaster, "Ultrapure Water: The Real Cost," Solid State Technology, p. 70, September 1996.
6. W. Kern, ed., Handbook of Semiconductor Wafer Cleaning Technology, Noyes Publication, New Jersey, 1993.
7. J.F. Mahoney, "Microcluster-surface Interactions: A New Method for Surface Cleaning," Inter. J. Mass Spectrom.Ion Processes, Vol. 174, p. 253, 1998.
8. J.F. Mahoney, J. Perel, C. Sujo, J. Andersen, "Removal of Particulate and Film Contaminants by Impacting Surfaces with Microcluster Beams," paper presented at the 28th Annual Meeting of The Fine Particle Society, Dallas, TX, April 1-3, 1998.
9. R.A. Bowling, "An Analysis of Particle Adhesion on Semiconductor Surfaces," J. Electrochemical Soc., Vol. 132, No. 9, p. 2208, 1985.
10. M.E. Kipp, D.E. Grady, J.W. Swegel, "Numerical and Experimental Studies of High-velocity Impact Fragmentation," Int. J. Impact Engng., Vol. 14, p. 427, 1993.
JOHN F. MAHONEY received his MS degree in physics from the University of Kentucky. He has been involved with ion and charged microcluster source development for more than 20 years, has numerous publications in the fields of charged particle beam sources and applications, and is a member of the American Vacuum Society. Mahoney is VP and senior research physicist at Phrasor Scientific Inc. 1536 Highland Ave., Duarte, CA 91010; ph 626/357-3201, fax 626/357-3203, e-mail [email protected].
JULIUS PEREL received his BS degree in physics from City College, NY, and his PhD in physics from New York University. As president and R&D manager of Phrasor Scientific Inc., he has extensive experience in alkali metal ion sources, molecular ion theory, and collision dynamics. He is author of more than 50 publications in these and related fields.
CARL SUJO received his BS degree in chemical engineering from the University of California, Los Angeles, in 1994. He works as a staff scientist for Phrasor Scientific Inc. and is involved in technical support of microcluster beam technology development.
JAMES C. ANDERSEN attended the University of Washington, majoring in mechanical engineering in 1981. Since 1990, he has worked at Applied Precision Inc. as a design engineer and project manager, developing metrology and test equipment used in the semiconductor industry. He is assigned to the Special Projects R&D Group, which focuses on leading-edge technology to solve semiconductor test problems. Andersen is a member of SEMI, SEMI/SEMATECH, and the AEA.