Basic Mechanisms and Models

REVIEWS
of Multi-Wire Sawing
By Hans Joachim Möller*

More than 80 % of the current solar cell production requires the cutting of large silicon crystals. While
in the last years the cost of solar cell processing and module fabrication could be reduced considerably,
the sawing costs remain high, about 30 % of the wafer production. At present the large crystals are cut
using the multi-wire slicing technology[2] which has the advantage of a high throughput (several
hundred wafers per day and machine), a small kerf loss of about 200 lm and almost no restrictions on
the size of the ingots. Basic knowledge about the microscopic details of the sawing process is required
in order to slice crystals in a controlled way. In the following the principles of the sawing process will
be described in this review article as far as they are understood today.

1. Introduction 2. Multi-Wire Wafering Technique

More than 80 % of the current solar cell production re- After crystal growth the silicon ingots are cut in a first step
quires the cutting of large silicon crystals. Multicrystalline in- by band saws into columns with a cross section that is deter-
gots grown by the Bridgman or gradient freeze technique mined by the final wafer size. Standard sizes are about 10 ”
reach cross sections now of more than 50 ” 50 cm2 and weigh 10 cm2, but larger wafers sizes up to 15 ” 15 cm2 become in-
over 250 kg, monocrystalline Cz-crystals have diameters of creasingly used in the solar cell technology. These columns
up to 20 cm today. While in the last years the cost of solar cell are glued to a substrate holder and placed in a multi-wire
processing and module fabrication could be reduced consid- saw which slices them into the final wafers. The principle of
erably, the sawing costs remain high, about 30 % of the wafer the multi-wire technology is depicted in Figure 1. A single
production.[1] The incentive to optimize the sawing technique wire is fed from a supply spool through a pulley and tension
for further cost reduction in mass production is thus high. At control unit to the four wire guides that are grooved with a
present the large crystals are cut using the multi-wire slicing constant pitch. Multiple strands of a wire net (known as a
technology[2] which has the advantage of a high throughput web) are formed by winding the wire on the wire guides
(several hundred wafers per day and machine), a small kerf through the 500 ± 700 parallel grooves. A take-up spool col-
loss of about 200 lm and almost no restrictions on the size of lects the used wire. The wire is pulled by the torque exerted
the ingots. Currently wafers between 250 and 350 lm are by the main drive and slave as shown. The tension on the
usually cut but a wafer thickness down to about 100 lm can wire is maintained by the feedback control unit at a pre-
be achieved by the technique. The sawing process depends
on several variable parameters as will be described later
which makes it difficult to optimize the process in view of
throughput, material losses, reduction of supply materials ±
and wafer quality. Basic knowledge about the microscopic [*] Prof. H. J. Möller
details of the sawing process is required in order to slice crys- Institute for Experimental Physics
tals in a controlled way. In the following the principles of the Technische Universität Bergakademie Freiberg
sawing process will be described as far as they are under- Silbermannstr. 1, 09599 Freiberg, Germany
stood today. E-mail: [email protected]

ADVANCED ENGINEERING MATERIALS 2004, 6, No. 7 DOI: 10.1002/adem.200400578  2004 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim 501
 Möller/Basic Mechanisms and Models of Multi-Wire Sawing

reduced in the future. The removal of oil from the wafer sur-
REVIEWS

faces requires comprehensive cleaning procedures. Since

large quantities of slurry and SiC are used during sawing, re-
cycling or the disposal of these materials has to be considered
as well. Water-based slurries or slurries which are very
hygroscopic have the problem that hydrogen gas is generated
from the interaction of water and silicon, which can cause the
hazard of explosion. From the environmental point of view
water-washable slurries may be however the choice of the fu-
ture.
Material is continuously removed through the interaction
of the SiC particles below the moving wire and the silicon
surface. The abrasive action of the SiC depends on many fac-
tors such as wire speed, force between wire and crystal, the
solid fraction of SiC in the suspension, the viscosity of the
Fig. 1. Schematic diagram depicting the principle of the multi-wire sawing technique.
The crystal (workpiece) is pushed against the wire web with a constant feeding rate
suspension, the size distribution and the shape of SiC parti-
which is equal then to the sawing rate. The wire bow determines the force on the wire. cles. The viscosity of the slurry depends on the temperature
The wire speed is maintained constant. and the solid fraction of particles. Since the temperature rises
due to the cutting process, the suspension has to be cooled
and the temperature controlled during sawing. The viscosity
scribed value. The silicon column on the holder is pushed
of the slurry also changes because of the continuous abrasion
against the moving wire web and sliced into hundreds of wa-
of silicon and iron from the wire. This gradually deteriorates
fers at the same time. The wire either moves in one direction
the abrasive action and the slurry has to be replaced or mixed
or oscillates back and forth. Solar cell wafers are mainly cut
with new slurry after some time.
by a wire that is moving in one direction whereas wafers for
The kerf loss and surface quality are determined by the di-
the microelectronic industry are cut by oscillating wires. Cut-
ameter of the wire, the size distribution of the SiC particles
ting in one direction allows higher wire speeds between 5 to
and the transverse vibrations of the wire. The amplitude of
20 m/s but yields less planar surfaces. Smoother and more
vibration is mainly sensitive to the tension of the wire, but
even surfaces are obtained by oscillating sawing. Depending
also depends on the damping effect of the slurry. Increasing
on the pulling speed the wires have a length between 150 ±
the tension will reduce the amplitude of vibration, hence the
500 km in order to cut a single column in one run. The wire
kerf loss.[9] Typical wire diameters are around 180 lm and
material is usually stainless steel.
the mean size of active particles can vary between 5 ± 30 lm.
Cutting is achieved by an abrasive slurry which is sup-
This yields kerf losses around 200 ± 250 lm per wafer.
plied through nozzles over the wire web and carried by the
The objective of efficient sawing is to slice with a high
wire into the sawing channel. The slurry consists of a suspen-
throughput, with a minimum loss of slurry and silicon, and
sion of hard grinding particles. Today SiC and diamond are
with a high quality of the resulting wafers. Since many pa-
the most commonly used abrasives. Both materials are very
rameters can be changed the optimization of sawing is a diffi-
expensive and account for 25 ± 35 % of the total slicing cost.
cult task and today mainly done by the wafer manufacturers.
The volume fraction of solid SiC particles can vary between
They are mostly guided by experience. In the following the
30 ± 60 % and the mean grain size between 5 ± 30 lm. For pol-
main results of investigations are summarized which describe
ishing smaller grain sizes below 1 lm are used. The main
the current understanding of the microscopic details of the
purpose of the slurry is to transport the abrasive particles into
wire sawing and yield some guidelines to optimize the pro-
the sawing channels and to the crystal surface. It also has to
cess.
keep the particles apart and must prevent their agglomera-
tion. The entry of the slurry is a result of the interaction be-
tween the wire and the highly viscous slurry. Normally only
a small amount of slurry enters the cutting zone. The factors
that are important here are the viscosity and the wire speed,
but a complex physical modeling is required to understand
the fluid mechanical problems that are involved. First at-
tempts of a description have been reported recently.[3±8]
Most of the commercial slurries are based on oil, water-
based or water-washable slurries based on ethylene glycol.
Oil slurries have several drawbacks. The wafers can stick to-
Fig. 2. Schematic diagram of wire, slurry with abrasive, and crystal in the cutting
gether and are difficult to separate after sawing. This problem zone. Under external force the wire bows and exerts forces on the particles and the slur-
becomes even more severe when the wafer thickness will be ry.

502  2004 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim http://www.aem-journal.de ADVANCED ENGINEERING MATERIALS 2004, 6, No. 7
Möller/Basic Mechanisms and Models of Multi-Wire Sawing

3. Basic Sawing Mechanism case here. The ªscratch-indentingº process occurs for instance

REVIEWS
when the abrasive grains are fixed at the surface of the wire
The main results of the experimental investigations are or of a cutting blade (e. g. for ID saws). A review of material
summarized here and describe the current understanding of removal mechanisms in FAM can be found in the litera-
the microscopic details of the wire sawing process. Figure 2 ture.[13,14]
shows schematically the situation in the sawing channel. The
space between wire and crystal surface is filled with slurry
and abrasive particles. The crystal is pushed against the wire 3.1. Micro-Indentation Experiments
web and causes a bow of the wire. The bow angle varies typi- The individual process of the interaction of a single parti-
cally between 1 ± 5. The resulting pressure of the wire varies cle with sharp edges and the surface of a brittle material has
along the contact area. The forces are maximum directly been studied by micro-indentation experiments. This is
below the wire and decrease towards the side faces (Fig. 3). shown in Figure 5 for a silicon surface. The damage structure
Because of the observed transverse vibrations of the wire ad- of several overlapping micro-indentations with a Vickers dia-
ditional forces may be exerted sideward. The cutting process mond indenter resembles the structure of an as-cut wafer.
at the side faces is important because it determines the final Numerous micro-indentation experiments on monocrystal-
surface quality of the sliced wafers. line silicon and other materials have been carried out in the
The interaction between the abrasive SiC particles and the past to investigate the damage structure quantitatively.[15±19]
crystal yields a distinct damage pattern on the surface that The main results are summarized schematically in Fig. 6 for a
can be analyzed by microscopic techniques. A typical surface ªsharpº Vickers indenter with a pyramid geometry. Loading
structure as seen under an optical microscope is shown in by sharp indenters first leads to the generation of a remnant
Figure 4. Similar structures are obtained along the entire con- plastic impression in the surface known as the elastic ± plastic
tact zone, which shows that the abrasive process is the same zone. Recent Raman investigations of this region have shown
in all directions. that under high pressures the silicon lattice transforms into
The surface structure consists of local indentations with a other crystal structures. Several phase changes have been ob-
mean diameter of a few micrometers. Such a uniform struc- served directly under the indenter, in particular a metallic
ture can be explained by the interaction of loose, rolling parti- high pressure phase.[20,21] Under loading at 11.8 GPa an en-
cles that are randomly indented into the crystal surface until dothermic transformation to metallic silicon (Si II) occurs
small silicon pieces are chipped away. Since SiC particles are (DG = 38 kJ/mol) which partly transforms back to another
facetted and contain sharp edges and tips, they can exert very high pressure phase (Si III at 9 GPa, (DG = -8.3 kJ/mol). In
high local pressures on the surface. This ªrolling ± indenting the metallic state the silicon can deform plastically and mate-
grainº model forms the physical basis of the wire sawing pro- rial can be removed by processes known for ductile metals.
cess. Similar surface structures also occur after lapping sur- This is however a slow but moderate process.
faces of brittle materials with loose abrasive particles.[10±12] In This load regime will be considered as the ªpolishing re-
contrast to the so called free abrasive machining (FAM) pro- gimeº contrary to the ªslicing regimeº for higher loads that
cess sawing with fixed grains yields a different surface pat- will be described next.
tern consisting of long scratches which is obviously not the With increasing pressure the material begins to break and
cracks are generated parallel to the load axis emanating from
the plastic zone. Median cracks are generated beneath the
plastic zone, where the tensile stresses are maximum, in the
form of full or truncated circles. At a critical size they become
unstable and extend towards the surface. In addition, shallow
radial cracks may be generated at the edges of the plastic
zone. Both radial and median cracks coalesce to form half-
penny shaped cracks which are visible at the surface (Fig. 5).
Upon unloading residual stresses from the elastic-plastic
zone can lead to lateral cracks parallel to the surface. When
these lateral cracks reach the surface material is chipped
away. This is the main process for material removal during
sawing. Chipping requires a certain minimum load to occur
(chipping threshold). When material is removed by chipping
only the median and radial cracks remain. This crack system
reaches into the surface and is part of the saw damage which
Fig. 3. Cross section of wire, slurry with abrasive, and crystal in the cutting zone. has to be removed for further processing of the wafers.

ADVANCED ENGINEERING MATERIALS 2004, 6, No. 7 http://www.aem-journal.de  2004 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim 503
 Möller/Basic Mechanisms and Models of Multi-Wire Sawing

pects of the process shall be described first. Further experi-

REVIEWS

mental and theoretical investigations are certainly necessary

to understand all details of the process.
The wire sawing process based on the rolling-indenting
model is shown schematically in Figure 7. It can be divided
in two different steps: the interaction of an individual particle
with the crystal surface from the viewpoint of the fracture
mechanics and the dynamical behavior of the abrasive parti-
cles in the moving slurry between wire and workpiece. Both
processes have to be combined to obtain a complete quantita-
tive description of the material removal rate. In the following
the first process shall be analyzed first because it can be com-
pared with a large number of experimental investigations of
indentation experiments on brittle materials.

3.3. Single Particle Interaction and Fracture Mechanics

As shown in Figure 2 free abrasive particles in the slurry

slide, roll and indent into the workpiece under the action of
the slurry and the wire which moves with the velocity v. The
driving forces for the motion of the particles can be divided
in two parts: the tangential force and the hydrodynamic pres-
sure that are due to the laminar flow of the slurry invoked by
the moving wire, and the normal force on the particle due to
the direct contact to the wire.
The abrasive process results from the indentation of the
particles into the surface under the force FN normal to the
surface. For brittle materials the formation of the plastic zone
and the different crack systems under the normal force can be
related to the hardness H and fracture toughness KIC. From
Fig. 4. Optical micrograph of the surface of an as-cut silicon wafer (left) and below the
wire in the sawing channel (right) showing the same damage structure.
standard indentation measurements the following relations
have been found[15]

FN FN
H ˆ a KIC ˆ b (1a,b)
a2 c3=2
3.2. Quantitative Model of Sawing
where c is the crack length of the half-penny shaped radial-
A first semi-quantitative model for slicing by the scratch- median crack system and a the diagonal of the square impres-
indenting process has been developed by Ermer et al. for sion of the plastic zone (Figs. 5,6). a and b depend in general
multi-blade saws.[22] Since the interactions of fixed and loose on the geometry of the indenter, residual stresses and other
grains with a crystal differ considerably the results cannot be factors.[28] For the Vickers indenter the parameter a = 2, for a
applied directly to the wire saw technique. Buijs and Korpel- Knoop indenter a = 4/p, while the best fit value b = 1/7 var-
van Houten derived a quantitative model for lapping of glass, ies considerably in published data. Measurements for mono-
which was already based on the assumption of rolling grains crystalline silicon at room temperature yield H = 10.6 ±
and the material removal by indentation and frac- 10.9 GPa[29,30] and KIC = 0.82, 0.9, and 0.95 MPa m1/2 in the
ture.[10±12,23,24] Because of the differences between the lapping {111}, {110}, and {100} plane, respectively. In large grained
and wire sawing process it remained unclear whether the re- polycrystalline silicon KIC = 0.75 MPa m1/2 has been ob-
sults could be directly applied to the wire sawing. tained.[31,32]
The first studies of the wire sawing process have been The material removal by chipping occurs during the un-
published during the last few years.[1,3±9,25±27] Combining the loading phase of the indenter by the formation of the lateral
rolling-indenting process of free abrasive grains with the frac- cracks parallel to the surface as discussed in the previous sec-
ture mechanics of brittle materials quantitative descriptions tion. The driving force for these cracks are the residual stress-
of the material removal process could be derived. Since cer- es of the plastic zone and the median / radial cracks. These
tain details of the process are not fully understood yet and stresses depend in a complicated way on the normal force FN
had to be replaced by assumptions a unique model cannot be that has been applied during the loading phase of the inden-
developed at present. Therefore in the following the main as- ter. Marshall et. al.[33] have investigated the stresses at the

504  2004 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim http://www.aem-journal.de ADVANCED ENGINEERING MATERIALS 2004, 6, No. 7
Möller/Basic Mechanisms and Models of Multi-Wire Sawing

with n = 5/8. The prefactor b* depends on the material con-

REVIEWS
stants E, H, KIC and two dimensionless constants that can be
fitted to experimental results. In the limit of large contact
loads FN >> Fo Eq. 2 reduces to the simple power law of cL.
The experimental measurements on a number of ceramic ma-
terials confirm this dependence and yield threshold values
between 0.05 N (MgF2) and 5 N (ZnS). Yang and Kao[8] de-
rived a similar dependence for high loads with the same ex-
ponent but a different prefactor b*. Further results from solid
particle erosion experiments on glass also confirm a similar
power law but with different exponents.[34] A recent direct in-
vestigation of the mean lateral crack extension in Vickers in-
dentation experiments for monocrystalline silicon[35] yielded
eq. 3 with b* = 13.76 lm/Nn and n » 0.85 which is slightly
higher than the theoretical exponent n = 5/8 = 0.625.
The experimental results also show that fracture requires a
minimal load.[17±19] Experimentally the threshold value has
been determined from lapping experiments on silicon sur-
faces using the same SiC slurry used for sawing.[26] The criti-
cal load Fc on a single SiC grain, below which no chipping is
observed in these experiments, turned out to be Fc = 0.03 N.
This value has been taken here as the lower limit for the be-
ginning of the abrasive action. The critical load depends how-
ever on many factors such as the crack system, the surface
orientation and roughness, or the indentation mode. There-
fore experimental values vary considerably in the literature:
between 0.003 ± 0.03 N. For loads lower than the critical load
the indentation only produces the elastic-plastic impression.
Material can still be removed in this case by plastic deforma-
tion processes as described before, but this is certainly a
slower process which yields however smoother and less dam-
aged surfaces.

3.4. Material Removal Rate

The material volume Vo that can be removed by lateral
cracks at the impression of a single grain is approximately
given by Vo = p c2 d if the lateral cracks are formed directly
beneath the plastic zone (see Fig. 7). The depth d of the im-
Fig. 5. Optical micrograph of individual indentations after a slight polish of an as- pression can be calculated from the shape of the plastic zone
sawn surface (left) and the damage structure of a single Vickers indentation (right). d = a / 2 tan j where j is the half-angle of the tip of the grain
Size and shape are similar. The typical damage structure by scratching is also shown
for comparison. that is indented into the surface. Combining this with Eq. 1, 2
yields

p a1=2 b 2
central plastic deformation zone theoretically and determined Vo ˆ F…4n‡1†=2 (4)
2 tan j H1=2 N
the following dependence of the extension (radius) of the lat-
eral cracks c on the normal force in the fully unloaded state With n = 0.85 as determined for silicon one obtains
Vo » FN2.2. The sawing rate or velocity vs, which is the thick-
1=4 1=2
c ˆ cL ‰1 …Fo = FN † Š (2) ness of a removed layer per unit time, can be calculated from
the number of indentation events m per contact area As and
where Fo represents an apparent threshold force, and cL a lim- time Dt multiplied by the material volume Vo that is removed
iting crack function that can be described by in a single event.

 n m Vo
cL ˆ b FN (3) vs ˆ (5)
As Dt

ADVANCED ENGINEERING MATERIALS 2004, 6, No. 7 http://www.aem-journal.de  2004 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim 505
 Möller/Basic Mechanisms and Models of Multi-Wire Sawing

If a rolling grain makes one indentation per cycle one can

REVIEWS

calculate the time interval for a single indentation from the

angular velocity of the rotation x = 2 p/Dt. For a laminar
flow of the slurry the mean angular velocity * x is given by
*
x = rot(v* )/2 and can thus be determined from the linear ve-
sl
locity profile *
vsl (z) between crystal surface and moving wire

z v
vsl ˆ v x ˆ (6)
Lo 2 Lo
where z is the distance from the crystal surface and Lo the dis-
tance between wire and surface. Combining Eqs. 4 ± 6 yields
for the sawing rate the fundamental relationship which forms
the basis for the following theoretical descriptions

…4n‡1†=2 p a1=2 b 2
vs ˆ vso v m FN vso ˆ (7)
4 tan j H1=2 As Lo Fig. 7. Indentation of a single particle into the surface. Under the action of the shear
and normal forces a plastic zone and cracks are formed. The extension 2c of the lateral
The remaining problem that has to be solved is to deter- cracks and the depth d of the plastic zone determine approximately the chipped volume.
mine the number m of indenting grains in the slurry and the
normal force FN acting upon each grain. Both factors can only
be average values that have to be determined from the global particles into the cutting zone and determines the force acting
stochastic behavior of many particles in the slurry under the upon the grains. Depending on the film thickness between
external sawing conditions such as wire velocity v, wire load wire and surface, the particles are either in direct contact with
Ftot, particle size distribution or slurry concentration. These both the wire and the surface (semi-contact case) or the parti-
conditions shall be analyzed in the following. cles are floating freely (non-contact case). In the first case the
force on the grains is exerted by the wire directly whereas in
the second case the force is supplied by the shear stress and
3.5. Hydrodynamic Behavior of Slurry and Abrasive
the hydrodynamic pressure in the moving slurry. It is also
Particles
evident that if the slurry film thickness is much lower com-
The global behavior of the abrasive particles is controlled pared to the average size of the abrasive grains less particles
by the hydrodynamic conditions in the slurry film between would enter the cutting zone, thus leading to very low saw-
ingot surface and wire. The slurry transports the abrasive ing rates and the risk of wire rupture due to dry friction.
It is therefore necessary to determine the slurry film thick-
ness under the various sawing conditions. The hydrodynamic
behavior of slurry films is also important in lubrication or
polishing processes, where many fundamental aspects have
been derived from experimental and theoretical results. From
the lubrication theory, the elasto-hydrodynamic behavior of
the system can be classified in different regimes.[36,37] Consid-
ering the conditions of wire sawing where the wire can de-
form elastically in response to the slurry pressure the system
most likely belongs either to the isoviscous-elastic (abbre-
viated as IE) or isoviscous-rigid regime (abbreviated as IR). If
the elastic deformation of the wire is negligible as compared
with the film thickness, the slurry is in the IR-regime. If the
elastic deformation is significant and the viscosity changes
are negligible the slurry is in the IE-regime. For constant slur-
ry temperature and perfectly smooth wire and crystal sur-
faces the lubrication theory yields for the minimum slurry
film thickness

hIE ˆ CIE v0:66 l0:66 = F0:21

tot
E0:44 (for IE-regime) (8)
Fig. 6. Schematic diagram of the development of the crack system below a sharp inden-
ter upon loading and unloading. Dark grey areas indicate the plastic zone below the in- hIR ˆ CIR v2 l2 = F2tot (for IR-regime) (9)
denter. Median and radial cracks form in the loading phase at the edges of the plastic
zone perpendicular to the surface. If they occur on the same plane they may coalesce
forming half-penny shaped cracks that can open to the surface and become visible then.
where v is the wire speed, l the slurry viscosity, Ftot the total
The lateral cracks form upon unloading and lead to chipping of volume. load on the wire and E an effective elastic modulus of the

506  2004 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim http://www.aem-journal.de ADVANCED ENGINEERING MATERIALS 2004, 6, No. 7
Möller/Basic Mechanisms and Models of Multi-Wire Sawing

sults are the following. The minimum film thickness increases

REVIEWS
both with wire speed v and viscosity l of the slurry. The
highly viscous slurry also generates a hydrodynamic pres-
sure which deforms the taut wire thus generating a elasto-hy-
drodynamic environment. The corresponding pressure which
is zero at the inlet increases gradually in the direction of the
moving wire and then decreases again towards the exit. The
maximum pressure is about 0.5 MPa for a 100 mm ingot, v =
10 m/s, a wire bow of 2 deg, and slurry viscosity of 1 Ns/m2.
An important result is that the film thickness in this calcu-
lation turns out to between 0.1 ± 0.4 mm which is much larger
than the average size of the abrasive grains of about 10 ±
30 lm for conventional SiC powders. Therefore it has been
concluded that the abrasive particles are not in direct contact
Fig. 8. Schematic particle size distribution g(l) for two different volume concentrations
with both wire and ingot but are floating freely. Such a large
cv In the semi-contact case grains with a diameter l > Lo(cv) are in contact both with
wire and workpiece surface. Lo(cv) is slightly dependent on the volume concentration. film thickness at the side faces of the sawing channel (Fig. 3)
Without applied force the total number of grains mo in contact is proportional to the appears however incompatible with experimental facts. In
shaded area.
most practical cases the kerf loss which is equal to the dis-
tance between wire and wafer surface corresponds to the size
workpiece and wire. CIE and CIE are geometry factors that de- of the largest grains in the particle size distribution. Therefore
pend sensitively on the geometries of the tool and workpiece one may have to distinguish between the sawing conditions
surface. Although the equations have been derived for slur-
ries without particles, it is maintained in the following that
these equations can also be fitted to the lubrication of wire a)
saws. The film thickness determines the shear stress in the
slurry that is given by

dv
s ˆ l (10)
dh
where dv/dh is the velocity gradient or shear rate in the slurry
between wire and crystal surface. Since the slurry films are
relatively thin one can expect laminar flow with a constant
and high velocity gradient, therefore one can write in this
case dv/dh = v/h. Replacing h by the minimum film thickness
in eq. 10 one obtains for the shear stress

sIE ˆ l0:34 v0:34 F0:21

tot
E0:44 = CIE (for IE regime) (11)

b)
sIR ˆ F2tot = CIR v l (for IR regime) (12)

Under the usual sawing conditions the slurry viscosity can

be assumed to be independent of the shear rate. It depends
however on the hydrodynamic pressure, the temperature and
the particle concentration. Therefore the shear stress in the
slurry and thus the dragging force acting on the abrasive par-
ticles may change considerably with the sawing conditions.
A more realistic analyses of the hydrodynamic behavior of
the slurry in a wire saw which takes into account the geome-
tries and boundary conditions of the system has been given
by Bhagavat et al.[6] They calculated the slurry film thickness
by a finite element simulation based on the solution of the
Reynold's equation[38] under the assumption that the system Fig. 9. Calculated diagram of the total wire force versus wire velocity for constant
slurry parameters using eq. 16 and 18. The results show the range of parameters for
is in the isoviscous-elastic state because of the elastic response the semi-contact and non-contact regime (a). In the contour diagram lines of constant
of the wire to the internal pressure.[39] The main numerical re- sawing rate are plotted (b).

ADVANCED ENGINEERING MATERIALS 2004, 6, No. 7 http://www.aem-journal.de  2004 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim 507
 Möller/Basic Mechanisms and Models of Multi-Wire Sawing

Experimental results from lapping experiments[11] show

REVIEWS

that with increasing total force Ftot the average force FN on

each particle and thus the average indentation depth d remain
constant. The same situation is encountered in the contact of
nominally flat rough surfaces, where the number of contact-
ing asperities (here m) increases with increasing load, such
that the average contact area and load per contact remain
constant. Values of FN v 0.3 ± 0.5 N have been determined. In-
serting eq. 14,15 in 7 yields the sawing rate vs for the direct-
contact case
 …4n 1†=2
p a1=2 b 2 …c d†
vs ˆ vsc v Ftot lo…4n 1†=2
vsc ˆ (16)
so so 4 tan j H1=2 As Lo
Fig. 10. Measurement of the sawing rate as a function of the applied wire load for two
different wire velocities. With n = 0.85 one obtains for the sawing rate vs »v Ftot lo0.2,
n = 5/8 yields vs » v Ftot./ lo0.25. This indicates a weak depen-
dence on the mean size of the largest grains in a distribution.
below the wire (non-contact case) and at the side faces (semi- The mean size of the largest grains which determines the dis-
contact case). One has to admit however that these details still tance Lo between wire and surface depends however weakly
have to be verified experimentally. In the following both situ- on the slurry concentration as discussed before.[35]
ations shall be analyzed separately.
3.5.2. Non-Contact Case
3.5.1. Semi-Contact Case
In the non-contact case the forces on the grains are exerted
In the semi-contact case only those grains are indented that by the shear stress in the slurry and the hydrodynamic pres-
are in contact both with the wire and the ingot surface. In sure that builds up in the slurry as discussed before. The
general the grain size in the slurry is not homogeneous but shear stress alone, however, causes the particles only to rotate
distributed over a certain range of diameters (Fig. 8). In addi- without any efficient energy transmission to the surface. In-
tion, both the surface of the crystal and the wire have a cer- homogeneities at the surface caused by the surface roughness
tain roughness. It is evident because of the stiffness of the ma- will cause some grains to transmit force to the surface. In gen-
terials that wire and surface must be kept at a certain distance eral it will be difficult to transfer energy to the surface and to
Lo that is determined by the largest grains. The slurry film deform or break the material particularly if the surface is
thickness h in Eq. 8,9 must be smaller then: h £Lo. With in- smooth. For a more efficient energy transmission a force nor-
creasing force Ftot on the wire the distance Lo decreases and mal to the surface will be required. The complete description
more grains come into contact with the surface. Neglecting of the process in the floating state has to take into account the
the surface roughness the number of particles m that are actu- combined action of shear stress and hydrodynamic pressure
ally indented into the crystal surface can be derived from the in the slurry. So far the process has only been described for
particle size distribution g(l) by the relation special cases.
Möller et al. assumed for simplicity that only the shear
R1
m ˆ g …l† dl (13) stress exerts force on the particles.[35] Since this force acts par-
Lo allel to the surface it can only lead to material removal at
where m is equal to the integral over the shaded area in Fig- steps and inhomogeneities of the surface. The results may
ure 8. One can see that for a given size distribution g(Lo) in- give however a first insight into the modifications of the
creases very weakly with the volume concentration cv of the semi-contact case. The shear stress exerts forces across the
particles in the suspension. This is also confirmed by experi- surface of the particle. Assuming spherical particles and tak-
mental results.[35] In the following it will be considered a con- ing their mean diameter lm the shear force on each particle is
stant. The force on each particle can be derived from Eq. 1a. given then by Ft = s plm2. If one considers the IE-regime only,
For a spherical particle with diameter l the mean contact area one obtains with Eq. 11
is a2 = 4 l d which yields
Ft ˆ p l2m l0:34 v0:34 F0:21
tot
E0:44 = CIE (17)
FN ˆ c l d (14)
Since the force acts mainly parallel to the surface only a
with c = 2 p H. For other shapes of the grains c adopts differ- fraction of it can contribute to the indentation process normal
ent numerical values. d is the average indentation depth. The to the surface. This can happen if the motion of the particles
total force Ftot can be calculated from deviates accidentally from the parallel flow or if the surface is
uneven. One can take this into account by a geometry factor
Ftot ˆ m FN (15) which determines the fraction of the shear force component

508  2004 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim http://www.aem-journal.de ADVANCED ENGINEERING MATERIALS 2004, 6, No. 7
Möller/Basic Mechanisms and Models of Multi-Wire Sawing

FN normal to the surface FN = sinu Ft. In general the average ones obtained here. Nonetheless the sawing rate equations in

REVIEWS
value of sinu will be small but may depend on other factors the semi-contact case (Eq. 6) and the non-contact case (Eq. 18)
such as the hydrodynamic pressure or the surface roughness. allow one to draw some general conclusions.
Assuming that the workpiece surface As is totally covered by The typical pressures in production saws exerted by a sin-
abrasive particles with an average diameter of lm the total gle wire on the workpiece surface lie below 0.1 MPa. The
number m is given by m = 4 As/plm2. Inserting m and Eq. 17 in shear stress and hydrodynamic pressure in the slurries can be
(7) the sawing rate in the IE-regime is given then by the gen- estimated from the numerical simulations and vary between
eral equation 0 ± 2 MPa. In the semi-contact case the pressure is exerted on
relatively few particles (the largest ones) which results in
n1 n2 n n
vs ˆ vnc
so
v l Ftot3 lm4 high forces on these particles whereas in the non-contact case
all particles contribute to the abrasive process but the force
…4n‡1†=2
!…4n‡3†=2 component normal to the surface is relatively small. There-

a1=2 b 2 …psinu† E0:44
vnc ˆ (18) fore the sawing rates in the semi-contact regime can be com-
so tan j H1=2 CIE
parable to the non-contact case. This makes it difficult to de-
where n1 = 0.68 n + 0.51, n2 = 0.68 n ±0.49, n3 = 0.42 n +0.315, cide in which regime the wire sawing process actually occurs
n4 = 4 n ± 1. With n = 0.85 one obtains for the sawing rate vs or whether a change from one regime to the other is possible
~ v1.1 l0.1 Ftot0.67 lm2.4, for n = 5/8 then vs ~ v0.94 l±0.07 Ftot0.51 lm1.5. for specific operating conditions. Experimentally the regimes
In comparison with Eq. 16 for the semi-contact case the de- can be in principle distinguished by the dependencies on the
pendence on the wire velocity is almost equal but weaker for sawing parameters but the experimental results so far are in-
the total force. There is almost no dependence on the viscosi- sufficient.
ty of the slurry. The main parameters in the non-contact case are the wire
Equations of the same type have also been developed by velocity, the viscosity, the total force and the mean size of
other authors based on different starting assumptions. Bhaga- slurry particles whereas in the semi-contact case the viscosity
vat et al.[6] derived the material removal rate for the IE-regime does not enter the equation and the mean particle size is re-
by considering the energy transfer to the surface. The result placed by the mean size of the largest particles in the distribu-
in combination with eq. 11 yields the following exponents n1 tion. Thus for a given slurry with a fixed grain size distribu-
= 0.68, n2 = ±0.32, n3 = 0.42, n4 = 1 in eq. 18. Yang and Kao[8] tion and concentration the viscosity and mean particle sizes
neglect the shear stress and assume that the main force on the are fixed. The sawing rate depends on the total force Ftot and
particles is supplied by the total force on the wire through the the wire velocity v only in both cases. An important fact is
hydrodynamic pressure. They also assume that the contact that these equations have only a certain range of validity and
area As between wire and workpiece is completely covered it appears useful to depict the different regimes in a two-di-
by abrasive particles. Then the force on each particle from mensional Ftot vs. v diagram from eq 8, 16 and 18 (Fig. 8a).
eq. 13 is given by FN = p Ftot lm2/4 As and the sawing rate by The semi-contact and non-contact regime are separated by a
Eq. 18 with the following exponents n1 = 1, n2 = 0, n3 = line which is defined by the condition that the slurry film
(4 n+1)/2, n4 = 4 n±1. With n = 0.85 one obtains for the sawing thickness is equal to the mean size of the largest particles in
rate vs » v Ftot2.2 lm2.4, n = 5/8 yields vs » v Ftot1.75 lm1.5. the size distribution. For a given wire speed the regime
In a general model for the sawing rate which includes both changes from non-contact to semi-contact with increasing to-
the shear stress, the hydrodynamic pressure and the variation tal force and the sawing rate changes abruptly when crossing
of force along the cross section of the wire one can expect that the line (Fig. 8b). The main features are the different depen-
the exponents of the parameters will differ again from the dencies on Ftot and v. In the semi-contact regime the sawing
rate is proportional to Ftot and v (Eq. 16), in the non-contact
regime the sawing rate is almost proportional to v but shows
a non-linear dependence on Ftot (Eq. 18).
The numerical results also show how the system behaves
when the slurry parameters are changed. For instance with
increasing slurry viscosity the non-contact regime extends to
higher forces because the separation line is pushed upwards.
Experimentally the position of the border line between the
two regimes could not be established yet. Therefore it is not
known at present whether sawing with typical industrial pa-
rameters occurs in the semi-contact or the non-contact regime
and whether the sawing rate increases or decreases when
crossing the line. These equations are therefore only a first ap-
proach to quantify the sawing rate and determine the impor-
Fig. 11. Measurement of the normalized total force on the wire as a function of the wire
velocity for two different sawing rates. The results have been fitted to a power law FN »
tant factors. In the next chapter some of the dependencies
vn. From eq. 16 one expects n = ± 1. shall be compared with available experimental results.

ADVANCED ENGINEERING MATERIALS 2004, 6, No. 7 http://www.aem-journal.de  2004 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim 509
 Möller/Basic Mechanisms and Models of Multi-Wire Sawing

4. Experimental Results
REVIEWS

Unfortunately only very few systematic results of sawing

experiments have been published that can be compared with
the sawing Eqs. 16, 18.[26] More results are available from lap-
ping[10±18,23] and lubrication experiments,[36±45] but in these
cases usually different slurries and abrasive materials are
used. Since the equations for material removal also depend
on the geometries of the tools and workpieces one has to be
careful when comparing these results. In most experimental
cases the slurry parameters are not known or have not been
published completely, therefore mainly the dependencies on
Fig. 12. Measurement of the sawing rate for different mean diameters of the SiC parti-
the parameters Ftot, and v, which determine the sawing re- cle distribution for a low wire velocity and constant load. The results have been fitted
gime for a given slurry, shall be analyzed. to eq 16 assuming that the mean particle size lm is proportional to the mean size lo of
the abrasive grains.
Figure 9 shows the sawing rate as a function of the applied
wire load for different wire velocities. A commercial oil slurry
with SiC particles has been used. In both cases the sawing also difficult to determine all parameters that are required to
rate is proportional to the load. One can also observe that the predict the performance of a commercial slurry quantita-
material removal starts only above a certain threshold value tively. One problem is to describe the particle distribution
of about 0.1 N. The threshold value can be explained with the and the morphology of the grains exactly, such as the correct
minimum load for an abrasive particle that is required for the determination of the mean size lo of the active grains. The
chipping mechanism to operate. The linear dependence on quantitative comparison with the experiments has shown that
the load indicates that the sawing occurs in the semi-contact in the semi-contact regime only the very large grains of a size
regime. distribution contribute to the sawing process, about twice as
Further experimental data however cannot always be large as the average size. This part of the size distribution is
reconciled with the sawing equation for the semi-contact very often not well characterized and may vary for commer-
regime. This is shown in Figure 10 where the wire load is cial products with the same specification. Commercial specifi-
depicted as a function of the wire speed for constant feeding cations yield for instance the l90-value (the size which 10 %
rate. The results show a power law dependence FN » vn with of grains exceed), but this value is not always sufficient to
two different exponents which differ from n = 1 as expected predict the sawing performance.
from eq. 16 for the semi-contact regime. If one compares the The increase of the sawing rate with the grain size suggests
experimental data with the numerical results of the FN vs. v to use coarser powders. There are limitations however due to
diagram in Figure 9a one can see that if the measurements the diameter of the wire. If the grains become too large com-
also cover data from the non-contact regime the interpolation pared to the wire diameter the sawing rate decreases again.
will give an intermediate value for the exponent. The results This may be due to the fact that not enough force can be ap-
therefore seem to indicate that both the sawing conditions of plied on the large grains by the wire. Thinner wires therefore
the semi-contact and the non-contact regime can occur. In fact also require smaller grain sizes. Coarser grains also lead to
if one considers that the forces exerted by the wire vary along more saw damage and roughness of the surface as will be
the cross section it is quite possible that different sawing described in the next chapter. This is not desirable for the
regimes may occur at the same time directly below the wire following solar cell process.
and at the side faces where the forces are much lower (see The shape of the particles may also have an influence on
Fig. 3). the transport properties of the grains in the slurry. For in-
Systematic experimental results for different slurries are stance elongated grains may rotate less in the laminar flow of
still less available. Fig 10 shows some results for the sawing the slurry than round shaped grains, which reduces the num-
rate as a function of the variation of the mean grain size of the ber of indentations. It is also known that new grains show a
slurry for the low velocity range. A weak dependence on the better sawing performance than used ones. The performance
mean size lo of large particles can be observed with an decreases with time perhaps because the grains loose their
exponent between 0.25 ± 0.5, which is also consistent with sharp edges due to fracture. The performance also decreases
the exponent derived for the semi-contact regime (Eq. 16). because the slurry is loaded with silicon and iron wear which
The few results indicate that at present a consistent picture changes the viscosity and the transport properties of the slur-
of the sawing mechanism cannot be derived. They point into ry. If the viscosity becomes too high and agglomeration of
the direction that the wire sawing occurs mainly in the semi- particles occurs, the sawing rate decreases. The slurry and the
contact regime but the non-contact case has to be taken into grains have to be replaced or recycled then after some runs.
account. Therefore it is still too early to predict the sawing In general the role of the slurry and the hydrodynamic prop-
behavior accurately, particularly if other parameters for in- erties are still poorly understood and a further development
stance the slurry composition are changed. In practice, it is of the model is required here.

510  2004 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim http://www.aem-journal.de ADVANCED ENGINEERING MATERIALS 2004, 6, No. 7
Möller/Basic Mechanisms and Models of Multi-Wire Sawing

REVIEWS
low pass: 2 mm

high / low pass:

200 µm - 2 mm

high pass: 200 µm

Fig. 13. Optical image of an as-cut wafer surface showing different types of surface damage (left). Vertical line scan measurements of the surface topography are shown right. The
surface profile has been filtered on different length scales showing coarse, medium and fine structures in the surface roughness.

5. Saw Damage and Wafer Surface Quality of the wire. They occur particularly under higher loads and
can be caused by a deficit of slurry, mechanical vibrations, or
After sawing the surface of the wafer is damaged from the inhomogeneities of the material. Mostly a large number of
fracture processes and contaminated with organic and inor- parallel wafers is affected then. Grooves on wafers cannot be
ganic remnants from the slurry. Therefore the wafers have to removed by etching and thus reduce the quality of a wafer.
be cleaned and the saw damage removed by etching before a On a length scale of about 100 lm the surface may have a
solar cell can be fabricated. In addition, the thickness and sur- wavy topology which is not detrimental unless sharp steps
face roughness of the wafer may vary which may be detri- do occur. The origin of these large scale inhomogeneities is
mental for some of the further processing steps. All of these still unclear.
factors are related to the sawing process. Figure 11 shows an On the lm-length scale the surface shows a certain rough-
example of the topology of an as-cut surface. It consists of ness which is directly related to the microscopic sawing
thickness variations on different length scales. On the scale of process as described before. The surface roughness Rz corre-
millimeters one can observe grooves parallel to the direction lates with the average indentation depth d of the grains: Rz »

ADVANCED ENGINEERING MATERIALS 2004, 6, No. 7 http://www.aem-journal.de  2004 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim 511
 Möller/Basic Mechanisms and Models of Multi-Wire Sawing

d. It can be derived from Eq. 14. Since FN is constant the ±

REVIEWS

roughness depends on the size distribution of the particles [1] A. Endrös, D. Franke, Ch. Häûler, J. Kaleis, W. Koch,
only and is independent of the other sawing parameters such H. J. Möller, in Handbook of Photovoltaic Engineering
as the wire velocity or the total wire stress. This is in agree- (Eds: A. Luque, S. Hegedus, Wiley-VCH, Weinheim,
ment with the experimental observation.[12] 2002).
The saw damage below the surface consists of the remnant [2] R. Wells, Solid State Technology 1987, 30, 63.
radial / median cracks. They have to be etched away before [3] R. K. Sahoo, V. Prasad, I. Kao, J. Talbott, K. Gupta,
solar cell processing. The extension of the damage zone is ASME J. Electron. Packag. 1998, 120, 131.
equal to the mean length c of these cracks. From Eq. 1b and [4] J. Li, I. Kao, V. Prasad, Proc. ASME ± IMECE, Dallas,
14 one can derive ASME Press 1997, 439.
[5] F. Yang, J. Kao, ASME J. Electron. Packag. 1999, 121, 191.
!2=3
c b lo d [6] M. Bhagavat, V. Prasad, I. Kao, J. Tribology 2000, 122,
c ˆ (19) 394.
KIC
[7] S. Wei, I. Kao, Trans. Int. J. Vib. Sound 2000, 231, 1383.
The crack length and thus the depth of the damaged zone [8] F. Yang, I. Kao, ASME J. Electron. Packag. 2001, 123, 254.
is therefore determined by the mean size lo of the active [9] I. Kao, S. Wei, P. Chiang, Proc. of NSF Design & Manu-
grains; it also does not depend on other factors.[46] The results fact. Grantees Conf. 1997, 239.
have been be confirmed by TEM and other experimental in- [10] M. Buijs, K. Korpel-van Houten, Wear 1993, 162, 954.
vestigations.[47±51] The extension of the saw damage lies typi- [11] M. Buijs, K. Korpel-van Houten, Wear 1993, 166, 237.
cally in the range of 5 ± 10 lm. [12] M. Buijs, K. Korpel-van Houten, J. Mater. Sci. 1993, 28,
Saw damage also occurs to the abrasive grains and the 3014.
wire itself. Although the fracture strength of the SiC particles [13] A. G. Evans, D. B. Marshall, in Fund. of Friction and
is higher than that of silicon, the grains eventually loose their Wear, ASM, Metals Park, OH, 1981, 441.
sharpness due to breakage which reduces their sawing per- [14] S. Malkin, J. E. Ritter, ASME J. Eng. Industry 1989, 111,
formance. To reduce the abrasion of the grains sawing should 167.
be done in a stress range where the load on the individual [15] B. Lawn, in Fracture of Brittle Solids, Cambridge Univer-
grains lies above the fracture strength of silicon but below sity Press 1993.
that of SiC. Excessive wear of the wire can lead to breakage [16] B. Lawn, A. Evans, J. Mater. Sci. 1977, 12, 2195.
which is undesirable during sawing because it is very time [17] A. G. Evans, E. A. Charles, J. Am. Cer. Soc. 1976, 59, 371.
consuming to build up the wire web inside of the machine. [18] J. Lankford, D. Davidson, J. Mater. Sci. 1979, 14, 1662.
[19] J. Hagan, J. Mater. Sci. 1979, 14, 2975.
[20] Y. Gogots, C. Baek, F. Kirscht, Semicond. Sci. Technol.
6. Future Developments 1999, 14, 936.
[21] E. Weppelmann, J. Field, M. Swain, J. Mater. Res. 1993,
The investigations of the microscopic processes of wire
8, 246.
sawing have laid the basis for the selection of the best range
[22] W. Ermer, D. Helmreich, D. Regler, D. Seifert, Techn. Di-
of parameters and for further modifications. It allows one to
gest 1st Int. PVSEC 1989, 781.
increase the sawing performance, to reduce the consumption
[23] K. Philips, G. M. Crimes, T. R. Wilshaw, Wear 1977, 41,
of slurry, SiC powder, wire material and etchant, and hence
327.
directly the costs of slicing. The technique becomes also en-
[24] C. Jeynes, Phil. Mag. 1983, 48 A, 178.
vironmentally more benign. Furthermore the quality of the
[25] S. Wei, I. Kao, Proc. Manufact. Eng. Div. ± IMECE, ASME
wafers such as roughness, flatness and saw damage of the
Press, 1998, 813.
surfaces can be improved. This is also important in view of
[26] C. Borst, H. J. Möller, VEDRAS ± Report, BMBF, Ger-
the development of thinner wafers for solar cells, a technolo-
many 1998, 23.
gy, which will use the expensive silicon material more effi-
[27] J. Kijlstra, D. Storch, VEDRAS ± Report, BMBF, Germany
ciently. The current sawing technique in production allows
1998, 53.
the sawing of wafers with thickness down to about 200 lm. It
[28] B. R. Lawn, D. B. Marshall, J. Am. Cer. Soc. 1979, 62, 347.
is the goal to further reduce the thickness down to about
[29] P. Feltham, R. Banerjee, J. Mater. Sci. 1992, 27, 1626.
100 lm in production. Sawing of thinner wafers is possible
[30] H. Li, R. Bradt, J. Mater. Sci. 1996, 31, 1065.
but at present still at the expense of more breakage. The prob-
[31] C. P. Chen, M. Leipold, J. Am. Cer. Soc. 1980, 59, 342.
lem becomes even more severe when the wafer size increases
[32] G. R. Anstis, P. Chantikul, B. R. Lawn, D. B. Marshall,
at the same time. In mass production such a development will
J. Am. Cer. Soc. 1981, 64, 533.
only become possible by a careful selection of the parameter
[33] D. B. Marshall, B. R. Lawn, A. G. Evans, J. Am. Cer. Soc.
range and an in-situ control of all factors that determine the
1982, 65, 561.
slicing process.
[34] S. Srinivasan, R. O. Scattergood, J. Mat. Science 1987, 22,
Received: March 24, 2004 3463.

512  2004 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim http://www.aem-journal.de ADVANCED ENGINEERING MATERIALS 2004, 6, No. 7
Möller/Basic Mechanisms and Models of Multi-Wire Sawing

[35] H. J. Möller, C. Funke, in Freiberger Forschungshefte B, [41] Y.-T. Su, S.-Y. Wang, J.-S. Hsiau, Wear 1995, 188, 77.

REVIEWS
Eds: H. J. Möller, G. Roewer, TU Bergakademie Frei- [42] Y.-T. Su, C.-C. Horng, Y.-D. Hwang, W.-K. Guo, Wear,
berg, 2003, 230. 1996, 199, 89.
[36] M. Verspui, G. de With, J. Europ. Ceram. Soc. 1997, 17, [43] T. Kasai, T. K. Doy, Jpn. J. Tribology 1992, 37 (5), 539.
473. [44] G. Siekman, Wear 1987, 117, 359.
[37] B. J. Hamrock, in Fundamentals of Fluid Film Lubrication, [45] L. M. Cook, J. Non-Cryst. Solids 1990, 120, 152.
NASA Reference Publication 1225, McGraw-Hill, New [46] R. Chauhan, Y. Ahn, S. Chandrasekar, T. Farris, Wear
York 1991. 1993, 166, 246.
[38] N. Gane, P. F. Pfalzer, D. Tabor, Proc. R. Soc. London Ser. [47] M. Verspui, G. de With, P. van der Varst, M. Buijs, Wear
A 1974, 340, 495. 1995, 188, 102.
[39] R. Gohar, in Elastohydrodynamics, Ellis Horwood Series [48] J. Larsen-Basse, Wear 1993, 166, 93.
in Mech. Engineering 1988. [49] M. Hamblin, G. Stachowiak, Wear 1995, 190, 237.
[40] S. Rohde, K. Oh, Proc. R. Soc. London Ser. A 1975, 343, [50] C. Poon, B. Bushan, Wear 1995, 190, 76.
315. [51] A. Misra, I. Finnie, Wear 1981, 65, 359.

______________________

ADVANCED ENGINEERING MATERIALS 2004, 6, No. 7 http://www.aem-journal.de  2004 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim 513