August 5, 2004 11:59 WSPC/140-IJMPB 02533

International Journal of Modern Physics B

Vol. 18, No. 15 (2004) 2091–2121
c World Scientific Publishing Company
by SEOUL NATIONAL UNIVERSITY on 07/25/22. Re-use and distribution is strictly not permitted, except for Open Access articles.

PERCOLATION AND TUNNELING IN COMPOSITE MATERIALS

I. BALBERG∗ , D. AZULAY, D. TOKER and O. MILLO†

Racah Institute of Physics, The Hebrew University,
Int. J. Mod. Phys. B 2004.18:2091-2121. Downloaded from www.worldscientific.com

Jerusalem 91904, Israel

∗[email protected]
†[email protected]

Received 22 June 2004

Classical percolation theory is concerned with the onset of geometrical connectivity and
the accompanied onset of electrical connectivity in disordered systems. It was found,
however, that in many systems, such as various composites, the geometrical and elec-
trical onsets of the connectivity are not simultaneous and the correlation between them
depends on physical processes such as tunneling. The difference between the above two
types of systems and the consequences for the electrical transport properties of the lat-
ter composites have been largely ignored in the past. The application of scanning local
probe microscopies and some recent theoretical developments have enabled a better un-
derstanding of the latter systems and their sometimes “strange” behavior as bona fide
percolation systems. In this review we consider the above issues and their manifestation
in three types of systems: Carbon Black–Polymer composites, metal–insulator cermets
and hydrogenated microcrystalline silicon.

Keywords: Tunneling; percolation; composites; non-universal behavior; eletro-thermal

switching; transport in multi-phase systems; scanning local probe microscopies and
spectroscopies.

PACS numbers: 68.37.Ps, 72.20.-i, 72.80.Tm, 81.05.Qk, 81.05Rm

1. Introduction
Percolation theory deals with the behavior of some geometrical and physical prop-
erties above and below the onset of a global connectivity in a system. In classical
percolation theory one follows these properties assuming a well defined-geometrical
criterion for the inter-element bonding, and thus the minimal concentration of bonds
that yields an infinitely connected network is defined as the percolation threshold.1
Such percolation systems are manifested in various porous media and other con-
nected collections of elements, and they are characterized in the continuum by
percolation thresholds that are given by a critical fractional volume of these ele-
ments, vc . The value of vc can be in the whole range between almost 0 and the
close packing values of the corresponding elements.2 – 4 However, in many systems,
in particular in various composites, there is no such obvious nearest-neighbor bond-
ing criterion, but there may still be collective phenomena associated with some of

2091
 August 5, 2004 11:59 WSPC/140-IJMPB 02533

2092 I. Balberg et al.

the physical properties. Even in such cases, while there was no a priori justifica-
tion to the application of percolation theory, this theory has been used in order to
explain or describe the behaviors of the systems. To illustrate the latter issue let us
consider first in some detail the electrical properties of two very well known types
by SEOUL NATIONAL UNIVERSITY on 07/25/22. Re-use and distribution is strictly not permitted, except for Open Access articles.

of composites.
The system of carbon Black-Polymer (CB-P) composites5 is of very wide in-
terest due to the many applications that it provides,6 such as the electro-thermal
switching and current regulating devices that are widely used commercially.7 It
turned out that in trying to develop these applications some fundamental physical
questions concerning electrical conduction in disordered systems should have been
Int. J. Mod. Phys. B 2004.18:2091-2121. Downloaded from www.worldscientific.com

raised.8 These questions, however, have hardly been asked so far and of course
have not been answered to this date. The first question is how is it possible that
for a system such as the one illustrated in Fig. 1 (i.e. a system of geometrically
disconnected particles in which there is an insulating layer between adjacent par-
ticles, on one hand, but electrically these are all connected via tunneling, on the
other hand) the electrical properties behave as predicted by classical percolation
theory. This is unexpected since we recall that in the latter theory (as in porous
media1 and speckled films9 ) the opposite assumptions are made (i.e., simultaneous
geometrical and electrical connectivity that is associated only with the “bonding”
of adjacent conducting elements). In fact, this difficulty was put aside in the early
stage of the study of composite materials, and it was even suggested10 that “there

Fig. 1. An illustration of a system of “metallic” particles that are embedded in an insulating

matrix. The particles are not connected geometrically but they are connected, via tunneling,
electrically. The sphere’s “hard” cores represent the particles and the halo around them represents
the range over which the tunneling is significant. The tunneling “nearest neighbors” connections
are marked by bold segments while the connections to farther neighbors are marked by thin
segments. The bias is applied in that case between a conducting “tip” and the ground counter
electrode.
 August 5, 2004 11:59 WSPC/140-IJMPB 02533

Percolation and Tunneling in Composite Materials 2093

exists an intimate connection between percolation and the onset of conductivity in

an inhomogeneous material as far as there exist no tunneling effects”. The second
question is, assuming a percolation behavior, why is it that the non-universal crit-
ical conductivity exponent observed11 – 14 experimentally was, in some cases, much
by SEOUL NATIONAL UNIVERSITY on 07/25/22. Re-use and distribution is strictly not permitted, except for Open Access articles.

larger than predicted by universal1 and mean field9 theories but much smaller than
predicted by the corresponding available non-universal theories.11,13,15 The third
question, which is also important for applications, is, assuming a percolation be-
havior, what is the interplay between the electrical current distribution and the
heat dissipation that yields the observed current–voltage (I–V ) characteristics in
the electro-thermal switches that are made of CB-P composites.7,16
Int. J. Mod. Phys. B 2004.18:2091-2121. Downloaded from www.worldscientific.com

The second type of systems that illustrate the difficulty in the application of
classical percolation theory to composite materials is that of granular metal cer-
mets such as Ni–SiO2 . In these systems there are metallic grains (usually few nm in
size) that are embedded in an insulating (usually glassy) matrix.17 For low metallic
content, where the metallic grains are isolated from each other, one finds a finite
conductivity at room temperature, which diminishes drastrically at low tempera-
tures. When the volume fraction of the metallic phase v is large enough (usually of
the order of 0.5) the coalescence of neighboring metallic grains yields a continuous
metallic network, which is of course conducting at both low and high temperatures.
The sharp rise of the conductivity with the increase of v at the corresponding onset
of the global geometrical connectivity vc is a manifestation of the simple classical
percolation theory.18 While this transition is well understood, the above-mentioned
behavior for v < vc is not, and it is still a subject of intensive debate.19 In fact the
most common approach to explain this regime was that there is a global tunnel-
ing network that does not have percolation characteristics.17 Rather, the limited
conductivity and its temperature dependence were attributed only to the preferred-
route selection (that depends on the temperature) due to the competition between
tunneling and Coulomb blockade in the transfer of electrons between one metallic
grain to another.17,20 Correspondingly, the predicted temperature and fractional
conducting-volume dependencies of the electrical conductivity, σ(T ) and σ(v), have
been discussed only within the framework of nearest neighbor tunneling, while the
context of global connectivity has only been discussed phenomenologically.20
Historically, one of the difficulties in tackling the above problems experimentally
was the lack of a technique that could locally monitor the current distribution
within composites such as CB-P and Ni–SiO2 . This difficulty can be overcome now
due to the development of various Scanning Probe Microscopy (SPM) methods,
in particular that of the Conductance Atomic Force Microscopy (C-AFM). On the
theoretical side, new particle distribution functions15 and recent developments13,21
in the theory of percolation in the continuum provide a more realistic picture of
percolation-tunneling systems. The above advancements enable now the derivation
of improved theoretical predictions and better understanding of the critical behavior
of the transport even in the more complicated tunneling-percolation systems,20,22
e.g. where the observed percolation threshold appears to diminish. In what follows
 August 5, 2004 11:59 WSPC/140-IJMPB 02533

2094 I. Balberg et al.

we will describe in more detail the above problems, and the progress we made
towards their solutions.
In our recent works we tried to resolve the above problems by the application of
SPM techniques that yield direct information on the current networks in the hetero-
by SEOUL NATIONAL UNIVERSITY on 07/25/22. Re-use and distribution is strictly not permitted, except for Open Access articles.

geneous systems that we study. Using such information and combining it with some
recent macroscopic data and theoretical advancements appears to provide a system-
atic physical picture of the electrical conduction processes in tunneling-percolation
composites. Extending the basic-physical understanding that we gained, we were
also able to better understand not only the electro-thermal switching in CB-P com-
posites but also the much more complicated systems of heterogeneous semiconduc-
Int. J. Mod. Phys. B 2004.18:2091-2121. Downloaded from www.worldscientific.com

tors. Our studies of the latter issue focused on the hydrogenated microcrystalline
silicon (µc-Si:H) system. In that system we were able to show that of the (at least,
a priori, five) possible electrical conduction networks, and their corresponding trans-
port mechanisms, only two are actually dominant. One is operational at low v, and
the other at high v values. Each of the pertinent conductivity transitions can be
well described by a percolation model, but the connectivity and the conduction
mechanism are very different in the two corresponding networks.
The structure of the paper is as follows. In Section 2, we analyze in some detail
the relation between percolation and tunneling in composites in general and in some
of the systems mentioned above, in particular. We turn then in Section 3 to the un-
derstanding of the non-universal critical behavior that was found in systems where
no geometrical connectivity exists. In Section 4, we apply the above understanding
to elucidate the so-far poorly understood electro-thermal switching in CB-P based
devices. Another application of the above understanding is presented in Section 5,
in which we identify the percolating networks that contribute to the conductivity
of µc-Si:H. The main conclusions of the present work are presented in Section 6.

2. Percolation-Like Behavior with Inter-Particle Tunneling

It is quite well known that in order to have electrical transport in a composite with
geometrically disconnected conducting particles embedded in an insulating matrix,
one must invoke tunneling between them. For the composites in which conducting
Carbon Black particles (100–300 nm in size) are embedded in an insulating polymer
this basic, early days, idea23,24 was confirmed by a series of works25 that have indi-
cated convincingly (by the study of the temperature and pressure dependence of the
electrical resistivity) that inter-particle tunneling is the dominant conduction mech-
anism. On the other hand, the fact that macroscopic conductance could be measured
indicated that these composites must constitute a (rather loosely-defined “perco-
lation”) connected network of resistors.26– 28 In fact it was quite at an early stage
that a qualitative combination of both effects has been mentioned.29,30 However, it
was not until our finding31 of a quantitative agreement with percolation theory that
established it as the common operational model of the CB-P composites.12,32 – 36
In particular, we found that the dependence of the macroscopically measured
 August 5, 2004 11:59 WSPC/140-IJMPB 02533

Percolation and Tunneling in Composite Materials 2095

conductivity σ on the carbon black (conducting phase) volume content behaves

as
σ ∝ (v − vc )t , (1)
by SEOUL NATIONAL UNIVERSITY on 07/25/22. Re-use and distribution is strictly not permitted, except for Open Access articles.

1
with a “universal” t value that was predicted by the classical theory of lattice
percolation. On the other hand, vc was some empirical critical volume (vc < v). In
fact, this quantitative agreement of the measured critical exponent of the conduc-
tivity with the predictions of the universal percolation theory, should not have been
expected since, as mentioned in Section 1, in the CB-P composites the CB parti-
cles are not connected geometrically,5,37 in contrast with the basic assumption of
Int. J. Mod. Phys. B 2004.18:2091-2121. Downloaded from www.worldscientific.com

percolation theory.1 Moreover, the corresponding electrical network in CB-P com-

posites is such that all the particles are interconnected electrically, and thus the
critical connectivity feature that is manifested by Eq. (1) is not expected to be
relevant. Correspondingly, the meaning of vc (that is usually associated with the
onset of geometrical connectivity) in such a system remained unclear and it is not
surprising that its computation was characterized as “impossible”.10 Surprisingly,
the very fundamental question of compatibility of tunneling and percolation has not
been raised until our recent work,8 and the above operational percolation approach
was used with no justification or hesitation by numerous authors.12,32,38 – 40 On the
other hand the above mentioned many experimental results suggested that “some-
how” there is an underlying electrical-percolation network that is responsible for
the corresponding observations. To check whether such a network exists we have
carried out twenty years ago41 a capacitance-probe microscopy study and found
that indeed the cluster size statistics that we monitored experimentally is equal to
the one predicted by percolation theory. A considerable improvement of that ap-
proach was recently provided42 for CB-P by using C-AFM,43 that enabled us44 not
only to further confirm our conclusion with a high degree of accuracy but also lead
to the following outlined understanding of the reasons for the observed behavior. 45
A schematic of the C-AFM method is given in Fig. 2. A conductive AFM tip
scans the surface of a sample and monitors (with a resolution of about ten nms) si-
multaneously, the topography and local conductance (current) with respect to a re-
mote electrode. The measured current images depict two-dimensional “cuts” of the
underlying three-dimensional conduction network, from which the corresponding
total “conducting area” At is recorded.44 As well known1 the fractal dimension of
a system is determined (as one does with a mass fractal1 ) by the relation At ∝ LD ,
where D is the fractal dimension of that system and L2 is the totally scanned square
area around a corresponding center. Assuming a statistically random isotropic net-
work, the corresponding fractal dimension of the conducting volume Df is given44
by Df = D + 1. We have found,44 from many measurements, such as the one shown
in Fig. 3(a), that the measured fractal dimension of the conducting cluster in the
CB-P composites is Df = 2.5 ± 0.15, in agreement the theoretical prediction1
for the fractal dimension of a percolation cluster, which is 2.53. The increase of
Df with v (above vc ) has further shown44 the expected transition towards the
 August 5, 2004 11:59 WSPC/140-IJMPB 02533

2096 I. Balberg et al.

by SEOUL NATIONAL UNIVERSITY on 07/25/22. Re-use and distribution is strictly not permitted, except for Open Access articles.
Int. J. Mod. Phys. B 2004.18:2091-2121. Downloaded from www.worldscientific.com

Fig. 2. An illustration of the C-AFM method. A conductive tip scans the surface of the sample
thus monitoring simultaneously the topography and the local electrical conductance (or current
with respect to a remote electrode). The current is recorded only if there is a continuous electrical
path (e.g. via an aggregate of particles that yield such a path) from the tip to the counter electrode.

Fig. 3. The dependence of the fractional surface area, that is covered by current carrying regions,
on the side of the scanned square area, for a few values of the “cut-off” currents that are indicated
in the figure: (a) for a CB-Polymer composites with v just above the “percolation threshold”:
(b) for a Ni–SiO2 cermet with v = 0.35, i.e. below geometrical connectivity. The slopes of these
lines yield the indicated fractal dimensions of the “current images” for the corresponding values
of Ico . A schematic illustration of the relative scales of the particle size and effective tunneling
ranges for the two types of composites are given in the figure.
 August 5, 2004 11:59 WSPC/140-IJMPB 02533

Percolation and Tunneling in Composite Materials 2097

Euclidean-homogeneous value of 3. In passing we note that we do not know of any

attempt to experimentally characterize the geometrical network of CB-Polymer
composites. Such a characterization was done for similar systems using optical
microscopy, also revealing46 an increase of Df as v was increased above vc . We
by SEOUL NATIONAL UNIVERSITY on 07/25/22. Re-use and distribution is strictly not permitted, except for Open Access articles.

note, however, that in that study and in another C-AFM investigation47 it was
suggested that the geometrical network in itself is not a “universal percolation”
system. While this subject will not be discussed here, it underlines our above finding
that the electrically connected network is a bona fide percolation system. While
our above finding of the Df behavior confirmed then the percolation nature of the
electrical network and indicated that vc is indeed a well-defined phenomenological
Int. J. Mod. Phys. B 2004.18:2091-2121. Downloaded from www.worldscientific.com

quantity, it did not clarify why such a percolation network exists and what is the
relation between the conductivity-determined vc and the particular geometrical
and/or electrical characteristics of the system.
In trying to address the latter questions we invoked the obvious electrical con-
nectivity of the system as well as our knowledge about the conduction mechanism
that controls this connectivity, i.e. the underlying network of tunneling controlled
resistors.25 We recall then that the inter-particle tunneling conductivity (or prob-
ability) σtun is given by:13
 
r − 2b
σtun ∝ exp − . (2)
d
Here r is the distance between the centers of the particles, b is the radius of the
particles and d is the typical tunneling range (or tunneling decay) parameter. As
clearly suggested by Eq. (2), every two particles (here assumed spherical) in the
system are electrically connected, but their contribution to the conduction of the
whole network diminishes as r increases. Considering the overall electrical connec-
tivity of the system it is expected that the conductivity will continuously decrease
as v goes to zero, and that the corresponding dependence will not necessarily con-
form with percolation theory [Eq. (1)]. This is of course since an added particle is
not only connected electrically to its geometrical nearest neighbors but also to all
farther neighbors.
Following the behavior offered by Eq. (2) and the above finding that CB-P com-
posites behave as percolation systems (not only by their macroscopic conductivity
but also by the “geometry” of the conducting network) suggested to us that only the
electrical contribution of the “nearest neighbor” (in the sense of effective tunneling
conduction, see below) counts, and that for all practical purposes the electrical tun-
neling connection to non-nearest neighbors yields a negligible contribution to the
conductivity of the system. This model seems reasonable in particular when noting
that for non-nearest neighbors r − 2b is larger than 2b. Now, if b  d, as is the case
in CB-P composites (b ≈ 125 nm and d is estimated13 to be at the nm scale), the
corresponding conduction will indeed yield a negligible contribution of the tunnel-
ing between non-nearest neighbors to the macroscopic conductivity. If this picture
is correct, then vc has a well defined meaning as a percolation threshold. This is the
 August 5, 2004 11:59 WSPC/140-IJMPB 02533

2098 I. Balberg et al.

threshold for the formation of a conduction network made of “effectively connected

nearest neighbors” (ECNN) where we define “nearest neighbors” as those for which
r−2b ≈ d  b. The network that results is reminiscent then of classical descriptions
of percolation systems, but vc is now not only determined geometrically but also
by SEOUL NATIONAL UNIVERSITY on 07/25/22. Re-use and distribution is strictly not permitted, except for Open Access articles.

physically (e.g. in our case by the value of d, see also Section 3 below). In fact,
the system we describe is reminiscent of doped semiconductors48 where a tunneling
range is considered as a “tunnelable halo” of an impurity atom. While apparently
sound, the above outlined picture needed a convincing experimental confirmation
in order to establish that our above interpretation of the electrical properties of the
CB-P and similar composites is well founded.
Int. J. Mod. Phys. B 2004.18:2091-2121. Downloaded from www.worldscientific.com

Following the above considerations we realized that one can demonstrate ex-
perimentally that the above interpretation is valid by showing the existence of an
“hierarchy” of networks that are electrically connected as follows. 45 We were look-
ing first for an ECNN network that is made of a relatively small subsystem of the
conducting particles, and then for subsystems that include also various “non-nearest
neighbor” conductors. Our experimental approach to carry out this discrimination
was to use the measured C-AFM current images (say, under a given applied bias)
and obtain from each of them a series of images for currents that are larger than
a pre-chosen cutoff value, Ico .44,45 Now, it is obvious that if we consider a large
Ico value we get an image of a sub-network that is dominated by the relatively
“large” nearest neighbor conductors, while the images with very low Ico values
are expected to contain, in addition, the current contribution of farther-neighbor
conductors. The quantity that we examined then was the fractal dimension of the
connected networks as reflected by the above computationally separated images. If
our above approach is appropriate, we do expect that the larger the value of I co
the closer will be the value of Df to the one predicted by (“the nearest neighbor”)
percolation theory,1 i.e. 2.5, and that this value will shift towards 3 as our pre-
chosen Ico value is lowered (and the contribution of “non-nearest neighbor” will
be revealed). As exhibited by the results shown in Fig. 3(a) no such “expected”
dependence on Ico was found. Still assuming that our above interpretation, that
the network conductivity is dictated by the nearest-neighbors tunneling, is correct,
we have rationalized our above results as follows: for the b  d systems, such as
the CB-P composites (b ≡ R(CB) in Fig. 3(a)), we observe the fractal dimension of
Df = 2.5 for the lowest values of Ico that we could obtain, since the contribution
of non-nearest neighbors is negligible to the degree of being undetectable with our
experimental apparatus. While reasonable, again, this argument does not prove or
disprove the above interpretation. We concluded then that in order to check our
above model conclusively, we must consider composites where b ≈ d. For those
composites we expect to see a Df = 2.5 value for high values of Ico , but also the
increase of Df towards the value of 3, as lower values of Ico are considered.
To test this idea we have measured a Ni–SiO2 composite45 in the so-called “di-
electric regime”,17 i.e. for v < vc (≈ 0.5), where the contribution of the non-nearest
neighbors is not expected to be negligible any more. Carrying out then the same
 August 5, 2004 11:59 WSPC/140-IJMPB 02533

Percolation and Tunneling in Composite Materials 2099

measurements as above on the Ni–SiO2 cermets for v = 0.35, where the condition
b ≡ R(GM) ≈ d is fulfilled, we found, as shown in Fig. 3(b), that unlike for the
CB-P composites, in this case Df increased, as predicted by our model, from 2.5
towards 3 with decreasing Ico . We note in passing that the better statistics (yielding
by SEOUL NATIONAL UNIVERSITY on 07/25/22. Re-use and distribution is strictly not permitted, except for Open Access articles.

the smaller error bars) in the present case (Fig. 3(b)) is provided by the much larger
“conducting” coverage of the surface in comparison with the former case (Fig. 3(a)).
However, for this case as well as in the former case we base our conclusion on the
reproducible findings that, within the experimental and computational accuracies,
the behaviors (on various composites and different areas on them) were always
the same, i.e. in the CB-P composites no dependence of Df on the Ico values was
Int. J. Mod. Phys. B 2004.18:2091-2121. Downloaded from www.worldscientific.com

found, while in the Ni–SiO2 cermets such a dependence was always observed. This
confirmation45 justifies our interpretation of the meaning of the percolation thresh-
old in the dielectric regime as the threshold of the network of nearest-neighbors
conduction which is a bona fide property of a percolation network. In fact, that

Fig. 4. The dependence of the macroscopic conductivity on the fractional Ni content in our Ni–
SiO2 cermets. The curves shown are fitted to the functional dependence of Eq. (1). Note the well
known classical percolation transition (at vc ≈ 0.52) and the newly discovered transition in the
“dielectric” tunneling-percolation regime (at vcd ≈ 0.18), for two different temperatures.
 August 5, 2004 11:59 WSPC/140-IJMPB 02533

2100 I. Balberg et al.

finding motivated us to look for a percolation-tunneling transition11 (as the one we

found in the CB-P composites) in the dielectric regime of the Ni–SiO2 cermets. This
was done by carrying out a careful examination of the v-dependence of the macro-
scopic conductivity σ. Indeed, our measurements revealed such a percolation-like
by SEOUL NATIONAL UNIVERSITY on 07/25/22. Re-use and distribution is strictly not permitted, except for Open Access articles.

behavior with a critical fractional Ni content of v = 0.18. This new transition in the
dielectric regime, that is exhibited in Fig. 4, has a percolation threshold that we
denote by vcd since it comes as an addition to the well known,17,18 bona fide, per-
colation transition at vc (i.e. for v = 0.52). The above described transition around
vcd = 0.18 suggests that we now have a continuous network of nearest-neighbor-
tunneling-resistors. This percolation cluster is probably similar to the skeleton of
Int. J. Mod. Phys. B 2004.18:2091-2121. Downloaded from www.worldscientific.com

the percolation network of the v = 0.35 sample, which constitutes the effective
percolation cluster at the highest Ico values (Fig. 3(b)). We note that as the tem-
perature decreases vcd is shifted to higher values (until for 4.2 K it coincides17 with
vc ) indicating, as stated above, that vcd or the establishment of ECNN, depends
not only on the geometry but also on the physical conduction mechanism, which is
tunneling under Coulomb blockade (TUCB) in the present case. Finally the values
we found for t (≈ 3–4) are also in excellent agreement with our above model and the
corresponding expectations from the non-universal theory for tunneling-percolation
systems as discussed below.11,13,15

3. Tunneling and Non-Universality

Now that we have confirmed that the CB-P system is, for all practical (electrical)
purposes, a percolation system, let us consider the fact that the value obtained for
t as defined by Eq. (1) was found13,14 in various CB-P composites to vary between
the well known universal value of tun ≈ 2 and values as high as 9. Such a deviation
from tun is known as the non-universal behavior in continuum percolation.4,11,21
The physical origin of this observation49 and its effect on the critical percolation
behavior50 were given many years ago, and can be described as follows.
Suppose that we have a resistor network on a lattice with a bond occupation
probability p and that it has a (normalized) distribution of the resistor values. We
know that in order to get connectivity a critical fraction pc of the resistors must
exist. If there is a larger fraction p1 (> pc ) of the resistors in this infinite system,
each random choice of pc resistors will provide an infinitely connected network. In
particular, this will be so for the pc resistors that have the lowest resistance values
in the sample, since there is no correlation between the position of the resistors and
their values. Correspondingly, that “pc sub-network” will yield the lowest resistance
possible in the sample. In a system that consists of a smaller resistors fraction, p2 ,
such that pc < p2 < p1 , the network of the lowest resistance that is made of a pc
fraction of the resistors must include resistors all of which are not smaller than those
used for the one in the p = p1 network. Thus, as the global system approaches the
percolation threshold from above, the resistance value of the “average” resistor in
the system increases. If the distribution does not diverge, there is a largest resistor
 August 5, 2004 11:59 WSPC/140-IJMPB 02533

Percolation and Tunneling in Composite Materials 2101

in the system, and thus the value of the average resistor will not be larger than
that of the largest resistor, and correspondingly, there will be no divergence in
the value of the average resistor’s as p approaches pc . On the other hand if the
resistors’ distribution diverges towards higher resistance values, at each step of the
by SEOUL NATIONAL UNIVERSITY on 07/25/22. Re-use and distribution is strictly not permitted, except for Open Access articles.

p2 approach to pc , larger and larger resistors must be involved in the network that is
necessary to yield the critical connectivity. Hence, we have a systematic divergence
of the “average” resistance as the threshold is approached. As this divergence comes
in addition to the increase of the network resistance due to the decrease of the overall
number of resistors in the system (as in universal percolation)1 we get a “stronger”
divergence of the conductivity as p → pc . The combination of the two effects yields
Int. J. Mod. Phys. B 2004.18:2091-2121. Downloaded from www.worldscientific.com

the behavior given by Eq. (1), but with a t value that will be larger than tun .
Of course, an appropriate diverging distribution that can account for a system
of the type described above must be normalizable. Such a “toy” distribution was
proposed first by Kogut and Straley50 who suggested the conductors’ distribution:
f (g) = (1 − α)g −α , (3)
where g is the conductance of a resistor (g > 0) and α is the parameter that
describes the “strength” of the divergence (0 ≤ α < 1). They found that for such a
distribution:
α
t = tun + , (4)
1−α
thus confirming the above intuitive suggestion and enabling a quantitative predic-
tion for the value of t. While Kogut and Straley gave the quantitative basis for the
expected t > tun values, it took six years until Halperin and co-workers51 were able
to show that such a distribution does indeed exist in porous media. The limits of the
t values in more general cases for which the geometry of the conducting elements
determines the value of α, were provided by one of us21 later.
Following the observation of non-universal t values in CB-P composites,
Balberg11 has mapped the “tunneling in a percolation system” model onto an f (g)
distribution such as in Eq. (3), where the Hertz distribution48 was assumed for the
nearest-neighbor distance r:
"   #
3
3r2 r
PH (r) = 3 exp − . (5)
a a

Here a is the average distance between the centers of two small (b  a) particles,
i.e. a = 2(3N/4π)1/3 , and N is the concentration of “metallic” particles in the
system. Since Eq. (5) does not allow a convenient mathematical treatment, the
above expression has been simplified phenomenologically to:11,13
 
1 r
P (r) = exp − , (6)
a a
which captures the essence of the problem, i.e. that the resistors that count are
those associated with the farther “nearest neighbor” particles of a given particle.
 August 5, 2004 11:59 WSPC/140-IJMPB 02533

2102 I. Balberg et al.

Under this distribution function and the tunneling conduction model that is given
by Eq. (2), one can apply the known general relationship for any two well behaved
distribution functions, F (x) and Q(y) that describe a system by two variables x
and y, when those are related to each other. This relationship can be written as
by SEOUL NATIONAL UNIVERSITY on 07/25/22. Re-use and distribution is strictly not permitted, except for Open Access articles.

dy
F (x) = Q(y) .
dx
In our case, the above variables are the inter-particle distance r, and the inter-
particle tunneling conductivity, g ∝ exp(−r/d). We thus obtain:
dr
f (g) = P (r) ∝ g (d/a−1) . (7)
Int. J. Mod. Phys. B 2004.18:2091-2121. Downloaded from www.worldscientific.com

dg
Following Eqs. (3) and (4), α is given by:
d
α =1− (8)
a
and consequently the non-universal value of t is given by:
a
t = tun + − 1 . (9)
d
More recently, we have justified the simplification given by Eq. (6) by deriving
the result given in Eq. (9) more rigorously.13 This was done by considering the
Nodes–Links–Blobs model. In that model the conducting system is composed of,
say, a cubic network of links, such that the basic resistor unit of the network is
the link and one determines the largest possible resistance value of the collection
of resistors in the link. In that approach the transition from Eq. (5) to Eq. (6) is
well justified physically, since we consider the inter-particle distances along a quasi-
one-dimensional chain rather than in a three-dimensional system. This transition
from a three-dimensional model to a quasi-one-dimensional one is manifested, of
course, by the decrease of the power of r from 3 to 1 in the transition from Eq. (5)
to Eq. (6).
In trying to compare the predictions of Eq. (9) to our experimental data on
CB-P composites we encounter two difficulties. The first is that if the condition
a  b is not fulfilled the distribution given by Eq. (6) in not necessarily applicable.
For example, considering that the fractional volume of the CB content is v and
that v = (4π/3)b3 N , we get that a = (2/v 1/3 )b. However this has meaning only for
v < 0.67 (that is the close packing density of spheres). If we apply this consideration,
the percolation threshold of vc = 0.4 that we found13 for the above CB-P composite
yields that, a = 2.7b. In particular, we noted13 that any distribution, between that of
Eq. (6) and that of a fixed inter-particle distance, can exist in real CB-P systems.
The second difficulty is that the value of d is not really known, but considering
its possible field dependence25 it is expected to be of the order of a nanometer.
This problem will be discussed elsewhere. In the comparison between the above
theory and our experimental results we found however a huge discrepancy.11,13
For example, for the almost spherical CB particles with a diameter 2b of about
 August 5, 2004 11:59 WSPC/140-IJMPB 02533

Percolation and Tunneling in Composite Materials 2103

250 nm37 we found13 experimentally that t = 6.4. Assuming13 as large a value of

d as 3 nm we get from Eq. (9) that t will be of the order of 75. This presents an
order of magnitude discrepancy with the experimental result. Not having another
distribution function the obvious improvement we can make within the framework of
by SEOUL NATIONAL UNIVERSITY on 07/25/22. Re-use and distribution is strictly not permitted, except for Open Access articles.

the above theory is to introduce the distribution function of the interface distances
between adjacent particles rather than the distribution function of the distances
between their centers. We replaced then Eq. (6) phenomenologically by:13
1 r − 2b
P (r − 2b) ∝ exp − . (10)
a − 2b a − 2b
Int. J. Mod. Phys. B 2004.18:2091-2121. Downloaded from www.worldscientific.com

With that modification our above phenomenological approach yields that α = 1 −

d/(a − 2b). This phenomenological approach was justified very recently by Grimaldi
and co-workers,15 who were able to derive such an expression rigorously. Following
the above we can predict then that
a − 2b
t = tun + −1. (11)
d
This is a general expression that for perfect spheres with no coalescence yields (if
the individual particle size is conserved with increasing v) that
a
t = tun + (1 − v 1/3 ) − 1 , (12)
d
or, more practically,
 
2b 1
t = tun + (1 − v 1/3 ) − 1 . (13)
d v 1/3
We should note here that from the theoretical point of view we can replace v in
Eq. (13) by vc since this expression is “exact” only when v approaches vc .
Returning to our experimental result of t ≈ 6.4, we find that while the above
modification of the theory (Eq. (13)) yielded a considerable improvement of the
theory (it predicts the value of t to be 26) the latter is still not in a reasonable
agreement with the experiment. In contrast with the large discrepancy of our theory
with our experimental results, the theory seems to fit well the experimental result
of Ali and Abo-Hashem14 who found in their samples a value of about 9 for t. This
result was obtained for CB particles with b = 15 nm and vcd = 0.17, that according
to our Eq. (13) should yield, with a d value of 3 nm, a value of t ≈ 8. While at present
this agreement cannot be considered “excellent”, since we do not know the exact
value of d (though it is expected13,25 to be of the order of 3 nm), the trend of a better
agreement with the theoretical predictions as the system is better described by the
theoretical model indicates that the essence of our model is confirmed. The fact that
the value of t varies so much from one CB-P composite to another indicates that the
inter-particle distance distribution probably depends on the preparation conditions
of the composite.11 In particular the proximity of the particles is not determined
statistically, as suggested by the “theoretical” distributions that were used above,
but rather by the rheological properties of the mold during the preparation of the
 August 5, 2004 11:59 WSPC/140-IJMPB 02533

2104 I. Balberg et al.

composite. For example, we may consider the polymer layer that coats the CB
particles to have an almost constant thickness that is of a couple of nm’s.32,52 This
brings about an almost constant tunneling distance between adjacent particles and
thus leads to the universal value for t. However, no direct proof was obtained to
by SEOUL NATIONAL UNIVERSITY on 07/25/22. Re-use and distribution is strictly not permitted, except for Open Access articles.

this suggestion13 thus far.

The situation with the estimate of d is somewhat better in the granular cermets
and a d value of about 1 nm is better established.17 Following these considerations,
and in order to further establish the validity of our model and explanation of the
above observations on CB-P composites,13 we turned to the Ni–SiO2 composites
since only now, after the conclusions of Section 2, we can consider the value of t in
Int. J. Mod. Phys. B 2004.18:2091-2121. Downloaded from www.worldscientific.com

this system within the framework of our above model. For the granular cermets it
was clearly shown17 previously by the temperature dependence of the conductivity
in the dielectric regime that the TUCB governs the transport. On the other hand
we have found, as shown in Fig. 4, that there is a percolation transition in this
“dielectric regime” with exponents of the order of 4. Looking at the TEM pictures 17
in the regime of 0.18 ≤ v ≤ 0.35, one sees that there is a considerable clustering
of the particles such that a − 2b is almost a constant in this regime and that it
is of the order of 2–3 nm. Of course, the crucial largest inter-particle distance in
the links of the percolation network, that are expected to connect the particles, is
always larger than a − 2b. Considering17 a d ≈ 1 nm value (that will be discussed
in detail elsewhere), Eq. (11) yields a value of t ≈ 4. This is in agreement with
the experimental values shown in Fig. 4. Further support to our theory comes
from the temperature dependent values of t, exhibited in Fig. 4. With TUCB, the
increase of temperature will make the preferred conduction route the one between
smaller particles (larger Coulomb blockade) but thinner barriers.17 This suggests
that for the same v, the a − 2b operational value, and consequently the value of t,
will decrease with increasing temperature. While preliminary in nature, the results
shown in Fig. 4 appear to follow this expected trend, yielding further support to
the theory13,15 and the above explanation for the large discrepancy that we found
between theory and experiment in some CB-P composites.

4. The Local Nature of the Electro-Thermal Switching Process

The huge increase of the resistance of CB-Polymer composites upon the increase of
temperature through the melting-softening point of the polymer Tsm is well known
and was extensively studied for many years.53,54 This “inverted” insulator-to-metal
transition that is characterized by a positive temperature coefficient of the resis-
tance (PTCR) is well correlated with the much larger thermal expansion coefficient
of the polymer in comparison with that of the CB particles in that temperature
range,38,53 and thus it is well accepted that this PTCR is associated with the greater
separation of the CB particles as the system is heated up through Tsm .53,54 Not-
ing that the composite goes through about a 10% linear expansion53 when heated
through Tsm , one popular explanation55 for the PTCR is that the above thermal
 August 5, 2004 11:59 WSPC/140-IJMPB 02533

Percolation and Tunneling in Composite Materials 2105

expansion causes simply the homogeneous decrease of the effective fractional vol-
ume occupied by the carbon black and this shifts the system below its “percolation”
threshold. Another explanation suggested is that of a pure tunneling effect, i.e. the
increase in the average inter-particle separation and thus the exponential increase
by SEOUL NATIONAL UNIVERSITY on 07/25/22. Re-use and distribution is strictly not permitted, except for Open Access articles.

in the inter-particle resistance of the individual resistors in the system.24,29 A third

possible explanation is essentially a combination of the above two.30,53,56 – 58 Also,
in these59 and similar60 systems, it was assumed that this adjacent-particle sep-
aration causes a drastic rupture of some critical resistors, thus jeopardizing the
electrical connectivity of the system.16 The percolation-tunneling model that we
proposed in Section 2 seems to provide a unified, though at this stage qualitative,
Int. J. Mod. Phys. B 2004.18:2091-2121. Downloaded from www.worldscientific.com

picture of the PTCR. The total volume expansion causes in particular the increase
of the nearest neighbors separation, which according to our model means that the
network of ECNN becomes diluted. In fact, this explanation is consistent with the
model of Section 3 as the corresponding t values in the CB-P composites were
found53 to increase with temperature, in agreement with the expected increase of
the corresponding a − 2b.
Turning to the electrically driven switch, we note that while the PTCR can
be described as “not fully” understood57 the understanding of the electro-thermal
switching, that is associated with the PTCR, is even at a more rudimentary level.
Moreover, while there are more questions to be answered in the latter case there
were (at least in the open literature) only very few published works on the switching
mechanism. While the origin of the phenomenon is generally agreed16,61 – 64 to be
the same as that of the PTCR effect, one must consider now also the generation
of the Joule heating in the CB particles and its dissipation in the polymer envi-
ronment. Most of the works that tackled the problem tried to explain the electrical
switching characteristics applying models of the heating by the macroscopic current
and the macroscopic-phenomenological PTCR effect.7,61 – 63 However, recent obser-
vations indicate54 that this is not enough and that these do not yield any insight
into the microscopic nature of the phenomenon. For example, one notes that the
macroscopic switching can be observed at low fields for which the global temper-
ature is still much below Tsm .61 The description of the corresponding microscopic
system is quite complicated since one has to take into account the balance between
the macroscopic current, the local heat supply and the local heat dissipation in the
system. Indeed, while the switching phenomenon has been discovered a long time
ago7,54 and in spite, or maybe because, of the (patented) commercial application,7
there were only handful of reports that tried to explain more deeply the opera-
tion of the device.16,64 These models, that are of a phenomenological nature, and
that considered local nonlinear resistors in a collective network did not make an
explicit association with the fundamental percolation and tunneling features that
we have described in the previous sections. On the experimental end the situation
appeared even worse as corresponding studies were limited to “quality control”-like
tests,65 and as there was so far no direct support for any of the existing macroscopic
or microscopic models of the effect. In particular the current network, the local
 August 5, 2004 11:59 WSPC/140-IJMPB 02533

2106 I. Balberg et al.

current values and the local heat dissipation on the individual CB particle scale
have not been monitored. Hence, even the most fundamental questions regarding
the local and global switched states have not been answered, and the time scales
involved in the local processes have not been estimated. Following these consid-
by SEOUL NATIONAL UNIVERSITY on 07/25/22. Re-use and distribution is strictly not permitted, except for Open Access articles.

erations we have carried out a microscopic local study66 of the switching process,
mapped the corresponding currents, evaluated the heat dissipation process and in-
terpreted the results in terms of a percolation-tunneling model. However, before
turning to our results let us describe briefly the basic features and previously given
explanations of the macroscopic switching phenomenon.
In general the electro-thermal switch consists of a CB-P composite layer (0.5 mm
Int. J. Mod. Phys. B 2004.18:2091-2121. Downloaded from www.worldscientific.com

thick) that is encapsulated between parallel laminated metallic foils that serve as
its electrodes.7,16,65,66 This is a very simple, low-cost, device that is used as a
series “resetable” fuse in various low current circuits such as computers and small
motors. Its main function is to limit the current through the corresponding circuit
in cases of a voltage or a current surge beyond preset values, Vc and Ic . The I–V
characteristics of the device, such as those shown for one of our devices in Fig. 5,
indicate that up to a certain applied bias, Vc , or an applied current, Ic , the device
acts as small Ohmic resistors. As the applied voltage V exceeds Vc the current

Fig. 5. Typical current-voltage characteristics of CB-Polymer based switching device. The cor-
responding macroscopic current values were measured at each bias; 1 second, 30 seconds and
300 seconds after the application of the voltage.
 August 5, 2004 11:59 WSPC/140-IJMPB 02533

Percolation and Tunneling in Composite Materials 2107

rather than exceeding Ic drops with the increase of the applied voltage7,16,61,65,66
(constant voltage conditions). If the device is current driven, one finds that for
I ≥ Ic the resistance of the system increases “indefinitely” with time, yielding a
current saturation that is usually described in terms of an electric breakdown or
by SEOUL NATIONAL UNIVERSITY on 07/25/22. Re-use and distribution is strictly not permitted, except for Open Access articles.

a switching-off behavior.16,67 The macroscopic qualitative explanation of the latter

behavior is quite clear; since the material has a very large PTCR around T sm , once
there is enough heat generation in the system, Ic2 Rc (where Rc = Vc /Ic ), the overall
temperature increases and so does the sample’s resistance R. Consequently for
I ≥ Ic , I 2 R will yield an avalanche-like breakdown of the conduction via the device.
The basic understanding of the drop of the current (when V exceeds Vc ) under
Int. J. Mod. Phys. B 2004.18:2091-2121. Downloaded from www.worldscientific.com

constant voltage conditions, with which we are concerned here, is somewhat more
complicated. The increase of power dissipation beyond Ic Vc causes a global net
heating of the device, under which the system (or, as we show below, a part of
the system) has been transformed to the melting-softening phase. At that point
the global resistance of the sample increases, but for a given V > Vc the current I
(< Ic ) is such that the melting-softening state is maintained. We note that in that
case the heat generation decreases with increasing V yielding a negative differential
resistance68,69 (NDR). The time constant involved in Fig. 5 is simply the time it
takes the sample to achieve its global stationary state under which no net heat is
supplied to the device. We note however that the behavior exhibited by the I–V
characteristic in Fig. 5 does not require a priori any macroscopic inhomogeneity in
the temperature of the system but the NDR behavior will become more pronounced
in the latter case.61 – 63 Physically, the essence of the latter models is that in a
PTCR sample the heat dissipation is least effective in the innermost part of the
device. For example, in a cylindrical device, the center of the cylinder axis (which
is the direction of the current flow) will be the hottest point in the device. Being
such, it will form a high temperature–high resistance region, that is known in the
literature as the “hot zone”,64,67 where the latter is believed to play a major role
in determining the characteristics of the switching phenomenon. This phenomenon
and its comparison with the NDR behaviors that were observed in semiconductor
devices68,69 will be considered elsewhere. In passing we note that experimental
evidence for the existence of such a macroscopic hot zone was found already in the
1970s by a thermo-chromatic paint53 and more recently by IR thermal imaging.66,67
In order to better understand the above described phenomenon we have tried in
our microscopic study to answer the following fundamental questions regarding the
electro-thermal switching in the CB-P composites: is that effect a global material
effect or is it a result of compounding local switching events that can be separated
from each other such that each of them can be evaluated individually? If the latter
is the case, what are the time scales involved in the local processes and what is
their relation to the time scale involved in the macroscopic processes? Recalling
then the macroscopic observations, does the collection of events represent, under a
given bias, a dynamic situation of ever opening and closing individual resistors in
the network, or is there a fixed network of switched-off inter-particle resistors?
 August 5, 2004 11:59 WSPC/140-IJMPB 02533

2108 I. Balberg et al.

To carry out our study we have applied our C-AFM to acquire local I–V char-
acteristics and to measure the voltage dependent maps of the current distribution
through the sample.66 In parallel, we mapped the topography and carried out me-
chanical sensitive measurements (friction and stiffness) that can distinguish between
by SEOUL NATIONAL UNIVERSITY on 07/25/22. Re-use and distribution is strictly not permitted, except for Open Access articles.

the CB and polymer phases at the surface. The good correlation found between the
latter material-sensitive measurements and the current maps, and the fact that
the tip is smaller than the CB particle, indicate that the current that reaches the
counter electrode in our measurements is always initiated at a single CB particle
that protrudes the surface of the sample.66 We have further certified that (unlike
the case in metal–polymer composites65 ) there is no junction that can switch be-
Int. J. Mod. Phys. B 2004.18:2091-2121. Downloaded from www.worldscientific.com

tween the tip and this particle,66 and thus, the corresponding current path starts
at that particle and continues to the electrically connected chain of CB particles
(thus constituting an “initial dead-end” of the percolating cluster, see Figs. 1 and
2). The dead-end merges then the backbone of the conducting network which is
terminated at the counter electrode.
We started out our study by taking images of the “current islands” that are
obtained as the tip scans the surface of the CB-P composite sample finding that
initially the number of the conducting islands and their areas increase considerably
with the increase of applied bias. Then, beyond a certain critical voltage Vcv , these

Fig. 6. (a) 5 × 5 µm2 current images taken on a CB-Polymer sample for four different volt-
ages. Gray scale is 0 to 30 nA. (b) The applied voltage dependence of the computed total area
that is covered by “conducting islands” in images such as those shown in (a). (c) Typical I–V
characteristics that are observed on various “current islands”.
 August 5, 2004 11:59 WSPC/140-IJMPB 02533

Percolation and Tunneling in Composite Materials 2109

two start to decrease with further increase of the bias. This behavior, that is exhib-
ited in Fig. 6(a), shows that for low values of the applied voltage the bias acts to
open local current routes in the system but then, beyond Vcv (∼ 5 V) it acts to close
them. As in the case of the two issues discussed in Sections 2 and 3, we tried to get
by SEOUL NATIONAL UNIVERSITY on 07/25/22. Re-use and distribution is strictly not permitted, except for Open Access articles.

the basic understanding of the phenomenon by carrying out similar measurements

on the Ni–SiO2 system. In the present context this was done in order to check
the most fundamental assumption regarding the PTCR and the above described
switching effect, i.e. to check if the above switching phenomenon is due indeed to
the induced thermal expansion of the insulating matrix when a voltage is applied,
or is it a more general voltage induced switching phenomenon in composites? We
Int. J. Mod. Phys. B 2004.18:2091-2121. Downloaded from www.worldscientific.com

note here that in contrast with the CB-P composites, in which a PTCR effect is well
known to exist, in the granular system there is no significant difference in the ther-
mal expansion coefficient of Ni and SiO2 and there is no PTCR of this or any other
source. Our measurements on the Ni–SiO2 composites revealed indeed a different
behavior of an ever increasing “conducting islands” area with increasing bias.
Turning back to the CB-P composites, we show in Fig. 6(b) the observed de-
pendence of the “conducting island area” At on the voltage applied at the tip. As
well seen, the shape of this dependence is very reminiscent of the macroscopic I–V
characteristic shown in Fig. 5. This behavior persists under the C-AFM constant-
voltage conditions, up to about 5.6 V where At increases again. Since the appear-
ance and disappearance of the “islands” represents an accumulation of uncorrelated
local effects (taking place during the scans of the surface at different times and for
CB particles far away from each other) it appears that the decrease of the global
current, for V > Vc (see Fig. 5), can be interpreted simply as a result of many inde-
pendent current closing routes rather than a global macroscopic effect such as that
of the macroscopic “hot zone”. Hence, at least for the global low-power generation
scenario, the boundary conditions of the sample have little effect on the switching,
and instead of hot zones in the sample we have a collection of local switches at
locations determined by their resistance and “position criticality”. Such critical po-
sitions are the “connecting dead-ends” or the effective singly-connected-bonds 1 in
the electrical backbone of the system. Considering the particular CB particle size
and geometry we estimate that the corresponding current densities within such a
local path in the “critical resistors” are of the order of 100 A/cm2 . Since the tip
scans a pixel within about one msec, we conclude that a local switching event takes
place within that time. Considering the fact that in the macroscopic I–V charac-
teristic the stationary state takes 30 sec to form, we conclude that this stationary
state is built from a collection of inter-particle resistors that switch-off individually
within a msec.
To further understand the local nature of the results in terms of the electro-
thermal switching effect we carried out local (or “one island”) I–V measure-
ments. The corresponding results, shown in Fig. 6(c), demonstrate that a complete
switching-off of the local current takes place in each switching event. This is an
even more direct evidence that the application of the voltage at the “beginning”
 August 5, 2004 11:59 WSPC/140-IJMPB 02533

2110 I. Balberg et al.

(CB particle) of a “dead end” is causing a complete switching-off event somewhere

along the path that connects it to the backbone, or somewhere at a “critical”
(e.g. singly-connected-bond1) resistor of the backbone. We also note that since the
corresponding I–V scan takes 0.1 sec, the local switching event can be estimated
by SEOUL NATIONAL UNIVERSITY on 07/25/22. Re-use and distribution is strictly not permitted, except for Open Access articles.

to take about a msec. This provides further support for the local switching time
that we have estimated above. In addition, as can be seen in Fig. 6, the switched-off
state, in some locations, is followed at some higher voltage, by an opening of an
alternative current path that starts at the same location. The latter phenomenon
suggests that the macroscopic state can be viewed as a dynamic “liquid flow pro-
cess” in a Cayley tree,1 in which there are junctions that are opened or closed even
Int. J. Mod. Phys. B 2004.18:2091-2121. Downloaded from www.worldscientific.com

for a constant value of the applied global bias.

To evaluate the above characteristics of the local switching phenomena within
the framework of our percolation-tunneling model we have measured the “local
switching threshold” Vcv (the voltage at the first peak in Fig. 6(b)) as a function of
the CB content in the composite. The basic expectation within this framework is
that the higher the CB content v the smaller the average inter-particle (in particular
the “nearest neighbor”) distance. Thus, not only will the tunneling current between
two neighboring particles be larger, but also that, since less polymer material is
confined between them, less heat will be required for the onset of the switching. In
other words, the threshold voltage is expected to decrease with an increase of v.
Indeed our observations have confirmed these expectations.66
All the above yields then the following picture of the operation of the CB-
P switch under the application of a bias. As tunneling is the major conduction
mechanism, above a certain applied tip voltage Vcv , a single individual tunnel-
ing barrier is completely switched off. This is due to the heat generation in the
corresponding conducting particle such that the inter-particle polymer layer, in
which the heat is dissipated, expands. The current carrying segment is then either
a “current-leading dead end” (that “begins” at a surface-protruding CB particle)
or a “singly-connected-bond-like” tunneling junction in the percolation backbone.
The well known global switching effect starts then with many such switching events
that are uncorrelated openings and closings of many tunnel junctions. Above some
threshold voltage the overall macroscopic current decreases when the effect of the
switching events dominates the global current-increase that is associated with the
opening of other junctions. The onset of tunneling is of course instantaneous, but
for the switching-off time that is associated with a thermal effect we have an ex-
perimental estimate of an upper bound, which is of the order of a msec. The time
evolution of the global stationary state that involves local cooling processes can be
estimated from the macroscopic behavior shown in Fig. 5, i.e. it is of the order of
tens of seconds (depending on the geometry of the device). This estimate of the
time it takes the global stationary state to form, is also in accordance with the
time that it takes for the global avalanche switching effect to occur under con-
stant current conditions.16 The fact that we see the local openings of alternative
current paths further suggests that the corresponding stationary state represents
 August 5, 2004 11:59 WSPC/140-IJMPB 02533

Percolation and Tunneling in Composite Materials 2111

a dynamic situation of different ensembles of switches at each time. We see then

that our results are well understood in terms of the tunneling-percolation model
and the corresponding local heating and local cooling events. In turn we can better
understand the huge macroscopic PTCR. This is in view of the fact that the local
by SEOUL NATIONAL UNIVERSITY on 07/25/22. Re-use and distribution is strictly not permitted, except for Open Access articles.

PTCR effect is associated with tunneling conduction, which is very sensitive to

the nearest neighbor inter-particle distance, and to the presence of “critical” local
resistors that are known to exist in percolation networks.

5. Percolation Phenomena in Microcrystalline Silicon

Int. J. Mod. Phys. B 2004.18:2091-2121. Downloaded from www.worldscientific.com

The above parts of our review dealt with systems that, for all practical purposes,
could be described as two phase systems where one phase is made of a metallic-like
material with a finite conductivity, and the other of an insulating-like material that
is, in practice, of zero conductivity. The next degree of complexity is percolation
systems in which the two phases have finite but different conductivities. The sim-
plest way to describe this case is by the effective medium approximation70 (EMA).
This approach becomes inappropriate if the system is not made of uniform regular-
shaped elements of the conducting phase. In particular it does not account for the
size and shape (such as sheet-like2 ) distributions of the conducting and/or insulat-
ing elements. Moreover, this approach does not consider the possible inter-particle
resistance distribution (such as of potential barriers in tunneling transport). All
the above mentioned properties determine the observed percolation threshold2,3,20
while the latter (see Section 3) may also affect the critical behavior.11 – 15 The situ-
ation becomes even more complicated if the above distributions themselves depend
on the relative content of the various conducting elements in the system such as
the fractional content of the conducting elements, and/or if there are more than
two competing conducting networks. Such complicated situations are encountered
in particular in composites where at least one phase is made of a semiconducting
material. In this section we would like to show that the understanding that we
have gained from the above described percolation-tunneling model,11,13,45 our “ex-
cluded volume” theory2,3 and the above described experimental techniques that we
applied44,45,66 can be useful in deriving the so far lacking basic understanding of
such complicated systems.
The system of hydrogenated microcrystalline silicon (µc-Si:H) has all the in-
gredients of the complexity described above.71 – 73 For this system the terms
percolation74,75 and tunneling76,77 have been very frequently used but usually in a
very loose way. On the other hand, almost every possible transport route and every
possible conduction mechanism were proposed to explain the various fragments of
the many macroscopic electrical data that were taken. In particular, the numerous
data and the many interpretations72,78 did not provide a comprehensive under-
standing of the dependence of the conductivity of µc-Si:H on the microcrystallites
volume fraction in that system. Following the above sections we note that by this
definition of v we introduce the hidden assumption that the microcrystallites are
 August 5, 2004 11:59 WSPC/140-IJMPB 02533

2112 I. Balberg et al.

the most conducting elements in the system. As we will show below this is not gen-
erally true and v serves here as a convenient, but not always justified, “percolation
proximity” parameter.
To appreciate the complexity of the problem let us describe briefly the structure
by SEOUL NATIONAL UNIVERSITY on 07/25/22. Re-use and distribution is strictly not permitted, except for Open Access articles.

of the µc-Si:H system.71 – 73 Under the common deposition conditions of the µc-
Si:H films one finds that usually up to about v ≈ 0.5 (depending on the deposition
conditions)79 the films contain small isolated spherical-like crystallites (10–100 nm
in diameter) that are embedded in an hydrogenated amorphous silicon (a-Si:H)
matrix.80 – 82 As v increases the crystallites aggregate in columns that have the
form of cones.72,73,80 The diameter of the cones is in the 0.1–0.3 µm range and
Int. J. Mod. Phys. B 2004.18:2091-2121. Downloaded from www.worldscientific.com

their concentration appears to increase with the film thickness.83 It appears then
that as v approaches the value of 1.0 the columns approach a cylindrical shape. 84 We
have then altogether that the µc-Si:H system consists of four different “phases”, mi-
crocrystallites, microcrystallites-columns, hydrogenated amorphous silicon (a-Si:H)
matrix and voids. Between any two of these “phases” there is a disordered silicon
tissue that may have its own (different or similar to a-Si:H) structural and electrical
properties. In addition there can be two more types of disordered silicon tissues, one
between adjacent crystallites and the other between adjacent columns,72,78 six dis-
ordered tissues altogether. The tissues that are known to be thin85 (in comparison
with the sizes of the elements of the phases involved) may act then, in principle, as
either, connected networks that form conduction routes of their own, or, as barriers
through which the charge carrier passes between the more conducting elements.
Let us review briefly but critically the expected evolution of the transport net-
works and the corresponding conduction mechanisms that can exist in µc-Si:H as a
result of the increase of v. For very low values of v we have isolated crystallites (IC)
embedded in the relatively insulating a-Si:H matrix. In the case of metallic-like IC
this system can be well described by the EMA approach,70 where the conductiv-
ities of the a-Si:H phase, σa , and the IC, σc , are combined to yield the resultant
conductivity, σe (v), of the system. Consider that a-Si:H and the IC are semicon-
ductors, there may be inter-crystallite potential barriers involved77,86 and then the
value of σe (v) is only the (high temperature) limit of the measured conductivity.87
Considering the fact that, as in the CB-polymer and the granular metals in the
dielectric regime (vcd < v < vc , see Sections 2 and 3), the silicon crystallites do
not coalesce (with increasing v) it is expected that the effect of potential barriers
in the system may be of importance. In some works the existence of inter-phase88
or inter-crystallite89,90 barriers was concluded and in others it was considered to
be negligible.91,92 These discrepancies can be explained by assuming the barri-
ers to depend on the particular conditions under which the µc-Si:H samples were
prepared.93
As v increases, the “touching” of two IC needs to be more carefully exam-
ined. If the IC are metallic-like, one would expect “continuous band transport”94
which essentially means a tunneling-percolation threshold and (depending on the
size of the IC and temperature) possibly a TUCB.81,95,96 Following Section 2,
 August 5, 2004 11:59 WSPC/140-IJMPB 02533

Percolation and Tunneling in Composite Materials 2113

we expect that the observed percolation threshold will be lower than the one
needed for a continuous path of IC that coalesce into large crystallites (see v c in
Section 2), or the one predicted by EMA,70 as was indeed observed.97,98 Comparing
these observations with the results mentioned above for granular metals it is not
by SEOUL NATIONAL UNIVERSITY on 07/25/22. Re-use and distribution is strictly not permitted, except for Open Access articles.

surprising that the large conductivity (percolation transition-like) jumps in doped

µc-Si:H systems were found in the 0.2 ≤ v ≤ 0.3 range and that the mechanism
of TUCB has been proposed.81,95,96 There were however two alternative explana-
tions for the percolation-like transition in that range. One of them considered EMA
and thermionic emission above the inter-crystallites barriers,87 and the other89 that
assumed a Scher and Zallen-like model3 where adjacent IC are “guaranteed”3 to
Int. J. Mod. Phys. B 2004.18:2091-2121. Downloaded from www.worldscientific.com

touch. These two models account however poorly for the above described structure
of the system. In the first model87 the definition of the connectivity of the system,
due to the presence of various size barriers (as we can conclude from Section 3)
is ill defined. In the other model89 the equal size cubic elements of both the IC
and the amorphous matrix, as well as the “guaranteed” touching of the adjacent
particles, provides a very idealized picture of the system at hand. On the other
hand, the experimental microscopic proof 99 that the current goes through the IC
in doped µc-Si:H and the corresponding sharp rise of the conductivity72,87,89 in
the 0.2 ≤ v ≤ 0.3 range, show that by a phenomenological extension, such as
the one proposed20 for a certain kind of granular metals, these models can serve
only as rough approximations to the more detailed TUCB model, which (following
Section 2) is re-endorsed here. In fact our tunneling-percolation picture should be
also appropriate for undoped µc-Si:H as long as σc  σa . The observation87,90
of the “jump” of the conductivity in that v range, that has not been discussed
previously in terms of the percolation-tunneling model, appears to be understood
now. In particular, the conductivity “jump” in the doped material should be, in
view of the known conductivities of the single phases, as observed, i.e. between σa
(10−8 (Ohm cm)−1 ) and σc (10−1 (Ohm cm)−1 )) in the doped87,100 µc-Si:H material
and between σa (10−11 (Ohm cm)−1 )) and σc (10−9 (Ohm cm)−1 ) in the undoped
case101,102 (see also below). We denote this tunneling-percolation component of the
conductivity by σtp (v).
All the above models assumed that except for the a-Si:H phase and the conduc-
tivity dominant IC-network, no other conducting phases are present in the system.
However, being concerned with a semiconductor, we must consider the possibil-
ity that if there is a percolating network of the crystallites there is also a per-
colating network of the crystallite–crystallite interface tissue. If this tissue (that
was considered above only as providing a potential barrier) has a conductivity σ t
that is large enough to yield a conducting network that has a higher conductance
than the above outlined inter-crystallite route, the above description may have to
be replaced by a porous media-like model where the conduction takes place in
sheet-like tissues.2 This possibility, which was suggested first by Schellenberg and
McLeod103 for µc-Si:H in 1988, and by one of us in the context of porous silicon ten
years later,104 was largely ignored for µc-Si:H. This was in spite of the numerous
 August 5, 2004 11:59 WSPC/140-IJMPB 02533

2114 I. Balberg et al.

macroscopic data that suggested conduction via a disordered amorphous-like tissue

in various µc-Si:H systems.72,78,105 On the other hand we note that the microscopic
experimental evidence99 that in doped µc-Si:H the current flows through the IC
indicated that at least for these systems an intergrain hopping, TUCB-like, mech-
by SEOUL NATIONAL UNIVERSITY on 07/25/22. Re-use and distribution is strictly not permitted, except for Open Access articles.

anism should be dominant. On the other hand, the above mentioned conductivity
transition in undoped µc-Si:H101,102 appears to be harder to resolve and in princi-
ple it may be associated with conduction in the disordered intercrystallite silicon
tissue. However, a closer examination of the available data72,78,105 indicates that
the extended states-like and variable-range hopping-like mechanisms were all ob-
served for v values larger than 0.5. Also the fact that the conductivity transitions
Int. J. Mod. Phys. B 2004.18:2091-2121. Downloaded from www.worldscientific.com

in both systems is in the 0.2 < v < 0.3 range (as we found above in Sections 2
and 3 and as expected from a EMA approach)20 suggests that in that range the
TUCB-like transport mechanism applies also to the undoped material. In contrast,
for a “tissue” network we would expect103 a different (probably a lower) percolation
threshold2,3 (that is determined by the tissue’s volume and not by the IC volume
as in the cases of Sections 2 and 3).
Let us turn now to the conductivity dependence on v above the v = 0.3 range.
Many data for undoped µc-Si:H in the latter range show72,106 a large conductivity
“jump” typically between 10−9 and 10−3 (Ohm cm)−1 around v ≈ 0.7. Consider-
ing the structural observation of the formation of the crystallites columns,73 it was
suggested that this conductivity transition is due to the formation of a continuous-
network of columns that yields a percolation-like behavior.72 The natural question
that was then raised107 is whether the conduction takes place through the columns
(i.e. through the encapsulated IC) or does it involve the continuous network of the
tissues that encapsulate the columns? In other words, which part of these column
networks carries the dominant contribution to the current flow? The above two al-
ternatives could not be simply decided between in view of the available conflicting
data. On the one hand the microscopic data of Resek et al.108 have been inter-
preted as showing currents through the columns (i.e. through the crystallites that
constitute them) while, on the other hand, the macroscopic data of that group,72
of ours105 and of many others78 indicated transport in a disordered tissue. Hence,
the question of where does the current flow in µc-Si:H has been left unresolved.
Following this puzzle we have carried out a comprehensive C-AFM and Scan-
ning Tunneling Spectroscopy (STS) study.109 We started by examining whether
the column tissue itself has a high conductivity in comparison with the crystal-
lite aggregates within the columns and/or the a-Si:H matrix and/or the (out of
the columns) IC network. This was done by using a thin (0.3 µm thick) film that
was deposited on a quartz substrate coated with a Cr/Au layer that served as the
counter electrode to the scanning tip. In this “vertical” transport configuration the
deposited film (that was prepared with the hot wire technique105,106 and under 80%
H2 dilution) was of low enough value of v (≈ 0.2) to avoid lateral conduction be-
tween the columns but high enough value of v to have some well formed (separated)
columns. The images shown in Fig. 7 demonstrate that indeed the current flows in
 August 5, 2004 11:59 WSPC/140-IJMPB 02533

Percolation and Tunneling in Composite Materials 2115

by SEOUL NATIONAL UNIVERSITY on 07/25/22. Re-use and distribution is strictly not permitted, except for Open Access articles.
Int. J. Mod. Phys. B 2004.18:2091-2121. Downloaded from www.worldscientific.com

Fig. 7. 2.5 × 2.5 µm2 topographic (right) and current (left) images measured on a µc-Si:H sample
with a few columns of silicon crystallites. The current image was acquired in the “vertical” config-
uration (current parallel to the columns growth). The correlation between the two images depicts
the dominant conduction via the disordered tissue that encapsulates the crystallites columns. The
gray scales are 100 nm (topography) and 0–0.03 nA (current).

the columnar-disordered tissue, as the “rings” in the current image and the columns
diameter, that we found independently from our Scanning Tunneling Microscopy
(STM) measurements,109 are both about 0.1 µm. In fact the size of the IC that
is about 20 nm (as estimated from our Raman spectroscopy measurements)105,106
was confirmed in our STM images.109 The results of Fig. 7 clearly show that the
contribution of the column’s tissue conductivity dominates that of the crystallites
within the columns and that of the a-Si:H matrix whether it does or it does not
contain IC.
In order to examine the current flow map for high v values and compare them
with the above mentioned macroscopic data that suggested transport through the
disordered tissue, we examined films, with v ≈ 1, in both the “vertical” and “lateral”
configurations. The results obtained in the “lateral” transport configuration, which
were qualitatively the same as those measured in the “vertical” configuration, are
exhibited in Fig. 8. Again, it is very apparent that the current flows through the
columnar tissues. However now, from the examination of the lateral configuration,
it is also clear that the “touching” columnar tissues form a “lateral” percolating
network. It is thus also apparent that the large “jump” observed82 in the lateral
conductivity for v ≈ 0.7 is associated with the formation of that connected network.
For samples with v values well below the onset of columnar tissue connectivity
(v < 0.5) the current in the lateral configuration was found to flow through the
a-Si:H matrix. The typical current values were, however, considerably smaller as
compared to those flowing through the disordered tissues that are shown in Fig. 8.
At this point we can summarize our findings for the v ≥ 0.5 regime as follows: the
macroscopic study105 concluded a conduction mechanism that is associated with
a disordered silicon tissue that is not a-Si:H-like, and the microscopic study109
 August 5, 2004 11:59 WSPC/140-IJMPB 02533

2116 I. Balberg et al.

by SEOUL NATIONAL UNIVERSITY on 07/25/22. Re-use and distribution is strictly not permitted, except for Open Access articles.
Int. J. Mod. Phys. B 2004.18:2091-2121. Downloaded from www.worldscientific.com

Fig. 8. 2 × 2 µm2 topographic (right) and current (left) images measured on a µc-Si:H sample
that did not contain an a-Si:H phase and on which the macroscopic phototransport measurements
were carried out. The “grains” in the topographic image were shown (by STM measurements) to
be the tops of the crystallites columns. The current is shown here to flow through the tissues that
encapsulate the columns. The electrical circuit is closed in the corresponding measurement in the
direction perpendicular to the columns’ growth. The encircled areas emphasize the close relation
between the two images indicating a lateral continuous-conducting network. The gray scales are
40 nm (topography) and (0–1.5 nA) current.

concluded that the dominant conduction route is the columns encapsulating

tissue. These two conclusions are consistent and they yield a full picture of the
high v percolation transition suggesting that the conductivity σct associated with
that columnar tissue, must be large enough such that it dominates the conductivity
that is associated with the networks (in or out of the columns) of the individual
crystallites σtp .
The above analysis can explain the fact that the transition at v ≈ 0.7 has not
been noticed in doped µc-Si:H, as due to the fact that in that case σtp > σct and
thus the contribution of the column’s tissue is not manifested in the σ(v) behav-
ior. More importantly, the above analysis may explain the discrepancy between
the macroscopic and microscopic results of Rezek et al. They found72 tissue-like
conduction in their macroscopic measurements, for regularly deposited samples,
but conduction that appeared to take place through the crystallites in their micro-
scopic measurements,108 for samples deposited at ultrahigh vacuum. Following the
previously suggested efficient and strong oxygen doping effect of, apparently, the
column’s encapsulating tissue74,110 it is very likely that in the regularly deposited
samples σct > σtp , while the opposite applies to the films prepared under ultra-high
vacuum conditions.
Our presently derived comprehensive picture of two subsequent percolation tran-
sitions (at v ≈ 0.3 and v ≈ 0.7) in µc-Si:H is well supported by the nicely resolved
two conductivity transitions that were reported, but not interpreted as such, by
 August 5, 2004 11:59 WSPC/140-IJMPB 02533

Percolation and Tunneling in Composite Materials 2117

Koynov et al.111 They found in undoped µc-Si:H a conductivity transition from

10−12 to 10−9 (Ohm cm)−1 at a low v value, and a subsequent conductivity transi-
tion between 10−8 and 10−3 (Ohm cm)−1 at a higher v value. Following our above
discussion one can now review the many data in the literature and find σ(v) features
by SEOUL NATIONAL UNIVERSITY on 07/25/22. Re-use and distribution is strictly not permitted, except for Open Access articles.

that are consistent with the two types of percolation transitions that we suggested
above. We note then that while usually these two transitions are not well resolved
due to experimental difficulties, they can still be revealed in careful studies.102,111
Such experimental difficulties include the fact that v changes substantially under
small-delicate variations of the films deposition conditions (in particular the amount
of the H2 dilution), the v-dependence of the evolution of the columnar structures
Int. J. Mod. Phys. B 2004.18:2091-2121. Downloaded from www.worldscientific.com

and the effect of the unintentional doping (in particular of oxygen).110 At present
is appears however that the latter makes the columnar encapsulating tissue the
dominant contributor to the conducting-percolation network.

6. Conclusions
Carbon black-polymer composites, Ni–SiO2 cermets and µc-Si:H are three types
of systems in which a percolation network and tunneling transport determine the
electrical conductivity. In the latter two systems, we were able to show that there
are two subsequent percolation transitions that have not been discussed before.
In particular, we found that for both systems a different mechanism controls the
transport in each of its two transitions. By comparing their local and macroscopic
electrical conductivity dependencies on the conducting phase content and the ap-
plied voltage we were able to derive new understanding on percolation-tunneling
systems in general and in the above three systems in particular. We could conclude
that:
(a) The percolation-like behavior in systems where tunneling is the transport mech-
anism is associated with a conductive network that is based on “nearest neigh-
bors” tunneling. In particular, the percolation-like behavior of the electrical
properties of CB-P polymer composites is a consequence of the fact that, for all
practical purposes, only “nearest neighbor” inter-particle tunneling contributes
to the electrical conduction.
(b) There are strong indications for the validity of the non-universal theory of
percolation-under-tunneling conduction and that the deviations in some cases
from its simple predictions, have to do with non-random or non-homogeneous
arrangements of the conducting particles.
(c) The PTCR effect can be described as a percolation threshold phenomenon
but this is in a system where this threshold is determined by the tunneling
mechanism. The corresponding electro-thermal switching observed in CB-P
composites is a global steady state manifestation of complete switching-off and
switching-on events of some individual critical tunneling junctions that play a
dominant role in the system’s connectivity.
(d) The high threshold percolation transition in both Ni–SiO2 cermets and µc-Si:H
can be understood within the framework of the classical percolation theory
 August 5, 2004 11:59 WSPC/140-IJMPB 02533

2118 I. Balberg et al.

while the low threshold percolation transitions in both systems seems to be

understood by the percolation-tunneling processes proposed in this review.

Acknowledgments
by SEOUL NATIONAL UNIVERSITY on 07/25/22. Re-use and distribution is strictly not permitted, except for Open Access articles.

This work was supported by the Israel Science Foundation, and benefited from
the collaboration with, M. Balberg, V. Chu, J. P. Conde, M. Eylon, C. Grimaldy,
M. B. Heaney, K. Schwartz, N. Shimoni and M. Wartenberg.

References
Int. J. Mod. Phys. B 2004.18:2091-2121. Downloaded from www.worldscientific.com

1. D. Stauffer and A. Aharony, Introduction to Percolation Theory (Taylor, London,

1992).
2. I. Balberg, Phys. Rev. B33, 3618 (1986).
3. I. Balberg and N. Binenbaum, Phys. Rev. B35, 8749 (1987).
4. For an early review see, I. Balberg, Phil. Mag. B56, 991 (1987).
5. J. B. Donnet, R. L. Bansal, M. J. Wang, editors, Carbon Black (Dekker, New York,
1993).
6. For recent review papers see, Carbon 40 (2002).
7. See for example, J. A. Oakes and C. Sanlberg, IEEE Trans. Ind. App. 1A9, 349
(1973); F. A. Doljack, IEEE Trans. Compon, Hybrides, Manuf. Tech. 4, 372 (1981).
8. I. Balberg, Carbon 40, 139 (2002) and references therein.
9. R. Zallen, The Physics of Amorphous Solids (John Wiley, New York, 1983).
10. F. Carmona, F. Barreau, P. Delhaes and K. Canet, J. Physique-Lett. 41, L-531
(1980).
11. I. Balberg, Phys. Rev. Lett. 59, 1305 (1987).
12. M. B. Heaney, Phys. Rev. B52, 12477 (1995).
13. Z. Rubin, S. A. Sunshine, M. B. Heaney, I. Bloom and I. Balberg, Phys. Rev. B59,
12196 (1999) and references therein.
14. M. H. Ali and A. Abo-Hashem, Polymer and Polymer Composites 4, 91 (1996).
15. C. Grimaldi, T. Maeder, P. Ryser and S. Strassler, Phys. Rev. B68, 024207 (2003).
16. L. Lamaignere, F. Carmona and D. Sornette, Phys. Rev. Lett. 77, 2738 (1998).
17. P. Sheng, B. Abeles and Y. Arie, Phys. Rev. Lett. 31, 44 (1973). For a review see,
B. Abeles, Appl. Solid State Sci. 6, 1 (1976).
18. B. Abeles, H. L. Pinch and J. I. Gittleman, Phys. Rev. Lett. 35, 247 (1975).
19. See for a recent example, C.-H. Lin and G. Wu, Physica B279, 341 (2000).
20. Q. Xue, Physica B325, 195 (2003).
21. I. Balberg, Phys. Rev. B57, 13351 (1998).
22. C. Chiteme, D. S. McLachlan and I. Balberg, Phys. Rev. B67, 024207 (2003).
23. See for example, L. K. H. Van Beek and B. I. C. F. Van Pul, Carbon 2, 121 (1964);
J. H. Smuckler and P. M Finnerty, Advances in Chemistry Series 134, 171 (1974).
24. W. F. Verhelst, K. G. Walthuis, A. Voet, P. Ehrburger and J. B. Donnet, Rubber
Chem. Technol. 50, 735 (1977).
25. P. Sheng, E. K. Sichel and I. I. Gittleman, Phys. Rev. Lett. 40, 1197 (1978). For a
review see also E. K. Sichel, J. I. Gittleman and P. Sheng, J. Elect. Mat. 11, 699
(1982).
26. J. Jensen, J. Appl. Phys. 46, 966 (1975).
27. D. M. Bigg, Polymer Eng. and Sci. 19, 1188 (1979).
28. J. D. Ajayi and C. Hepburn, Plastics and Rubber Processing and Applications 1, 317
(1981).
 August 5, 2004 11:59 WSPC/140-IJMPB 02533

Percolation and Tunneling in Composite Materials 2119

29. K. Ohe and Y. Naito, Jap. J. Appl. Phys. 10, 99 (1971).

30. A. Voet, Rubber Chem. Tech. 54, 42 (1981).
31. I. Balberg and S. Bozowski, Solid State Commun. 44, 551 (1982).
32. W. Y. Hsu, W. G. Holtje and J. R. Barkley, J. of Mater. Sci. Lett. 7, 459 (1988).
by SEOUL NATIONAL UNIVERSITY on 07/25/22. Re-use and distribution is strictly not permitted, except for Open Access articles.

33. L. Bengugui, J. Yacubowicz and M. Narkis, J. Polymer Sci. B25, 127 (1987).
34. F. Ehrburger-Dolles, J. Lahaye and S. Misono, Proc. of the 2nd Int’l Conf. on Carbon
Black (Mulhouse 1993), p. 223.
35. P. Hakizabera, E. Vanlathem and R. Deltour, Mater. Res. Soc. Symp. Proc. 367,
425 (1995).
36. A. Clezard, C. Deleuze, M. Dufort, E. Fudin, J. F. Mareche and E. McRae, Phys.
Rev. B53, 6209 (1996).
Int. J. Mod. Phys. B 2004.18:2091-2121. Downloaded from www.worldscientific.com

37. See A. I. Medalia, in Carbon Black-Polymer Composites, ed. E. K. Sichel (Dekker,

New York, 1992), p. 1.
38. See for example, W. Dafu, Z. Tiejun and X.-S. Yi, J. Appl. Poly. Sci. 77, 53 (2000).
39. F. Lux, J. Mater. Sci. 28, 285 (1995).
40. C. C. Chen and Y. C. Chou, Phys. Rev. Lett. 54, 2529 (1985).
41. I. Balberg and J. Blanc, Phys. Rev. B31, 8295 (1985).
42. R. Viswanthan and M. Heaney, Phys. Rev. Lett. 75, 4433 (1995); ibid. 76, 3661
(1995).
43. R. Wiesendanger, Scanning Probe Microscopy and Spectroscopy, Methods and Ap-
plications (Cambridge University Press, London, 1994).
44. For details see, N. Shimoni, D. Azulay, I. Balberg and O. Millo, Phys. Rev. B66,
620102(R) (2002).
45. For details see D. Toker, D. Azolay, N. Shimoni, I. Balberg and O. Millo, Phys. Rev.
B68, 041403(R) (2003) and references therein.
46. J. Fournier, G. Boiteux and G. Seytre, Phys. Rev. B56, 5207 (1997).
47. J. Ravier, F. Houze, F. Carmona, O. Schneegans and H. Saadaoui, Carbon 39, 287
(2001).
48. B. I. Shklovskii and A. L. Efros, Electronic Properties of Doped Semiconductors
(Springer-Verlag, Berlin, 1984).
49. V. Ambegaokar, B. I. Halperin and J. S. Langer, Phys. Rev. B4, 2612 (1971).
50. P. M. Kogut and J. Straley, J. Phys. C12, 2151 (1979).
51. B. I. Halperin S. Feng and P. V. Sen, Phys. Rev. Lett. 54, 2391 (1985).
52. B. Mikhnev, T. Yalamova and H. Berens, Elektroprom-st. Privovostr. 17, 407 (1982).
53. M. B. Heaney, Appl. Phys. Lett. 69, 2602 (1996); Physica A241, 296 (1997) and
private communications.
54. For earlier papers, see for example, F. Bueche, J. Appl. Phys. 44, 532 (1973); J.
Meyer, Polym. Eng. Sci. 13, 462 (1973), and for a recent review, see M. Hindermann-
Bisehoff and F. Ehrburger-Dole, Carbon 39, 375 (2001).
55. X.-S. Yi, L. Shen and Y. Pan, Composite Sci. and Tech. 61, 949 (2001).
56. R. D. Sherman, L. M. Middeleman and S. M. Jacobs, Poly Eng. and Sci. 23, 36
(1983).
57. G. Boiteux, J. Fournier, D. Issotier, G. Seytre and G. Marichy, Synthetic Metals 102,
1234 (1999) and references therein.
58. F. Carmona and C. Mouney, J. Mater. Sci. 27, 1322 (1992).
59. G. J. Lee, K. D. Suh and S. S. Im, Poly. Eng. Sci. 38, 471 (1998).
60. F. Carmona, E. Valot, L. Servant and M. Ricci, J. Phys. (France) 2, 503 (1992).
61. Y. Song, Y. Pan, Q. Zhang and X.-S. Yi, J. Poly. Sci. B, Polymer Physics 38, 1756
(1998).
 August 5, 2004 11:59 WSPC/140-IJMPB 02533

2120 I. Balberg et al.

62. W. Loser and C. Matthech, Phys. Stat. Sol. 18, 247 (1973); A. R. Dougal, IEEE
Trans. Power Electron. 11, 304 (1996).
63. R. Stumpler, G. Maidorn and J. Rhyner, J. Appl. Phys. 81, 6786 (1997).
64. J. E. Martin and M. B. Heaney, Phys. Rev. B62, 9390 (2002).
by SEOUL NATIONAL UNIVERSITY on 07/25/22. Re-use and distribution is strictly not permitted, except for Open Access articles.

65. A. R. Duggal and F. G. Sun, J. Appl. Phys. 83, 2046 (1998); F. G. Sun and A. R.
Duggal, ibid. 84, 5720 (1998).
66. For details see D. Azulay, M. Eylon, O. Ashkenazi, D. Toker, M. Balberg, O. Millo
and I. Balberg, Phys. Rev. Lett. 90, 236601 (2003).
67. L. Lamaignere, F. Carmona and D. Sornett, Physica A241, 328 (1997).
68. S. M. Sze, Physics of Semiconductor Devices (Wiley Interscience, New York, 1963).
69. B. K. Ridley, Proc. Phys. Soc. (London) 82, 954 (1963).
Int. J. Mod. Phys. B 2004.18:2091-2121. Downloaded from www.worldscientific.com

70. For a review see J. P. Clerc, G. Giraud, J. M. Laugier and J. M. Luck, Adv. Phys.
39, 191 (1990).
71. For a general review see R. E. Schropp and M. Zeman, Amorphous and Micro-
crystalline Silicon Solar Cells: Modeling Materials and Device Technology (Kluwer,
Boston, 1998).
72. For a recent review on the electrical transport in undoped µc-Si:H see, J. Kocka,
A. Fajfer, H. Stuchlikova, J. Fojtik, T. Mates, B. Rezek, K. Luterova, V. Svrcek
and I. Pelant, Solar Energy Materials and Solar Cells 78, 493 (2003), and references
therein.
73. For a detailed structure analysis see L. Houben, M. Luysberg, P. Hapke, R. Carious,
F. Finger and H. Wagner, Phil. Mag. A77, 1447 (1998) and references therein.
74. F. Liu, M. Zhu, Y. Feng, Y. Han, J. Kasouit and R. Vanderhagen, J. Non-Cryst.
Solids 299–302, 385 (2002).
75. S. Hamma and P. Roca I Cabarrocas, J. Appl. Phys. 81, 7282 (1997).
76. P. Hapke, U. Backhausen, R. Carius, F. Finger and S. Ray, Mater. Res. Soc. Symp.
Proc. 420, 789 (1996).
77. A. Di Nocera, A. Mittijo and A. Rubino, J. Appl. Phys. 78, 3955 (1995).
78. For a recent review see K. Lips, P. Kanschat and W. Fuhs, Solar Energy Materials
and Solar Cells 78, 513 (2003) and references therein.
79. See for example, D. Das, M. Jana and A. K. Barua, J. Appl. Phys. 89, 3041 (1995).
80. See for example, M. Fonrodona, D. Soler, J.M. Asensi, J. Bertomeu and J. Andreu,
J. Non-Cryst. Solids 299–302, 14 (2002).
81. T. Kamiya, K. Nakata, Y. T. Tan, Z. A. K. Durrani and I. Shimizu, J. Appl. Phys.
89, 6265 (2001).
82. A. B. Pevtsov, V. Yu. Davydov and N. A. Feoklisov, Phys. Rev. B52, 955 (1995).
83. M. Tzolov, F. Finger, R. Carius and P. Hapke, J. Appl. Phys. 81, 7376 (1997).
84. W. Fuhs, P. Kanschat and K. Lips, J. Vac. Soc. Tech. B18, 1792 (2000).
85. See for example, F. Diehl, M. Scheib, B. Schroder and H. Oechsner, J. Non-Cryst.
Solids 227–230, 973 (1998).
86. S. Brehme, P. Kanschat, K. Lips, I. Siebar and W. Fuhs, Mater. Sci. Eng. B69–70,
232 (2000).
87. K. Shimakawa, J. Non-Cryst. Solids 266–269, 223 (2000).
88. F. Siebke, F. Yata, S. Hishikawa and M. Tanaka, J. Non-Cryst. Solids 227–230, 977
(1998).
89. H. Overhof, M. Otte, M. Schmidtke, U. Backhausen and R. Carius, J. Non-Cryst.
Solids 227–230, 992 (1998).
90. T. Weis, S. Brehme, P. Kanschat, W. Fuhs, Limerheider and U. Wille, J. Non-Cryst.
Solids 299–302, 380 (2002).
 August 5, 2004 11:59 WSPC/140-IJMPB 02533

Percolation and Tunneling in Composite Materials 2121

91. R. Vanderhagen, S. Kasouit, R. Brenot, V. Chu. J. P. conde, F. Liu, A. de Martino

and P. Roca I Cabarrocas, J. Non-Cryst. Solids 299–302, 365 (2002).
92. R. Kankenhagen, M. Schmidt, S. Gerebner, M. Porchenriedes, W. Harrion, I. Siebker,
S. Koyanov and R. Schwartz, J. Non-Cryst. Solids 198–200, 923 (1996).
by SEOUL NATIONAL UNIVERSITY on 07/25/22. Re-use and distribution is strictly not permitted, except for Open Access articles.

93. S. Hamma and P. Roca I Cabarrocas, Appl. Phys Lett. 74, 3218 (1999).
94. F. Finger, C. Malten, P. Hapke, R. Carius, R. Fluckiger and H. Wagner, Phil. Mag.
Lett. 70, 247 (1994).
95. Y. He, C. Yin, G. Cheng, L. Wang, X. Liu and G. Y. Hu, J. Appl. Phys. 82, 3408
(1997).
96. S. K. Ram, S. Kumar, R. Vanderhagen and P. Roca I Cabarrocas, J. Non-Cryst.
Solids 299–302, 411 (2002).
Int. J. Mod. Phys. B 2004.18:2091-2121. Downloaded from www.worldscientific.com

97. R. Tsu, J. Gonzalez-Hernandez, S. S. Chao, S. C. Lee and K. Tanaka, Appl. Phys

Lett. 40, 534 (1982).
98. S. Hamma and P. Roca i Cabarrocas, J. Appl. Phys. 81, 7282 (1997).
99. C. Ross, J. Herion, R. Carius and H. Wagner, J. Non-Cryst. Solids 266–269, 69
(2000).
100. S. B. Concari, R. H. Buitrago, M. T. Gutierrez and J. J. Gandia, J. Appl. Phys. 94,
2417 (2003).
101. A. Baia Neto, A. Lambentz, R. Carius and F. Finger, J. Non-Cryst. Solids 299–302,
274 (2002).
102. T. Mates, A. Fejfar, I. Drbohlav, B. Rezek, P. Fojtik, K. Luterova, J. Kocka,
C. Koch, M. B. Schubert, M. Ito, K. Ro and H. Ujama, J. Non-Cryst. Solids
299–302, 767 (2002).
103. J. J. Schellenberg and R. D. McLeod, Solid State Commun. 66, 159 (1988).
104. I. Balberg, Phil. Mag. B80, 691 (2000).
105. I. Balberg, Y. Dover, R. Naides, J. P. Conde and V. Chu, Phys. Rev. B69, 035203
(2004) and references therein.
106. P. Brogueira, J. P. Conde, S. Arekat and V. Chu, J. Appl. Phys. 78, 3776 (1995).
107. D. Will, C. Lerner, W. Fhus and K. Lips, Mater. Res. Soc. Symp. Proc. 467, 361
(1997).
108. B. Rezek et al., J. Appl. Phys. 92, 587 (2002) and references therein.
109. For details see, D. Azulay, I. Balberg, V. Chu, J. Conde and O. Millo, to be published.
110. See for example, S. Veprek, Z. Iqbal, R. O. Kuhne, P. Capezzuto, E. A. Sarott and
J. K. Gimzewjki, J. Phys. C: Solid State Phys. 16, 6241 (1983); P. Torres, J. Meier,
R. Fluckiger, U. Kroll, J. A. Anna Selvan, A. Shah, S. D. Littlewood, I. E. Kelly
and P. Giannules, Appl. Phys. Lett. 69, 1373 (1996).
111. S. Koynov, S. Grebner, P. Radojkovic, E. Hartmann, R. Schwartz, L. Vasilev,
R. Krankenhagen, I. Sieber, W. Harion and M. Schmidt, J. Non-Cryst. Solids
198–200, 1012 (1996).