SANDIA REPORT

SAND2001-2160
Unlimited Release
Printed July 2001

Breakdown in ZnO Varistors by High

Power Electrical Pulses

Gordon Pike

Prepared by
Sandia National Laboratories
Albuquerque, New Mexico 87185 and Livermore, California 94550
Sandia is a multiprogram laboratory operated by Sandia Corporation,
a Lockheed Martin Company, for the United States Department of
Energy under Contract DE-AC04-94AL85000.

Approved for public release; further dissemination unlimited.

Issued by Sandia National Laboratories, operated for the United States Department
of Energy by Sandia Corporation.

NOTICE: This report was prepared as an account of work sponsored by an agency

of the United States Government. Neither the United States Government, nor any
agency thereof, nor any of their employees, nor any of their contractors,
subcontractors, or their employees, make any warranty, express or implied, or
assume any legal liability or responsibility for the accuracy, completeness, or
usefulness of any information, apparatus, product, or process disclosed, or represent
that its use would not infringe privately owned rights. Reference herein to any
specific commercial product, process, or service by trade name, trademark,
manufacturer, or otherwise, does not necessarily constitute or imply its endorsement,
recommendation, or favoring by the United States Government, any agency thereof,
or any of their contractors or subcontractors. The views and opinions expressed
herein do not necessarily state or reflect those of the United States Government, any
agency thereof, or any of their contractors.

Printed in the United States of America. This report has been reproduced directly
from the best available copy.

Available to DOE and DOE contractors from

U.S. Department of Energy
Office of Scientific and Technical Information
P.O. Box 62
Oak Ridge, TN 37831

Telephone: (865)576-8401
Facsimile: (865)576-5728
E-Mail: [email protected]
Online ordering: http://www.doe.gov/bridge

Available to the public from

U.S. Department of Commerce
National Technical Information Service
5285 Port Royal Rd
Springfield, VA 22161

Telephone: (800)553-6847
Facsimile: (703)605-6900
E-Mail: [email protected]
Online order: http://www.ntis.gov/ordering.htm
 SAND2001-2160
Unlimited Release
Printed July 2001

Breakdown in ZnO Varistors by High Power Electrical Pulses

Abstract

This report documents an investigation of irreversible electrical breakdown in ZnO varistors

due to short pulses of high electric field and current density. For those varistors that suffer
breakdown, there is a monotonic, pulse-by-pulse degradation in the switching electric field. The
electrical and structural characteristics of varistors during and after breakdown are described
qualitatively and quantitatively. Once breakdown is nucleated, the degradation typically follows
a well-defined relationship between the number of post-initiation pulses and the degraded
switching voltage. In some cases the degraded varistor has a remnant 20 mm diameter hollow
track showing strong evidence of once-molten ZnO. A model is developed for both electrical
and thermal effects during high energy pulsing. The breakdown is assumed to start at one
electrode and advance towards the other electrode as a thin filament of conductive material that
grows incrementally with each successive pulse. The model is partially validated by experiments
in which the varistor rod is cut at several different lengths from the electrode. Invariably one
section of the cut varistor has a switching field that is not degraded while the other section(s) are
heavily degraded. Based on the experiments and models of behavior during breakdown, some
speculations about the nature of the nucleating mechanism are offered in the last section.
 Contents

Page

Introduction 5

Materials and Electrical Tests 5

Pulse Test Results - Electrical 6

Pulse Test Results - Structural 8

Breakdown Model 10

Model Electrical Effects 11

Model Thermal Effects 12

Comparison of Model and Experiments 13

Conclusions 16

Speculations on Filament Initiation 18

Acknowledgements 18

Appendix A: Thermal Calculations 19

General Solution 19
Temperature Increases in ZnO Varistors 20

Appendix B: Electrical Protrusion Calculation 21

References and Notes 23

 Breakdown in ZnO Varistors by High Power Electrical Pulses

Introduction

Varistors are electronic devices with highly nonlinear current-voltage relationships. The
functional dependence of current on voltage is symmetrically bipolar and an inherent property of
the polycrystalline semiconductor from which it is made. At small applied electric fields,
varistors are insulating; but at a fairly well-defined, higher field, they switch to conducting and
maintain a nearly constant field over many decades of current. Most commercial, military, and
Sandia application varistors are based on polycrystalline, semiconducting ZnO with a variety of
other oxide additives typically in the molar range of 100 parts per million to several percent.
Their main application is in electrical circuits to limit or regulate the voltage that can be applied
to other devices or components. While it is not unusual for all varistors to operate with current
densities of 10-3 to 10 A/cm2, some Sandia applications also require unusually high electric fields
near 40 kV/cm. With these high power conditions, it is perhaps not surprising that small flaws in
the varistor result in breakdown, or a large irreversible change in their electrical and sometimes
structural properties. The purpose of this report is to document experiments and model
calculations that address the nature of varistor breakdown in the regime of high current density,
high electric field, and short pulse widths. Comments on the initiation mechanism are offered at
the end of the report.

Materials and Electrical Tests

All varistors in this study used ZnO as the nonlinear material. Most of the polycrystalline
ZnO elements were fabricated by sintering ZnO powder produced by the Sandia Chem-Prep
process1,2. In this process, a chloride solution containing all cation components of the varistor,
except bismuth and sodium, is prepared. NaOH solution is added to form a homogeneous
hydrous oxide slurry, followed by the addition of oxalic acid to convert the slurry to an oxalate
co-precipitate by a dissolution-re-precipitation reaction. A water wash, vacuum drying, and a
two hour calcine at 600ºC yields a “zincite” powder. Bismuth nitrate solution is added to a re-
slurry of the zincite powder. This causes hydrated Bi compounds to precipitate on the powder
particle surfaces, and these are converted to a layer of Bi2O3 by subsequent calcining at 400ºC
for two hours. Finally, sodium oxalate is added in solution to a slurry of the Bi-coated zincite
powder and freeze-dried to leave a layer of oxalate on the particles. Calcining again at 400ºC for
two hours dopes the powder with about 300 ppm sodium, which has been found to be beneficial
for improving reliability under high power pulses.

The powder is pressed into a cylindrical mold, then sintered at 732ºC. A pattern of opposing
silver paint electrodes is typically applied to the flat surfaces of the cylindrical sintered wafer so
the material can be electrically tested before further fabrication. The wafer is then cut along
sectors, and cylindrical rods of varistor material are cored-drilled parallel to the flat wafer
surface. Kovar electrodes are attached to each end of the cut rod using a thin silver-loaded epoxy
cloth preform. Then the entire varistor is surface ground to the desired diameter. Some of the
varistors were subsequently encapsulated in Epon 828 epoxy containing a high density of hollow
glass microballoons.
 Electrical pulse testing was performed using a high voltage tester of Sandia design as shown
schematically in Figure 1. The tester is intended to measure the voltage across the varistor at a
fixed current3 determined by the value of the current limiting resistor. Although not shown in
the figure, there were actually two resistors available in any given test series. This permitted the
two-point determination of the nonlinearity coefficient,

α ≡ ∆ℓn(J)/∆ℓn(E), (1)

where J is the current density and E

Pulsed I Current the electric field. The current pulse is
Voltage Limiting approximately 10 ms long with a rise
Generator Resistor time of about 1 ms. Pulses were
applied about 20 seconds apart.
Typically, a high/low current pulse
V pair would be applied at the
beginning of a testing sequence to
measure the nonlinearity coefficient,
which was usually 20 or slightly
larger. Subsequent pulses would be
only high current pulses to determine
the reliability of the varistors, and to
Varistor
examine the breakdowns under
constant current conditions when they
did occur. Pulsing was done to
varying degrees of electrical
Figure 1. Schematic diagram of the electrical tester breakdown to enable structural
for varistors. The generator voltage is several times examination of the stages. During the
higher than the varistor’s limiting voltage, thus the electrical testing, all varistors were
circuit is nearly constant current. immersed in a bath of Fluorinert FC-
77 (3M Corp.) to suppress surface
arcing.

Pulse Test Results – Electrical

The main purpose of the electrical tests was to determine the electrical performance
reliability of the varistors. Twenty-five pulses were standard for these tests, but some samples
were tested for hundreds of pulses. Many of the varistors survived the electrical reliability tests
with no discernable changes in properties. However, the topic of this report is the breakdown
behavior of those varistors that did not survive the testing without change. Because the pulse
tests were conducted as part of a Sandia varistor development program, the samples were
fabricated with several deliberate process variations, at several sites, and from several chemical
sources. While there were often large variations in reliability among different process lots, there
was a very consistent pattern for the electrical breakdowns that did occur.
 Varistors that suffered a permanent change in their electrical properties almost always
exhibited two regimes. First there was a regime in which the measured voltage appeared to be
constant within the noise-limited precision of the measurement which was roughly ±0.5%. The
number of pulses in this regime was quite variable from lot to lot, and among varistors within the
same lot. Of course, this was the only regime observed for varistors which passed the reliability
tests. Breakdown was always observed to be a decrease in the measured pulse voltage. Within
several pulses of the first noticeable decrease, a new trend was established. Each subsequent
pulse caused a decrease in voltage that was a constant fraction of the original voltage. For
varistors with switching fields of 40 kV/cm receiving current pulses of 11 A/cm2, each pulse in
the breakdown regime would cause an incremental decrease of about 4% of the original voltage.

Because of this common behavior, it is possible to overlay curves of measured voltage,

normalized by their original value, as a function of number of pulses, adjusted for the first
noticed voltage decrease, to get a single curve for many varistors. Of course, the slope of the
curve in the breakdown regime depends strongly on the pulse parameters, and to a lesser extent
on the process lot. An example of a single curve is shown for six varistors from 3 similar
process lots in Figure 2. Each of these varistors began to exhibit a voltage decrease at different

1.0

0.9
V/V0

0.8

0.7

0.6

0.5

0 5 10 15
Pulses after apparent start of breakdown

Figure 2. Electrical properties of ZnO varistors in breakdown. The ordinate scale shows the
voltage across each varistor subjected to a series of constant current pulses. The value is
normalized by the voltage of the first pulse, V0. Each of the six varistors had an apparent
start of breakdown at a different pulse number, but the curves were made coincident by
translating them along the abscissa. The actual pulse number corresponding to pulse zero in
the graph is indicated in the legend.
pulse numbers, but it is possible to translate each curve along the pulse number axis to create a
near correspondence for all. If the number of pulses in the apparently normal, original voltage
state are subtracted from the total number of pulses, this yields the number of pulses, Pa, received
in the apparent breakdown regime. An empirical fit to the breakdown curves is given by

V/V0 = 1 – 0.06ℓn{1 + exp[(2/3)(Pa - 6)]}. (2)

The form of this fitting equation was chosen because it has the correct limiting behavior early
and late in the breakdown characteristics.

Pulse Test Results – Structural

A few of the varistors that had electrical breakdowns were found to have holes on the
electrode surfaces of the ZnO. These holes were first noticed because of small holes at the
periphery of the silver-loaded epoxy preforms that were connected to the ZnO holes by a black
track in the preform. An example of a hole in the ZnO electrode surface is shown in Figure 3.
There is evidence not only of a hole, but also of once-molten flow
on the surface. Frequently these samples have a single crack which
passes through the hole. When the crack is forced opened, an
obvious track is exposed along the crack surface as seen in Figure 4.
Electrode Both the hole and the track are typically
Surface 20 µm in diameter. The tracks are
hollow, their surfaces are glassy and
occasionally they show a radial
crystalline re-growth region as shown in 50 mm
Figure 5. The hole and track structures in
the varistors strongly indicate that a
filament had formed with temperatures Figure 3 – Example of
high enough to melt the ZnO locally. It an electrode hole
was also hot enough to produce a vapor caused by high voltage
pressure sufficient to force much of the breakdown.
molten ZnO out through the electrode,
40 mm through the preform under the solid Kovar endcap, and out into the
Fluorinert.

6 mm
15mm
Figure 4. Example of
an internal track
caused by high
voltage breakdown.

The extent of track

penetration into the bulk of
varistor rods, without fracture or
surface tracks, was probed by
Figure 5. Example of radial recrystallization inside a
breakdown track.
cutting the rods at
various lengths. Initially the rods were just cut in half. Later, guided by results of a model
calculation, two badly degraded rods were cut into thirds. Only two rods of the 21 cut for track
examination showed a track both starting in the electrode surface and passing entirely through
the cut piece. One track ran through a half-length, the other through a third. Both samples had
been degraded to V/V0 = 0.59. Rods that had fractures or surface tracks did not follow the single
curve electrical breakdown and were not sectioned.

Another occasional feature of electrically degraded varistors

were 50 µm pits in the electrode or cut surfaces that were ringed
by larger halos of Bi-rich material as determined by EDS
(energy dispersive spectrometry) in the scanning electron
microscope (Fig. 6). EDS also indicated qualitatively high
concentrations of silicon at the bottom of the pits. These halo 300 mm
pits did not appear to be part of a penetrating hollow track. Of
the 21 cut samples examined, three showed these Bi-halos on
one electrode surface and three had them on interior cut Figure 6 - A backscattered
surfaces. electron image of a Bi halo.

Some varistors had more obvious structural damage in the form of tracks along the
cylindrical surface. These surface tracks did not always extend to an electrode surface, and it
was common to see a surface track emerge from and re-enter the cylindrical surface.
Electrically, these varistors did not follow the well-defined, single curve breakdown. Instead the
rate of voltage decrease per pulse was up to three times faster. An interesting hybrid is shown in
Figure 7 for varistor GB136B 60-1 A1. This varistor had a hole in one electrode, a short surface
track just below that electrode, and holes in both halved surfaces after it was cut. Notice that the
1.0
electrical properties
GB136B 60-1-B1 change more rapidly
GB136B 60-1-A1
GB140D 60-2-A1 than standard at first,
0.9 then join and follow
the common, single
curve for the remainder
0.8 of the pulse testing.
The interpretation is
V/V0

that the rapid changes

0.7 were associated with
the surface track, and
that the single curve
0.6 behavior occurred after
the track re-entered the
bulk of the varistor.
0.5
0 2 4 6 8 10 12 14
Apparent No. of High Voltage Breakdown Pulses, Pa Finally, I note that
while fractures and
tracks remnant of high
Figure 7. Comparison of electrical breakdowns between
temperature filaments
varistors with and without a surface breakdown track.
in the varistors are dramatic signs of electrical breakdown, many varistors showed none of these
features after breakdown. For a few varistors, an intense topographical examination was made of
electrode and cut surfaces using optical and electron beam microscopy without discovering
tracks or other breakdown remnant features. Nevertheless, subtle changes of grain texture or
intergranular phases may have been missed. Also, even in the presence of a high temperature
filament, the major structural changes may have been suppressed because of the strength of the
sintered ceramic body.

Breakdown Model

In this section, a model is presented to explain the essential electrical and structural features
of ZnO varistors in the breakdown regime during pulse testing. Although the model does not
explicitly address the materials issues associated with the initiation of breakdown, it does offer
some insights. The essential assumption of the model is that the voltage decrease is due to a
single conductive filament that forms at one of the electrodes and grows in length towards the
opposite electrode with each applied pulse.4 Figure 8 shows
the concept. The varistor length is L and the filament length
is Lf. In the pulsed, high current condition, the varistor
material everywhere, except in the conductive filament, obeys
the standard approximate relationship of J = pEα, where α is Lf
L
the nonlinearity coefficient and p is a constant. Because the
filament is conductive, the voltage at the bottom of the
filament is the same as the voltage on the top electrode. This
makes the electric field in the column below the filament
higher than the electric field in the remainder of the varistor5.
With α ~20, even a small increase in field will raise the local
current density in the filament and the column beneath it Figure 8. Schematic
substantially. As each pulse extends the filament deeper into illustration of a conductive
the varistor, the field and current density beneath the filament filament (shaded) in a
will increase until eventually the entire fixed current from the partial breakdown varistor.
tester will be flowing in the filament. This concentration of
field and current increases the Joule heating along the column below the filament. It is not
surprising that, under some circumstances, temperature rises can occur large enough to melt,
vaporize and fracture the ZnO.

Before quantifying the consequences of this model, I first examine the correctness of the
assertion that the enhanced current density at the filament tip will stay mostly contained in the
column beneath, rather than spread into the bulk of the varistor. Indeed, as shown in App. B, if
the varistor conductivity were independent of electric field, then the current would spread into
the bulk.6 However, in the switching regime, the conductivity of varistor material increases with
increasing current density. This is a positive feedback effect – increased current density locally
makes it more favorable to conduct current through the same locality. The effect is the same as
replacing the varistor material in the column with a rod of more conductive material. Current
localization under a conductive electrode protrusion has been graphically demonstrated using a
two-dimensional resistor lattice model in which the resistors were assigned nonlinear properties
similar to varistors.7 A small amount of current spreading is also seen in those calculated results.
Model Electrical Effects

While there may be some spreading of the current density beneath the filament tip, we will
ignore it to calculate approximately the voltage decrease due to the filament. Using the
subscripts b and c to denote the varistor bulk and column beneath the filament, respectfully, we
can first write the equation for the current distribution.

I = JbAb + JcAc = AbpEbα +AcpEcα = AbpEb0α + AcpEc0α, (3)

where A is the cross-sectional area and the subscript 0 denotes the initial, no filament condition.
Using a = Ac/Ab and ℓ = Lf/L, and noting that

Eb0 = Ec0 and Ec = EbL/(L – Lf),

the measured voltage as a function of filament length can be written as

V(ℓ)/V0 = (1 - ℓ){[1 + a]/[(1 - ℓ)α + a]}1/ α. (4)

The varistors tested had 5 mm diameters, so filament diameters of 20 µm and 200 µm yield
a = 1.6x10-5 and 1.6x10-3, respectively. The curves for these two ratios are given in Figure 9.
Notice that both have the same basic shape as the data in Figure 2. This suggests that they be fit
1.0

0.9
Calculated Fractional Voltage
Degradation, 20 micron diam.
Curve Fit, 20 micron
0.8
Calculated Fractional Voltage
Degradation, 200 micron diam.
V/V0

Curve Fit, 200 microns

0.7

0.6

0.5
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7
Fractional Filament Length, Lf/L

Figure 9. Calculated voltage degradation as a function of filament length

from Eq. (4). Also shown as data points are the fits to the calculations
represented by Eqs. (5) and (6).
to the same functional form as Equation (2). The calculated values of V/V0 versus ℓ are well
approximated by

V(ℓ)/V0 = 1 – 0.06ℓn{1+exp[(2/3)(42ℓ - 17)]}, for a 20 µm filament, (5)

and V(ℓ)/V0 = 1 – 0.06ℓn{1+exp[(2/3)(34ℓ - 9)]}, for a 200 µm filament. (6)

The two curve fits are shown as points in Figure 9.

Model Thermal Effects

The nucleation and propagation of a conducting filament with successive high power pulses
causes an increase in both electric field and current density at the filament tip, which increases
the local Joule heating per unit volume, J·E, significantly. The details of the calculation for the
corresponding temperature rise during a single pulse have been relegated to Appendix A. Here
we provide a summary of the pertinent issues for the varistors tested.

The standard high current for the reliability tests was I0 = 2.2 A, which corresponds to
J0 = 11 A/cm2 when distributed uniformly through undamaged varistors with no filaments. The
value of the varistor’s switching field in this uniform current density is E0 = 40 KV/cm. As
shown in Appendix A, this uniformly distributed electrical power heats the varistor at a rate of
0.16 K/µs, and thus ∆T = 1.6 K by the end of the 10 µs pulse. However, when a filament forms,
the local current and field at the tip increases and the local temperature rise is much higher. The
temperature increase in the column beneath the filament depends on the power per unit length,
Qf = E·I = E·J·Ac, where Ac is the cross-sectional area of the column. The factor by which Qf is
increased from the uniform value can be determined from the relations derived in the electrical
section above.

Qf(ℓ,a)/Qf0 = JcEc/Jc0Ec0 = (Ec/Ec0)α+1 = {Eb/[(1-ℓ)Eb0]}α+1. (7)

Using Eq. (4) this can be written as:

Qf(ℓ,a)/Qf0 = {[1 + a]/[(1 - ℓ)α + a]}1+1/α. (8)

As ℓ grows, the first term in the denominator begins to get small rapidly, which increases the
ratio. Using the calculated results from Table A1 and the thermophysical parameters of
Table A2, the temperature increase by the end of the 10 µs pulse can be estimated. Shown in
Figure 10 are the temperature increases at the tip after 10 µs of applied current for filament
diameters of 20, 100, and 200 µm. The dependence on filament length simply reflects the fact
that both E and J increase at the tip as the filament grows longer. The small filament diameters
concentrate J the most and thus their columns have the highest temperature rise. The saturation
of ∆T for the 200 µm filament reflects the fact that essentially all varistor current is flowing in
the filament for ℓ > 0.4. Note also that because the filament is assumed to be conductive, there is
little or no field in the filament, and thus its heating is negligible.
 2.5

2.0
20 micron diameter
Temperature Rise at 10ms (KK)

100 micron diameter

200 micron diameter
1.5

1.0

0.5

0.0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7

Fractional Filament Length, Lf/L

Figure 10. Filament tip temperature rise in kilokelvins at the end of a 10 µs high power pulse.
Calculated increases are shown for three filament diameters. The amount of increase depends
on the extent of filament penetration, Lf/L, because of the consequent concentration of Joule
heating at the tip.

For ZnO, the melting point is nearly 2000ºC (∆T = 2.0 kK) and the temperature at which the
vapor pressure8 reaches 1 atm. is about 1700ºC (∆T = 1.7 kK). The model calculations indicate
that both temperatures are exceeded for filaments less than 200 µm in diameter once they have
penetrated 30% of the varistor length.

Comparison of Model and Experiments

To connect the model electrical calculations with the single curve breakdown data, we
assume that the filament length is proportional to the number of high current pulses received
after the actual start of breakdown,

ℓ = kPb (9)

where k is the constant of proportionality. Note that this may not be a good assumption near the
beginning of degradation when the filament may propagate slower because of lower stimulus
(thermal or electrical). Next we note that because of the noise in the voltage measurements, we
may not observe the actual start of breakdown. It is only after the voltage has decreased by
about 1% that it becomes apparent that the monotonic breakdown is underway. Thus it makes
sense to write

Pb = Pa + Ph, (10)

where Pa is the number of pulses applied since the evidence of breakdown becomes apparent and
Ph is the “hidden” number of pulses applied after nucleation, but before breakdown is noticed
using the Sandia tester. With these two assumptions, we can now equate Equations (2) and (5) to
obtain k = 0.024 and Ph = 11 for a 20 µm filament. For a 200 µm filament, Equations (2) and (6)
yield k = 0.029 and Ph = 3. Taken literally, this means that each pulse during the breakdowns
shown in Figure 2 is advancing the filament length 2½% of the total varistor length, for either
diameter filament. The Ph value implies that the actual start of breakdown and filament
formation may be occurring 3 to 11 pulses before it is reliably noticed in the voltage output of
the tester. So without knowing the diameter of the filament, it is difficult to tell whether
different pulse breakdown delays among the tested samples are differences in nucleation time or
filament diameter.

As a test of the model, seven varistors were pulsed to various values of V/V0. All seven
varistors followed the single curve breakdown of Fig. 2, and none of them had surface tracks or
fractures. After the initial pulsing they were cut to different lengths to look for electrical and
structural evidence of breakdown. The idea is
illustrated in Figure 11. Depending on the degree of
breakdown, the rods were cut into either two or three
pieces with the intent to have one of the pieces not
penetrated by the filament. The cut surfaces were
examined for holes. Finally each cut face was given
an electrode of silver paint and each piece was
electrically re-tested using the same current. The
electrical results are summarized in Figure 12.

Figure 12 is a composite of data from this

sectioning experiment plus the 200 µm calculated
curve of Figure 9, both plotted as fractional degraded
Figure 11. Illustration of how rods voltage versus fractional length. The voltage axis of
were sectioned to determine filament V/V0 has the same meaning for each, but the abscissa
penetration. denotes different quantities. For the calculated
curve, it is the predicted length of the filament. For
the data, it is the fractional length of the cut(s) from
one varistor electrode. The superimposed drawings illustrate both the experimental cut locations
and the calculated filament length (associated with the arbitrary placement of the drawing on the
voltage axis).

Consider first the five varistors whose breakdown was limited to V/V0 > 0.75. These were
all cut approximately into halves. Four of the five varistors (triangles) each contained one half-
piece with very low voltage on the first re-test pulse. The other half-piece exhibited a constant
switching voltage through 25 additional pulses that corresponded to the initial electric field in the
 1.0

Rod Cut into 2 pieces

0.9

Experimental Cut Length, One

piece with original E field
0.8
Experimental Cut Length, Both
pieces with reduced E field
V/V0

Calculated Lf/L, 200 micron

Pieces with
0.7 Filament
reduced E field Rod cut into
Pieces with
3 pieces
original E field

0.6

0.5
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0

Fractional Cut Length from Start of Filament

Figure 12. Test of filament length prediction for degraded varistors. The solid curve is a
replot of the 200 µm calculation in Figure 9. Also plotted as open symbols are data from 7
varistors that were pulsed to various levels of breakdown as indicated by their position on
the voltage axis. The symbol location on the length axis shows the length of sections cut
subsequent to the breakdown. Circles indicate that pieces on both sides of the cut were
electrically degraded, and triangles indicate that a piece on one side of the cut was degraded
but the other piece was not. Superimposed are schematic drawings of the varistor rod
showing the two types of sectioning used. They are located in the voltage range appropriate
for each type, and the dashed line graphically shows the filament length of the calculation.
varistor. In other words, one half piece was catastrophically broken down electrically while the
other half was a perfectly good varistor with an un-degraded switching field. None of these
samples showed holes or halo pits on the electrode or cut surface prior to re-test. For the fifth
varistor in this group (circle) both halves exhibited catastrophic electrical breakdown upon re-
test, one on the first tester pulse, and one on the second. This sample had a halo pit on both cut
surfaces prior to re-test.

The remaining two varistors were degraded to V/V0 < 0.6 and were cut roughly into thirds.
Each of these varistors had two adjacent pieces that were badly degraded and broke down
immediately on re-test. The third piece of each varistor was just like new, exhibiting the original
switching field on re-test. However, after several pulses, each good third developed a
breakdown initiated at the edge of the silver paint spot that did not cover the entire cut surface.
Although these two varistors were electrically similar, there was a major structural difference
between them. The varistor that was degraded to V/V0 of 0.52 had no apparent sign of a
breakdown track or halo on its cut surfaces. The varistor degraded to 0.59 had a track that
completely penetrated the cathode third, from electrode to cut, and entered the middle third from
the cut.

Conclusions

A major assumption of the model is that breakdown is due to a conducting filament

originating at one of the electrodes. The strongest evidence in support of this assumption comes
from the sectioning experiments. These experiments show that the degraded, high conductivity
region is localized to one end of the rod, while the other end shows no evidence of degradation.
Also in all cases in which a breakdown track was observed, it had one end at an electrode. Thus,
confirmation of the electrode-initiated part of the assumption rules out a uniform bulk
degradation.

The filamentary nature of the breakdown is only partially proven. Electrically, the linear
decay of V/V0 with successive pulses is not proof of a thin filament since a conducting region
comprising the whole rod cross-section (a filament of the whole) would produce the same linear
decay. Only a small fraction of the varistors had tracks as visible evidence of a filament. In
addition, the observation of the Bi-halos in a few other rods might be interpreted as a remnant of
a hot filament. Eda, in his paper9 on varistor destruction (catastrophic breakdown), notes that
intergranular Bi2O3 melts at 820ºC. Thus a filament reaching this temperature might cause the
Bi2O3 to melt and agglomerate differently.

There may be several reasons why structural evidence for filaments has not been seen in
many of the electrically degraded varistors. One reason might be the diameter of the filament.
As seen from the curves in Figure 10, filaments with diameters greater than 200 µm are not
expected to raise the local temperature enough to vaporize or melt the ZnO and thus cause a
track. Even for smaller filaments, the large local temperature increases do not occur until the
filament has propagated 20% of the rod length. Since the filament itself is conductive, its Joule
heating is always small. So for penetration beyond 20%, the hot tip is capped by a cooler, solid
plug of filament, which may prevent venting or other gross structural changes. In fact, this latter
situation both explains why tracks often have associated fractures and confounds explanation for
the few cases in which tracks formed without fractures.

Nevertheless, a thin filament is the most likely form of the changed material. As noted at the
end of the Breakdown Model section, a conductive filament, once formed by any cause, will tend
to concentrate the current into the material beyond its tip because of the nonlinear properties of
the varistor. In that sense, filaments are stable, self-propagating structures. However, their
propagation through the varistor does not appear to be primarily driven by heating at the tip. As
noted in the last paragraph, short filaments of almost any sensible diameter are incapable of
concentrating enough current to provide significant heating during the pulse. If temperature is
not the primary cause, then the propagating degradation must be due to electrical effects.

Speculations on Filament Initiation

As stated earlier in the section titled Pulse Test Results – Electrical, most of the varistor
samples tested did not show electrical degradation or breakdown. Although 25 high voltage
pulses was a standard test, many were tested up to 50 pulses. A few were tested to hundreds of
pulses, one through 800 pulses, with no sign of degradation. From these results, one may
reasonably conclude that the inherent reliability of the varistor material is excellent for the
electric field and current density used. The observed failures were likely caused by a flaw in the
material as in other failure mechanisms, such as mechanical fracture. As argued above, the
initiating flaws are most likely on or near one of the electrode faces, and it is some type of
electrically conducting protrusion. This probability was also expressed by Vojta and Clarke7
who used a variable resistor network to do computational simulations of varistor conduction in
two dimensions. They showed that conductive inclusions at an electrode cause more current
concentration than the same inclusions in the body of a varistor but separated from an electrode
surface. They concluded that “electrode protrusions … are likely sources for failure.”

Mechanistically, how can an electrically conducting protrusion initiate degradation?

Evidence above shows that it is unlikely to be a thermal mechanism from Joule heating. A more
likely possibility is the increased local electric field associated with the shape of the protrusion.
As an example, the electric field due to a hemisphere-shaped protrusion is calculated in
Appendix B. At the tip of the protrusion, the local electric field is three times the uniform field
far away. This increase is independent of the diameter of the protrusion (for diameters small
compared with the lateral extent of the medium).

The increased electric field will not only increase the current during the applied pulse, but
also may change the permanent conductivity of the varistor material. The current increases
during the pulse are entirely electronic. The current-voltage characteristics of varistors are
determined by electrostatic potential barriers that form at the ZnO grain boundaries, as shown by
the simplified schematic of Figure 13. These barriers substantially impede electron flow at low
voltages. As the voltage is increased, the barrier lowers and
e -
electrons can more easily transit. Because the current depends
exponentially on the barrier height, small changes in voltage can
result in large current changes which gives rise to the switching
(or regulating) phenomena. However, the [energy] height and
V [depletion layer] width of these electrostatic potential barriers are
determined by the density of lattice defects or impurities near the
grain boundary that can be ionized to a positive charge state. If
Force on the density of these lattice sites changes, then the equilibrium
positive potential barrier and its response to voltage will change
charges
permanently. The positive charges may be induced to electro-
migrate due to the increased field and high current. Sensitivity of
the equilibrium barrier height, ΦB, to the density of positive
Figure 13 – Schematic
charge in the depletion region, Nd, can be estimated assuming that
diagram of a grain
the total charge per unit area at the boundary, Q, does not change.
boundary electrostatic
The relationship is10:
barrier under an applied
Q = 2(2εε0NdΦB)½.
voltage, V.
In this equation ε and ε0 are the relative and free space dielectric constants, respectively. So for
constant Q, if Nd doubles, then the barrier height reduces by a factor of two. This may account
for the higher conductivity of the varistor material at the tip of the protrusion or filament.
 What would constitute an electrically conductive protrusion at an electrode surface? Clearly,
a physical depression into which the silver-loaded epoxy was pressed would qualify. Such a
depression could form during surface grinding due to grain pull-out or preferential erosion in a
porous region. Possibly the depression could just be a pore exposed at the surface. Another,
more insidious, possibility for the protrusion might be the presence of excess Bi2O3 at the
surface. Bismuth oxide has been reported to be conductive11, so a continuous phase from the
surface into the bulk would also act as a protrusion. This type of concentrated Bi2O3 has been
observed in some batches of varistors in the form of “bismuth halos” discussed above in the
section titled Pulse Test Results – Structural and shown in Figure 6.

Acknowledgements

The author is pleased to acknowledge the assistance of Robert Jaramillo in the electrical
testing, and Lori Maestas for the electron microscopy and special sample preparations. Jim
Voigt and Steve Lockwood have been especially helpful in discussing powder and rod
preparation issues.
Appendix A: Thermal Calculations

This appendix describes the calculation of temperature rise for cylindrical filaments of
material, first generally and then specifically for ZnO. Here also are gathered various thermal
parameters for ZnO.

General Solution

The general problem solved here is the temperature rise in a cylinder of radius, Rf, heat
capacity per unit volume of ρfcf, and infinite thermal conductivity. The cylinder is embedded in
a semi-infinite medium of heat capacity per unit volume of ρmcm and thermal conductivity of κm.
Initially the filament and medium are at the same temperature. Beginning at time zero the
filament is heated by a power of Qf per unit length. Carslaw and Jaeger12 give the solution in
terms of a dimensionless temperature increase, (κm∆T/Qf), and time, (t/τm):

kmD T 2a 1 2
=
[ò 1 - e ( t )u ]du ,
¥
- t/ m 2

(A1)
Qf p3 0
u 3 D ( u, a 1)

where ∆(u, α 1 ) =[uJ0(u) – α 1 J1(u)]2 + [uY0(u) – α 1 Y 1(u)]2,

and α 1 = 2 ρ m cm/ ρ f cf = twice the ratio of heat capacities,

and τ m = ( ρ m cm/ κ m )Rf2.

The integral was solved numerically using numerical approximations to the various Bessel
functions. The results are tabulated below, and shown graphically in Figure A1.

t/τm (κm/Qf)∆T Adiabatic

0.01 2.65E-03 3.18E-03
0.02 5.07E-03 6.36E-03
0.0316 7.68E-03 1.00E-02
0.04 9.45E-03 1.27E-02
0.06 1.34E-02 1.91E-02
0.08 1.70E-02 2.54E-02
0.10 2.04E-02 3.18E-02
0.20 3.48E-02 6.36E-02
0.316 4.82E-02 1.00E-01
0.40 5.64E-02 1.27E-01
1.00 9.78E-02 3.18E-01
3.16 1.69E-01 1.00E+00
10.00 2.54E-01 3.18E+00
Table A 1 Dimensionless temperature increase in the cylinder vs dimensionless time from the
general equation and calculated adiabatically.
 Dimensionless Temperature
10.000

(k/Q)DT
1.000
Adiabatic
Rise

0.100

0.010

0.001
0.01 0.1 1 10
Dimensionless Time, t/tm
Figure A 1 Graph of filament temperature increase vs time for embedded and
adiabatic conditions. The embedded solution used α1 = 2.
The temperature increase of the filament heated under adiabatic conditions is given by:

∆T = (Qf/ρfcfπRf2)t. (A2)

Note that for t << τm, the general solution for the embedded filament is also linear in t, so

(κm/Qf)∆T = Xt/τm = Xtκm/(ρmcmπRf2),

where X is a constant of proportionality. Now in the limit of t → 0, the filament and surrounding
medium are at the same temperature. With no temperature difference, there is no heat flow to
the medium, and the temperature rise is essentially adiabatic. This fixes the constant of
proportionality to be X = ρmcm/πρfcf and allows the adiabatic rise to be tabulated and plotted with
the embedded data. This limiting condition simplifies the calculation of ∆T for t << τm.

Temperature Increases in ZnO Varistors

To use the calculations above for the case of ZnO varistors, the thermophysical parameters of
ZnO must be known. The Table A2 below lists them.
Material T (ºC) κ13 (W/m·K) c8 (J/g·K) ρ (g/cm3) Pvapor8
ZnO 20 0.50 5.6
200 17
227 0.58
627 0.64
800 5.3
1400 1
1500 12
FC-7714 20 0.063 1.05 1.78
Epon 82815 20 0.16 1.97 1.20
Epon 828 20 1.59 0.87
w/GMB15
Table A 2 Thermophysical parameters of ZnO, tester fluid, and encapsulants.
From these parameters one can compute the time constant, τm, from the expression below
Eq.(A1) above. It is difficult to give a precise meaning to the time constant, τm, because the

Filament Radius 10 µm 50 µm 100 µm 2.5 mm

τm (room temp.) 17 µs 0.43 ms 1.7 ms 1.0 s
τm (high temp.) 67 µs 1.7 ms 6.8 ms 4.2 s
Table A 3 Thermal time constants for ZnO surrounding filaments of various radii.
thermal problem involves diffusive heat transfer which is not a simple exponential relationship.
However, as a rough approximation, one can state that for t < 0.1τm heating is nearly adiabatic,
and that after the heating is stopped, the excess temperature is reduced by 50% at t = 0.1τm.

As can be seen from the two tables above, the temperature increase in all but the very
smallest filaments can be calculated adiabatically for pulses limited to 10 µs in duration. Using
the room temperature thermophysical parameters for the whole varistor rod (2.5 mm radius) as
the filament, in any of its various environments, the rise is 0.16 K/µs for conditions of the high
current pulse, J = 11 A/cm2 and E = 40 KV/cm.

Appendix B: Electrical Protrusion Calculation

Presented here is a simplified version of the more general solution for the effect of
inhomogeneities on electric field, current density, and power dissipation distributions in a
conducting medium6. For a perfectly conducting sphere of radius R in a medium of uniform and
isotropic conductivity σm, the electric field outside the sphere is given by16

E/E0 = ur[1 + 2(R/r)3]cosθ – uθ[1 – (R/r)3]sinθ, (B1)

where bold indicates vectors and ur and uθ are the radial and azimuthal unit vectors, respectively,
in spherical polar coordinates. The uniform field, E0, far from the sphere defines the direction of
the pole at θ = 0, and r is the distance from the center of the sphere. The field magnitude is

E = {1 + 2(R/r)3(3cos2θ - 1) + (R/r)6(3cos2θ + 1)}½E0, (B2)

At the surface of the sphere, r = R, the field is strictly radial and its magnitude reduces to
E = 3E0cosθ. The streamlines of the electric field, and hence the current, are found by solving
the vector cross-product equation

[E(r)/E(r)]Xdr = 0, (B3)

where dr is an incremental change in position of the vector r. This equation was solved
numerically with representative results shown in Figure B1.

The conducting sphere concentrates current 3.0

that eventually spreads into the surrounding E0

medium. The streamlines show the current

2.5
paths, but not the magnitude of the current
density. However, from Eq. (B2) one can
calculate that maximum current density drops 2.0
from 3σmE0 at r = R to only 1.07σmE0 at r = 3R
for this constant conductivity medium.
1.5

Although this calculation of electric field is Z

for a perfectly conducting sphere in a uniformly
conducting medium, the symmetry of the 1.0

configuration also allows interpretation as the

electric field produced by a perfectly conducting
0.5
hemispherical protrusion from an otherwise flat
conductive interface at Z = 0. For a varistor
rod, this corresponds to one electrode with a 0.0
hemispherical indentation that is filled with the 0.0 0.5 1.0 1.5 2.0 2.5

silver-loaded epoxy. X

Figure B1 – Streamlines from a perfectly

conducting sphere in a uniform electric
field. For this graph, R = 1, and only one
quadrant is shown because of symmetry
about the z-axis.