5 5gfc.2_
E>/e# W.33-2
BREAKDOWN PHENOMENA
IN
HIGH POWER KLYSTRONS"
SLAC-PUB-4546
March 1988
(A!
SLAC-HJB—4546
A-
E.
VLIEKS,
M.
A.
ALLEN,
R. S.
CALLI.N,
\V.
R.
FOWKKS,
DE88 007882
E.
W.
HOVT,
J.
V.
LEBACQZ.
T. G. !.E
E
C^tifSSO^^
'
~J^
Stanford Linear Accelerator Center
Stanford University, Stanford, California 9430S
ABSTRACT
In the course of developing new high peak power klystrons
at
SLAC,
high electric fields in several regions of these devices have
become an important source of vacuum breakdown phenomena.
In addition,
a.
renewed interest
in
breakdown phenomena
for
nanosecond pulse, multi-megavoll per centimeter fields has been
sparked by recent R&D work in the area of gigawalt RF sources
The most important regions
of
electrical breakdown
are in
the output c*vjiy gap area,
the RF
ceramic windows,
and the
gun ceramic insulator.
The details
of
the observed breakdown
in
these regions, ex-
periments performed
to
understand
the
phenomena arid solu-
tions found
to
alleviate
the
problems will
be
discussed.
Recently experiments have been performed
on a
new proto-
type R&D klystron. Peak electric fields across the output cavity
gaps
of
this klystron exceed
2
MV/cm. The effect
of
peak field
duration (i,e> pulse width)
on
the onset
of
breakdown havt- been
measured. The pulse widths varied from tens
of
nanoseconds
to
microseconds. Results from these experiments will be presented.
The failure
of
ceramic
RP
windows due
to
muHipaclor
and
puncturing was ?n important problem
to
evercome in order that
our high power klystrons would have
a
useful life expectancy.
Consequently many studies and tests were made
to
understand
and alleviate window breakdown phenomena. Some
of the re-
sults
in
this area, especially the effects
of
surface coatings, win-
dow materials
and
processing techniques,
and
their effects
on
breakdown will
be
discussed.
Another important source
of
Jyslron failure
in the
recent
past
at
SLAC
has
been
the
puncturing
of the
high vollage ce-
ramic insulator
in
the gun region. The breakdown phenomenon
occurs
in a
region of apparent low electric field
and
low potential
according
to
simulation studies (solutions to Laplace's equation)
and occurs despite
the
deposition
of a
thin metallic coating
an
the inner vacuum surface, "Treeing" phenomena were also
of-
ten present although
not
necessarily
in the
same region
as the
puncturing.
A
way
of
alleviating this problem
has
been found
although the actual cause of the puncturing is not yet clear.
The
"practical" solution
to
this breakdown process will
be
described
and
a
possible mechanism for
the
puncturing will
be
presented.
INTRODUCTION
Klystrons,
and
especially high power klystrons have the
re-
markable properly that besides being useful sources of RF power
also serve as
a
test bed
for
determining the electric standoff prop-
erties
of
various materials
and
geometric suapes. This paper
describes some
of
our experience
in the
area
of
breakdown phe-
nomena related
to
some
of
the more recent klystron designs
at
SLAC.
A klystron consists
of
several key components
and
regions
which
are
subject
to
high levels
of
electrical stress. Figure
1
shows
one of our
latest production klystrons.
H
operates
at
a beam voltage
of
350
kV and
delivers 67 MW
RF
power.
Its
operating frequency is 2.856 GHt. The regions where breakdown
phenomena have been
a
problem
in
this tube
are the gun
area,
the ceramic window area
and the
output cavity area.
Fig.
K
The
67
MW production klystron.
The
gun
consists basically
of a
thermionic cathode,
a fo-
cussing electrode (maintained
at
the same potential as the cath-
ode)
and an
anode.
The
cathode
is
pulsed
to a
voltage
of
-350 kV.
The
pulse repetition rate
is
180 times per second
for
pulse widths of 5.0 usee.
In
order
to
maintain the voltage stand-
off between
the
cathode and anode
a
ceramic cylinder
(or
seal)
connects
the
two structures.
The cavity and drift tube region serve lo convert the unmod-
ulated electron beam from
the
gun
to a
sharply bunched beam
modulated
at the
input driving frequency, Each
of
the resonant
cavities develop
RF
voltages across the cavity gaps (see Fig.
I).
Beginning with
the
input cavity, which develops
an RF
volt-
age
of
approximately 2.5
k\\
each
of
the succeeding cavities de-
velop
a
higher and highcr^pl^c dup to the increasing RF beam
*Work supported
by the
Department
of
Energy, contract DE-ACCKi.76SFlK)M V
dur to the
Inviled paper presented Lo
the ~^^
XINth International Symposium on Discharges and Electrical Insulation
i"
Vacuum
" .J
Paris,
France, June 27-30, 19SS
' , i o
fe'B3
T«
rv v-
K ilir-IIKENT
IS
»\HWA\C-
modulation. At the outpi't cavity the gap voltages reach a peal
value of 420 k V. This corresponds to a peak surface electric field
or 360 kV/cm. In some of the newer klystrons even higher fields
are required for proper operation.
The output waveguide and window region directs the RF
power generated at the output cavity to the external load which
is,
at SLAC, sections of a two-mile accelerator. Since the in-
side of a klystTon must be maintained under a good vacuum
at all times (typically 10"' Terr or better) while at the same
time permit RF power to reach the external Joed with minim*!
attenuation, i low-loss ceramic window i* used. In the case or
our latest production klystron we were forced to design a dual
output window configuration because of the high peak RF fields
and wide pulse widths involved.
BREAKDOWN PHENOMENA
High Voltage Seal Puncture
Approximately two-and-a-half yean ago as the production
rate of our klystrons reached a peak, a new failure mode began
to appear in an alarming number of klystrons. The failure was
an arc through the high voltage ceramic seal of the gun. Since
the gun area of the klystron is maintained in an oil bath for
voltage standoff any puncture or the sea] not only destroyed the
internal vacuum but coated the internal parts with oil. This
meant that no parts could be reused, Initial!., i; was assumed
that the oil was contaminated with water. Testing indicated
that indeed some of the failed tubes had oil contaminated with
water, This problem was quickly brought under control but
surprisingly the punctures continued to appear. Not only did
the punctures continue but as the statistics improved it became
apparent that the location of these punctures always occurred
at the same axial location of the ceramic.
The geometry of the gun is shown in Fig. 2. The cathode
and its support structure is located at the far right of the figure.
The inner and outer anode corona rings are also shown with the
Ceramic seal sandwiched between them, An external cathode
corona shield is also located at the base of the seal. To the right
of the ceramic seal is the oil bath. The oil is tested to withstand
an electric field strength of
ISO
kV/em. The separation between
the outer anode corona ring and the ceramic seal is 1.02 cm,
and the separation between the inner anode corona ring and the
Seal is 0.32 cm. The location of the punctures was always at
about the height of the outer anode corona ring. There was
also evidence of arcing activity on the vacuum side of the seal
opposite the inner corona ring but this did not appear correlated
with the puncture sites.
VI4 1114*2 j
Fig. 2. Gun field profile with original geometry.
A series of studies was begun to investigate the electric field
strength along the ceramic seal as well as the actual breakdown
strength of the seal under actual pulsed conditions. In addition,
since all high voltage seats which we use are coated with a T'N
coating the properties of this coating were also studied.
In order to study the field strengths along the ceramic a
series of computer simulations were performed using the pro-
gram POISSON. Using this program the electrostatic problem
could be solved for several boundary conditions using real val-
ues of th» ceramic and oil dielectric constants. Two important
ceramic boundary conditions were simulated. In one case the
effect of the TiN coating was not included and in the other the
coating was assumed uniform. Figure 2 shows the equipotential
lines for the case of a uniform coating. Each equipotential line
is separated from its neighbor by 10 kV. As can be seen from
the figure the maximum potential between ceramic and outer
anode corona ring is about 55 kV. This is insufficient for break-
down to occur. For the case af no TiN coating, the fields were
reduced to only 20 kV. We also studied the standoff capability
of the ceramic under normal pulsed conditions. A ceramic seal
was place in a bath of oil with a negative high voltage electrode
place against the inner surface of the seal. The grounding elec-
trode was placed at different distances from the outer surface of
the ceramic seal. For each distance the voltage was raised until
punch-through occurred. The results are shown in Table 1.
As can be seen, (as long as the oil break-down limit is ex-
ceeded) the ceramic breakdown potential is independent of
3
the ceramic Mil to corona ring separation. From this one can
draw the conclusion thai the ceramic should not puncture even if
the oil liu a somewhat lower standoff capability than expected,
since the point where the puncture occurs is much las (55 kV
according to the simulation) than the measured breakdown level
of 165-200 kV.
Table 1
Grounding Electrode
Separation (cm) Breakdown Voltafe (kV)
0.5
165
0.25
190
0.0 170-200 (several tests)
Investigations were performed on the resistivity changes to
the TiN coating as a junction of temperature. It was thought
that since the lower (towards the anode potential] portion
or
the
ceramic sea! was shielded from the radiant heat of the cathode
support structure the coaling might have a higher electrical re-
sistivity than the rest of the surface. This would permit a higher
than expected potential in the region of the puncture than ex-
pected. Results of measurements indicated that the resistance
did drop a factor of two for each increase in temperature of 25 C.
Unfortunately no sufficient temperature gradient could be found
on the ceramic seal during actual operation to allow this effect
to occur. Different types of conductive coatings were also found
to have no effect on the puncture probability or location.
The results of our measurements indicate that for the ex-
pected potential distribution no punctures should occur. We
are therefore led to the conclusion that the ceramic must ac-
quire an excess charge density and (because the inner ceramic
surface is coated) this excess charge must reside within the bulk
of the ceramic material. A possible mechanism for this excess
charge is by the emission of
a
small flow or charge [rom the cath-
ode support structure to the ceramic. This could occur either
by thermionic emission or (less likely) by field emission from the
cathode support structure. This structure is quite hot (500 C)
and most certainly would have a thin barium coating from initial
cathode processing. Electrons which arrive at the ceramic sur-
face near the corona ring would acquired a kinetic energy close to
350 keV, With this energy they would pass through the thin TiN
coating and enter the bulk of the „crimic. (One can calculate the
penetration of 350 keV electrons into aluminum oxide ceramic
and find the mean depth to be approximately 0.026 cm). Once
inside the bulk of the ceramic the charge would only slowly bleed
off to ground since the hulk resistance is approximately
10*
13
(\
and the fields are roughly axial. In this way it is possible for
a large accumulation of charge to build up within the ceramic
and then bfeed off slowly through the length of the ceramic or arc
to the closest metallic surface. These metallic surfaces are the
inner or outer corona rings. If this mechanism is correct then a
solution to our puncture problems would be to shorten the outer
corona ring. Any arcing that could take place would then only be
to the inner corona ring. We therefore reduced the outer corona
ring by 2.5 cm. Figure 3 shows the new design. After testing
several klystrons to determine that no negative effects to the
general performance resulted this change became the new design
for all klystrons. Fifteen months and 125 klystrons later no
klystron with this new design has had a punctured high voltage
seal.
Fig. 3. Gun field profile with modified outer anode corona ring.
Window Failure
In high power klystrons, the RF ceramic windows are prone
to several different types of breakdown phenomena because or
the high levels of RF power they must transmit. Because of
their susceptibility to failure, many studies and tests have been
performed to ensure a window design with a high probability
of long life. The ceramic window assembly is shown in Fig. 4.
One may classify the main breakdown phenomena into thrt^
categories:
(a) Dielectric fai/ure. This type of failure makes itself ap-
parent by a puncture through the bulk of the window.
It is caused by field gradients across the window which
exceed the standoff capability of the material. Mech-
anisms Tor this type or breakdown will be described
below.
(6) Thermal failure. This type of failure results from ex-
cessive differential heating of the window. It is ev-
idenced by cracking (usually radial) due to thermal
stresses. '
(cj Boundary
failure.
This type or failure is caused by ir-
regularities in the brazing interface resulting in ther-
mal stresses along the periphery of the window at the
copper/ceramic interface.
Fig. 4. Output ceramic window assembly.
Window Breakdown Mechanism
One of the principle mechanisms of dielectTic failure results
from the tendency of electrons within a waveguide to gain a
net drift velocity in the direction of the Poynting vector of the
RF fields. In klystrons this results in a flow of charge from the
klystron toward the output window resulting in a buildup of
negative charge on the upstream side of tiie window. On the
other side of the window, however, a net positive charge results
because of this same electron drift.
The resulting electric field across the window can become
strong enough to exceed the dielectric strength Dr the ceramic
resulting in a puncture. Impurities or voids within the window
can enhance the chances of puncture by serving as breakdown
centers.
Another phenomena which commonly causes windows to fail
is single-surface multipactor. This phenomena usually takes
place on the downstream side of the window because this side
becomes positively charged more easily.
If an electron leaves the positively charged surface of i win-
dow (or any other nearby surface) it will be attracted back,
toward* the window surface. At the same time it can gain a
great deal of kinetic energy from the transverse RF fields near
the window surface. This gain in kinetic energy can be used
to knock out secondary electrons from the surface. [Ceramics,
such as aluminum oxide, have the property that their secondary
electron yield is quite large (2.5-4.3).' If the transverse RF fields
reverse direction in about the time it takes the secondary elec-
trons to leave the surface of the ceramic and return, a rapid
buildup of apace charge can be built up around the ceramic sur-
face and multipactor results.
A large portion of the kinetic energy gained by the electron
from the RF fields is converted to heat energy on the ceramic
surface. This heating can cause surface melting, pitting and
eventually causes the window to crack. Arcing to nearby cop-
per walls can also occur if the multipaclor results in a localized
pressure rise due to the melting ceramic or desorption of gases.
In order to alleviate the problems of multipactor and di-
electric breakdown in our klystrons, several studies have been
performed to assess the qualities of various ceramics and surface
coatings.
The ceramics studied were sapphire, BeO, boron nitride,
95%
alumina (A1300) and 99.5SS alumina (4/995). The most
promising materials from the standpoint of dielectric strength
and resistance to cracking are BeO and alumina. Sapphire win-
dows are exceedingly susceptible to cracking and boron nitride
has too low a dielectric strength.
For our latest production klystron we use a pair of alumina
windows exclusively, although we have in the past used BeO
single windows. BtO is not used because of brazing difficulties
and its inherent health hazards. However, it does have the im-
portant quality that it has the best thermal conductivity of all
the ceramics tested.
In order to alleviate the problem of surface charge buildup
as well as reduce the high secondary electron production yield
(necessary for multipactor) various surface costings have been
tested. These coatings must be able to "bleed off" the surface
electrons and have a secondary electron production coefficient
<].
Not only is the material of the coating important but the
thickness of the coating is also critical. Too thin a coating will
result in it being inefficient while too thick a coating will result
in excessive RF heating.
We have found TiN and chrome oxide coatings to work quite
effectively when used in thicknesses of 2SA. For thickness much
greater than 40A, the windows become excessively hot.
To avoid klystron failures due to materia! imperfections (i.e.,
voids,
inclusions), all windows are currently tested in a traveling
wave resonant ring where they are subjected to twice the nor-
s
ma) RF power levels before being installed in a klystron. Tim
effectively identifies the mechanically weak or damaged windows
early in production
Besides thermal cracking due to excessive heating or the ce-
ramic, windows can crack if the copper/ceramic braxe fillet has
voids or extends into the ceramic where it enters a higher RF
field region. In both cases, boundary heating can crack the ce-
ramic. Much effort has been expended in developing methods
to ensure a uniform fillet at the boundary.
In an earlier window design used in the 5045, the RF elec-
tric field at the window was reduced by nearly 5096 by altering
the RF design so that the midplanc of the ceramic was located
at a voltage minima. This was accomplished by symmetrically
locating inductive irises on each side of the window. This field
reduction at the window; however, resulted in much higher fields
elsewhere and a drastic narrowing of the passband. The re-
sulting problems with Pe'd emission and dimensional tolerances
moved us to abandon this approach.
Finally, we have found that even small amounts of dust or
foreign particles on the ceramic surface can act as arc centers.
While cleanliness and smoothness of window surfaces can be
maintained during tube construction, it is much more difficult
during actual klystron installation (and perhaps removal) in the
accelerator gallery. We therefore made a change to the output
design which has effectively eliminated window failures due to
these foreign materials. We simply changed the output configu-
ration so the windows are vertical instead of horizontal In this
way any foreign material introduced into the output waveguide
will deposit itself harmlessly in a low-field region of the waveg-
uide.
INTERNAL ARCING
At SLAC we are currently investigating the feasibility of us-
ing high power X-band klystrons in future colliders. In these
devices the cavity dimensions and cathode/anode spacings be-
came quite small while the electric potentials, both RF and
beam, remain the same or are greater. Table 2 indicates our
experience so far with cathode/anode breakdown limits.
The first three tubes in Table 2 have run easily with the
stated pulse widths and gradients.
Table 2- Anode/cathode peak field gradients.
Tube
Type
Beam
Voltage
(kV)
1!
Comments
XK5
270
292
No breakdown with 3.5 fisec
pulses
5045 350
201 No breakdown with 5 fate
pulses
150 MW
450
270
No breakdown with l.B psec
pulses
SL-3
330
303
Breakdown limited with 3.3 usee
pulses
With tube type SL-3, however, anode/cathode arcing has
been a problem. Currently we have been able to run with a
beam voltage of 350 kV after extended conditioning, but it ap-
pears that a limiting gradient in the gun region of a klystron is
—320 kV/em for 1 /aec pulse widths. Part of the problem with
this tube is due to the larger than normal focus electrode and
anode surface, which increases the probability of breakdown.
We have also calculated peak fields in the output gap of
several klystrons and for several modes of operation. Some of
these results are listed in Table 3.
Table 3. Peak output gap field gradients.
Tube
Type
Output
Power
(MW)
Maximum
Surface Field
at Output Gap
(kV/cm)
Comments
5045
100
440 No breakdown at 2 usee
pulses
5045 67
360
No breakdown at 3.5 Msec
pulses
150
MW
150
275
No breakdown at I psec
pulses
SL-3 10
803
No breakdown at 2 jjsec
pulses
SL-3
15
990 No breakdown at > 0.7 (isec
pulse?
SL-3
17
1054
No breakdown at > 0.6 /isec
pulses
We see that for the first tube types operating at 2.85G GHz,
no outpuj gap breakdown is found for pulse widths of a few
(«ec. For tube type SL-3, we see that, because of the higher RF
frequency (6,568 GHi) and smaller dimensions, the peak fields in
the output gap become much higher for lower output RF power.
An interesting point is that output gap breakdown for the SL-3
wis observed for an output power of 15 MW for pulse widths
> 0.7 iaec and at 17 MW Tor pulse widths > 0.6jisec, indicating
the pulse width dependence of breakdown.
