IEEE TRANSACTIONS ON ELECTRON DEVICES, VOL. 67, NO. 11, NOVEMBER 2020
5125
Suspended Nanoscale Field Emitter Devices for
High-Temperature Operation
Lucia B. De Rose
, Axel Scherer, and William M. Jones
Abstract
—
In this work, we demonstrate suspended
two- and four-terminal field emission devices for high-
temperature operation. The planar structures were fabri-
cated with tungsten on a 200-nm silicon nitride membrane.
The insulator in the vicinity of the terminals was removed to
minimize undesirable Frenkel–Poole emission and increase
the resistance of leakage current pathways. The effects
of temperatures up to 450
◦
C on Fowler–Nordheim emis-
sion characteristics and parasitic leakage resistance were
studied. Turn-on voltages with magnitudes under 15 V that
further decreased as a function of increasing temperature
for the two-terminal device were reported. Gating at tem-
peratures of 150
◦
C and 300
◦
C was shown for the four-
terminal device, and corresponding transconductance and
cutoff frequency values were computed.
Index Terms
—
Electron emission, Fowler–Nordheim field
emission, Frenkel–Poole emission, high-temperature, leak-
age current, metallic emitter, vacuum microelectronics.
I. I
NTRODUCTION
V
ACUUM field emission devices have been investigated
for decades [1], [2] for various
applications, including as
sources in electron beam lithography [3], flat panel displays
[4], microwave power amplifiers [5], and space propulsion
systems [6]. Since electrons in vacuum can travel faster and
with less energy dissipation than in semiconductors due to
the absence of electron–phonon scattering [7], field emission
devices can operate at higher frequencies [8] and higher
power [9]. Moreover, it is often claimed that field emission
devices are more robust and could operate in higher tempera-
ture environments [10], [11] than their solid-state counterparts.
As temperature increases, semiconductor device operation is
limited by the increase of the intrinsic carrier population [12].
This eliminates the effectiveness of chemical doping in con-
trolling the carrier population. In this way, the semiconductor
industry has moved toward wide bandgap semiconductors,
Manuscript received June 26, 2020; revised August 20, 2020; accepted
August 24, 2020. Date of publication September 9, 2020; date of current
version October 22, 2020. This work was supported in part by the
NASA/ROSES HOTTECH Program 80HQTR17C0011 under a subcon-
tract from the Boeing Company and in part by the President and Director’s
Fund at the California Institute of Technology and the Jet Propulsion
Laboratory under Award 1614028. The review of this article was arranged
by Editor M. Blank.
(Corresponding author: Lucia B. De Rose.)
Lucia B. De Rose and Axel Scherer are with the Department of Applied
Physics, California Institute of Technology, Pasadena, CA 91125 USA
(e-mail: lderose@caltech.edu).
William M. Jones is with the Jet Propulsion Laboratory, California
Institute of Technology, Pasadena, CA 91109 USA.
Color versions of one or more of the figures in this article are available
online at http://ieeexplore.ieee.org.
Digital Object Identifier 10.1109/TED.2020.3019765
such as silicon carbide (SiC) and gallium nitride (GaN) [13].
Metal field emitters, on the other hand, do not rely on chemical
doping for their device operation. Thus, they are relatively
insensitive to temperature changes until the onset of thermionic
emission. High-temperature electronics are indispensable for
many industries, including aerospace, defense, automotive,
and deep oil and gas extraction [14]. In addition, since
elevated temperatures promote the desorption of surface
contaminants, field emission devices should have more stable
electrical performance in thes
e environments than at room
temperature [15].
However, several practical nanoscale vacuum devices have
insulators near where emission takes place [16]–[18]. The
combination of the large electric field required for field
emission with a high-temperature environment in an insulator
can be detrimental to device operation as it leads to unde-
sirable Frenkel–Poole current leakage [19], [20]. This is the
field-assisted thermal ionization effect by which an insulator
becomes electrically conductive
before reaching dielectric
breakdown [21], [22]. One approach to reduce the effect
of this parasitic leakage pathway that competes with the
Fowler–Nordheim field emission is to distance the insula-
tor from the emission area [23]. In this article, we will
show the fabrication and testing at various temperatures
of in-plane metallic vacuum field emitters suspended on a
silicon nitride membrane. The insulator in the vicinity of
the high electric field region will be removed to prevent
Frenkel–Poole emission. We will use tungsten, a refractory
metal, as our electrode material due to its low work function
(approximately 4.5 eV [24]) as well as its high-temperature
tolerance.
II. F
ABRICATION
The devices were fabricated on a suspended, stoichiometric
silicon nitride (Si
3
N
4
) membrane with planar dimensions
of 5 mm
×
5 mm and a thickness of 200 nm (Norcada
QX10500DS). The suspended membrane was laterally sup-
ported by a 200-
μ
m-thick silicon frame. Standard electron
beam lithography was used to pattern connected terminals,
interconnect lines, alignment marks, and contact pads. The
contact pads were written on the supporting silicon frame.
A conducting layer consisting of 150 nm of tungsten with a
7-nm titanium adhesion layer was sputtered in an ultrahigh
vacuum magnetron sputtering system (AJA). Subsequent lift-
off was performed in acetone. A
ligned electron beam lithog-
raphy was used to redefine the electrical contact pads, onto
This work is licensed under a Creative Commons Attrib
ution 4.0 License. For more information, see http
s://creativecommons.o
rg/licenses/by/4.0/
5126
IEEE TRANSACTIONS ON ELECTRON DEVICES, VOL. 67, NO. 11, NOVEMBER 2020
Fig. 1. (a)
Ion micrograph of suspended two-terminal diode-like device
and
(b)
four-terminal triode-like device. In
(a)
, the emitter–collector gap is
measured to be 95 nm, and in
(b)
, the emitter–collector gap is measured
to be 367 nm.
which a layer consisting of 150-nm gold with a titanium
adhesion layer was deposited. The thicker contact layer allows
ultrasonic wirebonding to connect the devices without punch-
ing through the silicon nitride l
ayer to the par
tially c
onductive
silicon frame.
The connected terminals were manually separated by means
of neon focused ion beam (FIB) patterning (Zeiss Orion
NanoFab). This allowed the creation of both two-terminal
diode-like [see
Fig. 1(a)
] and four-terminal triode-like lateral
devices [see
Fig. 1(b)
]. In both cases, one of the terminals,
the so-called “emitter,” was sharpened so that it has a larger
field-enhancement factor compar
ed with the other terminals
and, thus, emits at a smaller applied bias. This step yielded a
macroscopic tip diameter of approximately 10 nm in
Fig. 1(a)
and around 40 nm in
Fig. 1(b)
. In the two-terminal device,
our intention is that this asymmetric geometry created by
the sharp emitter and the blunt collector will allow us to
mimic current–voltage (
I
–
V
) curves’ characteristic to diodes.
In the four-terminal device, we focus on maximizing the field
enhancement of a single terminal to minimize the leakage of
field-emitted currents from either the on-axis collector or the
off-axis gates.
Before ion milling, the membrane was coated with a thin
conductive carbon layer for charge dissipation. In order to
prevent crack propagation and relieve stress, two stop holes
on the off-axis plane of the terminals were milled before
the terminals were separated. This removed the possibility of
sharp corners that could serve as nucleation sites for cracks
in the silicon nitride [25]. The FIB patterning was also used
to remove the silicon nitride membrane in the vicinity of the
terminals, thus minimizing a
potential leakage pathway across
the insulating substrate. Finally, the membrane was subjected
to oxygen plasma etching to remove the conductive carbon
layer.
III. R
ESULTS AND
D
ISCUSSION
The membrane was ultrasonically wedge wire-bonded with
aluminum wires to conductive strips on a 1-in
2
ceramic
substrate. The membrane and the substrate were placed on
top of a vacuum-safe heater (HeatWave Labs) with a thermo-
couple attached to the surface of the ceramic substrate inside
of a stainless-steel vacuum system. The membrane, ceramic
substrate, and thermocouple were held via compression by
temperature-resistant molybdenum springs. A photograph of
Fig. 2.
Photograph of wire-bonded membrane and ceramic substrate
sitting on top of heater.
this setup is shown in
Fig. 2
. The strips on the ceramic
substrate were connected to the electrical feedthrough of a
vacuum chamber that routed to au
tomated electrical picoam-
meters/voltage sources (Keithley 6487). Each terminal was
linked to a separate picoammeter so that the bias of each
terminal could be independently controlled, and the current
at each terminal could be measured. In addition, 10-M
current-limiting resistors were placed in series with each of the
terminals as ballast resistors to prevent both thermal runaway
and surge currents. The vacuum chamber was pumped to
∼
10
−
7
Torr and baked for several hours with the sample
temperature held at 150
◦
C before testing to assist with
the desorption of water vapor
and other contaminants from
the surface of the various terminals. After all tests were
completed on the membrane, the devices showed no visible
changes in morphology when inspected via scanning electron
micrography.
Multiple
I
–
V
scans per device with a forward and reverse
sweep were first taken at 150
◦
C in order to remove further
adsorbates from the emitting surfaces. This “conditioning”
[26] process was repeated until the turn-on voltage appeared
to be stable. Turn-on voltage was defined as the voltage
that yields a current above the noise floor (approximately
5 nA). We selected this threshold current value to be 15 nA.
Fig. 3(a)
shows the
I
–
V
plot for the diode-like device shown
in
Fig. 1(a)
tested from 150
◦
Cupto450
◦
C in increments
of 50
◦
C. The emitter to collector gap was measured to be
95 nm. The emitter voltage
V
is defined as
V
=
V
A
−
IR
,
where
V
A
is the applied voltage on the emitter,
I
is the
emission current, and
R
is the series resistance of the ballast
resistors. A current limit of 100 nA is imposed to prevent
excessive resistive heating and potential tip destruction. The
emission current from a metallic surface can be expressed
by the so-called elementary Fowler–Nordheim equation
DE ROSE
et al.
: SUSPENDED NANOSCALE FIELD EMITTER DEVICES FOR HIGH-TEMPERATURE OPERATION
5127
Fig. 3. (a)
I
–
V
characteristic for the two-terminal device removing the
series resistance and
(b)
its Fowler–Nordheim plot for currents over 15 nA
for temperatures between 150
◦
C and 450
◦
C. The lines correspond to
the least-squares regression.
given by [27], [28]
I
=
Aa
FN
φ
E
2
exp
(
−
b
FN
φ
3
/
2
E
)
(1)
with
a
FN
=
e
3
16
π
2
̄
h
≈
1
.
54
μ
A
·
eV
·
V
−
2
and
b
FN
=
4
3
e
(
2
m
̄
h
2
)
1
/
2
≈
6
.
83 V
·
nm
−
1
·
eV
−
3
/
2
where
A
is the effective emission area,
φ
is the local work
function,
E
is the local electric field at the tip, and
e
and
m
are
the elementary charge and mass of the electron, respectively.
In addition,
E
is related to the applied voltage by
E
=
β
V
,
where
β
is the field factor. We can also relate the local electric
field
E
to the distance between the emitter and collector
d
by the relation
E
=
γ(
V
/
d
)
,where
γ
is a dimensionless
TABLE I
T
WO
-T
ERMINAL
D
EVICE
D
ATA A S A
F
UNCTION OF
T
EMPERATURE
parameter called the field-enhancement factor. Since
γ
is
geometry-dependent, a way to maximize it is to sharpen the
emitter tip by reducing its radius of curvature [29].
The device displays turn-on voltages with a magnitude of
under 13 V. Low turn-on voltages are desirable since they open
the possibility for atmospheric pressure operation [30]. As the
energy of the electrons is not high enough to ionize the atoms
present in the atmosphere, ion bombardment of the terminals
can be diminished. Further reductions in the turn-on voltage
could be achieved by decreasing the size of the emitter to
collector gap [31].
Unfortunately, the emission from either terminal is sym-
metric. This could be explained by the fact that the collector
terminal is not completely blunt but rather has an undesired
sharp protrusion, which effectively yields a
γ
comparable to
the one at the emitter tip.
The inset of
Fig. 3(a)
shows a magnified view of the zero-
corrected
I
–
V
curve for voltages near 0 V. We can observe that
before the onset of field emission, current increases linearly
with increasing voltage. This illustrates that parasitic leakage
dominates at these small fields.
We performed a linear fit of the
small field data and extracted the values for the parasitic series
resistance at the various temperatures, which are presented
in
Table I
. This parasitic resistance decreases with increasing
temperature. Since we have removed the substrate in the
vicinity of the high fields, the
likely leakage pathway is via the
supporting silicon ring, which has a resistivity of 1–30
·
cm
at room temperature as quoted by the manufacturer.
Fig. 3(b)
shows the
I
–
V
characteristic in Fowler–Nordheim
coordinates for currents above 15 nA to focus on field emission
current exclusively. The Fowler–Nordheim coordinates corre-
spond to a linearization transformation applied to (1) given
by
ln
(
I
V
2
)
=−
(
b
FN
φ
3
/
2
d
γ
)
1
V
+
ln
(
Aa
FN
γ
2
φ
d
2
)
.
(2)
The experimental data for temperatures up to 300
◦
Cdis-
play linear behavior, in agreement with the Fowler–Nordheim
emission mechanism. For temperatures above 350
◦
C, the fit
deteriorates, and the measured current does not seem to follow
a purely Fowler–Nordheim behavior. We confirm that the
emission current is not thermionic by plotting the emission
current in Richardson–Dushman coordinates (ln
(
I
/
T
2
)
versus
1
/
T
) [32] for various applied voltages on the emitter, as shown
in
Fig. 4
. Since the resulting plot i
s clearly nonlinear in nature
for all potentials shown, we rule out thermionic emission.
Hence, this could suggest thermally promoted field emission.
5128
IEEE TRANSACTIONS ON ELECTRON DEVICES, VOL. 67, NO. 11, NOVEMBER 2020
Fig. 4.
Emission current in Richards
on–Dushman coordinates for
various applied voltages.
At higher temperatures, thermally excited electrons transmit
through a thinner potential barrier as these electrons have the
tail of the energy distribution above the Fermi energy. This
contribution of thermal electrons and resultant deviation from
merely Fowler–Nordheim emission is particularly noticeable
for low currents [33], as with higher fields Fowler–Nordheim
emission dominates [34]. In addition, this enhanced tunnel-
ing probability may also explain the lower turn-on voltages
observed at higher temperatures, as, for the same potential,
the emission current is larger [35], [36].
Table I
shows the values for the slope and
y
-intercept of
the least-squares regression line, along with the value for the
coefficient of determination for all measured temperatures.
Based on the reported literature work function of tungsten
(
φ
=
4
.
5 eV) and the slope of the Fowler–Nordheim plot,
a value of 133 for the field-enhancement factor
γ
is calculated
at 150
◦
C. We perform a 2-D finite element method (FEM)
simulation (COMSOL Multiphysics 5.4) based on the device
dimension values extracted from the ion micrograph to numer-
ically compute the normal component of the electric field at the
emission tip, as illustrated in
Fig. 5
. The simulation suggests
a more modest
γ
of 3.5, a value two orders of magnitude
smaller than what we obtain experimentally. This discrepancy
may be due to uncertainty in the efficient emitter tip radius,
as emission may be taking place locally at nanoprotusions,
grain boundaries, or groups of atoms [37], which may not be
accurately resolved via an ion micrograph.
Overall, the magnitude of the slope is decreasing as a
function of temperature. As the slope depends on both
γ
and
φ
, it could be the case that with increasing temperature,
either
γ
is increasing,
φ
is decreasing, or a combination of
both. An increasing
γ
could be a consequence of temperature-
dependent desorption of residual molecules that alter the
surface roughness of the emitter tip [38]. Temperature-induced
release of gaseous adsorbates
may also explain the potential
decrease in
φ
[39], [40]. Electronegative adsorbates in the
emitter surface, such as oxygen, have been shown to increase
the work function. In this way, traditional techniques to
analyze Fowler–Nordheim plots are challenging to apply [41],
Fig. 5.
COMSOL simulation for the diode-like device of the component
of the electric field normal to the emitting surface in V/m at
−
10 V applied
on the emitter terminal.
as it is hard to remove all contaminants and produce an
atomically clean surface.
Next, building upon our two-terminal device, we also fab-
ricate in-plane four-terminal triode-like structures, as shown
in
Fig. 1(b)
. The emitter to collector distance is 367 nm.
The two structures located off-axis will act as the gates by
modulating the electric field at the tip of the emitter, which
affects the tunneling probability [42]. Ideally, the
y
-component
of the electric field between the gates and the emitter cancels
out, which minimizes the leakage of field-emitted current
into the gates. Yet, asymmetries inevitably arise during the
fabrication process. In our case, the top gate is 442 nm away
from the emitter tip, while the bottom gate is separated by
462 nm. In order to account for this unevenness, the gates
are electrically separated, and their bias can be independently
controlled.
Fig. 6
shows the
I
–
V
curve of the triode-like device for
various gate biases tested at 150
◦
C[see
Fig. 6(a)
] and 300
◦
C
[see
Fig. 6(b)
]. The current shown is the emitter current.
In this way, we concentrate on Fowler–Nordheim currents
between emitter and collector and ignore leakage of field-
emitted currents to the gates that will be later addressed.
For
Fig. 6(a)
and
(b)
, the collector is kept at ground. Vari-
ous combinations for gate bias were tested to minimize the
preferential current leakage to an individual gate and achieve
as close to a symmetric behavior as possible. Ultimately, this
condition was achieved when the bias offset between them
was set to
+
10 V for
Fig. 6(a)
and
+
50 V for
Fig. 6(b)
,
such that the top gate was at a more negative potential
compared with the bottom gate. The gate voltage reported in
the plot corresponds to the top gate bias. The runs were not
performed in a decreasing gate bias sequence as the legend
suggests but in a randomized sequence. This was done to rule
out heating effects of the emitter tip as the justification for
any observed gating. The device displays turn-on voltages of
around
−
120 V for
Fig. 6(a)
and
−
80 V for
Fig. 6(b)
.Once
again, we observe thermally promoted field emission, as the
DE ROSE
et al.
: SUSPENDED NANOSCALE FIELD EMITTER DEVICES FOR HIGH-TEMPERATURE OPERATION
5129
Fig. 6.
I
–
V
characteristic for triode-like device at various gate bias for
(a)
150
◦
C and
(b)
300
◦
C, respectively. The insets present the data in
Fowler–Nordheim coordinates for currents over 15 nA. Added emitter and collector currents to illustrate leakage to the gate terminal for
(c)
150
◦
C
and
(d)
300
◦
C.
turn-on voltage becomes smaller at higher temperatures. Yet,
these devices display larger turn-on voltages compared with
the two-terminal devices previously discussed. This results
from the considerably larger emitter-to-collector distance. The
insets correspond to the data plotted in Fowler–Nordheim
coordinates. The linearity of the plots for higher emitter
voltages confirms Fowler–Nordheim current as the emission
mechanism. It can be seen that as the gate bias becomes more
negative, the measured current is reduced for a given emitter
potential. This illustrates the effect of gating at the emitter
tip—as the potential difference between emitter and gates
decreases, the magnitude of the e
lectric field at the emission
tip is reduced, and thus, the emitted current is diminished.
Fig. 6(c)
and
(d)
portrays the addition of the emitter and
collector currents (with signs) as a function of emitter bias for
the various gate biases for 150
◦
C and 300
◦
C, respectively.
The purpose of these graphs is to show the field-emitted
current leakage from either the emitter to the top gate or from
the top gate to the collector. We can ignore the leakage to
the bottom gate, as it was measured for all emitter voltages
to be consistently under 1 nA for 150
◦
C and under 7 nA for
300
◦
C (this value was measured
when the leakage to the top
gate was the largest). The smaller leakage for the bottom gate
may be a consequence of the larger distance to the emitter tip.
A positive value in the plots corresponds to collector leakage,
while a negative value corresponds to emitter leakage. When
the potential difference between the emitter and the gates is
increased, current leaks from emitter to gate. The leakage path
then changes, with most leak
age taking place from gate to
collector when the potential difference between the emitter and
gates is decreased. Since all cu
rrents are monitored, we ensure
that no current is unaccounted for. To reduce the emitter to gate
leakage, we should improve the symmetry of the gates so that
no gate wants to preferentially take any current. To avoid gate
to collector leakage, a potentia
l solution could be to fabricate
Fig. 7.
Calculated field factor for the four-terminal device as a function
of top gate voltage for 150
◦
C and 300
◦
C.
the gates with a metal of a higher work function and, thereby,
hinder the onset of exponential emission.
We fit a linear regression model to the data, and we extract
the vertical axis intercept and slope. We consider
φ
=
4
.
5eV,
and we calculate
β
,asshownin
Fig. 7
. For both temperatures,
β
is larger when the potential difference between the emitter
and the gates is increased. This agrees with the increase in
current for a given emitter voltage when the gate voltage
becomes more positive, which confirms that the gate electric
field modulates the emission current.
Next, we compare our measured data to a simulation of
our device operation. For the simulation, we select the lower
temperature data with the smalle
st average leakage current to
the gates (top gate at
−
70 V), and we calculate an emission
area
A
equal to 6.5 nm
2
. We use COMSOL to determine that
an emitter tip radius of 0.25 nm is necessary to achieve the
β
previously calculated (see
Fig. 7
) for the specific potential con-
figuration. Next, we simulate the expected gating, as shown in
Fig. 8(a)
and
(b)
. We calculate the expected transconductance
g
m
from the simulation at
−
153 V to be 24.5 nS. From the data