Sensors and Actuators B 145 (2010) 232–238
Contents lists available at ScienceDirect
Sensors and Actuators B: Chemical
journal homepage: www.elsevier.com/locate/snb
Palladium/silicon nanowire Schottky barrier-based hydrogen sensors
Karl Skucha
∗
, Zhiyong Fan, Kanghoon Jeon, Ali Javey, Bernhard Boser
Department of Electrical Engineering and Computer Sciences, University of California at Berkeley, Berkeley, CA 94720, USA
article info
Article history:
Received 25 September 2009
Received in revised form
16 November 2009
Accepted 27 November 2009
Available online 2 December 2009
Keywords:
Hydrogen sensor
Gas sensor
Schottky barrier
Nanowire
Hydrogen detection
H
2
sensor
abstract
This work presents the design, fabrication, and characterization of a hydrogen sensor based on a palla-
dium/nanowire Schottky barrier field-effect transistor that operates at room temperature. The fabricated
sensor consists of boron-doped silicon nanowire arrays that are contact printed on top of a SiO
2
/Si sub-
strate with subsequently evaporated Pd contacts. The fabrication process is compatible with post-CMOS
and plastic substrate integration as it can be completed at temperatures below 150
◦
C with good yield and
repeatability. The sensor can reliably and reversibly detect H
2
concentrations in the range from 3ppm
to 5% and has a sensitivity of 6.9%/ppm at 1000 ppm. A response distinguishable from drift and noise is
produced in less than 5 s for H
2
concentrations over 1000 ppm and less than 30 s for concentrations over
100 ppm. The sensor settles to 90% of the final signal value in about 1 h at lower concentrations and less
than 1 min at 10,000 ppm H
2
. Drift over an 87-h measurement period is below 5 ppm H
2
concentration.
© 2009 Elsevier B.V. All rights reserved.
1. Introduction
Sensors based on carbon nanotubes (CNT) and semiconductor
nanowires (NW) have been reported for the detection of various
gases, chemicals, and biomolecules. Many of these sensors achieve
excellent performance, attributed in part to the high surface to
volume ratio of nanostructures [1–6]. Highly uniform sensor fab-
rication, however, has been a challenge in many of these efforts,
especially those enabled by the bottom-up process. The solution
presented here overcomes this problem by utilizing the statistical
averaging from multiple parallel nanowires acting as one sensor
element. The fabrication process is based on a NW contact print-
ing technology described recently [7–9]. In this approach, NWs
are grown in a CVD system on a donor substrate and are subse-
quently transferred to the sensor substrate. The separation of NW
growth from sensor fabrication imposes minimum constraints on
the growth process and the choice of substrates available for the
sensors. Since printing is a room temperature process, it is compat-
ible with a very wide selection of substrates including fabricated
semiconductors and flexible materials such as plastics or paper [7].
The sensors presented here are fabricated on Si dies as a preliminary
step towards fabricating arrays of NW sensors on active electronic
substrates with embedded evaluation circuits.
∗
Corresponding author at: BSAC UC Berkeley, 497 Cory Hall #1774, University of
California, Berkeley, CA 94720-1774, USA. Tel.: +1 510 643 6690;
fax: +1 510 643 6637.
E-mail address: kskucha@eecs.berkeley.edu (K. Skucha).
This work presents contact printed silicon nanowires (SiNW)
applied to high-sensitivity hydrogen gas detection. Hydrogen
is used in many industrial processes such as hydrogenation,
petroleum transformation, cryogenic cooling, and chemical pro-
duction. However, because this gas is odorless, colorless, and
flammable at concentrations over 4%, it poses safety concerns and
creates a need for effective H
2
leakage sensors. These applications
frequently demand sensors with a lower limit of detection (LOD)
of several ppm to identify small leaks that may be located far
from the detector. High sensitivity and low drift are also required
to reliably meet this requirement. Other important requirements
include fast response time, good linearity, long lifetime, insensitiv-
ity to environmental changes including temperature and humidity,
selectivity over other gases, and room temperature operation.
Many H
2
sensors reported to date are based on the selective
absorption of H
2
by palladium, which results in the reversible
formation of palladium hydride (PdH
x
). Several properties of Pd,
including its work function, conductivity, lattice constant, and opti-
cal properties are modulated by this absorption. H
2
sensors rely on
one or more of these effects to detect H
2
gas dissolved in ambient
air.
Several H
2
nanosensors have been developed over the past
decade that operate at room temperature and can be fabricated or
assembled using a low temperature process. To compare the per-
formance of these sensors, the sensitivity and LOD are reported.
Sensitivity is defined as the percent change in a physical param-
eter (such as resistance, current, etc.) per additional ppm of H
2
.
Hydrogen sensitive CNT transistors are created by evaporating
or electrodepositing non-continuous Pd beads onto a CNT film
0925-4005/$ – see front matter © 2009 Elsevier B.V. All rights reserved.
doi:10.1016/j.snb.2009.11.067
K. Skucha et al. / Sensors and Actuators B 145 (2010) 232–238 233
[10–12]. Upon exposure to H
2
, the workfunction change of Pd
causes an increase in resistance of the CNT. Kong et al. reported a
sensor based on this principle with a LOD below 40 ppm [11] while
Mubeen et al. reported a sensitivity of 0.42%/ppm up to 1000 ppm
H
2
concentration [12]. Nano-cluster-based H
2
sensors take advan-
tage of the H
2
-induced lattice expansion (HILE) in the Pd lattice,
which causes the metal particles to swell and consequently fill
open junctions in the cluster, increasing its conductance. Their high
surface area to volume ratio results in a response time under 1 s,
albeit at the expense of low sensitivity. Xu et al. reported a nano-
cluster-based sensor with 0.007%/ppm sensitivity at 1000 ppm H
2
concentration [13]. The reported LOD of 25 ppm corresponds to
only a 2% resistance change, less than the typical sensor drift in
many situations. Pd NW resistors exploit the natural decrease of
the conductance of Pd when it absorbs H
2
gas, but they also suf-
fer from low sensitivity. Yang et al. reported a sensitivity below
0.002%/ppm at 1000 ppm H
2
concentration and an LOD of 2 ppm
[14]. As before, the corresponding resistance change of 0.15% is dif-
ficult to detect due to intrinsic sensor drift and potential changes
in ambient conditions.
The sensitivity and LOD of reported nanosensors is vastly
inferior to Pd/bulk-semiconductor Schottky barrier-based sensors
presented in literature. In 1981, Ruths et al. presented a Pd/SiO
2
/Si
sensor with sensitivity of over 600%/ppm at 154 ppm H
2
concentra-
tion [15]. Pd/InP and Pd/SiO
2
/AlGaN diode-based sensors achieved
even higher sensitivity of approximately 1000%/ppm [16,17].At
15 ppm, the reported response of over 3800% can be easily dis-
tinguished from drift and noise [16], suggesting that the LOD is
significantly lower. Additionally, Schottky barrier-based sensors
showed a nearly linear relationship between the sensor response
and H
2
concentration over a range from 15 ppm to 1% [16].
The objective of this work is to combine the benefits of NW
sensors, including their flexible manufacturing processes, with the
high performance reported for Schottky-based bulk sensors. The
proposed sensor is based on a Schottky barrier nanowire field-effect
transistor (SBNWFET) with Pd contacts. Although Pd/Si sensors
exhibit lower sensitivity than other semiconductors due to Fermi
level pinning and undesired palladium silicide (PdSi) formation
[15], SiNWs are used in this work because of their well-understood
characteristics, tunability via doping or gate voltage, and well-
developed growth and functionalization chemistries. Moreover, Si
SBNWFETs have shown excellent electrical characteristics without
needing the high temperature anneals usually required for good
ohmic contacts in standard FETs [18].
Modulation of the Schottky barrier contacts by H
2
has already
been demonstrated for CNTs [19] and Wong et al. reported a
Schottky barrier-based CNT H
2
sensor, but it achieved a sensitivity
several orders of magnitude lower than bulk sensors [20]. The NW
sensor reported here achieves a drift limited LOD of only 5 ppm
and a sensitivity significantly exceeding that of nanosensors and
several recently reported bulk diode designs [21–23].
2. Sensing principle
Fig. 1a shows the structure of the sensor consisting of a Pd/Si/Pd
layers. As opposed to the standard metal–semiconductor architec-
ture used in diode-based sensors, the metal–semiconductor–metal
architecture was used here because it is simpler to fabricate (since
an ohmic contact is not needed), and several other benefits, such
as bi-directional sensing and potentially higher sensitivity [24].A
native SiO
2
layer exists at each Pd/Si interface and serves as a diffu-
sion barrier against PdSi formation while concurrently reducing the
effect of Fermi level pinning [15], but is too thin to inhibit carrier
transport. As seen in Fig. 1b, under ambient conditions an ohmic
contact forms between Pd and Si. As H
2
is introduced, it dissoci-
Fig. 1. (a) Schematic of the SBNWFET sensor. (b) Energy band diagram before H
2
is introduced; holes see no barrier and the contact is ohmic. Native SiO
2
layer is
omitted in this diagram because it is too thin to affect transport. (c) Energy band
diagram after hydrogen is introduced, showing a formation of a Schottky barrier.
ates into H
+
ions which thereafter diffuse into Pd and settle on the
Pd/NW interface. These ions induce a dipole layer and cause the
workfunction of the metal to effectively decrease [25]. As seen in
Fig. 1c, this eventually leads to the formation of a Schottky barrier
which impedes holes from crossing over from the metal to the NW
at the reverse biased source contact, which limits the current flow.
If this current is primarily due to thermionic transport, it has an
exponential dependence on the Schottky barrier given by:
I
s
∝ e
−q
B
/k
B
T
(1)
where I
s
is the current,
B
is the Schottky barrier height (SBH), k
B
is
the Boltzmann constant, q is the electron charge, and T is the tem-
perature. Correspondingly, the change of the barrier height from
ambient air to a certain H
2
concentration at atmospheric pressure
can be approximated with the Temkin model [26],
B
−
B
0
=
B
= V
p
ln
F
[H
2
]
[H
2
]
0
(2)
where V
p
is a fitting parameter, F is an equilibrium constant asso-
ciated with the H
+
ion coverage on the Pd/SiO
2
interface,
B
represents the change in barrier height, [H
2
] is the H
2
concentration
and [H
2
]
0
is the intrinsic H
2
concentration in ambient air, which
is approximately 0.5 ppm. The parameter V
p
captures effects that
depend on the method of Pd deposition [16] and the surface prop-
erties of the semiconductor. Substituting Eq. (1) into (2), and noting
that sensor resistance is inversely proportional to the current for a
constant bias V
ds
, yields the sensor response:
R
f
− R
0
R
0
=
R
R
0
= C × [H
2
]
n
(3)
where R
f
the resistance of the sensor at concentration [H
2
], R
0
is
the resistance in ambient air, C is a constant that equals (F/[H
2
]
0
)
n
and n = V
p
/V
T
, where V
T
= kT/q. This relationship indicates that the
change in the resistance is dependent on a power of the H
2
concen-
tration. For V
p
=V
T
the response is linear to first order. However,
second order effects such as tunneling transport and the initial
234 K. Skucha et al. / Sensors and Actuators B 145 (2010) 232–238
Fig. 2. (a) An optical image of four sensors in a Wheatstone bridge configuration. (b) A zoomed in portion of two sensors that shows a parallel array of hundreds of nanowires
contacted between two Pd terminals.
ohmic region may cause R/R
0
to deviate from linearity. In addi-
tion, because of the finite number of interface states for H
+
ions
[27], the sensor response saturates for H
2
concentrations exceeding
a few percent [26].
3. Experimental
3.1. Fabrication
SiNWs with an average diameter of 30 nm and a boron doping
concentration of 10
19
cm
−3
were grown on a separate Si substrate
using a vapor–liquid–solid (VLS) process. They were subsequently
contact printed at room temperature on a second Si substrate with
a 200 nm SiO
2
on top that was pre-patterned with lift-off photore-
sist. A rinse in octane, acetone and IPA removed the resist, leaving
regions of well-aligned, dense arrays of NWs. After printing of the
NWs, the dies were placed in a petri dish and kept at room tem-
perature in air with 40 to 55% humidity for at least one week to
facilitate formation of approximately 8 Å of native SiO
2
[28].A5s
dip in 100:1 HF, used to remove contaminants from the NW sur-
face, removed about 2–3 Å of this oxide [29], leaving the rest as a
barrier against PdSi formation. This etch was followed immediately
by the evaporation of 50 nm of Pd and a rapid thermal anneal for
30 s at 150
◦
C. The Pd contacts were made 200 m wide in order to
reduce variability due to varying NW density. Although the aver-
age length of the printed NWs is typically over 20 m, the contacts
were separated by only 2 m, ensuring that about 90% of all printed
wires form contacts between the Pd junctions. Finally, the dies
were assembled into standard IC packages and were wire bonded.
Fig. 2 shows the completed sensor. The entire fabrication process
Fig. 3. Comparison of the responses of Ni and Pd contacted NWFETs to 1% H
2
.
requires only two lithographic steps to pattern the NW print area
and the Pd contacts.
3.2. Measurement setup
All H
2
sensing experiments were conducted in a 0.5 l chamber
by mixing the appropriate amounts of Praxair 5% H
2
/95% N
2
gas
mixture and compressed air with a series of calibrated Dwyer VFA
series flow meters and Wilkerson R03 series pressure regulators.
The temperature and humidity in the chamber were maintained
between 22–27
◦
C and 18–23%, respectively. An Agilent B1500A
Semiconductor Device Analyzer was used to set the voltage bias
and measure the current through the sensor.
4. Sensor characterization and discussion
4.1. Sensing mechanism
Several gases, such as NO
2
and NH
3
, modulate the conductance
of the NW by adsorbing to its surface and acting as oxidizing or
reducing agents [5,30]. To ensure that the effect from H
2
that is
measured in this work stems primarily from the Pd/NW Schottky
barrier modulation, and not a NW surface interaction, a second
NWFET with nickel contacts was fabricated in the same process,
except it was annealed at 450
◦
C to form good ohmic contacts. Fig. 3
compares the response to 1% H
2
concentration of two NWFETs with
either Ni or Pd contacts. The virtually absent response of the sen-
sor with Ni contacts compared to that with Pd contacts confirms
that the primary transduction mechanism is due to Schottky barrier
modulation at the Pd/NW interface.
4.2. Sensor bias
Fig. 4 shows the current I
s
as a function of the drain voltage
V
ds
for different H
2
concentrations. As expected, the formation of
a Schottky barrier decreases the conductance of the sensor as H
2
is introduced. The response resembles that of a H
2
sensitive diode
in reverse bias, which generally achieves high sensitivity over a
wider range than diodes biased in forward bias [16,23,24]. Since a
SBNWFET has two back-to-back junctions, one of which is always
reverse biased, the current is symmetric around V
ds
= 0 V. Modulat-
ing the backgate voltage can also be used to tune the sensitivity. Low
backgate voltages, for example, increase the initial Schottky barrier
height of the sensor and generally reduce the extent of the ohmic
region, which is less sensitive to H
2
.AtV
bg
= −20 V, the maximum
R/R
0
achieved at 1% H
2
was over 1400, compared to approxi-
mately 500 at V
bg
= 0 V. Since a three order magnitude response is
K. Skucha et al. / Sensors and Actuators B 145 (2010) 232–238 235
Fig. 4. I
s
vs. V
ds
at different H
2
concentrations with V
bg
=0V.
not possible under tunneling transport at this NW’s doping level,
1
the current must be dominated by thermionic transport and hence
our previous use of Eq. (1) is justified. In subsequent measurements
the sensor is biased at V
ds
= 2 V, a value that represents a trade-
off between current and sensitivity, and V
bg
= 0 V, which maintains
high sensitivity over a wide range and has a low drift.
4.3. Sensitivity
Fig. 5a shows the measured sensor response as a function of H
2
concentration. The sensor response is reversible and does not sat-
urate for H
2
concentrations up to 5%. Fig. 5b shows the responses
to two consecutive 3 ppm H
2
pulses, which are clearly above the
intrinsic sensor noise. To establish the relation between the elec-
trical sensor output and measured H
2
concentration, the data from
Fig. 5a is fitted to the model specified by Eq. (3), yielding n = 0.71 and
C = 13,100. The result is shown in Fig. 5c and is in excellent agree-
ment with the model for concentrations in the range from 13 ppm
to 1%. For lower concentrations the sensor is in the ohmic region of
operation, whereas at 5% it approaches saturation. Measured data
from two additional sensors tested confirm the power law, yielding
a similar fit with values for n of 0.71 and 0.79 and values for C of
12,000 and 13,700, respectively.
Sensitivity S of the sensor, defined as
S =
d(R/R
0
)
d([H
2
])
= nC × [H
2
]
n−1
(4)
is 6.9%/ppm at 1000 ppm, while for the other two sensors it
is 6.3%/ppm and 4.6%/ppm, respectively. This represents more
than one order-of-magnitude improvement over the best result
reported previously by H
2
nanosensors based on other sensing
principles.
4.4. Response time
Fig. 6 shows the response of the sensor to an increase of the
H
2
concentration from ambient to 13 ppm. After a fast initial tran-
sient, the output changes very slowly. Common measures such as
90% settling time require knowledge of the final value the output
settles to. An accurate measurement of this value would take sev-
eral hours and exceeds the drift caused by our experimental setup.
Instead we found that the data is well modeled with the sum of two
1
Tunneling probability at a boron doping level of 10
19
cm
−3
decreases by less
than five times per 1 V of Schottky barrier change, about the maximum possible in
100% H
2
conditions. The response of our sensor is over two orders of magnitude
higher at 1% H
2
.
exponentials of the form
R(t)
R
0
=
R
final
R
0
[
1 − ae
−t/
a
− (1 − a)e
−t/
b
]
(5)
where
a
and
b
are first and second order time constants, respec-
tively, and a is a fitting parameter. Fig. 7 shows a plot of
a
and
b
obtained from fitting this model to measured data at concentra-
tions from 3 ppm to 5% H
2
concentration. The time for the signal
to settle to 50% and 90% of the final value based on this fit are also
shown. Since a ≈ 0.5, independent of the H
2
concentration, the set-
tling time to 50% is dominated by
a
while the much slower
b
dominates settling to 90% of the final value. At low concentrations,
the 90% settling time is on the order of an hour and decreases to
below 1 min at 1%, comparable to the times reported for bulk diode
sensors [16,17].
The relevant sensor response time strongly depends on the
application. Whereas in some industrial process control applica-
tions the exact H
2
concentration is needed, for leak detection it
is often sufficient to report a sudden rise. Fig. 8 reports the time
required for the sensor output to reach R/R
0
= 0.95, a value that
corresponds to 5 ppm H
2
and is clearly above the noise and drift
of this sensor. Concentrations above 25 ppm are detected in less
Fig. 5. (a) R/R
0
for H
2
concentrations between 13 ppm and 5% as a function of time.
(b) Response to two consecutive pulses of 3 ppm H
2
concentration. (c) Measured vs.
actual H
2
concentration. The dotted line represents 100% accuracy.
236 K. Skucha et al. / Sensors and Actuators B 145 (2010) 232–238
than 100 s while those above 1000 ppm are detected in less than
5s.
Fig. 9 shows the recovery time of the H
2
sensor, defined as the
time the sensor settles to a resistance corresponding to a partic-
ular H
2
concentration after exposure to 1% H
2
concentration. The
1 ppm point, for example, corresponds to a recovery of approxi-
Fig. 6. Measured sensor response to 13 ppm H
2
with fitted second order exponential
model.
Fig. 7. First order time constant (
a
), second order time constant (
b
), and the 50%
and 90% settling times vs. H
2
concentration.
Fig. 8. Response time to a detectable leak vs. applied H
2
concentration. A detectable
leak is defined as the 5 ppm H
2
level, which corresponds to a R/R
0
of 95%. For
concentrations above 2%, the actual response time is less than 1 s; the reported
value is limited by the resolution of the measurement.
Fig. 9. Sensor recovery time from exposure at 1% H
2
to R/R
0
values correspond-
ing to various H
2
concentrations. The 1 ppm concentration point corresponds to a
recovery of approximately 10% within R
0
.
Fig. 10. R/R
0
vs. time for two sensors measured over a time span of 87 h. At all
times the 5 ppm H
2
level is above the sensor drift.
mately 10% within R
0
. Depending on the resolution requirement
for the application, the relevant response time varies from minutes
to seconds. For example, in scenarios that need detection of H
2
con-
centrations of 100 ppm and above, the relevant recovery time is less
than 30 s.
We have observed that response and recovery times are strongly
affected by humidity and other contaminants, such as sulfur and
carbon monoxide, present in the compressed air that was used.
These contaminants are known to degrade response and recov-
ery times of Pd-based H
2
sensors [15,31]. In order to remove most
of these impurities, sensors were heated to 150
◦
C for 1 min in N
2
ambient before each measurement was performed. We found this
improved response times and recovery times by as much as 10-fold
in some cases.
4.5. Drift
To accurately measure the LOD and accuracy of the sensor
in presence of drift and other effects, the sensor response was
measured over an 87h long period by cycling high and low con-
centrations of H
2
in ambient air. The sensor was not heated before
this experiment so that the amount of adsorbed water vapor and
other contaminants stabilized. Fig. 10 shows an experiment in
which two sensors are alternately exposed for two hours to 1%
and 3 ppm H
2
, respectively. Each exposure is followed by a 12-h
