chemical sensing and catalysis by one - dimensional metal - oxide nanostructures

15 Mar 2004 17:40 AR AR218-MR34-05.tex AR218-MR34-05.sgm LaTeX2e(2002/01/18) P1: FHD
AR REVIEWS IN ADVANCE10.1146/annurev.matsci.34.040203.112141
(Some corrections may occur before final publication online and in print)
R
E
V
I
E
W
S
I
N
A
D
V
A
N
C
E
Annu. Rev. Mater. Res. 2004. 34:151–80
doi: 10.1146/annurev.matsci.34.040203.112141
Copyright
c
 2004 by Annual Reviews. All rights reserved
CHEMICAL SENSING AND CATALYSIS
BY
ONE-DIMENSIONAL METAL-OXIDE
NANOSTRUCTURES
Andrei Kolmakov and Martin Moskovits
Department of Chemistry and Biochemistry, University of California,
Santa Barbara, email: ,

Key Words one-dimensional nanostructures, sensors, catalysis
■ Abstract Metal-oxide nanowires can function as sensitive and selective chemical
or biological sensors, which could potentially be massively multiplexed in devices of
small size. The active nanowire sensor element in such devices can be configured
either as resistors whose conductance is altered by charge-transfer processes occurring
at that their surfaces or as field-effect transistors whose properties can be controlled
by applying an appropriate potential onto its gate. Functionalizing the surface of these
entities offers yetanother avenue forexpanding their sensingcapability. In turn,because
charge exchange between anadsorbate and thenanowire can change the electrondensity
in the nanowire, modifying the nanowire’s carrier density by external means, such as
applying a potential to the gate, could modify its surface chemical properties and
perhaps change the rate and selectivity of catalytic processes occurring at its surface.
Although research on the use of metal-oxide nanowires as sensors is still in early stages,
several encouraging experiments have been reported that are interesting in their own
right and indicative of a promising future.
INTRODUCTION
Chemical and biological sensors have a profound influence in the areas of per-
sonal safety, public security, medical diagnosis, detection of environmental toxins,
semiconductor processing, agriculture, and the automotive and aerospace indus-
tries (1–4 and references therein). The past few decades has seen the development
of a multitude of simple, robust, solid-state sensors whose operation is based on
the transduction of the binding of an analyte at the active surface of the sensor to
a measurable signal that most often is a change in the resistance, capacitance, or
temperature of the active element.
The evolution of gas sensors closely parallels developments in microelectronics
in that the architecture of sensing elements is influenced by design trends in planar
electronics, and one of the major goals of the field is to design nano-sensors that
could be easily integrated with modern electronic fabrication technologies. For
1531-7331/04/0804-0151$14.00 151
15 Mar 2004 17:40 AR AR218-MR34-05.tex AR218-MR34-05.sgm LaTeX2e(2002/01/18) P1: FHD

AR REVIEWS IN ADVANCE10.1146/annurev.matsci.34.040203.112141
152 KOLMAKOV

MOSKOVITS
Figure 1 A cartoon of a nanowire-based electronic nose. The nanowire surfaces are
functionalized with molecule-selective receptors. The operation is based on molecular
selective bonding, signal transduction, and odor detection through complex pattern
recognition.
example, the current goal is to replace the large arrays of macroscopic individual
gas sensors used for many years for multicomponent analysis, each having its
associated electrodes, filters, heating elements, and temperature detection, with
an “electronic nose” embodied in a single device that integrates the sensing and
signal processing functions in one chip (5–8). Multicomponent gas analysis with
these devices is accomplished by pattern recognition analogous to odor identifica-
tion by highly evolved organisms (Figure 1) (9–11). By increasing the sensitivity,
selectivity, the number of sensing elements, and the power of the pattern recogni-
tion algorithms, one can envision a potent device that can detect minute quantities
(ultimately one molecule) of an explosive, biohazard, toxin, or an environmentally
sensitive substance against a complex and changing background, then signal an
alert or take “intelligent” action. However, this requires an increase in the sensitiv-
ity and selectivity of active sensor elements despite the loss of active area and the
increased proximity of neighboring individual sensing elements as the individual
components are miniaturized. Recent progress in materials science and the many
new sensing paradigms originating out of nanoscience and technology, particu-
larly from bottom-up fabrication, makes one optimistic that these goals are within
reach.
Metal oxides possess a broad range of electronic, chemical, and physical prop-
erties that are often highly sensitive to changes in their chemical environment.
Because of these properties metal oxides have been widely studied, and most
commercial sensors are based on appropriately structured and doped oxides.

Nevertheless, much new science awaits discovery, and novel fabrication strate-
gies remain to be explored in this class of materials by using strategies based
15 Mar 2004 17:40 AR AR218-MR34-05.tex AR218-MR34-05.sgm LaTeX2e(2002/01/18) P1: FHD
AR REVIEWS IN ADVANCE10.1146/annurev.matsci.34.040203.112141
NANOWIRES, SENSORS, AND CATALYSTS 153
on nanoscience and technology. Traditional sensor fabrication methods make use
of pristine or doped metal oxides configured as single crystals, thin and thick
films, ceramics, and powders through a variety of detection and transduction prin-
ciples, based on the semiconducting, ionic conducting, photoconducting, piezo-
electric, pyroelectric, and luminescence properties of metal oxides (4, 12–14).
Chemical and biological sensors having nanostructured metal oxides and espe-
cially metal-oxide nanowires benefit from the comprehensive understanding that
exists of the physical and chemical properties of their macroscopic counterparts
(15).
This review is limited primarily to semiconducting devices with quasi-one-
dimensional nanostructures such as nanowires and nanobelts. Likewise, we restrict
ourselves to two related device configurations: conductometric elements and field-
effect transistors. A few issues relating to real-world sensors and sensor arrays are
also discussed.
Numerous quasi-one-dimensional oxide nanostructures with useful properties,
compositions, and morphologies have recently been fabricated using so-called
bottom-up synthetic routes. Some of these structures could not have been created
easily or economically using top-down technologies. A few classes of these new
nanostructures with potential as sensing devices are summarized schematically in
Figure 2. These achievements in oxide one-dimensional nanostructure synthesis
and characterization were recently reviewed by Xia et al. (16) and others elsewhere
(17–19). Much work has also been published on the use of carbon nanotubes,
individually or as arrays, as sensors (20–25). Although we do not refer to this
work (which has also been thoroughly reviewed) (26–30), the great progress
made to date in understanding the electronic properties of carbon nanotubes, their

Figure 2 A schematic summary of the kinds of quasi-one-dimensional metal-
oxide nanostructures already reported (see reviews 16, 17). (A) nanowires and
nanorods; (B) core-shell structures with metallic inner core, semiconductor, or
metal-oxide; (C) nanotubules/nanopipes and hollow nanorods; (D) heterostructures;
(E) nanobelts/nanoribbons; (F) nanotapes, G-dendrites, H-hierarchical nanostructures;
(I) nanosphere assembly; (J) nanosprings.
15 Mar 2004 17:40 AR AR218-MR34-05.tex AR218-MR34-05.sgm LaTeX2e(2002/01/18) P1: FHD
AR REVIEWS IN ADVANCE10.1146/annurev.matsci.34.040203.112141
154 KOLMAKOV

MOSKOVITS
reactivity toward gases, photochemical properties, junction effects, and perfor-
mance when configured as transistors certainly informs the discussion of all quasi-
one-dimensional systems. We therefore acknowledge the great debt we owe to that
literature in establishing and clarifying many of the key questions pertaining to
quasi-one-dimensional nanostructures.
The properties of bulk semiconducting oxides have been extensively studied
and documented. Not so those of quasi-one-dimensional oxide nanostructures (i.e.,
systems with diameters below ∼100 nm), which are expected to possess novel
characteristics for the following reasons:
(a) A large surface-to-volume ratio means that a significant fraction of the
atoms (or molecules) in such systems are surface atoms that can participate
in surface reactions.
(b) The Debye length λ
D
(a measure of the field penetration into the bulk)
for most semiconducting oxide nanowires is comparable to their radius
over a wide temperature and doping range, which causes their electronic
properties to be strongly influenced by processes at their surface. As a result,
one can envision situations in which a nanowire’s conductivity could vary

from a fully nonconductive state to a highly conductive state entirely on the
basis of the chemistry transpiring at its surface. This could result in better
sensitivity and selectivity. For example, sensitivities up to 10
5
-fold greater
than those of comparable solid film devices have already been reported
for sensors on the basis of individual In
2
O
3
nanowires (31). The signal-to-
noise ratio obtained indicates that ∼10
3
molecules can be reliably detected
ona3-µm-long device. By shortening the conductive channel length to
∼30 nm, the adsorption of as few as 10 molecules could, in principle, be
detected.
(c) The average time it takes photo-excited carriers to diffuse from the interior
of an oxide nanowire to its surface (∼10
−12
–10
−10
s) is greatly reduced
with respect to electron- to-hole recombination times (∼10
−9
–10
−8
s). This
implies that surface photoinduced redox reactions (Figure 3) with quan-
tum yields close to unity are routinely possible on nanowires (assuming

reactants reach the surfaces rapidly enough and interfacial charge transfer
rates are not limiting). The rapid diffusion rate of electrons and holes to the
surface of a nanostructure provides another opportunity as well. The recov-
ery and response times of conductometric sensors are determined by the
adsorption-desorption kinetics that depends on the operation temperature.
The increased electron and hole diffusion rate to the surface of the nanode-
vice allows the analyte to be rapidly photo-desorbed from the surface (∼a
few seconds) even at room at temperature.
(d) Semiconducting oxide nanowires are usually stoichiometrically better de-
fined and have a greater level of crystallinity than the multigranular oxides
currently used in sensors, potentially reducing the instability associated
with percolation or hopping conduction.
15 Mar 2004 17:40 AR AR218-MR34-05.tex AR218-MR34-05.sgm LaTeX2e(2002/01/18) P1: FHD
AR REVIEWS IN ADVANCE10.1146/annurev.matsci.34.040203.112141
NANOWIRES, SENSORS, AND CATALYSTS 155
Figure 3 A summary of a few of the electronic, chemi-
cal, and optical processes occurring on metal oxides that
can benefit from reduction in size to the nanometer range.
(e) Nanowires are easily configurable as field-effect transistors (FETs) and
potentially integratable with conventional devices and device fabrication
techniques. Configured as a three-terminal FET, the position of the Fermi
level within the bandgap of the nanowire could be varied and thus used to
alter and control surface processes electronically.
(f) Finally, as the diameter of the nanowire is reduced, or as its materials prop-
erties are modulated either along its radial or axial direction, one can expect
to see the onset of progressively more significant quantum effects (32).
Surface Reactions on One-Dimensional Oxides,
Gas Sensing, and Catalysis
The exploration of the metal-oxide nanostructures as a platform for chemical sens-
ing is a recent event. Yang and coworkers fabricated and tested the performance

of individual SnO
2
single-crystal nanoribbons configured as four-probe conduc-
tometric chemical sensors both with and without concurrent UV irradiation (33).
Photoinduced desorption of the analyte can lead to rapid detection and reversible
operation of a sensor even at room temperature. A detection limit ∼3 ppm and re-
sponse/recovery times of the order of seconds were achieved for NO
2
. Comparing
the performance of the ohmic nanoribbon sensors with those that showed rectifi-
cation led the authors to conclude that the nanoribbons themselves dominate the
photo-chemical response and not thephenomena occurring at the Schottky barriers.
15 Mar 2004 17:40 AR AR218-MR34-05.tex AR218-MR34-05.sgm LaTeX2e(2002/01/18) P1: FHD
AR REVIEWS IN ADVANCE10.1146/annurev.matsci.34.040203.112141
156 KOLMAKOV

MOSKOVITS
Figure 4 Top: TEM, HRTEM, SEM images of an individual SnO
2
nanoribbon: (A)
low magnification, (B) atomically resolved, and (C) deposited on previously prepared
Au electrodes. Bottom: The conductance response of the nanoribbon to NO
2
pulses in
air with simultaneous 365 nm irradiation (after Law et al. 33).
A wide array of potentially useful one-dimensional metal-oxide nanostructures,
including nanobelts, were synthesized and characterized in Wang’s group (19, 34)
and in other laboratories (see 16, 17 and references therein). Comini et al. (35)
configured groups of the SnO
2

nanobelts between platinum interdigitated elec-
trodes and assessed their behavior at 300–400
◦
C under a constant flux of synthetic
air. The nanobelt sensors showed excellent sensitivity toward CO, ethanol, and
NO
2
.NO
2
could be detected down to a few parts per billion.
Individual SnO
2
and ZnO
2
single-crystalline nanobelts (30–300 nm width and
10–30 nm thickness) (34) were configured as FETs and studied by Arnold et al.
15 Mar 2004 17:40 AR AR218-MR34-05.tex AR218-MR34-05.sgm LaTeX2e(2002/01/18) P1: FHD
AR REVIEWS IN ADVANCE10.1146/annurev.matsci.34.040203.112141
NANOWIRES, SENSORS, AND CATALYSTS 157
(36). The electrical properties of these individual nanobelts in vacuum, in air,
and under oxygen, as a function of thermal treatment, suggested that the oxygen
adsorption and desoption dynamics depends sensitively on the concentration of
surface oxygen vacancies, which, in turn, affect the electron density in the nanobelt.
CdO nanowires, nanobelts, and nanowhiskers are prospective active elements
for LEDs and lasers from nanostructures. The Zhou group (37) showed that
in vacuum, as-prepared CdO nanowires have a carrier concentration of
∼1.3 × 10
20
cm
−3

arising from oxygen vacancies and interstitial Cd. Temperature-
dependent conductance measurements indicate an activated process with E
a
∼ 13.3
meV at high temperature, switching to tunneling conductance below 30 K. The
conductance of single nanowires exposed to 200 ppm of NO
2
(an oxidizing gas)
at room temperature dropped by ∼30%.
Kolmakov et al. used nanoporous alumina as a template for synthesizing arrays
of parallel Sn nanowires, which were converted to polycrystalline SnO
2
nanowires
of controlled composition and size (38). Conductance measurements on these in-
dividual nanowires were carried out in inert, oxidizing, and reducing environments
in the temperature range ∼25–300
◦
C (39). At high temperatures and under an inert
or reducing ambient, the nanowires behaved as highly doped semiconductors or
quasi-metals with high conductances that depended weakly on temperature. When
exposed to oxygen, the nanowires were transformed to weakly doped semiconduc-
tors with a high conductance activation energy. The switching between the high
and low conductance states of the nanowires was fully reversible at all tempera-
tures. Configured as a CO sensor, a detection limit of ∼a few 100 ppm for CO
in dry air and at 300
◦
C was measured with these SnO
2
nanowires, with sensor
response times of ∼30 s.

The above observations can be largely accounted for in terms of mechanisms
developed over many years to explain the function of polycrystalline metal-oxide
gas sensors (40–43). This mechanism is outlined below, using SnO
2
nanowires
in the presence of oxygen (an electron acceptor) and CO (an electron donor) as a
model system for oxide semiconductor systems moregenerally. Specific departures
from this general picture are pointed out for individual cases and for other surface
adsorbate molecules when necessary.
The surface of stoichiometric tin oxide (a large bandgap semiconductor) is rel-
atively inert. Even moderate annealing in vacuum, or under an inert or reducing
atmosphere, causes some of the surface oxygen atoms to desorb, leaving behind
oxygen vacancy sites (Figure 5). Likewise, exposure to UV results in oxygen pho-
todesorption (or of other surface species) even at low temperatures. Essentially, all
experiments carried out to date on metal-oxide nanowires (or other nanostructures)
indicate that the role of oxygen vacancies dominates their electronic properties
along much the same lines as they do in bulk systems. Each vacancy results in the
formation of a filled (donor) intragap state lying just below the conduction band
edge (Figure 5c). The energy interval between these states (or at least some) and
the conduction band is small enough that a large fraction of the electrons in the
donor states is ionized even at low temperatures, thus converting the material into
15 Mar 2004 17:40 AR AR218-MR34-05.tex AR218-MR34-05.sgm LaTeX2e(2002/01/18) P1: FHD
AR REVIEWS IN ADVANCE10.1146/annurev.matsci.34.040203.112141
158 KOLMAKOV

MOSKOVITS
Figure 5 (a) Stoichiometric SnO
2
(110) surface, (b) partially reduced SnO
2

with
missing bridging oxygens. Molecular oxygen binds to the vacancy sites as an electron
acceptor. CO molecules react with preadsorbed oxygens. Electron are released back
to the nanowire [a,b after Kohl (14) with modifications]. (c) Oxygen vacancies make
SnO
2
into an n-type semiconductor. (d ) When the Debye length is comparable to the
radius of the nanowire, adsorption of electron acceptors shifts the position of the Fermi
level away from the conduction band.
an n-type semiconductor. At a given temperature the conductance of the nanowire,
G = πR
2
eµn/L, is determined by the equilibrium conditions determining the rel-
ative concentrations of (singly or doubly) ionized surface vacancy states, which
determine the electron concentration in the bulk of the material. (Surface defects
can also migrate into the interior resulting in bulk defects that are clearly much less
responsive to surface processes, and their low diffusion constant implies that they
are normally not important participants in the material’s sensing action, which re-
quires a response time faster than the inverse diffusion rate. However, bulk defects
do contribute to a sensor’s long-term stability.)
The conductance of SnO
2
changes rapidly with gas adsorption as a result of a
(usually) multistep process wherein the first is the adsorption of a molecule (for ex-
ample, with O
2
, might dissociate into two surface oxygen ions after chemisorption)
with a consequent molecule-to-SnO
2
charge transfer (or vice versa). With oxygen

as the adsorbate, the afore-mentioned surface vacancies are partially repopulated,
which results in ionized (ionosorbed) surface oxygen of the general form O
−α
β S
.
The resulting (equilibrium) surface oxygen coverage, θ, depends on the oxygen
partial pressure and the system temperature through the temperature-dependent
adsorption/desorption rate constants, k
ads/de
, on the concentration of itinerant
15 Mar 2004 17:40 AR AR218-MR34-05.tex AR218-MR34-05.sgm LaTeX2e(2002/01/18) P1: FHD
AR REVIEWS IN ADVANCE10.1146/annurev.matsci.34.040203.112141
NANOWIRES, SENSORS, AND CATALYSTS 159
electrons, n, and the concentration of unoccupied chemisorption (vacancy)
sites, N
s
.
β
2
O
gas
2
+ α · e
−
+ N
s


O
−α

βs
k
ads
· N
s
· n · p
β/2
O
2
= k
des
· θ
(where α, β = {1,2} accounts for the charge and molecular or atomic nature of the
chemisorbed oxygen (44)). In forming ionosorbed oxygen, electrons become lo-
calized on the adsorbate, creating a ∼30–100-nm-thick, electron-deficient surface
layer corresponding approximately to the Debye length for SnO
2
(in the tempera-
ture range 300–500 K), which results in band bending in the surface region of bulk
samples. For 10–100 nm diameter nanowires, the charge-depletion layer encom-
passes the entire nanowire resulting in a so-called flat-band conditions wherein
the relative position of the Fermi level shifts away from the conduction band edge
not only at the surface but throughout the nanowire (Figure 5d). Ultimately, a new
kinetic equilibrium among the free electrons and the neutral and ionized vacancies
is re-established. Under these nearly flat-band conditions at moderate tempera-
tures and for electron momenta directed radially, electrons can reach the surface
of the nanowire with essentially no interference from the low electrostatic barrier.
As a result, the electrons become distributed homogeneously throughout the entire
volume of the nanowire. Accordingly, the charge conservation condition simplifies
to

N
s
· θ =
R
2
· (n − n
m
),
where n
m
is the density of itinerant electrons remaining in the nanowire after
exposure to the adsorbate. The accompanying electron depletion n = 2N
s
θ/R
results in a significant drop in conductance:
G =
π R
2
eµ
L
·
2N
s
θ
R
and the corresponding depopulation of the shallow donor states results in an in-
crease in activation energy (39). [We neglect the dependence of the mobility on
the surface coverage, a reasonable approximation at small bias voltages and when
the electron diffusion length (∼1 nm) is much smaller than the diameter of the
nanowire (∼50 nm) (45)].

Upon adsorbing a reducing gas such as CO, the following surface reaction takes
place with the ionoadsorbed oxygen
β · CO
gas
+ O
−α
β S
→ β · CO
gas
2
+ α · e
−
which results in the reformation of the adsorption (defect) sites and the redonation
of electrons to the SnO
2
(Figure 5b). It can be shown that under flat-band condi-
tions the increase in electron concentration, n
CO
∼ p
β
α+1
CO
, and therefore in the
15 Mar 2004 17:40 AR AR218-MR34-05.tex AR218-MR34-05.sgm LaTeX2e(2002/01/18) P1: FHD
AR REVIEWS IN ADVANCE10.1146/annurev.matsci.34.040203.112141
160 KOLMAKOV

MOSKOVITS
conductivity of the nanowire, G
CO

∼ e · n
CO
(T ) · µ(T ), increases mono-
tonically with CO partial pressure (44). This was confirmed experimentally on
nanowires assuming O
−
(α, β = {1}) to be the dominant reactive surface species
(39). The foregoing simple mechanism is able to account for the operation of
tin-oxide nanowire sensors under ideal ambients consisting of dry oxygen and a
combustible gas such as CO. In a real-world environment, a large array of other
molecules (chief among them, water) complicates the picture. Surface hydrox-
yls and hydrocarbons can temporarily or permanently react with adsorption sites
modifying or adding to the possible reaction pathways.
A consequence of being able to shift the position of the Fermi level of the oxide
nanowire by applying an external field or by doping the nanowire is the possibility
of controlling molecular adsorption onto its surface (resulting in the oscillation of
the adsorbate between an electron donor and acceptor). An interesting instance of
this was reported recently (46) with In
2
O
3
nanowires exposed to NH
3
. For nano-
wires with a low density of oxygen vacancies (corresponding to a Fermi level
lower in energy within the bandgap), the adsobate behaved as an electron donor
causing the resistivity of the nanowire to increase upon exposure to ammonia.
With a higher oxygen vacancy density (the Fermi level nearer to the lower edge of
the conduction band) the NH
3

behaved as an acceptor, quenching the nanowire’s
conductance (Figure 6).
Single Nanowire FETs
The architecture of a typical nanowire-based FET is shown in Figure 7. The
nanowire acts as a conducting channel that joins a source and drain electrode. The
entire assembly rests on a thin oxide film, which, itself, lies on top of a conducting
(in this case p-type Si) gate electrode. (This is a so-called back gate configuration.
A top gate can also be deposited on the nanowire as an alternative.) Tuning a
nanowire’s properties by configuring it as the conductive channel of a FET was
Figure 6 Alternating donor (right plot) versus acceptor (left plot) behavior of
NH
3
adsorbate as a function of the doping level of an In
2
O
3
nanowire (taken from
Zhang et al. 46).
15 Mar 2004 17:40 AR AR218-MR34-05.tex AR218-MR34-05.sgm LaTeX2e(2002/01/18) P1: FHD
AR REVIEWS IN ADVANCE10.1146/annurev.matsci.34.040203.112141
NANOWIRES, SENSORS, AND CATALYSTS 161
Figure 7 A schematic traces the response of the nanowire’s conductance
to the charge state of the adsorbed molecules [adapted with permission from
Nano Lett. 2002. Copyright Am. Chem. Soc. (49)].
recently reported with single metal-oxide nanowires and nanobelts (36, 47). The
pioneering effort in this regard is from Lieber, who has also reported the opera-
tion of nanowire sensors in an aqueous medium (48). This FET device consists
of an individual Si nanowire acting as the conductive channel whose thin, native
oxide skin, used as the gate oxide, is functionalized with target-specific receptors.
These receptors change their charge state when bonded to their target species. The

layer of molecular or ionic receptors essentially acts as a polarized gate electrode
modifying the carrier density inside the Si nanowire and, therefore, its conduc-
tance. By terminating the surface with 3-aminopropyltriethoxysilane, calmodulin,
or biotin receptors, the nanowire was used, respectively, as a pH monitor, a Ca
2+
ion concentration detector, and as a sensor for a variety of biomolecules. pH sens-
ing resulted from the change in the charge state of the amine as it gained or lost
protons in response to the pH of the surrounding medium. A similar approach was
15 Mar 2004 17:40 AR AR218-MR34-05.tex AR218-MR34-05.sgm LaTeX2e(2002/01/18) P1: FHD
AR REVIEWS IN ADVANCE10.1146/annurev.matsci.34.040203.112141
162 KOLMAKOV

MOSKOVITS
used by this group to create bistable switches (49). By covering a semiconducting
nanowire with an oxide layer of variable thickness (Figure 7, top panel) and then
bonding cobalt phthalocyanine, which is capable of existing in two or more redox
states to the oxide, the layer of molecules acted as a virtual gate electrode altering
the carrier concentration in the nanowire by changing the redox state of the cobalt
phthalocyanine (Figure 7a,b). These molecules bonded to the oxide could also be
made to accept or donate electrons either by varying the potential applied to a
global back gate or by pulsing the bias voltage of the nanowire.
Although the primary goal of this particular study was to create a nanoscale
logic device, the results demonstrate the subtle interplay between the chemistry
at the surface of the oxide and the nanowire conductance. Sweeping the voltage
from negative to positive values then back again often shows significant hysteresis
and other memory effects. These effects normally arise from trapping of charges
at various interface (and other) sites, frequently residing there for a long time, or
(in the absence of heating or irradiation with light) indefinitely. This also leads to
a better appreciation of the variety of effects that a nanowire sensor must contend
with when operating in a real-world environment.

Arnold et al. (36) succeeded in creating working FETs out of pristine, individual
SnO
2
and ZnO
2
single-crystalline nanobelts and investigated their operation in
air, in vacuum, and after admitting low concentrations of oxygen and nitrogen
in vacuum chamber. The performance of the devices as a function of thermal
pretreatment in air and in vacuum was also reported. Nanobelts with conducting
channels as short as 100 nm and as long as 6 µm were used. These devices exhibited
excellent switching ratios (the resistance ratios between the ON and OFF states)
up to 10
6
and electron mobilities as high as 125 cm
2
V
−1
s
−1
at room temperature
in air. The channel conductance and the threshold voltage (defined as the value of
the gate potential required to turn the device on) of the devices were found to be
sensitive to the gas environment and thermal pretreatment.
For example, nanobelts annealed in vacuum were found to have such high
electron densities that they could not be gated at reasonable values of the gate
voltage. However, exposure to even low concentrations of oxygen dramatically
depleted the electron density in the nanobelt and shifted its threshold potential
toward more positive V
g
values, as previously indicated should be the case for an

n-type semiconductor. The switching ratio of the nanobelt FET was also found to
be strongly dependent on the channel length. For channel lengths in the range of
100 to 500 nm, no significant modulation of the conductance with gate voltage
was observed, implying some, as yet, poorly understood size dependence on the
performance of nanowire-based FET devices.
Zhou and coworkers (31) characterized and explored the room-temperature
electronic properties of individual In
2
O
3
nanowires (∼10 nm diameter) configured
as FET sensors. I
DS
(V
DS
) measured in atmospheres consisting of trace amounts
of NO
2
or NH
3
mixed in Ar showed significant conductance decreases when the
target gas was introduced (Figure 8). Both trace gases behaved as oxidizers. Apart
from the dramatic increase of the resistance, exposure to oxidizing gases induced
15 Mar 2004 17:40 AR AR218-MR34-05.tex AR218-MR34-05.sgm LaTeX2e(2002/01/18) P1: FHD
AR REVIEWS IN ADVANCE10.1146/annurev.matsci.34.040203.112141
NANOWIRES, SENSORS, AND CATALYSTS 163
Figure 8 To p:I
DS
(V
DS

) measured in pure Ar before and after 1% NH
3
is admitted
at room temperature. Note different left and right scales. Bottom: the corresponding
I(V
G
) (after Li et al. 31).
positive shifts in the threshold voltage to values exceeding 15 V, consistent with an
appreciable reduction in the carrier density inside the nanowire. The dynamic range
of the resistivity change, R
gas IN
/R
inert
, due to 100 ppm of NO
2
, was measured to be
∼10
6
, an impressive increase. Concentrations as low as 500 ppb and response times
of the order of a few seconds (with 100 ppm of NO
2
) were reported. Recovery of
the nanowire’s original conductance following the normally irreversible adsorption
of the target molecule was accomplished using UV irradiation. At high values
of V
DS
,I
DS
is often observed to saturate, occasionally with portions of the I(V)
curve showing negative differential resistance (Figure 8 top panel). The observed

current saturation at high-bias voltage (the so-called pinch-off effect) is typical
for FETs. It results from the electron depletion near the drain electrode when it
is at a high potential. The negative differential resistance, on the other hand, is an
intriguing observation. The authors tentatively ascribed this effect to bias-induced
redistribution of electron density in the conduction channel closer to the nanowire
surface, where increased scattering probability degrades the electron mobility.
15 Mar 2004 17:40 AR AR218-MR34-05.tex AR218-MR34-05.sgm LaTeX2e(2002/01/18) P1: FHD
AR REVIEWS IN ADVANCE10.1146/annurev.matsci.34.040203.112141
164 KOLMAKOV

MOSKOVITS
One can see from above discussion that setting the gate potential at an appro-
priate value increases the sensitivity S = R
analyte IN
/R
ambient
of a nanowire sensor
configured as a FET. As an example, for the In
2
O
3
nanowire sensor shown in
Figure 8, setting V
g
to −10 V results in a value of S ∼ 2 for 1% of NH
3
, whereas
under the same conditions the sensitivity of the device increases 10
5
-fold at V

g
=
−30 V. This electronic control over sensitivity is one of the major promising
characteristics of nanowire-based FET sensors, especially when they are incor-
porated into arrays in their eventual real-world applications. The question arises:
Can one control the selectivity of a FET sensor using the same general approach?
If so, a more subtle, and potentially chemically more interesting effect might be
observable. Because many target gases are detected through a catalytic reaction
(for example, CO detection is carried out using the catalytic oxidation of CO to
CO
2
on the surface of SnO
2
), gate control of sensor selectivity implies control
of surface reactivity, including catalysis, by changing the potential applied to the
gate. Recent experiments on the oxidation of CO on SnO
2
nanowires suggest that
this is indeed possible (50, 51). This effect is not unique with nanowires. Indeed,
such concepts are integral aspects of the electronic theory of catalysis and gas
sensing by semiconductors (52). The effect of an electrostatic field on the sur-
face chemistry has also been reported for thin film semiconductors and FET-based
chemical sensors fabricated by traditional technologies (53–57). The difference
between what one expects with bulk systems and with nanowires, however, is
one of extent. The combination of factors such as the comparability of the Debye
length and nanowire radius, the small number of carriers present in the nanowire
(typically ∼10
5
electrons), the high surface-to-bulk ratio, and its small capacitance
suggests that a relatively small number of adsorbed molecules can alter the car-

rier concentration significantly. Reciprocally, the removal or addition of a small
number of electrons can alter its surface chemistry measurably. For example, be-
cause chemisorbed oxygen intermediates are required in the catalytic oxidation of
CO, and a small number of electrons is required to create these adsorbed oxygen
species, the catalytic oxidation of CO should then be significantly controllable
using relatively small values of the gate voltage (Figure 9).
The source-drain current, I
SD
,ofaSnO
2
nanowire configured as a back-gated
FET was measured at constant V
SD
as a function of the gate-to-source voltage
under flowing gas with various partial pressures of nitrogen, oxygen and CO.
(Figure 9, top panel). Baseline values of I
DS
were established through prolonged
exposure of the system to dry N
2
while maintaining the system at the selected
gate potential and temperature. At time t
1
, 10 sccm of oxygen gas were mixed into
the 100 sccm nitrogen flow. This was followed at time t
2
by the addition of CO
(5 sccm) into the gas flowing into the cell. The steady-state value of the source-
drain conductance in the dry nitrogen atmosphere, G
N2

, decreases monotonically
and significantly as the gate potential becomes more negative and increases a
little and then saturates at positive values of the gate potential (Figure 9, bot-
tom panel). This behavior is expected. SnO
2
is an n-type semiconductor hence
a negative gate voltage will cause the electron density in the nanowire to de-
crease. Also plotted in Figure 9 is the conductance decrease, G
oxy
, following
15 Mar 2004 17:40 AR AR218-MR34-05.tex AR218-MR34-05.sgm LaTeX2e(2002/01/18) P1: FHD
AR REVIEWS IN ADVANCE10.1146/annurev.matsci.34.040203.112141
NANOWIRES, SENSORS, AND CATALYSTS 165
Figure 9 (a) Response of a SnO
2
nanowire’s current to the sudden addition of oxygen
to the flowing nitrogen gas at time t
1
followed by the addition of CO at time t
2
at
various values of the gate potentials at 553 K. G
N2
is the steady-state current in a dry
nitrogen environment; G
O2(CO)
, the values of the conductance decrease (increase)
when O
2
(CO) gas is sequentially admitted into the gas cell. (b) The reactivity of

oxygen, G
oxy
, and CO, G
CO
, as a function of gate voltage as measured by the
total change in conductance determined from the response curves shown in (a). The
nanowire conductance G
N2
under dry nitrogen is included for comparison. Also shown
is the extent of reaction of CO through a putative second-reaction channel that does
not involve ionosorbed oxygens as a reagent (51).
15 Mar 2004 17:40 AR AR218-MR34-05.tex AR218-MR34-05.sgm LaTeX2e(2002/01/18) P1: FHD
AR REVIEWS IN ADVANCE10.1146/annurev.matsci.34.040203.112141
166 KOLMAKOV

MOSKOVITS
exposure to oxygen and after steady state is once again achieved. Interestingly,
G
oxy
follows the gate dependence of the conductance closely. At sufficiently
negative values of the gate potential, G
oxy
drops to zero (Figure 9, bottom
panel). If one assumes that the reduction in electron density is proportional to
the coverage of ionosorbed oxygen, θ
oxy
(see above), the reduction of G
oxy
to
zero implies that at these values of the gate potential, ionosorption no longer takes

place. (Of course, one cannot be sure that physisorbed oxygen is absent from
the surface.) The CO-induced conductance increase of G
CO
(which we asso-
ciate with the catalytic oxidation of CO at the tin oxide surface) begins almost
immediately upon the introduction of CO and achieves steady state slowly. The
latter is not a monotonic function of the gate potential but shows a maximum for
gate voltage values in the range –2 to 0 (Figure 9, bottom panel). Interestingly,
the addition of CO to the gas flowing over the nanowire increases its conductance
even at V
g
=−6 V, when no oxygen ionosorption takes place. This behavior can be
reconciled if one assumes that there are at least two CO reaction channels
(Figure 10). The first channel (B in Figure 10) is the chemical reaction of CO
with the preadsorbed (inosorbed) oxygens, which depends on the availability of
free electrons and should therefore depend on gate voltage. The second channel
(A in Figure 10) is believed to be the interaction of the CO molecule with lattice
oxygen (or with some other species) that donates electrons back to the nanowire.
This reaction channel, which is assumed to be independent of the pre-existing elec-
tron density, will not be affected by altering the value of the gate potential. If this
simple model describing the two-step process—oxygen chemisorption followed
by CO oxidation—is assumed to apply, one can determine from the experimental
curves shown in Figure 9 (bottom panel) the gate–potential dependence of the
equilibrium coverage of ionosorbed oxygen before [θ
1
(V
g
)] and after CO is ad-
mitted into N
2

+ O
2
mixture [θ
2
(V
g
)]. Using this model, the two coverage values
can be obtained from the experimental data as G
oxy
(V
g
) = C · [θ
1
(V
g
)] and
G
co
(V
g
) ≈ A − C · [θ
2
(V
g
) − θ
1
(V
g
)]. The results are shown in the Figure 10 (the
constant A represents here the gate-independent reaction channel). According to

this treatment, the major effect of admitting CO is to shift the equilibrium oxygen
coverage to lower values at any given value of V
g
. At negative values of V
g
, the
combined effect of the low-electron density and reaction with CO totally elimi-
nates ionosorbed oxygen from the surface of the nanowire (triangles in Figure 10
bottom). When electrons (and therefore ionosorbed oxygens) become plentiful as
V
g
becomes large and positive, the role of CO decreases and the equilibrium oxy-
gen coverage at high-gate potential values is no longer sensitive to the presence of
CO. This phenomenon can therefore account for the amonotonic dependence of
G
co
(V
g
) (Figure 9). It is important to note that the drop in the value of G
co
(V
g
)
does not mean that the CO oxidation process is slowing down but rather indicates
the limitations in associating conductance values, proportionally, with surface cov-
erage without taking into account changes in the surface-chemical mechanism.
The above two-channel model describing CO oxidation on the tin dioxide
nanowire has an interesting implication for nanowires as sensor (or, reciprocally,
15 Mar 2004 17:40 AR AR218-MR34-05.tex AR218-MR34-05.sgm LaTeX2e(2002/01/18) P1: FHD
AR REVIEWS IN ADVANCE10.1146/annurev.matsci.34.040203.112141

NANOWIRES, SENSORS, AND CATALYSTS 167
Figure 10 Top: Schematic diagram of the major electron donor-acceptor reaction
channels: (A) interaction of CO molecule with lattice oxygens (or some other acceptor
site); (B) the same with ionosorbed oxygens; (C) oxygen ionosorption (the reverse,
i.e., desorption process is not shown here). For simplicity, we imply O
−
S
to be the
dominant chemisorbed species, although both single atom and diatomic species are
known to exist on the surface under appropriate conditions. Bottom: The calculated
equilibrium relative oxygen coverage as a function of gate potential before (squares)
and after (triangles) CO gas is admitted (51).
as a CO oxidation catalyst). Because the first channel depends directly on the
ionosorbed oxygen coverage, increasing this coverage by applying a large positive
potential on the gate results in increased sensitivity toward CO. At low (nega-
tive) gate potentials, the reaction mechanism is independent of the ionosorbed
oxygen, causing the CO-sensing mechanism to switch to another mode. This il-
lustrates the prospect of a highly gate-tunable sensor (or nano-catalyst) whose
sensitivity (reactivity) and selectivity is tunable using the gate potential as a
15 Mar 2004 17:40 AR AR218-MR34-05.tex AR218-MR34-05.sgm LaTeX2e(2002/01/18) P1: FHD
AR REVIEWS IN ADVANCE10.1146/annurev.matsci.34.040203.112141
168 KOLMAKOV

MOSKOVITS
surface-chemistry determining parameter. The nanowire FET then becomes, in
essence, a sensor (nano-reactor) with electronically controllable selectivity and
sensitivity (reactivity).
PHOTOCHEMICAL PROPERTIES OF INDIVIDUAL
METAL-OXIDE NANOWIRES
The photochemical and photophysical properties of metal-oxide nanoparticles

and materials fabricated from them have been studied extensively, especially
in the context of solar energy conversion. In a similar vein, small diameter,
quasi-one-dimensional oxide nanostructures are promising photocatalysts on ac-
count of the efficient migration of electrons and holes to the nanostructure surface
where they can participate in chemical reactions before recombining. Although
these sorts of applications of oxide quasi-one-dimensional nanostructures are in
their infancy, a few compelling studies have already appeared. Dai and coworkers
recently reported molecular photodesorption from single-walled carbon nanotubes
and its dramatic influence on the electronic properties of a FET fabricated from
it (58). Yang and coworkers observed a record six-order-of-magnitude increase
in conductance when ZnO nanowires were exposed to 0.3 mWcm
−2
of 365 nm
UV light (whose photon energy just exceeds the material’s bandgap cut off) (59).
Similar results were reported for SnO
2
nanoribbons (33) whose sensing properties
were investigated under the influence of photoexcitation. The photoresponse of
the nanowire and its associated time constants were found to depend sensitively
on the nature of the gas environment. On the basis of this observation, the au-
thors suggested that the photoresponse of nanoribbons and nanowires depended
on two contributions: (a) direct excitation of electron-hole pairs, which produce
a photocurrent that is influenced by the bias, and (b) photoinduced desorption of
ionosorbed species through a photochemical reaction of the form h
+
+ A
−
→ A
0
(A

−
= O
−
,O
2−
,NO
2−
etc.), which eliminates the adsorbate-induced gating of the
nanoribbon thereby increasing the conductance of the nanowire. These effects are
exploitable in low-temperature sensing where many molecules stick irreversibly
to oxide surfaces. For example, room temperature detection of NO
2
at concentra-
tions as low as 3 ppm and with response/recovery times of the order of seconds
was reported with concurrent UV illumination. Although comparable sensitivity
was reported for thin film semiconducting oxide sensors at room temperature (and
under UV illumination) (60, 61), their recovery times are of the order of hours.
Avouris and coworkers reported similar observations on ZnO nanobelt FETs
(36). On shutting off the 350 nm excitation, the current decay kinetics showed
two characteristic time constants: a fast process, which was likely due to electron
recombination, and a slow decay process, which was ascribed to the readsorption
of the (electron acceptor) molecules.
On illuminating an In
2
O
3
NW in air at room temperature with 3 mW/cm
2
of
254 nm UV, Zhou et al. (62) observed a 10

4
-fold increase in conductance
(Figure 11, top). The negative shift of the threshold voltage of the nanowire
15 Mar 2004 17:40 AR AR218-MR34-05.tex AR218-MR34-05.sgm LaTeX2e(2002/01/18) P1: FHD
AR REVIEWS IN ADVANCE10.1146/annurev.matsci.34.040203.112141
NANOWIRES, SENSORS, AND CATALYSTS 169
Figure 11 To p: Photoresponse of the current through a In
2
O
3
nanowire to photoex-
citation, in turn, with 365 and 254 nm radiation. Bottom:TwoNO
2
-sensing cycles:
(a) 254 nm light is ON; (b) light is OFF and Ar flux is ON; (c) 0.1% NO
2
is ON;
(d)NO
2
is OFF; (e) Ar is OFF; ( f ) 254 nm light is ON followed with send cycle (from
Zhang et al. 62).
configured as a FET upon UV exposure suggested a release of electrons into its
conduction band with electron concentration increases ∼1.8 nm
−1
and ∼7.7 nm
−1
of nanowire length for 365 nm and 254 nm light, respectively. The measured
transconductance of the FET indicated not only an increased electron
density but also 7-fold and 80-fold respective increases in mobility for the two
15 Mar 2004 17:40 AR AR218-MR34-05.tex AR218-MR34-05.sgm LaTeX2e(2002/01/18) P1: FHD

AR REVIEWS IN ADVANCE10.1146/annurev.matsci.34.040203.112141
170 KOLMAKOV

MOSKOVITS
wavelengths used. (The increased electron mobilities might be due to photon-
assisted electron excitation out of traps.) Because all measurements were per-
formed in air, the results were explained in termsof the conventional modelwherein
increases in conductance are linked to photoinduced quenching of O
−
2
and OH
−
electron acceptors and the photogeneration of electron-hole pairs. The desorption
of adsorbed NO
2
(an electron acceptor) was also studied (Figure 11, bottom). The
same group investigated the infrared photodetection properties of CdO nanowires
(37). Promising results were obtained with pulsed 60 mWcm
−2
of 950 nm light at
1.2 K, which caused the photoconductance to switch with an on/off ratio of 8.6.
[CdO is an indirect bandgap semiconductor (E
indirect
g
∼ 0.55 eV).]
OXIDE NANOWIRES IN A REAL-WORLD ENVIRONMENT
Most nanowire sensor studies are carried out in idealized atmospheres. Although
studies with nanowire sensors are rather recent, it is not too early to consider
some of the challenges associated with fabricating sensing devices and carrying
out electron transport measurements under more realistic conditions (63). Nan-

odevice fabrication includes a number of steps (such as the nanowire growth,
application and removal of a photoresist, deposition of electrodes) where the sur-
face (and bulk) of the nanowires and the surrounding area are exposed to reactive
species. In addition, in an ambient environment the nanostructure surface or its
junctions can adsorb reactive species capable of altering its device performance.
This likely accounts for the significant scatter in the electron transport values,
the sensing behavior, and the performance as a FET reported on closely similar
systems by different groups. Even presumably pristine individual nanostructures
manifest memory effects (that is, electron transport that apparently depends on
the order in which measurements are carried out) or hysteresis effects in the
I
DS
(V
DS
)orI
DS
(V
G
) measurements (63). In an attempt to reduce these effects,
several empirical cleaning procedures and measurement protocols have been pro-
posed, which usually include preannealing the nanowire device in oxygen or in
vacuum, UV irradiation, plasma or ozone surface cleaning, the use of initial high-
current pulses and so on. The origin of the current instabilities observed with oxide
nanowires fall into three groups: (a) contacts effects, (b) adsorbed contaminants,
and (c) impurities in or on the support layer (which is often SiO
2
) in the proximity
of the nanostructure. As was shown for carbon nanotubes, Schottky barriers formed
at the carbon/metal contacts and their change in properties with gate potential or
as a result of gas adsorption can dominate the measured transport properties of the

device (30). Using a reactive metal (e.g., Ti) underlayer in forming Ti/Au contacts
and isolating the contacts from the gaseous environment by covering them with an
inert layer reduces but does not fully eliminate these effects. This long-standing
problem was recently revisited for macroscopic sensors where the effect of the
applied potential, the influence of the ambient on the state of the Schottky barrier,
and the role of poisoning catalytic reactions occurring at the contact electrodes
15 Mar 2004 17:40 AR AR218-MR34-05.tex AR218-MR34-05.sgm LaTeX2e(2002/01/18) P1: FHD
AR REVIEWS IN ADVANCE10.1146/annurev.matsci.34.040203.112141
NANOWIRES, SENSORS, AND CATALYSTS 171
Figure 12 (a) The response of the source-drain conductance (I
DS
) of a SnO
2
nanowire
resulting from the slow dissipation of accumulated surface charge in the vicinity of
the nanowire, which acts as parasitic gate. The charge was induced by the imposi-
tion of a negative potential V
g
on the back-gate electrode (A. Kolmakov, Y. Lilach
& M. Moskovits, unpublished data); (b) Surface potential map image of a biased
nanowire connecting two electrodes, obtained using AFM.The halo-like areas along the
nanowires are likely from induced surface charge. (G. Cheng, K. Jones & M. Moskovits,
unpublished data).
were discussed (44). No corresponding systematic studies have been done so far
for oxide nanowires where such effects are expected to be at least as signi-
ficant as in macroscopic systems. The contamination of the nanowire surface
or the gate oxide in the proximity of the nanowire, with water or certain other
molecules, can act as a (poorly controlled) external gate and alter the measured
conductance so that it shows memory or hysteresis effects, which can take a long
time to dissipate in the absence of treatments for discharging the trapped charges

using UV irradiation or thermal treatments. This issue was recently explored with
carbon nanotube FETs (63). Memory effects whose characteristics depend criti-
cally on how the nanowires were fabricated, stored, and operated are also often
observed with metal-oxide nanostructures. For example, Figure 12a shows the time
evolution of the source-drain current through a SnO
2
nanowire FET following a
sudden change in the gate potential (from large negative values to zero) under dry
N
2
and at 300 C (A. Kolmakov, Y. Lilach & M. Moskovits, unpublished data).
The large observed decrease in I
DS
followed by a slow decay is likely due to the
accumulation of positive surface charge in the proximity of the nanowire (or on its
surface) when the gate was negatively biased. When the gate is suddenly grounded,
this surface charge does not dissipate immediately but acts toward the nanowire
as an effective positive gate. That is, the electron density in the SnO
2
nanowire
remains high as long as these surface charges remain in place, adding their influ-
ence to that of the potential applied to the gate. When the initial gate voltage is less
negative (see second cycle in Figure 12a), less charge is induced, resulting in a
diminished transient current response and one of shorter duration. These induced
15 Mar 2004 17:40 AR AR218-MR34-05.tex AR218-MR34-05.sgm LaTeX2e(2002/01/18) P1: FHD
AR REVIEWS IN ADVANCE10.1146/annurev.matsci.34.040203.112141
172 KOLMAKOV

MOSKOVITS
surface charges can be seen as halos (dark or bright depending on sign) in the

neighborhood of the biased nanowire in images obtained using scanning potential
microscopy (Figure 12b).
POSSIBLE ARCHITECTURES OF QUASI-ONE-DIMENSIONAL
METAL-OXIDE NANODEVICES
Among the many synthetic strategies reported for creating nanowires, those based
on bottom-up synthesis are normally regarded to present the most cost-effective
means for producing nanowires in large quantities. Nevertheless, most synthe-
sis techniques, such as VLS growth, produce randomly oriented assemblies of
nanowires. Most studies to date have been carried out on individual nanowires
selected by some means from these assemblies, then wired up with appropriate
contacts appropriate for transport measurements. This is done largely because,
at present, it is the properties of individual nanowires that offer the most direct
route to new fundamental science. However, if quasi-one-dimensional nanosys-
tems are to be exploited in massively parallel sensor applications, one needs
to consider engineering strategies for producing ordered structures consisting
of a large number of individual cells that can be regionally functionalized and
individually read. Several promising techniques have been reported for align-
ing or otherwise imposing a level of architecture on nanowires, including the
use of microfluidics, electrostatic or magnetic fields, surface prepatterning, self-
assembly, and templating. These architecture principles through which ptoential
prototype devices can be developed have been comprehensively reviewed recently
(16, 64–66).
Many applications including chemical sensing and catalysis rely on achieving a
high surface- to-bulk ratio in the active nanosystem and therefore do not necessarily
require high crystallinity. This potentially expands the range of strategies for pro-
ducing technologically viable, low-cost devices based on quasi-one-dimensional
oxide nanostructures. Here we discuss a few recent accomplishments having in
common the fact that they combine thin film deposition technology with the ten-
dency of the underlying nanostructure to self-organize.
Arrays of free-standing metal/metal-oxide nanowire arrays were fabricated

using porous alumina (PAO) templates. Briefly, highly ordered and periodic
nanopores in PAO were produced by a two-step anodization (67). By varying the
anodizing conditions, the electrolyte temperature, the anodizing voltage, and/or the
nano-texture of the aluminum substrate, the diameter, length, and the density of
the parallel nanopores can be tuned in the ranges, respectively, ∼10–300 nm,
∼300 nm to 100 µm, and ∼5 × 10
9
to 10
11
pores/cm
2
. Metal, metal-oxide, or
compound semiconductor nanowires and nanotubules of the required length, di-
ameter, and composition (including heterostructures) can then be grown inside the
nanoporous template using electrochemistry, vapor deposition, sol-gel, or VLS
synthesis (68–72). After growing the nanowires, annealing the system, removing
15 Mar 2004 17:40 AR AR218-MR34-05.tex AR218-MR34-05.sgm LaTeX2e(2002/01/18) P1: FHD
AR REVIEWS IN ADVANCE10.1146/annurev.matsci.34.040203.112141
NANOWIRES, SENSORS, AND CATALYSTS 173
Figure 13 Left top: Cartoon of a planar nanowire array sensor with individually
addressable (multiple-nanowire) sensing elements based on PAO template synthesis.
Bottom: an intermediate step in the fabrication of a planar nanowire array sensor before
oxidation of the Sn nanowires (G. Cheng, A. Kolmakov & M. Moskovits, unpublished
data). Right top: the sensing performance of an individual SnO
2
nanowire (39).
the underlying aluminum layer, and etching down the oxide matrix to expose the
nanowire tips at both ends, a system of patterned metal electrodes is
lithographically deposited top and bottom to act both as addressable pads for
electrically contacting the array of wires and as the anchoring structure that main-

tains the integrity of the nanowire device (Figure 13, left top graph). Much of the
alumina matrix is then removed by wet etching, leaving behind a planar array of a
large number of nanowires (Figure 13, bottom), most of which are fully exposed
to the ambient medium (G. Cheng, A. Kolmakov & M. Moskovits, unpublished
data). A typical device fabricated in this manner would contain hundreds of indi-
vidually functionalized cells of nanowires that can be interrogated using cross-bar
electrodes. The bulk composition can be tuned during the nanostructure growth
(for example via changing reactants). Sub-micrometer cell sizes and distances
over which the functionalization varies are routinely possible. This has resulted in
multisensor arrays with a large number of individual sensors in an area a few
microns across, which creates a so-called electronic nose with thousands of indi-
vidually readable receptors.
15 Mar 2004 17:40 AR AR218-MR34-05.tex AR218-MR34-05.sgm LaTeX2e(2002/01/18) P1: FHD
AR REVIEWS IN ADVANCE10.1146/annurev.matsci.34.040203.112141
174 KOLMAKOV

MOSKOVITS
Arrays of nanotubules were also successfully grown inside the channels of
porous anodic oxide films (73, 74). Such devices have the added advantage that
the gas or fluid carrying the analyte can flow through the interior of the nanotubules,
potentially improving the device’s detectivity (or if used as a catalyst, its catalytic
efficiency). Moreover, the supporting oxide matrix need not be removed for the
sensing surface to be able to contact the ambient carrying the analyte. In principle,
large areas could be covered with such nanostructures.
Because titanium metal is one of the metals capable of growing porous ox-
ides, the Ti nanoporous template itself can be used as a sensor platform. This
was successfully reported recently by Egashira’s and Grimes’s groups, who fab-
ricated planar array sensors based on titania nanotubules and the nanostructured
porous titania itself (75–77). Arrays of parallel titania nanotubules with diame-
ters in the range 20–100 nm were obtained by anodizing Ti foil in hydrofluoric

acid solution (Figure 14, top left). As with alumina, control over the diameter
Figure 14 TiO
2
nanotubules (top left) and planar sensor (top right) based
on self-organization of oxide during the anodization of Ti metal foils. Bottom:
The sensing performance at 290
◦
C toward H
2
(after Varghese et al. 77).
15 Mar 2004 17:40 AR AR218-MR34-05.tex AR218-MR34-05.sgm LaTeX2e(2002/01/18) P1: FHD
AR REVIEWS IN ADVANCE10.1146/annurev.matsci.34.040203.112141
NANOWIRES, SENSORS, AND CATALYSTS 175
and length of the nanotubules was achieved by varying the anodization voltage
and time. The stoichiometry and crystallinity were improved by postannealing
at high temperatures in oxygen. Impedance measurements were carried out by
depositing two platinum electrodes, as shown in Figure 14 (top right). Record
values for hydrogen sensitivity with TiO
2
were reported for this nanotubule ar-
ray sensor (Figure 14, bottom). The superior sensitivity was explained in terms
Figure 15 Silica helices: (a) pillars, (b) zig-zag/chevron structures (d) grown using
ballistic glancing angle deposition (GLAD) (79). (c) Top view of the pillar structure
grown on a lithographically preseeded support (80). (e) The response of a humidity
sensor determined by SiO
2
pillars fabricated in this manner (84).