H₂S sensing for breath analysis with Au functionalized ZnO nanowires

The ZnO NWs used here were grown by high temperature CVD using the vapour liquid solid (VLS) method [23] in a three zone tube furnace. For the growth, $300\,{\rm{mg}}$ of a mixture of ZnO ($99.99 \%$ purity) and carbon powder (molar ratio of ZnO to C was 1:1) was placed inside a quartz boat and positioned inside the furnace at a nominal temperature of $1045\,^\circ {\rm{C}}$ at a pressure of $900\,{\rm{mbar}}.$ The substrate, a $1\,{{\rm{cm}}}^{{\rm{2}}}$ silicon (100) wafer piece with a $3\,{\rm{nm}}$ thick catalytic gold (Au) layer on top, was positioned $25\,{\rm{cm}}$ downstream of the source material inside the liner tube of the furnace at a temperature of $1057\,^\circ {\rm{C}}.$ During the $60\,{\rm{\min }}$ of the growth process, evaporated ZnO(g) was transported via an argon carrier gas (Ar, $190\,{\rm{sccm}},$ $99.998 \%$ purity) towards the oxygen inlet (O₂ $99.995 \%$ purity diluted in Ar down to $5 \% ,$ $1.14\,{\rm{sccm}}$) which was positioned $1\,{\rm{cm}}$ upstream of the substrate. The catalytic gold film melts at the beginning of the process and forms Zn saturated Au/Zn droplets. Ongoing oxidation of the Au/Zn alloy then forms a dense forest of 80–100 nm thin and up to $40\,\mu {\rm{m}}$ long nanowires. Scanning electron micrograph in figures 1(a)–(b) shows an example of such a dense ZnO NW forest.

Zoom In Zoom Out Reset image size

A detailed analysis of the ZnO NW growth process and an in-depth analysis of the ZnO NW properties were reported by Li et al [24].

Sensors were fabricated by deposition of ZnO NWs on top of a planar titanium/gold (Ti/Au) contact structure on a silicon substrate with $900\,{\rm{nm}}$ of insulating SiO₂ on top. The two parallel electrodes were $1.5\,{\rm{cm}}$ long and separated on average by a $5\,\mu {\rm{m}}$ gap. This gap was bridged with pristine ZnO nanowires, typically 40 μm long and 50–160 nm in diameter with a peak diameter of 80–100 nm (figure S1 (available online at stacks.iop.org/NANO/32/205505/mmedia)), which were scratched from the silicon substrate and dispersed in isopropanol using a sonication bath. The dispersed nanowires were then drop coated onto the planar contacts and aligned by dielectrophoresis. All specimens were stored in and exposed to identical environmental conditions simultaneously within each measurement series. Figure 1(c) shows a SEM picture of a typical gas sensor used in our experiments. A regular perpendicular alignment of the nanowires relative to the Ti/Au electrodes can be observed. The as-prepared sensors were then tested in our gas sensing setup by applying a constant voltage of $1\,{\rm{V}}$ leading to a typical total channel current of $1\mbox{--}10\,\mu {\rm{A}}$ through the ensemble of contacted ZnO NWs directly after fabrication. After electrical and gas sensitive testing of sensors with pristine ZnO NWs, sensors with comparable channel current level at $1\,{\rm{V}}$ and sensing characteristics towards H₂S were selected and surface functionalized in the next step. In order to achieve a reproducible and uniform distribution of Au catalyst on the ZnO NW surface, magnetron sputtering was applied. A Bal-Tec Med 020 high vacuum coating system was used to deposit thin nanoparticle layers over the entire surface of each gas sensor unit with nominal layer thicknesses ranging from $1\mbox{--}10\,{\rm{nm}}.$ The sputtering condition for Au was $5\cdot {10}^{-2}\,{\rm{Pa}}$ base pressure and $7\cdot {10}^{-3}\,{\rm{Pa}}$ sputtering pressure in pure argon atmosphere at room temperature. Setting the sputtering power to a low value of only $7\,{\rm{W}}$ on a $2^{\prime\prime}$ Au target allowed for a controlled surface functionalisation of pristine ZnO NWs with a distinct range of catalyst loading thickness via control of the sputtering time. For a $5\,{\rm{nm}}$ thick Au nanoparticle film a sputtering time of $90\,{\rm{s}}$ was acquired. The thin film thickness was monitored during the sputter process via a quartz crystal microbalance placed besides the gas sensor samples.

Optical properties of the pristine ZnO NWs were tested by low-temperature ($T\approx 5\,{\rm{K}}$) and room temperature PL spectroscopy using a solid-state laser operated at $320\,{\rm{nm}}$ wavelength for excitation, and a CCD camera attached to a $f=1\,{\rm{m}}$ monochromator for detection. The surface morphology and composition of our Au functionalized ZnO NWs was investigated by scanning electron microscopy (SEM) and energy-dispersive x-ray spectroscopy (EDX), comparing scans before and after successful sputtering. The noise power spectra of the sensor structures were measured using a dual channel digital lock-in amplifier and showed the expected $1/f$ dependence over the whole range from $1\,{\rm{Hz}}$ to $100\,{\rm{kHz}}.$

All gas sensors were measured in a sealed circular dynamic flow chamber with a total gas volume of $11\,{\rm{ml}}$ with space for 2 sensors at a time. The small volume allowed for a fast exchange of different test gas atmospheres when the gas testing chamber is flushed with a flow of $42\,\mathrm{sccm}\,.$ The operating temperature of the sensors was kept at $(20\,\pm \,1)\,^\circ {\rm{C}}$ by a water bath. Via a computer controlled dilution stage, consisting of various valves, a pressure controller, and four mass flow controllers, the initial $1\,{\rm{ppm}}$ of H₂S ($98 \%$ purity) diluted in nitrogen ($99.9999 \%$ purity) were mixed with nitrogen ($99.9999 \%$ purity) or synthetic dry air and further diluted to H₂S concentrations in the low ppb range. Synthetic air was mixed within the dilution stage itself in a ratio of 4:1 of nitrogen ($99.9999 \%$ purity) and oxygen ($99.995 \%$ purity). For the electrical measurements, a constant voltage of $1\,{\rm{V}}$ was applied to the gas sensors. The current $I$ of each gas sensor was measured by a combination of switchable low-noise transimpedance amplifiers and a 12 bit analog-digital converter (ADC), recording the actual $I$ values each second. The resulting $I\left(t\right)$ traces were analysed in terms of a set of evaluation parameters, which were estimated from the quiescent current level in nitrogen or synthetic air, and the current measured in the presence of the target gas H₂S. The main focus of our evaluation is on the response $R,$ response rate $\dot{E},$ sensitivity $S,$ and limit of detection $LOD.$ The response $R$ was defined as

$$\begin{eqnarray}&&R\left( \% \right)=\left({I}_{g}-{I}_{0}\right)/{I}_{0}\cdot 100 \% \end{eqnarray} \tag{ 1 }$$

here ${I}_{g}=I({t}_{g})$ is the current level of the ZnO NWs in the presence of the target gas H₂S, and ${I}_{0}=I({t}_{0})$ is the quiescent current level of the ZnO NWs in nitrogen or synthetic air [25], with ${t}_{g}$ and ${t}_{0}$ as the times at the end of the target gas interval and quiescent gas interval respectively.

Because the sensors in this work did not show signal saturation, even after being exposed to a target gas atmosphere for several hours (figure S2), we introduce a new figure of merit, the response rate $\dot{E},$ defined as

$$\begin{eqnarray}&&\dot{E}=\displaystyle \frac{I\left({t}_{g}\right)-I\left({t}_{0}\right)}{{t}_{g}-{t}_{0}}=\displaystyle \frac{{\rm{\Delta }}I}{{\rm{\Delta }}t}.\end{eqnarray} \tag{ 2 }$$

Here, ${\rm{\Delta }}t$ is the duration of the flushing interval, and ${\rm{\Delta }}I$ is the increase of the initial current level during this interval with the ZnO NWs in target gas atmosphere. The response rate does not depend on the initial base current level ${I}_{0}$ and thus is not influenced by possible underlying slow current drifts. Moreover, the response rate enables the comparison of all sensors with linear non-saturating gas sensing behaviour, which cannot be characterized in the usual way by determining a reaction or response time. The response rate $\dot{E}$ can already be estimated within the first minute of exposure to a distinct target gas concentrations, and is representative for a set of different sensors. The classical measure for sensitivity $S$ is defined as the slope of the response $R$ versus the detected target gas concentration $c$

$$\begin{eqnarray}&&S=\displaystyle \frac{{\rm{\Delta }}R}{{\rm{\Delta }}c},\end{eqnarray} \tag{ 3 }$$

which is mandatory in order to calculate the $LOD$ of a gas sensor according to the definition

$$\begin{eqnarray}&&LOD=3\displaystyle \frac{RMS}{S},\end{eqnarray} \tag{ 4 }$$

with $RMS$ as the root mean square of the response $R$ [26].

3.1.1. PL characterisation and noise properties

The PL spectroscopy results (figure 2) for pristine ZnO NWs at low temperature ($T\approx 5\,{\rm{K}}$) show several distinct emission lines, which can be identified as neutral donor bound exciton transitions. The dominating contributions are the gallium-related transition D₀X I₈ at $3.3604\,{\rm{eV}}$ and the yet still not identified donor contribution I_8a at $3.3598\,{\rm{eV}}$ [27], followed by the indium-related transition D₀X I₉ at $3.3566\,{\rm{eV}}.$ These contributions, which were also observed in our previous works [28], are typical for our CVD grown ZnO NWs on silicon and are most likely caused by residual impurities in the ZnO source material.

Zoom In Zoom Out Reset image size

The inset in figure 2 shows two PL spectra recorded at room temperature with the near-band-edge emission (NBE) at ${\rm{3.21}}\,{\rm{eV}}$ for ZnO NWs [29] and the broad green band emission centred at $\sim 2.45\,{\rm{eV}}$ [30]. The green emission band was commonly assigned to the presence of zinc or oxygen vacancies [31]. Both room temperature spectra belong to the same set of ZnO NWs, but were recorded at two different positions on the sample. The varying intensity ratio of NBE and green luminescence is typical for nanowires and indicates, especially in samples with low impurity concentrations, an emission effect closely connected to the bulk-to-surface ratio of the NWs and hence a varying nanowire diameter [32].

3.1.2. SEM and EDX analysis

A first morphological study of the gold nanoparticle layer was performed by SEM imaging. Nominal thicknesses of $1\,{\rm{nm}},$ $3\,{\rm{nm}},$ $5\,{\rm{nm}},$ $7\,{\rm{nm}}$ and $10\,{\rm{nm}}$ of catalyst layer on the ZnO NW surface, as measured by a crystal quartz microbalance, were deposited and thereafter tested in the gas sensing setup. The crystal quartz microbalance for Au nanoparticle layer thicknesses meassurements was calibrated with x-ray reflectometry (XRR) data. The respective SEM images (inLens, 5 kV) of the sensor surface before and after sputtering are displayed in figure 3 for Au thicknesses of 1 nm, 3 nm, 5 nm and 7 nm and figures S3(a)–(b) for 10 nm. In the SEM images, the pristine ZnO NW surface appears to be very smooth in comparison to the insulating sensor substrate. After successful deposition of the catalyst at room temperature without additional annealing, an increased roughness of the ZnO NW surface appears to be visible. The corresponding image contrast becomes more pronounced with increasing Au layer thickness. We conclude that the surface functionalization with Au results in a discontinuous nanoparticle layer with Au islands and in-between gaps (indicating uncovered substrate regions). For Au catalyst layer thickness of 10 nm we still observe remaining gap regions. However, the I(V)-curve of a sensor with a surface functionalization of 10 nm clearly shows ohmic behaviour in contrast to a surface functionalisation of 7 nm and less which exhibits Schottky contact behaviour (figures S3(c)–(d)). Evaluating the resistance of sensors before R_NWs(i) and after R_NWs(Au) their functionalisation reveals an electrical resistance decrease by more than 4 orders of magnitude for the 10 nm Au sensors (table S1). Therefore, we conclude that the sensor with 10 nm Au layer thickness is covered by a Au layer showing percolation behaviour (i.e. a continuous current path through the metallic Au connecting both electrodes), which acts as an electrical shortcut between the electrodes. The typical H₂S sensing mechanism of ZnO NWs, as discussed in section 3.5, is based on conductivity changes in the metal oxide caused by ongoing adsorption and desorption processes on its surface. When a shortcut is present, the highly conductive metal catalyst layer dominates the electrical properties of the sensor. We do not expect a pronounced effect on the metal layer conductivity by H₂S adsorption/desorption at room temperature. Hence, samples for which the resistance measurement indicates the existence of such a shortcut behaviour have been excluded from further gas sensing measurements.

Zoom In Zoom Out Reset image size

Further analysis of the nanoparticle layer was performed by EDX, as displayed in figures 4 and S4–S9. Figures 4(a) and (d) show single nanowires randomly distributed on a Si substrate without electrodes before and after Au catalyst deposition of $5\,{\rm{nm}}$ thickness. The SEM images once again display a smooth ZnO NW surface before the sputtering, and an almost closed catalyst layer of Au islands separated by various gaps after the sputtering. The respective EDX spectra were obtained by focusing on a bigger aggregation of ZnO NWs as seen in figures 4(b) and (e). EDX results in table 1 confirm Si, O and Zn contributions for both spectra, and an additional Au contribution only after the sputtering process. No further elements were identified in both spectra. Figures S4–S9 display EDX data of ZnO NWs on sensors without any catalyst functionalization and with 1 nm, 3 nm, 5 nm, 7 nm and 10 nm thick Au functionalization, respectively. The elemental maps include Zn, O, Au and Si contributions. The Au contribution in the elemental maps is clearly visible for sensors with a 3–10 nm thickness. In these samples the largest Au signal is found at the ZnO NW locations which may be a result of Au accumulation at the NWs or enhanced scattering of electrons from the primary beam. The EDX results of each sensor (tables S2–S7) reveal an increasing Au contribution for sensors with increasing Au thickness. Sensors with no functionalisation and with only 1 nm of Au thickness show no clear Au contribution in their spectra and elemental maps.

Zoom In Zoom Out Reset image size

Table 1. Contribution of oxygen, silicon, zinc and gold in the EDX spectra before Au sputtering (displayed in figure 4(c)) and after sputtering (displayed in figure 4(f)). An additional Au contribution only appears after successful sputtering.

Element	Weight% (before Au)	Atomic% (before Au)	Weight% (after Au)	Atomic% (after Au)
O	6.81	12.51	4.79	10.37
Si	76.15	79.65	64.54	79.62
Zn	17.65	7.93	13.04	6.91
Au	−0.61	−0.09	17.64	3.10
Totals	100.00		100.00

3.1.3.  I(V) characteristic of a non-functionalized gas sensor based on pristine ZnO NWs

The sensors consist of two metallic Ti/Au contacts, which are connected by roughly ${N}_{{\rm{NWs}}}={10}^{4}$ pristine ZnO NWs, estimated from scanning electron micrographs. This metal–semiconductor–metal (M–S–M) structure typically shows a distinct non-linear and almost symmetric I(V)-characteristic, which is displayed in figure 5. The electron affinity of n-type ZnO is $\chi =4.5\,{\rm{eV}}$ [33], whereas the work function of gold is ${\phi }_{{\rm{Au}}}\,\gt \,5.2\,{\rm{eV}}$ [34, 35]. Hence, the M–S–M structure consists of two Schottky barriers with parallel shunt resistances ${R}_{{\rm{Sh}}},$ both antiparallel in series with a resistor ${R}_{{\rm{NWs}}},$ representing the ChemFET formed by the ZnO NWs. By applying a constant bias V to this structure, one Schottky contact operates in forward direction and the other in reverse direction. The bias V is divided into three contributions

$$\begin{eqnarray}&&V={V}_{{\rm{F}}}+{V}_{{\rm{NWs}}}+{V}_{{\rm{R}}},\end{eqnarray} \tag{ 5 }$$

with ${V}_{{\rm{F}}},$ ${V}_{{\rm{R}}}$ the voltage drop across the forward- and reverse-biased Schottky barrier, respectively, and ${V}_{{\rm{NWs}}}$ the voltage drop over the ZnO NWs. The equivalent current I flowing through the forward- and reverse-biased Schottky contact ${I}_{{\rm{F}}}$ and ${I}_{{\rm{R}}}$ as well as the ZnO NWs ${I}_{{\rm{NWs}}}$ can be described by

$$\begin{eqnarray}&&I={I}_{{\rm{F}}}={I}_{{\rm{NWs}}}={I}_{{\rm{R}}},\end{eqnarray} \tag{ 6 }$$

with

$$\begin{eqnarray}&&{I}_{{\rm{F}}}={I}_{0}\left[{e}^{\left(\displaystyle \frac{q{V}_{F}}{{k}_{B}T}\right)}-1\right]+\displaystyle \frac{{V}_{F}}{{R}_{{\rm{Sh}}}},\end{eqnarray} \tag{ 7 }$$

$$\begin{eqnarray}&&{I}_{{\rm{NWs}}}=\displaystyle \frac{{V}_{{\rm{NWs}}}}{{R}_{{\rm{NWs}}}},\end{eqnarray} \tag{ 8 }$$

$$\begin{eqnarray}&&{I}_{R}={I}_{0}\left[1-{e}^{\left(-\displaystyle \frac{q{V}_{R}}{{k}_{B}T}\right)}\right]+\displaystyle \frac{{V}_{R}}{{R}_{{\rm{Sh}}}}.\end{eqnarray} \tag{ 9 }$$

Here, we assume the ideal case in which both Schottky contacts have an identical Schottky barrier height ${\phi }_{{\rm{SB}}}={\phi }_{{\rm{Au}}}-\chi$ and contact area $A.$ According to the thermionic emission theory [36], the saturation current I₀ for both contacts is:

$$\begin{eqnarray}&&{I}_{0}=AA* {T}^{2}{e}^{\left(-\displaystyle \frac{q{\phi }_{{\rm{SB}}}}{{k}_{B}T}\right)},\end{eqnarray} \tag{ 10 }$$

where $A* \,=\,4\pi em* {k}_{B}^{2}/{h}^{3}$ is the Richardson constant for ZnO with the effective electron mass $m* \,=\,0.22\,{m}_{0}$ for ZnO.

Zoom In Zoom Out Reset image size

The constant resistance ${R}_{{\rm{NWs}}}$ can be estimated directly by linear regression of the measured I(V)-characteristic in the region $0.5\,-\,1.0\,{\rm{V}}$ in the form

$$\begin{eqnarray}&&{R}_{{\rm{NWs}}}={\left(\displaystyle \frac{dI}{dV}\right)}^{-1}.\end{eqnarray} \tag{ 11 }$$

On the other hand, the resistance of an ensemble of NWs connected in parallel can be estimated from:

$$\begin{eqnarray}&&{R}_{{\rm{NWs}}}=\displaystyle \frac{l}{A\cdot q\mu {N}_{D}\cdot {N}_{{\rm{NWs}}}}.\end{eqnarray} \tag{ 12 }$$

Here, $l=\left(15\pm 5\right)\cdot {10}^{-6}\,\mu {\rm{m}}$ is the average length of harvested NWs, and $A=\left(8\pm 3\right)\cdot {10}^{-15}\,{{\rm{m}}}^{2}$ is their average cross-sectional area. The electrical conductivity of the dispersed NWs $\sigma =q\mu {N}_{{\rm{D}}}$ is calculated by using the donor density ${N}_{{\rm{D}}}=1.7\cdot {10}^{15}\,{{\rm{cm}}}^{-3}$ [37] for ZnO grown in our setup as reported in our earlier works, and the mobility $\mu \approx 100\,{{\rm{cm}}}^{2}\,{{\rm{Vs}}}^{-1}$ for unintentionally doped ZnO at room temperature [38].

As a result, the directly measured resistance is ${R}_{{\rm{NWs}}}=\left(1.41\pm 0.03\right){\rm{M}}{\rm{\Omega }},$ while the calculated estimation suggests a resistance of ${R}_{{\rm{NWs}}}=\left(70\pm 50\right){\rm{k}}{\rm{\Omega }}.$ This difference by a factor of 20 suggests that only a small fraction of ZnO NWs is properly connected to the contact structure. In the presented case, only ${N}_{{\rm{NWs}}}=\left(500\pm 300\right)$ ZnO NWs out of the original estimation seem to contribute to the sensor signal.

The main focus of this work lies in the investigation of the effect of the catalytic surface functionalisation of pristine ZnO NW surface by gold on the detection mechanism of H₂S sensing with ZnO NWs. In order to separate the role of Au on H₂S sensing, the initial pristine ZnO NW sensors in question need comparable electronic and sensor characteristics, e.g. a similar quiescent current level, response $R,$ response rate $\dot{E},$ signal-to-noise ratio ($SNR$) and $LOD$.

Hence, in the first series of H₂S sensing measurements the comparability of sensors in the same batch was investigated. Figure 6(a) displays the I(V)-characteristics of two nominally identical gas sensors manufactured from ZnO NWs shown in figure 1. Both sensors were fabricated by dropping $30\,\mu {\rm{l}}$ of ZnO NWs freshly dispersed in $1000\,\mu {\rm{l}}$ isopropanol onto two pairs of $1.5\,{\rm{cm}}$ long Ti/Au electrodes separated by a $5\,\mu {\rm{m}}$ wide gap. The I(V)-characteristic curves are overall rather similar, and both sensors reach a current level of ${I}_{0}\,=\,0.4\,\mu {\rm{A}}$ at $1\,{\rm{V}}.$

The initial H₂S sensing behaviour of both sensors was then investigated in a dynamic flushing measurement. Here, the surrounding atmosphere was periodically switched in $10\,{\rm{\min }}$ intervals between pure oxygen or nitrogen, and $1\,{\rm{ppm}}$ of H₂S diluted in pure nitrogen respectively. As figure 7(b) shows, introducing $1\,{\rm{ppm}}$ of H₂S to both sensors increases the current level linearly by $\dot{E}=0.05\,\mu {\rm{A}}\,{{\rm{\min }}}^{-1}$ and $0.04\,\mu {\rm{A}}\,{{\rm{\min }}}^{-1}.$ Switching back from diluted H₂S to pure oxygen decreases the current mostly back to its initial level ${I}_{0}$ for both sensors.

Because the H₂S is diluted in nitrogen, a current increase between nitrogen and oxygen atmosphere could be expected to partially contribute to the detected signal. To investigate this contribution, during the last intervals both sensors, were flushed with nitrogen instead of oxygen. As discussed in our previous work [39], pure nitrogen does not reset the sensor to the initial current level, which results in an undesirable background current level drift, and makes nitrogen an unsuited candidate for flushing gas. Nonetheless, the inset in figure 7(b) shows that the response rate for 1 ppm of H₂S remains rather constant among several flushing intervals and hence is representative for the detection of 1 ppm of H₂S. This leads to the conclusion that possible differences in the response of pristine ZnO NWs to pure nitrogen versus pure oxygen can be neglected for the detection of high H₂S concentrations in the range of $1\,{\rm{ppm}}$ against oxygen. Initial testing and comparison of the sensors displayed in this work are done in H₂S in nitrogen versus oxygen measurements if not mentioned otherwise.

Both sensors start at a quiescent current level of ${I}_{0}=0.4\,\mu {\rm{A}}$ at $1\,{\rm{V}}$ in air and an average of ${I}_{0}=0.3\,\mu {\rm{A}}$ or $0.25\,\mu {\rm{A}}$ at $1\,{\rm{V}}$ in pure oxygen. The second sensor with the slightly lower current exhibits a lower response rate, which results in an almost identical response for both sensors of $R=(140\pm 20) \%$ and $(135\pm 30) \%$ for $1\,{\rm{ppm}}$ H₂S, or a $LOD$ of $(120\pm 30)\,{\rm{ppb}}$ and $(150\pm 30)\,{\rm{ppb}},$ respectively, as displayed in figure 6. We conclude that identical fabrication leads to sufficient comparability of different samples, what should allow for a straight-forward comparison and interpretation of the possible sensory effects assigned to different Au loadings on ZnO NWs.

Zoom In Zoom Out Reset image size

To investigate the catalytic effect of a discontinuous Au nanoparticle layer on the H₂S sensing performance of ZnO NWs, multiple series of modified gas sensors with different metal loading on the NW surface were fabricated and compared.

In a first experiment, the previously discussed highly reproducible pristine ZnO NW sensors were loaded with $5\,{\rm{nm}}$ and $7\,{\rm{nm}}$ of Au catalyst, respectively. After the successful surface functionalisation, a first very noticeable change for both sensors is the increased quiescent current, as displayed in the characteristic I(V)-curves in figure 7(a). Here, the $5\,{\rm{nm}}$ sample reached a current of ${I}_{0}=0.6\,\mu {\rm{A}}$ at $1\,{\rm{V}},$ and the $7\,{\rm{nm}}$ sample reached a current of ${I}_{0}=0.95\,\mu {\rm{A}}$ at $1\,{\rm{V}}$ compared to the initial value of ${I}_{0}=0.4\,\mu {\rm{A}}$ for pristine NWs. This first trend, an increase of the base current level along with an increase of metal particle loading, appears to contradict the effect of Au catalyst loading on metal oxide semiconductors as postulated in literature [40], where surface modification with Au particles is expected to lead to catalytic active oxygen dissociation [40]. When actually more oxygen is adsorbed on the ZnO NW surface, more electrons are trapped on the metal oxide surface, which in turn should result in a thicker non-conductive depletion layer underneath the surface and thus an increase in resistivity of the Au covered ZnO NWs as compared to pristine ZnO NWs. The observed decrease of resistivity can be attributed to two other effects. Before Au deposition, the NWs on our sensors are loosely distributed on top of our contact structure. An additional Au nanoparticle layer possibly improves the quality of the contact between the ZnO NWs and the Ti/Au contact pads. A second more likely explanation is attributed to the morphology of the surface modification itself, and the parallel conduction path via Au islands in a percolation-style situation. It was mentioned that Au nanoparticle films with a too high thickness can even lead to a shortcut between the Ti/Au contact pads. Here, ZnO NWs covered with a $10\,{\rm{nm}}$ thick Au layer show already ohmic conduction, which is in accordance to literature where sputtered Au structures show ohmic characteristics above $10\,\mathrm{nm}$ film thickness and non-ohmic characteristics below $10\,{\rm{nm}}$ [41]. With more Au deposition, the nanoparticles grow in size and undergo a transition from the isolated nucleation stage to lateral Au nano islands growth. The gaps in between the islands start to vanish, and the resistance of the catalyst layer rapidly decreases until the high ohmic current outweighs the catalytic effect of oxygen dissociation.

Zoom In Zoom Out Reset image size

The current increase of both modified sensors during exposure to $1\,{\rm{ppm}}$ H₂S at $1\,{\rm{V}}$ is shown in figure 8. Because unexpectedly the current limit of our experimental setup was already reached after 3 min, the measurement in the target gas interval had to be interrupted. The $5\,{\rm{nm}}$ and the $7\,{\rm{nm}}$ sensor started at a quiescent current of ${I}_{0}=0.32\,\mu {\rm{A}}$ and $0.58\,\mu {\rm{A}}$ in pure oxygen atmosphere, and reached a current of $37\,\mu {\rm{A}}$ and $100\,\mu {\rm{A}}$ after 3 min exposure to H₂S. During the first 1.5 min of the target gas interval, the previous flushing gas atmosphere is completely replaced. First, the quiescent current is kept, and then is followed by a linear current increase with approximately $\dot{E}=23\,\mu {\rm{A}}\,{{\rm{\min }}}^{-1}$ and $110\,\mu {A{\rm{}}\min }^{-1}$ for sensor 1 and sensors 2, respectively. Seemingly, the detection signal never saturates during the $10\,{\rm{\min }}$ long exposure to $1\,{\rm{ppm}}$ H₂S at $1\,{\rm{V}},$ which leads to an approximate response of $R=60\,000 \%$ and $150\,000 \%$ and a $LOD$ of $4\,{\rm{ppb}}$ and $2\,{\rm{ppb}}$ of H₂S in nitrogen for sensors 1 and 2, respectively. These values only represent a rough estimate and hence no error margins are quoted. Apparently, additional surface functionalisation of pristine ZnO NWs on the on hand leads to a quiescent current increase due to the additional conduction path via the Au nanoparticle layer, and on the other hand the ZnO NWs now show a response rate significantly enhanced by three orders of magnitude as compared to before functionalisation. The $RMS$ noise for both sensors appears to be in the same order of magnitude and doesn't show clear tendencies related to the different Au loading. However, due to the enhanced response, the resulting $SNR$ is largely increased, and the $LOD$ drops by two orders of magnitude. These effects appear to be more pronounced for samples with thicker catalyst loading.

Zoom In Zoom Out Reset image size

In order to validate the measurement results of this first catalytic surface modification, a second series of H₂S sensors was investigated with $1\,{\rm{nm}},$ $3\,{\rm{nm}}$ and $5\,{\rm{nm}}$ thick Au catalyst loadings. Because the $1\,{\rm{nm}}$ thickness is hard to control and shows higher variance due to the short sputtering time, two sensors with $1\,{\rm{nm}}$ loading were manufactured. The characteristic I(V)-curves of the sensors before loading are presented in figures 9(a)–(b) alongside with the ratio between the quiescent current ${I}_{i}$ before and ${I}_{{\rm{Au}}}$ after surface functionalisation in air for an applied voltage of $0.1\,{\rm{V}}.$ For this series, a lower forward voltage was chosen in order to prevent the sensing signal from exceeding the current limit of our setup. For this series similar trends were found, i.e. the quiescent current level increases after successful Au nanoparticle film functionalisation of any thickness (figure 10(a)). In addition, the response is increased for thicker loading. In this measurement series the sensor with $5\,{\rm{nm}}$ loading shows the largest enhancement of the signal when subject to $1\,{\rm{ppm}}$ of H₂S at $0.1\,{\rm{V}}$ for the first time. However, the catalytic enhancement of the response reduces with every subsequent interval of exposure to H₂S, which was especially observed for the $5\,{\rm{nm}}$ coverage, and less drastically for $3\,{\rm{nm}}$ or 1 nm. This phenomenon strongly suggests a possible contamination of the Au nanoparticle layer by reaction with sulphur from the interaction with H₂S molecules (see section 3.5).

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Because the noise value for all sensors appears to be identical after Au functionalisation, the $LOD$ is improved through the enhanced response, as shown in figure 10(b). The response for the first measurement interval with $1\,{\rm{ppm}}$ H₂S was determined as $R=(350\pm 50) \% ,$ $(480\pm 50) \% ,$ $(620\pm 50) \%$ and $(1030\pm \,50) \%$ and the $LOD$ as $(69\pm 10)\,{\rm{ppb}},$ $(66\pm 8)\,{\rm{ppb}},$ $(35\pm 4)\,{\rm{ppb}}$ and $(18\pm 2)\,{\rm{ppb}}$ for $1\,{\rm{nm}},$ $3\,{\rm{nm}}$ and $5\,{\rm{nm}}$ Au loading, respectively (table 2). The results support the previously observed trend that a higher catalyst loading improves the H₂S sensing ability of ZnO. Here, it appears to be initially possible to improve the H₂S sensitivity of ZnO NW strong enough to detect concentrations even in the low ppb region.

Table 2. Direct display of response $R$ and resulting $LOD$ for sensors with $1\,{\rm{nm}},$ $3\,{\rm{nm}}$ and $5\,{\rm{nm}}$ Au loading. A higher catalyst loading improves the H₂S sensing ability of ZnO NWs.

Sensor	Response ${\boldsymbol{R}}( \% )$	${\boldsymbol{L}}{\boldsymbol{O}}{\boldsymbol{D}}(\text{ppb})$
1 (5 nm Au)	1030 ± 50	18 ± 2
2 (3 nm Au)	620 ± 50	35 ± 4
3 (1 nm Au)	480 ± 50	66 ± 6
4 (1 nm Au)	350 ± 50	69 ± 10

The rapid contamination process of the catalyst for $1\mbox{--}5\,{\rm{nm}}$ Au nanoparticle layer thickness was confirmed by tracking the response and response rate towards $1\,{\rm{ppm}}$ H₂S for all sensors over the duration of 3 weeks, as displayed in figure 11. At first the superior initial sensing ability of the sensor with $5\,{\rm{nm}}$ Au loading clearly stands out. However, with increasing number of detection cycles, all four sensors undergo a decrease in response and response rate until after 3 weeks it is impossible to distinguish the sensors solely by their sensing ability towards H₂S. After 3 weeks, the sensor response becomes stable and is finally comparable to the initial sensing abilities of pristine ZnO NWs towards gaseous H₂S.

Zoom In Zoom Out Reset image size

In addition, comparison of the gas sensing performance of pristine ZnO NWs to H₂S with the gas sensing performance to other gases, i.e. methane (CH₄) and hydrogen (H₂), as shown in figure S10, reveals a clear affinity towards H₂S. While the response for 1 ppm H₂S is $R=(25\pm 7){\rm{ \% }}$, the response for high concentrations of 500–10000 ppb of CH₄ and H₂ stays below 10%. After functionalisation of ZnO NWs with Au the sensor signal shows a response of $R=(320\pm 20) \% ,$ for 1 ppm H₂S, whereas no enhanced response is measured for the detection of CH₄ and H₂ with Au functionalized ZnO NWs. Here, the response towards both reducing gases stays still below 10%.

Summarizing both measurement series, we conclude that ZnO NWs with a thick Au loading led to an overall improved H₂S sensing performance, i.e. an enhanced response and a decreased $LOD$. The positive enhancements due to the deposition of the Au catalyst increase with increasing thickness, but for 10 nm Au layer thickness electrical short-cut behaviour is found. For the samples investigated in the present study, the best results are found for 7 nm Au layer thickness on ZnO NWs.

So far, we have compared the detection of H₂S with Au modified ZnO NWs to the detection with pristine ZnO NWs. However, H₂S was always only diluted in nitrogen. A more realistic scenario, which is closer to medical breath analysis with focus on detection of target gas concentrations in the low ppb range, is a measurement series for H₂S diluted in air.

Hence, one sensor with pristine ZnO NWs and for comparison another sensor with surface modified ZnO NWs were prepared for H₂S sensing in synthetic air. The ZnO NWs, which were selected for these sensors, were harvested from samples with very homogeneous and particularly thin ($\sim 60\,{\rm{nm}}$ diameter) nanowires. Moreover, the resulting sensors showed very low noise and hence appeared to be very promising for the detection of extremely low H₂S concentrations. In order to further enhance the detection signal, the applied voltage was raised to $1\,{\rm{V}},$ and the gas switching intervals were extended from $10\,{\rm{\min }}$ to $20\,{\rm{\min }}.$ The detection of every concentration was repeated twice before a higher H₂S concentration was selected. To make sure that the sensors were exposed towards different H₂S concentrations with comparable starting conditions, a $30\,{\rm{\min }}$ long flushing interval with pure oxygen was included in between each concentration step.

As shown in figure 12(a), a clearly visible response of $R=26 \%$ by pristine ZnO NWs was observed for $63\,{\rm{ppb}}$ of H₂S in synthetic air. The response increased linearly with increasing H₂S concentrations and led to a theoretically estimated $LOD$ of $(30\pm 5)\,{\rm{ppb}}.$ The $LOD$ results of nominally identical ZnO NWs with additional $5\,{\rm{nm}}$ Au surface modification are displayed in figure 12(b). For these, an experimentally visible response of $R=80 \%$ was measured for $10\,{\rm{ppb}}.$ With a $SNR$ of 67 this leads to a theoretically calculated $LOD$ of approximately $\left(500\pm 200\right)\,{\rm{ppt}}.$ In comparison to recently reported works on H₂S gas sensors (table 3), Au functionalized ZnO NWs achieve high response and extremely low $LOD$ at low working temperature and hence low energy consumption. This outstanding result highlights the tremendous potential of surface modification with metal nanoparticles on metal oxide semiconductors. Here, the surface modification with Au enables the detection of H₂S in the ppt—ppb range even in the presence of $20 \%$ highly oxidative oxygen. The selective catalytic effect of Au towards H₂S has been attributed to the strong chemical affinity between gold and sulphur (see section 3.5) [42]. This measurement also demonstrates the deterioration of the catalytic effect of the Au nanoparticle layer functionalisation. Whilst the first exposure to H₂S yields a response of $R=80 \% ,$ the response during the second interval for H₂S diluted in air already decreases to $R=40 \% .$ This observation underlines the need for a better understanding of stability and degradation mechanisms in the Au functionalized sensors.

Zoom In Zoom Out Reset image size

The gas sensing mechanism of ZnO is based on the continuous adsorption and desorption of molecules on the metal oxide surface. ZnO is inherently a n-type metal oxide semiconductor. Deep acceptor-like surface states on the ZnO NWs lead to an accumulation of negative charges and finally to the formation of a permanent thin surface depletion layer with low carrier density, while the nanowire core remains n-type conducting.

When oxygen is interacting with the ZnO NW surface, it will ionosorb in molecular (O₂ ⁻) and atomic (O⁻, O²⁻) form on unoccupied chemisorption sites by trapping electrons from the conducting core of the ZnO NW to its surface [50] (figure 13(a)). This charge transfer leads to an accumulation of negative charges on the surface, and thus an upward band bending at the ZnO NW surface and hence an increase of the thickness of the surface depletion region relative to the conducting nanowire core. The resistance of the NWs increases, and the measured current at constant voltage decreases accordingly. For temperatures below $150\,^\circ {\rm{C}}$ the molecular form of oxygen dominates, and the chemisorption of oxygen can be described as [50]:

$$\begin{eqnarray}&&{{\rm{O}}}_{2}\to {{\rm{O}}}_{2({\rm{ads}})},\end{eqnarray} \tag{ 13 }$$

$$\begin{eqnarray}&&{{\rm{O}}}_{2}+{{\rm{e}}}^{-}\to {{{\rm{O}}}^{-}}_{2({\rm{ads}})}.\end{eqnarray} \tag{ 14 }$$

Zoom In Zoom Out Reset image size

When the reducing gas H₂S is introduced to the electron depleted ZnO NW surface, the previously adsorbed oxygen species will interact with the target gas, which leads to the desorption of oxygen by charge transfer, as described in [51, 52]:

$$\begin{eqnarray}&&2{{\rm{H}}}_{2}{{\rm{S}}}_{({\rm{g}})}+3{{{\rm{O}}}^{-}}_{2({\rm{ads}})}\to 2{{\rm{SO}}}_{2({\rm{g}})}+2{{\rm{H}}}_{2}{{\rm{O}}}_{({\rm{g}})}+3{{\rm{e}}}^{-}.\end{eqnarray} \tag{ 15 }$$

By finally forming H₂O and SO₂, the chemisorbed oxygen is removed and trapped electrons are released back into the conducting NW core. As a result, the surface band bending is relaxed and the width of the depletion region decreases. This is the sensing mechanism observed for pristine ZnO surfaces.

Once Au is dispersed on the ZnO NW surface, the catalytic activity of Au will strongly influence the sensing mechanism. Because the work function of Au [35] is larger than the work function of ZnO [53], electrons will transfer from the metal oxide to the noble metal [42]. These nano-Schottky barriers lead to the formation of an additional depletion region in the contact area between ZnO and Au (figure 13(b)). An additional nanoscopic initial depletion region just underneath the Au islands, which will be modulated by oxygen and H₂S adsorption, leads to an enhanced resistance change within the nanowires in comparison to the pristine ZnO NW case. This enhanced modulation depth can easily be observed via the increased response.

Another very beneficial effect of Au modification of the ZnO NW surface is the catalytically activated dissociation of gas molecules. Due to the strong catalytic affinity between Au and sulphur [54], Au acts as a highly active adsorption site for gaseous H₂S. H₂S interacts rather selectively with the Au nanoparticles and transfers to the ZnO NW surface (figure 13(c)). This spill-over effect results in a quantitatively increased interaction of H₂S with adsorbed molecular oxygen, as well as the increased rate at which the release of previously trapped electrons into the conduction band of ZnO occurs.

A similar effect is expected for the catalytic dissociation of oxygen [40], although this is difficult to verify in our measurements due to the enhanced quiescent current in the Au/ZnO structures and the preferential interaction of H₂S with Au/ZnO instead of oxygen with Au/ZnO for H₂S diluted in synthetic air. As was mentioned before, this is contradicting an enhanced oxygen dissociation.

While both effects, the additional depletion region and the catalytic adsorption of H₂S, tremendously enhance the sensing ability of Au/ZnO NWs towards H₂S, the affinity between Au and sulphur most likely leads to an irreversible contamination of the catalyst at room temperature. Considering the strong affinity of both elements, a reactive interaction is rather likely [42, 55, 56]:

$$\begin{eqnarray}&&{{\rm{H}}}_{2}{{\rm{S}}}_{({\rm{g}})}+{{\rm{Au}}}_{({\rm{s}})}\to {{\rm{AuS}}}_{({\rm{s}})}+{{\rm{H}}}_{2({\rm{g}})}.\end{eqnarray} \tag{ 16 }$$

H₂S adsorbed on Au reacts to Au-SH and Au-S type species, which results in the Au nanoparticle layer being covered by a sulphide shell over time [42]. With ongoing exposure to H₂S, the Au particles lose their catalytic adsorption activity and the spill-over effect vanishes. In the present work, this behaviour was observed for all Au/ZnO samples. As an outlook, we want to perform a detailed investigation of the surface composition of Au functionalized ZnO NWs before and after H₂S detection with additional x-ray photoelectron spectroscopy (XPS) data.