Layered Heterostructure of Graphene and TiO2 as a Highly Sensitive and Stable Photoassisted NO2 Sensor

As an atomically thin electric conductor with a low density of highly mobile charge carriers, graphene is a suitable transducer for molecular adsorption. In this study, we demonstrate that the adsorption properties can be significantly enhanced with a laser-deposited TiO2 nanolayer on top of single-layer CVD graphene, whereas the effective charge transfer between the TiO2-adsorbed gas molecules and graphene is retained through the interface. The formation of such a heterostructure with optimally a monolayer thick oxide combined with ultraviolet irradiation (wavelength 365 nm, intensity <1 mW/mm2) dramatically enhances the gas-sensing properties. It provides an outstanding sensitivity for detecting NO2 in the range of a few ppb to a few hundred ppb-s in air, with response times below 30 s at room temperature. The effect of visible light (436 and 546 nm) was much weaker, indicating that the excitations due to light absorption in TiO2 play an essential role, while the characteristics of gas responses imply the involvement of both photoinduced adsorption and desorption. The sensing mechanism was confirmed by theoretical simulations on a NO2@Ti8O16C50 complex under periodic boundary conditions. The proposed sensor structure has significant additional merits, such as relative insensitivity to other polluting gases (CO, SO2, NH3) and air humidity, as well as long-term stability (>2 years) in ambient air. The results pave the way for an emerging class of gas sensor structures based on stacked 2D materials incorporating highly charge-sensitive transducer and selective receptor layers.


■ INTRODUCTION
Nitrogen dioxide NO 2 can be singled out as one of the principal air pollutants that is extremely harmful to humans and the environment. 1,2The level of this toxic gas in the air is regulated, with the EU standard limiting the NO 2 annual mean value to 40 μg/m 33 and the US standard to 53 ppb (100 μg/m 3 at normal conditions). 4The limit recommended by WHO is even stricter, being recently reduced to 10 μg/m 3 . 5−8 Concentrations ranging from a few μg/m 3 in the natural background up to several hundred μg/m 3 at coal-burning industrial sources or traffic hotspots have to be measured. 6,9he device sensitivity must cover this concentration range, and a low susceptibility to interference from moisture and other pollutant gases and high stability are required.−13 Graphene is a unique two-dimensional conducting material with an ultimate area-to-volume ratio.Its electrical conductivity is highly responsive to environmental perturbations, because the density of its highly mobile charge carriers is low.Therefore, graphene is considered a very promising candidate for gas sensing applications. 11,14Extremely high sensitivity to different gases (down to single molecule detection) has been achieved using graphene in laboratory tests. 15,16In addition to the outstanding gas sensitivity of graphene and related twodimesional (2D) materials, their relative ease of integration into CMOS devices is extremely important. 17,18echnological developments in semiconductor pilot lines are promising; however, in parallel, it is also necessary to open the sensing potential of 2D materials for applications.The demonstrations of the ultimate sensitivity of graphene were made in a vacuum or inert gas, 15,16 but a critical criterion that a gas sensor material should fulfill is its ability to function in a natural air environment. 19In contrast to perfect transducer properties, pristine graphene is a poor, nonspecific receptor for small gas molecules.The electrical conductivity of graphene becomes sensitive to NO 2 gas only in the presence of defects or impurities in the carbon lattice or contaminants on its surface. 15,20,21This is not unexpected because the binding energy of NO 2 on the graphene surface is only 50−70 meV, 22 while it can be several eV at the defect sites. 23The defects could be due to fabrication conditions, 21 ozone treatment, 24 grain boundaries, 25 patterning, 26 or impurity doping. 27During the past decade, several research groups have confirmed that introducing defects into the graphene lattice improves sensor performance. 14−33 We have demonstrated that pulsed laser deposition (PLD) of silver nanoparticles or zirconia nanolayer on graphene significantly increases its sensitivity to NO 2 gas. 30PLD is a universal method for deposition of various materials, such as metals, metal oxides, and other compounds, with exceptionally high precision, equal to only 1/100 of a monolayer per laser pulse. 34This allows us to explore an approach in which adsorption takes place on the nanolayer of another material on top of graphene but charge interchange with graphene still occurs through the interface.In this case, the sensor material is not defective graphene but a heterostructure, where graphene is an electronic transducer, and the added layer provides the receptor sites for gas molecules.
A substantial enhancement in the sensor behavior can be predicted for the TiO 2 overlayer, because of the reduced bandgap and the variable valence states of Ti (2+,3+,4+) as compared to heavier 4d-element Zr. 35 The bandgap values of the TiO 2 and ZrO 2 films are estimated to be 3.2 eV 36 and 5.0 eV, 35 respectively.TiO 2 has a conduction band derived from the empty Ti 4+ 3d orbitals, whereas the filled valence band mainly corresponds to the 2p levels of O 2− .It is possible to optically excite TiO 2 at ultraviolet (UV) wavelengths <390 nm, whereby the photogenerated charge carriers may invoke different surface processes (leading, e.g., to the formation of surface oxygen vacancies and Ti 3+ ions with a lower oxidation state), and alter the chemical reactivity of the oxide surface with respect to adsorption. 36,37The sensitizing effect of UV light has been demonstrated in various gas-sensitive materials. 12,38UV-assisted sensing of NO 2 has been investigated with sputtered TiO 2 thin films, and a dramatic improvement in the response to 500 ppm of NO 2 has been observed. 39We have previously tested this approach with other transition metal oxides (V 2 O 5 and CuMn 2 O 4 ), preferentially rendering graphene sensitive to the reducing gases NH 3 and H 2 S. 40,41 With TiO 2 placed on graphene, the photoinduced effects could be more prominent and more easily achievable owing to the well-known photosensitizer properties of this oxide. 36,37From a practical point of view, on the one hand, the light source increases the complexity and cost of the sensor.On the other hand, without additional energy provided by a microheater or light source, the room temperature sensors 10 have frequently very slow signal recovery and may not function stably enough in outdoor conditions.
In this study, we investigated the influence of PLDfabricated TiO 2 nanolayers on the electronic and gas sensing properties of chemical vapor deposited (CVD) graphene, the sensing mechanisms in the graphene-TiO 2 (Gr/TiO 2 ) heterostructure, and optimized its fabrication and operating conditions for the best performance in NO 2 environmental monitoring.
■ RESULTS AND DISCUSSION Characterization of Materials.The sensors were built on 10 × 10 mm 2 SiO 2 (300 nm)/Si substrates.A sheet of CVD graphene was transferred to cover the slit between the two prefabricated gold electrodes and a TiO 2 layer was deposited on it using pulsed laser deposition.The details of the sample fabrication, gas sensing setup, and characterization methods are described in the Materials and Methods section.
Figure 1(a) shows the Raman spectra of pristine and TiO 2coated graphene (Gr/TiO 2 ) samples obtained after the deposition of different amounts of the target material.The sample used in this series was modified as follows: after each deposition, it was removed from the PLD chamber, characterized, and loaded again into the PLD chamber for the next deposition.The cumulative number of laser pulses N used for TiO 2 deposition is shown at each spectrum.Before PLD, pristine graphene had a minimal number of defects, indicated by missing D-peak at ≈1350 cm −1 .The ratio of G and 2D band intensities (ca.1:3), combined with band peak positions (1590 and 2690 cm −1 , accordingly), 42 prove that graphene was single-layered.
After PLD, the 2D band intensity decreased, with the emergence of defect-related bands D and D′.The ratio of the D-to G-band intensities was equal to 1.25 after the first deposition and increased to approximately 1.75 at N = 135 or higher.All Raman bandwidths increased by ∼50% up to the TiO 2 layer thickness obtained with N = 135 laser pulses and changed only slightly (5−15%) thereafter.The lower frequency part of the spectrum contained only the Raman bands of the Si substrate (see Figure S1 of the Supporting Information), the characteristic features of TiO 2 crystalline phases were not visible, which indicates the amorphous structure of the material.
Figure 1(b) presents the transconductance curves of the graphene sensor before and after the deposition of the TiO 2 layer.To reduce the Fermi level shift induced by oxygen and water molecules adsorbed from ambient air, 43 measurements were conducted under dry nitrogen flow after annealing for 2 h at 150 °C.For pristine graphene, the characteristic V-shaped curve shows the Dirac point at the gate voltage U g = 8 V and the field effect hole mobility of 2035 cm 2 /(V s) at U g = 0 V.After deposition of ∼0.6 nm of TiO 2 , the mobility was drastically reduced to 29 cm 2 /(V s), whereas the Dirac point shifted out of the measurement scale, due to doping.For the different devices tested, the mobility was in the range of 1000− 4000 cm 2 /(V s) before and 25−30 cm 2 /(V s) after the TiO 2 deposition.
Scanning electron microscopy (SEM) images of pristine graphene and Gr/TiO 2 (sample with N = 675) are displayed in Figures 1(c) and 1(d), respectively.The structure of pristine graphene is primarily defined by the topography of the Cu foil used during the growth phase, 44 which explains the darker lines in both subpanels.The size of the crystallites in large-area polycrystalline graphene was between 1 and 10 μm (see also additional SEM images in Figure S2 of the Supporting Information).As shown in Figure 1(d), the deposited material uniformly covered the graphene, producing a granular structure on top of the graphene.The uniform coverage by C and Ti was confirmed by mapping with energy dispersive Xray (EDX) spectroscopy (see Figure S3 of the Supporting Information).
The X-ray photoelectron spectroscopy (XPS) spectrum of a Gr/TiO 2 sample is presented in Figure 2(a), where the lines corresponding to Ti, O, C, Si (substrate), and Au (contacts) can be observed.Generally, the XPS spectrum of TiO 2 gives symmetric peak shapes for Ti 2p. Figure 2(b) confirms that the main form of Ti films is Ti 4+ , and the obtained spin−orbit splitting of the Ti 2p peak equals the value characteristic of TiO 2 at 5.7 eV. 45Consequently, despite using the TiN target material in the PLD process, the material on top of graphene in the tested devices was TiO 2 .Presumably, oxidation occurred during exposure to air, which was facilitated by the thinness of the layer.A similar complete oxidation has been observed for a 3 nm Ti metal layer deposited on graphene. 46his was confirmed by the X-ray fluorescence (XRF) measurement, which resulted in mass areal densities of 0.74 mg/cm 2 for Ti and 0.00 mg/cm 2 for nitrogen for the N = 675 sample.Assuming a density of 4.0 ± 0.2 g/cm 3 for TiO 2 (average between rutile and anatase 36 ), the thickness of the TiO 2 layer was 3.1 nm.The ellipsometry data also agreed with the overlayer being TiO 2 and not TiN, whereas the thickness estimate was 6.2 ± 0.4 nm at 51% porosity.These data allowed us to calibrate the growth rate at the given process parameters to 4.6 pm per laser pulse.The C 1s peak of our lab-made graphene was recently carefully analyzed, whereas the deconvolution resolved mainly sp2 carbon (87%), with some sp3 carbon (7.5%), and oxidized (C−O, C�O, O�C−O) species (5−6%). 47as Responses.Figure 3 represents the core results of the work: it shows the drastic improvement in the gas responses of TiO 2 -coated graphene, as compared to pristine graphene, in the dark and under UV illumination.The experiments were performed in synthetic air with 20% (see Figure 3(a)) or 0% (see Figure S4 of the Supporting Information) relative humidity (RH); the UV light had an intensity of 0.2 mW/ mm 2 at 365 nm.
As shown in the inset of Figure 3(a), pristine graphene is completely unresponsive to a relatively high NO 2 concentration (4 ppm) but acquires some sensitivity under UV illumination.An increase in the sensor's signal (i.e., electrical current at 0.1 V supply voltage) under the influence of oxidizing gas agrees with the transconductance measurements that showed the material to be p-type, that is, hole-conducting.Graphene has been found to be p-doped by adsorbed water and oxygen molecules in air; 48,49 our samples retained this type of conductivity to some extent even after heating in a nitrogen environment (see Figure 1(b)).Because NO 2 acts as an electron acceptor, hole doping is promoted further, leading to increased conductivity.
The gas responses maintain their sign but are significantly amplified when the TiO 2 overlayer coats the graphene.First, as shown in Figure 3(a), the Gr/TiO 2 structure is already responsive in dark conditions.In this experiment, within the time interval of 62 to 102 min, four sequential exposures to NO 2 gas at concentrations between 40 and 1000 ppb were made for 5 min each, separated by 5 min intervals of pure synthetic air.At 144 min, UV illumination was initiated, causing a significant drop in the signal.Previously, a similar  effect has been observed on pristine graphene and explained by the photoinduced removal of oxygen and the associated decrease in the density of electron holes in graphene. 49,50oping by oxygen has been found to depend on RH, which is ascribed to the stabilizing effect of water on oxygen anions on SiO 2 substrates. 49This interpretation is consistent with the similarity of the photoinduced resistivity in pristine CVD graphene 50 and Gr/TiO 2 heterostructure (see Figure 3(a)).
After stabilizing the signal, a similar series of NO 2 gas exposures were performed (see data from 200 to 250 min in Figure 3(a)).Significantly stronger and faster responses were observed with light stimulation: the relative signal change in the presence of 40 ppb test gas was approximately 10% in the dark and 40% under UV light.The repeatability of the signals under UV illumination was very good, and the responses did not differ significantly when measured in dry air (see Figure S4).The baseline returned to its former level in the dark within 2 days.
Figure 3(b) shows the effect of subsequent laser deposition (thickness of the TiO 2 layer) on the gas response and conductivity of the Gr/TiO 2 structure.Relative responses to NO 2 were improved with every new PLD sequence until the titania thickness reached 0.6 nm, dropping thereafter with every new deposition.This thickness value is close to that of single-layer TiO 2 (d = 0.7 nm). 51At the same time, a continuous decrease in conductance occurs, according to the power law G ∼ d 0.27 (see Figure 3(c)).This decrease was due to the increased number of defects created in the graphene lattice during the PLD process and the interactions of graphene with the deposited material.The dependence was strongly sublinear because the already deposited material shielded graphene from the bombardment by the PLD plasma plume, and the interactions with the deposited material decreased rapidly as the distance from the graphene increased.
DFT Simulations.To model the graphene−TiO 2 interaction and NO 2 gas adsorption on TiO 2 -coated graphene, we performed quasi-two-dimensional density functional theory (DFT) calculations using periodic boundary conditions.The VASP program package 52,53 was used together with the potential projector augmented-wave (PAW) method 54,55 and PBEsol functional approach 56 with DFT-D3 corrections. 57The Bader method 58 was employed to analyze the charge distribution in the system under consideration.First, we determined the ground states of the Ti 8 O 16 C 50 cluster, consisting of the graphene sheet of 50 atoms and a TiO 2 layer in a box, and an isolated NO 2 molecule in the same box.A plane-wave basis set with a 500 eV cutoff and a Γ-pointcentered mesh for k-point sampling were chosen in our calculations.For atom relaxation, k-grid sampling was taken as 4 × 4 × 1.The details of the electronic structure were investigated using a denser 25 × 25 × 1 k-point grid.All titanium, nitrogen, and oxygen atoms were fully relaxed during geometry optimization, whereas the carbon atoms were fixed in the c-direction.Next, the ground state of the complex NO 2 @ Ti 8 O 16 C 50 was determined.The final relaxed structure is shown in the inset of Figure 4 (the lattice parameters were 12.2, 12.2, and 34.3 Å).A chemical bond had formed between the oxygen belonging to the adsorbate and one of the titanium atoms.The obtained energy of the NO 2 adsorption on Ti 8 O 16 C 50 cluster was 2.65 eV.This value is much higher than the corresponding value of 0.055−0.067eV in the case of pristine graphene 22 but similar to that found for NO 2 adsorbed on graphene defects/impurities. 23The Ti 8 O 16 C 50 formation and adsorption of NO 2 on its top led to the redistribution of the electron charge between different subsystems of the complex.Using Bader's approach, we evaluated the changes in the Coulomb charge over the course of the reactions: (1) (2) It was found that the electron charge transferred from the Ti and C subsystems to the adsorbate and oxygen subsystems.The values of the component charges are listed in Table 1.The charge transfer from the carbon subsystem occurred during both the formation of the complex Ti 8 O 16 C 50 and NO 2 adsorption.This resulted in an increase in the partial density of the electronic states (PDOS) of the carbon subsystem at the Fermi level during the adsorption process.The PDOS of all atomic subsystems of the considered complexes is shown in Figure 4.
As can be seen in Figure 4(a), there is no Ti−C or O−C hybridization in the system under consideration: the shape of the graphene partial density of states does not change its dependency on the energy in the presence of TiO 2 overlayer.Thus, owing to the small charge transfer between the graphene and TiO 2 subsystems (0.24e, where e is the elementary charge; see the decrease in the C 50 valence electron content in   1) and the absence of hybridization, this structure is mainly stabilized by a van der Waals interaction, which is considered in our calculations by the DFT-D3 correction. 57The band diagram is given in Figure S5 of the Supporting Information.
The energy scales of the two subpanels in Figure 4 are shifted with respect to each other so that the minima of the carbon subsystem PDOS (Dirac points) coincide.It is easy to see that adding a TiO 2 cluster shifts the Fermi level toward lower energy by about 0.4 eV and the adsorption of NO 2 by another 0.4 eV.This explains the Fermi level shift in the transconductance curves (see Figure 1(b)).The accompanying increase in the carbon PDOS at the Fermi level also explains the conductivity increase during the experimental adsorption process.The charge transferred from graphene to the adsorbed NO 2 molecule was approximately 0.5e (see Table 1).
Discussion of Molecular Mechanisms.First, we considered the gas sensor response as a function of the NO 2 concentration.To obtain the main parameters−response amplitudes and rates−from the data shown in Figure 3(a) and its analogs (e.g., Figure S4), each response and recovery curve was approximated with an exponential function.The dependence of the relative response amplitudes (defined as the conductance change with respect to the initial conductance) is shown in Figure 5(a).It follows that UV light improved the sensitivity by a factor of 4, and the concentration dependence resembles in both cases Langmuir isotherms, 59 expressed as a dependence of surface coverage θ on the gas concentration c: In eq 3, the quantity c 0 can be considered as the inverted adsorption−desorption equilibrium constant and, in practical terms, the concentration at which the response amplitude is half of its maximum.
To understand the key factors determining the sensitivity enhancements of graphene sensors by the TiO 2 overlayer and UV illumination, we considered the basics of sensor signal formation.In the case of hole-dominated conductivity in graphene, the conductivity σ can be described by the following equation: = en where n and μ are the hole density and mobility, respectively.Under the influence of NO 2 , conductivity is mainly affected by the charge transfer between the adsorbed NO 2 molecules and graphene. 11As can be seen in Figure 3(a), the process is mainly reversible, but a fraction of the signal does not recover in the dark or has a prolonged recovery.The latter phenomenon is commonly observed in chemiresistive materials and can be explained by slow desorption from adsorption sites with high binding energies.The reversible chemisorption reaction via electron transfer is described by the following equation: (5 where A is the vacant adsorption site on the surface (e.g., Ti 3+ in titania). 60Consequently, the conductivity is increased by an approximate amount (neglecting the effect of mobility 61 ): where N A is the density of adsorption sites and η is the fraction of electron charge extracted from graphene at the absorption of a single NO 2 molecule.The quantity we would like to compare (and maximize for practical applications) is the relative amplitude of the gas response, which can now be expressed by the equation: Let us define sensitivity S as the derivative of the relative response to gas concentration c.The sensitivity at c 0 (gas concentration in the middle of the detection range) can be calculated by the following equation: This quantity can be enhanced by increasing the density of adsorption centers N A or decreasing the density of charge carriers n in the initial state.Generally, graphene can have high gas sensitivity owing to the low value of n when the Fermi level is close to the Dirac point.
However, the sensitivity is virtually absent in the case of pristine graphene (see the inset of Figure 3(a)), and in the Gr/ TiO 2 structure, the Fermi level is significantly shifted toward a lower energy below the Dirac point, which increases n (i.e., PDOS, contributing to conductivity).Consequently, the significantly higher sensitivity of the Gr/TiO 2 structure can only be explained by the increased N A due to the new adsorption centers created by the deposition of TiO 2 .The Gr/ TiO 2 heterostructure has many adsorption centers located at Ti cations, where efficient charge transfer from graphene occurs during adsorption, as confirmed by DFT simulation.Therefore, the TiO 2 overlayer acts as a receptor, and graphene,  1 (the rates are defined as 1/t in and 1/t out , respectively).
with its conductivity generating the sensor signal, acts as a transducer.
Of course, the largest possible degree of electron transfer occurring during the elementary adsorption process is also necessary; that is, the value of η must be close to unity.The simulation using a small TiO 2 cluster yielded a reasonably high value η = 0.5 (see Table 1).It is obvious that this factor decreases when the adsorption center is located further away from the graphene; indeed, the gas response started to decrease when the thickness of the TiO 2 layer exceeded that of a monolayer (see Figure 3(b)).
The final parameter in eq 8, c 0 , is determined solely by the sensor application -its value should be in the middle of the required sensing range.For a given application, this parameter should be chosen by adequately selecting the sensor material and may be adjusted further by choosing the temperature and properties of exciting light.For environmental and air quality monitoring applications, the value of c 0 corresponds to a NO 2 gas concentration of about 50 ppb, as described in the Introduction.
The sensitivity S is the derivative of the curve in Figure 5(a), and it is approximately equal to 1%/ppb at the lowest gas concentrations.Another important sensor parameter, the limit of detection (LOD), can be determined from the following formula: where δI is the measurement noise, and I 0 is the baseline signal.
Because of the good electrical conductivity of graphene, the noise is relatively small; the typical ratio δI/I 0 equal to 10 −4 , which leads to the estimation of LOD = 0.03 ppb.A comparison of the sensitivity and LOD of different graphenebased materials is given in Table S1.
To understand the underlying mechanisms, the effects of UV light on the response and recovery curves should be considered.Under UV illumination, the response rate increased by an order of magnitude.Fitting the temporal response curves in Figure 3(a) using an exponential function yielded a response time t in of 256 s in the dark vs 24 s under UV light, at 40 ppb of NO 2 .The recovery times t out , determined in a similar way, decreased from 1300 to 200 s, in the order of measurements from 40 ppb to 1 ppm in the dark, but were equal to 96 ± 6 s for all four curves recorded under UV light (see Table 2 and Figure 5(c,d)).
In the case of mutually noninteracting adsorbates on a homogeneous surface (i.e., the Langmuir model), the recovery rates (1/t out ) should ideally depend only on the activation energy of desorption and not on the gas concentration. 59It follows from Table 2 and Figure 5(d) that this is not the case in dark conditions where the desorption rates increase with the gas concentration.In addition, the linearity of the adsorption rates (equal to 1/t in if t in ≪ t out ) on the gas concentration, characteristic of the Langmuir process, 59 was also not observed.Instead, a square-root dependence of 1/t in on the gas concentration was observed (Figure 5(c)).A similar sublinear dependence in the case of NO 2 adsorption on CVD graphene was ascribed to the increasing adsorption barrier with increasing coverage owing to mutual interactions between the adsorbed molecules. 62Such interactions also reduce the adsorbent binding energies, leading to faster desorption (higher 1/t out ) at a more extensive coverage. 63This is exactly the behavior shown in Figure 5(d).The difference with the UV illumination where the desorption rate is higher and constant at all gas concentrations can be explained if the amount of energy released by absorbed quantum light allows molecules with even the strongest binding to cross the desorption energy barriers.The light-induced acceleration of absorption should have a different origin, as the curves under dark conditions and under UV illumination in Figure 5(c) are shifted in parallel toward higher values under illumination.This behavior is consistent with a scenario in which the light quantum does not act imminently at the adsorption, but before it liberates the adsorption sites occupied by other (competing) molecules.For example, continuous exposure of CVD graphene to UV light significantly improved the level of NO 2 detection; a sub-ppt concentration range was achieved in an inert (ultrapure Ar or N 2 ) atmosphere. 16The appearance of reversible oxygen sensitivity of CVD graphene under UV illumination in air, interpreted as the result of partial liberation of oxygen-binding sites, is another supporting example. 50e also examined the effect of visible light at wavelengths 436 and 546 nm.The respective light quanta have energies of 2.84 and 2.27 eV, smaller than the TiO 2 bandgap (≥3 eV 36 ).The behavior of the gas sensor signals under the influence of light with different wavelengths but similar photon flux is presented in Figures 5(b) and S6.The sequence of the experimental steps for the data shown in Figure 5(b) is as follows.First, the sample was kept in the dark under a constant flow of synthetic air (RH = 20%).Then, 546 nm light was applied to Gr/TiO 2 , which caused a slight decrease in the current.After an hour, the wavelength changed to 436 nm, which caused a further drop in the current at a slightly faster rate.Finally, 365 nm UV light was applied, which, in contrast to the previous two wavelengths, caused a much quicker decrease in the current and signal stabilization at about 8.75 μA.The response to 300 ppb of NO 2 was measured under each illumination.The use of 546 and 436 nm shows smaller gas responses to 300 ppb NO 2 compared to 365 nm (10% vs 50% signal change for 30 min).Compared with longer wavelength light, the effect of UV light on NO 2 response and recovery times was much greater, up to 10 times.
Consequently, the onset of illumination efficiency lies in our structure between 365 and 432 nm, indicating the role of TiO 2 as a light absorber and energy channeler.Such large differences in the influence of UV and visible light cannot be ascribed to light absorption by graphene, which covers a wide range from far-infrared to UV. 64 We also analyzed the possibility of the direct excitation of nitrite anion NO 2 − , which is the most likely product formed when NO 2 binds to the Gr/TiO 2 surface, and found it to be negligible because of the small absorption crosssection of only about 20 M −1 •cm −1 . 65In contrast, the TiO 2 pigment strongly absorbs UV radiation above 370 nm, whereas photoinduced charge carriers can both reduce NO 2 and oxidize Ti 3+ or NO 2 − . 36,37In pristine CVD graphene, the sensitizing effect of UV light has been attributed to the excitation of impurities at the Si/SiO 2 substrate or to the removal of surface contaminants and competitors (e.g., O 2 molecules). 66,67electivity and Stability.To determine the selectivity of the optimized Gr/TiO 2 structures, their gas responses were measured with respect to other prominent pollutants (NH 3 , CO, and SO 2 ).It follows from Figure 6(a) that the produced material can detect small NO 2 concentrations more effectively than much higher concentrations of NH 3 and other pollutants.These cross-sensitivity measurements were performed under UV illumination in synthetic air at a relative humidity of 20%.
The gas sensitivity can be compared with data from previous publications on Gr/TiO 2 structures, particularly in the case of ammonia.Previous studies have reported a signal change of approximately 20% at an NH 3 concentration of 400 ppm under visible light 46 or at an NH 3 concentration of 25 ppm in a biased field effect transistor. 68Our structures, fabricated by a different route, had a relative signal change of only a few percent for an NH 3 concentration of 20 ppm (see the dynamic gas responses at different NH 3 concentrations in Figure S7 of the Supporting Information).Obviously, gas sensitivity is strongly dependent on the manufacturing method.
The cross-sensitivity to the RH change was also relatively small, as shown in Figures 6(a) and S8.In Figure 6(a), the change for a 50% humidity difference is shown, whereas the supplementary Figure S8 provides a more detailed view of the signal changes compared to the NO 2 response.Small changes in the background signal owing to humidity variations facilitate the practical use of sensors at NO 2 concentrations below 100 ppb.A comparison of Figures 3(a) and S4 reveals that NO 2 response amplitudes depend somewhat (up to 50%) on RH: the sensitivity at low NO 2 concentrations is higher in humid air.The response rates exhibited a similar trend, as shown in Figure 5(c).
Long-term stability and repeatability of responses to the target gas are essential for reliable sensors.Figure 6(b) demonstrates that even without preconditioning (e.g., heating in an inert gas or vacuum), the sample exhibited excellent stability and repeatable responses to NO 2 at concentration ranging from 40 ppb to 1 ppm over 29 months.Between the experiments, the sample used in the long term tests was held in ambient air in a laboratory room without special humidity/ temperature controls.
The largest variations in the long term measurement data were observed in the initial baseline of the signal and its drift.These changes did not behave monotonically in one direction or another and were likely related to seasonal variations.The response amplitudes, as measured from the beginning to the end of each 300 s gas exposure cycle, were more stable; in the data of Figure 6(b), the variations in the amplitudes are below 10%.Another essential sensor characteristic, the limit of quantization (LOQ), can be determined based on the longterm accuracy estimate.Assuming that the baseline variability can be corrected, we estimated that the LOQ = 5 ppb.

■ CONCLUSIONS
We fabricated gas sensor heterostructures based on CVD single-layer graphene and a laser-deposited TiO 2 layer.In such a structure, the semiconducting metal oxide fulfills the function of a redox-active receptor, whereas graphene is an electronic signal transducer.By varying the deposition conditions we were able to find an optimal TiO 2 thickness and a trade-off between the degree of graphene functionalization and the partial damage to the 2D lattice of graphene during deposition.Optimal heterostructures are well suited for detecting NO 2 gas at concentrations below one ppm that must be monitored in the environment.The sensitivity, defined as the relative signal change per unit gas concentration, reached 1%/ppb under UV light (365 nm, 0.2 mW/mm 2 ) at room temperature.Exposure to relatively weak UV light increased the sensor signal amplitude by several times and, more importantly, accelerated both the reaction and recovery kinetics by an order of magnitude.The response times were between 24 and 5 s, at 40 ppb and 1 ppm of NO 2 , respectively, whereas the recovery times were approximately 100 s at all tested gas concentrations.The effect of visible light (436 and 546 nm) turned out to be much weaker, indicating that the excitations due to light absorption in TiO 2 play an important role, whereas the characteristics of the gas responses imply the involvement of both photoinduced adsorption and desorption.The sensor structure was highly selective to NO 2 , as compared to CO, SO 2 , and NH 3 , had a low (<5%) cross-sensitivity to humidity, and retained high sensitivity to NO 2 for over two years.The sensing mechanism is ascribed to NO 2 adsorption on the surface of TiO 2 , which results in electron transfer from graphene to the adsorbed gas molecules.Consequently, the Fermi level of graphene moves further away from the Dirac point, and the conductivity of the initially p-doped graphene increases.This was confirmed by DFT simulations with Ti 8 O 16 clusters on graphene, showing both the Fermi level shift of graphene and an efficient charge transfer between the graphene and NO 2 molecule adsorbed on the Ti cation.Experimentally, it was shown that the optimal thickness of the TiO 2 receptor layer is about 0.6 nm, corresponding roughly to the oxide monolayer.Although it is not a crystalline monolayer but a granular coating, the result paves the way for a novel class of gas sensor structures based on 2D heterostructures with an ultimately sensitive transducer and selective receptor layer.

■ MATERIALS AND METHODS
Sample Fabrication.The CVD graphene used in this study was grown on 25-μm thick polycrystalline copper foil (99.5%, Alfa Aesar) in a laboratory-built hot-wall quartz tube CVD reactor or purchased from Graphenea.To transfer it onto substrates (Si with 300 nm SiO 2 ; area 10 × 10 mm 2 ) with two predeposited Ti/Au (7 nm/80 nm) electrodes, graphene on Cu foil was first coated a with poly(methyl methacrylate) (PMMA; MW ∼ 996 000 g/mol, Sigma-Aldrich) layer, and then Cu was dissolved in 1 M ammonium persulfate solution.Graphene attached to the PMMA overlayer was thereafter transferred in deionized water (DI) onto the electrode substrate.Finally, PMMA was dissolved in dichloromethane (Alfa Aesar), and the sample was washed with DI and dried.
Pulsed laser deposition (PLD) was utilized to produce oxide nanolayers on top of the graphene.Prior to deposition, the sensor substrates with graphene were heated in a PLD chamber at 150 °C at a base pressure of 10 −6 mbar for 2 h.PLD was performed at 35 °C in 0.05 mbar of N 2 gas using a KrF excimer laser (COMPexPro 205, Coherent) with a pulse frequency of 5 Hz at an energy density of 7 J/ cm 2 for target ablation.A hot-pressed 1-in.diameter TiN sputtering target disc (Goodfellow) was used as the PLD target.
The gas pressure of 0.05 mbar in the PLD process was found to be optimal based on the maximal sensitivity of the Gr/TiO 2 heterostructure.Graphene was significantly damaged if lower pressure was used in the PLD process.At a gas pressure of 0.01 mbar in the deposition chamber (which can be referred to as "harsh" deposition conditions), even a single laser pulse produced the D-line in the Raman spectrum with I D /I G > 2. In refs 46 and 68 a 3−10 nm thick TiO 2 layer was created on graphene by electron-beam evaporation (EBE; actually, Ti was deposited that later oxidized in the air), while the D-band did not appear in the Raman spectrum as a result of this process.Consequently, EBE did not produce any defects in graphene, which can be explained by the lower energy of the deposited particles than that in the PLD process used in our study.The energy of particles landing on the substrate can be reduced in PLD by increasing the gas pressure in the deposition chamber, which leads to the thermalization of the initially highly energetic plasma plume.At higher gas pressures ("soft" deposition conditions), the number of created defects decreased, as indicated by the small I D /I G ratio and relatively small decrease in conductivity (see Figure S9).However, the relative gas response was smaller than that obtained at an optimal process pressure of 0.05 mbar.This can be explained by the fewer adsorption centers created or less effective charge transfer between graphene and adsorbed gas molecules; seemingly, certain defects created during the deposition in graphene may promote this transfer.
Characterization Methods and Experimental Details.All gassensing measurements were performed using graphene-based resistors at room temperature.The volt-ampere characteristics were linear (see the example in Figure S10 of the Supporting Information).For the transconductance measurements, the back gate was provided by the Si substrate.To contact the gate, the oxide layer was removed from the back side of the Si plate using HF, which was then coated with a Ti(7 nm)/Au(80 nm) contact layer.The field-effect mobility was estimated at U G = 0 using the equation μ FE = g × (L/W)/(C ox × V SD ), where g is the transconductance, L and W are the length and width of the channel, respectively, and C ox is the capacitance of the oxide layer (300 nm thick SiO 2 ) per unit area.
The gas-sensing measurement scheme is shown in Figure S11.The electrical conductance was recorded with a Sourcemeter Keithley 2400 at a fixed supply voltage of 100 mV applied to the electrodes with a gap length of L = 4 mm and width of W = 1 mm.In transconductance measurements, the second Sourcemeter provided the gate voltage between the upper Au electrode and the Si substrate.In all experiments, a constant gas flow of 200 sccm was maintained through a 7 cm 3 sample chamber.The purity of all gases used was 99.999% (Linde Gas).All measurements were performed in synthetic air (21% O 2 and 79% N 2 ), whereas the nitrogen flow was partially directed through a water bubbler, thus allowing the regulation of relative humidity (RH) levels between 0% and 50%.The ratio between the flow rates from different gas cylinders was changed using mass flow controllers (Brooks models SLA5820) to adjust RH and NO 2 concentrations.Gas concentrations and humidity levels were regularly calibrated by measuring the gas mixer output with an APNA-370 NO x monitor (Horiba) and a Hygropalm 1 humidity and temperature indicator (Rotronic), respectively.A Xe−Hg lamp (L2422, Hamamatsu) with different sets of filters was used as a light source: 1) a water-filled filter to cut off the infrared part of the light; 2) a narrow-band interference filter to select the wavelength (Andover 2-in.models with central wavelengths of 365, 436, and 546 nm); and 3) grayscale filters to set the intensity of light on the sample.The same photon flux of 3.7 × 10 20 photons/m 2 •s was maintained for all wavelengths, corresponding to 0.2 mW/mm 2 at 365 nm.During the continuous illumination, the light power on the sample was 2.4 mW.We determined the temperature rise due to illumination to be 1.1 ± 0.3 °C at a light power of 10 mW, which allowed us to neglect the thermal effects.
Structural characterization of graphene was performed using a micro-Raman spectroscopic system Renishaw inVia at an excitation wavelength of 514 nm.The morphology of the samples was characterized using a high-resolution SEM Helios NanoLab 600 (FEI).The elemental mapping of the samples was performed using the EDX spectrometer from Oxford Instruments (Abingdon, UK), which was coupled to SEM.TiO 2 layer thickness was deduced from XRF and ellipsometry data, recorded with a dispersive spectrometer ZXS-400 (Rigaku) and a spectrometric ellipsometer GES-5E (Semilab, Sopra), respectively.XPS measurements were conducted under ultrahigh vacuum conditions, using Mg K a X-rays from a nonmonochromatic twin anode X-ray tube (Thermo XR3E2) and an electron energy analyzer SCIENTA SES-100.
Full range Raman spectrum of Gr/TiO 2 on Si/SiO 2 .Wider area SEM images of lab-made and commercial graphene, EDX mapping of Gr/TiO 2 on Si/SiO 2 , gas responses and repeatability in dry air (RH = 0), photoresponses of the baseline current in air at different light wavelengths (365, 436, and 546 nm), Gr/TiO 2 band diagram, dynamic responses to NH 3 and H 2 O, gas response of the sensor with TiO 2 deposited at higher pressure, volt-ampere characteristic, scheme of gas measurement setup, and comparison table of graphenebased NO 2 gas sensors (PDF)

Figure 1 .
Figure 1.(a) Raman spectra of graphene samples with different amounts of deposited TiO 2 (N denotes the number of laser pulses used in the PLD process).(b) Transconductance before and after the TiO 2 deposition.SEM images of (c) pristine graphene and (d) TiO 2coated graphene.

Figure 3 .
Figure 3. (a) Response of TiO 2 -coated graphene to different NO 2 concentrations in the dark and under continuous UV illumination (λ = 365 nm).Measurements were performed with a supply voltage of 0.1 V at room temperature in synthetic air with RH = 20%.The thickness of TiO 2 on single-layer CVD graphene was 0.6 nm.Inset: relative response of unmodified pristine graphene to 4 ppm of NO 2 .(b) Dependence of the relative response to 300 ppb NO 2 and (c) the normalized conductance G/G 0 (G 0 is the conductance of the sample before PLD) on the thickness of the deposited TiO 2 layer.

Figure 4 .
Figure 4. Partial densities of the states of complexes (a) Ti 8 O 16 C 50 and (b) NO 2 @Ti 8 O 16 C 50 were obtained using a k-mesh 25 × 25 × 1.The dashed lines indicate the position of the Fermi level.The dotdashed line shows the position of the minima of PDOS of the carbon subsystem in the considered complexes.The inset in the lower panel shows a schematic view of the ground state geometric structure of the complex NO 2 @Ti 8 O 16 C 50 used in DFT simulations.

Figure 5 .
Figure 5. (a) Dependence of relative response on NO 2 concentration in the dark and under UV light.The red solid line represents the approximation of UV data to the Langmuir isotherm with c 0 = 65 ppb.(b) Response of Gr/TiO 2 to NO 2 under illumination by 546, 436, and 365 nm light.Measurement was performed in synthetic air at room temperature (RH = 20%).Dependence of (c) response and (d) recovery (desorption) rates on NO 2 concentration as determined from the data shown in Figure 3(a) and Figure S4(a).Part of the parameters obtained from data fitting are given in Table1(the rates are defined as 1/t in and 1/t out , respectively).

Figure 6 .
Figure 6.(a) Relative conductive response of Gr/TiO 2 to different pollutant gases and humidity level changes.(b) Stability and repeatability of Gr/TiO 2 sensor electrical conductance signal and gas responses over 29 months.Samples were stored under ambient conditions between measurements.

Table 1 .
Number of Valence Electrons in the Subsystems Calculated by the Bader Method a The number of valence electrons considered in the PAW pseudopotentials.Ti 8 O 16 C 50 in Table

Table 2 .
Relative Response Amplitudes a and Characteristic Times for Response (t in ) and Recovery (t out ) is the conductance in the presence of NO 2 gas, and G air is the baseline conductance in air.
a G g