A CuO/TiO2 Heterojunction Based CO Sensor with High Response and Selectivity

The use of heterojunctions is a promising solution to the problem of cross‐sensitivity in gas sensors. In this work, a carbon monoxide sensor based on the CuO/TiO2 heterojunction was designed and fabricated. Due to the good adsorption properties of CuO materials to CO, and the heterojunction interface charge transfer, the CuO/TiO2 thin film sensor exhibits high sensitivity to CO at room temperature. The response is as high as 10.8–200 ppm CO, about 10 times its response to H2. Interference from H2 is greatly reduced by optimizing the structure of the CuO/TiO2 heterojunction. This reliable detection of carbon monoxide with excellent discrimination against H2 is of great significance for the development of CO gas sensors.


Introduction
Carbon monoxide is a flammable and explosive toxic gas, so effective detection and monitoring of the concentration of carbon monoxide is paramount to human health and safe industrial production. [1,2] Consequently, developing efficient carbon monoxide gas sensors is a safety necessity. Semiconductor based CO sensors have been widely used due to their advantages of low cost and good stability. [3][4][5][6][7] Lee et al. reported that multi-walled carbon nanotubes (MWCNTs)-doped TiO 2 xerogel composite films, synthesized by the sol-gel method, can achieve high response to 50 ppm CO gas at 300°C. [8] Pan et al. reported a fast response Pd-TiO 2 thin film sensor, in a temperature range of 573-753 K, with a response time, to 80-800 ppm CO, of 10-15 s. [9] A specific morphology of nanostructured spinel cobalt oxide (Co 3 O 4 ) has been reported to exhibit excellent performance towards CO, with a 360% response to 200 ppm CO at 300°C. [10] Hsu et al. prepared SnO 2 /In 2 O 3 nanofibers, which had the best adsorption performance for CO and had a response of 9.47% to 50 ppm CO at 200°C. [11] In particular, room temperature operation of CO sensors is very important for the safety applications. [12] Wang et al. prepared a carbon monoxide gas sensor based on ZnO/SnSe 2 composite film operating at room temperature, which showed a 14.8% response to 200 ppm CO and good dynamical characteristics. [13] Zhang et al. prepared a Pd-WSe 2 sensor with a response of 9.25% to 5 ppm CO at room temperature. [14] Naganaboina et al. used a solvothermal method to synthesize GNPs-CeO 2 nanocomposites, which can detect 2-10 ppm CO gas at room temperature (27°C), but with a response of only 1.5% for 10 ppm CO. [15] Although the effective detection of CO can be achieved at room temperature, there is still an urgent demand to further enhance the response whilst simultaneously ensuring high specificity. Among the various interfering gases, H 2 is the most serious for CO. The very similar chemical properties to CO lead to difficulties in distinguishing the two gases with resistance type semiconductor gas sensors. [16] Commercial sensors often document this interference in their specifications for CO and H 2 sensors. For example a CO detector exposed to hydrogen at a concentration of 7% lower explosive limit, sounded an alarm within 3 min and had a sensitivity of 10%. [17] The cross-sensitivity reported of the carbon monoxide detectors to hydrogen is about 20%. [18] There are few reports in the literature for differential detection of CO and H 2 sensors at room temperature. The underlying mechanism for the selective detection of CO and H 2 also remains largely unexplored. The development of sensors that can selectively detect CO and exclude H 2 interference is an urgent area for scientific research and commercial applications.
The construction of heterojunction based gas sensors is one of the promising ways to improve both the response to a specific gas and to restrain interference from other gases. Yin et al. successfully detected H 2 by preparing n-SnO 2 /p-Co 3 O 4 composite nanomaterials. The sensor demonstrated reduced response to CO, making a major breakthrough in the problem of poor selectivity. [19] Yin et al. prepared an xSnO 2 -yCr 2 O 3 nanocomposite, which exhibited an n-type behavior for H 2 and an opposite p-type behavior for CO by regulating the operating temperature. [20] Sharma et al. prepared electro-spun nanofibers for H 2 and CO gas sensors based on polyaniline (PANI)/SnO 2 composites. The response of the sensor to 1000 ppm H 2 was much higher than to the same concentration of CO. [21] Moon et al. prepared SnO 2 -ZnO composites with surface coated CuO. High selectivity to CO was obtained between 150°C and 250°C, while high selectivity to H 2 gas was shown between 310°C and 400°C. [22] Table 1 compares the performance of reported sensors which could detect both H 2 and CO. An interference factor, IF(CO, H 2 ), expressed by S(CO)/S(H 2 ) is adopted to indicate the sensor selectivity, where S (CO) is the response to CO, and S(H 2 ) is the response to H 2 . An IF (CO, H 2 ) value bigger than 1 indicates a better response to CO than to H 2 and vice versa. A heterojunction based sensor which can effectively detect CO at room temperature without interference from H 2 remains undeveloped.
The aim of this work is selecting suitable materials to construct a heterojunction with both a high response and high interference factor IF(CO, H 2 ) to CO at room temperature. TiO 2 is a good base for constructing heterojunctions for detecting CO. [28] Hsu et al. prepared a TiO 2 /perovskite pn-type heterojunction gas sensing material with a response value of 38.41% at 200°C to 400 ppm CO. [29] Looking for optimum materials to form the heterojunction would further enhance the gas sensing performance of TiO 2 . Oxide materials which are highly sensitive to CO are perfect candidates. Luo proposed that well-dispersed CuO is capable of adsorbing CO at low temperature, and the reducing nature is the reason for the lowtemperature CO oxidation, whereas bulk CuO cannot adsorb CO at high temperature and the reducing nature contributes little to the oxidation activity. [30][31][32] It has been reported that the exposure of CuO to CO induces reactive adsorption, surface reduction and carbonate formation, while the subsequent Cu + is able to weakly readsorb/desorb CO at low temperature. [33] Zhang et al. synthesized a series of nanocomposite catalysts with different copper contents by a simple low-temperature hydrothermal method. This demonstrated that copper supported on CeO 2 can be used as a chemisorbed CO site and efficiently convert CO to CO 2 . [34] Fine control of the microstructure of the CuO will help improve the sensing performance to CO.
In this work, a heterojunction sensor with a CuO/TiO 2 bilayer thin film structure is constructed for CO detection and prevention against H 2 interference. Self-assembled [002]-oriented TiO 2 nanorod films were used as the bottom n-type layer and CuO nanosheets with (111) exposed facets were adopted as the top p-type layer. Gassensing properties of the CuO/TiO 2 heterojunctions to CO and H 2 were measured at room temperature. A mechanism for detecting CO with high response and effectively excluding the interference from H 2 is proposed based on the heterojunction structure formation, surface adsorption and the matching of the space charge regions. The interference factor IF of the optimized sample is 8.9, which shows an excellent CO selectivity against H 2 even when working at room temperature.

Microstructural Characterization
Morphologies of the single layer TiO 2 and CuO thin films are shown in Figure 1a-d. The TiO 2 layer is composed of TiO 2 nanorods, as shown in Figure 1a Figure 1b, which is about 3.4 μm, while the gaps between the TiO 2 nanorods are filled by CuO nanosheets. The top layer is the CuO film, and the surface morphology of CT1, as shown in Figure 1e, is very similar to the pure CuO ( Figure 1c). With the increase in the precursor concentration, the size of the nanosheets increased from about 600 nm to 1 μm and then increased to about 1.5 μm, as shown in Figure 1e,g,i. The CuO nanosheets started to accumulate on the surface of CT3. The thickness of the CuO layers in CT1, CT2 and CT3 are 3.511, 5.172 and 5.952 μm, due to the difference in the concentration of copper precursor.
TEM was carried out to study the interface of the CuO/TiO 2 heterojunction. Figure 2 shows the TEM image of sample CT2 and the corresponding EDS results. It can be seen from Figure 2a,b that the interface is composed of TiO 2 nanorods and CuO nanosheet. The diameter of TiO 2 nanorods is 100-200 nm. The CuO nanosheets, with a thickness of about 0-100 nm, grow on the surface of the TiO 2 nanorods. The corresponding high-resolution TEM image of the interface is shown in Figure 2c. For the upper left region of the image in Figure 2c, the lattice spacing is estimated to be 0.2516 nm, corresponding to the D spacing of the (−111) crystal plane of CuO. The lattice spacing of the lower right region in the HRTEM image in Figure 2c is 0.3235 nm, which is consistent with the D spacing of the (110) crystal plane of TiO 2 . The elemental mapping results in Figure 2d-h further confirm that the CuO nanosheets initially grow from the space between the TiO 2 nanorods. These form a close contact with the surface of the nanorods, and then grow continuously on the surface of the TiO 2 nanorod layer, consistent with the SEM results. TEM results confirmed the successful construction of a CuO/TiO 2 heterojunction and a considerable increase in the interface area due to the intimate contact between CuO/TiO 2 .
The XRD spectra of the as-prepared samples are shown in   [21] Pt/SnO 2 400 2000-8000 0.2 [23] ZnO@ZIF-8 250 50 0.35 [24] SnO 2 nanoparticle 150 500 0.59 [25] SnO 2 /Co 3 O 4 350 500 1-2 [19] SnO 2 /xCuO 350 400 3.44 [26] Rb (5 wt.%)-In 2 O 3 300 1000~10 [27] CuO/  Figure 3b. The peaks at around 458.6 and 464.4 eV correspond to the 2p3/2 and 2p1/2 peaks of Ti 4+ , no low-valence states of Ti were found in the samples, indicating that there are few Ti 3+ defects on the surface of the samples. Four characteristic peaks can be found in the XPS spectra of CuO and CT2. The peaks at 933.8 and 953.7 eV correspond to the 2p3/2 and 2p1/2 peaks of Cu 2+ , the peaks at 942.3 and 962.3 eV are the satellite peaks of Cu 2+ , which indicates that pure CuO phase is present in the sample. The O1s spectra of TiO 2 and CT2 are shown in Figure 2c, d. In Figure 2c, the core energy levels can be divided into three peaks, O lat (529.95 eV), O vac (531.71 eV) and O ads (532.51 eV), attributed to lattice oxygen, oxygen defects and surface adsorption, respectively. The three peaks could also be found in sample CT2, the difference is that the O vac and O ads peaks are much higher than that in pure TiO 2 , indicating many more oxygen vacancies and adsorbed oxygen species on the surface of sample CT2.

Sensing Properties
The carbon monoxide sensing curves of the samples are shown in Figure 4a-e. Figure 4a is the carbon monoxide sensing curve of TiO 2 . The sample shows a typical n-type response to reducing gas. When the reducing gas (carbon monoxide) was introduced, the resistance decreased and then increased after the CO was released. The resistance reaches a stable state giving a certain response and recovery time while the recovered resistance was smaller than the original one. Figure 4b is the carbon monoxide sensing curve of CuO. The sample shows a typical p-type response to reducing gas. When the reducing gas (carbon monoxide) was introduced, the resistance increased and then the value decreased after the CO is released. The recovered resistance became larger than the original value. Figure 4c-e is the carbon monoxide response curves of CT1, CT2 and CT3, respectively, all of which show a p-type response. The starting resistance of the CT1, CT2 and CT3 gradually decreased with the increase in CuO thickness. The recovered resistances of CT1, CT2 and CT3 are very close to the starting resistances, indicating the samples are more stable than pristine TiO 2 and CuO. The CO response of the samples are calculated based on the resistance change as described in Section 4 and the comparison of the gas responses of the five samples are shown in Figure 4f. As can be seen from the figure, the CO response of the TiO 2 /CuO composite bilayer heterostructure is significantly higher than that of the single layer samples, especially for sample CT2. The difference between the CuO sample and CT2 is that there is a TiO 2 layer at the bottom of CT2. This suggests that the heterojunction formed between CuO and TiO 2 has a strong impact on the CO sensing performance of the sample. The response of samples TiO 2 , CuO, CT1, CT2 and CT3 to H 2 is tested and the results are shown in Figure 5. When the samples are put in the H 2 environment, a similar n-type response was observed on the TiO 2 sample, and p-type responses were observed on CuO, CT1, CT2 and CT3. The recovered resistance of TiO 2 , CT2 and CT3 are quite stable, while for CuO and CT1 in different concentrations of H 2 it gradually decreased as shown in Figure 5a. The dynamic testing curves of CT2 and CT3 to H 2 seemed very similar to that to CO, while the change in resistance to H 2 as shown in Figure 5a is much smaller than to CO as shown in Figure 4. The response of samples TiO 2 , CuO, CT1, CT2 and CT3 to CO and to H 2 are compared in Figure 5b. All of them increased with increase in gas concentration, especially at low concentrations. The response of the single layer TiO 2 sample to H 2 is significantly higher than that of CO. The pure CuO sample shows the opposite response with a higher response to CO than to H 2 . When the heterojunction formed in CT1, CT2 and CT3, the response to CO is significantly enhanced, and the increase in the response to H 2 is much smaller than for CO. The results show that constructing TiO 2 /CuO heterojunctions, offers a promising way to strongly detect CO while suppressing the interference from H 2 .
The sensing mechanism of semiconductor metal oxide gas sensors can be explained as the redox reaction between the oxygen adsorbed on the surface of the sensing material and the test gas molecules, resulting in the change in the carrier concentration. Since the electron affinity of the oxygen molecule is higher than the work function of the sensing material, the oxygen molecule will capture electrons from the conduction band of the semiconducting oxide and form different oxygen species (O À 2 , O − , and O 2− ). CuO is a typical p-type semiconductor in which holes are the main conductive carriers. The CO sensing mechanism can be explained by a modulation of the hole accumulation layer (HAL) induced by oxygen adsorption. The sensor is tested at room temperature (25 AE 1°C). When the sample is exposed to the air environment, oxygen molecules can adsorb on the surface of the sample to form oxygen anions (O À 2 ) by capturing electrons. This process can be expressed by Equations (1) and (2). The electrons in the conduction band will be transferred to the oxygen species, resulting in an increase in hole carrier concentration, forming a hole-rich hole accumulation layer, which ultimately reduces the resistance of the sensor material. When the sensor is placed in a CO environment, CO molecules will be adsorbed on the sample surface and react with surface oxygen species, this process can be represented by Equation (3). The trapped electrons will be released back to the surface and recombine with holes, resulting in a thinner hole accumulation layer and increased sensor resistance.
As the p-type CuO and n-type TiO 2 are combined together, a space charge region will form at the interface of the two layers. When the  Figure 5c. The resistance of the TiO 2 /CuO heterojunction samples consists of the resistance of the TiO 2 layer, the junction resistance, the resistance of pristine CuO and the resistance of the HAL. They are connected parallel as shown in Figure 5d. With a thin layer of CuO as in CT1, the HAL layer is small, with a resistance comparable to the TiO 2 layer (3.6 kΩ). Consequently, the starting resistance of CT1 (3.4 kΩ) is comparable to that of pristine TiO 2 . With the increase in CuO thickness, the HAL layer in CT2 and CT3 increased, and the resistance of this layer gradually decreased. The parallel resistances of CT2 and CT3 are largely determined by the HAL resistance. Consequently, the starting resistances of CT2 and CT3 decreased to 0.9 and 0.5 kΩ, as shown in Figure 4d,e.
When the samples are exposed to CO, electrons were injected in to the CuO layer, recombining with the holes in the HAL and so the resistance of the film increased. When CT1 was exposed to 200 ppm CO, the recovered resistance is about 12.1 kΩ, very close to that of the pristine CuO layer. As CT1 has similar CuO thickness to that of pristine CuO, we can deduce that electrons returned by 200 ppm CO could recombine almost all the accumulated holes in the CuO layer, leading to the recovery of the layer. When CT2 is exposed to 200 ppm CO, the resistance is smaller than that of CT1, 10.1 kΩ, which means that the HAL layer in CT2 has only partially recombined. Further increase in the CuO layer in CT3 results in an even thicker  HAL layer. The resistance of CT3 is mainly determined by the unrecombined HAL layer and the resistance of CT3 in 200 ppm is about 3.2 kΩ. The enhanced response of the heterojunction in CO can also be attributed to the increase in oxygen vacancies. Comparing with the single layer sample, more oxygen vacancies were observed as shown in the XPS results in Figure 3. Thus the heterojunction thin film demonstrated bigger changes with exposure to CO. The TiO 2 layer modifies the space charge region and changes the resistance of the heterojunction accordingly. The width of the space charge region in the CuO layer will decrease, and in the TiO 2 layer it will increase after adsorption of oxygen species in air, and the width change reverses in CO or H 2 gas. This change in the space charge region further enhanced the resistance change in CT2, giving the biggest response. The existence of the space charge region also promotes the adsorption and desorption of CO on the heterojunction film surface, resulting in a faster response and more stable baseline resistance.
The different sensing behaviors of the samples in CO and H 2 can be further explained by the adsorption mechanism of the gas on the materials and the diffusion length of the two gases. We calculated the adsorption of CO and H 2 on the CuO surface with oxygen vacancies, the adsorption energy E ads is defined by: where E all is the total energy after adsorption, and E substract and E gas represent the total energy of CuO-O vac and free gas molecules, respectively. Higher adsorption energy values indicate that more gas molecules can be adsorbed, and negative E ads indicate that the binding process is exothermic. The E ads of CO is calculated to be −0.6482 eV, and the E ads of H 2 in the sample is −0.0276 eV. The adsorption energy of CO on CuO is about 24 times that of H 2 . This calculation proves that CO can be better adsorbed on the surface of CuO than H 2 . Another factor is the diffusion length of the two gases. The CO gas has larger molecular size than H 2 , which means that it has shorter diffusion length in the heterojunction, so the CuO layer plays the main function for sensing CO. When H 2 is introduced, a small amount of H 2 can be adsorbed on the surface of CuO due to the small adsorption energy of H 2 on CuO.
The reaction between H 2 and oxygen species on CuO surface releases electrons, leading to the reduction of hole accumulation layer, but with less effect on the resistance change in CuO due to the small adsorption energy. The small size of the H 2 molecule means it can easily diffuse into the TiO 2 layer through the gaps of the CuO nanosheets and react with the surface of TiO 2 . The release of electrons leads to the resistance decrease of the TiO 2 layer as well as the junction resistance, which compensates the resistance increase of the CuO layer, so the device has poorer sensitivity to H 2 than CO. CT2 was further tested in NO 2 and NH 3 , and the test results are shown in Figure 6. Figure 6a is the test result of CT2 to NO 2 ; a decrease in resistance was observed when NO 2 was introduced, agreeing well with the behavior of p-type CuO to oxidizing gases. The NO 2 captured the electrons on the surface of CuO, increased the concentration of holes and so the resistance of the p-type CuO decreased. However, the resistance change in CT2 was largely constant and independent of the different concentrations of NO 2 . The low response of CT2 to NH 3 is shown in Figure 6a. Figure 6b compares the gas sensing curves of CT2 to one cycle of CO, H 2 , NO 2 and NH 3 , at a concentration of 200 ppm. Clearly the response of CT2 to CO is significantly higher than that of the other three gases, indicating that CT2 has good selectivity to CO. The response of TiO 2 , CuO, CT1, CT2 and CT3 to 200 ppm CO, H 2 , NO 2 and NH 3 are compared in Figure 6c. It can be seen from the figure that TiO 2 has extremely low responses to CO and the response to H 2 is the highest. CuO has the best response to NO 2 , and a smaller response to CO. By combining the two materials TiO 2 and CuO together as CT1, CT2 and CT3, the response to CO is much improved. The value of the interference factors IF(CO, H 2 ) are calculated and the result is shown in Figure 6c by the red stars. Clearly when the heterojunction was formed, the value of IF(CO, H 2 ) is significantly increased. In particular, the IF(CO, H 2 ) of CT2 reaches the largest value, nearly 8.9. This demonstrates that the prepared CuO/TiO 2 heterojunction thin film sensor is able to detect CO with high selectivity against H 2 . Humidity resistance is also one of the important performance indicators of a sensor. The response curves of CT2 samples tested under humidities of 45%, 55% and 70% are presented in Figure 6d. The similar curves indicate the samples were independent of humidity, showing excellent humidity resistance.

Conclusions
In this article, a heterojunction gas sensor composed of CuO nanosheets and TiO 2 nanorods was prepared for CO detection and to improve the selectivity of the sensor. The results show that at room temperature (25 AE 1°C), the sample with the best performance, CT2, has a response value of 10.8-200 ppm CO, a great improvement compared with the pure sample. The sensor was tested in several gases and showed excellent selectivity. In particular, the response to H 2 was only 1.2, this is more than 10 times less than for CO than the pure sample. The response to CO is opposite, with the calculated IF(CO, H 2 ) value for CT2 reaching a maximum value of 8.9. This article also discusses the mechanism for improving the CO response and suppressing the H 2 response. The CuO layer plays an important role in the improvement in CO performance, mainly because the surface of the CuO layer provides more oxygen vacancies and high adsorption energy. The difference in the adsorption and the diffusion lengths of the two gases also determine the different responses. When p-type CuO and TiO 2 are combined, a space charge region is formed at the interface of the two layers, and the TiO 2 layer influences the space charge region and changes the resistance value of the resistive heterojunction. The change in the space charge region adds to the more pronounced change in the resistance of the device to CO than for H 2 . This work provides a practical solution for the selective detection of CO without H 2 interference.

Experimental Section
Synthesis and characterization of CuO/TiO 2 bilayer thin films: The CuO/TiO 2 bi-layer thin films were prepared by a two-step method on F-doped SnO 2 (FTO) substrate. The TiO 2 layer was prepared by a hydrothermal method as described in our previous study. [35][36][37] The prepared TiO 2 films were annealed at 400°C for 20 min in air, and then used as the substrate to grow the CuO layer. CuO thin films were prepared by a water bath method. Copper nitrate trihydrate (Cu(NO 3 ) 2 Á3H 2 O) was used as a precursor and dissolved in 80 mL of deionized water. Four to seven millilitres of condensed ammonia water was slowly added into the precursor, and then stirred evenly to obtain the precursor solution. Two FTO substrates or TiO 2 thin film substrates were put into the precursor with the conducting side of the FTO or TiO 2 side facing down. The water bath reaction was carried out at 75°C for 3 h. The obtained samples were washed with deionized water and dried at 60°C. The as-prepared CuO and CuO/TiO 2 films were then annealed in air at 200°C for 1 h. The CuO/TiO 2 samples are labeled as CT1, CT2, CT3, corresponding to different precursor concentrations. The concentrations of Cu(NO 3 ) 2 Á3H 2 O in samples CT1, CT2 and CT3 were 0.05, 0.1 and 0.15 M, respectively. The concentration of Cu(NO 3 ) 2 Á3H 2 O for the CuO single layer is 0.1 M, the same as that in CT2.
The prepared samples were characterized by X-ray diffraction (Bruker D8-Advance) using Cu Kα radiation (λ = 0.15418 nm) with a scan rate of 4°min −1 over a 2θ range of 20-70°. Planar and cross-sectional morphology of the films was examined by field emission scanning electron microscopy (Sigma 500; Zeiss) and TEM (Tecnai G2 F30; FEI). Chemical compositions and valence band structure were characterized by X-ray photoelectron spectroscopy (ESCALAB 250 Xi; Thermo Fisher).
Fabrication of CuO/TiO 2 gas sensor: Platinum interdigitated electrodes were deposited onto the CuO/TiO 2 sample surface for the gas sensing test. The specific devices size are the same as those in our previous reports. [38] Carbon monoxide and hydrogen sensing of the samples were measured at a constant bias voltage of 1 V. The prepared samples were exposed to CO and reference gases at different concentrations for 5 min and then re-exposed to air for 5 min. The sensing response S of the samples was calculated by the comparison of the resistance in air and gas: where R air and R gas denote the resistances of the sensor in air and in the testing gas, respectively. The response and recovery times were defined as the time taken for the resistance variation to reach 90% of the final ΔR (ΔR = R air − R gas ). The value of the interference factor (IF(gas1, gas2)) of a certain concentration of CO to H 2 was defined by the equation: where S(CO) and S(H 2 ) are the response to CO and H 2 , respectively. Calculation of the adsorption energy of CO and H 2 on CuO surface was carried out by DFT modeling similar to that as described in our previous study. [35,38]