Atomic layer deposition to heterostructures for application in gas sensors

Atomic layer deposition (ALD) is a versatile technique to deposit metals and metal oxide sensing materials at the atomic scale to achieve improved sensor functions. This article reviews metals and metal oxide semiconductor (MOS) heterostructures for gas sensing applications in which at least one of the preparation steps is carried out by ALD. In particular, three types of MOS-based heterostructures synthesized by ALD are discussed, including ALD of metal catalysts on MOS, ALD of metal oxides on MOS and MOS core–shell (C–S) heterostructures. The gas sensing performances of these heterostructures are carefully analyzed and discussed. Finally, the further developments required and the challenges faced by ALD for the synthesis of MOS gas sensing materials are discussed.


Introduction
Toxic and harmful chemicals that poses a serious threat to the human health and the environment are ubiquitous in our daily life [1]. The current demand for high performance gas sensors to detect poisonous gases is growing fast. Metal oxide semiconductors (MOSs) sensing layers display high sensitivity and fast response and recovery, and can meet the current demand for gas sensors. Other types of semiconductor materials, such as metal nitrides, also show some prospects in gas 3 Contributed equally to this work. * Author to whom any correspondence should be addressed.
Original content from this work may be used under the terms of the Creative Commons Attribution 4.0 licence. Any further distribution of this work must maintain attribution to the author(s) and the title of the work, journal citation and DOI. sensing [2,3]. However, metal nitrides are often sensitive to air and moisture [2], and their synthesis usually requires harsh synthesis conditions such as high temperature. Besides easy synthesis, the low cost, small size and easy operation of MOS afford a great potential to be utilized into various sensor devices [4][5][6].
In the past decade, nanostructured MOS materials such as nanoparticles [7], nanowires [8,9], nanosheets [10], thin films [11], as well as 3D hierarchical structures [12] have been applied in gas sensing. However, when used as the sensing layer, pristine MOS materials often suffer from low response, poor selectivity and insufficient reliability due to their surface or structure properties. In order to overcome these problems, researcher have made many efforts in designing and manufacturing heterostructures by depositing sensitizers on the MOS hosts. Noble metals and metal oxides are typical sensitizers. They can modify either the chemical or the electronic properties of the MOS supports to make the sensing performances optimized [13,14]. For example, numerous studies have reported improved sensitivity, selectivity or lower working temperature through the formation of heterostructures [15,16].
Since the surface structure and properties of MOS influence the gas sensing properties, a general and efficient method for the surface engineering of MOS is highly desirable. Atomic layer deposition (ALD) is a deposition technique basing on self-limiting surface reactions [17,18]. In this manner, the deposited layer is able to be controlled at the atomic level and the film thickness is determined by the number of ALD cycles [19,20]. Therefore, ALD is particularly suitable for the deposition of thin films in the nanometer range. For MOS sensors, when the film thickness is close to the Debye length of the material (λ d = √ ε0εKBT q 2 nc ), where ε 0 is the dielectric constant of free space, ε is the dielectric constant of the material, K B is Boltzmann constant, T is temperature, q is the charge of charged particles, and n c is the concentration of charge carriers [21,22], the entire film will be completely depleted of electrons. This will result in a most variation of the conductivity, therefore the sensor sensitivity will be significantly improved. In addition, due to the sequential selflimiting reactions, the deposited thin films are extremely conformal over large area and high aspect ratio nanostructures [21,[23][24][25]. Compared with other technologies such as sputtering and evaporation, ALD can achieve uniform coating on complex substrates and large areas. Moreover, ALD is also very effective in depositing metal catalysts, which can be uniformly loaded onto various sensor materials and improve the gas sensing performance [26]. Due to these unique characteristics, ALD is capable of not only synthesizing novel sensor materials, but also manufacturing gas sensor devices with high reproducibility and uniformity [27][28][29][30]. Materials deposited by ALD on a MOS host can be noble metals or MO, either as nanoparticles anchored on the MOS surface or as a conformal coating layer over the MOS structures. It is worth mentioning that ALD has recently been used to create single atom catalysts, which greatly develops the application perspective of materials including gas sensors [31,32].
In this review, we will discuss the recent progress of MOSbased heterostructures synthesized or modified by ALD. The utilization of ALD in the field of gas sensing allows to establish clear structure-property correlations in complex heterostructures. Since the principles of ALD and the chemical mechanism involved in the ALD process have been described elsewhere [33,34], here we will focus on how to optimize the gas sensing performance through the engineering of heterostructures by ALD. The main objectives are the improvement of the sensor selectivity, moisture resistance, and sensitivity. Finally, we will discuss the perspectives of ALD in the future development of practical sensors.

Principle of ALD
ALD is based on self-limiting surface reactions and is therefore quite different from most of the available thin film deposition technologies. The ALD precursors are introduced into the reactor sequentially, exposed and reacted on the substrate surface separately. As shown in figure 1(a), each ALD cycle includes four steps, including two precursors adsorbed on the substrate surface ((a1) and (a3)) to react with functional groups, and two purges ((a3) and (a4)) with typically high-purity N 2 which is used as purging gas to remove the reaction products and unreacted precursors [35]. Since after one ALD cycle only at most one monolayer can be deposited, the film thickness can be controlled at atomic level by the number of ALD cycles [17,33,[36][37][38]. Moreover, ALD is capable to coat high aspect ratio substrates conformally and can be easily applied to various types of substrates [36,37].
A general ALD process is performed within a suitable temperature range, which depends on the reactivity and stability of the precursors and surface functional groups. Temperatures outside this range lead to poor film quality. The condensation of precursor at low temperature, thermal decomposition and rapid desorption of surface reactive groups at high temperature will lead to non-ALD deposition [35]. In some particular cases, low temperature ALD might be required. For example, when the substrate is temperature sensitive like for most of flexible substrates. In these cases, a plasma source can be used to reduce the temperature of the ALD reaction without compromising the film quality ( figure 1(b)). The energy required for the surface reaction of conventional ALD comes entirely from the thermal energy of the substrate. On the other hand, in plasma enhanced ALD (PE-ALD), plasma species with high reactivity can promote the chemical reaction, making the required temperature of the substrate to be significantly lower than in thermal ALD. As a result, PE-ALD has developed into a highly attractive solution for nanodevices. For example, Hong et al [40] prepared staggered bottom-gate thin-film transistors using PE-ALD at low temperature (100 • C) by depositing 20 nm thick InO x layers on the polyethylene naphthalate substrate as channels.
Conventional ALD may suffer from low deposition rate [39,41]. To realize high throughput deposition on planar substrates, special ALD technology such as roll-to-roll ALD has been developed (figure 1(c)) [39]. In a typical ALD cycle, the time of reactant filling and gas purging is the factor restricting ALD efficiency, and the actual chemical reaction time in the ALD cycle is actually very short. Roll-to-roll ALD optimizes this process, saving the time needed for filling and purging mentioned above, making the deposition reaction a continuous process, so it can effectively improve the deposition rate and the production efficiency [42]. For example, Ali et al [43] used a roll-to-roll ALD process to grow uniform Al 2 O 3 thin films on polyethylene terephthalate (PET).

Sensing mechanism of MOS heterostructures
Heterostructures based on MOS have been comprehensively studied for gas sensors due to the synergistic effects that might arise from the combination of different materials [44]. ALD allows the preparation of heterogeneous structures by the deposition of either a particulate-like or a conformal film which depends on the nature of the substrate and of the ALD process. In addition, the morphology, crystallinity and uniformity of MOS films can be improved by designing the intermediate layer with high surface energy through ALD. For example, Wang et al [45] have found that the Al 2 O 3 and ZnO intermediate layers could facilitate the growth of conformal and crystalline TiO 2 films on CNTs. Figure 2 shows some representative MOS heterostructures synthesized by ALD, which have demonstrated improved sensor properties. The enhancement of the sensor performance are generally ascribed to the altered chemical or electronic properties originating from the heterostructures, such as the generation of more active sites and larger specific surface area [46,47], the spillover effect from noble metals [48][49][50], or the modulation of the electronic resistivity of the materials [51,52]. The morphology of the heterostructures may have an influence on the sensor properties. It has been reported that the confined space within hollow SnO 2 nanocoils after functionalization of ALD NiO enhances the gas adsorption on the surface and improves the sensing performance [30,53]. In air, oxygen chemisorbs onto the surface of MOS materials to form surface oxygen species (O 2 − , O − and O 2− ) [54,55]. An electron depletion layers forms on the surface of n-type MOS materials and a hole accumulation layers on the p-type materials, which causes the surface resistance of the materials to change. When the surface of the MOS becomes in contact with a target gas the surface resistance of the material changes due to the competition between the analyte and chemisorbed oxygen, thus the resistivity of the MOS layer is affected [56,57]. The electronic change in MOS materials is controlled by gas adsorption and redox reactions on the surface, leading to a modification of the resistance of the MOS [58]. Let us consider n-type MOS ZnO as an example [26,59]. In air, oxygen molecules are adsorbed on the surface of the material, and oxygen molecules capture free electrons from the conduction band of the material, resulting in an electron depletion layer as shown in figure 3(a). When the ZnO sensor is exposed to a reducing analyte such as trimethylamine (TMA), TMA molecules are oxidized by oxygen ions, and the reaction releases electrons back to the conduction band of ZnO. This process leads to the reduction of the thickness of the electron depletion layer and the resistance of the sensor.
Different MOS materials have different reactivity due to their unique chemical properties, surface reactivity and gas adsorption properties [60]. Similarly, different MOS nanostructures also have different responses and selectivity to gases due to the differences in porosity, crystallinity and surface morphology, for example. The receptive function in MOS can be improved by increasing adsorption and reactivity through formation of heterostructures. The formation of a junction also affects the transduction mechanism, i.e. electron transfer and charge accumulation in the material [44]. For example, a p-n junction produces an electron depletion layer around the interface between the two MOS [56,61]. These processes affect the electron exchange between gas and MOS surface and improve the response level of the sensor. Therefore, it is particularly crucial to design MOS materials with different heterostructures to enhance electron capture for the development of gas sensors. For example, when noble metals are loaded on the surface of MOS materials to form a heterostructure, the unique catalytic effect of noble metal atoms can regulate the gas adsorption, thus regulating the electronic properties of MOS surfaces and enhancing the gas sensitivity [62]. At the same time, the noble metal itself may undergo oxidation-reduction reaction with the gas on the surface of the MOS material, they result in improved gas selectivity or sensitivity. For example, when Pd are loaded on the surface of MOS materials, when they are exposed to H 2 , Pd reacts with hydrogen to form PdH x , while the lattice parameters of Pd nanoparticles will increase, resulting in the expansion and phase change of Pd particles [63][64][65]. When a MO/MOS heterostructure is formed between MOS materials, their different Fermi levels lead to electron transfer between the two materials. This process provides more electrons to the sensing process, resulting in a higher change in the resistance of the MOS materials when reacting with the target gas [66,67]. Furthermore, the modulation of atomic defects such as oxygen vacancies in MOS also proves to be effective to enhance the sensor responses [68]. Oxygen vacancies can serve as active sites to promote the adsorption of charged oxygen species [31]. ALD allows not only the regulation of the size and thickness of the sensing layers [4,29], but is also efficient in creating oxygen vacancies in MOS materials [69][70][71].

ALD of metal catalysts on MOS
Studies have proven that noble metal loading can enhance the sensitivity of gas sensors. For example, the modification of MOS with noble metal particles such as Ag, Au, Pt can enhance the adsorption of oxygen and target gases due to their catalytic spillover effect [29,59,72], thus improving the sensitivity of MOS materials. Some metals react with specific gas molecules, thus modifying the selectivity of MOS materials. For example, Rh has been shown to improve the moisture resistance of MOS materials [26,73] as Rh is easily react with water molecules to.
Chinh et al [74] synthesized ZnO thin films with different thicknesses on alumina substrate by ALD. The ZnO thin film with a film thickness of 37 nm exhibited the best gas sensing performance. The SEM morphology is shown in figure 4(a), and the thin film is uniform and compact. When the device was immersed in a colloidal gold solution, Au nanoparticles with a diameter of 10 nm were loaded on the surface of the thin film. As shown in figure 4(b) the response to 10 ppm NO at room temperature with the assistance of blue light was higher than without Au loading, and the response and recovery times were improved. The improved sensor properties originates from the catalytic effect of Au nanoparticles, which increases the active sites for the oxidation reaction of NO.
Lin et al [58] deposited Pt nanoparticles on SnO 2 nanowires by ALD as a catalyst to improve the gas sensing performance of SnO 2 nanowires. As shown in figure 4(c), after 200 ALD cycles, island-like Pt nanoparticles uniformly distributed on the nanowires were observed. As shown in the response curves to different concentrations of ethanol (figure 4(d)) the Pt-modified SnO 2 nanowires obtained after 200 ALD cycles show an ultra-high sensitivity of 8400 to 500 ppm ethanol at 200 o C, which is almost 7000 times that of the original SnO 2 nanowires. The high sensitivity is due to the Pt catalyst effect and the change of Schottky barrier junction at the Pt/SnO 2 interface.
Xu et al [75] prepared W 18 O 49 nanospheres by a hydrothermal method, and then loaded Pt nanoparticles on their surface by ALD. As shown in figure 5(a), Pt nanoparticles with an average size of 0.5 nm are uniformly distributed on  figure 5(c). The cross-section diameter of the ZnO nanowires is about 100 nm, and the Pd nanoparticles are about 6 nm in diameter (darker contrast), which are homogeneously distributed on the surface of the ZnO nanowires. As shown in figure 5(d), ZnO nanowires loaded with 100 Pd ALD cycles have good selectivity to H 2 , with a response to 50 ppm of H 2 close to 9, which is three times that the one of unmodified nanowires. The optimal response temperature is reduced to 200 • C by the Pd loading. The improvement of the sensing performance lies in the different work functions of Pd and ZnO. When contacting, electrons transfer to ZnO, which increases the depth of the electron depletion layer. When the sensor is exposed to H 2 , Pd is converted into PdH x , and some surface ZnO is reduced to Zn because of the relatively high sensing temperature.
Kim et al [77] prepared reticular SnO 2 nanowires by vaporliquid-solid (VLS) growth method, coated them with ZnO by ALD, and then modified them with Au nanoparticles by γray radiolysis. As shown in the TEM image of figure 6(a), SnO 2 -ZnO C-S nanowires have a rough surface with a diameter of about 220 nm, the thickness of ZnO shell is 85 nm and the diameter of Au nanoparticles is about 20 nm. As shown in figure 6(b), the response of the SnO 2 -ZnO-Au heterostructure to 100 ppb CO at 300 • C is 26.6, but there is no response in the unloaded original nanowire, which is attributed to the catalytic effect of the Au nanoparticles towards CO oxidation. Weber et al [78] used VLS growth method to grow ZnO nanowires, then used 40 ALD cycles to prepare BN films on the surface of the nanowires, and then used 100 ALD cycles to deposit highly dispersed Pd nanoparticles. The TEM image (figure 6(c)) shows that the diameter of ZnO nanowires is about 100 nm, the thickness of BN films is about 5 nm, and the diameter of the Pd nanoparticles is about 10 nm. The Pd/BN/ZnO nanowires show the highest response to H 2 , and the response to 50 ppm H 2 is five times that of the unloaded nanowires (figure 6(d)). The optimal sensing temperature is reduced to 200 • C, thanks to the catalytic effect of Pd and the resistance modulation of the ZnO nanowires.
Kondalkar et al [79] sputtered a ZnO seed layer on interdigital electrodes then prepared a ZnO nanorod array by simple solution method, deposited a 2 nm thick Al 2 O 3 layer by ALD at 200 • C, and loaded Pt nanoparticles by electron beam evaporation and photochemical deposition. SEM shows that the diameter of the nanorods is about 50 nm, forming a uniform and compact nano array (figure 6(e)). The Pt/Al 2 O 3 /ZnO heterostructures showed a 96.46% high response δR/R(%) to 200 ppm acetylene at 120 • C, and with a detection limit of 1 ppm. It is worth noting that the response of multicomponent Pt/Al 2 O 3 /ZnO to 20 ppm acetylene (figure 6(f)) changes only slightly in the relative humidity range of 20%-70%, which proves its excellent moisture resistance. Noble metal supported MOS heterostructures deposited by ALD for gas sensing application have been already widely investigated and in some cases they have made important contributions towards the development of high-performance sensors (cf table 1) [80].
In addition, noble metal single atom catalysts are attracting much attention. They can adjust the coordination environment around a single atom, and exhibit extremely high activity. The element utilization rate of single atom catalysts is very high, which is beneficial to preparation of gas sensors at a lower cost [31]. Single atom catalysis in gas sensing reactions has been reported in a few works. Xu et al synthesized SnO 2 ultra-thin film functionalized with single Pt atoms by ALD, and the film thickness was controlled to be equal to the Debye's length. In the HAADF-STEM image in figure 7(a), the existence of Pt single atom can be clearly seen on the SnO 2 film. The oxygen activation of a single-atom Pt catalyst and the synergistic    effect of oxygen vacancies in SnO 2 film made the sensor highly sensitive to triethylamine. As shown in figure 7(b), the response of the optimized sensor to 10 ppm TEA is increased six times [29]. It should be emphasized that ALD turned out to be very efficient in the deposition of single atom catalysts [32].

ALD MO on MOS
In addition to metal catalysts, metal oxides either as a conformal thin layer or nanoparticles have also been loaded on MOS sensing materials by ALD to obtain improved gas sensing performance. The synergistic effect and different binding energy of the two MOS materials, as well as the catalytic effect of the loading layer on the sensing layer, will optimize the gas sensing performance of the MOS sensing layer [81]. Lou et al [82] prepared Co 3 O 4 nanospheres by hydrothermal method, and then functionalized them by ALD to form Co 3 O 4 /NiO x heterostructures. As shown in figure 8(a), low coordination atoms are existing in the Co 3 O 4 /NiO x heterostructure, which facilitates the controllable adjustment of oxygen vacancy concentration and electronic and energy band structure of the material. Therefore, as shown in figure 8(b), compared with the pristine Co 3 O 4 , the sensor device based on the optimized Co 3 O 4 /NiO x heterostructure has significantly enhanced response to triethylamine, including a lower operating temperature, a higher sensitivity and a lower detection limit. In addition, Xie et al [83] designed a unique NiOfunctionalized macroporous In 2 O 3 film by ALD. The continuous film structure is shown in the SEM image in figure 8(c). The optimized sensor showed excellent gas sensing performance, and showed a very high response of 532.2-10 ppm NO 2 at a relatively low working temperature of 145 • C, which was 26 times higher than that of the pristine In 2 O 3 sensor ( figure 8(d)).
Ko et al [81] used the VLS method to grow highly reticulated SnO 2 nanowires in a tube furnace, and then used ALD to uniformly deposit V 2 O 5 nanoparticles on SnO 2 nanowires, which could easily aggregate into isolated island structures by forming supercooled droplets at a relatively low temperature. The test results show that the SnO 2 /V 2 O 5 heterostructure with 50 ALD cycles of V 2 O 5 exhibits the best performance. The shape of the nanoisland is rectangular (figure 9(a)), with an average size of several tens of nm. It can be seen in figure 9(b) that the response of SnO 2 /V 2 O 5 sensor to 5 ppm NO 2 is more than 50 times higher than that of the original SnO 2 nanowire. Two MOS materials with different work functions, SnO 2 and V 2 O 5 , contact each other to form a heterojunction, and the mutual diffusion of electron and hole carriers at the interface produces a depletion region, which effectively reduces the initial concentration of electrons, increases the response, modulates the conduction channel of SnO 2 and improves the sensing performance. Lou et al [84] synthesized SnO 2 nanosheets by hydrothermal method, and then deposited Fe 2 O 3 on the SnO 2 nanosheets by ALD to form SnO 2 /Fe 2 O 3 heterostructures. The SEM image is shown in figure 9(c). The thickness of the SnO 2 nanosheets is about 8-9 nm, and the SnO 2 /Fe 2 O 3 composite synthesized with 20 Fe 2 O 3 ALD cycles exhibits the best performance. As shown in figure 9(d), it shows the highest response to 4.5-20 ppm formaldehyde at 220 • C, which is better than for pristine SnO 2 nanosheets. Excellent sensing modulation lies in the potential barrier change caused by the electron transfer between SnO 2 and Fe 2 O 3 .
Yuan et al [85] synthesized highly ordered SnO 2 nanobowls on the surface of a MEMS substrate by sacrificial template method and modified them with ZnO ALD. The SEM image is shown in figure 10(a). The diameter of the SnO 2 nanobowls is about 660 nm and exhibit a uniform and conformal ZnO nanowire seed layer with a thickness of 20 nm. As shown in figure 10(b), the response of the SnO 2 /ZnO heterostructure to 1 ppm H 2 S at 250 • C is as high as 6.24, which is 2.6 times that of pristine SnO 2 , and exhibit fast response and recovery times of 14 and 39 s, respectively. The excellent sensing performance is ascribed to the increase of specific surface area caused by the formation of the heterojunction and homojunction, and the additional reaction between ZnO and H 2 S. Kei et al [86] used ALD to fabricate Al-doped ZnO (AZO) nanotube templated by tris(8-hydroxyquinoline) gal- lium nanowire. The TEM image in figure 10(c) shows that the thickness of the AZO nanotube after 400 ALD cycles is about 41 nm. In figure 10(d), it is seen that the AZO nanotube sensor exhibits good response to various flow rates of O 2 . The synthesis of ALD MOS functional heterostructures has developed rapidly in the field of gas sensors and made important contributions towards the improvement of the surface activity of MOS materials, the increase of the sensing area and the enhancement of the gas sensing performance.

C-S structure gas sensing material.
C-S nanomaterials are the most extensively used heterostructures in gas sensing. The core material is covered by a nano-sized thin shell, and interfaces are formed between the core and the shell materials, forming various heterojunctions. This results in the radial modulation of the electron depletion shell and electric field smearing effect [51,85,87,88], which directly affect the gas sensing performances. By ALD the thickness and conformality of the shell can be accurately controlled, so that the whole shell can be completely depleted of electrons, resulting in a large resistance modulation effect, thus improving the response level and sensitivity [4,[89][90][91].
Raza et al [92] synthesized a series of NiO/SnO 2 C-S heterostructure supported on hollow carbon nanofibers by ALD, and explained the electronic conduction of the C-S structures through a series of resistance elements ( figure 11). The response of the sensor will only be determined by the resistance change of the material surface facing the atmosphere. Only when the surface change extends to the whole heterostructure (i.e. the shell thickness ⩽ λ d ), the heterojunction can play an important role in the sensor response. In addition, a uniform and compact shell structure can be grown on the core material by ALD, which ensures the feasibility of manufacturing and the stability of the sensor [88,93]. The C-S heterostructures often have new chemical and electronic properties. At the interface of the C-S structure, band bending will occur and this would have an influence on the charge transport [77,94]. Many studies have proved the practical application of C-S heterogeneous materials in gas sensing. For example, Xu et al [21] synthesized ZnO nanorods as core materials by hydrothermal method at 140 • C for 12 h, and then used TiO 2 ALD to deposit shells with different thicknesses and form ZnO/TiO 2 C-S structures. Its morphology is shown in figure 12(a). ZnO nanorods are randomly distributed, with diameters ranging from 50 to 100 nm and average lengths of about 1 µm. The gas sensing response is the highest when the thickness of the TiO 2 layer is 6.4 nm. The ZnO/TiO 2 C-S heterostructure exhibits the best sensitivity to 5 ppm nbutanol at 200 • C, with a response value of 4, which is twice as large than that of pure ZnO. The improved sensing performance is due to the different work functions of the two MOS materials, which promotes electron transfer from TiO 2 to ZnO. The energy band shift at the radial heterojunction brings an effective barrier for the electrons and holes moving through the interface, which leads to a larger resistance change. Moreover, the thickness of the TiO 2 shell is close to the Debye thickness (6.5 nm) of TiO 2 , and TiO 2 shell is completely depleted. Kim et al [51] synthesized SnO 2 nanofibers as core materials by electrospinning, and then coated with Cu 2 O with different shell thicknesses (15-80 nm) by ALD. The morphology of the SnO 2 /Cu 2 O C-S fiber heterostructure is shown in figure 12(c). It exhibits a long and continuous fiber morphology with a cross-sectional diameter of about 50 nm. Cu 2 O shell uniformly covers the surface of the SnO 2 core, and the performance is the highest for a shell thickness of 30 nm. As shown in figure 12(d), the response of the SnO 2 /Cu 2 O heterostructure to 10 ppm CO at 300 • C is 5, which is four times than that of pristine SnO 2 nanofibers. The response and recovery times are also greatly improved. The resistance modulation caused by the p-n junction formed between the n-type SnO 2 core and the p-type Cu 2 O shell, together with the optimized Cu 2 O shell thickness, regulates the gas sensing performance. Zhao et al [95] prepared In 2 O 3 nanofibers as the core structure by electrospinning, and then deposited ZnO shells with different ALD cycles to form In 2 O 3 /ZnO hollow porous C-S structures. The TEM image is shown in figure 13(a). The average grain size of In 2 O 3 is 40 nm. The dependence of the gas sensing properties with the shell thickness was studied. The sensor loaded with 50 ZnO cycles has a response of 78.6-50 ppm NO 2 at a low working temperature of 200 • C ( figure 13(b)). It was also found that the sensor loaded with 300 ZnO cycles had excellent ethanol sensing response at 320 • C, and the selectivity of the sensor changed with different ZnO cycles. Raza et al [96] synthesized SnO 2 nanowires by gas-liquid-solid deposition. A NiO shell was deposited by ALD, forming a SnO 2 /NiO C-S structure. A TEM image is shown in figure 13(c). The diameter of the SnO 2 nanowires was 50-70 nm. The 100 NiO ALD cycles produced a polycrystalline shell of 4.1 nm in thickness. The response of four sensors with different number of NiO ALD cycles are shown in figure 13(d). At 500 • C the sensing response of SnO 2 /NiO-100 towards 500 ppm H 2 is 114, which is about four times higher than that of the pristine SnO 2 nanowires. The improvement of the response is ascribed to the formation of a p-n heterojunction at the interface of SnO 2 core and NiO shell, and the adjustment of the energy band structure. Table 2 summarizes the properties of some selected heterostructure materials synthesized by ALD during at least one of the preparation steps. The formation of heterostructures often leads to improved sensor properties, however only for optimized shell thicknesses and charge carrier types of the core and shell layers. For an exhaustive discussion we refer the reader to a recent publication on the role of heterojunctions on in C-S heterostructures [92]. The application of ALD to C-S MOS heterostructures by uniformly coating a core MOS material with another MOS shell material has now reached maturity, and several studies have shown clear progresses compared to state of the art sensors based on only one MOS [21,51,[95][96][97][98][99].  [95]. Copyright (2018) American Chemical Society. (c) TEM images of SnO 2 /NiO-100, (d) dynamic response of the sensors fabricated with pristine SnO 2 nanowires and the SnO 2 /NiO-X CS nanowires at 500 • C toward various concentrations of H 2 . Reprinted with permission from [96]. Copyright (2020) American Chemical Society.

Conclusion and perspectives
ALD has demonstrated great potential in fabricating MOSbased heterostructures for improved gas sensors. ALD is capable of growing complex and well-controlled heterostructures that cannot be easily obtained by other methods. On the one hand, the thickness of the MOS films, which is one of the crucial parameters influencing the response of the sensor, can be accurately controlled by the number of ALD cycles. On the other hand, the self-limiting reactions of the ALD process allows deposition of metals or metal oxides on various MOS sensing materials. The formed heterostructures can often deliver higher response, faster response and recovery speed, higher sensitivity and better selectivity due to the improved receptor and transduction functions. In addition, ALD is highly compatible with current nanofabrication techniques. For example, it is superior to other methods such as solution chemistry for the modification of MOS sensing layers with catalysts in a batch-production mode, greatly facilitating the fabrication of sensors devices at industrial scale. Sensors based on ALD will have a broad development space in the field of integrated sensing circuits. With the development of integrated circuit and MEMS, the sensor components are continuously miniaturized and the power consumption is constantly reduced. Accurate control of the growth of thin films at the atomic scale is needed to manufacture the required semiconductor and nanoscale devices. ALD allows an easy control of the sensing film at nanometer level to produce low-power and micro-sized sensors.
Although ALD hold a great promise for fabrication of nanosensors, some challenges remains to prevent this technology in sensors research. MOS thin films grown by conventional ALD processes often need post-annealing to improve the crystallinity and conductivity. Post-annealing may bring minimum influence if the sensor substrates used are inorganic such as glass, ceramics or silicon, but may damage organic substrates such as PET that are used in flexible devices. Flexible sensors are currently attracting increasing attention due to the bending and deformable functions that are required in wearable devices and electronic skins [100,101]. However, the development of flexible and wearable sensors requires new ALD processes compatible with temperature sensitive substrates. While recent reports have shown that ALD of MOS films on a flexible substrates are possible, real sensors from low temperature ALD has been not reported. Endeavors are still needed to explore the potential of ALD in designing flexible devices.