The effect of Ag doping on the structure, optical, and electronic properties of ZnO nanostructures deposited by atmospheric pressure MOCVD on Ag/Si substrates

Atmospheric pressure metal–organic chemical vapour deposition was used to synthesize Ag-containing ZnO nanostructures of different morphology on Si substrates coated by Ag. Ag from Ag/Si substrates and Ag from silver acetylacetonate after its decomposition were used as a catalyst for ZnO nanocrystal growth for deposition of ZnO nanostructures with different morphologies. We investigated the relation of the structural parameters and chemical composition probed by x-ray diffraction (XRD), x-ray photoelectron spectroscopy (XPS), and Raman spectroscopy with the photoluminescence (PL) properties and electron–phonon coupling (EPC) reflected in the multi-phonon Raman spectra. The Raman and PL spectra were studied at different powers of the exciting laser radiation (P exc). The spectral position and width of the phonon Raman peaks and the near bandgap PL band at low P exc are supposed to be determined by the structural quality of the surfaces/boundaries of the crystallites. The intensity of the near-bandgap and defect-related PL and the magnitude of the EPC are additionally affected by the dopant concentration. Because of the large crystallite size (>30 nm, determined from XRD), the effects of phonon or electron confinement are negligible in these nanostructures. The behaviour of the position and width of phonon and PL bands with increasing P exc indicates that the heat dissipation in the film, which is dependent on the nanostructure morphology and Ag content, plays an important role. In addition, the cytotoxicity of ZnO:Ag nanostructures was investigated by using monolayer cell cultures of epithelioid origin Madin-Darby bovine kidney and Madin-Darby canine kidney cells at a MTT assay revealing that the level of silver doping of ZnO nanostructures, their morphology, and geometric dimensions determine their toxic effects.


Introduction
ZnO has been attracting a steady interest of researchers and technologists for decades [1][2][3][4]. Its advantages are the wide bandgap (E g ≈ 3.4 eV) and large exciton binding energy (BE) (60 meV) that enable electro-and photoluminescence (PL) emission in the UV range at room temperature [5][6][7]. Intrinsic n-type doping and energy band levels make it suitable for substitution of indium tin oxide as transparent electrode material [2][3][4]8]. Besides, ZnO is considered as a candidate for photovoltaic applications in the near UV spectral region, for sensing [4,9,10], as well as a promising photocatalyst, because of its strong oxidation ability and photocatalytic activity comparable to the benchmark TiO 2 [11,12]. The properties of ZnO can be further tuned by numerous dopants [13][14][15][16].
The variety of perspective applications defines the need for different approaches to the synthesis of ZnO microand nanostructures [1][2][3][4][17][18][19][20][21], as well as the formation of its heterostructures with semiconductors or metal nanostructures [22][23][24]. Atmospheric pressure metal-organic chemical vapour deposition (APMOCVD) is one of the popular methods of fabrication of oxide thin films, micro-and nanostructures, including ZnO [25][26][27][28][29][30][31][32], because it is relatively simple and non-expensive and can provide high-quality material without using catalysts or prepatterned substrates [25]. The possibility to vary the substrates and other deposition parameters allows ZnO with various morphologies, structural, optical, and other properties to be obtained and investigated. Unlike most other reports, where smooth or textured films were obtained for ZnO by this growth technique [28][29][30][31]33], our synthesis results in very homogeneous arrays of nanorods (NRs) and nanowires (NWs) with tunable parameters [25,34,35]. Therefore, the properties of this type of ZnO nanostructures need to be investigated in detail. Furthermore, doping ZnO with Ag during the synthesis was very rarely studied yet for this growth technique [36]. Among different dopants used so far [31], silver is one of the most promising, since the determinant factor of Ag in the photocatalytic activity of doped ZnO nanostructures is the charge carrier recombination rate. Reducing this rate is the main challenge on the way to improving photocatalytic and antibacterial performance. In works [25,34], were demonstrated the influence of Ag on the microstructure as well as electrical and PL properties of Agdoped ZnO nanostructures grown by chemical vapour deposition from metal-organic compounds, and performed a preliminary study of their cytotoxicity [35].
In the present study, we focus on PL and vibrational Raman spectroscopy of Ag-containing nanostructures, performed at various powers of laser excitation and on establishing the relation between the structure, thermal, and photoluminescent properties. PL and Raman spectroscopy are among the most broadly used characterization tools for ZnO, as they provide information on the lattice structure, electronic states, and electron-phonon coupling (EPC) [22][23][24][37][38][39], and are complementary to the basic structural data obtained from x-ray diffraction (XRD), SEM, and x-ray photoelectron spectroscopy (XPS).

ZnO nanostructure growth
Ag-containing ZnO nanostructures (Ag-ZnO:Ag NS) with different Ag content were grown on polished Si substrates coated by an Ag layer by atmospheric pressure MOCVD. The details of the growth technique can be found elsewhere [25,26]. A 120 nm thick Ag layer was deposited on Si substrates by thermal evaporation and subject to a thermal annealing to induce island-like morphology needed for subsequent nucleation of the ZnO nanostructures [37]. Before loading this Ag/Si substrate into the quartz tube reactor for ZnO growth, it was cleaned in acetone, ethanol, and deionized water for 10 min each. After that, the substrates were dried in a nitrogen flow. Mixtures of zinc acetylacetonate (Zn(AcAc) 2 ) and silver acetylacetonate (Ag(AcAc) 2 ) precursors (Sigma Aldrich, purity better than 99%) with 1 wt. % or 10 wt. % of silver acetylacetonate in the mixtures of precursors were used for the growth of ZnO samples at substrate temperatures (T s ) of 400 • C or 500 • C. At temperatures higher than the Ag-Zn eutectic temperature (465 • C) the vapour-liquidsolid or Volmer Weber mechanism governs the growth of ZnO nanocrystals [40][41][42]. In our experiments, the role of a catalyst for ZnO nanocrystal growth is taken by (i) Ag from Ag/Si substrates as well as (ii) Ag from silver acetylacetonate (Ag(AcAc) 2 ) after its decomposition. The parameters of growth and characteristics of the obtained ZnO nanostructures are summarized in table 1.

Characterization
The crystal structure was investigated by XRD using a DRON-4 diffractometer, utilizing Cu-Kα radiation (λ = 0.1542 nm). The morphology and elemental composition of the samples were studied by scanning electron microscopy (SEM). For this purpose, a TESCAN MIRA3 LMU microscope (Czech Republic) equipped with Oxford Instruments energy dispersive and wavelength dispersive spectrometers for x-ray spectrometry and with a detector for electron back-scattered diffraction was used. The XPS spectra were measured with an UHV-Analysis-System (SPECS, Germany) with the residual pressure in the chamber equal to 9.4 × 10 -10 mbar using a x-ray Mg Kα source (E = 1253.6 eV) and a pass energy of 32.0 eV. The spectrometer energy scale was calibrated employing gold and copper etalons. In order to account for possible charging effects, the C1s signal of the hydrocarbon surface adsorbates was used as an additional reference, with its C1s peak position at 284.6 eV [43]. Raman and PL spectra were acquired at the room temperature. Raman spectra were excited using a 514.7 nm diode pumped solid state laser (cobolt) or the 325 nm line of a He-Cd laser and registered at a spectral resolution of about 2 cm −1 for visible and 5 cm −1 for UV excitation using a LabRam HR800 micro-Raman system (Horiba Scientific) equipped with a liquid nitrogen cooled charge-coupled device detector. The incident laser power under the microscope objective (100× for 514.7 nm and 40× for 325 nm) was 0.1 mW for 514.7 nm and 10-0.001 mW for 325 nm. PL spectra were excited with the 325 nm line under the same conditions as Raman spectra, but recorded with lower spectral resolution of about 10 cm −1 .
MTT-assays (3-(4,5-dimethylthiazol-2-yl)-2,5-diphenyl tetrazolium bromide) were used for the analysis of cell viability [44,45]. After 24 h of growth in the growth medium, monolayers of MDBK and MDCK cells in 96-multiwell plates were incubated with the medium at a dilution 1:10 and nondiluted. Control cells were incubated with fresh medium for 24, 48, and 72 h. A total of 20 µl of MTT solution (BioFroxx, Germany) was added into the wells, and cells were incubated at 37 • C and 5% CO 2 for 3-4 h, then the medium was removed and 150 µl of 96% ethanol was added. The plates were read using a Multiskan FC (Thermo Scientific, USA) with a 538 nm test wavelength. The percentage decrease of cell viability subject to the influence of the ZnO nanostructures was calculated by the following formula: % decrease of cell viability =100 − (A/B × 100), where A is the mean optical density of the studied samples at a certain concentration and B is the mean optical density of the control cell samples. Figure 1 shows SEM images of the three types of ZnO samples studied in this work. The different morphologies obtained at varied growth parameters are summarized in table 1 along with some of the parameters of the nanostructures obtained from energy-dispersive x-ray spectroscopy (EDX), XPS, and XRD measurements. In particular, well-defined ZnO NRs with diameters of about 40-60 nm were grown at T s = 400 • C and 1 wt.% Ag (sample #1) (figure 1(a)). Increasing the nominal Ag content to 10% at the same T s (sample #3) results in a significant increase of the NRs diameter, to 100-130 nm, and a tower-like shape of the nanorods (TNR) (figure 1(c)). Increasing the growth temperature to T s = 500 • C at the same doping 10 wt.% Ag (sample #2) leads to the formation of randomly oriented ZnO NWs with diameters in the range of 40-50 nm ( figure 1(b)). Therefore, we can conclude that not only the temperature of the substrate has a crucial effect on the morphology of ZnO nanostructures grown by APMOCVD, but also the concentration of the dopant. Note that obtaining three-dimensional nanostructures in the present work is a very unusual result for APMOCVD growth of ZnO, because in most other reports, where this growth technique was used, smooth or textured films were obtained [28][29][30][31]33]. Regarding the Ag doping, it has an influence not only on the morphology of the structures obtained, but also on their structural, thermal, and optical properties as shown below.

SEM and EDX results
The EDX measurements confirmed that all three samples contained no other elements except zinc, oxygen, silver, and carbon. The Zn/O ratio was obtained equal to 1.2 for sample #1 and 1.3 for samples #2 and #3 (table 1). As ZnO is known to form stoichiometrically even at much lower temperatures and ambient atmosphere [46], the large deviation from stoichiometry in our nanostructured films can be due to the contribution of zinc and oxygen from the residuals of the synthesis. The ZnO nanostructures themselves are supposed to be very stoichiometric, as concluded below based on the results of XRD and Raman spectroscopy. The minor amount of carbon found obviously is the result of the decomposition of acetylacetonate precursors or surface organic contaminations. Due to the large probing depth of EDX, the detected content of Ag (⩾4.0 at.% for all samples) is dominated by the Ag layer on Si substrate, therefore, XPS was used to determine the Ag content in the ZnO nanostructures as described below.
ZnO NRs, obtained in samples #1 and #3, have the same set of crystallographic planes, (100), (002), (101), (102), (103), (004), and (202), but have notably different intensities of XRD lines (figures 2(a) and (c)). Observing this set of diffraction peaks indicates a rather random orientation of rods in both samples, while for the ZnO NWs (#2), on the contrary, observing only intense (002) and (004) diffraction peaks (figure 2(b)) indicates texturing (preferred orientation) along the c-axis. The XRD peak parameters of the ZnO nanostructures calculated using Gaussian functions are summarized in table 2. The small magnitude of the full width at half maximum (FWHM) of the ZnO (002) diffraction peak of 0.22-0.25 degrees (table 2) confirms the high crystallinity of all ZnO nanostructures. The mean crystalline size was evaluated using the Scherrer equation L = K·λ/(β·cosθ) [48]. The value of the shape factor K was chosen as 0.9 as an appropriate value in the absence of detailed shape information. The calculation of crystalline size was carried out for the most intensive XRD (002) peak of the ZnO crystal lattice. The crystallite size is assumed to be the size of a coherently diffracting domain and it is not necessarily the same as the particle size [49]. Taking into account that lattice period c = 2·d (002) (where d (002) is lattice spacing of (002)) for hexagonal structure of ZnO [50], the lattice period c was calculated from Bragg's law: 2·d (002)· sinθ = λ (where λ = 0.1542 nm is the wavelength of Cu-Kα radiation and θ is the Bragg angle (half of the peak position angle)). There are no significant changes in the lattice constant c, which obviously indicates low Ag ion incorporation into the ZnO crystal lattice. Khranovskyy et al [51]

XPS characterization
The effect of Ag incorporation into the ZnO lattice on the chemical bonding and the electronic structure was studied by measuring XPS core-level and valence-band spectra. In the survey XPS spectra (figure 3(a)), the intense peaks of O 1s and Zn 2p core-level electrons are observed for all three samples. Weak XPS peaks at 367 and 373 eV belong to Ag 3d 5/2 and Ag 3d 3/2 core-level electrons [52], respectively, and are better seen in the high-resolution spectra (figure 3(d)). No other elements were detected, except for adventitious carbon (C 1s peak), i.e. common hydrocarbon species adsorbed on the surface of a sample stored in air between fabrication and XPS measurement.
The high-resolution core-level spectra of Zn 2p (figure 3(b)) reveal the Zn 2p 3/2 and Zn 2p 1/2 components at BEs of 1022.0 ± 0.05 eV and 1045.0 ± 0.05 eV, respectively, and correspond to the Zn 2+ state [53][54][55], in particular in the ZnO wurtzite structure [38,55,56]. The charge state of Zn atoms does not change with Ag content and substrate temperature. The increase of the nominal Ag concentration from 1 wt.% to 10 wt.% only leads to a minor increase of the FWHM, from 2.43 ± 0.05 eV to 2.51 ± 0.05 eV. This fact can be explained by the slight distortion of the tetrahedral environment of Zn 2+ ions by Ag ones [55]. The different intensities of XPS lines for different samples can be related with different morphology.
The XPS O 1s spectra (figure 3(c)) exhibit an asymmetric shape often observed for ZnO nanostructures and consist of a major component at 531.8 ± 0.1 eV, attributed to adsorbed hydroxyl groups [38,[54][55][56][57], and a shoulder around 530.1 ± 0.1 eV, attributed to O 2ions in the ZnO lattice, i.e. fully coordinated to Zn atoms [48,53,54]. It was shown earlier that the presence of hydroxyl groups plays a major role in increasing the photocatalytic activity due to preventing electron-hole recombination [58]. In our ZnO nanostructures, this component dominates and is apparently beneficiary for their cytotoxicity properties (discussed below). Upon increasing the nominal silver content from 1 to 10 wt.% this component shifts slightly downwards, from 531.8 ± 0.1 eV to 531.4 ± 0.1 eV, with a concomitant increase of FWHM, from 2.96 ± 0.1 eV to 3.5 ± 0.1 eV. A larger effect of Ag doping on the oxygen than on zinc spectra is not surprising, because of the expected Ag-O bonds formed.
The XPS spectra in the BE energy range of Ag 3d corelevel electrons are shown in figure 3(d). For the ZnO:1 wt.% Ag NR sample Ag 3d core-level emission is hardly detectable, obviously because the actual Ag content in the topmost surface layer, that is probed by XPS, is much smaller than the nominal value of 1% and is below the sensitivity of the XPS method. For the samples with 10 wt.% of Ag the characteristic spin-orbit split doublet of Ag 3d 5/2 and Ag 3d 3/2 is observed, with a BE of Ag 3d 5/2 electrons of 367.1 ± 0.1 eV and 366.8 ± 0.1 eV in the ZnO:10 wt.% Ag NW and ZnO:10 wt.% Ag TNR samples, respectively. The commonly reported value of BE of Ag 3d 5/2 for bulk metallic Ag is about 368.2 eV [52,59,60], while for Ag 2 O and AgO much lower values were typically reported, 367.3-367.7 eV [57,59]. Therefore, the BE values observed in our case of Ag 3d 5/2 , 367.1 eV and 366.8 eV, confirm that Ag ions are introduced into the oxide lattice of our ZnO nanostructures. The silver content in ZnO nanostructures calculated from the ratio of the integrated intensities of Ag 3d and Zn 2p peaks gives 3.1 at.% for sample #2 (ZnO:10 wt.% Ag NWs) and 5.6 at.% for sample #3 (ZnO:10 wt.% Ag TNR). Both these values are much lower in the topmost surface layers analysed by XPS than the nominal 10 wt.% Ag concentration added during the synthesis. Therefore, the absence of the Ag signal for sample #1 is likely due to the actual Ag content being also 2-3 times lower than the nominal one (1%) and thus below the sensitivity of the method. Based on the analysis of the ZnO:Ag XRD results and the Ag concentrations obtained from XPS we can suggest that Ag is partially incorporated into the ZnO lattice, and the Ag concentration in ZnO increases with increasing Ag content in the mixture of precursors, hence, we deal with Ag-containing ZnO (Ag-ZnO:Ag) NS.
The XPS valence band spectra are presented in figure 3(e). They are formed by O 2s, Zn 3d, and O 2p states [54,61] as marked in figure 3(e). The feature in the energy range of 2-8 eV below the Fermi level is formed by the O 2p states hybridizing with the Zn 4s and Zn 4p states.

Raman spectra
Raman spectra of the studied samples excited with resonant λ exc = 325 nm ( figure 4(a)) reveal the characteristic vibrational pattern of highly crystalline ZnO, consisting of the longitudinal optical (LO) (fundamental) peak at about 580 cm −1 and its overtones at about 1155 cm −1 (2LO) and 1738 cm −1 (3LO). The relatively high intensities of overtones along with the relatively small FWHM of the LO peak, 20 cm −1 , indicate that the crystallite size is in the range of tens of nanometres in agreement with XRD results. The distinct surface optical (SO) phonon feature around 470 cm −1 is a common attribute of the phonon spectra of crystalline semiconductor structures with a relatively large portion of surface [18,23,24]. The relative intensity of the LO phonon peak and its overtones (nLO, n = 1,2,3…) in the resonant Raman spectra at 325 nm excitation is known to be a qualitative measure of the electron phonon coupling (EPC) in polar semiconductors [23,62,63]. In addition, the lineshape of the nLO band and its frequency position with respect to the LO mode can indicate peculiarities of the EPC. In bulk ZnO crystals, due to the anisotropic short-range forces in the uniaxial lattice, the LO frequency varies from 574 cm −1 (E-phonon mode) to 583 cm −1 (A-phonon mode), depending on the orientation of the crystal with respect to the polarization of the electric field of the exciting light. In the ensembles of randomly orientated micro-or nanostructures this effect leads to a corresponding inhomogeneous broadening of the LO mode and its overtones [24]. Moreover, this results in an asymmetric line shape of the overtone features (figures 5(b) and (d)). Besides the aforementioned anisotropy factor specific for ZnO, the LO phonon peak width is determined by other factors that are common for all nanocrystalline materials: structural imperfections, strain, and phonon confinement. As the effect of phonon confinement induces measurable broadening only when one of the dimensions of the nanostructure is smaller than 10 nm [18,23,24,64], we can exclude this effect from consideration for the samples studied here. The effect of strain of free-standing ZnO nanostructures may only arise in the bending points of some of the NWs in sample #2 ( figure 1(b)) and therefore is not supposed to have any significant effect on the phonon peak position of the ensemble spectra.
From the dependence of the Raman spectra on P exc of different samples one can judge about qualitative differences in the thermal conductivity and relative thermal expansion coefficients of the components in the nanostructure. The downward shift and broadening of the phonon Raman peaks upon increasing laser power is a known temperature effect related with the increase of the lattice constant and the phonon anharmonicity [39]. The very first observation that can be made from comparing the spectra of samples #1-#3 at P exc = 1 mW and P exc = 10 mW is a noticeable enhancement of the SO mode of sample #3 at higher P exc ( figure 4(b)). This additional effect is discussed in more detail below. The dependence of the position and relative intensity of the LO peak and its overtones on P exc is presented in figure 5.
From figure 5(a), we can notice that at lower laser power the LO band of sample #1 is broader and slightly shifted downwards compared to that of samples #2 and #3. As the crystallite size in all three samples (>30 nm) is above the values, for which phonon confinement can occur (<10 nm), the difference in the phonon peak width is more likely determined by the structure/ordering of the boundaries/surfaces of the crystallites. The lower intensity of the 2LO and 3LO peaks for sample #1 ( figure 5(b)) compared to other two samples is also likely to be related with poorer interface/surface quality. The difference in crystallite size is unlikely to have an effect on the EPC strength (and thus on the overtone intensity), because neither electronic states nor phonons are affected by changes in the crystallite size above 30 nm [64]. At increased laser power, the LO peaks of samples #2 and #3 shift stronger downwards than that of sample #1 ( figure 5(a)). The stronger photo-induced heating of the samples #2 and #3 may be due to either (i) worse thermal contact of the ZnO nanostructures with the substrate; (ii) weaker coupling between electrons and acoustic phonons responsible for the heat dissipation in crystals; or (iii) effect of larger Ag doping. Option (i) is not very likely, because from the SEM images one may conclude that the NWs (sample #2) may indeed have poorer contact with the substrate, but for the rods of both samples #1 and #3 such a difference is not obvious. The larger content of Ag in samples #2 and #3 may be responsible for their stronger photoinduced heating compared with #1. Especially a pronounced heating effect is observed for sample #3, with the largest actual Ag content according to XPS, 6%. For the latter sample not only the LO lineshape is affected but a strong enhancement of the SO mode is observed at higher laser power ( figure 4(b)). Notably, the relative intensity of this mode reflects neither the different 'external' surface/volume ratio of nanostructures, revealed by SEM (figure 1) nor the different ordering/structure of the surface/boundaries of the crystallites, as judged from the LO lineshape above. These observations are in agreement with previous works, which found no unambiguous relation between the SO mode intensity and the size and morphology of ZnO or other nanostructures [18,23,24,64]. An important finding of this work is that the relative intensity of the SO band at photo-induced heating is sensitive to internal microscopic structure of ZnO nanostructures, in particular doping ( figure 4(b)).
By analysing the overtone-to-fundamental intensity ratio as a function of laser power one can deduce information about the dependence of the EPC on laser power and local heating. This fact can be well illustrated by comparing the Raman spectra in figures 5(a) and (b). At lower laser power (1 mW) the 2LO intensities of samples #2 and #3 are equal and higher than that of sample #1. As we do not expect electron or phonon confinement effects in these nanostructures to be noticeable for EPC, the 2LO/LO intensity ratio seems to be governed not by size but another factor in our structures. In particular, the larger intensity ratio in #2 and #3 can be due to larger concentration of the dopant, as compared to those in sample #1.
When the exciting laser power is increased by a factor of 10 (10 mW), the 2LO intensity (ratio to LO) gets lower for sample #1 and to lesser extent for #2, while some increase is observed for sample #3. Taking into account that samples #2 and #3 showed a larger downward shift of the LO peak at increased P exc ( figure 5(a)), indicating stronger heating, the P exc -dependence of the EPC coupling strength is affected by Ag doping not only via heating but also via changes in the electronic properties. Besides, the change in the relative intensity of the overtones can (partially) be due to improved resonance conditions of the band gap shifted by the temperature change. The inset shows the spectrum of sample #1 in a broad spectral range to demonstrate the presumable PL background (see text for details). Because of the strong PL background this spectrum was accumulated for 1 s, while for the other two spectra 30 s accumulation was used.
At non-resonant excitation with λ exc = 514.7 nm (figure 6), the common Raman pattern is observed for samples #2 and #3 consisting of E 2 -symmetry modes at 98 and 437 cm −1 [38,65,66], as well as their difference mode around 340 cm −1 [65,67], while no Raman features are detected for sample #1 ( figure 6). As we did not observe metallic Ag in XPS, we can exclude the formation of Ag NCs in #2 and #3 and a related enhancement of the ZnO Raman signal [68]. So far, there were no reports that silver in the form of dopant ions may lead to an enhancement of the (host) lattice modes. Just on the contrary, any local lattice disorder caused by the dopant would lead to a weakening of both E 2 modes [69]. A more likely effect is that the dopant reduces radiative recombination responsible for the PL at this λ exc , thus facilitating the observation of the Raman peaks. As the actual content of Ag in sample #1 is at least one order of magnitude lower than in samples #2 and #3, it may be not sufficient to reduce the PL intensity strongly enough to observe Raman peaks. The inset in figure 6 shows the spectrum of sample #1 in a broad spectral range to reveal the PL background excited at this λ exc = 514.7 nm. Because of the strong PL background, this spectrum was accumulated for 1 s, while for the other two spectra 30 s accumulation was used. Figure 7 shows the PL spectra of the samples obtained at above-bandgap excitation, λ exc = 325 nm, measured at low laser power (0.01 mW), which was proved not to induce any noticeable effect on the structure during the spectra acquisition. The spectra consist of a sharp UV band peaked at 375 nm and a broad visible emission centred around 550 nm. According to the previous works, the UV band is assigned to near bandgap PL (denoted here as NBPL), while the broad visible band is associated with recombination via deep defect levels (DPLs), presumably oxygen vacancies (V O ) on the NC surface [1,2,5,18,31,39]. In the studies of different ZnO nanostructures, the PL spectra were reported to have very different intensity ratios of the NBPL and DPL bands, up to the presence of only NBPL [7] or only DPL [18] bands. The intensity ratio of these two bands is determined by the concentration of radiative and non-radiative defects, the surface-tovolume ratio of the NC, as well as by the NC environment or matrix [70].

PL spectra
The red shift of the NBPL band maximum from sample #1 (375 nm) to #2 (377 nm) and to #3 (379 nm) cannot be explained by the confinement effect, because the size of crystallites in all three samples, 33-37 nm, is beyond the quantum confinement range. The residual strain is also not likely to be the governing effect, because from this view point the PL and Raman data would be contradictory. In particular, the #3 has higher phonon frequency (which means more compressive strain or less tensile strain) but its PL band is shifted to the red, which indicates the opposite assumption about strain. The effect of the dopant on the position or filling of the emitting state may therefore be assumed. The poorer surface/boundary in sample #1 (assumed from the Raman data) may explain also the weak overall PL intensity at the above-bandgap excitation with the λ exc = 325 nm, because of the higher probability of the photo-excited charge carriers to be trapped by non-radiative surface defects [18,24]. In the particular case of our Ag-ZnO:Ag nanostructures, one may assume that both the absolute overall intensity of PL and the NBPL/DPL intensity ratio are determined by the interplay of at least three factors: (i) quality of the surface/boundary between crystallites, (ii) concentration of point defects inside crystallites, and (iii) concentration of Ag dopant. In particular, the much larger Ag content in samples #2 and #3 (according to XPS) may be partially responsible for stronger overall PL in these samples as compared to sample #1. The underlying mechanism may be an elimination of non-radiative defects by Ag substitution [71].
It was shown that the intensity of light used for PL excitation (P exc ) has a significant effect on the intensity ratios of the NBPL and DPL [70]. In particular, in the latter work on colloidal ZnO NCs we observed the NBPL band intensity to increase notably with P exc , while the DPL band intensity remained unaffected. We explained this effect by the saturation of the trap states at an increased concentration of photogenerated charge carriers, resulting in a higher probability of interband recombination (resulting in NBPL band). We suggested that saturation of both non-radiative and radiative traps takes place, because the DPL intensity is not affected, which can be the case if the radiative traps involved in DPL are saturated already at low excitation powers. A similar effect of P exc is observed also in the present work-at low P exc the NBPL/DPL intensity is different for the three samples ( figure 7(a)), while at high laser power (10 mW) the spectra of all samples look very similar and are dominated by the NBPL band (figures 8(a)-(c)).
Along with intensity effects, increasing P exc up to 10 mW results in a (reversible) red shift of the NBPL band of samples  #2 and #3 (figures 8(e) and (f)), which can be explained by lattice expansion because of laser induced heating. Interestingly for sample #1 no shift of the near bandgap PL peak is observed even at one order of magnitude higher P exc (10 mW, figure 8(d)). This effect is in agreement with the weak effect of the laser power on the phonon frequencies in the Raman spectra of this sample, and may be related either with a better thermal contact of ZnO nanostructures with the substrate in sample #1 or by low concentration of Ag dopant in this sample.

Cytotoxicity study
The cytotoxicity of the obtained Ag-containing ZnO nanostructures on the living of MDBK and MDCK cells at different exposure times are shown in table 3. A pronounced cytotoxic effect was determined for the Ag-ZnO:Ag NS samples deposited with high silver content (sample #2 and sample #3) in MDBK cells and consists of 76%-83% after 24 h of exposure time without dilution. Opposite to the toxic influence of Ag-ZnO:Ag nanostructures on the living of MDBK cells, the toxic effect of Ag-ZnO:Ag NS on the living of MDCK cells was not observed. The cytotoxic effect was determined to a maximum of 39% for ZnO samples doped with the highest concentration of silver. Also, the difference between the toxicity of sample #2 and sample #3 can be explained by the difference in the size of nanostructures as well the specific surface areas as shown by SEM ( figure 1). Generally, the nanostructures having a smaller size, and thus a large surface-to-volume ratio, are chemically more active (e.g. in catalysis or photocorrosion). Besides, Agdoped ZnO nanostructures have a size-dependent biological effect, especially cytotoxicity [72]. Thus, size might play an important role in toxicity and in general for biological activity. For example, smaller nanoparticles are known to have higher abilities for cell invasion and disrupt cellular homeostasis, resulting in lysosomal and mitochondria damage and ultimately cell death [72,73]. Hence, sample #3 with a nanostructure diameter of 100-130 nm has a less cytotoxic effect on both cell lines, inhibition of mitochondrion activity was 69% on 72 h of exhibition. While sample #1 and sample #2 with diameters in the range of 40-60 nm and 40-50 nm were more toxic and inhibited mitochondrion activity from 74% to 88% (table 3). The cell membrane on its surface contains many thiol-containing molecules that are targets for the silver ions of Ag NPs, the interaction of which leads to lipid peroxidation, which increases its permeability. As a result of this process, the cytoplasmic contents of the cell leak and, accordingly, leads to its destruction. These processes correlate with the silver content contained in the sample, the strength of its bond with the surface, the morphology of its placement on the surface of the nanostructure, and the contact time. In addition, the results of the study show that the cytotoxic effect of highly reactive Ag + , which can elute from the surface into the solution at certain concentrations, can easily penetrate the cell membrane causing cell death, while these effects may differ depending on the way of surface formation [74][75][76]. In [35] an investigation of the toxic effects of pure ZnO films on the living of MDBK and MDCK cells was carried out. Hence, we concluded that the Ag-ZnO:Ag samples have about three times more higher toxic effect in comparison with undoped ZnO films. Therefore, the amount of silver impurity in ZnO nanostructures, their morphology, and geometric dimensions determine their toxic effects.

Conclusions
Ag-containing ZnO nanostructures of different morphology and Ag doping were synthesized on Si substrates coated by Ag by atmospheric pressure MOCVD. It was shown that both the growth temperature and Ag content influence the outer dimensions and shape of the ZnO nanostructures. The size of the crystallites determined from XRD is rather close for all samples, in the range of 33-37 nm, and beyond the values where confinement effects can occur for electrons (20 nm) and phonons (10 nm). Therefore, the differences in the phonon Raman and NBPL peak positions, as well their behaviour with increasing excitation laser power (P exc ) are attributed to the effect of different Ag content in the samples and quality of the surfaces/boundaries of the crystallites. In particular, we observe stronger heating of the nanostructures with higher Ag content. The 2LO/LO intensity ratio, reflecting the strength of the EPC, decreases at higher P exc in the sample with the lowest Ag content. However, the decrease of EPC strength becomes smaller and turns into an increase as the Ag content in the sample increases. From this trend we can conclude that the thermally induced decrease of the EPC strengths is offset by some other (presumably electronic) effect of the Ag doping that enhances EPC. The determined indicators of cytotoxicity of Ag-ZnO:Ag nanostructures reveal the dependence of this effect on the structure of the studied samples, the amount of silver doping, and the morphology of the ZnO nanostructures, and also show different sensitivity for MDBK and MDCK cell lines. It should be noted that Ag-ZnO:Ag nanostructures do not demonstrate a highly toxic effect, which is important for humanity in the industrial production of these materials.

Data availability statement
All data that support the findings of this study are included within the article (and any supplementary files).

Acknowledgments
This work was partially supported by the research projects of NAS of Ukraine 'The development of photocatalytic nanocomposites for viruses inactivation in the air' and 'Development of innovative photocatalytic nanostructured materials based on ZnO and TiO 2 ' (528/IPM-11/20).