Raman and X-ray Photoelectron Spectroscopic Study of Aqueous Thiol-Capped Ag-Zn-Sn-S Nanocrystals

The synthesis of (Cu,Ag)-Zn-Sn-S (CAZTS) and Ag-Zn-Sn-S (AZTS) nanocrystals (NCs) by means of “green” chemistry in aqueous solution and their detailed characterization by Raman spectroscopy and several complementary techniques are reported. Through a systematic variation of the nominal composition and quantification of the constituent elements in CAZTS and AZTS NCs by X-ray photoemission spectroscopy (XPS), we identified the vibrational Raman and IR fingerprints of both the main AZTS phase and secondary phases of Ag-Zn-S and Ag-Sn-S compounds. The formation of the secondary phases of Ag-S and Ag-Zn-S cannot be avoided entirely for this type of synthesis. The Ag-Zn-S phase, having its bandgap in near infrared range, is the reason for the non-monotonous dependence of the absorption edge of CAZTS NCs on the Ag content, with a trend to redshift even below the bandgaps of bulk AZTS and CZTS. The work function, electron affinity, and ionization potential of the AZTS NCs are derived using photoelectron spectroscopy measurements.


Introduction
Thin film solar cells based on Cu2ZnSn(S,Se)4 (CZTSSe) absorber layers have gained increasing attention due to their suitable absorption spectrum, high absorption coefficient (10 4 -10 5 ), and nontoxic earth-abundant components [1,2]. However, the progress with improving the cell efficiency has stopped at a value of about 13% since it has not been possible to further increase the open circuit voltage 3 . The latter problem is generally believed to originate from band tails caused by Cu-Zn antisite defects [4]. Partial substitution of Cu for Ag is supposed to be a promising approach to reduce the antisite defect density and thus the band tailing [5,6]. Theoretical calculations demonstrated that due to a substantially larger ionic radius of Ag + (1.14 Å) in comparison with Cu + (0.74 Å) or Zn 2+ (0.74 Å), the formation energy of the AgZn defect (at 0.2 eV above the valence band edge) is much larger than that of the CuZn defect (0.12 eV). Many reports indeed showed an improved lattice (cationic) order in (Cu1-xAgx)2ZnSn(S,Se)4 (CAZTS) or Ag2ZnSn(S,Se)4 (AZTS), as well as a concomitant increase of the photovoltaic device efficiency [5,7,8], although also adverse effects of the substitution were reported [9]. It should be noted that embedding a high Ag concentration into the (intrinsically p-type) CZTS alters the conductivity to n-type [8]. Besides, for AZTS, the stannite structure seems to be more favorable than the kesterite one that is more common for CZTS [10]. Unlike CZTS, AZTS NCs can be fluorescent [11]. Therefore, despite all presumable similarities between CZTS and AZTS, the Cu-for-Ag substitution brings both new opportunities and challenges.
One of the routes to obtaining a thin absorber layer, especially for flexible photovoltaics (PV), is the synthesis of nanocrystals (NCs) in a colloidal solution, with subsequent deposition of a NC layer on a (flexible) substrate by printing, spin-or spray-coating, or other techniques [1,12]. Due to cost-effective roll-to-roll processes, flexible PV devices have their advantages for their potential applications compared to PV cells on glass [13].
Raman spectroscopy has proved to be a generally accepted structural characterization technique for CZTS and many related compounds, including colloidal NCs [14][15][16][17][18][19][20][21][22][23][24][25][26][27]. Besides high sensitivity to secondary (impurity) phases [20,[28][29][30][31][32], Raman spectroscopy requires only tiny amounts of the material for analysis (unlike XRD) and no sample preparation (unlike TEM, which is in addition quite challenging for tiny aqueous NCs). Furthermore, it can probe NCs directly in the as-synthesized solutions. For these reasons, Raman spectroscopy permits a broad screening of the synthesis conditions of new NC compounds, particularly by a detailed variation of each of the components in the whole compositional range. Therefore, Raman spectroscopy was chosen in this work as a primary characterization technique of the CAZTS and AZTS NCs synthesized by a method that we recently proved to be successful for obtaining good quality CZTS NCs [33].
Most of the Raman studies of CAZTS and AZTS were performed on poly-/microcrystalline films [34][35][36][37][38][39], with only a few reports on colloidal NCs [9,11,40]. There is a noticeable discrepancy in the literature regarding the dependence of the phonon Raman peak position and width on the Ag content in CAZTS, the NC size, or non-stoichiometry [5,39]. With the growing number of reports on the synthesis and applications of colloidal AZTS NCs [6,9,11,33,40,41], the physics and chemistry of defects and structural phases in such NCs need to be explored.
In this work, a series of CAZTS and AZTS NCs samples were prepared in aqueous solutions under mild conditions and systematically investigated in detail by resonant Raman spectroscopy, supported by X-ray photoemission spectroscopy (XPS), X-ray diffraction (XRD), IR absorption on phonons and UV-vis absorption spectroscopies. The experimental results are supported by ab initio DFT calculations.
Basic AZTS NC synthesis. The AZTS colloids were prepared from a mixture of mercaptoacetate (MA) complexes of silver, tin, and zinc reacting with sodium sulfide in aqueous alkaline solutions in the presence of ambient air, at 22-24 °C, and under normal pressure. The synthesis procedure is similar to the CZTS synthesis reported earlier [33].
In a typical synthesis, to 3.0 mL deionized (DI) water 0.3 mL aqueous 0.5 M SnCl2 solution (containing 4 M NaOH), 3.0 mL aqueous 1.0 M MAA solution, 0.1 mL aqueous 5 M NH4OH, 0.15 mL aqueous 1.0 M Zn(CH3COO)2 solution, and 3.0 mL aqueous 0.1 M AgNO3 solution were added dropwise at intense stirring. Afterwards, 0.5 mL aqueous 1.0 M Na2S solution was added, and the final mixture was heated at 96-98 °C (in a boiling water bath) in cylindric glass vials without reflux for 5 min. The as-prepared colloidal AZTS solution was subjected to purification by precipitating NCs with 2-propanol (5.0 mL 2-propanol to 10.0 mL colloidal solution), centrifugation of the water/alcohol mixture for 3 min at 6000 rpm, collecting the precipitate and redissolving it in 2 mL of DI water.
Variation of the NC composition and stoichiometry. Several sample series differing in composition and ratios of the components were prepared to probe the effects of stoichiometry and Ag-to-Cu substitution on optical and vibrational properties, including a series with a varied Ag/Cu ratio (denoted as AgxCu1-xZnSnS4 or (ACZTS NCs) and AZTS series with a varied silver content (denoted further as AgxZnSnS4 NCs), a varied zinc content (Ag2ZnxSnS4 NCs), a varied tin content (Ag2ZnSnxS4 NCs), and a varied sulfur content (Ag2ZnSnSx NCs). In the AgxCu1-xZnSnS4 series the Ag-to-Cu ratio was varied keeping the total Ag+Cu amount constant, while in other series, the content of one of the components was varied, maintaining constant the ratios between all other components. Details of preparation procedures for these series can be found in the Supporting Information (SI) file.
In a separate series of colloids, one or several constituents were not introduced into the system to let possible binary (such as Ag2S) or ternary (Ag-Sn-S) phases form for the purpose of comparison. To check possible effects of Ostwald ripening and re-crystallization of AZTS NCs a series of solutions was prepared with different heating periods from 0 to 50 min maintaining all other parameters constant.
For the spectral studies, the freshly synthesized and purified samples were drop-casted on cleaned substrates, bare Si for Raman, Si or glass for XRD, Au-covered Si for XPS, and placed for drying in a desiccator under dynamic vacuum.
XRD patterns were taken with a Rigaku SmartLab X-ray diffractometer equipped with Ni filtered Cu Kα X-ray source. The measurements were performed in θ-2θ geometry with a step of 0.05° 2θ and acquisition speed of 5°/min. The sharp lines from the Si(100) substrate were identified and subtracted.
AFM images were acquired with an AFM 5500 from Keysight (Agilent). The AFM tip had a radius of 10 nm, and the Si cantilever a resonance frequency of 273 kHz. The samples were prepared by drop-casting of very diluted (to the order of 10 -5 M in terms of Ag concentration) colloids on a freshly cleaved mica surface and dried in a nitrogen stream at room temperature. No ultrasound treatment of the colloidal samples was performed. The NC size distribution charts were plotted based on AFM measurements of the height profiles for several hundred separate QDs. The AFM images were processed using the Gwyddion software (with "plane substraction" and "aligning rows" filters and zero leveling) and marked by using the edge-detection grain marking tool of Gwyddion and the distributions of mean height were plotted for each image.
Raman spectra were excited using 514.7 nm, 532 nm, 638 nm, and 785 nm solid state lasers or the 325 nm He-Cd laser line and registered at a spectral resolution of about 2 cm -1 for visible and 5 cm -1 for UV excitation using a LabRam HR800 or Xplora micro-Raman systems equipped with cooled CCD detectors. The incident laser power under the microscope objective (50) was in the range of 0.1 -0.01 mW.
XPS measurements were performed with an ESCALAB 250Xi X-ray Photoelectron Spectrometer Microprobe (Thermo Scientific) equipped with a monochromatic Al Kα (hν = 1486.68 eV) X-ray source. A pass energy of 200 eV was used for survey spectra, 40 eV for Auger spectra, and 20 eV for high-resolution core-level spectra (providing a spectral resolution of 0.5 eV). Spectra deconvolution and quantification were performed using the Avantage Data System (Thermo Scientific). The linearity of the energy scale was calibrated by the positions of the Fermi edge at 0.00 ± 0.05 eV, Au4f7/2 at 83.95 eV, Ag3d5/2 at 368.20 eV, and Cu2p3/2 at 932.60 eV measured on in situ cleaned metal surfaces. The NCs samples were measured with a built-in charge compensation system to prevent charging. Finally, the spectra were corrected to C1s sp 3 peak at 284.8 eV as the common internal standard for binding energy (BE) calibration [42].
UV-vis absorption spectra were registered using a Black Comet CXR-SR UV/vis/NIR spectrometer (StellarNet Inc.) equipped with miniature deuterium/halogen lamps as an excitation source and 100 μm slits in the range of 220−1100 nm. The spectra were recorded in standard 10.0 mm optical quartz cuvettes.
Lattice dynamics calculations. Ab initio linear response (density functional perturbation theory) lattice dynamics calculations of both CZTS and AZTS were performed within the generalized gradient approximation using the Perdew-Burke-Ernzerhof local functional [43] as implemented in the CASTEP code [44]. Norm-conserving pseudopotentials were used. Before performing calculations, the structures were relaxed while keeping lattice parameters fixed and equal to the experimentally determined ones so that forces on atoms in the equilibrium position did not exceed 2 meV/Å and the residual stress was below 0.01 GPa. A self-consistent field (SCF) tolerance better than 10 −a and a phonon SCF threshold of 10 −an were imposed. For the electronic ground state calculations, integration within the Brillouin zone was performed over a 3×3×4 Monkhorst-Pack grid [45] in reciprocal space. As it is known, such an approach is proven to provide reliable results for quaternary semiconducting materials with various crystallographic structures [46,47].

AgxCu1-xZnSnS4 (ACZTS) NC series
Based on our recent experience of successfully synthesizing CZTS NCs in water [33], we started a transition from the copper-based kesterite to the silver-based one through a series of mixed Cu1-xAgxZnSnS4 (ACZTS) NCs using the same set of synthesis parameters. By tuning the group-I cation content x from Cu to Ag, a series of NCs with the nominal composition AgxCu1-xZnSnS4 was obtained. Their Raman spectra at exc = 514.7 nm excitation are shown in Figure 1a. The spectrum of pure CZTS NCs is dominated by the main peak due to a A1 mode vibration at about 332 cm -1 and corresponding higher-order scattering features at 660 cm -1 and 990 cm -1 . This is in agreement with previous reports on CZTS NCs obtained by the same synthesis route and indicates the formation of 3-4 nm kesterite NCs of good crystallinity [33]. Another evidence that the observed phonon Raman spectrum is characteristic of a single quaternary ACZTS structure, rather than a combination of the spectra of ternary and/or binary secondary phases, is the observation of higher-order scattering features ( Figure 1a). Noteworthy is the fact that the kesterite structure of ACZTS is distinctly preserved despite a pronounced Sn deficit ( Figure 1b).
With an increase of the nominal Ag content in the NCs, the main phonon peak gets broader and weaker, while bands at about 210 and 260 cm -1 arise, with comparable intensity to the A mode in the pure AZTS sample (Figure 1a). An additional feature about 160 cm -1 can also be distinguished for AZTS samples. The actual Ag/Cu ratio in the whole ACZTS NC series was confirmed by XPS ( Figure 1b); therefore, the evolution of the Raman spectra should be explained by the increased amount of Ag entering the CZTS NC lattice. Weakening of the main kesterite mode intensity (at 332 cm -1 ) without its noticeable shift and broadening (except for pure AZTS sample) can indicate a smaller volume of the kesterite phase formed with increased Ag content, while the simultaneous enhancement of new features at about 210 and 260 cm -1 can be indication of impurity (secondary) phases being formed. It is most straightforward to consider first the compounds that contain Ag, i.e. Ag-S, Ag-Zn-S, and Ag-Sn-S. Silver sulfide has Raman modes at 200 and 220 cm -1 [48], thus providing the lowest-frequency feature in our spectra, centered at about 210 cm -1 , but no mode around 260 cm -1 . To the best of our knowledge, for Ag-Sn-S only Ag6SnS8 is known as a stable phase at room temperature and normal pressure [49]. However, the only reported Raman spectrum of this compound is most likely just a combination of Ag2S and SnS2 modes [49], while for Ag-Zn-S no report on phonon Raman spectra could be found. Therefore, to establish the vibrational Raman fingerprint of the Ag-S, Ag-Zn-S, and Ag-Sn-S compounds, we synthesized and studied Raman spectra of Agcontaining ternary phases and Ag-S, as discussed later in this manuscript.

Heating of the NCs in solution.
In order to check whether the weakening of the main kesterite peak and odd features at lower frequencies in the spectrum of AZTS NCs (Figure 1a) can be influenced by postsynthesis thermal treatment, the as-synthesized AZTS NC solution was subjected to heating at 96-98 C for 5 to 50 minutes. This led to only a slight narrowing of the Raman peak without a noticeable change of its peak position, even after 50 min treatment (Figure 2a). At the same time, the XRD pattern changes already after 5 min, and after 50 min the reflexes become very sharp, with several new features appearing (Figure 2b). The transformation of the XRD patterns due to crystallization of initially amorphous (or poorly crystalline) NCs is not likely, because the as-synthesized NCs are already crystalline, according to TEM data (Figure 2b, inset). The size distributions of AZTS NCs were probed by atomic force microscopy (AFM). In these experiments, the original purified AZTS NC colloids were diluted with DI water by a factor of 200-300, drop-casted on a freshly cleaved mica and dried in a nitrogen flow. Figure  3a shows a typical AFM image of AZTS NCs synthesized with 5 min heating. The sample contains 3-5 nm NCs along with occasional NC aggregates as large as 20 nm. The NC size distributions were produced using Gwyddion software for 5 different areas and averaged, resulting in a symmetric distribution with a center at 3 nm (Figure 3b). Additionally, an exemplary image in Figure S1 with profiles shows the presence of 3-5 nm AZTS NCs as well as NC aggregates clearly evidencing clustering of separate NCs. We found that the variation of the thermal treatment duration from 5 to 50 min does not affect the size distribution of AZTS NCs noticeably. The fairly weak effect of the thermal treatment of the NC solution on the Raman spectra may indicate the fact that the Raman bandwidth is dominated not by the lattice crystallinity but by other factors, which are not determinant for forming the XRD pattern and TEM images, in particular by a short phonon lifetime due to the spatial variation of the nearest neighbor interaction caused by an inherent cationic disorder. It is well-reported that the frequency and FWHM of the main Raman peak of a broad range of multinary metal chalcogenides (CZTS, Cu-In-S, Ag-In-Se, Cu-Sn-S, CuxS) is rather insensitive to the sample non-stoichiometry, because it is mainly due to the vibration of Sulphur atoms with cations being immobile [18,24,[50][51][52][53][54]. Based on the Raman and XRD data in Figures 1 and 2, we can assume that at a small extent of the Cu-for-Ag substitution (below 15-20 %), the kesterite structure is well preserved. At larger Ag content, secondary phase segregation becomes significant, deteriorating the quality of the main (ACZTS) phase. Moderate thermal heating can improve the crystallite structure of the ACZTS NCs as seen by XRD but has only a minor effect on the vibrational (phonon) spectra probed by Raman spectroscopy. The latter is known to be more sensitive than XRD to the arrangement of the cation sublattice of chalcogenide compounds [14,16,27].
The results presented above allowed us to conclude that the synthesis conditions adopted from the protocol optimized for CZTS NCs [33] might not be optimal for ACZTS and AZTS NCs. Therefore, we synthesized and investigated several series of AZTS NCs with varying nominal compositions of each of the four constituents. The results are analyzed in the following subsections.
Variation of silver content -AgxZnSnS4 NCs.  The spectrum at xAg=0 coincides with the Zn-Sn-S NC spectrum reported by us for the first time recently [29]. With an increase of the nominal Ag content in AZTS NCs, the features at 210 and 260 cm-1 increase in intensity with respect to the 336 cm -1 one. Note that according to XPS data, with an increase of the nominal Ag content, the real Zn content decreases from 37 to 23 % and the Sn content is constant and low (< 5%) (Figure 5a).

Variation of zinc content -Ag2ZnxSnS4 NCs.
The Raman spectrum of the sample with xZn=0 is likely to be characteristic for a certain Sn-poor Ag-Sn-S structure, in particular Ag6SnS8 [49], because it cannot be attributed to a combination of Ag-S and Sn-S phases, and no other Ag-Sn-S stable phase at room temperature and normal pressure has been reported so far, to the best of our knowledge. The Raman spectrum reported in [49] looks more like a combination of Ag2S and SnS2 modes. Therefore, the present work is the first that reports a characteristic Raman spectrum of an Ag-Sn-S compound. It should be noted that the Ag-Sn-S phase is relatively stable, based on the Raman spectrum (Figure 4b), because it does not decay into Ag-S-and Sn-S-type Raman spectra. The characteristic spectra of Ag-S, SnS, and SnS2 are analyzed for comparison later in this manuscript (Figure 7).
The changes in the spectra of the Zn-series are most likely not (only) due to an increase of Zn content but (also) due to a concomitant drastic drop of the Ag and Sn contents ( Figure  5b). According to XPS data, the sample with a nominal content of xZn=15 should contain almost pure ZnS. However, the Zn-S peak in the UV Raman spectra is nearly of the same intensity in the rest of the samples in the series (Figure S2-S3). Therefore, segregation into Ag-S and ZnS is not likely to occur here, and we can attribute the observed Raman spectra to a certain Ag-Zn-S structure, as discussed in the next section.

Variation of tin content -Ag2ZnSnxS4 NCs.
From the Raman spectra of the Sn-series (Figure 6a) it is obvious that the main feature at 340 cm -1 gains in intensity with increasing Sn content, while the intensity of the features at about 210 and 260 cm -1 remain unchanged. Raman spectra of the Ag-Zn-S and Ag-S NCs at exc=638 nm; the Ag-Zn-S NC spectrum is also presented (thick red curve) normalized to that of the Ag-S, in order to underline the different lineshapes for two compounds. The spectrum of ZnS NCs is shown for comparison.
Therefore, the latter features can originate from Ag-Zn-S, because the spectrum of ZnS at this exc looks different (Figure 7a) and Ag2S has either a single peak at 200 cm -1 (at red excitation) or peaks at 200 and 220 cm -1 (at blue-green excitation) [48]. There has been no Raman spectrum of the Ag-Zn-S compound reported so far to the best of our knowledge. Therefore, it is crucial to identify the inherent spectrum of Ag-Zn-S and distinguish it from the Ag-S one. This will allow the latter two phases to be distinguished in the spectra of AZTS NCs. For this purpose, we measured Raman spectra of Ag-Zn-S and Ag-S NC samples at different exc ( Figure S4) and detected a pronounced difference between the two spectra at exc=638 nm (Figure 6b). Moreover, the intensity in the Raman spectrum of Ag-Zn-S is stronger than that of the Ag-S NCs at any excitation ( Figure S4), providing additional proof that our Ag-Zn-S NCs are not a combination of Ag-S and ZnS phases but possess their own, intrinsic electronic and vibrational structure. This conclusion corroborates the different spectral line shapes of the Ag-Zn-S and Ag-S NCs samples. Furthermore, the spectrum of Ag-Zn-S NCs is not a combination of Ag-S and ZnS spectra because the relatively sharp peak of ZnS NCs is unlikely to account for a broad feature in the Ag-Zn-S spectrum. The spectrum of Ag-Zn-S NCs is most likely a combination of Ag-S vibrations (resulting in the 210 cm -1 feature) and Zn-S vibrations (resulting in the 260 cm -1 feature) of a joint/alloyed ternary compound/lattice. Earlier we observed Zn-S vibrations in the alloyed NCs of other multinary chalcogenides: CuInS2-ZnS [21,22,55], Cu-Zn-In-S [55], and Zn-In-S [56]. The spectrum of ZnS NCs at UV excitation (325 nm) reveals a sharp first-order LO phonon peak at 345 cm -1 and a relatively strong second-order feature at 790 cm -1 (Figure 7b), which is also distinct from the spectrum of Ag-Zn-S NCs at this excitation showing only a weak first-order LO peak (Figure 7b). Note that the reference Raman spectrum of ZnS NCs shown in Figs. 6b and 7 was measured on the sample consisting solely of ZnS NCs of relatively high crystallinity. Comparing the signal-tonoise ratio of the phonon peak in the latter reference spectrum and the Ag-Zn-S spectrum at exc=325 nm, one can hardly expect any noticeable contribution of a minor content of pure ZnS phase (concluded from UV Raman) to the spectrum of Ag-Zn-S NCs at visible excitation (Figure 6b).
It should be noted that in the above-discussed series with the varied nominal Sn content, the real Sn content varies not very much and does not exceed 5 % in the whole series. In contrast, the Zn content decreases significantly with an increase of the nominal Sn composition (Figure 5c). Therefore, the evolution of the Raman spectra and of the underlying NC structure is most likely determined by the Zn deficit at larger nominal Sn composition. This allows the conditions for AZTS phase formation to be able to compete with the formation of Ag-Zn-S. The variation of real Zn composition in the Sn-series explains the similarity of the evolution of the Raman spectra to that of the Zn-series (Figure 4b).

Variation of sulfur content -Ag2ZnSnSx NCs.
It was expected that predominantly silver sulfide or silver zinc sulfide would form at a low sulfur content, while an increase in the sulfur content would favor the formation of the kesterite phase. In line with these expectations, we observe that at low nominal S content the Ag-Zn-S features dominate the Raman spectra, while the AZTS feature at 336 cm -1 builds up with an increase of the nominal S content ( Figure S5). This behavior corroborates the chemist's expectations in the previous paragraph. Furthermore, the increase of the nominal S content is accompanied by an increase of the real Sn content (from 0 up to 10%) and a decrease of the real Ag content (Figure5d), which is in agreement with enhancing the AZTS feature in the Raman spectra.

Resonance effects in Raman spectra of AZTS NCs.
Using Raman spectroscopy with different exc has proven to be a powerful tool for selectively probing both compositional and structural heterogeneities of small colloidal NCs [20,[28][29][30][31][32]57]. In the previous subsections (Figure 6b and Figure 7), we already discussed some of the resonant effects and referred to Raman spectra at different exc ( Figure S2-S3). Here, we would like to complete this discussion by considering the evolution of a representative AZTS NC spectrum (Figure 8) with exc tuned from UV (325 nm) to IR (785 nm) and discuss two additional effects not mentioned above. The first effect that can be concluded from Figure 8 is that the relative intensity of the features in the range of Ag-S and Ag-Zn-S modes, i.e. at 220 and 260 cm -1 , gets stronger at red and IR excitation. Based on the fact that at visible exc Ag-Zn-S has stronger Raman peaks than Ag-S (Figure 6b), and at exc =785 nm there is hardly any Raman feature seen in the Ag-S spectrum ( Figure S4), the resonant behavior in Figure 8 can be attributed to the increased contribution of Ag-Zn-S, but not of Ag-S.
The second effect is that similarly to previous reports on other multicomponent sulfides [55,58], Raman spectra excited with the 325 nm laser line selectively detect minor contents of the ZnS phase in AZTS NC samples. It can be distinguished from the spectral feature of AZTS, which occurs in the same frequency range, by observing the second-order (2LO) feature at 790 cm -1 for ZnS NCs (Figure 7b).

IR phonon spectra.
Infrared absorption by phonons is a complementary technique to phonon Raman scattering. It provides useful structural information also on multinary and hetero-NCs [59][60][61][62] , although it is still rarely used compared to Raman spectroscopy [57]. IR absorption does not have the advantage of Raman spectroscopy of selective probing different compounds in the NCs. Instead, being a first-order optical process, it profits from higher sensitivity (the probability of the absorption of an IR photon by lattice phonon is orders of magnitude higher than the scattering of a visible light photon by the same phonon). Moreover, the low spectral density of the exciting light in IR spectroscopy excludes any photoinduced processes in the NCs, including photocorrosion or photooxidation. Only few publications exploring IR absorption spectroscopy by phonons can be found for the entire broad family of I2-II-IV-VI6 compounds, including no reports on AZTS and only two works on CZTS [63,64] . No reports of the IR phonon spectra of the related Ag-based ternary compounds could be found as well.
Representative IR phonon spectra of several selected NС samples from this work are presented in Figure 9. As in case of the Raman spectra, one can establish a characteristic spectral lineshape of the AZTS compound, namely a sharp single band at about 360 cm -1 and a broad, apparently multicomponent, feature in the range of 150-300 cm -1 . The former vibration is absent in the samples that did not show the characteristic AZTS peak 330 cm -1 in the Raman spectra. Apparently, the vibrational IR pattern of the Ag-Zn-S structure encompasses the broad spectral range of about 170-280 cm -1 , similarly to the Raman spectra. The spectrum of the Ag-Sn-S NCs (Ag30Sn15S25) is similar to that of the AZTS NCs, except for the much weaker absorption component about 280-300 cm -1 in the former. Therefore, the absorption around 280 cm -1 may be related to vibrations strongly involving Zn atoms, while the sharp peak at 360 cm -1 most likely invokes Sn, because it is absent in Ag-Zn-S. However, more Ag in the lattice enhances this vibration. The latter assumption is supported by the absence of the latter strong feature in the spectrum of CZTS NCs (Figure 9).

DFT calculations.
Our ab initio DFT calculations of the phonon spectrum of kesterite AZTS show the main phonon (A1 mode) frequency to be by 14 cm -1 higher than that of the kesterite CZTS ( Figure  10). In the experimental Raman spectra ( Figure 10) one can see only a slight upward shift of the main phonon peak with increased Ag content. This result is in agreement with some literature reports [5,65], although a downward shift was reported by others [9,39,66]. What both groups of reports have in common is the small difference in the peak position compared to the CZTS one. This can be explained by the fact that the most intense, A1 mode peak is due to a symmetric vibration of the anion without involving any cation motion. The modes that involve cation vibration are not resolved in the present work, but in studies where it was observed as a separate peak, it exhibited a continuous shift toward the low energy side with increasing Ag content, well reflecting the effect of the substitution of Cu + by the heavier and bigger Ag + [65]. Consequently, most of the works on ACZTS nano-or microcrystals reported the main Raman peak in the same frequency range as for CZTS ones, namely 332-338 cm -1 [36,40].

UV-vis absorption.
The UV-vis absorption spectra are one of the key characteristics of an absorber material for photovoltaic applications. Even though the optical bandgap of AZTS and CZTS is due to direct interband transitions, the absorption edge of these materials, especially of NCs, is always much "smeared". The ACZTS NCs studied here also possess the same typical lineshape ( Figure 11). The absorption spectra were plotted in (D*hv) 2 vs. h coordinates to determine the direct bandgaps from the intercept of the linear part with the h-axes. Besides the expected transitions a noticable absorption was observed in the range h<EG. Such a "tail" originates from the contribution of sub-bandgap states or/and bandgap fluctuation due to chemical composition variations and structural disorder at the nanoscale. This lineshape is usually explained by an Urbach tail formed by a high density of gap states, which are present even in the samples with high crystallinity in terms of XRD or TEM data [67]. The dominance of the gap/defect states in determining the absorption onset of CZTS and AZTS NC samples may account for the scatter of the literature results and nonmonotonous behavior of the bandgap value as a function of AZTS NCs size or Ag content in the CAZTS. In particular, the Eg of 1.48-1.65 eV was reported for highly crystalline 5-7 nm AZTS NCs in [11], which is 0.5 eV below the bandgap of bulk Ag2ZnSnS4, 2.0 eV [10]. Note that both the Raman and XRD peaks of the latter NCs were as broad as those of the NCs in the present work, indicating that it is not purely the degree of an "extended" crystallinity of the NCs that governs XRD and Raman peak broadening. The cationic disorder that causes the absorption (Urbach) tail can lead to an apparent red shift of the absorption edge. No significant change or non-monotonous variation of the bandgap value with Ag content in the CAZTS NCs, up to 20% or even more, has been reported in other works [10,66,68,69]. For the spectra of ACZTS NCs of the present study, an increase of the absorption in the NIR tail with an increase of the Ag content may also reflect the enhancements of the structural disorder of the ACZTS phase. However, the main reason of the non-monotonous behavior of the absorption edge with Ag content in ACZTS is the interplay of two competing factors: (i) the bandgap increase due to Ag entering the Cu sites in the kesterite lattice of CZTS, and (ii) the red shift of the absorption related with an increasing portion of the Ag-Zn-S phase (as confirmed by Raman spectroscopy, Figure 1), because the absorption edge of pure Ag-Zn-S NCs is at longer wavelengths than that of AZTS or CZTS ( Figure 11). These results also reveal the reason why the presumable contribution of the Ag-Zn-S phase to the Raman spectra, i.e. the features at 210 and 260 cm -1 , gets stronger at red/IR excitation (see Figure 8).

Conclusions
Owing to the coupling of the vibrational and optical properties in the resonant Raman scattering phenomenon, unique information about the structural composition and possible heterogeneity can be obtained even for ultrasmall and non-stoichiometric NCs of multinary compounds. In particular, one can conclude from the current study that the structure of the (Cu,Ag)-Zn-Sn-S and Ag-Zn-Sn-S NCs obtained by a facile and scalable aqueous synthesis in this work is generally a combination of the main quaternary (AZTS or ACZTS) and secondary Ag-Zn-S phases. The formation of the latter ternary compound is facilitated by an inherent Sn-deficiency unavoidable for the synthesis in water. Based on specific resonant behavior of the phonon Raman spectra, one may assume that the two phases are not completely isolated from each other but have coupled electronic structures. However, the frequency position of the main kesterite feature at about 336 cm -1 is not sensitive to the degree of AZTS NCs non-stoichiometry or amount of Ag-Zn-S phase in them. In the Raman spectra, the AZTS phase can be reliably distinguished from the main impurity phases due to significant differences in the frequency of characteristic vibrational fingerprints of Ag-Zn-S (210 and 260 cm -1 ) and Zn-Sn-S (344 and 380 cm -1 ). The identification of the Ag-Zn-S phase allows explaining consistently the non-monotonous behavior of the bandgap of CAZTS NCs with Ag/Cu content. An alternative explanation of similar anomalies in the UV-vis spectra of the CAZTS system is thus suggested, which was earlier most frequently attributed to the effect of Urbach tail due to structural disorder. The Raman and IR vibrational fingerprints of Ag-Zn-S and Ag-Sn-S, established in this work, along with the detected resonant behavior of the former phase, will facilitate the compositional analysis of AZTS and related complex compounds in the future.