Solution Processed Single-Phase Cu2SnS3 films: Structure and Photovoltaic Performance.

High quality microcrystalline tetragonal p-Cu 2 SnS 3 (CTS) ﬁ lms are prepared by spin-coating a single precursor of metal salts and thiourea on to 5 (cid:1) 5 cm 2 Mo substrates. The best of the solar cells completed on these ﬁ lms with a substrate con ﬁ guration: glass/Mo/CTS/CdS/i-ZnO/Al:ZnO/Ni – Al and a total area of 0.5 cm 2 displays an open-circuit voltage of 206 mV, a ﬁ ll factor of 34.5%, a short circuit current density of 27.6 mA cm (cid:3) 2 and a power conversion e ﬃ ciency of 1.9% under simulated AM1.5 illumination. This the best performance reported for such solar architectures obtained by solution processing, with dispersion below 20% for 24 devices. For the ﬁ rst time, the key factors limiting CTS device e ﬃ ciency are quantitatively established based on temperature dependent current – voltage curves and impedance spectroscopy, namely: (i) carrier recombination at the CTS/CdS interface, (ii) MoS 2 non-ohmic back contact, formed due to sulfurization of the top Mo layer, with a barrier height of up to 216 (cid:4) 14 meV and (iii) the presence of two trap levels with activation energies 41 (cid:4) 0.4 meV and 206 (cid:4) 7 meV. The shallower trap is linked Cu vacancies, while the deeper trap is associated with Sn in Cu antisite defects based on DFT supercell calculations.


Introduction
Cu 2 S absorbers initiated the eld of thin lm photovoltaics (PVs) with power conversion efficiencies up to 10%, although the cell performance rapidly degraded due to the migration of labile Cu vacancies. 1The eld evolved towards chalcopyrite structures with the inclusion of elements such as In and Ga which signicantly increase the stability of Cu-S phases. 2 Recently, Cu(In,Ga)Se 2 solar cells have achieved a record 22.6% efficiency, exceeding those of single-junction Si devices. 3owever, In and Ga are rare and expensive elements which can limit the expansion of this technology from gigawatt (GW) to terawatt (TW) installed capacity. 4,5Isoelectronic substitution of In and Ga with earth abundant elements led to the development of materials such as Cu 2 ZnSn(S,Se) 4 , 6 with devices fabricated via solution processing reaching efficiencies up to 12.6%. 7owever, the complexity in the preparation of these materials without elemental disorder, structural defects and compositional inhomogeneity has hindered the process towards improving efficiencies towards the 20% mark. 8ese studies have reinvigorated research in similar materials from the Cu-Sn-S family.The Cu-Sn-S phase diagram is rich with many stable phases: Cu 2 SnS 3 , Cu 3 SnS 4 , Cu 4 SnS 4 , Cu 2 Sn 3 S 7 , and Cu 5 Sn 2 S 7 ; among which only Cu 2 SnS 3 (CTS) has shown photovoltaic potential. 9Theoretically, CTS is shown to have a wide stability range and be devoid of Fermi level pinning, while the other forms exhibit poor hole-mobility, Fermi level pinning or high conductivity. 10Cu 2 SnS 3 has a band gap of approximately 1 eV, an absorption coefficient above 10 5 cm À1 and conductivity between 0.5 and 10 S cm À1 , which correspond to a hole concentration of $10 18 cm À3 and mobility between 1 and 80 cm 2 V À1 s À1 . 11,12The present state of the art includes solar cells, featuring materials obtained by physical vapor deposition, with efficiency up to 4.63% for pure CTS and up to 6% for Ge-alloyed CTS. 13,14So far, reports have primarily focused on the material synthesis and assessment of PV properties of CTS and very little knowledge has been gathered with regard to the phenomena limiting the power device conversion.
A variety of physical deposition methods have been employed to prepare CTS including sputtering, 11,[14][15][16] thermal or e-beam evaporation, 11,13 and pulsed laser deposition. 11,17Wet chemical methods, involving single or sequential steps, have also been implemented such as spin/dip coating, 12,18,19,[22][23][24][25] electrodeposition 20,21 and nanoparticle inks. 18,19,227][28][29] In the context of CTS based devices, the vast majority of the work has focused on physical vapor deposition, with only two studies employing solution processing generating efficiencies up to 2.1%. 23,24n this contribution, we assessed the PV performance of phase pure CTS lms obtained by spin coating a single solution based precursor.The interactions between the metal and sulfur components of the precursor are investigated by IR spectroscopy, while thermally induced crystallization and stability of the CTS phase are probed by thermo-gravimetric analysis and differential scanning calorimetry.X-ray diffraction and Raman spectroscopy conrm the generation of polycrystalline CTS lms with a tetragonal phase.Solar cells in the superstrate conguration: glass/Mo/CTS/CdS/i-ZnO/Al:ZnO/Ni-Al were fabricated featuring power conversion efficiencies close to 2%, which in itself is the highest reported for solution processed solar cells in the industrially adopted substrate conguration.Temperature dependent J-V and impedance measurements show that carrier recombination at the CTS/CdS junction is a dominant factor in the cell performance, as well as the presence of two different bulk defect states with activation energies 41 AE 0.4 meV and 206 AE 7 meV.Employing DFT supercell calculations of defect formation energies identies Sn on Cu antisites as the key bulk recombination state.

Experimental and computational procedure
CTS thin lms are processed employing a single solution containing chloride salts of Cu(II) and Sn(II) and thiourea in a dimethyl formamide and isopropanol solvent mixture (1 : 1).The solution is spin-coated on to 5 Â 5 cm 2 Mo coated glass and heated on a hot-plate maintained at 300 C in air for 2 minutes.This process was repeated 6 times to obtain the desirable lm thickness.No blocking layer between Mo and glass is employed in order to minimize Na diffusion from the substrate to the absorber.Finally, the lms are annealed in a graphite box with S powder using a MTI-OTF1200X furnace at 550 C for 30 minutes.Subsequently, the lms are etched in 10% KCN (aq.) and then immediately transferred to an aqueous chemical bath for CdS buffer layer deposition, following the methodology described previously for Cu 2 ZnSnS 4 lms and devices. 30The bath consists of CdSO 4 , thiourea and ammonium hydroxide, maintained at 70 C. Deposition of an i-ZnO and aluminum doped ZnO window layer was performed by sputtering.Finally, a Ni/Al contact grid on top of the solar cell was deposited by evaporation using a shadow mask.Solar cells with an area of 0.5 cm 2 are scribed mechanically.No antireection coating is employed.J-V characteristics of the completed device are measured in the dark and under illumination using an in-house class A solar simulator with a simulated AM 1.5 G spectrum and an integrated power density of 100 mW cm À2 at 23 C. External quantum efficiency of the cells is attained using dual illumination from halogen and xenon lamps and a Bentham TM 300 monochromator (Bentham instruments).Calibrated Si and Ge photodiodes are used as references for the illumination source in J-V characteristics and quantum efficiency measurements.Low temperature impedance measurements are carried out in the dark using a Solartron Modulab impedance analyzer, interfaced with a Linkam HFS 600PB4 cooling stage, in the frequency range of 0.5 Hz to 1 MHz with no applied DC bias and an AC stimulus of 25 mV.
The enthalpy of formation of different point defects is calculated employing DFT in a 96 atom supercell.Calculations are performed with the CASTEP code employing a DFT pseudopotential approach.A generalized gradient approximation functional: PBESOL with ultraso pseudopotentials, an energy cutoff of 500 eV and a Monkhorst-Pack grid with a spacing of $0.02A is used for geometry optimization and energy calculations.The atomic positions are optimized using a BFGS scheme with convergence tolerances of 1.0 Â 10 À7 eV per atom for energy, 0.01 eV A for maximum force, 0.02 GPa for maximum stress, and 5.0 Â 10 À4 A for maximum displacement.

Results and discussion
CTS lms are deposited from a precursor solution containing thiourea as well as Cu(II) and Sn(II) chloride salts dissolved in a dimethyl formamide and 2-propanol solvent mixture.Fig. 1a contrasts the FT-IR spectra of pure thiourea and the precursor solution at room temperature.The key vibrational modes of thiourea undergo a signicant change aer complexation with metal ions in the precursor.In particular, a red shi of the C]S stretching mode (700-740 cm À1 ); and a blue shi of N-H stretching modes (3100-3300 cm À1 ) and coupled modes of C-N stretching and N-H bending (1300-1500 cm À1 ), are observed.A signicant change in the vibrational frequencies of these modes implies a strong interaction between thiourea and metal ions via the S atom.Similar types of complexation have been recently reported in the precursor solution for Cu 2 ZnSnS 4 thin lms. 30 Mo coated glass substrate is spin coated with the precursor solution and heated at 300 C. Aer the desired thickness is achieved by repeating the last steps, the lms are annealed in a S atmosphere at 550 C. Further details of deposition and processing are provided in the Experimental section.Thermogravimetric analysis of the precursor is shown in Fig. S1, † featuring a weight loss between 180 C and 320 C which corresponds to the decomposition of the metal-thiourea precursor to the corresponding sulphide.A small mass loss is also observed between 380 C and 600 C, most probably associated with SnS or S. Based on these results, a thermal procedure is established involving a heating step at 300 C aer spin-coating followed by annealing at 550 C.
Fig. 1b shows a characteristic X-ray diffraction (XRD) pattern of the CTS lm on a Mo substrate.2][33][34][35][36][37] Cubic CTS has a zinc blende type of structure.The monoclinic form is essentially a superstructure of the cubic arrangement, while the tetragonal structure derives from the cubic form in case there is a random distribution of Cu and Sn.This canonical relation leads to a similar diffraction pattern for these different CTS phases. 37In order to establish the type of crystal structure, analysis of diffraction patterns employing Rietveld renement is performed using the Fullprof suite. 38alues of the correlation coefficients R wp and R p of 6.48   A typical Raman spectrum of the CTS thin lm is displayed in Fig. 1c.Raman spectroscopy is particularly useful for distinguishing the various polymorphs of CTS, with characteristic bands reported at 290 cm À1 and 352 cm À1 for monoclinic; at 303 cm À1 and 355 cm À1 for cubic and at 336 cm À1 and 351 cm À1 for the tetragonal forms. 15,39,40The spectrum of the thin lm shows a broad peak with a shoulder between 300 and 380 cm À1 , which can be deconvoluted into two Voigt functions centered at 334 and 352 cm À1 .These two modes are consistent with a tetragonal lattice.Based on the XRD and Raman analysis, a schematic of the rened unit cell is shown in Fig. S2.† Each metal site is tetrahedrally coordinated to S atoms and vice versa with absence of a direct S-S bond.The Cu-S distance is estimated to be 2.322 A while the distances between mixed metal atom sites M(I) and M(II), i.e.Wyckoff positions 4d and 2b, and S are 2.354 A and 2.364 A, respectively.The bond angles S-Cu-S, S-M(I)-S and S-M(II)-S vary between À0.62 and +1.16 around the tetrahedral angle.These results are in agreement with previous studies. 22,23,32,35The metal atom distribution on site M(I) is dominated by Cu (56.39% occupancy) while the M(II) site is Sn dominated (52.53% occupancy).The overall unit cell composition reects a Cu/Sn ratio of 2.12, which is in agreement with the elemental ratio obtained from EDAX.
XPS analysis of the Cu 2p, Sn 3d and S 2p core levels is shown in Fig. 2. The binding energies of Cu 2p 3/2 and Cu 2p 1/2 are observed to be 933.2 and 953.1 eV, with a full width at half maximum of 1.90 eV and 2.24 eV, respectively, consistent with a Cu + oxidation state. 22,23,32,35The absence of any satellite or shake-up peak around 942 eV further conrms the absence of the Cu 2+ state.In suldes, the tetragonal coordination of Cu typically promotes the +1 oxidation state, as the Cu d 10 s 1 conguration is the most stable for this geometry.The binding energies of Sn 3d 5/2 and 3d 3/2 are 486.47 and 494.99 eV, respectively, closely matching those reported for SnS 2 (i.e.+4 oxidation state). 22,23,32,35The S 2p 3/2 and 2p 1/2 are closely located at 161.26 and 162.37 eV, respectively, which is similar to the spectra of CuFeS 2 .The XPS data allow estimating the relative atomic ratios of Cu/Sn/S of 30.6%/14.7%/50.6%,while the ratios of C and O are 1.5% and 2.6%, respectively.These compositions are consistent with EDX analysis as well as the results from the XRD renement.
The morphology of lms as probed through scanning electron microscopy of the CTS lm on a Mo coated substrate is presented in Fig. 3a.The lms are homogenous and compact with grain sizes between 500 and 1100 nm, which are appropriate for device fabrication.Fig. 3b shows the cross-sectional image of the lms which reveals a uniform growth of the adherent lm with a thickness of 1.2 mm.The contrast between CTS and the Mo layer indicates the partial sulfurization of the Mo layer at the interface.Formation of this MoS 2 would implicate the device performance which will be discussed in the later section.Fig. 3c shows the diffuse reectance spectrum of the CTS lm, featuring a large change in reectance between 1050 and 1350 nm corresponding to the band-to-band transition.The inset in Fig. 3c corresponds to a modied Tauc's plot using the Kubelka-Munk transformation of the diffuse reectance, from which an optical band gap of 1.1 eV can be estimated.This value is consistent with previous optical studies of tetragonal CTS. 15,23,40our probe conductivity and Hall measurements of the CTS lms revealed p-type conductivity with a resistivity and hole mobility of 2.226 U cm and 4.581 cm 2 V À1 s À1 , respectively.These values allow estimating a room temperature hole concentration of 6.12 Â 10 17 cm À3 , which is appropriate for The performance of 0.5 cm 2 devices with the substrate architecture glass/Mo/CTS/CdS/i-ZnO/Al:ZnO/Ni-Al, with no antireective coating, under a standard simulated AM1.5 spectrum is displayed in Fig. 4. The CdS layer (70 nm) was grown by chemical bath deposition, while the i-ZnO and Al-ZnO layers (400 nm total) were deposited by RF sputtering.The J-V characteristics of the best cell in the dark and under illumination are shown in Fig. 4a, featuring a power conversion efficiency of 1.92%, an open-circuit voltage (V OC ) of 206 mV, a ll factor (FF) of 34.5% and a short circuit current density, J SC of 27.6 mA cm À2 .These gures of merit are the highest reported for solution processed CTS on a substrate architecture.The relatively narrow dispersion of key parameters for 24 solar cells is summarized in Table S2.† The spectral response of the best cell is shown in Fig. 4b, showing a maximum external quantum efficiency (EQE) of 70% at around 560 nm.At shorter wavelengths the device performance is restricted by the CdS (540 nm edge) and ZnO layers (400 nm edge).A band gap of 1.1 eV can be estimated from the onset of the EQE spectrum, which corroborates the value estimated from diffuse reectance (Fig. 3c).The integrated value of the photocurrent over the entire spectrum is found to be 25.8 mA, falling close to the J SC calculated from J-V curves.Fig. 4c shows the temperature dependence of the key device performance in the range of 300 to 80 K.As the temperature increases, the V OC linearly decreases over large portions of the temperature range, while the J SC slightly increases reaching a maximum at 240 K.The non-monotonic temperature dependence of J SC is a manifestation of a non-linear series resistance, most probably associated with a MoS 2 rectifying back contact. 42he device efficiency increases with decreasing temperature reaching a maximum of 4.8% at 120 K.The temperature dependence of the V OC can be expressed in terms of: [42][43][44] where, n is the diode or ideality factor, E A,V OC is the activation energy for recombination, and J 00 is a weakly temperature dependent pre-factor of the reverse saturation current density, J 0 .Ignoring the temperature dependence of J 00 , a plot V OC vs. T allows estimating a E A,V OC of 610 meV.This value is considerably smaller than the band gap (1.1 eV), which provides a strong indication that recombination predominantly takes place at the CTS/CdS interface.The origin of the interfacial recombination is most likely connected to the cliff like band alignment between CTS and CdS which has been measured using photoelectron spectroscopy. 45ecombination can also be linked to clusters of cation disorder and stacking faults, which has been recently theoretically postulated and experimentally observed employing electron microscopy. 46,47Further studies are required in order to clearly identify the nature of the interfacial recombination site associated with the V OC deciency shown in Fig. 4c.
As discussed in the ESI, † quantitative analysis of carrier dynamics can be extracted from the temperature dependence of the device impedance spectra (Fig. S3 †).A systematic analysis using different equivalent circuits reveals contributions from two RC time constants associated with the CdS/CTS and CTS/ MO interfaces, as well as the dynamic responses of two defect sites (Fig. S3a-c †).][50] The temperature dependent relaxation frequency (u T ¼ 1/ RC) of the two defect states measured by impedance spectroscopy is displayed in Fig. 5a, based on the following expression: where, x 0 is the thermal emission factor and E A,D is the activation energy of the corresponding defect.Both frequencies show a clear Arrhenius behavior over the entire temperature range investigated, providing activation energy values of E A,D1 ¼ 41 AE 0.4 meV and E A,D2 ¼ 206 AE 7 meV.Interestingly, E A,D1 is very similar to the values associated with a Cu vacancy in chalcopyrites and kesterite cells. 42,43,48,50The nature of E A,D2 is discussed further below.The temperature dependence of the back contact R b is displayed in Fig. 5b.In this case, the temperature dependence can be described in terms of, [42][43][44] where, A* is the effective Richardson's constant and f b is the back contact barrier height.In order to identify the nature of the deep defect state exhibiting an activation barrier of 216 meV, we have performed DFT calculations on a 96 atom monoclinic CTS supercell displayed in Fig. 6a.The use of a monoclinic unit cell as a rst approximation can be justied considering that the tetragonal polymorph of CTS is a special case of the cubic form with a random cation distribution, while the cubic form is in itself a superstructure of the monoclinic polymorph. 37Furthermore, the latter is the ground state structure of CTS at 0 K. Calculations on a tetragonal unit cell would require extensive computational resources which are beyond the scope of this work.Details of calculations are included in the Experimental section.
The formation energies of various types of neutral point defects are calculated at ve different points in the Cu-Sn-S ternary phase diagram schematically shown in Fig. 6b.The CTS phase lies at the center of the pentagon, while the ve corners represent the points at which CTS is in equilibrium with stable metal, binary or ternary phases.The point defects investigated were Cu (V Cu ), Sn (V Sn ), and S (V S ) vacancies as well as Cu on Sn (Cu Sn ) and Sn on Cu (Sn Cu ) antisites.The monoclinic structure consists of two different types of Cu sites and three S sites, thus defect formation energies of all of the various sites are calculated.
Fig. 6c shows the formation energies of the various defect sites, revealing that V Cu has the lowest formation energy.
Consequently, V Cu is considered as the main acceptor state of CTS responsible for its p-type conductivity.On the other hand, the key donor state consists of a Sn Cu defect, exhibiting the second lowest formation energy.V Sn and Cu Sn states appear thermodynamically unfavorable based on these calculations.In view of these ndings, the deep state associated with E A,D2 ¼ 206 meV can be attributed to Sn Cu antisite defects.Interestingly, recent TEM studies with atomic-scale resolution have shown clear evidence of the presence of Sn Cu antisite domain boundaries in Cu 2 ZnSnS 4 kesterite nanoparticles. 51Consequently, optimization of the CTS preparation is required in order to simultaneously control the doping density via V Cu and suppress the Sn Cu bulk recombination sites.Solution based precursor methods as described in this work are uniquely suited to explore these conditions by: (i) controlling the composition ratio of the elemental precursors and (ii) introduction of dopants which can decrease atomic disorder. 30

Conclusions
The present report unveils the key factors limiting the efficiency of thin-lm PV devices featuring phase pure polycrystalline CTS lms.We describe a new methodology for preparing high  quality CTS lms featuring a tetragonal structure and metal poor composition.The lm exhibits a band gap of 1.1 eV (direct transition) as well as micron scale grain sizes which are ideal for PV applications.The best solar cell device features a power conversion efficiency of 1.9%, with a V OC of 200 mV, ll factor of 34.5% and J SC of 27.6 mA cm À2 .These gures of merit are amongst the highest reported for CTS devices.Temperature dependent J-V and electrical impedance measurements were carried out in order to assess the key parameters limiting the efficiency of the devices.Extrapolating the V OC to 0 K provides a value of 610 mV, which is signicantly lower than 1.1 V as expected from the band gap.This behavior points towards interfacial recombination losses at the CTS/CdS interface, which is most probably connected to the misalignment of the band edge energies.The generation of a MoS 2 layer at the back contact during the lm formation also generates an electronic barrier with an activation energy as high as 206 meV, while impedance spectroscopy allowed estimating two characteristic frequencies associated with defect states.The shallower one features an activation energy of approximately 40 meV, which is consistent with states generated by Cu vacancies as seen in related materials such as CuInSe 2 and Cu 2 ZnSnS 4 .DFT supercell calculations of the formation energies of different defects suggest that the deeper state with an activation energy just over 200 meV corresponds to a Sn Cu antisite.In addition to investigating alternative absorber layers that can offer a more appropriate band alignment, our work shows that manipulating the composition of the molecular precursor solution and/or adjusting the annealing conditions in order to minimize structural disorder can generate signicant improvement in cell efficiencies.

Fig. 1
Fig. 1 (a) FT-IR spectra of pure thiourea (black) and the precursor (red); (b) X-ray diffraction with Rietveld fitting and (c) Raman spectrum of the CTS film on a Mo substrate.

Fig. 2 X
Fig. 2 X-ray photoelectron spectra of the Cu 2p, Sn 3d and S 2p core levels in the CTS films.

Fig. 3
Fig. 3 Top (a) and cross-sectional (b) SEM images of a CTS film on a Mo substrate.Diffuse reflectance spectrum of the CTS film, the inset shows a plot for a modified Tauc plot using the Kubelka-Munk transformation of diffuse reflectance to determine the band gap (c).
Fig.5bshows two different slopes which can be linked to two different barriers in series.The data are consistent with an energy barrier of 216 AE 14 meV at temperatures above 250 K, while a second barrier of 66 AE 0.21 meV emerges at a lower temperature.This complex behavior has been recently observed in Cu 2 ZnSnS 4 solar cells.43These two distinct barrier heights could be attributed to CTS/MoS 2 and Mo/MoS 2 junctions.This rectifying back contact barrier and the low shunt resistance due to recombination at the CTS/CdS interface are the key contributors to the low device FF.Similar device losses have been seen in kesterite solar cells.8

Fig. 4
Fig. 4 Performance of CTS solar cells with the structure: glass/Mo/ CIS/CdS/i-ZnO/ZnO:Al/Ni-Al, and a total area of 0.5 cm 2 : J-V characteristics of the best cell in the dark and under simulated AM1.5 illumination (a); external quantum efficiency spectra of the best device under short-circuit conditions (b); open circuit voltage (V OC ), short circuit current (J SC ) and power conversion efficiency (h) as a function of temperature (c).Data in (a) and (b) were measured under front illumination at 23 C.

Fig. 5
Fig. 5 Temperature dependence of the characteristic time constants associated with defect sites obtained from impedance spectroscopy (a).The data are plotted following the Arrhenius formalism (see text).Temperature dependence of the series back resistance (b).Impedance spectra were recorded in the dark between 0.5 Hz and 1 MHz and temperatures ranging from 80 to 370 K.

Fig. 6
Fig. 6 DFT supercell calculation of point defect formation energies: the 96 atom monoclinic CTS supercell used in calculations (a); simplified schematic of the Cu-Sn-S phase diagram showing the five points at which formation of defects is assessed (b) and formation energies associated with Cu (V Cu ), Sn (V Sn ) and S (V S ) vacancies as well as Sn on Cu (Sn Cu ) and Cu on Sn (Cu Sn ) antisites at various points on the phase diagram (c).