The role of nanoparticle inks in determining the performance of solution processed Cu2ZnSn(S,Se)4 thin film solar cells

Cu2ZnSnS4 (CZTS) nanoparticle inks synthesized by the injection of metal precursors into a hot surfactant offer an attractive route to the fabrication of Earth‐abundant Cu2ZnSn(S,Se)4 (CZTSSe) thin film photovoltaic absorber layers. In this work it is shown that the chemical reaction conditions used to produce CZTS nanoparticle inks have a fundamental influence on the performance of thin film solar cells made by converting the nanoparticles to large CZTSSe grains in a selenium rich atmosphere and subsequent cell completion. The reaction time, temperature and cooling rate of the nanoparticle fabrication process are found to affect doping level, secondary phases and crystal structure respectively. Specifically, prolonging the reaction offers a new route to increase the concentration of acceptor levels in CZTSSe photovoltaic absorbers and results in higher device efficiency through an increase in the open circuit voltage and a reduction in parasitic resistance. Quenching the reaction by rapid cooling introduces a wurtzite crystal structure in the nanoparticles which significantly degrades the device performance, while elevating the reaction temperature of the nanoparticle synthesis introduces a secondary phase Cu2SnS3 in the nanoparticles and results in the highest cell efficiency of 6.26%. This is correlated with increased doping in the CZTSSe absorber and the results demonstrate a route to controlling this parameter. © 2016 The Authors. Progress in Photovoltaics: Research and Applications published by John Wiley & Sons, Ltd.


INTRODUCTION
Cu 2 ZnSn(S,Se) 4 (CZTSSe) is a promising photovoltaic material and experiencing rapid improvements in recent years because of its Earth-abundant constituents and outstanding optoelectronic properties [1][2][3][4][5][6]. Among the variety of techniques employed for the preparation of the absorber films, solution-based deposition and processing have the potential to provide low-cost scalable routes to produce photovoltaic devices with high efficiency. Such results include the current record efficiency for CZTSSe solar cells at 12.6% using a hydrazine-solution based method [7]. On the other hand, nanoparticle inks offer a non-toxic alternative. Deposited from the Cu 2 ZnSnS 4 (CZTS) nanoparticle inks, CZTS precursor thin films annealed in the presence of Se can provide devices with efficiency as high as 9.0% [8].
Both CZTS and CZTSSe are known to be highly disordered as a consequence of high doping levels from native defects resulting in a band structure that can be described by a fluctuating potential model [9]. The doping levels are of fundamental importance as they determine the optoelectronic behavior of devices. We have recently developed an experimental procedure to tune the properties of CZTS nanoparticles via the chemical reaction conditions [10]. Wurtzite CZTS was formed at lower temperature whereas Cu 2 SnS 3 (CTS) was present in the nanoparticles at higher temperature. The reaction time was also found to have a strong effect on the nanocrystals which became increasingly copper poor and zinc rich as the reaction evolved. Additionally, the existence of wurtzite structure depends critically on the reaction cooling rate.
CZTS nanoparticle inks are the basis of the photovoltaic absorber layer; their impact on device performance has not yet been reported or investigated. In most cases it is assumed that the majority of CZTS nanoparticles have been converted into large grains and lose their direct influence on the device performance. However, below the top micronscale grains, the metal-chalcogenide nanoparticles are still found present in the grain sub-layer creating a bi-layer absorber [11]. To the best of our knowledge there has been no detailed report linking the properties of the nanoparticle inks with the completed solar cell performance. In this work, we demonstrate that we can fabricate large grain CZTSSe thin films and control the concentration of dopants. This is shown to have a significant effect on the solar energy conversion efficiency of photovoltaic devices made from the films. Furthermore, the device performance is also found to be sensitive to the atomic arrangement of the CZTSSe crystal structure which is strongly influenced by the rate of reaction cooling.

Nanoparticle synthesis and ink preparation
CZTS nanoparticles were fabricated using a hot-injection method described previously [10]. Four types of nanoparticle inks were prepared by varying key reaction parameters: temperature, time or cooling mode. The control sample (sample A) was obtained by setting the reaction temperature to 225°C for 30 min followed by slow cooling (~5°C/min) under ambient laboratory conditions. The remaining three nanoparticle inks were fabricated by systematically changing one of the key reaction parameters as indicated in Table I. For quenched sample B, the reaction was quenched by rapidly cooling the reaction vessel in ice-water.
The as-synthesized CZTS nanoparticles were dispersed in toluene with a concentration of~10 mg/ml. However, the ink needed to be concentrated further to yield efficient deposition. The CZTS precipitation was therefore dispersed into hexanethiol (95%, Sigma Aldrich) with the aid of sonication and then filtered through a 1 micron syringe filter (Whatman) to remove large agglomerates. The filtrate was then sonicated for 10 min to yield the final concentration of~100 mg/ml.

Thin film deposition
CZTS thin films were prepared by spin-coating~30 μl of the ink onto a square Mo-coated glass or bare glass substrate (2.5 cm × 2.5 cm) at a rate of 1200 rpm for 5 s. The samples were then dried on a hot plate in air at 150°C for 30 s and then at 300°C for 30 s (hereafter "softbaking") to remove the residual solvent. The thickness of the deposited thin films could be accurately controlled and reproduced by repeated spin coating and soft-baking procedures. A thickness of 1 μm was set for efficient light absorption in all thin films.

Photovoltaic device fabrication
In order to produce an efficient absorbing layer, the CZTS nanocrystal thin films need to undergo a recrystallization process where small nanoparticles are converted into large grains. This is achieved via a selenization process where the formation of the sulfoselenide compound CZTSSe through the Se-S anion exchange induces large-scale grain growth by dissolution of the parent CZTS and cation reordering in the selenium-based lattice. The as-deposited CZTS/Mo/glass films were placed inside of a graphite cylinder with 300-mg selenium pellets placed directly beneath the substrate. The furnace was evacuated and heated to 500°C (~20°C/min) where the temperature was held for 20 min in an argon atmosphere (~10 mbar).
The resulting CZTSSe thin films were converted into solar cell devices with the configuration: Mo/CZTSSe/ CdS/i-ZnO/ITO/Ni-Al. The CdS buffer layer was deposited using a chemical bath. Deionized water was first poured into a double-walled beaker. After the temperature inside the beaker stabilized at 70°C, CdSO 4 (2 mM) and ammonium (1.5 M) were added successively with a magnetic stir bar continuously stirred to form the solution. The CZTSSe films were then soaked in the solution for 2 min before thiourea (12 mM) was mixed into the solution. After the reaction, the samples were removed from the bath, rinsed with deionized water and dried under a nitrogen stream and then annealed at 200°C for 10 min in air. The transparent oxide layers were deposited by sputtering~60 nm-thick insulating ZnO (i-ZnO) and 200 nm-thick indium tin oxide (ITO). Finally, the front contact grid was deposited by electron beam evaporation of Ni (~50 nm) and Al (~1 μm) through a shadow mask. Each substrate was defined by mechanical scribing into nine 0.16 cm 2 devices.

Experimental methods
The morphology of CZTS precursor thin films was investigated using tapping mode atomic force microscopy (AFM) using a Veeco Nanoscope multimode system. The structure of the CZTS nanocrystals was examined Cu 2 ZnSnS 4 nanoparticle inks Y. Qu, G. Zoppi and N. S. Beattie using X-ray diffraction (XRD) carried out with a Siemens D-5000 diffractometer using a Cu K α radiation source (λ = 0.15406 nm for K α1 ). The film morphology and composition were determined using a FEI Quanta 200 scanning electron microscope (SEM) equipped with Oxford Instruments energy dispersive X-ray spectroscopy (EDS). K α X-ray emission lines for Cu, Zn, S and L α for Sn were used for quantification. Raman spectroscopy was performed with a Renishaw inVia microscope using a 514-nm argon ion laser. The valence states of the constituent elements in the CZTS thin film were assessed using a Thermo Scientific K-Alpha X-ray Photoelectron Spectrometer (XPS) System with a monochromatic Al K α X-ray source (1486.6 eV). A short sputter cleaning was applied before the scanning process to remove possible contamination on the top surface. Elemental depth profiling was performed by secondary ion mass spectroscopy (SIMS) using a Hiden Analytical gas ion gun and quadrupole detector. A primary Ar beam with an impact energy of 4 keV and beam current of 200 nA was used to sputter over a 500 μm × 500 μm rastered area. A gating area of 10% was used to remove side wall effects. Current density-voltage (J-V) measurements were performed in a four-point probe configuration using a Keithley 2400 series source meter. Samples were illuminated with an Abet Technologies Sun 2000 solar simulator with an AM1.5 spectrum set (100 mW/cm 2 ). The external quantum efficiency (EQE) measurements were operated using a double grating monochromator (Bentham Instruments, M300) with illumination normalized against calibrated silicon and germanium detector. Capacitance-voltage (C-V) measurements of the finished device were made in the dark using an Agilent E4980A Precision LCR Meter operating at 100 kHz frequency and 100 mV step with bias voltage from 0 to À1 V.

CZTS precursor thin films prepared from nanoparticle inks
The as-deposited precursor control thin film appears smooth and is formed of densely packed, uniform spherical CZTS nanoparticles with diameter around 20 nm as shown in Figure 1 (a). Figure 1 (b) shows a representative crosssectional SEM image of the 10-layer coating CZTS nanocrystal film on Mo/glass substrate. It can be seen that the film is uniform with a thickness of~1 μm.
XPS was used to confirm that the chemical synthesis method resulted in the formation of Cu (I) valence states in the CZTS precursor thin film synthesized using control conditions. As shown in Figure 2  The SIMS depth profiles of a CZTS precursor control thin film are shown in Figure 3. Except at the very surface of the thin film where the oxygen increases the ion yield of positive ions, the constituent elements of CZTS exhibit uniform distribution throughout the film thickness. The depth profile S/metals ratio is also shown as the dark solid line at the bottom of the graph which further demonstrates the uniformity of S and metals. The significant increase of Mo signal around 2100 s indicates the start of the CZTS/Mo interface. Combining the curves of Mo and S/metal ratio, the spectra can be divided into three compositional zones marked by the vertical lines. A MoS 2 layer is formed during the soft-baking process between the CZTS layer and Mo substrate. The C-rich solvent molecules are loosely bounded to the CZTS nanoparticles, and the softbaking process allows to partially evaporate those molecules. However a diffusion process is required within the precursor film for this to happen, and it is conceivable that some of the C will diffuse towards the surface and evaporate while C diffusing towards the back of the film will start accumulating at the CZTS/Mo interface and result in the increase of C signal towards the interface with the substrate.
The crystal structure of the thin films fabricated using the four different nanoparticle inks was studied by XRD and Raman spectroscopy ( Figure 4). As shown in Figure  4 (a), the major XRD diffraction peaks of the control appear around 28°, 47°and 56°, which can be attributed to the (112), (220) and (312) planes of kesterite CZTS (PDF 026-0575), respectively. The quenched sample exhibits three weak diffraction peaks marked by asterisks shown in Figure 4 (a) in addition to the primary peaks. These belong to wurtzite CZTS co-existing in the nanoparticle ink because metastable wurtzite CZTS has insufficient time to complete the transformation to the more stable kesterite phase when the reaction temperature cools rapidly [10]. XRD patterns of the hot and prolonged samples have a similar pattern to kesterite CZTS with no obvious secondary phases observed.
Raman spectroscopy was used to further confirm the structure of the samples and rule out other binary or ternary compounds that have a similar XRD pattern to CZTS (Figure 4 (b)). For the control, the main peak located at 329 cm À1 is likely to be the A 1 mode of kesterite CZTS shifted downwards from 338 cm À1 because of the cation sublattice disorder for non-stoichiometric CZTS material. Because of the mixed atomic arrangement of kesterite and wurtzite phase, the main peak of the quenched sample broadens and shifts to 335 cm À1 . For the hot sample, a strong peak at 329 cm À1 corresponds to the A 1 mode of CZTS, although there is also a nearby shoulder peak at 301 cm À1 . This peak matches with either cubic CTS at 303 cm À1 or tetragonal CTS at 297 cm À1 [12]. For the prolonged sample, however, only one characteristic peak located at 329 cm À1 is observed, and has a similar spectrum to the control.

Preparation of CZTSSe photovoltaic absorber thin films
The selenization process was applied to the CZTS precursor thin films to convert the nanoparticles into micronsized grains and minimize grain boundaries.
XRD was used to confirm the crystal structure of the selenized films. As shown in Figure 5 (a), the selenization process induces a sharpening of the diffraction peaks. All the peaks can be assigned to CZTSe (PDF 052-0868) slightly shifted to higher angles because of residual sulfur present in the lattice (Table II). As shown in Table II   composition of the selenized film is zinc rich and copper poor and about 90% anion exchange between sulfur and selenium occurs during the selenization process. A zinc loss was observed in the selenization process as the Cu/(Zn + Sn) ratio increased from 0.79 to 0.89 and the Zn/Sn ratio decreased from 1.15 to 1.
Raman spectroscopy was also performed on the selenized films and as shown in Figure 5 (b), the two sharp peaks at 172 and 195 cm À1 , and weak peaks at 232 cm À1 correspond to the literature values of CZTSe [13]. However, the shoulder peaks at the high frequency side of the 232 cm À1 are difficult to determine with accuracy. In addition, the wide peak at 327 cm À1 can be attributed to the A 1 mode of residual CZTS [14]. The~10 cm À1 shift from the reported peak position of CZTS (~338 cm À1 ) is likely because of the disorder or stress induced by the exchange between sulfur and selenium. In order to analyze the region close to the absorber/Mo interface, the measurement was repeated on the substrate after mechanically removing part of the absorber. As shown in Figure 5 (b), the spectrum from the substrate shows clear peaks at 241 and 253 cm À1 which can be identified as the main mode of MoSe 2 and ZnSe, respectively [15]. In addition, weaker peaks at 171, 290 and 350 cm À1 can all be indexed to MoSe 2 .
As shown in Figure 6 (a), the cross-sectional image of the selenized thin film reveals a four-layer structure. The bottom layer is typical of Mo with a columnar grain structure. Under the selenium-rich annealing conditions, an approximately 0.5 μm thick layer of Mo(S,Se) 2 forms on top of the Mo layer by replacing part of the sulfur in MoS 2 . Above the substrate, the CZTSSe layer is composed of a residual fine grain sublayer (~0.5 μm thick) and a large grain upper layer (~1 μm). Compared to the CZTS precursor thin film, the thickness of the CZTSSe thin film is increased because of the Se-S anion replacement and volume expansion of the CZTSSe unit cell during the selenization process.
The elemental distribution in the selenized film was studied using SIMS and shown in Figure 6 (b). Combined with the cross-sectional SEM image, the SIMS spectra can be divided into the four sections mentioned previously with different compositions. The large-grain layer is composed of high purity CZTSSe with low carbon content. The residual fine grain layer is rich in carbon, especially close to the substrate interface confirming our previous hypothesis of C being driven towards the back of absorber   layer upon thermal annealing. The Raman analysis of the substrate after absorber removal also supports the presence of binary ZnSe in the fine grain layer [11]. A Mo(S,Se) 2 layer is formed above the Mo substrate because high levels of S and Se signals are found between the CZTSSe film and Mo substrate. Even in selenium-rich conditions, some residual sulfur is still present in the lattice. This agrees with the Raman spectra shown in Figure 5 (b) where a slight Raman shift is observed compared with literature position of the pure MoSe 2 [15].

Photovoltaic device performance
The CZTSSe thin films were converted into solar cell devices following the deposition of CdS, i-ZnO, ITO and Ni/Al contact layers. A series of four substrates was prepared (nine 0.16 cm 2 devices per substrate) using the four types of nanoparticle inks: control (sample A), quenched (sample B), hot (sample C) and prolonged (sample D). Devices differ only in the precursor ink used, and all other conditions were kept identical during the fabrication. Note that the devices do not include an anti-reflective coating. J-V curves of the best device built on each substrate are shown in Figure 7 (a) with the device parameters extracted from the J-V curves summarized in Table III. Additionally, the distributions of solar energy conversion efficiency (η), open circuit voltage (V oc ), short circuit current density (J sc ) and fill factor (FF) of the nine devices on each substrate are shown in Figure 8 demonstrating the high uniformity of device performance across the substrates.
Baseline photovoltaic parameters for the control device can be derived from Figure 7(a): η = 5.41%, V oc = 0.36 V, J sc = 29.6 mA/cm 2 and FF = 50.7%. The most striking contrast to this set is found for the quenched reaction conditions as the presence of wurtzite phase CZTS has a profound effect on the efficiency of device B where η = 2.30%, indicating that the reaction cooling mode which affects the atomic arrangement has a direct negative influence upon the device performance. The wurtzite phase in  the nanoparticles is thermodynamically unstable and tends to convert into stable kesterite phase during the selenization process [16]. The crystal structure difference between the wurtzite and kesterite structure may hinder the grain growth in the selenization process and results in small grains as observed in Figure 9. This is a new and important insight not least because this fabrication technique for synthesis of CZTS nanocrystals is widely reported in the literature. As shown in Table III, the reduced efficiency of this device arises not only from reductions in V oc and J sc , but also a large drop in FF caused by higher series resistance (R S ) and lower shunt resistance (R SH ). From the top view SEM images shown in Figure 9 (a), the selenized thin film made from sample A is composed of densely packed grains that can be up to 2 μm, with no obvious voids or cracks on the surface. However, as can be seen in Figure 9 (b), the grain size of the CZTSSe film made from the quenched sample is significantly smaller than that of the control. Considering the sheet resistance of the Mo and ITO is 0.2 Ω/□ and 35 Ω/□, respectively, the increased parasitic resistance of the quenched device is likely derived from increased density of grain boundaries which act as recombination centers. It is possible that wurtzite phase CZTS undergoes a phase transition to kesterite during the selenization stage [16]; however, this wurtzite CZTS may hinder the grain growth process and even present a residual component that significantly degrades device performance. Figure 7 (a) also shows the J-V characteristic for the devices with hot and prolonged reaction conditions (C and D respectively). These devices have higher solar energy conversion efficiencies relative to the control, and this is manifested through higher fill factors which in turn are determined by R S and R SH . The former is likely to be influenced by the unconverted nanocrystals and the Mo(S,Se 2 ) layer at the back of the device [8] while R SH is influenced by the CZTSe grain size. Analysis of Figure 9 reveals that the prolonged reaction time results in a selenized thin film that has slightly larger grains (Figure 9 (d)) compared with the hot reaction condition (Figure 9 (c)). This reduces the number of grain boundaries and, in turn, the number of shunt pathways [17] and is evidenced by the higher R SH value.  The EQE spectra of the devices (Figure 7 (b)) show behavior that is consistent with the J-V characteristics. All devices show less ideal EQE toward longer wavelengths which is a result of recombination losses that can attributed to the fine grain and MoSe 2 layers [14], and future work will focus on reducing the influence of these on performance. In addition, the spectrum for the quenched device (E g = 1.12 eV) is exhibiting a strong triangular shape indicative of severe current collection losses. This could arise from short electron diffusion length and high doping density (see analysis of capacitance measurements). The variations in energy band gap derived from the band tail of the EQE spectra are consistent with the fluctuations in Se/(Se + S) ratios as measured by EDS (Figure 9 insets). It is interesting to note that sample B yields less Se incorporation into the CZTS lattice compared to the other three processes which have similar Se/(Se + S) ratio at around 0.90.
The hot device (device C) exhibits the highest overall efficiency of η = 6.26%, with V oc = 0.37 V, J sc = 29.7 mA/cm 2 and FF = 57.0%. Raman spectroscopy (Figure 4 (b)) revealed that this film contained secondary phase CTS, and this champion efficiency correlates with the current world record efficiency for CZTSe thin films fabricated using selenization of CZTS nanoparticles which were also found to contain CTS [8]. The reason for this is not currently well-understood and merits further investigation.
C-V measurements were performed to estimate the doping density (N A ), the built in potential (V bi ) and depletion layer width (W) in the CZTSSe absorber layer of the completed devices. Figure 10 shows a plot of 1/C 2 versus reverse bias voltage for the devices and from the slope and intercept of the linear fits, N A and V bi can be determined respectively. Additionally, the width of the depletion region in each region W can be deduced from the data. These values for all four samples are listed in Table IV where the values are comparable to vacuum [18] or hydrazine-based CZTS [7] fabrication methods confirming that there is no loss of junction quality using the nanocrystal fabrication route. CZTSSe is usually reported to be p-type arising from a large number of Cu vacancies V Cu À , although antisite defects such as Zn Cu and Cu Zn are also common [19].
Comparisons between all four devices in terms of doping are difficult because the crystal structure has been shown to be markedly different between the samples and therefore cannot easily be attributed to the density of V Cu À (despite all absorbers being Cu poor). However, we have demonstrated a high degree of similarity between the control and the prolonged samples and devices, and the most obvious discrepancy is indeed N A which is an order of magnitude higher for the prolonged sample. It is this parameter that results in a higher solar energy conversion efficiency and to the best of our knowledge the first experimental demonstration that the doping density can be increased without changing the crystal structure of a CZTSSe absorber synthesized from nanoparticle inks. The increase in efficiency derives from small increase in average V oc for the prolonged device and a reduction in fill factor. Although the interpretation of the C-V data for the quenched and hot devices is more difficult, we suggest that the increased values of N A for these devices relative to the control are a consequence of increased disorder in the crystal structure. The case of the hot device is particularly interesting because it results in the highest efficiency of all the devices. This is consistent with the earlier analysis of the doping concentration between the control and the prolonged device. The key difference here however, is that we have correlated the higher reaction temperature with the presence of secondary phase CTS. These findings are important in the context of work by Miskin et al. who recently demonstrated a CZTSSe nanoparticle ink solar cell with η = 9% (with an anti-reflection coating) [8] but do not account for an experimentally observed CTS phase which has previously been thought to be detrimental to solar cell performance because of a narrower energy band gap. Our results indicate that the apparent positive effects of CTS may be because of an effective increase in the absorber doping which in turn enhances V oc in the device. The apparent increase in V oc is seen more clearly by considering that the voltage deficit (E g -qV oc ) (also shown in Table III) is minimized for the hot device. It has been shown that CTS may precipitate at the CZTS grain boundaries [20], where the minority carrier collection can also be enhanced [21]. Additionally, the CTS precipitation could provide some grain boundary passivation and thus explain the increase in device performance observed.
Solving Poisson's equation for the electric field in the n-type and p-type regions of a pn junction yields a dependence on the density of donors and acceptors respectively. Continuity of the field requires that the width of the depletion region varies inversely with doping concentration, i.e. that the depletion region extends into the side with the lowest doping. This is consistent with the control device having the largest value of W in Table IV and the highest EQE in the 700-1150 nm range testimony of the better charge collection in that device. The values of V bi are similar apart from the hot sample, and this is likely attributable to the presence of CTS which reduces the energy band gap of the absorber.

CONCLUSIONS
High quality CZTSSe photovoltaic absorber layers were fabricated from CZTS nanoparticle inks by annealing stacked layers of nanoparticles in a selenium rich atmosphere. These were subsequently incorporated into thin film solar cells, and the device performance has been found to depend critically upon the chemical reaction conditions used to synthesize the nanoparticles. Crucially, rapid thermal quenching of the reaction results in wurtzite CZTS in the nanoparticles. This results in smaller grains and an increase in the number of grain boundaries in the CZTSSe thin film absorber layers which are strongly detrimental to the device performance. Prolonging the reaction offers a new route to increasing the concentration of acceptor levels in CZTSSe photovoltaic absorbers and results in higher device efficiency relative to a control device. The presence of CTS correlates with increased efficiency, and the apparent positive effects of this phase are also accompanied by an increase in the doping levels. To fully exploit doping control in these devices, we propose that it will be essential to minimize any negative effects of the fine grain sub-layer below the CZTSSe grains that are commonly found in this type of solar cell.