Effect of the KF post-deposition treatment on grain boundary properties in Cu(In, Ga)Se2 thin films

Significant power conversion efficiency improvements have recently been achieved for thin-film solar cells based on a variety of polycrystalline absorbers, including perovskites, CdTe, and Cu(In,Ga)Se2 (CIGS). The passivation of grain boundaries (GBs) through (post-deposition) treatments is a crucial step for this success. For the case of CIGS, the introduction of a potassium fluoride post-deposition treatment (KF-PDT) has boosted their power conversion efficiency to the best performance of all polycrystalline solar cells. Direct and indirect effects of potassium at the interface and interface-near region in the CIGS layer are thought to be responsible for this improvement. Here, we show that also the electronic properties of the GBs are beneficially modified by the KF-PDT. We used Kelvin probe force microscopy to study the effect of the KF-PDT on the CIGS surface by spatially resolved imaging of the surface potential. We find a clear difference for the GB electronic properties: the KF-PDT increases the band bending at GBs by about 70% and results in a narrower distribution of work function values at the GBs. This effect of the KF-PDT on the GB electronic properties is expected to contribute to the improved efficiency values observed for CIGS thin-film solar cells with KF-PDT.

Harvesting of abundant sun light through photovoltaic (PV) power conversion is considered to be a cornerstone of the renewable energy supply for a sustainable future. Presently, Si wafer-based solar modules dominate the market. Nevertheless, polycrystalline thin-film PV technologies present substantial advantages, including potential for lower costs, for flexible substrates, and for reduced material consumption and weight. Despite the abundance of grain boundaries (GBs) in thin-film solar cells based on Cu(In,Ga)Se 2 (CIGS), CdTe, and the recent perovskite absorbers, all of these materials have achieved power conversion efficiencies above 22% 1,2 , within reach of that of single crystalline silicon. The physics and the role of GBs in these materials has therefore received considerable attention. In CdTe solar cells, it has been shown that the CdCl 2 post-deposition treatment leads to the passivation of GBs 3,4 . For organometal trihalide perovskite solar cells, treatment with a PCBM/C 60 double fullerene layer was shown to passivate defects at GBs, thereby eliminating photocurrent hysteresis effects 5 . GB passivation has been shown to reduce non-radiative recombination and improve device efficiencies 6,7 .
For polycrystalline CIGS solar cells the role and physics of GBs has been intensively investigated. Cu depletion at GBs has been proposed to cause a lowering of the valence band edge, resulting in a reduced hole concentration at GBs [8][9][10] . The potential landscape around GBs at the surface of CIGS films has been measured extensively by Kelvin probe force microscopy (KPFM), finding band bending of different sizes and directions around GBs, mostly attributed to charged defect states [11][12][13][14][15] . Other scanning probe microscopy studies have explained electronic GB properties by a valence band offset [16][17][18] . The role of Na at GBs has received particular attention, and its presence has been directly observed at GBs 19 . It has long been known that the alkali element Na is essential for obtaining high efficiencies 20,21 , where Na can be introduced by various strategies 22 . Interestingly, the recent efficiency increases in CIGS solar cells have been achieved by a potassium fluoride post-deposition treatment (KF-PDT) applied after the deposition of the CIGS absorber 23,24 . The reasons for the beneficial effect of the KF-PDT are currently being investigated. An increase in the net carrier concentration as a result of the presence of K has been reported 25,26 , leading to an increase in the open circuit voltage. This has been attributed to the reduction of compensating donors, similar to what has previously been described for Na 27 . A strong Cu and Ga depletion resulting from the KF-PDT has been observed at the surface of CIGS by x-ray photoelectron spectroscopy (XPS) 23,[28][29][30] . An ion exchange mechanism has been proposed, where K replaces Na in the near surface region of the CIGS 29 . These observations led to the proposal of a modified surface, consisting in In 2 Se 3 or KInSe 2 30 . A widened band gap at the surface has been reported, where a lowering of the valence band edge has been assigned to result from the Cu depletion 28,30 , in agreement with theoretical predictions 8,31 .
While most studies to date have focused on the surface and close to surface effect of the KF-PDT, it had been speculated that K might also be found at grain boundaries (GBs) 28,29 . Very recently, atom probe tomography measurements have in fact observed an increased K concentration at grain boundaries 32,33 . Here we report the first study of the effect of the KF-PDT on the electronic GB properties. We present a comparative KPFM study of CIGS with and without the KF-PDT. Furthermore, we also studied the initial formation of CdS on the CIGS surface and its effect on the surface electronic properties.

Experimental
We studied a set of four different samples, consisting in CIGS without and with a KF-PDT and without and with a thin CdS layer, denominated: CIGS, KF-CIGS, CIGS/CdS and KF-CIGS/CdS, respectively. CIGS absorber layers were co-evaporated on Mo-coated soda-lime glass (SLG) following the 3-stage process 34 . A standard KF-PDT, consisting in evaporating 150 Å of KF at 0.1 Å/s under Se atmosphere directly onto the CIGS absorber heated at 350 °C, was applied for KF-CIGS and KF-CIGS/CdS samples. The CdS buffer layer was grown by chemical bath deposition in an open reactor heated at 60 °C, kept opened and containing ammonia (20 ml, 1 mol/l), cadmium acetate dihydrate (12 ml, 2.6 × 10 −3 mol/l) and thiourea (12 ml, 9.5 10 −2 mol/l). CIGS/CdS and KF-CIGS/CdS were dipped for one minute in the bath and then rinsed with deionized water and dried with N 2 gas. CIGS and KF-CIGS were also used to complete solar cell devices to verify the beneficial effect of the treatment. 60 nm-thick CdS (7 minutes in the same bath), as well as ZnO/ZnO:Al bilayers and contact grids were deposited onto these absorbers following the baseline of the IMN 35 . Table 1 shows the photovoltaic parameters of these devices. The KF-treated device exhibits a higher open circuit voltage (V oc ) than the KF-free one, while fill factor (FF) and short circuit current density (J sc ) have similar values, yielding an important increase of the conversion efficiency ( + 8 rel%).
All samples for the KPFM study were kept under N 2 atmosphere after the growth to reduce any uncontrolled modification of the surfaces. They were taken from the N 2 atmosphere immediately prior to the KPFM experiments. KPFM measurements were performed in a Bruker Dimension Icon atomic force microscope (AFM) operated in ambient environment. We used Pt/Ir-coated cantilevers (PPP-EFM Nanosensors TM ) with a nominal tip radius of 25 nm, spring constant of 2.8 N/m, and 75 kHz resonance frequency. To ensure comparability of the results the tips were calibrated using an Au-coated Si sample. The surface potential was measured using amplitude-modulated KPFM in a dual pass technique. In the first pass the topography line profile is mapped by tapping mode AFM. The topography is traced at a lift height of 5 nm in the second pass, during which an AC-bias voltage (V AC = 500 mV) is applied to the tip at the mechanical resonance f 0 of the cantilever. The oscillating electrostatic forces are minimized by applying a compensating V DC to the tip. The contact potential difference (CPD) is then determined as V CPD = − V DC = e(Φ sample − Φ tip ), where Φ is the work function. For the analysis of the CPD variation across GBs, we identified the GB based on the topography image 15 and analyzed a CPD line profile perpendicular to the GB averaging over an about 150 nm wide segment of the GB.

Results
The topography of all samples appears similar (Fig. 1). Some structures of order of a few tens of nanometers can be seen in all samples, decorating mostly the surface/facets of the grains (see Supplementary Information). These small structures/clusters appear to be more abundant in the CIGS samples in comparison with the CdS-coated ones. The CPD images also provide a fairly similar contrast between the samples.
To compare the electronic surface properties of the samples, the distributions of CPD values are shown as histograms in Fig. 2. The CPD histograms correspond to the initial KPFM measurement of each sample, taken immediately after removal from the N 2 atmosphere. To ensure the stability of the surface electronic properties, we monitored the development of the CPD with time over a period of ~10 hours by repeated KPFM experiments with intermediate calibrations of the AFM tip on an Au/Si reference sample (Fig. S2, Supplementary Information).
For quantitative evaluation, each histogram in Fig. 2 was fitted by a Gaussian distribution, from which an average CPD and a distribution of CPD values (full-width at half maximum (FWHM)) are determined. The values are summarized in Table 2  To analyze in detail the CPD variation at GBs, we considered 14 GBs for each sample (Fig. 3). The location of the GBs (indicated in Fig. 1) was identified based on the topography images 15 to ensure their selection independent of specific CPD changes. Nearly all analyzed GBs show an increase of the CPD at the GB (Fig. 3a-d). The    Table 2. CPD average and FWHM values obtained from the Gaussian fits to the CPD histograms of the four studied samples. Using the calibration on an Au/Si sample, the work function was estimated, assuming Φ Au = 5.1 eV. The average CPD changes at the grain boundaries, < ΔCPD GB > , are also given (see Fig. 4a).

CIGS KF-CIGS CIGS/CdS KF-CIGS/CdS
variation Δ CPD GB was determined by analyzing the difference between the CPD value at the GB and the average CPD value of the grain surfaces on both sides of the GBs; these values are illustrated in Fig. 3e-h. Two observations can be made from these plots, the short CdS deposition attenuates the CPD variation at GBs and the samples with KF-PDT show a smaller spread of the CPD variations at the GBs. These observations are analyzed statistically in Fig. 4a, showing Fig. 3a-d, it can be observed that the variation of the CPD value at the GBs is smaller for the KF-CIGS samples compared with the CIGS samples without the treatment. This observation is quantified in Fig. 4b, showing the spread of the CPD values at the GB and those on the grain surfaces to the left and right side of the GB. It is clearly observed that the spread of the CPD values at the GBs is smaller for the KF-CIGS samples. The observations from the results in Fig. 4 indicate that the KF-PDT leads to more homogeneous GB electronic properties.

Discussion
The observed overall increase in the work function due to the KF-PDT (about +160 meV for CIGS and +280 meV for CIGS/CdS) can result from (i) a surface dipole (by means of a change in electron affinity), (ii) a different surface material (by means of a direct change in the work function), (iii) a different band bending (by means of different surface defects or surface charging), (iv) a different charge carrier concentration (leading to a change in the Fermi level), or (v) a band gap widening (in the case of shift of the conduction band minimum). The increase in CPD is in agreement with the observations of Handick et al. 30 , who report indications of an In 2 Se 3 or a KInSe 2 surface composition and a significantly increased band gap. In any case, a change in the work function of the CIGS layer should affect the band alignment to the CdS layer. The short deposition of CdS for 1 min does not lead to the growth of a CdS film sufficient to form a pn junction. For a sufficiently thick CdS deposition, a lowering of the work function could be expected in accordance with the n-type character of CdS. The increase of the work function upon the 1-min deposition of CdS therefore indicates a surface modification; this might be a surface cleaning or the formation of a different surface compound, e.g. CdIn 2 S 4 , as proposed recently by Lepetit et al. 36 .
The main result of our work relates to the electronic properties of the GBs. We observed that the KF-PDT increases the upward band bending at the GBs (Fig. 5). This increase is compatible with the observed general increase of the work function and indicates that the effect of K at the GBs is enhanced. In fact, recent atom probe tomography experiments have directly observed an increase in the K concentration at GBs 32,33 . Two-dimensional device simulations have shown that sufficient band bending at GBs can lead to improved device characteristics 37,38 . We note that the measured Δ CPD GB values need to be corrected considering the well-known averaging effect of KPFM 39,40 . According to 15,41 , a correction factor of about 2-5 should be applied considering the lift height of 5 nm. Nevertheless, it should be noted that the increase of band bending at the GBs is smaller than the overall increase of the CPD on the surface. We would also like to point out that while KPFM is a surface characterization technique, the potential variations measured do reflect the potential in the near-surface region. KPFM measures and compensates the local electrostatic field. In the case of charges localized below the surface, their electric field will be screened by the surrounding material and the screening length will depend on the charge-carrier concentration. Therefore, the observed potential variations are expected to be representative for the near-surface region, which is the location of the charge-separating junction.
In addition to the increased upward band bending at the GBs, we consider it at least equally important that the KF-PDT results in a more homogeneous CPD distribution at the GBs (Fig. 5). A narrow distribution of CPD values should improve the junction properties at the GBs and could correspond to reduced voltage losses due to different band alignments at different GBs, in agreement with the improved V oc of the KF-treated CIGS device (Table 1). Moreover, the initial CdS layer formation results in a further reduced variation at the GBs, making the band alignment even more homogeneous. This observation is in agreement with the favorable Cd diffusion into the CIGS layer and GBs, controlling the band bending by the formation of Cd point defects, as has been proposed by Pianezzi et al. 26 . Theoretical studies support such n-type doping of Cd in CIGS 42,43 .

Conclusion
So far, the positive effect of the KF-PDT on CIGS solar cells was attributed to modifications of the doping concentration and the interface forming the pn-junction. Our results indicate that the KF-PDT additionally contributes to more homogeneous electronic properties of the GBs at the CIGS surface, which should improve the band alignment at the GBs, reducing losses in the open circuit voltage and leading to an efficiency enhancement. In view of the general relevance of the electronic activity in GBs and its impact on the efficiency of polycrystalline solar cell materials (e.g. CdTe and perovskites), an improved understanding of the beneficial modification of GB electronic properties might lead to the development of strategies to improve GB properties also in other solar cell and electronic materials.