Perovskite-polymer blends influencing microstructure, non- radiative recombination pathways and photovoltaic performance of perovskite solar cells

Solar cells based on organic-inorganic halide perovskite are now leading the photovoltaic technologies due to their high power conversion efficiency. Recently, there have been debates on the microstructure related defects in metal halide perovskites (grain size, grain boundaries, etc.) and a widespread view is that large grains are a prerequisite to suppress non-radiative recombination, and improve photovoltaic performance although opinions against it also exist. Herein, we employ blends of methylammonium lead iodide perovskite with an insulating polymer (polyvinylpyrrolidone, PVP) that offer the possibility to tune the grain size in order to obtain a fundamental understanding of the photoresponse at the microscopic level. We provide, for the first time, spatially-resolved details of the microstructure in such blend systems via Raman mapping, light beam induced current (LBIC) imaging and conductive atomic force microscopy. Although the polymer blend systems systematically alter the morphology by creating small grains (more grain boundaries), they reduce non-radiative recombination


Introduction
Organic-inorganic halide perovskite solar cells (PSCs) now demonstrate photovoltaic performance comparable to state-of-the art silicon and thin film solar cells. Their certified power conversion efficiency (PCE, 22.7%), 1 achieved in merely six years since their first high efficiency solid-state cell report in 2012, 2 together with their added functionality such as transparency, flexibility and low temperature compatible roll-to-roll processing put them at forefront of future photovoltaic technologies. 3 After the initial efficiency driven research, the focus nowadays is to understand what makes these devices so well performing and the potential to further improve their photovoltaic (PV) performance. Interfacial engineering to manipulate recombination kinetics at the device interfaces, 4-7 compositional engineering to extend the perovskite absorption region 8 or to enhance its structural stability, [9][10] as well as morphology tuning to obtain a pin-hole free layer or to control non-radiative recombination pathways [11][12][13] are the key research areas that enabled the remarkable rise in device performance in PSCs.
An important subject in PSCs are their point and structural defect properties that influence band alignment, charge transport/transfer and their stability. These defects -whether present within the bulk, at boundaries between two neighboring grains or at the device interfaces -are influential to performance and their control is the key to further alleviate the PCE close to its thermodynamic limit. 14 As perovskite films are typically solution-processed at low temperature, a significant density of defects cannot be avoided (~10 16 cm -3 ). [15][16] Particularly interesting are the defects related to grain boundaries (GBs) as there are reports in favor and against their major role in creating deep trap states. A widespread understanding is that perovskite films comprising large grains (in the range of a few micrometer) are crucial to high performing devices as they contains lesser density of trap states thus leading to lesser non-radiative recombination, enhance light harvesting efficiency and improve charge transport properties. [17][18][19][20][21] Herein, the GBs are considered as a recombination center prevailing non-radiative recombination losses that eventually influence the PV parameters, particularly, the open circuit voltage (VOC). This is well explained by Rau using a reciprocity relationship between luminescence properties of solar cells and the VOC, showing that VOC could reach its radiative limits (VOC rad ) if the recombination current at a given voltage is zero. 22 Recently, significant reduction in non-radiative losses in a mixed halide perovskite ((Cs,MA,FA)Pb(I0.4Br0.6)3) based PSCs has been reported when the defects related to GBs are passivated via potassium halide layers. 23 The defect passivation is reflected in the remarkable open circuit voltage (VOC) 1.23 V, ~110 mV higher than a reference device employing pristine perovskite without passivation. An enhancement in performance is also noted for metal halide perovskite based LEDs and optical amplifiers by the GBs passivation. 24 However, contrarily, high charge collection and high performance is also reported using small grains. For example, PCE above 20% is reported in hysteresis-free inverted planar PSCs made using small perovskite grains (100 -200 nm). 25 Vacuum-based processing of perovskite films results in uniform small crystals that can reach a PCE as high as 20%. 26 Therefore, highperforming PSCs may not always need large grains, and it is rather the nature of GBs that plays a significant role. This is supported by the findings that the perovskite GBs are benign and only generate shallow in-gap defect states, [27][28] or in some cases are reported to facilitate charge dissociation rather than acting as a recombination center. 29 Chu et al. 30 demonstrated via photoconductivity imaging a spatially homogeneous photoconductivity across the methylammonium lead triiodide films irrespective of the location of GBs, evidencing that the microstructure alone does not significantly influence the photoresponse.
Herein, we employ methylammonium lead triiodide (MAPbI3) perovskite-polymer (Polyvinylpyrrolidone) blend films (hereafter termed as MAPbI3-PVP blend, a method to indirectly tune the morphology (grain size and GBs etc.) and investigate its contribution towards the various optoelectronic and device performance properties. We only employ a small amount of PVP into the MAPbI3 bulk to engineer the morphology and to avoid any significant effect on the conductivity. Such blend systems are reported in PSCs, [31][32] and very recently, making breakthroughs in LEDs, [33][34] however, details on the morphology of the blend systems are rather missing. In this article, we provide details on films morphology via scanning electron microscopy (SEM) and atomic force microscopy (AFM) together with Raman imaging.
Light beam induced current (LBIC) and conductive atomic force microscopy (cAFM) imaging allows us to spatially map the microstructure. To the best of our understanding, this is the first report providing spatially resolved characterizations of such systems. We then detail the effect of polymer inclusion in the perovskite bulk by comparing its photoluminescence properties and current-voltage characteristics. Our results show that although polymer additives systematically alter the morphology of perovskite films by creating smaller grain sizes and a higher density of GBs, these blends surprisingly reduce non-radiative recombination within the film.

Perovskite synthesis and device fabrication
The PSCs were made in glove box following a procedure reported before 35

Morphological, structural and optical characterizations of the blend films
The film morphology of the MAPbI3-PVP blend systems was studied by scanning electron microscope (Zeiss CrossBeam 1540XB FESEM) equipped with an in-lens detector. The working distance was 5 mm and the acceleration voltage 5 kV was used for imaging. The crystallography of the films was investigated using X-Ray diffraction (Bruker D8 Discover with Lynxeye XE detector, Cu Kα 0.15418 nm). A FluoTime 300 from PicoQuant was used for photoluminescence (PL) spectroscopy. Perovskite films were loaded into a cryostat (Oxford Instruments) and measured under vacuum (5×10 -5 mbar). The samples were excited with a 405 nm laser (pulse energy ~3 j•cm -2 ). To get insights into the lifetime, time-resolved PL decays were recorded using time-correlated single photon counting (TCSPC) mode.
The Raman analysis on MAPbI3-PVP samples was carried out by employing a Jobin-Yvon-Horiba micro-Raman system (LabRAM ARAMIS) in a backscattering geometry and by focusing the green laser beam (wavelength λ = 514.7 nm) on the perovskite surface. In order to avoid severe laser induced sample degradation occurring during the Raman measurements, we kept the laser power density lower than 26 W.cm -2 . The spot size was ~1 μm 2 while a 1800 lines/mm diffraction grating ensured a frequency resolution of about 0.2 cm -1 . Generally, the Raman signal of the MAPbI3 system presents several limitations 36 such as intrinsic low signal, strong background, complex broad patterns and strong overlapping of the modes with the possible secondary phases (most prominently PbI2). Thus, the acquired raw spectra underwent postacquisition processing, consisting in subtraction of the fluorescence background performed by a polynomial fitting, while spectral de-convolution was carried out by nonlinear leastsquares fitting of the Raman peaks with Gaussian lines shapes.

Characterization of solar cells and stability measurements
The light beam induced current (LBIC) 37 was measured by using a chopper that modulates the light at 133 Hz and via two lock-in amplifiers (to enable measurement of low currents limiting the impact of the noise). The light beam is moved across the sample to scan a 1 mm × 1 mm area of the cell. Spatially resolved photocurrent maps were measured by means of an inverted microscope (Leica DMI 5000) coupled with a monochromator (Cornerstone 130) illuminated by a 200 W Xenon Lamp. The wavelength was fixed to 530nm (±2 nm). A long working distance objective with 100x of magnification yielded a 50 µm × 50 µm spot area. The device area was scanned in steps of 25 µm by an x-y motorized stage. The short circuit photocurrents of the devices were measured by a phase sensitive detection system composed by an optical chopper (133 Hz of modulation) and two digital lock-in amplifiers (EG&G 7265).
The current-voltage characteristics of our solar cells (see Figure S1 for the device schematic) were measured in a N2-filled glovebox with a Fraunhofer ISE certified Si reference diode and a KG5 filter, using a Keithley 2410 source meter. For a reliable measurement reporting, we employed a current voltage (J-V) measurement tracking algorithm to track the temporal maximum power point (MPP) under illumination as described by Zimmermann et al. 38 The shelf-life of the PSCs was measured for 100 days. The devices were stored at ambient (room temperature 22±5  C) and at relative humidity that varied between 50 -70% during the measurement time. For each measurement, only the stabilized PCE value was noted after MPP tracking. which is due to the higher viscosity of the solution due to the long chain polymer, as also reported in literature. 32 Interestingly, we note a reduction in film roughness when the polymer is incorporated in the perovskite films (Figure 1f & SI 3). The root mean square roughness showed a slight drop (from 12.1 for pristine MAPbI3 film to 11.1 for 1 mg•ml -1 sample) due to the small grains of the blend system.

Results and Discussions
The crystal structure of the perovskite films on glass is investigated via XRD (Figure 1 g-j). The XRD patterns of all the samples are identical and all the reflexes could be attributed to tetragonal MAPbI3 perovskite. 39 The prominent reflexes of the (110), (220) and (330) lattice planes were found at 14.13°, 28.45° and 43.22°. The broad feature between 15° and 32° is attributed to the amorphous glass substrate underneath the perovskite film. A comparison of pristine perovskite film with those employing polymer shows broadening of the (220) reflex when PVP is included, suggesting a slight reduction in crystallinity upon polymer incorporation which has already shown to reduce the grain sizes. We note an enhanced full width at half maxima of XRD peaks indicating a reduction in crystallite size (calculated using Scherrer equation, see Table S1) that is in good agreement with SEM micrographs. We note the discrepancy between SSPL and TRPL for the 0.6 and 1 mg•ml -1 samples. We emphasize that term carrier lifetime is rather poorly defined as suggested by Tress 41 and it should be carefully interpreted when referring to a longer (shorter) lifetime. Tress explained that a shorter radiative lifetime is not detrimental and it rather the competition between radiative (RR) and non-radiative recombination (NRR) that matters (ideally the NRR should be as long as possible leading to high radiative yield). This is the case for most efficient perovskite based LEDs in literature. Another important consideration is that the SSPL only provides information on the total RR (at a given time only), whereas the TRPL depicts relaxation behavior of the initial photogenerated charge carrier density (which is a superimposition of both the RR and NRR). 42 It should be noted that while the TRPL measures only the charges recombining radiatively on a purely time-dependent scale, the shape of the transient is dominated by non-radiative recombination, the initial fast decay being assigned to charge trapping whereas the longer tail could be dominated by a slower charge relaxation (probably from the shallowly trapped charges being emitted, reabsorbed and re-emitted). The latter is particularly the case when the slower decay component of TRPL is non-linear and accompanied by a drop in PL intensity (evidencing a reduction in RR). However, when a slower (longer) lifetime accompanied by an increase in the PL intensity is an indication of the dominant RR. Since the TRPL spectrum is typically plotted on a normalized scale, the total PL intensity (as measured in SSPL) should not be anticipated as an indication of longer carrier diminishing time. 43 Whereas, the initial fast decay in the TRPL spectra is dominated by trapassisted recombination, the slower long tail could also arise delayed PL from the shallowly trapped charges. 44 Taking these considerations into account, we suggest that the difference between the slower decay for 0.6 and 1 mg•ml -1 samples should be understood as the different trap density of the two films, being more dominant for the 0.6 mg samples, as also evident from the its lower PL intensity in SSPL (Figure 2a). In order to investigate any structural changes in the MAPbI3-PVP blend systems, we performed a Raman analysis of the films. Prior to obtaining Raman maps, we record optical images of all the MAPbI3 films. Notable is the difference in the morphology of the films depending on the PVP concentration (Figure 3a & Figure S 7). The highest PVP concentration (1 mg•ml -1 ) showed darker extended domains unlike a homogeneous film texture in a pristine MAPbI3 film. We, however, cannot rule out a contribution due to the height difference in the films that may leads to contrast in the images. In order to confirm whether the difference in the contrast arises from height difference or a different MAPbI3-PVP ratio, we recorded Raman spectra of a pristine MAPbI3 film and compared it with the spectra taken at a dark and faded region of the 1 mg•ml -1 sample (Figure 3

Figure 3: (left) Superimposition of scanned area optical image and 130 cm -1 Raman band intensity map of a
MAPbI3-PVP blend sample (1 mg.ml -1 ) taken from 50X ultra long working distance objective, and (rights) Raman spectra acquired on MAPbI3 films without (curve A) and with PVP (1 mg.ml -1 ) at two different regions acquired by focusing the laser beam on faded and dark areas respectively, as indicated in the optical images (curve B and C, respectively).
All the reported spectra showed the typical MAPbI3 vibrational modes, in agreement with recently published Raman investigation on perovskite layers. 36 In particular, at lower wavenumbers, three main vibrational modes are evident (89, 105, and 130 cm −1 ) while the higher wavenumbers spectral region is dominated by the peak at 245 cm −1 and five less intense contributions (210, 290, 355, 380, and 446 cm −1 ). We cannot distinguish any PVP Raman band in all the MAPbI3-PVP blend films, probably due to the low polymer concentration used in the perovskite precursor solution. Theoretical calculations of the MAPI vibrational modes assigned the 89 and 105 cm -1 peaks to the vibrations of the Pb-I cage while the broadest band at 245 cm -1 to the MA cation torsional modes. 46 Differently from the spectra reported in literature, we observed a prominent peak at 130 cm -1 that could be a superimposition between the two peaks usually observed at 122 cm -1 and at 138cm -1 ascribed to the Pb-I cage vibration 47 and to the liberation of the organic cation respectively. 36     In order to further investigate the spatial PVP distribution within the perovskite films, the lateral current distribution was studied using LBIC. Figure 5 (a -d) shows the local current mapping over an area of 1 × 1 mm 2 . The reference device (without PVP) shows a spatial homogeneous current in the range of 52 -55 nA. Surprisingly, the addition of PVP led to a notable increment in the local current; the current raises to 62 -67 nA for 0.3 mg•ml -1 and 0.6 mg•ml -1 based devices, although few spots with reduced current (45 -55 nA) are also visible. The low current spots are between 0.05 mm and 0.1 mm which could be attributed to intermixed MAPbI3-PVP phases. Those local patches of reduced current are even more pronounced at 1 mg•ml -1 PVP concentration, stretching over several hundred microns.
However, we note that PVP does not form agglomerates but is rather disperses in the MAPbI3 films, as in the case of PVP agglomerates, completely insulating domains should have been formed contributing to regions with no current. LBIC measurements ( Figure 5) suggest that the obtained spatial photocurrent is slightly higher when PVP is added to the perovskite. To further investigate spatial conductivity of the films, we recorded cAFM maps of samples with much higher polymer concentrations. We intentionally prepared samples with a higher PVP concentration in order to trace the PVP in MAPbI3 films, and to see if it forms any insulating patches. The images of samples with much higher PVP concentration (5 and 10 mg . ml -1 ) further confirm that PVP does not form large insulating patches ( Figure S 9) and the conductivity is homogeneously distributed across the film, even at a nanometre scale. This proves that no insulating areas are formed upon incorporation of polymer. In order to accurately compare the photovoltaic performance of PSCs made using different concentration of polymer, we carried out a MPP tracking protocol as suggested by Zimmermann et al. 38 Figure 6a displays the PCE of the champion device using PEDOT:PSS as a hole transport layer while 6b shows the average PCE made from at least 20 selected solar cells.
The PCE showed little increase in the beginning but it stabilizes quickly and remained stable during the rest of the tracked time. We note that the average PCE of the lower concentrations of PVP (0.3 and 0.6 mg•ml -1 ) remains almost similar to a reference cell employing pristine  Table S2). The higher VOC in the PSCs made using the blend system affirms a reduction in bulk defects, which act as a recombination center. This high performance in PSCs employing blend systems, despite larger number of GBs is remarkable. It confirms that the number of GBs alone may not contribute to defects density (and non-radiative recombination) but the nature of GBs is more crucial. In the case of pristine MAPbI3 films (with no polymer inside), although large grains are visible (SEM images, Figure 1), a clear deep grain boundary between neighboring grains is distinguishable. Whereas in the case of the blend system, a highly homogeneous microstructure is observed. Finally, in order to compare the shelf-life of the various PSCs, we carried out their shelf-life stability testing. The cells were stored in dark in ambient and measured at room temperature at a relative humidity between 50 -70%. The PSCs employing PEDOT:PSS layer showed a ~ 80% drop in the initial PCE in merely 20 days, whereas the PCE of a reference cell made using NiO (without PVP) dropped only by 25% after 100 days. These results highlight the importance of choosing selective contacts towards long-term stability of the performance. 4 Interestingly, the PSCs made using blend system retained >92% of the initial PCE at similar experimental conditions. Despite the smaller grain size in these devices, their stable performance compared to a NiO reference rival (with large perovskite grains) is remarkable.
Our results suggest high performance can also be obtained using small MAPbI3 grains. The pursuit of large grains is because of the understanding that the grain boundary is a recombination center where dangling bonds form deep levels within the band gap resulting in charge traps. However, an efficient passivation can reduce the deleterious effect 24 . At the same time such passivation can also effect positively the device lifetime. Therefore grain boundaries need to be reconsidered, as it was recently even shown that GBs are not always undesirable. [29][30] For example, Yun et al. demonstrated that GBs are in fact beneficial for charge separation and collection. 48 Similarly, recently shown highest performance (>20%) in planar PSCs employ perovskite films made of small grains. 25 Our results further confirm these latest trends in PSCs.

Conclusions
In conclusion, we provide a systematic investigation of the role of the grain size and grain boundary towards recombination, photovoltaic performance and stability of perovskite solar cells by employing MAPbI3-PVP blend systems. Interestingly we note that the MAPbI3-PVP blend films featuring small sized grains show reduced non-radiative recombination compared to their pristine MAPbI3 rivals. The light beam induced current (LBIC) mapping of the devices measuring photoresponse down to microscopic level shows a higher local current in films employing small grains, corroborating the trends of the photoluminescence transient decay curves. This is in accordance with few recent reports in literature that grain boundaries in halide perovskite are in fact benign and favor charge dissociation and that high efficiency devices do not necessarily require large grains.
In addition, we also, for the first time, provide spatially resolved insights into the morphology of such blend systems via Raman mapping, LBIC and conducting atomic force microscopy.
These spatially resolved techniques lead us to conclude that the polymer disperses in the bulk of the MAPbI3 film and also that it does not form agglomerates nor does it remain on the surface (which would otherwise hinder charge collection). Interestingly, we note that although the addition of low concentrations of an insulating polymer creates small grains (more GBs), it also reduces non-radiative recombination within the film which we attribute to the reduction in the density of bulk by the polymer passivation. We further support our arguments of the reduction of the trap states by an order of magnitude higher photoluminescence yield and a remarkable open-circuit voltage (1.11 V) in the MAPbI3-PVP blend films, ~80 mV higher than a device employing pristine MAPbI3 film. Finally, we also show that the solar cells employing MAPbI3-PVP blend films demonstrate a stabilized nearly hysteresis-free power conversion efficiency of ~17.8% and significantly enhance the long term stability. Our results pave the way for further enhancement of stability without affecting the solar cell performance.

Supporting Information
The Supporting Information is available free of charge on the ACS Publications website at DOI….
Schematic of solar cell architecture, cross sectional SEM of perovskite films, AFM topography of the various perovskite films, absorbance and emission spectra, steady-state and time resolved photoluminescence spectra of the various perovskite films excited through film and glass side as well as a time dependent-PL intensity tracking, superimposition of optical image and Raman mapping of the perovskite films, cAFM images showing local conductivity, currentvoltage characteristics curves, statistics and hysteresis index of solar cells, continuous light soaking test results, and J-V parameters of the champion devices.