Alumina Nanoparticle Interfacial Buffer Layer for Low‐Bandgap Lead‐Tin Perovskite Solar Cells

Mixed lead‐tin (Pb:Sn) halide perovskites are promising absorbers with narrow‐bandgaps (1.25–1.4 eV) suitable for high‐efficiency all‐perovskite tandem solar cells. However, solution processing of optimally thick Pb:Sn perovskite films is notoriously difficult in comparison with their neat‐Pb counterparts. This is partly due to the rapid crystallization of Sn‐based perovskites, resulting in films that have a high degree of roughness. Rougher films are harder to coat conformally with subsequent layers using solution‐based processing techniques leading to contact between the absorber and the top metal electrode in completed devices, resulting in a loss of VOC, fill factor, efficiency, and stability. Herein, this study employs a non‐continuous layer of alumina nanoparticles distributed on the surface of rough Pb:Sn perovskite films. Using this approach, the conformality of the subsequent electron‐transport layer, which is only tens of nanometres in thickness is improved. The overall maximum‐power‐point‐tracked efficiency improves by 65% and the steady‐state VOC improves by 28%. Application of the alumina nanoparticles as an interfacial buffer layer also results in highly reproducible Pb:Sn solar cell devices while simultaneously improving device stability at 65 °C under full spectrum simulated solar irradiance. Aged devices show a six‐fold improvement in stability over pristine Pb:Sn devices, increasing their lifetime to 120 h.


Introduction
Hybrid organic-inorganic metal halide perovskites, as an emerging class of semiconductors, have shown promise as next-generation materials for photovoltaics (PV). [1][2][3][4][5] 10 to 30 meV. [23,24] The consequence of this is a much weaker absorption coefficient near the band edge, and a requirement to have a relatively thick Pb:Sn absorber layer in order to absorb sufficient infrared (IR) light. [11][12][13][14][15] Although the reason is currently unknown, the presence of MA enables thicker and smoother Pb:Sn perovskite films to be processed. However, MA is known to be chemically and thermally less stable than formamidinium (FA) when incorporated into metal halide perovskites, therefore it would be favorable to have an MA-free Pb:Sn perovskite device that could also deliver high efficiency. [25][26][27][28] It is much more challenging to obtain high efficiency with MAfree Pb:Sn perovskites, with only two reports reporting over 20% efficiency. [29,30] For reasons that remain unclear, Pb:Sn PSCs operate much more efficiently in the positive(p)-intrinsic(I)-negative(n) (p-i-n) configuration than in the n-i-p configuration. In the p-i-n configuration, PSCs have the electron-transport layer (ETL) processed on top, which has been optimized to be an extremely thin (<30 nm) layer of phenyl-C 61 -butyric acid methyl ester (PCBM) for solution-processed cells, to negate transport losses, alternatively evaporated C 60 is employed, which also needs to be extremely thin. Even for the Pb-based perovskites, this necessitates extremely smooth underlying perovskite films in order to avoid pinholes in the PCBM layer and consequently direct contact between the metal electrodes and the perovskite film. [31][32][33][34] The direct contact between the top metal electrodes and the top of the perovskite film can lead to a relatively low resistance "shunt pathway" through the device, which can lead to substantial losses in fill factor (FF) and open-circuit voltage (V OC ), respectively. [35][36][37][38] Furthermore, since the dark conductivity in Pb:Sn perovskites is usually orders of magnitude higher than in neat-Pb perovskites, the shunting issue is expected to be exasperated for these narrow-bandgap cells. In addition, this problem is expected to be more problematic for large-area cells with some degree of roughness, where there is a larger probability of shunts forming over the device area. [32,39,40] Therefore, preventing these shunts will be important for scaling-up.
We have historically struggled with obtaining reproducibility in device performance, and suspect that the very thin top layers, coated upon the rough perovskite underlayer could be the primary issue. Here, we investigate the impact of adding a thin layer of alumina nanoparticles (Al 2 O 3 -NPs) inserted in between the thick, rough perovskite (FA 0.83 Cs 0.17 Pb 0.5 Sn 0.5 I 3 ) and the ETL in a p-i-n device structure. Such mesoporous alumina layers have previously been used as "buffer layers" in dye-sensitized and early n-i-p PSCs, to inhibit metal migration from the top-metal contact into the solar cell active layer during aging, [41,42] and alumina nanoparticle layers have also been employed to improve the "wettability" of the perovskite absorber layer coating upon the HTM in PSCs. [43][44][45][46] Following the "buffer layer" approach, our initial intention was to employ a meso-porous layer of Al 2 O 3 -NPs to physically block the direct contact between the rough perovskite layer and the top metal electrode. However, we show that an extremely thin non-continuous alumina nanoparticle layer leads to enhanced conformality of the subsequently coated ETL, leading to increases in V OC , PCE and device stability under continuous illumination.

Results and Discussion
We used a common one-step anti-solvent quench solution deposition technique to fabricate films of the MA-free narrowbandgap perovskite, FA 0.83 Cs 0.17 Pb 0.5 Sn 0.5 I 3 (PV bandgap [E g,PV ] ≈1.27 eV; Figure S12b, Supporting Information [SI]). The bottom half of our solar cell device stack consists of ITO/poly(3,4-ethylenedioxythiophene) polystyrene sulfonate (PEDOT:PSS)/perovskite, with the subsequent electrontransport layer and metal contact being phenyl-C 61 -butyric acid methyl ester (PCBM)/bathocuproine (BCP)/Ag. In an attempt to improve the interface between the perovskite and topcontact layers, we solution-deposit an alumina nanoparticle layer by spin-coating from dispersions on top of the fully annealed perovskite films. The equivalent thicknesses of the resulting alumina nanoparticle layers depend upon the dilution-ratio of a stock dispersion (<50 nm particle size [DLS], 20 wt.% in isopropanol). We investigated different volumetric dilution ratios of the stock Al 2 O 3 dispersion in order to determine the optimal layer thickness: 1:x in IPA, for x = 4, 10,15,20,30,40,50,75, and 100 (the corresponding samples are termed as Al 2 O 3 _x). The details of the spin-coating recipe are shown in Section S1 (Supporting Information) of the SI.
In complete PSCs, we found that the optimum dilution for Al 2 O 3 dispersions was 1:50, corresponding to an equivalent film thickness of only ≈30 nm. We note that we use the term "equivalent thickness" since these films are porous as well as non-continuous at the lower concentration ranges. In Figure 1, we show scanning electron microscopy (SEM) images of complete devices, and atomic force microscopy (AFM) images of perovskite films coated with PCBM, with and without the inclusion of the Al 2 O 3 nanoparticle interlayer with the 1:50 dilution. Figure 1a shows the cross-sectional SEM image of control devices that had the following structure: ITO/PEDOT:PSS/FA 0.83 Cs 0.17 Pb 0.5 Sn 0.5 I 3 /PCBM/BCP/Ag.
The SEM images reveal that there is significant thickness variation in the PCBM layer coating the top of the perovskite film and indicate regions of very close contact between the top metal contact and the polycrystalline perovskite film. In contrast, the cross-sectional morphology of full devices with the Al 2 O 3 -NPs interlayer, ITO/PEDOT:PSS/FA 0.83 Cs 0.17 Pb 0.5 Sn 0.5 I 3 /Al 2 O 3 /PCBM/BCP/Ag, (Figure 1b) exhibits a relatively uniform layer thickness of PCBM between the perovskite film and top metal electrode, with many fewer points of very close contact. To illustrate that this is typical, we show further SEM cross-sectional images of both control and alumina coated devices in Figures S1 and S3 (Supporting Information). We show top-view AFM images of device stacks up to and including the PCBM layer, without and with the inclusion of the Al 2 O 3 -NPs interlayer in Figure 1c,d respectively. In the control films, absent of Al 2 O 3 , there appear to be circular "craters" distributed across the film surface. For the films containing the Al 2 O 3 -NPs interlayer, the craters are not discernible, however, a distribution of features with tens of nanometer length scale are present. In Figure S2   tens of nanometer scale distributed in a non-continuous manner over the perovskite film surface. We identify these to be the Al 2 O 3 -NPs, consistent with the features we observe in the AFM image of Figure 1d.
Furthermore, we performed AFM measurements to examine the surface roughness. First, we compared the root means square (RMS) surface roughness (S q ) of both thin and thick layers of Pb:Sn perovskite films (476.98 and 709.95 nm, respectively, Table S1, Supporting Information) to see how surface roughness varies with perovskite film thickness, noted that thicker narrowbandgap absorber layers are required for all-perovskite tandem solar cells. [11][12][13][14][15] The RMS surface roughness increases from 27 to 43 nm as the thickness of the perovskite layer increases, which is shown in Figure S4 (Supporting Information).
We also measured the surface roughness of thick perovskite films coated with PCBM using AFM. The perovskite films coated with PCBM (Pero/PCBM) are smoother than the neat perovskite film, with RMS surface roughness reducing from 43 to ≈16 nm. This indicates that the PCBM is "planarizing" the surface, which by itself could be beneficial. However, since the roughness reduction is on the same order of magnitude as the PCBM film thickness, it is likely that very thin regions of PCBM coating exist. In contrast, the perovskite films coated with Al 2 O 3 and PCBM (Pero/Al 2 O 3 /PCBM) show a much smaller reduction in RMS surface roughness from 43 to ≈32 nm, consistent with more conformal coating of the underlying perovskite surface. We show the Table S2 (Supporting Information). Figure 1c shows many crater-like shapes on the AFM images. These features are possibly related to bubbles forming during the annealing process of PCBM and may arise from the inhomogeneous thickness of PCBM due to the rough perovskite films. Conversely, when there is an Al 2 O 3 -NPs layer between the perovskite and the PCBM, the crater-like shapes are no longer observed by AFM (Figure 1d). In Figure 1e, we show a simplified illustration of how PCBM may cover the rough perovskite surface with and without Al 2 O 3 -NPs.

RMS roughness values in
Encouraged by the improved conformality of the PCBM coating, we investigated the influence of the Al 2 O 3 -NPs layer on the optoelectronic properties of the perovskite layer and device stack. To do this, we fabricated neat perovskite films and "half-stacks" of perovskite films with hole and electron transport layers, with and without Al 2 O 3 , and measured their photoluminescence quantum yield (PLQY), time-resolved PL (TRPL) decays, and captured PL images (shown in Figure 2; Figures S5-S7). From these results, we can estimate the quasi-Fermi Level splitting (QFLS) in the perovskite absorber layer, which represents the maximum opencircuit voltage the perovskite absorber layer or device stack could generate. The PLQY was measured in an integrated sphere as detailed in Section S1 (Supporting Information) of the SI.
We (PCBM/BCP); glass/pero/Al 2 O 3 /ETL; glass/pero/ETL/Ag; glass/pero/Al 2 O 3 /ETL/Ag. In Figure 2b, the average QFLS, derived from the PLQY, [47,48] for perovskite films on glass with and without Al 2 O 3 -NPs layers were 0.86 and 0.83 eV respectively, which indicates that the coating of the perovskite films with the alumina oxide nanoparticles results in a marginal increase in non-radiative recombination at the exposed surface. As previously reported, [49] adding interfaces with charge transport layers can induce a substantial reduction of the PLQY compared to neat perovskite films, indicating recombination losses. The average QFLS for perovskite films with ETL (pero/PCBM/BCP) and films with HTL (PEDOT:PSS/pero) were 0.76 and 0.75 eV, respectively, denoting presence of significant interface recombination in both cases. When combined with Al 2 O 3 -NPs layers, the QFLS's were also 0.76 and 0.75 eV, respectively for the pero/ETL and HTL/pero stacks, respectively. These findings show that the addition of Al 2 O 3 layers does not significantly impact the PLQY of the perovskite layers when integrated into the complete device stacks. [50] Therefore, from the luminescence studies of the isolated perovskite films on glass, and when contacted by the charge extraction layers, we expect to observe a negligible change in V OC for the cells with the inclusion of the alumina particles.
In order to investigate if the presence of metal electrodes introduces further non-radiative recombination, we assessed QFLS data on half-stack samples of the pero/ETL layer with Ag metal back contacts evaporated on top, which we show in Figure 2b. The control samples with metal back contacts (pero/ETL/Ag) exhibited significant reduction in QFLS in comparison with the pero/ETL half-stacks, whereas this additional QFLS loss is largely reduced in the samples containing Al 2 O 3 -NPs. The significant drop in QFLS following the deposition of the Ag contacts, is likely to arise from some regions of direct contact of the Ag electrode with the perovskite absorber layer. These results are consistent with the presence of the Al 2 O 3 -NP layer inhibiting direct contact of the metal electrode with the perovskite absorber layer, limiting charge recombination with the metal. In Figures S6 and S7 (Supporting Information), PL imaging was used to quantify macroscopic inhomogeneities on perovskite films on glass with and without Al 2 O 3 -NPs. [51] We observe that there are some inhomogeneities in the QFLS maps after Al 2 O 3 -NPs coating. There is a general small reduction in average QFLS on the order of 20 meV, which is similar to what we infer from the macroscopic PLQY results in Figure 2. For the alumina coated samples, there appear to be two distinct regions. As seen in the hysterograms of Figure  S6 (Supporting Information), there are some regions that have a similar QFLS distribution to the uncoated perovskite films, and some have lower QFLS. Since we do not know explicitly why the average QFLS is reduced upon alumina coating, it is difficult to postulate as to the origin of this small but observable increase in inhomogeneity. We tentatively interpret this to indicate that the higher QFLS regions are less well-coated with the alumina.
In order to test the functionality of the alumina nanoparticles interlayer in complete devices we fabricated PSCs in a p-i-n architecture: ITO/PEDOT:PSS/FA 0.83 Cs 0.17 Pb 0.5 Sn 0.5 I 3 / Al 2 O 3 /PCBM/BCP/Ag, as shown in Figure 3a.
We optimized the photovoltaic performance of PSCs based on different dilution ratios of Al 2 O 3 -NPs to IPA, as shown in Figure 3f and Figure S8 (Supporting Information). For our device parameters, we present the maximum-power-point-tracked efficiency and the steady-state J SC and V OC , the latter two determined by holding the cell at zero volts, and recording the current density over time, or by holding at zero mA cm −2 and recording the voltage over time, respectively. The q-FF is an effective FF calculated using, V OC,SS , and J SC,SS since a FF cannot be measured in steady-state, as is shown in Equation 1.  Our control devices have significant V OC loss from this limit, with an average steady-state V OC (V OC,SS ) of 0.58 V. The V OC,SS increases when Al 2 O 3 -NPs layers are incorporated, with the average V OC,SS equal to ≈0.74 V ( Figure S8a, Supporting Information). The maximum-power-point-tracked efficiency ( mpp ) has a similar trend (Figure 3f). Beyond a certain thickness, the insulating Al 2 O 3 -NPs layer leads to the reduction of steady-state J SC (J SC,SS ) and quasi-steady-state fill factor (q-FF) indicating poor charge extraction In Figure 3b-e, we show box plots comparing the mpp , V OC,SS , J SC,SS , and q-FF of devices with and without an optimized thickness alumina nanoparticles layer. In Table 1, we report all photovoltaic parameters of champion and average devices with and without Al 2 O 3 -NPs layer (steady-state, reverse, and forward J-V scan). The average mpp of Al 2 O 3 _50 devices is almost twice as large as that of the control devices, which mainly results from the contribution of the significant increase in V OC and FF. The average V OC,SS increases remarkably from 0.58 to 0.74 V in optimized PSCs, and the maximum V OC,SS we achieved was 0.78 V. For reference, the detailed balance limit of V OC for solar cells with a bandgap of 1.27 eV is 1.02 V. [20,52] Similarly, the average q-FF of the control devices increased from 0.55 to 0.65 after applying the insulating buffer layer.
For completeness, we verified that the effect was not a consequence of the IPA solvent within which the Al 2 O 3 -NP was diluted. We do observe a small, but much less significant improvement upon the solar cell performance after a simple IPA rinse, as shown in Figure S9 and Table S3 (Supporting Information). [53] To examine reproducibility, we fabricated several devices in different batches. As seen in the histogram in Figure S10 (Supporting Information) and the distribution of data points in the box plots in Figure 3b-e, the devices with optimized Al 2 O 3 layers show better reproducibility, i.e., their parameter distributions are narrower, than those of control devices.
The difference between the QFLS inferred from the PLQY of either the pero/ETL or HTL/pero stacks, with the average V OC of the complete solar cells, is 180 mV for the control devices and only 10 mV for the devices with the optimized alumina nanoparticle interlayers. This difference in open-circuit voltage loss of the complete solar cells is consistent with the change in estimated QFLS when metal electrodes are deposited on top of the pero/ETL half-stacks. This indicates that the imperfect top contact region is indeed the main origin of voltage loss in the con-trol devices, which is overcome by the more conformal coating of PCBM facilitated by the Al 2 O 3 -NP buffer layer.
Most control devices have a significant degree of hysteresis, which can be seen in the J-V curve of champion control device in Figure 4a. All else being equal, significant hysteresis is often correlated with lower mpp compared to devices with low hysteresis. [54] Figure S11a  Interestingly, devices fabricated with Al 2 O 3 -NPs layer display reduced hysteresis under identical test conditions. In the early days of investigating hysteresis in ni-p perovskite cells, it was observed that cells that had a complete absence of an electron transport layer and consequently had direct contact between the lead-based perovskites and the electron collection metallic contact (fluorine doped tin-oxide) exhibited extremely severe hysteresis. [55] We thus infer here, that the severe hysteresis in the control devices is due to direct contact between the perovskite absorber layer and the metal top electrode. Figure S12 (Supporting Information) presents the external quantum efficiency (EQE) spectrum and integrated current density of a control device and a device with the optimized thickness of Al 2 O 3 -NP interlayer. As shown in Figure S12  Having demonstrated a clear improvement in the photovoltaic performance of our devices by incorporating alumina buffer layers, we tried to further optimize their efficiency. One method that has been successfully demonstrated for Pb:Sn perovskites uses ethylenediamine (EDA) passivation, which passivates the perovskite surface and decreases the V OC loss. [56] Here we add EDA directly into the Al 2 O 3 -NP dispersion so both could be processed in one-step. Figure 4b shows additional J-V curve data  [57,58] All photovoltaic parameters (steady-state, reverse, and forward J-V scans) from devices with an area of 1.00 cm 2 are tabulated in Table S5 (Supporting Information). All photovoltaic parameters of devices (steady-state, reverse, and forward J-V scans) including average values are shown in Table S4 (Supporting Information). Figure S13 (Supporting Information) shows that applying only the EDA layer for chemical passivation also improves pristine Pb:Sn devices, but the size of the improvement is smaller than using EDA and Al 2 O 3 -NPs together. The EDA and Al 2 O 3 -NPs mixture takes on the role of both a chemical passivation and "conformality boosting interlayer". Therefore, using both EDA and Al 2 O 3 -NPs together synergistically boosts V OC , FF, and PCE.
The long-term stability of PSCs is a key consideration for enabling their real-world use. However, the stability of narrow-bandgap PSCs is limited since mixed Pb:Sn PSCs undergo rapid decay under elevated temperatures and continuous illumination. [25][26][27][28]59] To test the impact of the Al 2 O 3 interlayer upon the long-term stability under light and elevated temperature, we compared three different types of Pb:Sn PSCs: control devices; devices with optimal thickness of Al 2 O 3 layer (Al 2 O 3 _50); and devices with Al 2 O 3 layers as thick as possible that still results in operational cells (Al 2 O 3 _15). We encapsulated 4 cells for each configuration with epoxy-resin and glass cover-slides, and then subjected them to 0.76 mW cm −2 simulated AM1.5 irradiance (generated from a xenon lamp without UV filtering) at 65°C under open-circuit conditions for 120 h (Figure 4c). We implemented light soaking at 65°C in an industry-standard aging box, and detailed information is shown in Section S1 (Supporting Information) of the SI. In Figure 4c we plot the normalized mpp as a function of time. It clearly shows that pristine Pb:Sn devices degraded substantially during the first 6 h of stressing, and the mpp of the control Pb:Sn devices maintained only 10% of their initial mpp after 20 h of light soaking. In contrast, devices with Al 2 O 3 layer demonstrated a six-fold slower degradation rate over 120 h. Al 2 O 3 _50 devices retained >50% of their initial mpp after 20 h of light soaking. These devices showed complete degradation after ≈120 h of light illumination. Moreover, Al 2 O 3 _15 devices showed improved light illumination stability, in comparison with the Al 2 O 3 _50 devices, retaining 60% of the initial mpp after 20 h of continuous light illumination. Both types of devices with Al 2 O 3 layer showed a large improvement of light illumination stability at elevated temperature, compared to pristine devices.
We note that it is quite common to employ evaporated C 60 in place of PCBM in Pb:Sn PSCs, and often also in conjunction with a metal oxide buffer layer, such as SnO X , between the C 60 and the metal top contact. [60][61][62] One may assume that a thermally evaporated C 60 layer should be able to conformally coat the top of the rough perovskite layer and be an alternative solution to the issue we are addressing here. However, we test our devices in air, and have found that Pb:Sn PSCs with thermally evaporated C 60 top contacts drop in performance rapidly when tested in air. Investigating the reason for this stark difference between PCBM and C 60 is beyond the scope of our study here, but it is clearly an important area to address in future studies.

Conclusion
In summary, we have shown through careful optimization that the application of an alumina nanoparticle interlayer, directly between the narrow-bandgap Pb:Sn perovskite absorber layer and PCBM ETL layer, dramatically improves the solar cell device performance and reproducibility. We show compelling evidence that the sparse distribution of alumina nanoparticles on the perovskite surface result in improved conformality of the subsequently coated PCBM layer, reducing the occurrence of close contact between the metallic electrodes and the perovskite absorber layer. As a result, devices that employ this layer achieved a champion mpp of 15.0% versus 10.3% for the control device. With further passivation using a mixture of EDA and Al 2 O -3 NPs, we achieved mpp of 16.5% using a 0.25 cm 2 active area, and up to 15.4% with 1.00 cm 2 active area. The performance spread also showed vast improvement with a median efficiency of 11.5% for devices with Al 2 O 3 -NPs layer versus 7.0% for controls. Device stability under 65°C, simulated full spectrum solar irradiance at open-circuit are also greatly improved. This study demonstrates the importance of morphology at device interfaces and presents a facile method to improve interface conformality and inhibit nonradiative recombination losses and electronic shunts occurring in perovskite PV cells.

Supporting Information
Supporting Information is available from the Wiley Online Library or from the author.