Two‐terminal Perovskite silicon tandem solar cells with a high‐Bandgap Perovskite absorber enabling voltages over 1.8 V

Perovskite silicon tandem solar cells are a promising technology to overcome the efficiency limit of silicon solar cells. Although highest tandem efficiencies have been reported for the inverted p‐i‐n structure, high‐efficiency single junction perovskite solar cells are mostly fabricated in the regular n‐i‐p architecture. In this work, regular n‐i‐p perovskite solar cells with a high‐bandgap mixed cation mixed halide absorber suitable for tandem solar cells are investigated by compositional engineering and the open‐circuit voltage is improved to over 1.12 V using a passivating electron contact. The optimized perovskite solar cell is used as a top cell in a monolithic perovskite silicon tandem device with a silicon heterojunction bottom cell allowing for voltages up to 0.725 V. The tandem solar cells with an active area of 0.25 cm2 achieve record open‐circuit voltages of up to 1.85 V and have efficiencies over 20%. Analyzing the perovskite absorber by spatially resolved photoluminescence measurements shows a homogenous and stable emission at ~ 1.7 eV which is an optimal value for tandem applications with silicon. The tandem solar cells are mainly limited due to a low current. A spectrometric characterization reveals that the perovskite solar cell is current limiting which could be improved by a thicker perovskite absorber.

solar cells, thermalization losses can be reduced. In consequence, for an optimal top cell bandgap of 1.72 eV, the Shockley Queisser limit for a silicon based tandem solar cell is increased to over 42%. 4 Perovskite solar cells are promising as a tandem partner for silicon.
Since their emergence in PV in 2006 5 efficiency has continuously improved and has reached more than 20%, [6][7][8][9][10][11][12][13][14] with a record efficiency of 25.2%. 15 They can be fabricated using solution- 16,17 as well as vacuum- 18,19 and hybrid [20][21][22] deposition techniques. Low-temperature processes are available for all layers, [23][24][25] making perovskite solar cells also interesting in combination with a silicon heterojunction bottom solar cell which would degrade at temperatures over 200°C. Annealing above 200°C results in the effusion of hydrogen, thereby depassivating interface defects at the amorphous/crystalline silicon interface and thus, increasing recombination losses of the silicon heterojunction bottom cell. 26,27 With a strong absorption coefficient and a sharp absorption edge, 28 also the optical properties of perovskites are highly suitable for tandem application. Additionally, the bandgap can be adjusted in a wide range by compositional engineering. [29][30][31] Since the report of the first monolithic perovskite silicon tandem device in 2015 with an efficiency of 13.6% 32 tandem performance was substantially increased to up to a certified efficiency of 28% reported by Oxford PV in the end of 2018. 33 The highest efficiency with published solar cell structure is 26.0%. 34 The decisive factors for reaching high efficiencies with perovskite silicon tandem solar cells are the overall device structure, the absorber material and the adaption of the contact layers.
The highest efficiencies over 25%, so far, have been reached with an inverted p-i-n structure, [34][35][36][37][38] but also more than 22% has been achieved with the regular n-i-p architecture. 39,40 The actual top value for the open-circuit voltage (V OC ) of 1.83 V was first reported for a regular structure. 41 In the meantime, 1.83 V have also been reported for the p-i-n structure. 38 As highest efficiencies for single junction perovskite solar cells are reported for the n-i-p structure, 7-10,14 tandem research on this device polarity remains a relevant research topic.
While the first publications on perovskite silicon tandem solar cells mostly used methylammonium lead triiodide (MAPbI 3 ) as perovskite absorber, 32,[42][43][44][45] recent studies propose perovskites with a higher bandgap approaching 1.7 eV as more suitable. 37,41,46 However, for common high-bandgap compositions based on halide mixing, phase segregation under illumination and low V OC was reported. [47][48][49] Better stability was reached with mixed cations using formamidinium (FA) and caesium (Cs). 29 0 mV compared to control devices without passivation. 53 In this work, we present monolithic perovskite silicon tandem solar cells combining the two technologies allowing for highest efficiencies in single junction solar cells: a silicon heterojunction solar cell and a perovskite solar cell with n-i-p architecture. We use a top solar cell absorber with a tandem-relevant optical bandgap around 1.7 eV. As high-bandgap mixed cation mixed halide perovskites usually face voltage losses due to halide segregation, we focus on voltage improvement as a first step towards high efficiencies in this paper. First, we present tandem-relevant high-bandgap absorbers with varying Cs to Br ratios and implementing a PCBM/PMMA passivation layer in single junction perovskite solar cells. Combining the optimized top solar cell with a silicon heterojunction bottom solar cell with passivating contacts leads to a new record V OC of 1.85 V and an efficiency over 20% on an active area of 0.25 cm 2 . A detailed examination of our tandem device using spatially resolved photoluminescence confirms a homogeneous and remarkably photo-stable perovskite absorber. However, the current of the cells is still relatively low. We further analyze our tandem devices with spectrometric characterization revealing that the perovskite top solar cell limits the current and has a low parallel resistance and discuss issues of the front side contact limiting the overall current.  Figure S1). 54 In the tandem device, this value is slightly lowered to0 .69 V -0.70 V as not the whole light intensity reaches the silicon bottom solar cells and a shadow mask is used to define active cell areas in currentvoltage (IV) measurements ( Figure S2). The bottom solar cell is connected to the top cell via an indium doped tin oxide (ITO) recombination layer. The ETL of the perovskite solar cell consists of compact and mesoporous TiO 2 , both processed at low temperature in order not to degrade the bottom solar cell. The low-temperature process is described in our previous publications. 24,55 PCBM/PMMA is used as a passivation layer. The perovskite absorber is FA 0.75 Cs 0.25 Pb(I 0.8 Br 0.2 ) 3 .
An ITO front side contact was directly sputtered on the Spiro-OMeTAD without any buffer layer. A magnesium fluoride (MgF 2 ) layer was applied to reduce front side reflection. Both ITO layers (interconnection and front side contact) are sputtered through a mask with four openings  Further film characterization is given in the SI ( Figure S4).

| Perovskite absorber composition and electron contact passivation
The composition with 25% Cs and 20% Br shows a~1.5 mA cm −2 higher short-circuit current density (J SC ) than the other compositions.
With PCBM/PMMA, a clear passivation effect can be seen for all absorbers resulting in a significantly higher voltage for the solar cells

| Tandem solar cell results
The overall best performing tandem device achieves a stabilized effi-  Figure S7.
The values shown so far were obtained from laboratory measurements using a non-spectrally adjusted LED sun simulator. After six weeks storage in a nitrogen filled glovebox, the best performing tandem solar cell presented in Figure 3a was measured in the accredited solar cell calibration laboratory at Fraunhofer ISE, CalLab PV Cells. The stabilized efficiency has been determined to (19.91 ± 0.65)% (Figure 4 a). The corresponding calibration report can be found in the SI. This value is clearly lower than in the results presented above. The difference is originating from a by~0.6 mA cm −2 lower J SC and a by3 % reduced FF. As the CalLab PV Cells measurements were carried out over one month later than the measurements shown before, one can assume that the solar cell slightly degraded over that time. We noticed the same trend (lower FF and lower J SC ) also for perovskite single junction solar cells after a few weeks ( Figure S5). To ensure that no significant mistake was made in the laboratory measurements shown in Figure 3 we measured the tandem cell again at the laboratory setup after the CalLab PV Cells measurements and retrieved similar values than the CalLab values shown in Figure 4 (Table S1). This shows that our internal laboratory measurements are well aligned with the calibration standard and thus that the initially measured efficiency of 21.6% is reliable.
From the IV curves one can see that the tandem solar cell has a comparatively low parallel resistance R P . For further investigation a spectrometric characterization was carried out (Figure 4c). A detailed description of this method which is known from the characterization of III-V multijunction cells can be found in literature. 60,61 Basically the spectrum used for IV scans is varied from a red rich to a blue rich spectrum, while the overall effective irradiance is constant, which means that the photo-currents of the individual sub-cells change, but their sum remains constant. For the red-shifted spectrum the top cell is expected to limit the current of the tandem device, and for the blueshifted spectrum the bottom cell. At each measurement point a value can be defined for both sub-cells by the corresponding ratio of the photo current density under the actual spectrum J photo, subcell, actual to the photo current density under the reference spectrum (AM1.5 g spectrum) J photo, subcell, AM1.5g . During the spectrometric characterization the simulator spectrum is varied in a way that the sum of the two subcell ratios J photo, subcell, actual /J photo, subcell, AM1.5g always equals 2.
A clear current mismatch with the top cell limiting the device can be seen from this evaluation as the maximum J SC is reached for a spectrum with an increased blue part (J photo, bottom, actual /J photo, bottom, AM1.5g = 0.9 and J photo, top, actual /J photo, top, AM1.5g = 1.1). The V OC is only slightly affected by the spectral variation. The FF shows an asymmetric behavior with a strong increase with increasing blue part of the spectrum and only a slight increase for red rich spectra. While normally a rather symmetric behavior would be expected, an asymmetry occurs when the current limiting sub-cell has a comparatively low R P . 61 Thus, from the spectrometric characterization it can be concluded that the perovskite solar cell has the lower R P . The photo current of the perovskite solar cell J photo, top is therefore slightly lower than the J SC of the tandem solar cell. This difference is schematically illustrated in Looking at the power output at maximum power point P MP in the spectrometric characterization, one can see that the highest power is generated for current matching conditions. Thus, to improve the tandem solar cell it would be necessary to enhance the current in the   higher J SC up to 17.6 mA cm −2 . 39,40 With a p-i-n structure tandem efficiencies over 25% were achieved. [35][36][37][38] It is noticeable that in all these cases the V OC is indeed lower (between 1.75 V and 1.80 V) than in our case, but the J SC is again clearly higher with values between 17.8 mA cm −2 and 19.5 mA cm −2 , 35-38 although Bush et al. also used a high-bandgap absorber (1.68 eV). 37 Thus, in comparison, the overall current of our tandem devices is rather low, even if current matching could be reached. Here, the front side contact with Spiro-OMeTAD and ITO is the main reason: Incoming light first faces the ITO with a refractive index n of~2 determined by spectral ellipsometry, afterwards the Spiro-OMeTAD (n~1.6 65 ) before entering the perovskite absorber with n~2.4. 66 The reflectance can significantly be lowered in a wide spectral range with the MgF 2 antireflection coating used in our samples ( Figure 5). The integrated reflected current density decreases from 8.1 mA cm −2 to 5.0 mA cm −2 . However, the unfavorable order of refractive indices still leads to strong front side reflection. To further lower front side reflection, a textured foil could be used as in the case of other publications about perovskite silicon tandem solar cells. 40,42,43,67,68 Another option would be the use of silicon bottom solar cells with a textured front side. 69 However, the top cell cannot be deposited by solution processes anymore, but vacuum techniques like evaporation or hybrid processes would be required for all layers.
Additionally, parasitic absorption, especially in the Spiro-OMeTAD featuring the highest extinction coefficient of the whole layer stack, 65 lowers the current of the tandem device. Thus, for significant decrease of optical losses, the hole transport material has to be exchanged. Also current matching could be improved by replacing Spiro-OMeTAD as most of its parasitic absorption occurs in the short-wavelength range. 69 A promising candidate might be copper thiocyanate (CuSCN) which was used in dye-sensitized solar cells before. 70 Perovskite single junction solar cells in the regular n-i-p structure with CuSCN as HTL were reported to achieve 20% efficiency. 71 Pattanasattayavong et al. published a refractive index of~1 .9 and an extinction coefficient < 10 −4 for this material, 72 which is more promising than Spiro-OMeTAD considering both values. The implementation of CuSCN in our perovskite solar cells is currently under investigation.

| Photoluminescence measurements
To further characterize the tandem solar cells, subcell-selective and spatially resolved photoluminescence (PL) measurements were carried out on all four solar cells (cells 1-4) of one substrate. The approach has been reported before 73  Concerning the PL intensity corresponding to the perovskite, a distinct difference can be observed between the four cells. One cell (cell 1) shows a higher intensity than the others (Figure 6a, second row; Figure 6b) whereas the PL intensity corresponding to the silicon solar cell shows the reverse trend (Figure 6a, third row). The difference correlates with the IV parameters as follows: the FF is reduced for cell 1 mainly due to a higher series resistance compared to the other three cells, resulting in a 3% abs lower efficiency (see Figure   S7). An anti-correlation between PL intensity and performance of perovskite solar cells has been reported before. 24,82 In case of contacts with sufficient carrier selectivity, generated charge carriers might be extracted from the perovskite absorber lowering the PL signal.
However, we do not see a difference in the current, only in FF, while in previous publications the PL signal was compared to the current obtained by light beam induced current measurements. 24,82 Regarding the layer homogeneity, the PL intensity maps for both the perovskite and the silicon cells reveal local features. The points with a low PL intensity in both materials correspond to the areas with a high reflectance. The spatially resolved reflectance of the laser used for exciting the silicon shows a clear pattern originating from a spin coating process (Figure 6a, fourth row). Thus, non-uniform reflectance and transmittance in a spin coated layer such as Spiro-OMeTAD might be responsible for the local features in the PL intensity maps.   Table 1 for each composition.
After Spiro-OMeTAD spin coating, the solar cells were finished by evaporation of 80 nm thick gold contacts.

| IV measurement
Current-voltage characteristics of perovskite silicon tandem solar cells were measured using a LED sun simulator (Wavelabs, Sinus-220); the perovskite single junction solar cells presented in Figure 2 were measured with a sun simulator equipped with a xenon short arc lamp and a Keithley 2651A source meter. In both cases light intensity was calibrated to 1 sun under the AM 1.5 g spectrum using a silicon reference cell. All solar cells were first measured in forward scan direction (from −0.1 V to 1.9 V for tandem solar cells and − 0.

| CalLab PV cells measurements
A shadow mask was used during all measurements to define an active area of (0.249 ± 0.013) cm 2 . The spectral response of the perovskite silicon tandem solar cell was measured with a laser-based measurement setup. 83 Two different selective bias-illuminations and required bias-voltages were used, to resolve both sub-cells. 84 The bias-settings were adjusted to minimize measurement artefacts. [85][86][87] The For an efficiency certification we measured the IV curve in both directions (V OC ➔ I SC and I SC ➔ V OC ) within approximately one minute per IV curve. Then the cell was set to the determined P mp and the maximum was followed by variation of the voltage. The current and voltage was logged for 5 min (steady state measurement). At the end we repeated the IV scan in both directions. The IV-measurement and the 5 min P mp -Log are stated in the certificate.
For the spectrometric characterization the mean value of the IVscans in both directions was chosen, as the focus was on analyzing the characteristic behavior regarding spectral variations. The metastable behavior changed slightly with the spectral variation, but is of minor importance on the outcome showed here.
A detailed description of the measurement uncertainties can be found in literature. 88 Reflectance measurements were carried out using a Lambda 950 spectrophotometer from PerkinElmer equipped with an integrating sphere to be able to detect also diffusely reflected light. Samples were measured in a wavelength range of 250 nm -1200 nm with a step size of 2 nm.

| Photoluminescence spectroscopy mapping
SEM pictures were taken using a Schottky emission scanning electron microscope (Zeiss, Auriga 60) at 3 kV. In advance samples were broken and a thin platinum layer was sputtered on the cross section to have a good conductivity.
Grazing incidence X-ray diffraction was performed with a X'Pert MRD system (PANalytical) equipped with a copper anode X-ray source (Kα1 = 1.5406 Å). Scans were recorded within the 2θ-range of 5°to 65°with a step size of 0.05°and an integration time of 1 s.
Refractive index n of ITO was determined by spectral ellipsometry using a J. A. Woollam M-2000 ellipsometer. An oscillator model containing a Gaussian and a Lorentz oscillator was fitted to the measured data using the CompleteEASE software.