Nanoscale Size Control of Si Pyramid Texture for Perovskite/Si Tandem Solar Cells Enabling Solution‐Based Perovskite Top‐Cell Fabrication and Improved Si Bottom‐Cell Response

A monolithic perovskite/silicon tandem solar cell architecture is employed to surpass the single‐junction efficiency limit. Recently, there is an increasing need for the double‐sided textures in the Si bottom cell to be compatible with the solution‐processed perovskite top cell from an industrial perspective. Herein, a silver‐assisted alkaline etching method is applied to fabricate nanoscale Si pyramid textures, and the influence of varying pyramid size (400–900 nm) on the interface morphology and the performance of perovskite/Si tandem cells is investigated. It is demonstrated that electrical shunting starts to increase, and the open‐circuit voltage (VOC) decreases when the texture size exceeds the perovskite thickness (~500nm) due to the non‐uniform top‐cell formation on a rough Si surface. However, when the texture size is reduced to 400–500 nm, all spin‐coated perovskite top‐cell component layers exhibit an even form over the nanopyramid Si, resulting in a high VOC and an enhanced Si bottom cell current (≈1.0 mA cm−2) due to the suppressed reflectance at the top/bottom cell interface without using optical couplers. The double‐sided nanopyramid Si texture offers opportunities to increase tandem cell efficiency while reducing its production cost compared with the commonly used single‐sided textured Si.


Introduction
Currently, crystalline silicon (Si) photovoltaic (PV) technology accounts for ≈ 95% of the worldwide PV market due to DOI: 10.1002/admi.202300504 the abundance of Si, long-term reliability, relatively high power conversion efficiency (PCE), and the continuous reduction in production cost. [1]Thus, Si-PV is expected to provide an economical and sustainable solution as a major renewable energy resource toward a decarbonized society.Meanwhile, the PCE of the state-of-the-art Si solar cell (>26%) [2,3] is approaching its single-junction efficiency limit (29.4%) [4,5] as a result of continuous research in recent years.Stacking a wide bandgap material on top of the Si subcell to form a tandem device architecture is one of the routes to overcome the fundamental efficiency limitations of single-junction Si devices.Lead halide perovskites [6,7] are a promising class of materials for tandem integration with Si in terms of bandgap matching and tunability (1.5−2.3 eV), high efficiency (25.7% in single-junction devices [8,9] ) and potential for low-cost production using solution-based deposition processes.Recently, a PCE of 33.7% has been achieved experimentally in a perovskite/Si tandem solar cell, [10] which far exceeds the limit of Si solar cells and is a great leap toward the realization of highenergy density PV.
Amongst various tandem design architectures, (i.e., two-, three-, and four-terminal devices), [11,12] the two-terminal tandem solar cell-a monolithic series connection of two subcells via a recombination junction-is the most compatible architecture for the current Si PV technology.In two-terminal architecture, most of the currently reported perovskite/Si tandems are based on Si bottom cells featuring a front-planar rear-textured structure to ensure the growth of high-quality perovskite layers by solution-based processes such as a spin coating method.Besides, an indium-tin-oxide (ITO) layer is commonly used as a recombination junction layer for electrical interconnection between the top and bottom cells, which often causes a high interfacial reflection and thus an optical loss in the bottom cell. [13,14]o mitigate the interfacial reflection, an optical coupler based on ITO/hydrogenated nanocrystalline silicon oxide (nc-SiO x :H) (≈100 nm) bilayer design was proposed, [14][15][16] leading to a PCE of 29.7%. [17]However, depositing such a thick nc-SiO x :H layer is a challenge in the industrial production due to its low deposition rate.The use of an ITO interlayer is also an issue because it requires an additional physical vapor deposition (PVD) process as well as involving the use of a scarce metal (i.e., indium).
[20][21][22][23][24][25][26] This is also motivated by the fact that the double-sided textured Si is widely used in industrial scale production.There are two approaches of implementing double-sided Si textures in tandem devices; one is a micron-sized Si pyramid (micro-pyramid) texture prepared by the standard anisotropic etching in alkaline solutions.[21][22] Alternatively, the perovskite layer can be formed on such a micropyramid texture by solution processes including spin coating [23] and blade coating [24] if the perovskite layer is sufficiently thick (typically > 1 μm), whereas this might suffer from a trade-off between the absorber thickness and the carrier diffusion length.
[27] Tockhorn et al. demonstrated a PCE of 29.8% in a perovskite/Si tandem cell by introducing sinusoidal periodic textures with a size ≈500 nm fabricated by nanoimprint lithography. [26]Ying et al. presented a 28.2% efficient perovskite/Si tandem solar cell by using a very small sized Si texture (≈<100 nm) fabricated by the metal-assisted etching technique followed by the surface reconstruction process. [27]However, these nanoscale texturing techniques are applied only for the front side Si wafer while the rear side micro-pyramid texture needs to be prepared in a separate process.Thus, an asymmetric Si texturing approach as such might not be suitable for industrialscale production.
Given the situation mentioned above, a Si pyramid texture with a size <1 μm which can be fabricated by the simple alkaline etching process is highly desired to make the Si bottom cell compatible with the subsequent solution-based perovskite deposition processes, which is economically viable and is attractive for largescale industrial production.However, there are limited reports on the practical applications of the solution-processed Si nanostructures in perovskite/Si tandem solar cells. [28]In particular, there is still a lack of detailed research regarding the actual impact of the size and morphology of nanoscale Si textures on the interface properties and performance of perovskite/Si tandem solar cells.
Recently, we have developed an original methodology to fabricate Si nanopyramid textures using a silver(Ag)-assisted one-step etching process. [29]This allowed us to vary the average size of the Si pyramids from >1 μm down to ≈500 nm by adjusting the concentration of silver nitrate (AgNO 3 ) in the etching solution.The double-sided textured Si wafers were subjected to our standard Si heterojunction (SHJ) cell process and an improved solar cell performance was demonstrated using the nanopyramid texture with an average size of ≈600 nm.In this work, we extend our methodology to apply the nanostructured SHJ bottom cells to perovskite/Si tandem solar cells that feature an interlayer-free design and all spin-coated top cells.The results show that the highly efficient tandem solar cells are obtained when the Si pyramid size is ≈ 400-500 nm, showing a significant improvement in the bottom cell response compared with the devices with frontplanar rear-textured Si bottom cells.This technology enables both the uniform deposition of perovskite top cell component layers by solution processing and the significant improvement in light absorption of tandem solar cells without implementing the complex and thick optical multilayers at the top/bottom cell interface, which is expected to provide guidance for the interface design of double-sided textured Si for industrial perovskite/Si tandem solar cells.

Fabrication of Si Nanopyramid Texture
We used an Ag-assisted etching process to fabricate Si nanopyramid textures. [29]This process allows us to control the size of the Si pyramids at the nanoscale by adjusting the concentration of AgNO 3 in the alkaline solution.We fabricated differently sized Si nanopyramid textures under three surfactant (Hayashi Pure Chemical Ind. Ltd., TK81) concentrations of 50%, 30%, and 10%.The AgNO 3 concentrations of 1.0 or 1.2 mm were added to the solution.As shown in the top-view scanning electron microscope (SEM) images in Figure 1a (additional SEM images are shown in Figure S1 (Supporting Information), the average size of the Si pyramids decreases from ≈900 to ≈400 nm with increasing TK81 concentration.A conventional micro-pyramid texture was also prepared under the standard etching condition using pottasium hydroxide (KOH), TK81, and an additional additive (Hayashi Pure Chemical Ind. Ltd., TT72). [29]Here we term these samples and corresponding devices by referring to the average size of each texture (A 420nm , B 530nm , C 680nm , D 950nm , and E 1050nm ).It is noted that the homogeneity of the size distribution is improved, as shown in the corresponding histogram in Figure 1b, as the texture size becomes smaller.
The following is a brief explanation of the etching mechanism of the Ag-assisted alkaline etching method used in this study.First, the Si surface is plated with Ag nanoparticles which act as a uniform and dense etching mask.At the same time, anisotropic etching of Si occurs at the Ag-Si contact site, forming a Si nanopyramid texture.Hydrogen (H 2 ) bubbles formed during etching are preferentially generated on the Ag particle site due to their high electronegativity, leaving the Si-etchant contact area and enhancing the etching rate.Such an intensive reaction and generation of H 2 bubbles are controlled effectively by adding TK81.The TK81 acts as a surfactant to promote the detachment of H 2 bubbles and also suppress the excess reaction.From our observation, we conclude that the combination of fine and dense masking and an appropriate etching rate is crucial to obtaining the precise size control of Si pyramids with good uniformity.It should be mentioned that the Ag consumption for Si nanotexturing used in this work is estimated to be 8 mg W −1 , which is lower than that for the Ag screen printing (25 mg W −1 ). [30]We expect that this amount can be further reduced by process optimization.In addition, Ag could be replaced with other base metals like copper and nickel, which are known to be applicable to metal-assisted Si etching.The size of the pyramids influences the spectral reflectance (R) of the Si wafer, as shown in Figure 2a.As the texture size decreases, the R in the wavelengths of 400-1000 nm first decreases slightly and then increases.It is noted that nanopyramid textures C 680nm and D 950nm outperform the standard micropyramid texture E 1050nm in terms of the absorbance of the Si wafer (1-R) among the samples investigated.This is in good agreement with our previous work and attributed to the improved light incoupling effect by the sharp nanopyramids with uniform size distribution. [27]For smaller pyramids (A 420nm and B 530nm ), the R starts to increase when the wavelength is greater than the average size of the Si pyramids particularly in the long wavelength region.This is because light in-coupling by random pyramids is based on ray optics and works well only if the size of the pyramids is sufficiently large with respect to the wavelength of the incident light.If the size of the pyramids becomes smaller than the wavelength of the light, the light in-coupling efficiency therefore decreases significantly which results in less absorption within the Si wafer. [31]In addition, the light trapping effect becomes less efficient with the decrease in texture size, resulting in the blue shift of the absorption edge at ≈1100 nm as seen in Figure 2a.Nevertheless, the R of the samples A 420nm and B 530nm is still much lower than that of the double-sided planar Si as well as the front-planar rear-textured Si.
To investigate the influence of texture size on the spectral sensitivity of Si solar cells, external quantum efficiency (EQE) measurement was performed for the SHJ bottom cells without forming any top cell component layers.Instead, an ITO layer and  an Ag-grid were deposited on top of the cell precursors that have n-Si absorber sandwiched by undoped/doped hydrogenated amorphous Si (a-Si:H) and a following doped hydrogenated nanocrystalline Si (nc-Si:H) layer on the front side, forming a test cell structure of Ag-grid/ITO/n-nc-Si:H/p-a-Si:H/i-a-Si:H/n-Si/i-a-Si:H/n-a-Si:H/ITO/Ag. [32]Figure 2b shows the EQE curves of the SHJ solar cells with differently textured Si wafers.The solar cell parameters of the representative devices (A 420nm , E 1050nm , reference) are given in Table S1 (Supporting Information).In Figure 2b, it is evident that the double-sided textured Si provides a much higher spectral response in the wavelengths of 600-1200 nm compared with that of the reference front-planar rear-textured Si because of the reduced reflection.For doublesided textured Si cells, in contrast to the reflectance spectra, the variation in the EQE spectra is less pronounced, regardless of the size of the pyramid textures.This is because the top ITO layer on textured Si works as an ideal broadband antireflection (AR) layer which suppresses the variation of the optical properties in these wafers.However, a slight redshift of the EQE curve can be seen for wavelengths of >1050 nm with an increase in the size of the pyramid texture, indicating that the light trapping effect near the bandgap wavelength is enhanced by enlarging the pyramid size.On the other hand, the EQE in the short wavelengths ≈ 300-650 nm decreases with decreasing the texture size, which probably originates from the absorption loss due to the thickening of the component layers (a-Si:H, nc-Si:H, ITO) because the deposition time of these layers was not adjusted with respect to the texture size.

Tandem Cell Performance
Next, we fabricated monolithic two-terminal perovskite/Si tandem solar cells using the differently textured Si bottom cells shown above.The structure of our perovskite/Si tandem solar cells is shown schematically in Figure 3a.The bottom cell was prepared in the front-junction configuration.On top of the p-a-Si:H hole contact layer, an n-nc-Si:H recombination layer was grown in situ by plasma-enhanced chemical vapor deposition (PECVD), enabling transparent conductive oxide (TCO)-free interconnection between the top and bottom cells. [33]Then, the perovskite top cell was deposited in the n-i-p deposition sequence starting from tin oxide (SnO 2 ) (≈50 nm) as electron transport layer (ETL), perovskite as light absorber layer (500-600 nm) and doped 2,2′,7,7′-Tetrakis [N,N-di(4-methoxyphenyl)amino] −9,9′spirobifluorene (spiro-MeOTAD) (≈200 nm) as hole transport layer (HTL).The bandgap of the perovskite absorber layer used in this study is ≈1.63 eV which provides reasonable current matching in the present tandem device design.These n-i-p stacked layers were deposited entirely by spin coating.Finally, ITO and Aggrid front electrodes were made by sputtering through shadow masks.No buffer layer was used between the spiro-MeOTAD and ITO layers.The details of our tandem device design and fabrication process can be found elsewhere. [33]igure 3b shows the current density-voltage (J-V) curves of the perovskite/Si tandem cells with differently textured Si bottom cells.For comparison, a tandem cell with a front-planar rear-textured Si bottom cell was also prepared as a reference.It can be seen from the J-V curves that the solar cell performance is significantly affected by the texture size of the Si bottom cell.In particular, there is a substantial decrease in V OC from 1.75 to 0.89 V with enlarging the texture size from ≈500 to ≈1000 nm.The anomalously high short-circuit current density (J SC ) (>19 mA cm −2 ) seen for devices D 950nm and E 1050nm can be attributed to both the electrical shunting through the top cell and the non-uniform perovskite layer formation on large textures as discussed below.In contrast, the open-circuit voltage (V OC )of device B 530nm remains as high as that of the reference sample.The J SC of device B 530nm is higher by 0.8 mA cm −2 compared to the reference cells.Overall, this results in an increase in the solar cell efficiency from 19.0% to 20.3%.Unexpectedly, device A 420nm exhibits poorer performance in both V OC and J SC .This originates from the degradation of the bottom cell rather than the decrease in the top cell performance.It was found that the a-Si:H passivation quality is strongly dependent on the PECVD conditioning particularly when being deposited on such a small-sized pyramid texture. [34]As shown in Table S1 (Supporting Information), we have attained a reasonably high performance singlejunction device using the Si substrate A 420m .However, after making a minor change in the deposition reactor of the intrinsic a-Si:H layer, we have subsequently been unable to reproduce goodquality a-Si:H passivation in our nanotextured devices.Although we have not yet found which PECVD deposition parameter is key in passivating the nanopyramid surface like A 420nm , such a technical issue can be solved by the minor tuning of the deposition parameters.
The dependence of the device performance on texture size can be understood by analyzing the cross-sectional SEM images of these devices shown in Figure 3c.For devices A 420nm and B 530nm , it is seen that these nanopyramids allow the perovskite layer to be uniformly deposited on the textured Si surface by the spin coating method.It is also found that the top of the perovskite layers is planarized when formed by such a solution-based deposition process, which is not particularly ideal from an optical point of view.However, an AR monolayer coating can effectively reduce the optical reflection at the top-cell surface, as shown later.On the other hand, there seems to be a local thickness non-uniformity in the SnO 2 layers, which will be evidenced by a transmission electron microscope (TEM) analysis below.For device C 680nm , the peak of the large pyramid almost reaches the top surface of the perovskite layer, at which the thickness of the perovskite layer is very thin (≈50 nm).This might cause a local electrical shunting path, accounting for the decrease of all solar cell parameters.For devices D 950nm and E 1050nm , due to the excessive size of the pyramids, none of the constituent layers of the top cell uniformly covered the substrate.The perovskite and spiro-MeOTAD layers are punched through by some of the large pyramid peaks.This results in considerable electrical shunting and thus the photogenerated carriers of the Si bottom cell can be collected by the top electrode without carrier exchange via the recombination junction layer.Also, the average thickness of the perovskite layer becomes thinner, explaining the very high J SC (≈20 mA cm −2 ) solely generated by the Si bottom cell, as shown in Figure 3b.As shown in the top-view SEM images in Figure S2 (Supporting Information), it is confirmed that the density of the uncovered pyramid tip increases from D 950nm to E 1050nm while no uncovered area is identified for samples A 420nm , B 530nm , and C 680nm .
In Figure 3c, it can be seen that the height of the pyramids of each sample is smaller by a factor of ≈0.6-0.7 compared with their average lateral size as measured by the top-view SEM images, which originates from the pyramid angle between the Si {111} facet and the Si {100} plane.Given that device B 530nm shows the best solar cell performance, the height of the pyramids being as small as the thickness of the perovskite layer (500 nm) and its tight size distribution are the prerequisites to benefit from texturing Si while preserving the good electrical properties of the top cell.We note that the solar cell performance of devices C 680nm , D 950nm , and E 1050nm is expected to be improved by simply thickening the perovskite layer well above the texture size.However, there must be a trade-off between the thickness of the perovskite layer and the carrier diffusion length.
Figure 3d compares the EQE spectra of the tandem cells made with a front-planar rear-textured Si (reference) and with an optimally textured Si featuring an average size of 530 nm (B 530nm ).It is clarified that the J SC gain by applying the nanopyramid texturing originates from the EQE improvements of the bottom cell in the wavelength range of 800-1050 nm thanks to the elimination of the optical interferences in the reflectance spectrum.The increase of the bottom-cell current J bottom , which can be calculated by integrating the product of the EQE curves and the air mass 1.5 global solar spectrum, is 0.9 mA cm −2 .On the other hand, no obvious change is found in the top cell spectral response or top-cell current J top by the Si texture although the interference bumps are slightly reduced in the wavelengths of 400-750 nm.As a result, the increase of J bottom is roughly matched with that of J SC , indicating that J SC of the reference tandem solar cell is limited by the J bottom .
Figure 4 shows the cross-sectional high-angle annular field scanning transmission electron microscopy (HAADF-STEM) images of device B 530nm .Similar to the SEM images shown in Figure 3c, the perovskite layer is formed on a rough surface while its top surface is planarized.The cracks found in the perovskite layer and the voids (black holes) seen in both the perovskite and spiro-MeOTAD layers are most likely created by mechanical damage during the TEM sample preparation, as no such features are found in SEM images.In the magnified images, it can be seen that both i-p-a-Si:H (≈10 nm) passivating contact layers and n-nc-Si:H (≈25 nm) recombination junction layer grown by PECVD provide excellent conformal coverage on the Si nanopyramid surface even at the peak and valley positions.It is also clearly seen that surface roughing occurs in the n-nc-Si:H layer with a roughness height of ≈10 nm.This small surface roughening was found to provide an antireflection effect at the SnO 2 /Si interface in the case of a planar Si surface. [33]We observed a preliminary result that this favorable effect is still present in the case of depositing n-nc-Si:H layer on the nanopyramid Si, although the effect becomes less pronounced (Figure S3, Supporting Information).On the other hand, the SnO 2 layer formed by spin coating shows relatively poor surface coverage.In particular, the thickness of SnO 2 at the valley position is ≈100 nm which is thicker by a factor of two than that on the planar surface.In contrast, almost no SnO 2 layer is deposited at the tip of the pyramid.This poor coverage of SnO 2 layer can also be identified by the top-view SEM images of the Si bottom cells with a SnO 2 layer spin-coated on top (Figure S4, Supporting Information).As the size of the Si pyramid texture increases, the area without SnO 2 coverage gradually increases, which potentially results in creating shunting paths or uneven electron collection in the top cell.Nevertheless, at least for samples B 530nm , it is worth noting that this locally uncovered SnO 2 layer does not result in significant shunt or charge collection loss judging from the J-V and EQE results.Since we do not use a highly conductive recombination junction layer like ITO, this might help in preventing the local shunt creation.Alternatively, the application of a vacuum-processed SnO 2 such as by atomic layer deposition [35] would be a solution to achieve complete coverage of the SnO 2 layer over the textured Si.
Next, we applied an AR coating on top of the tandem solar cells to clarify if the gain in the J bottom by Si texturing can still be the effect of the Si nanopyramid texture on the light in-coupling in the bottom cell can still be preserved after applying an AR layer on top of the ITO layer.As a result, a gain in the J bottom by ≈1 mA cm −2 is obtained, leading to a 22.1% efficient tandem solar cell (J SC = 18.8 mA cm −2 , V OC = 1.693V, fill factor (FF) = 0.692, area = 1.0 cm 2 ).In terms of PCE, this tandem cell performs slightly less efficiently compared to the reference cell made in the same batch (J SC = 17.5 mA cm −2 , V OC = 1.812V, FF = 0.730, PCE = 23.2%,area = 1.0 cm 2 ) because the overall loss in V OC and FF is more than the gain J SC .The V and FF losses come from the inferior top cell performance due to the excessive size of the nanopyramids used in this particular sample (B 530nm ).In fact, V OC as high as 1.81 V has been obtained using the nanopyramid texture A 420nm , when the superior a-Si:H passivation was achieved (not shown).Therefore, it is believed that the tandem cell performance can be further improved by using a nanopyramid texture like A 420nm in the future.In addition, since our baseline process for the perovskite top cell is not yet well-established, even higher efficiency is expected with further optimization.
The tandem cell performance presented here is still limited by the low J SC (18.8 mA cm −2 ) compared to the J SC (≈20 mA cm −2 ) of the state-of-the-art tandem cell. [22]This gap mainly originates from the different top-cell configurations.In the n-i-p configuration used in this study, the refractive index mismatching at the interface between the ITO (n ≈ 2.0) and the thick spiro-MeOTAD (n ≈ 1.6) layers causes a large reflection loss, whereas the recent tandem cell based on inverted p-i-n configuration has no such an optical issue.Despite the inferior optical properties, the n-i-p configuration presented here still offers highefficiency potential by implementing a thinner HTL layer.For example, the replacement of spiro-MeOTAD with similar derivatives such as evaporated 2,2′,7,7′-tetra(N,N-di-p-tolyl)amino-9,9spirobifluorene (spiro-TTB) with much lower thickness (≈25 nm) has been demonstrated in perovskite/Si tandem solar cells with J SC ≈ 19.5 mA cm −2 and efficiencies ≈27%. [36]The simpler cell design and deposition process for n-i-p configuration than p-i-n configuration would be advantageous in industrial applications.
In terms of the nanoscale texture design, a similar approach has been reported where the Si nanopyramid texture with an average size of 500 nm was used for solution processing of perovskite top cells in monolithic two-terminal perovskite/Si tandem solar cells. [28]The size of the pyramid texture they used agrees well with our optimal texture size (i.e., A 420nm or B 530nm ) for perovskite deposition.Nevertheless, their results show no distinct improvement in either top-cell or bottom-cell current, which is different from our observation.This is probably because they have employed refractive-index matching layers comprised of ITO (20 nm) and n-nc-SiO x :H (90-110 nm) double interlayers which effectively suppress the internal reflection at the interface between the top and bottom cells.In contrast, our results prove that the nanopyramid texture provides a similar AR effect without using such a multilayer coating that requires an additional PVD process and a very long deposition time for a thick n-nc-SiO x :H layer by PECVD.Thus, the biggest advantage of our device architecture is the TCO-free interface design which offers the opportunity toward realizing industrially feasible high-efficiency tandem solar cells.The use of nanopyramid texture on both sides of the Si wafer is also beneficial in improving tandem solar cell performance without additional cost, if compared to the use of single-sided textured Si or asymmetrically textured Si that has been used for perovskite/Si tandem solar cell development at a laboratory scale.
Finally, we address the possible industrial solution processes for the fabrication of the perovskite layer.In this study, we employed the spin coating method as it is most widely used in the laboratory.However, it is known that spin coating is not industrially feasible because of the large waste of the precursor materials and the limitation for the large-area uniform deposition.Instead, various industrial solution processes have been proposed such as blade coating, [37] spray coating, [38] inkjet printing, [39] etc.Although the compatibility of these processes needs to be investigated, it is expected that the nanoscale texture developed here can be used for these industrial solution processes if the same/similar precursor solution is used.

Conclusion
We explored the effect of size variation of Si nanopyramid textures of the bottom cell on the overall perovskite/Si tandem cell performance.Double-sided Si nanopyramid textures having different average sizes of 400-900 nm with an improved size distribution were fabricated using an original Ag-assisted alkaline etching method and applied in the bottom cell of perovskite/Si tandem cells.As the size of the Si pyramid increases, the light absorption of the bottom cell gradually increases, while excessive pyramid size (>600 nm) causes severe electrical shunting and thickness inhomogeneity in the perovskite absorber layer, resulting in the degradation in the performance of tandem cells, particularly in the V OC .We find that the optimum Si texture size is ≈ 400-500 nm, by which the perovskite top cell can be processed entirely by the conventional spin-coating method without creating an electrical shunting path in the top cell.Compared to the front-planar rear-textured reference Si bottom cell, the doublesided textured Si bottom cell with an average size of 530 nm provides a J SC improvement by ≈1.0 mA cm −2 and a higher PCE as a result of the increased spectral response in the Si bottom cell due to the suppressed interfacial reflection between the top and bottom cells.The results obtained in this study show the great potential for cost-effective tandem cell manufacturing using the solution-based top cell process without the costly multilayer optical coating nor the lithography patterning step at the top/bottom interface.

Experimental Section
Fabrication of Si Nanopyramid Texture: Mirror-polished n-type float zone (FZ)-grown monocrystalline Si wafers (1-5 Ω • cm, <100>-oriented) with a thickness of ≈280 μm were used for the Si bottom cells.After cleaning processes with acetone and deionized water, the planar Si wafer was textured by dipping in a mixed solution containing the following components: KOH, surfactant (Hayashi Pure Chemicals Ind. Ltd., Pure Etch TK81), AgNO 3 solution of hydrofluoric acid (HF) (46%−48%): AgNO 3 : deionized water (10 ml: 0.1 g: 40 ml).For the conventional micro-pyramid textures, a planar Si wafer was textured by the mixed solution using an organic masking material (Hayashi Pure Chemicals Ind. Ltd., Pure Etch TT72,) instead of AgNO 3 solution.After etching, the samples were immersed in HNO 3 solution to remove the residual Ag.The surface morphology of the Si substrates was characterized by field emission SEM (Hitachi High-Tech S-4300 SEM).In this study, the bottom width of over 150 pyramids was measured and used their average value to represent the pyramid size of each texture.To characterize the optical performance, the reflectance of the samples was measured by UV-vis-NIR spectrophotometer (PerkinElmer, Lambda 950) equipped with an integrating sphere (Labsphere, 150 mm RSA ASSY).
Fabrication of the Si Bottom Cells: SHJ bottom cells were fabricated using the same Si wafers.Si nanopyramid textures with different sizes were formed on both sides of the Si wafer by the Ag-assisted alkaline etching process described above.The Si wafers were subjected to our standard cleaning process. [40]No special cleaning was done after the Ag-assisted alkaline etching process.After dipping the cleaned Si wafer into a dilute HF solution, i-a-Si:H/n-a-Si:H and i-a-Si:H/p-a−Si:H layers were deposited on the rear and front sides of the Si wafer, respectively, in a multichamber PECVD system. [41,42]In this study, phosphorous doped n-nc-Si:H was deposited as a recombination junction layer on top of the p-a-Si:H layer. [33]rior to depositing n-nc-Si:H, a carbon dioxide (CO 2 ) plasma treatment was carried out on the underlying p-a-Si:H layer for 10 s to facilitate the nucleation of the nc-Si:H growth. [43]Then, ITO−Ag stacked layers were deposited on the rear side of the wafer without a mask by magnetron sputtering.
Fabrication of the Perovskite Top Cells: The front surface of SHJ bottom cells was treated by UV-ozone for 3 min at the maximum available power.An SnO 2 nanoparticle colloidal solution was diluted with deionized water, and then 300 μL of the solution was spin-coated (2000 rpm for 30 s) followed by annealing at 100°C on a hot plate in dry air for 1 h.The substrates were then treated with UV-ozone again for 10 min.Rb 0.05 (FA 0.83 MA 0.17 ) 0.95 PbI 0.83 Br 0.17 perovskite [44] was prepared by dissolving 1.4 m PbI 2 , 0.25 m PbBr 2, 0.09 m RbI, 0.25 m methylammonium bromide (MABr), and 1.26 m formamidinium iodide (FAI) in a 4:1 (V:V) mixture of dimethylformamide (DMF) and dimethylsulfoxide (DMSO).After mixing, the perovskite precursor solution was shaken overnight (≈20 h) at room temperature and then filtered with a 0.45 μm pore size PTFE filter immediately before use.An amount of 40 μL of the solution was pipetted onto the substrate and spin coated in a two-step program: 1300 rpm for 5 s with 200 rpm s −1 ramp and then 5000 rpm for 30 s with 2000 rpm s −1 ramp.Ten seconds before the end of the second step, 300 μL of anhydrous anisole was pipetted onto the rotating substrate.The substrates were then annealed at 110°C for 15 min.The HTL was prepared by dissolving 69 mg of spiro-MeOTAD in 800 μL of chlorobenzene.The spiro-MeOTAD solution was doped by adding 14 μL of Li-TFSI (574 mg mL −1 in acetonitrile) and 27.1 μL of 4-tertbutylpyridine to the spiro-MeOTAD solution.The solution was shaken at room temperature for 3 h prior to use.An amount of 50 μL of the doped spiro-MeOTAD solution was pipetted onto the substrate and spun at 4000 rpm for 30 s. Devices were then placed in dry air (<1% relative humidity) overnight.Next, a 130-nm-thick ITO (In 2 O 3 /SnO 2 90/10 wt.% in target) was sputtered through a shadow mask in an argon(Ar)-oxygen(O 2 ) gas mixture (O 2 : 0.5%) using a magnetron radio-frequency-sputtering system.To mitigate the sputter-induced damage in perovskite top cell, a long target-substate distance of 250 mm was used.The sheet resistivity of the ITO layer was ≈ 40 Ω/sq.Then, Ag (250 nm) finger and busbar contacts were sputtered in the same system.The Ag finger and busbar were then capped by an ITO (20 nm) thin layer to protect underlying Ag from corrosion during storage.The resulting device area was ≈1 cm 2 .The devices were then placed in a nitrogen-purged oven at 50°C for 1 h.Some of the devices were coated with a 110 nm-thick MgF 2 AR layer using a vacuum evaporator.The thicknesses of ITO and MgF 2 layers were roughly optimized using an optical simulator (OPAL 2) for maximizing the J SC in tandem devices. [45]In optical simulation, the optical constants of each layer were taken from the simulation database.
Characterization: The J-V characteristics were evaluated using a duallight solar simulator (Wacom, WXS-50S-L2) under standard test conditions (100 mW cm −2 , air mass 1.5 global, 25 °C).A shadow black mask was used to define an illumination area of 1.0 cm 2 which includes the Ag finger but excludes the Ag busbar (see Figure S5, Supporting Information).The J−V parameters of solar cells were recorded in both forward and reverse scan directions using a Keithley 2400 source meter in the voltage range from −0.1 to 1.9 V.The scan rate used was 0.15 V s −1 .EQE spectra were measured with an EQE setup (Bunkou Keiki, CEP-25BXS).For tandem devices, red ( > 750 nm) or blue ( < 650 nm) light bias was applied using optical filters to saturate photocurrent of the unmeasured subcell.No electrical bias was applied.
The surface morphology of the Si substrates was analyzed by scanning electron microscopy (SEM).The cross-section of the devices was captured by field-emission SEM at AIST and by HAADF-STEM (JEOL JRM-ARM200F) at JFE Techno-Research Corporation.

Figure 2 .
Figure 2. a) Spectral reflectance of the bare Si wafers with differently sized pyramid textures fabricated by Ag-assisted alkaline etching (A 420nm , B 530nm , C 680nm , D 950nm ) and by the conventional alkaline etching (E 1050nm ).For comparison, the reflectance spectra of the double-sided planar Si and frontplanar rear-textured Si are shown.b) EQE spectra of the SHJ single-junction solar cells employing Si wafers used for the reflectance measurement in (a) except double-sided planar Si.For clarity, 1-R spectra are shown for the representative solar cells (B 530nm and reference).Note that the thickness of the front component layers (ip-a-Si:H, n-nc-Si:H, ITO) for the double-side textured cells were not optimized.

Figure 3 .
Figure 3. a) Schematic diagram of the monolithic two-terminal perovskite/Si tandem solar cells fabricated in this study.b) Current density-voltage (J-V) curves (solid lines: forward scan, dashed lines: reverse scan) and c) The corresponding cross-sectional SEM images of the perovskite/Si tandem cells with differently textured Si bottom cells.Note that these SEM images are taken from the samples without the front ITO layer, while the device B 530nm has an ITO and magnesium flouride (MgF 2 ) layers on top of the spiro-MeOTAD HTL layer.d) EQE and 1-R spectra of the tandem cells made with a front-planar rear-textured Si (reference) and with an optimally textured Si featuring an average size of 530 nm (B 530nm ).

Figure 4 .Figure 5 .
Figure 4. a) Cross-sectional HAADF-STEM images of perovskite/Si tandem solar cell fabricated with the nanopyramid Si texture B 530nm .The highmagnification images at the valley and peak positions of a Si pyramid are shown in (b,c), respectively.