Ligand-bridged charge extraction and enhanced quantum eﬃciency enable eﬃcient n–i–p perovskite/silicon tandem solar cells †

Translating the high power conversion eﬃciencies of single-junction perovskite solar cells in their classic, non-inverted (n–i–p) architecture to efficient monolithic n–i–p perovskite/silicon tandem solar cells with high current densities has been a persistent challenge due to the lack of low-temperature processable, chemically-insoluble contact materials with appropriate polarity and sufficient optical transparency. To address this, we developed sputtered amorphous niobium oxide (a-NbO x ) with ligand-bridged C 60 as an efficient electron-selective contact, deposited on the textured-silicon bottom cell. For the sun-ward, hole-selective contact we implemented a stack of molecularly doped broadband transparent evaporated 2,2 0 ,7,7 0 -tetra( N , N -di- p -tolyl)amino-9,9-spirobifluorene (spiro-TTB) and atomic layer deposited vanadium oxide, which further enhances the device quantum efficiency. Combining these contact materials with two-dimensional perovskite passivation on the micrometer-thick solution-processed perovskite top cell yields 27% efficient monolithic n–i–p perovskite/ silicon tandem solar cells, which represents one of the highest power conversion efficiencies reported on pyramidal textured crystalline silicon bottom cells, and the highest with this polarity.


Introduction
Metal halide perovskite solar cells (PSCs) have been reported with many different material combinations and device configurations. However, in silicon-based tandem architectures, most of the studies in literature have been in the p-i-n configuration. [1][2][3][4][5][6][7][8] Although single-junction n-i-p devices have systematically proven higher power conversion efficiencies (PCEs), [9][10][11] their translation to the perovskite/silicon tandem configuration has been hindered by (i) a lack of lowtemperature processable, chemically insoluble, and conformal electron-selective layers (ESLs), (ii) a lack of efficient, conformal, and optically transparent hole-selective contact stacks, and (iii) an inability to fabricate the perovskite top cell on micron-sized pyramidal textured interfaces, as employed in high-efficiency silicon solar cells. 12 These challenges have limited the performance of n-i-p tandems to B22% (41 cm 2 ) on silicon heterojunction (SHJ) cells, and B24% (1 cm 2 ) on poly-silicon passivating-contact silicon solar cells, 13 even though all initial perovskite/silicon tandems were in the n-i-p configuration. [14][15][16][17] In the single-junction n-i-p configuration, several wellestablished ESLs are available, such as the archetypal compact and mesoporous titanium dioxide (c-TiO 2 /m-TiO 2 ) stack, which requires a high-temperature sintering step (4450 1C), however. 11,18,19 This temperature requirement excludes the applicability of this ESL stack for tandems built onto highperformance SHJ bottom cells, 20 whose temperature processing window is o250 1C due to their amorphous-silicon passivated surfaces. Thin films of tin oxide nanoparticles (np-SnO x ) are also increasingly used as an ESL on PSCs, 21 and these are usually deposited at lower temperatures (o200 1C). 22,23 However, np-SnO x ESLs are solution-processed. As a result, achieving conformal deposition on rough, complex surfaces, such as pyramidal-textured silicon, is difficult. Furthermore, thermallyevaporated C 60 is a successful ESL, but is soluble by subsequent perovskite solution processing.
In the search for alternative ESLs, thin-film niobium pentoxide (Nb 2 O 5 ) stands out as it combines favorable properties such as a low work function (WF), a wide bandgap, high chemical and ultraviolet (UV) light stability, as well as compatibility with various processing techniques. [24][25][26][27][28] To adopt such Nb 2 O 5 layers to n-i-p tandems as an ESL, we employed the RFsputtering technique. Using stoichiometric targets under an argon atmosphere, we achieved a slightly oxygen (O) deficient and amorphous film structure. Notably, we find that such O vacancies in the amorphous phase do not induce deep-level trap states, in contrast to np-SnO x , but rather act as shallow donor-like defects that improve the conductivity and mobility in the bulk of the film, and promote charge extraction through the contact structure. Modification of the charge transport layer/perovskite interface with C 60 derivatives is a successful strategy to further enhance electron extraction. [29][30][31][32] To apply this at a-NbO x /perovskite interface, we utilize C 60 with a functionalized pyrrolidine tail to create a conformal self-assembled molecular monolayer (SAM), anchored on the a-NbO x surface, which makes it insoluble to the aprotic solvents used in perovskite solution processing. To support our findings, we theoretically underpin this self-anchoring interaction by density functional theory (DFT) calculations. On the device-level, we show high PCE values for both perovskite single-junction and perovskite/silicon tandem solar cells utilizing C 60anchored a-NbO x ESLs, thanks to reduced non-radiative recombination at the interface.
Another major factor hindering the progress of n-i-p tandems has been the lack of efficient, optically transparent, and conformal hole selective layers (HSLs) and buffer layers. Such buffer layers protect the soft perovskite absorber from sputter damage during the deposition of the top transparent conductive oxide (TCO). In the more widespread p-i-n tandem configuration, a bilayer consisting of an evaporated C 60 ESL, followed by an ALD SnO x buffer layer, is employed, onto which a TCO is sputtered. So far, the most widely reported HSL for n-i-p devices is spiro-OMeTAD, but this material suffers from strong parasitic absorption in the blue part of the solar spectrum, and requires post oxidation with metalorganic complexes for improved hole mobility. 33 Nevertheless, several thermally evaporable small-molecule HSLs could be used. 34,35 However, the common issue with these molecules is their low hole mobility, low electrical conductivity, and unfavorable ionization potential. 36 To address this, we evaporated small molecule HSLs [2,2 0 ,7,7 0 -tetra(N,N-di-ptolyl)amino-9,9-spirobifluorene (spiro-TTB) 2 and N4,N4,N4 00 , N4 00 -tetra([1,1 0 -biphenyl]-4-yl)-[1,1 0 :4 0 ,1 00 -terphenyl]-4,4 00 -diamine (TaTm) 37 ] and doped them using the strong electron-accepting molecule 2,2 0 -(perfluoronaphthalene-2,6-diylidene) dimalononitrile (F6-TCNNQ) 38 to enhance the hole mobility and conductivity. Finally, we tested these HSLs in bilayer stacks with vanadium oxide (VO x ) 39 and tungsten oxide (WO x ) as buffer layers, first in single-junction devices, and then in monolithic perovskite/silicon tandems. On our optimized devices, we used doped spiro-TTB/VO x bilayers as an HSL stack.
To minimize reflection losses from the silicon bottom cells, [1][2][3]7,8 we utilized textured SHJ bottom cells, onto which micrometer-thick perovskites were solution-processed. In addition, we utilized a mixed 2D/3D passivation treatment on the perovskite top surface for improved charge extraction, and reduced hysteresis, which have been implemented usually in n-i-p configuration devices to date. Combining minimized reflection losses with reduced parasitic absorption by evaporated HSLs enabled one of the highest PCEs for tandems on textured interfaces. Combined, these systematic improvements in the n-i-p tandem device design solve the persistent issues that have hindered their progress on SHJ bottom cells, resulting in an improvement in PCE from B22% 17 to over 27%, a performance that is commensurate with its state-of-the-art p-i-n tandem counterparts.

Characteristics of pristine and C 60 -anchored a-NbO x films
Scanning electron microscope (SEM) and atomic force microscopy (AFM) top-view images reveal the amorphous morphology of our sputtered NbO x films ( Fig. 1a and Fig. S1c, respectively, ESI †). Unlike spin-casted np-SnO x films (Fig. S1b, ESI †), our a-NbO x films show a high conformality on the textured-Si surface, which is critical for tandem integration (Fig. 1b). High-resolution transmission electron microscopy (HR-TEM) (Fig. 1c), and surface-sensitive grazing incidence X-ray diffraction (GIXRD) measurements of a-NbO x films (Fig. 1d) confirm the amorphous nature of the material. In addition, postdeposition annealing of the films at elevated temperature (up to 500 1C) does not lead to crystal formation, as shown by the XRD graphs in the inset figure of Fig. 1d. 40 The multi-valent nature of niobium (Nb) leads to various stable oxidation states (+2, +4 and +5), resulting in different polymorphs such as NbO, NbO 2 , and Nb 2 O 5 . 41 Here, our amorphous films are sputtered from a Nb 2 O 5 target in a pure Ar atmosphere (without intentional oxygen flow), yielding nearly stoichiometric Nb 2 O 5 films, which is thermodynamically the most stable phase. 42 Nevertheless, our a-NbO x films are slightly O-deficient, according to our X-ray photoelectron spectroscopy (XPS) analysis (Note S1, ESI †). The O-deficient character of the a-NbO x films is further demonstrated by a two orders of magnitude decrease in electrical conductivity of the films after annealing in ambient air (Note S2, ESI †). 40,43 We investigate the chemical structure of the a-NbO x (Fig. 1e) and np-SnO x (Fig. S2, ESI †) films using high-resolution XPS spectra, which show that Nb ions exist in single metal-oxide environments of Nb 5+ with Nb 3d 5/2 (207.3 eV). The elemental composition of the films by wide survey XPS scans ( Fig. S2a and b, ESI †) revealed an atomic percentage ratio of Nb/O metaloxide = 1/2.2 and Sn/O metaloxide = 1/1.8 for a-NbO x and np-SnO x , respectively (see Note S1, ESI † for a detailed analysis). These results indicate a slightly O-deficient nature of the films, resulting in improved conductivity. We confirm the C 60 -anchoring on a-NbO x layers via XPS analysis, as the metal oxide core lines are strongly attenuated with a concomitant increase in the peak at B284.7-285 eV, corresponding to C 60 sp 2 carbon (Fig. 1f). The increased peak intensity further indicates efficient coverage of the C 60 -SAM molecule on a-NbO x films, as discussed in detail in Note S1 (ESI †).
Based on DFT calculations, we find a strong energetic preference for the O vacancies of a-NbO x to aggregate at the (100) surface of a particular crystal polymorph, which was employed as an approximate description of the material (as explained in the Note S3, ESI †). As a result, C 60 -pyrrolidine molecules can self-assemble on this surface by forming bonds between pyrrolidine terminating O and the surface Nb sites. On the other hand, the C 60 moiety interacts with the perovskite. The contact angle on a-NbO x films reveals that the surface becomes more hydrophobic after C 60 -anchoring; the AFMmeasured roughness remains the same, suggesting conformal coverage ( Fig. S8 and S1c, respectively, ESI †). C 60 -anchored random-pyramid textured silicon wafers also show hydrophobicity, which indicates good coverage over this complex surface (Fig. S8, ESI †).
The measured WF, valence band maximum (VBM), and conduction band minimum (CBM) of the a-NbO x layers, determined by ultraviolet photoelectron spectroscopy (UPS), analysis are À4.32 eV, À7.50 eV, and À3.85 eV (VBM + E g ) from the vacuum, respectively (see Note S4, ESI †). This suggests that the a-NbO x Fermi level is located close to its CBM, indicative of its strong n-type character. The relative band edges and WF values of all the layers in single-junction and tandem devices are shown in Fig. S9 (ESI †). We estimate the optical band gap (E g ; derived from Tauc plots) of the a-NbO x films is 3.65 eV, which is sufficiently high to be applied as an ESL in PSCs ( Fig. S11c and d, ESI †).

Single-junction solar cells with C 60 -anchored a-NbO x
To investigate the performance of the C 60 -anchored a-NbO x layers as ESL, we first fabricated single-junction n-i-p PSCs based on triple-cation perovskites (1.61 eV), 44,45 as sketched in   (Fig. 2b) while SEM and XRD results confirm the high-crystallinity of the perovskite layer ( Fig. S12a and b, ESI †) on C 60 -anchored a-NbO x . The J-V characteristics of the fabricated devices (with np-SnO x layers as a control) are shown in Fig. 2c. As reported in Table S4 (ESI †), C 60 -anchored a-NbO x based devices reach a 21.3% PCE for a 1.61 eV perovskite absorber, with high batch-to-batch reproducibility and negligible hysteresis (Fig. S13, ESI †).
Taking the advantage of low parasitic absorption of a-NbO x layers, ESL-side illuminated single-junction devices show a high short-circuit current density (J SC ) of 24 mA cm À2 , which is higher than their np-SnO x -based counterparts. These results are confirmed by external quantum efficiency (EQE) measurements, as shown in Fig. 2d (23.3 mA cm À2 from the integrated current density from EQE, within a 5% margin of equipment error). The average statistical distribution of the J SC results also confirms this observation, as shown in Fig. S13 (ESI †). Spectrally weighted absorptance values reveal that C 60anchored a-NbO x and np-SnO x has negligible absorption within the 290-850 nm range (0.65 and 1.02 W m À2 nm À1 , respectively). 1 À R (R is reflectance) measurements from the whole device stack show that the PSCs based on such stacks suffer from similar R losses (112.6 and 116.9 W m À2 nm À1 , respectively). In conclusion, high J SC values is mainly due to efficient charge collection as evidenced by EQE measurements (Fig. 2d).
Further, C 60 -anchored a-NbO x devices provide B4% absolute higher fill factor (FF) values on average compared to C 60anchored np-SnO 2 counterparts. To understand this difference, we simulated the electric-field distribution across the device profile, which is governed by the built-in voltage (V bi ), with different biasing voltages, as discussed in Note S6 (ESI †). Considering that the perovskite/HSL interface is identical for all cases, the low voltage drop at this interface can be attributed to a high shunt and low contact resistivity of the overall stack and a higher V bi (Fig. S15, ESI †). In an ideal case, the charge separation process is evidently more efficient if the internal electric field is stronger. Accordingly, we observed that a-NbO x based devices show a higher electric-field ratio between the ESL/perovskite and the perovskite/HSL interface, and overall lower voltage losses at the perovskite/HSL interface (see Fig. S16, ESI †). This implies an improved contact quality at the ESL/perovskite interface, which we assign to the high negative fixed-charge density of a-NbO x . 46 C 60 -anchoring further improves the electric-field ratio and provides effective passivation at the perovskite/ESL interface. 32 The relatively lower diode ideality factor of a-NbO x based devices (n = 1.23)  than that of np-SnO x -based devices (n = 1.35) suggests that in the former case the recombination occurs predominantly in the quasi-neutral region rather than in the space-charge region of the device. (Fig. S18, ESI †). Here, the high negative fixed-charge density of a-NbO x may enhance the drift region in ESL/perovskite devices, which in turn leads to enhanced charge extraction and less recombination. We attribute this behavior to the high dipole moment at the a-NbO x /perovskite interface. Specifically, C 60 -SAM anchoring on the a-NbO x surface increases the negative charge density at the ESL/perovskite interfaces which leads to a further increases in dipole moment and electric field (see Note S6, ESI †). Surprisingly, the FF values of the devices are not significantly affected by increasing the thickness of the a-NbO x layers up to 75 nm (Fig. S14, ESI †). Typically, metal oxidebased charge transport layers give the best PCE values when their thickness is limited to o30 nm due to their poor conductivity. [47][48][49][50] We hypothesize that the absence of deep-level defect states in the bulk of the a-NbO x films, together with its thickness-independent high negative fixed-charge density allows the use of thicker films. Indeed, DFT calculations of the electronic density of states (DOS) as a function of energy demonstrate that O vacancies do not create states within the bandgap of a-NbO x . Instead, the defects shift the highest occupied states at the bottom of the CBM of the material, resulting in enhanced n-type conductivity (Note S3, ESI †). In contrast, O vacancies in tetragonal np-SnO x introduce energy levels within the bandgap of the material (Fig. S7, ESI †) and can thus act as carrier trap sites. All these results underline the efficient electron selective character of the C 60 -anchored a-NbO x films.
Differences between the reverse (V OC to J SC ) and forward (J SC to V OC ) scan are a persistent problem of the n-i-p PSCs. The potential reasons for such observed hysteresis in the case of our devices are investigated in detail in Note S8 (ESI †). More generally, we mitigated this issue by balancing the charge extraction speed at both ends of the contacts by implementing 2D/3D passivation on top of the perovskite surface (i.e. holecollecting side), and C 60 -anchoring to the a-NbO x (i.e. electroncollecting side), as well as improving the perovskite crystallinity by using urea as an additive. We verified our results with stabilized power output values measured by the three-point perturbation MPP tracking method (see algorithm in Note S7, ESI †), shown in Fig. 2e (21.3% for a-NbO x and 19.9% for the np-SnO x based devices). Finally, we demonstrate the implementation of large-area deposited C 60 -anchored a-NbO x layers by fabricating a mini-module with a 6.8 cm 2 active area, which consists of four series-connected cells. The mini-module delivered a PCE of 19.5% (active area) with a FF of 77.7%, (the J-V curve is shown in Fig. 2f), demonstrating the scalability of the C 60 -anchored a-NbO x layers. 51

Perovskite/silicon tandem solar cells on double-side textured silicon cells
We fabricated micrometer-thick perovskite films via solution processing on the C 60 -anchored a-NbO x ESL (see processing details in Fig. S21, ESI †). The high perovskite crystal quality and uniformity is evidenced by the high spatial uniformity of the current collection from the tandem devices by light-beaminduced-current (LBIC) imaging (Fig. 3a). As can be seen from the cross-sectional SEM images in Fig. 3b, the perovskite layer has full end-to-end coverage on the micron-scale pyramids  without leaving any uncovered pyramid or void at the interface. We verified the homogenous distribution of the perovskite elements by energy-dispersive X-ray spectroscopy (EDX) mapping from a slice of cross-sectional TEM images shown in Fig. 3c. Notably, by utilizing large-area uniform sputtered a-NbO x ESLs (as shown in Fig. S22, ESI †) with self-anchored C 60 , we eliminated the chemical solubility challenge of the ESL during perovskite processing as shown in Fig. S23 (ESI †). The carboxylic-acid-terminated anchoring group of the C 60 -SAM interacts with the a-NbO x surface thanks to its strong affinity and prevents the washing away of the C 60 layer during the perovskite deposition. 52 Also, the high lateral resistivity of the a-NbO x film impedes the flow of shunt currents on textured interfaces.
For the HSLs, we first tested conventional Li-doped spiro-OMeTAD but expectedly, this resulted in considerable parasitic absorption in the range of 300-430 nm and a modest tandem J SC of 17.5 mA cm À2 (see Fig. S24, ESI †) due to its considerable required thickness (conventionally B150 nm) (Fig. 4a). To circumvent this issue, we used thermally evaporated, conformal, and thin HSLs such as spiro-TTB (B25 nm) and TaTm  (B50 nm). Here, we successfully adopted a strong electron acceptor dopant, F6-TCCNQ to the evaporated HSLs (host), which is evident from the absorption peak in the 450-570 nm range (see Fig. S25a, ESI †) -due to the p-p*-transition of the cations/charge transfer (CT) states within the HSL. 53,54 The overall parasitic absorption of the optimally doped HSLs is much less than for conventional spiro-OMeTAD (Fig. 4a). The observed near infrared (NIR) absorption peak of the spiro-TTB at B950 nm can be attributed to the F6-TCNNQ anion; this peak is not seen for TaTm. 55 However, TaTm strongly absorbs in the blue and limits the response of the perovskite subcell (Fig. 4a). Table 1 summarizes the device characteristics of the single-junction PSC (based on 1.68 eV) with various ESLs. Upon  doping, the ionization potential of the HSLs is shifted down as shown in Fig. 4b and this leads to an increased V OC . Overall, with these improvements, using doped spiro-TTB, we fabricated n-i-p tandems with higher perovskite subcell J SC values (20.2 mA cm À2 by EQE) compared to their p-i-n counterparts (19.1 mA cm À2 ) (Note S11, ESI †) with V OC values of 41.8 V.
We found that a-NbO x does not induce a considerable contact resistivity as confirmed by the J-V curves of the SHJ bottom cells and the contact resistivity values by transmission line measurements (TLM, see Fig. S27, ESI †), which facilitates high FF values. Also, among various HSL stacks, spiro-TTBbased ones give the lowest contact resistivity, as shown in Fig. 4c. The J-V characteristics of the single-junction devices for various HTLs (Fig. 4d) and monolithic tandems with optimized spiro-TTB are shown in Fig. 4e which shows high batchto-batch reproducibility (see Fig. S28, ESI †). Notably, we enhance the NIR response of the tandems by reducing the reflection losses owing to the double-side textured nature of the SHJ bottom cells (Fig. S29, ESI †). To balance the J SC of top and bottom cells, we use a 1.68 eV bandgap perovskite obtained by increasing the Br percentage from 15% to 25%, which is in the range of an ideal top cell bandgap. 1 To minimize the hysteresis behavior, which is discussed in Note S8 (ESI †), we utilized a mixed 2D/3D passivation treatment strategy (see structural details in Fig. S30, ESI †) since this enables non-continuous 2D layers on the perovskite surface by passivating the surface defects, while still allowing charge transport via 3D perovskite pathways. By applying iso-butylammonium iodide, (iBAI) and formamidinium iodide (FAI) mixed passivation, we decreased the number of surface defects and mitigated ion migration, thereby significantly decreasing the hysteresis of the device (see Fig. 4e and Fig. S31, ESI †). 56 Furthermore, we minimized the parasitic absorption losses in the blue range of the solar spectrum by combining thermallyevaporated spiro-TTB with atomic layer deposited VO x in the hole selective contact stack (see discussion on buffer layer selection in Note S13, ESI †) as well as by replacing the rear ITO with Zr-doped indium oxide (IZRO) 57 (for a lower extinction coefficient) and by thinning down the front indium zinc oxide (IZO) from 100 to 70 nm (Fig. S33, ESI †). As a consequence of improved light management (Fig. 4f), combined with enhanced charge extraction, we record a significant improvement from the last reported perovskite/silicon tandem in the n-i-p configuration on a SHJ bottom cell (which was B22%) and achieved a 27% efficient monolithic tandem performance from such devices, which is one of the highest PCEs reported in either the n-i-p (see Table S7, ESI †) or p-i-n tandem configurations. [2][3][4][5] We confirmed this PCE value by MPP measurements under continuous illumination (inset of Fig. 4e).
We also performed stability tests of the tandem devices under 1-sun continuous illumination and intermittent J-V scans. For this, we first encapsulated devices between two glass sheets with butyl rubber as edge sealant, shown in Fig. S34 (ESI †). The optimized devices maintained 90% of their initial V OC and FF values for 150 h (Fig. 4g). In the literature, n-i-p devices are frequently reported to have short operational stability due to fast degradation of the doped HSL. 33,36,37 To identify potential degradation mechanisms of our HSL stacks, we performed a set of stability measurements. We found that the spiro-TTB/VO x interface appears to be sensitive to degradation due to Lewis acid quenching of radical anions and di-anions of the F6-TCNNQ, with a consequence of the free electron lone pairs that act as a Lewis base. At the HSL/VO x interface, vanadium centers act as a Lewis acid, and attract and coordinate highly electron-rich nitrogen atoms of F6-TCNNQ, thereby suppressing its activity. From our stability test with singlejunction test structures, we found that elevated temperature is the primary reason for this accelerated degradation (see Fig. S35, ESI †). This stems from the accumulation of quenched dopants at the VO x interface at elevated temperature. 58 To mitigate this issue, we introduced a 2 nm thin layer of 2,2 0 ,2 00 -(1,3,5-Benzinetriyl)-tris(1-phenyl-1-H-benzimidazole), TPBI, which effectively prevented the interaction between F6-TCNNQ and VO x without jeopardizing the charge transfer due its ultra-thin nature.

Efficient charge extraction at C 60 -anchored a-NbO x and perovskite interfaces
We performed time-resolved photoluminescence (TRPL) spectroscopy measurements to analyze the charge carrier dynamics at the perovskite/ESL interface. We found that the PL lifetime of the perovskite in the quartz/a-NbO x /perovskite sample is significantly longer than that of both the quartz/np-SnO x /perovskite and the quartz/perovskite sample structure (Fig. 5a). We note that the substrate could affect the perovskite film formation and thereby the perovskite defect distribution which in turn could alter the PL lifetime of the perovskite films, as shown in Fig. 2b. We hypothesize that the reason for the longer PL lifetime is two-fold: (i) the absence of deep-level defects at the a-NbO x /perovskite interface, and (ii) the dependency of the PL lifetime of the absorber on the carrier concentration (N e ) of the charge transport layers, where low N e values lead to less quenching of the PL lifetime. 19,59 Here, compared to np-SnO x , a-NbO x films have a seven orders of magnitude lower carrier concentration than np-SnO x , which appears to be the reason for this behavior (Table S1, ESI †).
We note that the PL lifetime of C 60 -anchored a-NbO x /perovskite sample is almost three times that of the C 60 -anchored np-SnO x /perovskite sample. The shorter PL lifetime of the latter can be caused by either efficient charge extraction from the perovskite to the ESL or increased recombination at the interface. 19 To clarify the origin, we determined the k 1 values, which are related to trap-assisted recombination, for samples with and without C 60 -anchored ESLs. We found that C 60anchoring on np-SnO x and a-NbO x resulted in lower k 1 values than observed for neat ESLs and overall less trap-assisted recombination. All values are listed in Table 2.
To evidence the C 60 -anchoring on the a-NbO x films on textured substrates, we performed TRPL measurements also on wide bandgap (1.68 eV) perovskite absorbers deposited on textured c-Si/SiN x /perovskite stacks, where SiN x is silicon nitride, which is used to electrically isolate the Si wafer ( Fig. 5b). Here, we found the PL lifetime to be reduced from 140 ns to 57 ns in the case of C 60 -anchoring, confirming the enhanced charge extraction properties of the self-assembled layer on textured interfaces, which is in line with the measurements performed on the 1.61 eV bandgap perovskites on quartz substrates.
For the C 60 -SAM anchoring, the pyrrolidine group of the C 60 -SAM molecule strongly interacts with the metal oxide ESLs (Fig. 5c), while its C 60 component interacts with (negativelycharged) iodide ions, which originate from non-coordinated Pb or Pb-I anti-site defects. To further elaborate on the role of the   50 case, the C 60 -related state is also below a gap level associated with the common Pb-I anti-site perovskite defect.
In agreement with experiments, our DFT calculations (Note S3, ESI †) suggest that the empty C 60 -related gap states can extract electrons from the conduction band or electron traps (related to, e.g., Pb-I antisite defects) of the perovskite (Fig. 5d) and pass them on to the NbO x side through the firmly attached pyrrolidine tails. On the ESL side, the role of O vacancies (O v ) is crucial for the efficient extraction of electrons. In the SnO x case, O v defects create charge trap states (Fig. S7, ESI †) that can hinder carrier transport, which is in line with space-chargelimited-current (SCLC) and impedance spectroscopy measurements (Note 14 and 15, ESI †). On the other hand, in the NbO x case, O v shifts the highest occupied states into the bottom of the conduction band. This effect is also found for O-deficient NbO x terminations (Fig. S6, ESI †), and persists even after C 60anchoring. At this level of theory, O v enhances the n-type conductivity of the material, in agreement with the aforementioned experimental observations.
To support our V OC results, we performed absolute PL measurements (Fig. S38, ESI †) to determine the quasi-Fermilevel-splitting (QFLS), which represents the maximum achievable V OC from the perovskite absorbers with and without contact stacks. Herein, C 60 -anchored a-NbO x based stacks show 1.24 eV, while np-SnO x devices show 1.21 eV for 1.61 eV perovskites (Fig. 5e), in line with the results of the TRPL measurements (Fig. 5a), and also the V OC values of the devices (Table S4, ESI †). The presence of the C 60 -anchoring improves the charge extraction in both a-NbO x and np-SnO x samples but it lowers the QFLS values due to energetic issues, as discussed in Note S4 (ESI †). However, in devices, the measured V OC values are lower than the QFLS values. According to our measured UPS and IPES data, both a-NbO x and np-SnO x layers add a 20 meV potential barrier at the interface to perovskite after C 60anchoring. However, several studies reported that this level of CB offset is needed to suppress interfacial recombination due to the reduced back-transfer recombination and improved FF. [60][61][62]

Conclusion and outlook
Here, we overcome the persistent processing limitations of n-i-p configuration perovskite/silicon tandem solar cells by combining on pyramidal textured SHJ bottom cells a RFsputtered a-NbO x with conformal C 60 -anchoring as ESL, micrometer-thick solution-processed perovskites and a highly transparent conformal HSL stack. All these advancements enabled 27% efficient perovskite/silicon monolithic tandems in n-i-p configuration, which is a 45% absolute increment over the prior state of the art on SHJ bottom cells. Our demonstration of an efficient n-i-p tandem device platform opens new pathways to propel the efficiency of perovskite/ silicon tandems beyond the thermodynamic limit for singlejunction solar cells. For instance, thermally-evaporated C 60 ESL for p-i-n configuration stands out unrivaled so far, even though this material causes strong parasitic absorption. On the other hand, there are many small molecule HSLs in the literature which have remained unexplored in n-i-p tandem solar cells to improve the device optics further and overall reaching to 420 mA cm À2 tandem devices. Furthermore, to overcome the limited long-term operational stability of the perovskite/silicon tandems, 2D-perovskite passivation strategies can be utilized which has been a challenge with p-i-n devices due to energetic alignment problems of the 2D perovskites with the widely used C 60 ESLs. Also, bifacial tandem devices can be more relevant in the n-i-p configuration, since the bottom cell utilizes an n-type contact at its back, high refractive index but lower conductivity TCOs can be utilized. Beyond these opportunities, our developed hole selective contact stack will also enable progress in perovskite/perovskite tandems by eliminating the air exposure requirement of moisture-sensitive Sn-based perovskites during HSL processing, which has been an issue so far. Lastly, the n-i-p tandem platform enables utilization of tandems as a photocathode in efficient solar-driven water splitting applications for hydrogen generation since this application requires electrons to be extracted from the cathode side with voltage values of 41.4-1.8 V, as well as high current densities.

Experimental section
Deposition of a-NbO x and np-SnO 2 layers a-NbO x layers were deposited by radio frequency (RF, 13.56 MHz) magnetron sputtering from the stoichiometric a-NbO x target (Plasmaterials, 99.95% purity) at room temperature using Angstrom Engineering EvoVac system at 1.97 W cm À2 power density. The Argon gas was supplied to the chamber with 20 sccm. The deposition rate of the films was 0.15 Å s À1 . The base pressure for deposition was o2 Â 10 À6 Torr, and the deposition pressure was 3 mTorr. Before each deposition, the target was pre-sputtered for 10 min to remove the contamination layer, if any, on the surface of the target. For homogenous deposition, substrate rotation was provided. The substrate was not heated or cooled intentionally. np-SnO 2 layers were deposited by spin-casting of aqueous Alfa Aesar nanoparticle suspension at 3000 rpm for 30 s (with 3000 rpm s À1 ramp rate), and resultant films were annealed in ambient air at 150 1C for 30 min. Prior to np-SnO x deposition, 10 min UV-ozone treatment was applied on ITO/glass substrates. However, this step was not used prior to the deposition of a-NbO x layers.

Energy & Environmental Science Communication
Open This journal is © The Royal Society of Chemistry 2021 C 60 -SAM modification of a-NbO x layers C 60 -based self-assembled monolayer (C 60 -COOH-SAM, CAS No.: 631918-72-4, Luminescence Technology Corp.) dissolved in chlorobenzene and stirred vigorously overnight. Later, the stock solution filtered using a 0.2 mm PTFE filter. The C 60 -SAM was spun on a-NbO x or np-SnO x at 3000 rpm for 60 s with a 3000 rpm s À1 ramp rate. Before this, the surface of the a-NbO x layer is treated with 1 mg mL À1 concentrated aqueous potassium chloride solution. After that, films were transferred immediately onto 100 1C preheated hotplate and annealed for 30 min. 25 nm Spiro-TTB was deposited by thermal evaporation as defined below. 10 nm ALD deposited vanadium oxide (VO x ) was deposited using via the Picosun ALD tool using vanadyl-tri-isopropoxide (VTIP, precursor was kept 80 1C) and H 2 O as precursors and substrate at 90 1C. 100 nm of Ag layers were thermally evaporated on top of HSL stack by thermal evaporation in an Angstrom Engineering EvoVac deposition system. 100 nm of MgF 2 is evaporated (if stated) as an anti-reflective layer with 1 Å s À1 deposition rate to minimize the reflection losses for both np-SnO x and a-NbO x layers.

Fabrication of the silicon bottom cell
SHJ bottom cells are fabricated on float-zone double-sidetextured four inches wafers (n-doped, resistivity 1-5 O cm À1 , thickness 250-280 mm). The texturing process is done in an alkaline solution to get randomly distributed pyramids and cleaned in RCA1 and RCA2 solution. The size of the pyramids is controlled by adjusting the alkaline concentration and the process temperature. Before the PECVD depositions, the wafers are dipped in 5% hydrofluoric acid solution to remove the native oxide layer. The intrinsic, p, and n silicon amorphous layers are deposited in a PECVD cluster (Indeotec Octopus -2) with a thickness of 8, 6, and 13 nm, respectively. The back electrode is realized by sputtering sequentially ITO and Ag (150 and 250 nm, respectively) in the PVD part of the Octopus cluster. In case of IZRO rear TCO, the recipe given in our previous publication is used. 57 The ITO recombination junction (15 nm) is sputtered in the same tool. Prior to the a-NbO x deposition, and after a-NbO x deposition, the wafers are annealed at 190 1C for 10 min to recover the sputtering damage.

Fabrication of the perovskite top cell
To achieve n-i-p configuration perovskite top cells, a-NbO x layers (20 nm on the textured surface) is deposited by RF sputtering as the ESL and modified by C 60 -SAM as defined above.  Fig. 2c). Note that the reverse scan is from V OC to J SC (1.25 V -À0.1 V), and the forward scan is from J SC to V OC (À0.1 V -1.25 V). No preconditioning protocol has been used before the measurement. J-V measurements of the tandem solar cells were performed using Wavelabs Sinus 220 LED-based solar simulator. To calibrate spectrum, we used Fraunhofer ISE certified calibration cell (without a filter). We did not apply a spectral mismatch factor. The illumination area of the devices was 1.03 cm 2 , which is determined by the laser cut shadow mask coated with black paint. Maximum power point tracking (MPPT) measurements

UPS/XPS analysis
All the films were prepared on pre-cleaned ITO substrates as used in devices. The measurements were carried out with a UHV 0.5-1 10 À9 mbar Omicron spectrometer equipped with a monochromatic Al K a X-ray Omicron XM1000 X-ray source (hn = 1486.6 eV) operating at 389 W and a Sphera II spectrometer with a 7-channel channeltron. The survey and high-resolution spectra were collected at constant analyzer pass energies (CAE) of 50 and 15 eV, respectively. Samples (B0.5 Â 1.0 cm) were mounted in electrical contact to the Omicron style sample plate by clamping with Ta clips in contact with the sample surface, binding energy calibration and sample charging were accounted for by reference to the ambient contamination C 1s adventitious carbon, B284.7 eV. Samples were introduced to the chamber via a N 2 filled UHV compatible transfer system from an N 2 glove box where the samples were mounted to minimize ambient exposure. The spectra were analyzed with Casa XPS software. The individual peak envelopes were fitted by a Gaussian (70%)-Lorentzian (30%) (GL30) function after Shirley-type background subtraction. UPS measurements were performed in the same instrument on the same samples with a (CAE) of 10 eV, using a He (I) excitation discharge lamp (hn = 21.22 eV) (Focus). A bias of À9.97 V was applied to the sample. The photoelectrons were collected at an angle of 901 between the sample and analyzer.

Time-resolved photoluminescence (TR-PL) spectroscopy
Perovskite films were deposited either on pre-cleaned quartz substrates or on ESLs (np-SnO x and a-NbO x ) using the same recipe as used for device fabrication. After deposition of the layers, samples were placed in a sealed N 2 chamber to avoid any changes due to photo-induced degradation. All measurements were performed at room temperature. The samples were excited with the Coherent Helios 532 nm nanosecond (ns) laser with a pulse width of 0.85 ns and a peak with a repetition rate of 1 kHz. The excitation beam at 532 nm was focused on pumping the samples, mounted inside the sealed chamber. The PL of the samples was collected by an optical telescope (consisting of two plano-convex lenses) and focused on the slit of a spectrograph (PI Spectra Pro SP2300) and detected with a Streak Camera (Hamamatsu C10910) system. The data was acquired in photon counting mode using the Streak Camera software (HPDTA). The photogenerated carrier concentration at t = 0 was calculated using the formula 64 of n 0 = aÁf/E photon where a is the absorption coefficient at the excitation wavelength, f is the fluence of the laser and E photon is the excitation photon energy. The PL lifetimes were determined by using the general rate equation (Àdn/ dt = k 1 n + k 2 n 2 ) however, excluding Auger recombination, since the initial carrier concentration was low (n 0 = 4 Â 10 16 cm À3 ), where Auger recombination is negligible in perovskites. 65

Absolute PL measurements
Perovskite films were deposited on pre-cleaned glass substrates or on ESLs (np-SnO x and a-NbO x ) using the same recipe as used for device fabrication. The absolute PL spectra of the samples were collected using a hyperspectral imaging system coupled to a microscope with 2 nm spectral resolution (Photon etc. IMA). Samples were excited with 532 nm laser at B1 sun illumination condition (before the measurement, the power density of the laser was adjusted with a reference perovskite solar cell). The absolute calibration procedure of the setup is reported in detail elsewhere. 66 The collected data was analyzed via home-built MatLab code using modified generalized Plank law to get quasi-Fermi level splitting using the formula: 67 j ðE;yÞ ¼ A ðE;yÞ cos y 4p 3 h 3 c 0 2 E 2 1 exp E À Dm k B T À 1 where A (E,y) is the absorption probability of a photon with energy E, incident at an angle y with respect to the surface normal. Dm and k B T are quasi-Fermi level splitting and Boltzmann constant and the temperature (equivalent to 25.7 meV at room temperature), respectively. Here, cos y factor appears as the emission from the surface following Lambert's law.

Device simulations
The optical response of the single-junction PSCs was calculated based on the finite element method (COMSOL Multiphysics), where periodic boundary conditions were implemented in a unit cell. 68 The linearly polarized plane wave was illuminated normally to the cell top surface. The PSC was illuminated from the top with a linearly polarized plane wave impinging normally to the cell top surface, covering the wavelength ranges from 300 nm to 800 nm (AM1.5G illumination), comprising the bandgap of PSCs. Two-dimensional device simulation of NbO x and SnO x PSCs was based on a classical drift-diffusion and Poisson's equation and solved by the finite element method (COMSOL Multiphysics) based semiconductor device simulator. 69 The simulation was carried out at 300 K under one sun (AM1.5G, À100 mW cm À2 ) condition. The optical bandgap, thickness, and energetics (bandgaps, electron affinity) related to the active and other layers used in the simulation were obtained from the UPS/IPES measurements and listed in Note S4 (ESI †).

DFT analysis
Calculations were performed with the DFT code Quantum Espresso, 70 the Perdew-Wang generalized-gradient approximation exchange-correlation functional (xc), 71 projectoraugmented waves, 72 [3] and the DFT-D2 method 73 to include van der Waals interactions. The energy cutoff for the planewave basis was set at 100 Rydbergs. The calculation of densities of states (DOS) employed the Monkhorst-Pack scheme with Gaussian broadening and 4 Â 4 Â 1 k-grids. 74 In total energy calculations on large supercells, only the gamma point was used for k-sampling. Structures were rendered with VESTA. 75 To model O vacancies in the bulk of SnO 2 and Nb 2 O 5 we used supercells with, respectively, 71 and 111 atoms. The calculations pertaining to adsorption of pyrrolidine or C 60 -pyrrolidine on the (001) surface of SnO x , the (100) surface of Nb 2 O 5 , and the (001) surface of the halide perovskite used supercell slabs with, respectively, 102, 142, and 197 atoms. The large size of the supercells precluded the possibility of using post-DFT corrections in the form of a hybrid xc-functional.