Scalable Triple Cation Mixed Halide Perovskite–BiVO4 Tandems for Bias‐Free Water Splitting

Strong interest exists in the development of organic–inorganic lead halide perovskite photovoltaics and of photoelectrochemical (PEC) tandem absorber systems for solar fuel production. However, their scalability and durability have long been limiting factors. In this work, it is revealed how both fields can be seamlessly merged together, to obtain scalable, bias‐free solar water splitting tandem devices. For this purpose, state‐of‐the‐art cesium formamidinium methylammonium (CsFAMA) triple cation mixed halide perovskite photovoltaic cells with a nickel oxide (NiOx) hole transport layer are employed to produce Field's metal‐epoxy encapsulated photocathodes. Their stability (up to 7 h), photocurrent density (–12.1 ± 0.3 mA cm−2 at 0 V versus reversible hydrogen electrode, RHE), and reproducibility enable a matching combination with robust BiVO4 photoanodes, resulting in 0.25 cm2 PEC tandems with an excellent stability of up to 20 h and a bias‐free solar‐to‐hydrogen efficiency of 0.35 ± 0.14%. The high reliability of the fabrication procedures allows scaling of the devices up to 10 cm2, with a slight decrease in bias‐free photocurrent density from 0.39 ± 0.15 to 0.23 ± 0.10 mA cm−2 due to an increasing series resistance. To characterize these devices, a versatile 3D‐printed PEC cell is also developed.


Introduction
As society's energy demands increase, the production and storage of sustainable energy becomes a critical issue. [1][2][3][4] While photovoltaic (PV) modules may cover this demand on which was later improved to a biasfree value of 0.57% using a pSi nanoarray|Pt-Mo:BiVO 4 |Co-Pi tandem. [48] Due to the rapid growth in the efficiency of lead halide perovskite photovoltaics, rising from the initial 3.8% reported by Miyasaka and coworkers [51] to beyond 22% nowadays, [52] these photoabsorbers have also attracted attention from the solar fuel community. The initial prototypes involved wired systems, where several solar cells were connected in series to separate electrocatalysts, resulting in successful proton, [53][54][55][56][57] CO 2 , [58] or H 2 S reduction. [59] Since the use of wired systems provides several disadvantages by requiring gas separating membranes, wiring, external connections and additional device packaging, [13,60] several recent reports proposed more compact designs for PECPV tandem devices. [61,62] Nevertheless, to pre vent degradation of the perovskite layer, the photovoltaic com ponent was again physically separated from the solution, with a conductive wire ensuring the connection to the electrocatalyst. Such devices mainly combined a perovskite cell with a wide band gap oxide layer (i.e., BiVO 4 , [61,[63][64][65][66] WO 3 , [67] TiO 2 , [68] or hematite [62,69,70] ) to drive water splitting, [63] or CO 2 reduction. [71] In those cases, biasfree devices could be achieved due to the high open circuit voltage (V OC ) of the perovskite component, which enabled a good overlap between the cathodic and anodic photocurrents.
While previous attempts to directly interface the regular struc ture perovskite surface to an electrolyte solution through a thin nickel layer have produced a moderate stability of 10-30 min with photocurrents varying between 2 and 17 mA cm −2 for the oxygen evolution reaction, [72][73][74] a significant improvement was recently demonstrated in our group by employing the low melting point Field's metal (FM) instead. [75] This InBiSn alloy could simulta neously provide encapsulation and electrical contact to a plat inum nanoparticle (PtNP) catalyst, sustaining an encouraging hydrogen generation photocurrent of -6.9 ± 1.8 mA cm −2 at 0 V against the reversible hydrogen electrode (RHE) beyond one hour. [75] In this case, an inverse structure perovskite cell was necessary to collect electrons for the proton reduction at the outer device surface (see schematic depictions in Figure 1). With an onset potential of 0.95 ± 0.03 V versus RHE, this initial system represented a promising example of a perovskitebased photocathode, on the basis of which wireless tandems for solar fuel production could be developed.
While most of these initial results employed the moisture, air, and temperature sensitive methylammonium lead triiodide (MAPbI 3 ) perovskite, [51,76] more recent photovoltaic reports have shown that improvements in both efficiency and stability can be obtained when using complex precursor solutions. In par ticular, mixtures including the formamidinium (FA) and meth ylammonium cations (MA), together with iodide and bromide anions, achieved efficiencies beyond 20%, [77,78] whereas the addition of further (earth) alkali cations (e.g., cesium, magne sium, or potassium) have increased both efficiency (above 21%) and stability (beyond a few weeks). [79][80][81][82][83][84][85] Still, the limited scal ability of perovskite cells remained an issue until very recently, when large scale deposition procedures such as spray coating, doctor blading or vacuum techniques [86] started to be optimized alongside spin coating [82,87] for larger photovoltaic modules, leading to efficiencies between 10% and 16% for areas above 36 cm 2 . [16,82,88,89] Accordingly, although organic-inorganic lead halide perov skite photovoltaics have experienced a rapid development in performance within the last decade, the field has mostly con tinued to suffer from the same challenges of device upscaling as photoelectrocatalysis. To address some of those issues, in this work we investigate the scalability of tandem PEC devices for water splitting, combining a cesium formamidinium methylammonium (CsFAMA) triple cation perovskitedriven photocathode with a bismuth vanadate photoanode. To gain a complete understanding of the limiting factors, the per formance and scalability are investigated separately for the photovoltaic cells and for the derived photocathodes. In order to evaluate devices of various sizes, we also propose a straightfor ward design for a modular 3Dprinted PEC cell, which can be easily assembled or adapted for large scale studies.

Perovskite Photovoltaic Cells
A NiO x hole transport layer [90] was deposited on top of the FTO coated glass substrate by spincoating a Ni(NO 3 ) 2 ·6H 2 O precursor solution followed by annealing. A stateoftheart CsFAMA triple cation perovskite photoabsorber was deposited by spincoating using the established antisolvent method. [78,79,85,90] Polyethylen imine (PEIE) [91] was spincoated on top of the thin [6,6]phenyl C 61 butyric acid methyl ester (PCBM) electron selective layer, to pre vent interfacial degradation [92] from the reaction of silver with the halide ions. The additional physical separation provided by the PEIE layer is noticeable when comparing Figure 1a of a complete device with Figure S7a (Supporting Information) of the bare PCBM on a perovskite substrate. This design provided a per formance improvement over our previously reported devices, [75] where a poly(3,4ethylenedioxythiophene):poly(styrenesulfonate) (PEDOT:PSS) hole transport layer (HTL) and a MAPbI 3 photo absorber were used.
To draw an accurate comparison between the performance of a photocathode and the underlying photovoltaic component, all devices have been first investigated as solar cells. In this way, thorough batch statistics of the photocathodes could be made, by taking already faulty PV devices into account.
The average photovoltaic parameters are depicted as a func tion of the device area in Figure 2, with the exact values given in Table 1. The corresponding histograms for the 0.045, 0.25, and 1 cm 2 devices can be found in Figures S2-S6 of the Supporting Information. Examples of typical I-V curves are also presented in Figure 3a.
In general, the small 8pixel devices (area 0.045 cm 2 ) present the best performance, with the champion device reaching an efficiency of 16.4% in backward scan direction, with an open circuit voltage (V OC ) of 0.99 V, a short circuit current density (J SC ) of 19.7 mA cm −2 and a fill factor (FF) of 83.1%. The improved V OC of 1.00 ± 0.02 V is also con sistent with previous reports, which indicate a better align ment between the work function of NiO x and the perovskite's valence band edge level than in the case of PEDOT:PSS, which typically only delivers V OC values between 0.5 and 0.9 V. [93][94][95][96][97] As observed from Figure 2d, a roughly logarithmic decrease in the photovoltaic cell efficiency (PCE) with the active area occurs above a surface of 0.25 cm 2 , due to resistive losses. [98,99] These losses manifest mainly through a decrease in the  fill factor (see Figures 2c and 3a), which induces a similar behavior in the J SC (Figure 2b). This leads to a limitation of the maxi mum short circuit current below 100 mA, as observed for the 4 and 10 cm 2 devices. The less pronounced decrease in the V OC (Figure 2a) may be due to an increasing shunt resistance originating from more defects and pinholes in the layer struc ture of the larger devices. Nevertheless, our study on the scal ability of single pixel perovskite PV cells (as opposed to serially connected modules) reveals that efficiencies above 1% can be obtained even for 10 cm 2 devices, whereas the Field's metal is also first demonstrated as a valid encapsulant for photovoltaic applications. These observations indicate that metal fingers at distances between 0.5 and 1 cm from each other are required for maintaining a high performance in large scale perovskite applications, similar to the ones available for silicon panels.
Another interesting aspect of our findings is that the NiO x layer can also effectively cover the rougher surface of bare FTO (Figure 4a,b). In this case, a very thin coating of NiO x nano clusters is formed, which does not present additional peaks in the Xray powder diffraction (XRD) spectra ( Figure S8, Sup porting Information). This smoothing of the underlying sur face appears to result in more homogeneous and better packed perovskite grains (Figure 4c,d). While the higher surface rough ness of FTO has made indium tin oxide (ITO) glass the pre ferred choice for planar inversestructure perovskite cells, [100,101] these results show that the former can be also successfully employed. Since FTO is chemically more robust to aggressive cleaning methods (e.g., Piranha solution) and has a typically lower sheet resistance, its use has a potential advantage for large scale, industrial applications.  Table 1. Numerical data of the photovoltaic parameters illustrated in Figure 2. The open circuit voltage, short circuit current density, fill factor and photovoltaic cell efficiency are given as a function of the photoactive area for both forward (f) and backward (b) scans, whereas the number of working devices from the total amount produced is also shown.

Perovskite-Based Photocathodes
After the photovoltaic characterization, the 0.25 cm 2 perovskite PV devices (abbreviated PVK) were encapsulated by briefly melting a thin Field's metal (FM) sheet on top of the silver con tact via a Peltier thermoelectric element and sealing the edges with epoxy resin. By maintaining the Field's metal in a liquid state for only 20-40 s, a good adhesion to the underlying silver contact was ensured, while preventing a degradation of the perovskite layer through exfoliation of the Ag layer. Following the electroless Pt nanoparticle deposition, the performance of the resulting PVK|FM|Pt photocathodes was investigated for H 2 evolution. Typical examples of the results are depicted in Figure 5a,b, where the sign of the photocathode traces is reversed for convenience. Further data from all devices can be found in Figures S16-S18 of the Supporting Information. In order to determine the reliability of the device fabrication procedures, performance statistics have been performed on an initial batch of eight 0.25 cm 2 PV devices (V OC, b 1.00 ± 0.06 V, J SC,b 18.9 ± 0.8 mA cm −2 , FF b 72.3 ± 3.8%, PCE b 13.7 ± 0.9%), from which two were shorted. Due to the improved encapsu lation technique using thin Field's metal foils and a Peltier thermoelectric element, all six remaining devices could be investigated for photoelectrochemical studies.
While the favorable position of the NiO x energy levels pro vide an improvement of around 0.1 V in the V OC of the PV devices, a slightly more negative onset potential of around 0.6-0.8 V versus RHE was obtained for all photocathodes as compared to the previous 0.95 ± 0.03 V versus RHE. [75] However, the use of the CsFAMA triple cation perovskite resulted here in an almost doubling of the photocurrent over our initial MAPbI 3 devices, [75] with 5 out of 6 samples deliv ering an average value of -12.1 ± 0.3 mA cm −2 at 0 V versus RHE (see Figure 5a, red curves, and Figure S16 of the Sup porting Information). This high current density is compatible with stateoftheart PVelectrolyzer and PV-PEC systems, which reach between 4 and 12 mA cm −2 under biasfree conditions. [21,39,40,53,65] Thus, the encapsulated perovskite photo cathodes are valid alternatives to existing components in the wired designs for solar fuel generation. Moreover, by replacing the acidic, hydrophilic PEDOT:PSS HTL [100][101][102] with the robust NiO x , [90,95,96] a significant improvement in the device stability has been obtained from the initial 1-2 h, [75] with 4 out of 6 devices lasting for at least 4 h under use (up to 7 h stability for our champion device, see Figure 5b and Figure S16 of the Supporting Information). For those devices, the average fara daic yield (FY) amounted to 78.8 ± 3.5 % after 4 h. Losses are possibly caused by the entrapment of small H 2 bubbles along the measuring cell's glass walls or by leakage due to overpres sure in the cathodic compartment.
Similar results have also been recorded for four 1 cm 2 PV devices (V OC, b 0.867 ± 0.002 V, J SC, b 21.1 ± 1.3 mA cm −2 , FF b 50.2 ± 4.7%, PCE b 9.2 ± 1.2%, no shorted cells). In this case, a comparable average photocurrent of -9.3 ± 1.1 mA cm −2 at 0 V versus RHE was determined for all four corresponding photo cathodes, and the more noisy voltammetric signal was likely caused by the vigorous hydrogen evolution (see Figure S17 and Video S1 of the Supporting Information). Two devices lasted beyond 2 h (FY 74.7 ± 3.0%, see Figure S17 of the Supporting Information), with the slight loss in stability due to a higher probability of water leakage with increasing area, com bined with additional mechanical stress during the transport of the encapsulated devices for PV characterization.

BiVO 4 |TiCo Photoanodes
The BiVO 4 |TiCo photoanodes of corresponding sizes were prepared similarly to reported procedures. [47,103] BiOI was elec trodeposited onto FTO glass and annealed in the presence of a vanadium precursor. The TiCo catalyst was spincoated from a single source precursor solution. [47] A photoelectrochemical batch analysis has also been conducted for the BiVO 4 |TiCo photoanodes; results for all the devices are presented in Figures S19-S21 of the Supporting Information and typical linear sweep voltammetry (LSV) scans are depicted in Figure 5a (blue curves). For a batch of six 0.25 cm 2 devices, the average photocurrent density amounted to 1.61 ± 0.28 mA cm −2 at an applied potential of 1.23 V versus RHE (see Figure S19 of the Supporting Information), whereas it reached 1.31 ± 0.10 mA cm −2 at 1.23 V versus RHE for three 1 cm 2 devices ( Figure S20, Supporting Information).
Since the comparatively high stability of BiVO 4 under use is well known, [104][105][106][107] the chronoamperometric measurements at 1.23 V versus RHE have only been conducted for 4 h, to match the stability of the perovskitebased photocathodes. An example is given in Figure S12   leakage (see Figures S19 and S20 of the Supporting Information for the linear fitting of the raw fluorescence lifetime τ signals, and Scheme S2 of the Supporting Information for a clarifica tion on the extracted information). The resulting faradaic yield only amounted to 38.4 ± 10.7% for the 0.25 cm 2 devices, due to the particular affinity of the O 2 bubbles to the glass walls of the measuring cells. The oxygenglass affinity also explains why a slow equilibrium is reached after the end of the chrono amperometry, since the oxygen needs to first dissolve from the bubbles into the solution, then diffuse to the upper gas space ( Figure S12, Supporting Information). A similar value of 31.2 ± 9.2% was obtained for the 1 cm 2 devices.

Perovskite-BiVO 4 |TiCo PEC Tandem Devices
For the tandem systems, a backtoback configuration has been preferred over a wired PEC design due to several reasons. The backtoback monolithic design is closer to the idealized artificial leaf and also provides sufficient spatial separation between the H 2 and O 2 evolution sites. Thus diffusion of the produced oxygen to the cathodic compartment and its sub sequent reduction is less likely even in the absence of an ion exchange membrane. High performing BiVO 4 photoanodes are designed to maximize scattering. [103,[107][108][109][110] Therefore, the small gap between the BiVO 4 |TiCo front side and the perovskite back side ensures that most of the scattered light actually reaches the perov skite photoactive area. The use of 3D printed sample holders of custom size also ensures that no additional scattered light reaches the perovskite component (i.e., by circumventing the BiVO 4 active area).
Examples of typical cyclic voltammetry scans and long term stability tests of the 0.25 cm 2 photoelectrochemical tandems are found in Figure 5c,d, respectively. The data for all tandems can be found in Figures S22-S25 of the Supporting Information.
As seen from Figure 5c, using the BiVO 4 |TiCo photoanode enables the construction of a biasfree water splitting PEC tandem with an inversestructure perovskitebased photo cathode. Four of the five 0.25 cm 2 devices investigated present an average photocurrent density of 0.39 ± 0.15 mA cm −2 at zero applied potential bias (see Figure S22 of the Supporting Information). This corresponds to a biasfree photontocurrent efficiency of 0.49 ± 0.18%, which lies close to the maximum achieved at around 0.2 V (see the calculated applied bias photontocurrent conversion efficiency (ABPE) curve from Figure S13 in the Supporting Information). The onset poten tial for water splitting lies around -0.6 V, the most negative reported for photo electrochemical devices, which means that  the tandems can provide enough driving force for simulta neous solar fuel and electricity production. The tandem devices possess an excellent stability of up to 20 h (see a 1 cm 2 device in Figure S23a of the Supporting Information), with an average faradaic yield toward hydrogen production of 71.4 ± 13.0% after 12 h. The slower increase in the FY value may be caused by the larger volume of the 3Dprinted PEC cell and the tex tured nature of its polylactic acid (PLA) walls, which favors gas leakage and the entrapment of bubbles for the smaller samples even more. Accordingly, the solartohydrogen conversion effi ciency of the devices amounts to 0.35 ± 0.14%.
The BiVO 4 light filtering and scattering is responsible for a lower perovskite photocathode response of -1.73 mA cm −2 at 0 V versus RHE, as exemplified in Figure 5a (green curves). This shifts the intersect between the perovskite and BiVO 4 |TiCo photosignals (i.e., the ideal biasfree photocurrent) from 1.13 to 0.85 mA cm −2 . Assuming ideal conditions (100% FY, no ohmic or optical losses), the biasfree 0.85 mA cm −2 would cor respond to a STH efficiency of 1.05%.
The 1 cm 2 devices ( Figure S23, Supporting Informa tion) reached a similar biasfree photocurrent density of 0.43 ± 0.08 mA cm −2 , corresponding to a photontocurrent efficiency of 0.53 ± 0.10% and an STH efficiency of 0.37 ± 0.08%. Three out of the four investigated devices delivered an average hydrogen faradaic yield of 69.9 ± 5.2% within 12 h.
The excellent stability of the tandem devices also gives a hint toward the degradation pathways. Since the back side of the electrodes is completely sealed by epoxy resin in the backto back configuration, this suggests that water is infiltrating the encapsulated devices from the glass/epoxy interface. Another explanation for the improved stability may be found in the lower photocurrent density, which allows the gases to diffuse away from the electrode surface without vigorous bubbling. The gas bubble formation may also affect the stability of the Field's metal/epoxy interface, as observed from a decreased stability when purging the measuring cell with the perovskite photo cathodes immersed in solution.

Scalability Studies
In order to test the scalability of the PEC tandems and corre sponding photoelectrodes beyond the commonly reported sizes of up to 1 cm 2 , [21,48,65,107] larger devices of 4 and 10 cm 2 were also prepared using the same deposition techniques (see the Supporting Information for particular fabrication details). All larger devices were characterized within the 3Dprinted PEC cell, in either a 2 or 3electrode configuration. A comparison of typical photocurrents is found in Figure 3 for the PV and PEC devices of various sizes. The raw data recorded for the perovskitebased photocathodes, BiVO 4 |TiCo photoanodes, and PEC tandems can be found in Figures S16-S18, S19-S21, and S22-S25 (Supporting Information), respectively.
The photocurrent response of the perovskite photocathodes follows a decreasing trend similar to the one observed for the PV components in respect to their size (Figure 3a,b). As with the efficiency of the photovoltaic cells (Figure 2), the absolute value of the photocathode current density at 0 V versus RHE decreases exponentially with the photoactive area ( Figure S11a, Supporting Information), reaching only -2.68 ± 0.80 mA cm −2 for the 10 cm 2 devices (22% of the 0.25 cm 2 photocathodes' signal). The photocathodes also show a lower fill factor than the PV cells, due to a combination of the series resistance with the solution resistance and the electrochemical overpotential for H 2 evolution.
While the photocurrent density of the BiVO 4 |TiCo elec trodes is lower than that of their perovskite counterparts in the 3electrode configuration, the former appear to scale up better, with values stabilizing around 1.2-1.4 mA cm −2 at 1.23 V versus RHE for the backside irradiated photoanodes (see Figure 3c and Figure S11b, Supporting Information). The finding is consistent to previous reports, [107,111] which indicate that higher photocurrent densities are obtained when irra diating a small area. This observation may be explained by a higher overall homogeneity for larger devices (see higher error bar for the 0.25 cm 2 photoanodes; Figure S11b, Supporting Information), as well as by lower kinetic limitations through ionic diffusion to the electrode surface. [112] The influence of the mass transport limitations on the shape of the cyclic vol tammograms and absolute photocurrents are clearly visible from Figure S14 of the Supporting Information, which com pares the voltammetric traces recorded in a stagnant solution to the ones under stirring. Accordingly, we report all PEC char acterization of the individual BiVO 4 |TiCo photoanodes under stirring.
The perovskiteBiVO 4 |TiCo PEC tandems also demonstrated a remarkable scalability, as shown by the roughly similar bias free photocurrents of 0.2-0.5 mA cm −2 plotted in Figure 3d. However, a combination of greater series resistance and mass transport limitations have an increasing impact especially for the 10 cm 2 devices, which present a lower biasfree photo current of 0.23 ± 0.10 mA cm −2 (Figure 3d and Figure S11c, Supporting Information) and a current peak around 0.6 V (Figure 3d and Figure S25, Supporting Information). While the BiVO 4 performance commonly depends on the irradiated side, the backtoback tandems with frontside irradiated BiVO 4 |TiCo still present similar photocurrents to the ones of the character ized photoanodes (≈0.8 mA cm −2 at 0.9 V applied bias, a close estimate of 1.23 V versus RHE). Nevertheless, the low trans parency of Perspex acrylic glass at wavelengths below roughly 380 nm (see Figure S10 in the Supporting Information) may also explain the lower photocurrents.
Due to a higher probability of defects and pinholes in the perovskite deposition, Field's metal film, or epoxy encapsula tion, the larger area devices presented a lower stability in use, which only reached up to 6 and 14 h for the 4 and 10 cm 2 tan dems, respectively (see Figures S24 and S25 in the Supporting Information). Nevertheless, the results obtained for our hand made devices remain very promising, showing potential for further stability improvements in an automated fabrication process.

External Quantum Efficiency (EQE) Spectra
To gain a deeper insight on the processes influencing the light conversion to fuel, EQE spectra are also recorded for both photovoltaic and photoelectrocatalytic components www.advenergymat.de www.advancedsciencenews.com with the averaged results given in Figure 6. The spectra are recorded at applied biases of 0 V for the perovskite PV cells, 0 V versus RHE for the perovskite photocathodes, 1.23 V versus RHE for the BiVO 4 |TiCo photoanodes, and 0 V for the PEC tandems. As expected, the average EQE spectrum of the six 0.25 cm 2 BiVO 4 |TiCo electrodes plateaus around 30% at wavelengths below 500 nm, whereas the perovskite PV cells harvest light with an efficiency around 80% over the entire vis ible range (the upward drift in the lower wavelength region is due to instrumental limitations). The integrated photocurrents obtained from the EQE spectra amount to 1.66 ± 0.44 mA cm −2 at 1.23 V versus RHE and −18.8 ± 1.3 mA cm −2 at short cir cuit conditions, respectively, which are consistent to the IV data obtained under 1 sun irradiation.
Surprisingly, the EQE spectrum of the perovskite based photocathodes was similar to that of the under lying solar cells. Its integration results in an ideal J SC of -19.3 ± 1.9 mA cm −2 at 0 V versus RHE, which differs sig nificantly from the -12.1 ± 0.3 mA cm −2 recorded by cyclic voltammetry. This indicates that the maximal performance of the photocathodes is limited by kinetic effects, namely the fast depletion of protons and the vigorous formation of bub bles in the vicinity of the electrode surface, which decreases the active electrochemical area and inhibits a fast diffusion of protons from within the solution. The finding is consistent to reports of similar limitations in BiVO 4 photoanodes [112] and water splitting systems. [113] The mass transport limitation is also observable when recording the cyclic voltammograms in a 0.1 m potassium borate buffer solution (KBi, pH 8.50) without the 0.1 m K 2 SO 4 electrolyte salt, as shown in Figure S15 of the Supporting Information. This observation also contributes toward explaining why the performance of various wired perov skitebased systems [53,[55][56][57]72,114] is similar regardless of their photo voltaic efficiency, since the concentration and nature of the electrolyte solution actually plays the major limiting role.

Comparison with State-of-the-Art and Outlook
Overall, these findings reveal the scalability of the perovskite BiVO 4 |TiCo photoelectrochemical tandem devices. While the STH efficiency of around 0.3% is lower than the 2-6% reported for PV and PEC multijunction systems (i.e., overall ⩾3 photo absorbers), [21,22,57,113] its value compares favorably to state oftheart devices containing oxide, [45,115] or singlejunction siliconphotocathodes. [47,48,50,116] Moreover, the stability reaching 20 h and scalability of up to 10 cm 2 presented in this work counts among the highest reported values, even surpassing most PV-PEC and wired systems (see Table 2 for a detailed literature comparison), which emphasizes the relevance of our findings as an early step toward commercial implementation.
From this point of view, a lower biasfree photocurrent can even prove beneficial for commercial systems by preventing bubble formation. [117] In this case, the dissolved gas could be removed from the PEC cell and separated in a recirculating system, which can be powered by the additional driving force of up to 0.6 V of the tandems. To mitigate the effect of device deg radation and resistive losses on the overall performance of such assemblies, a tiled design consisting of smaller wireless tandems could be constructed, where any faulty device could be simply replaced. For larger tandem devices, the lateral resistive losses caused by the FTO sheets [118,119] may be again avoided by intro ducing metal fingers. [120] In those cases, the total footprint area of the module would also need to be taken into account. [57,86] Alternatively, a low temperature HTL [121][122][123] and BiVO 4 [124,125] deposition on flexible, thinfilm substrates [126][127][128] would provide further commercial advantages, by obtaining lighter devices which may be massproduced by scalable rolltoroll tech niques. [86,89] Concerning the device encasing, modular designs similar to the one presented here could be easily scaledup.
Beyond the practical side, the facile 3Dprinted design could also enable widespread research on solar fuels in developing countries, since the raw materials (PLA, Blu Tack, acrylic glass) cost only a fraction of the price of highly specialized commer cial PEC cells. Accordingly, a few national 3Dprinting work shops could provide small and mediumscale reactors for local laboratories, bringing this fundamental science closer to the communities most in need.

Conclusions
In conclusion, we have investigated the potential of using per ovskiteBiVO 4 photoelectrochemical tandems for biasfree solar water splitting. By employing the CsFAMA triple cation mixed halide perovskite as the photoabsorber and NiO x as the hole selective layer, substantial improvements have been achieved in the performance (-12.1 ± 0.3 mA cm −2 at 0 V versus RHE) and stability (up to 7 h) of 0.25 cm 2 photocathodes, with the cor responding PV devices reaching an efficiency of 13.0 ± 1.2%. The 1 cm 2 backtoback PEC tandems presented a remark able stability of up to 20 h for the biasfree water splitting, with a corresponding solartohydrogen conversion efficiency of 0.37 ± 0.08%. Their very negative onset bias of around -0.6 V enables simultaneous solar fuel and electricity produc tion, provi ding also flexibility for either operation mode.  Due to significant progress in production techniques, an excellent reproducibility and reliability have been obtained for both single pixel solar cells and photoelectrodes. Those advantages enabled a further comprehensive study on the device upscaling from 0.25 to 10 cm 2 , which revealed valu able insights into the resistive and kinetic limitations affecting both photovoltaics and photoelectrocatalysis. The excellent scal ability of our perovskite and BiVO 4 electrodes allowed us to obtain a comparable activity for 10 cm 2 PEC tandems, which are to the best of our knowledge the largest reported devices of their kind. To characterize the performance of the larger bias free PEC tandems, an affordable 3Dprinted measuring cell was also designed, which may be easily adjusted for laboratory research or consumerbased applications.
Our results reveal that encapsulated perovskite photocath odes and the corresponding photoelectrochemical tandems can compete in terms of stability and scalability with more estab lished PV-PEC or PVelectrolyzer wired systems. More gener ally, this study indicates that both perovskitebased photovoltaic and photoelectrocatalytic systems have potential for large scale applications, as long as lowcost designs and the series resist ance of the substrates are taken into account. Inverse-Structure Perovskite Cells: The inverse-structure perovskite cells were prepared using a modified procedure based on previous reports. [75,78,85,90] The FTO layer was selectively etched away with Zn dust and 2 m HCl before cleaning the glass slides in a Piranha solution. A 1.0 m nickel oxide precursor solution was prepared by dissolving Ni(NO 3 ) 2 · 6H 2 O (1.454 g) in 5 mL ethylene glycol and Adv. Energy Mater. 2018, 8, 1801403 Table 2. Comparison between our results and several state-of-the-art PEC, PV-PEC, and PV-electrolyzer water splitting devices. Data is reported under 1 sun simulated irradiation (100 mW cm −2 , AM 1.5G), for bias-free systems only. 334 µL ethylenediamine (1.0 m in the final solution). [90] The solution was filtered twice through a 0.20 µm polytetrafluoroethylene (PTFE) Millipore Millex-FG filter and spin-coated on the FTO glass slides at 5000 RPM for 45 s. The samples were annealed in two steps, at 373 K for 30 min, and at 573 K for 60 min, to give the NiO x HTL. Full details of the NiO x preparation can be found in ref. [90] by Hoye et al. The hot samples were next transferred inside the glovebox for the perovskite and electron transport layer (ETL) deposition. A two-step spin-coating program was employed for the perovskite layer, with an initial spreading at 1000 RPM for 10 s, followed by spinning at 6000 RPM for 35 s. For the 8-pixel and 0.25 cm 2 devices, a smooth perovskite film was achieved by dropping 75 µL chloroform onto the spinning sample 7 s before the end of the program, followed by tempering at 373 K for 30-70 min. [78,79,85,90] A PCBM (99%, Solenne BV) solution in chlorobenzene (CB, 35 mg mL −1 ) was stirred at 343 K for 30 min and filtered through a 0.2 µm PTFE filter before use. The PCBM solution was spin-coated at 3000 RPM for 45 s to form the ETL. 35 µL of a PEIE (5.8 µL) solution in isopropanol (IPA, 1.5 mL) was next spin-coated under ambient conditions (3000 RPM, 30 s), before returning the samples under inert atmosphere for storage. Lastly, a 100 nm silver layer was evaporated to form the electrical contacts of the solar cells. The deposited layers are visible from the magnified cross-section of a 0.25 cm 2 inverse-architecture device in Figure 1a and from its schematic depiction in Figure 1b. Specific details regarding the preparation of larger devices are given in the Supporting Information.

Experimental Section
Field's Metal-Epoxy Encapsulation: To prevent degradation of the perovskite layer by the aqueous electrolyte, a Field's metal-epoxy encapsulation was employed. [11,75] In this case, 0.5-1 mm thick protective foils were obtained by melting the bulk alloy onto a clean hotplate at 363 K and spreading the liquid metal using common glass slides. After cooling below 323 K, the large area foils could be peeled off the hotplate, and cut into the desired dimensions. A small piece of the metal foil was first melted on top of the silver contact, by applying a 2.4 A current for 30 s to a Peltier thermoelectric element, and then solidified by applying −2.4 A for 30 s (2-step chronopotentiometric sequence on a IviumStat instrument). After sealing the edges with Araldite 5-Minute Rapid two component epoxy, the photoelectrodes were stirred in a K 2 PtCl 4 (21.4 mg, 51.6 mmol) aqueous solution (5.16 × 10 −3 m, 10 mL) for 20 s and left immersed for a further 40 s, resulting in an electroless deposition of catalytically active Pt nanoparticles on the Field's metal. All encapsulation steps were conducted at ambient conditions. BiVO 4 |TiCo Photoanodes: BiVO 4 photoanodes were prepared similarly to previous reports. [47,103] In case of the 0.25 cm 2 devices, a solution was prepared by dissolving Bi(NO 3 ) 3 · 5H 2 O (0.194 g, 0.4 mmol) and NaI (1.199 g, 8.0 mmol) in Milli-Q water (20 mL). After sonicating with an ultrasonic probe (Fischer Scientific Model 120 Sonic Dismembrator) for 3 min at 100% amplitude, the pH was adjusted to 1.20 using concentrated nitric acid. A second solution consisting of p-benzoquinone (0.292 g, 2.7 mmol, 0.3 m) in absolute ethanol (9 mL) was also sonicated for 3 min. The two solutions were mixed and stirred for 30 min at room temperature to obtain a dark brown BiOI precursor solution. After cleaning the FTO glass slides for 10 min in Piranha solution, and masking the active area with adhesive tape, the orange BiOI layer was electrodeposited onto the active area of the FTO slides, by maintaining a potential of -0.3 V against a Ag/AgCl reference electrode for 5 s, followed by -0.1 V for 180 s. A vanadyl acetylacetonate solution was prepared by sonicating VO(acac) 2 (0.530 g, 2.0 mmol) in 5 mL DMSO for 5 min. 40 µL cm −2 of the VO(acac) 2 solution was drop-casted onto the BiOI active areas, before heating the glass slides at 723 K for 60 min with a ramp rate of 1 K min −1 . The brownish V 2 O 5 crust was dissolved from the photoanode surface by gentle stirring for 30 min in a NaOH (0.2 m) aqueous solution.
A [Ti 4 O(OEt) 15 (CoCl)] (TiCo single source precursor, 0.024 g, 24.5 µmol) solution in dry toluene (5 mL) was prepared under inert atmosphere. 20 µL cm −2 of the single source precursor solution were spin coated four times at 2000 RPM for 10 s under air, to obtain a transparent TiCo catalyst layer. [47] The amount of reagents employed for the preparation of the larger electrodes can be found in the Supporting Information.
Characterization: Photovoltaic Cells: The simulated solar irradiation was achieved using a Sun 2000 Solar Simulator (Abet Technologies), which was calibrated to 100 mW cm −2 (1 sun) using a silicon reference solar cell RS-OD4, and by taking a mismatch factor of 1/1.12 into account. The 8-pixel and 0.25 cm 2 devices were investigated under inert atmosphere using gas-tight sample holders, whereas the I-V curves of the larger devices were recorded in their encapsulated form, using the Field's metal as an electrical contact. All devices were characterized without additional masking. The applied potential was varied by a Keithley 2635 source meter, which also recorded the generated current. Reverse and forward I-V curves were determined separately between −0.1 and 1.2 V, with a scan rate of 100 mV s −1 . The EQE of the working devices was determined using a Newport Oriel 66881 setup, an Oriel 74000 Cornerstone monochromator, the Keithley 2635 source meter, a Keithley 2000 multimeter for the Thorlabs SM05PD1A reference photodiodes, a LIA-MV-200-H lock-in-amplifier and a Thorlabs PDA200C photodiode amplifier. The spectra were recorded between 375 and 900 nm, with a 5 nm step size, a 50 ms delay and at 0 V bias.
Characterization: Photoelectrochemical Devices: To investigate the performance of the 0.25-4 cm 2 photoelectrodes, a Newport Oriel 67005 solar light simulator was employed with an Air Mass 1.5 Global (AM 1.5G) solar filter. A LOT-QD LS0816-H large area solar simulator was instead used for the 10 cm 2 devices. The light intensity was calibrated to 100 mW cm −2 (1 sun) using a Newport 1916-R optical power meter. Electrochemical measurements (e.g., cyclic voltammetry, chronoamperometry) were conducted on Ivium CompactStat.e potentiostats, with a sample as the working electrode, a Ag/AgCl/ KCl (sat) (Basi) reference electrode and a platinum mesh counter electrode. The electrodes were submerged in a 0.1 m KBi solution of pH 8.50, which contained 0.1 m K 2 SO 4 as supporting electrolyte. In the 3-electrode configuration, the anodic and cathodic compartments were separated by Selemion (AGC Engineering) or Nafion ion exchange membranes for the 0.25 and 1 cm 2 photoelectrodes, respectively. Due to mass transport limitations, the solution was only stirred in case of the photoanodes, and left unstirred otherwise. The cyclic voltammetry measurements were conducted between −0.2 and 1.2 V versus RHE for the photocathodes, and between 0.1 and 1.4 V versus RHE for the photoanodes, at a scan rate of 10 mV s −1 . The potential versus RHE was determined using Equation (1) at a temperature of 298 K [75] [ versus RHE] [ versus Ag/AgCl] 0.059 pH 0.197 Similar conditions were also employed for the tandem devices, where the scans were conducted in a two electrode configuration, between −0.6 and 1.4 V. To allow ionic movement, a narrow opening was made below the sample instead of using a separating membrane (see Figure 7). The front BiVO 4 active area of the PEC tandems was additionally surrounded by opaque adhesive tape, to prevent any unfiltered light from reaching the perovskite photocathode (see Figure 7b,c).
The EQE was determined using a LOT MSH-300 monochromator, a Thorlabs PM100D power meter with a Thorlabs S302C thermal power sensor, and an IviumStat potentiostat. The wavelength λ (full width at half maximum of 15 nm) was typically varied between 300 and 800 nm in 25 nm steps every 30 s. The resulting EQE was determined using Equation (2), where h is the Planck constant, c is the speed of light, J is the photocurrent density, e is the electron charge, and Pλ is the wavelength-dependent light intensity flux (or irradiance) [129] EQE hcJ e P λ = λ (2) A Shimadzu GC-2010 Plus gas chromatograph (GC) was employed for H 2 evolution measurements. The gas-tight electrochemical cells were purged with nitrogen containing 2% methane as an internal standard. The amount of H 2 was determined by manually injecting 100 µL from the PEC cells' headspace using gas-tight syringes. A NeoFox-GT fluorometer and Fospor-R fluorescence oxygen sensor probes from Ocean Optics were used to determine the amount of produced O 2 . The faradaic yield (FY) of the photoelectrodes was determined by comparing the amount of evolved gas (n gas ) to the total charge passed (Q), as shown in Equation (3). The equivalent charge used per gas molecule amounts to z = 2 for the H 2 generation, and to z = 4 for the O 2 evolution, while F is the Faraday constant The total charge was obtained by integration of the recorded current (I) over the duration (t) of chronoamperometry. This is illustrated in Equation (4), where t 0 and t e represent the beginning and end-points of the measurement Henry's law was employed to compensate for the small effect of dissolved gases, [130] although it is worth noting that the concentration in solution of an evolving gas is probably higher than its equilibrium level.
In order to characterize the performance of tandem devices, the ABPE was additionally calculated using Equation (5), where U bias is the applied bias, and P total is the total recorded light intensity flux (100 mW cm −2 ) [47] ABPE (1.23 ) bias total J V U P = − Similarly, the AB-STH efficiency can be also calculated as the product of ABPE and FY at a given applied bias (in our case, U bias = 0 V).
Characterization: Statistics: The reported values were averaged from the data of at least three PV or PEC devices unless otherwise noted. The given errors correspond to the calculated standard deviation. Examples of typical data are taken from stable devices which follow the general trend, having around or above average performance.
Characterization: Materials: An FEI sFEG XL30 scanning electron microscope (SEM) was employed to investigate the surface and cross-plane topology of the devices. Energy-dispersive X-ray (EDX) spectra were recorded using an Oxford Instruments X-Max silicon drift detector to confirm the elemental composition of various samples. The formation of the polycrystalline perovskite phase was confirmed by XRD measurements, which were conducted on a Panalytical Empyrean X-ray diffractometer. The spectra were recorded using Cu Kα radiation, in a θ-2θ configuration. The angle of the incident beam was varied between 10° and 90° with a step size of 0.008. The thin film samples were rotated at a speed of 30 RPM.
3D-Printed PEC Cell: Common glass PEC H-cells with membrane separation were employed for the characterization of 0.25 and 1 cm 2 photoelectrodes (see ref. [75] and Video S1 in the Supporting Information). A 3D-printed cell was developed for the larger electrodes and all back-to-back tandems, to maintain a constant sample position and ensure a certain separation between the cathodic and anodic compartments. In comparison to other recent models which have employed 3D-printing for electrodes, [131][132][133] electrochemical flow reactors [134][135][136][137] or biased photoelectrodes, [138] this represents the first reported 3D-printed PEC cell design for bias-free tandems.
While existing (mainly glass) PEC cell models are often made to accommodate samples of fixed dimensions, our modular design provides a facile way to measure samples of different sizes in various configurations (i.e., 3-electrode, monolithic, wired, or even colloidal systems). For this purpose, replaceable sample holders are employed which can slide along the side rails. The holders cover the nonactive area of the devices, to prevent additional unfiltered light from reaching the back-side of the tandems. The working principle is illustrated by the 3D model in Figure 7a, where the thin slides supporting the sample are highlighted in green and the lateral rails are depicted in yellow. Photographs of the 3D-printed PEC cell during use can be also found in Figure 7b,c, revealing the back-and front-side of a 4 cm 2 tandem, respectively.
In order to obtain a light-weight, inexpensive, and easily adjustable PEC cell, the 3D components were designed in SolidWorks, and created from PLA using an Ultimaker 2 Extended+ 3D printer. Sketches of the main components and their assembly are given in Figure S1 of the Supporting Information. The PEC cell's windows consisted of 3 mm thick Perspex acrylic glass, whereas Blu Tack reusable adhesive was used as a sealant.

Supporting Information
Supporting Information is available from the Wiley Online Library or from the author. The raw data that support the findings of this study are available from the University of Cambridge data repository, https://doi. org/10.17863/CAM.24084.