Nanoporous anodic alumina with ohmic contact between substrate and infill: Application to perovskite solar cells

The use of a nanostructured aluminum substrate as the cathode for a perovskite photovoltaic device is described. This cathode consists of an aluminum substrate onto which a highly ordered array of aluminum oxide tubular pores was grown via anodization. The 1‐μm thick pores – arranged in a closed‐packed hexagonal pattern – were subsequently selectively etched at the bottom to remove the so‐called aluminum oxide barrier layer. The subsequent infiltration of the pores with the components of a methylammonium lead iodide perovskite solar cells, and the completion with a semi‐transparent anode, have shown to allow the establishment of an ohmic contact between the substrate itself and the components infiltrated into the Al2O3 pores. Indeed, a clear rectifying behavior was observed on the full devices, as well as modest photovoltaic conversion efficiencies. This paper demonstrates that an ohmic contact can be established between the aluminum substrate from which nanoporous anodic alumina was grown, and that the pores can be used to compartmentalize the infill material down to the nanoscopic level.


| INTRODUCTION
Nanoporous anodic alumina (NAA) has attracted a great deal of interest in the last two decades. Since the discovery of the two-step anodization procedure by Masuda et al. allowing the fabrication of highly ordered pores at the nanoscopic scale, 1-3 NAA has found applications in various fields of research. 4,5 Indeed, the ability to fabricate layers or laminas with pores having a wide range of thicknesses, diameters, and interpore distances allows them to easily adapt to new applications. [6][7][8][9] To date, NAA has been applied to an extremely vast field of research, such as optical sensors, [10][11][12][13] biosensors, [12][13][14][15][16] cell culture, 17 drug delivery, [18][19][20] crystallization enclosures, 21 energy storage, and energy generation 22-28 devices among others. However, NAA still suffers from the presence of an insulating barrier layer oxide at the bottom of the pores, which does not allow for the establishment of an ohmic contact between the material infiltrated in the pores and the substrate. Consequently, NAA has had limited applications for energy storage devices (batteries, supercapacitors) or energy generation devices (photovoltaic cells), electrical sensors, where the materials infiltrated in the NAA layer would need to be electrically connected to the aluminum electrode. Some research groups have described procedures based on successive reanodization and chemical etching steps, that partly address these setbacks. [29][30][31][32][33][34] However, these procedures are usually tedious and rarely lead to the complete removal of the barrier layer. They include reanodization steps in sulfuric or oxalic acid using incremental voltage stepdown, 34 incremental current stepdown, 29,34 or constant current successive reanodization 30 and cathodic polarization 31,33 procedures. We have recently reported a procedure to selectively remove the barrier layer on mild anodization NAA with interpore distances superior to 300 nm. 35 That method allows for the complete removal of the barrier layer, allowing, in turn, the establishment of an ohmic contact between the material the pores are filled with and the aluminum substrate.
In order to assess the applicability of the aforementioned method to the manufacture of devices in which an ohmic contact is required, herein, we describe a perovskite solar cell (PSC) in which the nanostructured alumina is used as an enclosure for the perovskite absorber layer (methylammonium lead iodide, MAPbI 3 ) and where the aluminum substrate functions as the cathode. The confinement of the material in the pores is expected to have a significant impact on the crystallization behavior of the MAPbI 3 , as previously demonstrated. 28,[36][37][38][39] Additionally, PSC devices in which the absorber perovskite was partly infiltrated in a nanoporous scaffold were shown to have higher stability with respect to standard architectures, since the encapsulation of the perovskite absorber has been shown to retard the diffusion of moisture through the active layer. 28,38,40,41 Moreover, the crystallization of organic materials from nanoscopic enclosures has been demonstrated to follow alternative thermodynamic and kinetic pathways than those from open-air crystallization methods. As such, crystallizations performed in NAA pores often allow the formation of otherwise unstable polymorphs. They can favor the crystallization toward a preferential direction of the lattice and can alter the kinetics of the crystallization process. [42][43][44][45][46] A barrier layer-free 1-μm thick NAA layer was fabricated as described in our previous report, 35 and the constituent of the perovskite device where successively infiltrated in the pores. The completed device comprised a semi-transparent anode consisting of alternating thin layers of Ag and WO 3 , 47 deposited via thermal evaporation on top of the hole transport material (HTM). The electrical and photovoltaic properties of the device demonstrated that an ohmic contact was established between the aluminum substrate and the materials infiltrated into the pores. It additionally demonstrated the compatibility of the NAA with the manufacture of complex electrical devices, as the study expounded herein showed a clear rectification behavior and a somehow modest photovoltaic conversion.

| RESULTS AND DISCUSSION
The fabrication of the barrier layer-free NAA (f-NAA) substrates was based on a recently published work by our group 35 in which the initial array of ordered pores was submitted to consecutive reanodization and etching steps (see Supporting Information). The pores were subsequently filled with the standard materials used in methylammonium lead iodide perovskite solar cells (MAPbI 3 ) to demonstrate its applicability to photovoltaic devices fabrication. [48][49][50][51][52][53][54] The procedure followed to fabricate such a device is depicted in Figure 1B and relies on wet deposition techniques and on the thermal evaporation of a WO 3 /Ag semi-transparent anode. 47 A picture depicting a real device is as well shown in Figure 1C. The band diagram of the complete device is depicted in Figure 1A, and shows, as expected, a nearly perfect band alignment of the conduction and valence bands of the absorber with the electron-and hole transport materials (ETM and HTM). Moreover, the work function of the aluminum (from the substrate) and the WO 3 from the anode are, according to literature values, compatible with efficient electron and hole injection.
The choice of the pore size and NAA layer thickness was made in accordance with the PSC architecture prerequisites, that is, optimum MAPbI 3 domain size, optimum active layer thickness, as well as the minimum thickness practically attainable for NAA layers. 50,55,56 Indeed, the optimum domain size of MAPbI 3 PSC devices is in the range of 200-500 nm; therefore, an alumina layer with pore width in the 350-400 nm appeared optimum to us, as it could accommodate a single (crystalline) domain in each pore. Since the optimum active layer thickness is in the 300-500 nm range, and the optimum thickness of the top HTM layer is around 100 and 200 nm, an NAA thickness of 500-600 nm should, therefore, be sufficient to accommodate all afore-mentioned layers and confer added protection toward the decomposition of the MAPbI 3 due to the encapsulation effect of the NAA pores. However, we turned our focus to NAA layers with slightly higher thicknesses (approximately 1000 nm) since NAA layers with aspect ratio over 2 are usually easier to manufacture and characterize, even though NAA substrates with a thickness of 600 nm are in principle achievable. We believe that the optimization of the NAA thickness lies beyond the scope of this study. The fabrication of the barrier layer-free NAA (f-NAA) substrates consisted of the sequential anodization of aluminum substrates at a constant voltage, chemical etching, and reanodization under a constant current. The initial array of ordered nanopores was grown using the Masuda two-step method at a constant voltage of 195 V, in a 1% phosphoric acid solution at −5ºC, producing NAA with an interpore distance of 350-400 nm. The substrates were subsequently submitted to a partial chemical etching in a 5% phosphoric acid solution at 35°C (see Supporting information for details). This last step, as well as etching the pores' walls, etches the barrier layer to a final thickness of about 80 nm ( Figure S1B). NAA was subsequently submitted to a constant current reanodization, which results in the formation of nanoscopic channels through the barrier layer, reaching the aluminum surface ( Figure S1C). Upon submitting the NAA layer to a final chemical etching step, the now weakened barrier layer dissolves at a faster rate than the pores' walls, leaving the bottom of the pores free of alumina. 57 Importantly enough, this final etching step, in addition to achieving the complete removal of the barrier layer, also leaves the bottom of the pores with nanoscopic longitudinal openings ( Figure 2B). These openings play a crucial role in the infiltration of the layer in direct contact with the aluminum substrate as it allows the liquid to channel through the openings and allows overcoming the force exerted by the air being trapped at the bottom of the nanoscopic pores if they were sealed. The effect is seen visually, as when a drop of liquid is cast in the middle of the substrate, the liquid diffuses over the entire layer through capillarity. Therefore, the first layer of the assembly was readily infiltrated into the pores via spincoating, in air, of a sol-gel precursor of titanium dioxide. Calcination of the substrates in a furnace at 500°C yielded the TiO 2 layer with no damage to the NAA porous structure nor to the aluminum substrate. Due to the thinness of the resulting nanostructured TiO 2 layer, obtaining a precise measurement of the layer thickness is complex using scanning electron microscopy (SEM), however, a rough estimate was possible using cross-section SEM images, as seen in Figure 2C. The image shows a TiO 2 layer with an approximate thickness of 40 nm.
The infiltration of the subsequent MAPbI 3 material into the NAA required careful optimization of the deposition conditions since, after the infiltration of the TiO 2 layer, the openings at the bottom of the pores are sealed to a significant extent, by the latter material. Our first attempts to drop-cast diluted solutions (0.25 and 0.5 mol L −1 ) of MAPbI 3 precursors and placing the substrates on a hot plate, did not provide the expected homogeneous crystal growth originating from the bottom of the pores. Indeed, due to the highly crystalline nature of MAPbI 3 , any sharp edge provides for a nucleation site that triggers the crystallization of MAPbI 3 from random locations on the NAA layer. We, therefore, turned our focus to spin-coating to achieve a more homogeneous filling of the pores. To facilitate the displacement of the air trapped at the bottom of the pores upon casting the solution of precursors, the substrates were placed under vacuum for a minute before being spun (see Supporting Information for details). A preliminary result of spin-coated-filled NAA in which a high concentration of precursors was used is depicted in Figure 3A,B. Similar to the results obtained from drop-casting (vide supra), one can observe from the SEM images, that large crystals have formed from random nucleation sites on the surface of the NAA layer. This effect seemed to be predominant regardless of the spinning velocity, implying that MAPbI 3 material is pulled out of the pores during the crystallization process, presumably by diffusion from a semiliquid state during the growth step. This is clearly evidenced from the voids (dark areas in Figure 3A,B) left after the film was dried.
The optimization of the precursors' concentration was, therefore, carried out, and the best results were obtained with a 2 mol L −1 concentration of PbI 2 and MeNH 3 I. Lower concentrations of precursors led to partially filled pores and the formation of lumps on the surface of the NAA, an effect clearly seen on the image from Figure 3C,D.
Contrary to the 2.5 and 1.5 mol L −1 solutions, the 2 mol L −1 concentration showed to yield MAPbI 3infiltrated layers with no lumps nor large crystalline aggregates on top of the NAA, as well as providing a high degree of pore-filling. The top views in Figure 3E,F show only a few areas with unfilled pores, while cross-section images recorded with a backscattered electrons detector ( Figure 3G,H) confirm that the pores are on average filled to about half of their height. These images show how the filling is homogeneous across large areas of the substrate and can be extrapolated to the whole NAA layer. Noteworthy, the vast majority of the pores are filled from the bottom-up, with no significant voids at the bottom of the pores as observed in the layers manufactured from 1.5 and 2.5 mol L −1 solutions (see Figure S2). The spinning velocity was carefully optimized to provide for layers of an approximate 500 nm thickness; however, spinning speeds could only be varied within a very limited range without significantly impacting the homogeneity of the filling.
Additional to the SEM images, the NAA layers were further characterized using X-Ray diffraction using both Bragg-Brentano geometry and grazing incidence angles with an area detector. The conventional diffractograms clearly show that the MAPbI 3 layer is crystalline, and the crystals are from the typical tetragonal structure from the I4-cm space group, as expected for this combination of lead and iodine precursors. 58,59 A comparison with a layer deposited in identical conditions of concentration (2 mol L −1 ) and spinning velocity on a bare aluminum substrate reveals that the crystallinity of the MAPbI 3 layer is somewhat similar in terms of crystallite size and volume, however, and quite importantly, the MAPbI 3 precursor solution has very poor film-forming properties on bare aluminum substrates, and the resulting films show extremely poor homogeneity up to the macroscopic level. A Pawley profile adjustment of the diffraction patterns allowed establishing that the MAPbI 3 crystallites are on average 208 nm large in the case of the NAA and 210 nm in the case of the bare aluminum (see Figure S3). These findings strongly support the evidence that the nucleation step is predominant in both cases, rather than the growth step, contrary to the NAA filled with 2.5 mol L-1 MAPbI 3 precursors where growth seems to be predominant. Due to the fact that the crystallites' size is similar in both NAA and bare aluminum substrates, we concluded that the concentration of the MAPbI 3 precursors is determinant for the crystalline properties of the MAPbI 3 layer, rather than the confining effect of the pores. However, a comparison of the area images and diffractograms of both the perovskite deposited on bare aluminum substrates and on the NAA show some differences in texture. That is, both the MAPbI 3 deposited on bare aluminum substrates and that infiltrated in the NAA pores show evidence of preferential crystalline orientation as seen by the mismatch between the peaks' intensity of the theoretical diffraction pattern and the experimental diffractograms ( Figure 4A,D,E). In the case of the MAPbI 3 deposited on aluminum substrates, a correlation with the corresponding 2D diffractogram (area image in Figure 5A) confirms a preferential orientation as seen by the stronger intensity of the Debye rings in the out-of-plane direction for the 004/220-diffraction planes. The additional calculated 1D diffractograms with a preferential orientation toward the (004) or (220) planes show a fair match between the latter and the experiential diffractogram, advocating, therefore, that the crystallites are oriented with the 220-diffraction plane in the out-of-plane direction ( Figure 4B-D).
A similar analysis applies to the MAPbI 3 layer infiltrated into the NAA. However, although 2D diffraction patterns in Figure 5B show a clear preferential orientation of the 220 or 004 planes in the out-of-plane direction, in this case, the 1D diffractogram ( Figure 4E) does not allow to discriminate one preferential orientation with respect to the other since the intensity ratio of the corresponding diffraction peaks in not significantly changed compared to the calculated diffractogram of randomly oriented crystals ( Figure 4A). It, therefore, results that the MAPbI 3 layer is composed of at least two groups of crystallites each having a preferential orientation. A closer look at the 2D diffractogram showed additionally a significant orientation toward the 202 plane, supposedly corresponding to a third group of crystallites with preferential orientation. To verify this hypothesis a calculated diffractogram corresponding to the sum of the diffractograms with a preferred orientation toward the 004, 220, and 202 planes was compared with the experimental diffractogram of the MAPbI 3 infiltrated in the NAA. Adjusting the ratio between the 3 fractions to 34%, 55%, and 11%, respectively, produced a fairly accurate correlation between the two sets of data ( Figure 6).
To conclude, these results show that, presumably due to the fast kinetics of the nucleation and growth processes of the MAPbI 3 , the NAA does not play a significant role as a scaffold for the homogeneous growth of large MAPbI 3 crystals within the cavities of the pores. However, the NAA has been shown to have a significant impact on the orientation of the crystallites with respect to the bare aluminum surface. This is in accordance with a previous study from Lee and coworkers where MAPbI 3 was shown to grow crystallites in the preferential <001> plane direction when infiltrated in NAA with up to 200 nm diameter. 39 The orientation is in our case less pronounced than in Lee and coworkers' work; however, it seems to follows their trend in which the orientation's magnitude decreases with the pore's diameter. Similarly, the orientation toward three different crystallographic planes may well be linked to the larger pore diameter of our NAA with respect to Lee's work. Despite these properties, the crystalline nature of the material fits with the prerequisites for a functional PSC device. The next step is, therefore, the infiltration of the HTM, which was chosen to be spiro-OMeTAD for its widespread usage and effective hole transport properties even when thick layers are employed. 60 Indeed, in this particular F I G U R E 4 Calculated diffractograms of MAPbI 3 from the tetragonal structure (I4cm, a 8.849; b 8.849, c 12.642; at 293K) with no preferential orientation (red line), preferential orientation toward the (004) plane (blue line), preferential orientation toward the (220) plane (green line), and experimental diffractogram of the MAPbI 3 layer deposited on bare aluminum substrates (black line) case, the spiro-OMeTAD layer was deposited to provide -additional to the hole conduction properties-for a complete leveling of the NAA pores top openings. This is a prerequisite for the subsequent deposition of the thin WO 3 / Ag alternating layers composing the anode, to grant electrical continuity over the whole device surface. Therefore, the overall thickness of the spiro-OMeTAD layer resulted to be somewhat higher than optimum (ie >800 nm vs. 150 nm) and is expected to incur some resistive losses. However, these losses are a trade-off accounted for in our study, and are not expected to interfere with the operation of the device, since state-of-the-art PSC devices with up to 700 nm thick spiro-OMeTAD have been reported with high fill factors and efficiencies approaching 11%. 61 The deposition of the spiro-OMeTAD layer was carried out via spin-coating as for the previous layers; however, it was found that the infiltration into the pores and the optimization of the layer thickness was less complex than in the case of the MAPbI 3 . This is most probably due to the amorphous nature of the material, and to the fact that the pores were already partially filled with MAPbI 3 prior to the deposition of the HTM. Figure 7B,D show the effectiveness of spiro-OMeTAD in filling the pores and leveling the NAA layer before the cathode deposition. The impact of incomplete leveling is also shown, in Figure 7A-C, where the subsequently deposited anode forms bumps, whose shape follows the hexagonal pattern of the pores. This further highlights the importance of the HTM in leveling the surface of the NAA before the deposition of the anode. Consequently, when the spiro-OMeTAD is sufficiently thick the multilayer cathode displays a very flat surface over the area of the substrate with no obvious discontinuities, as seen in the SEM images ( Figure 7E,F). Figure 7G shows a cross-section SEM image (from a backscattered electrons detector) of a complete device where all the layers are stacked above one another. Each material is highlighted with a different color and the thicknesses of the MAPbI 3 , spiro-OMeTAD, and semitransparent electrode are estimated to be 500, 900, and 230 nm, respectively. The cathode shows to be completely homogeneous and without disruption, guarantying electrical conduction over the whole surface of the device.
Finally, the electrical properties of the full devices were characterized. The current-voltage (J-V) characteristics of the best performing PSCs devices measured in the dark and under simulated AM1.5G illumination are shown in Figure 8. The main properties of the devices are summarized in Table 1. All in all, the devices displayed modest values of open-circuit voltage (V OC ), short-circuit current density (J SC ), fill factor (FF), and power conversion efficiency (PCE) with respect to state-of-the-art PSC devices, however, and most importantly, the shape of the J-V characteristics provide unequivocal confirmation that an ohmic contact is established between the aluminum substrate and the successive layers. Indeed, the rectifying behavior demonstrates that electrons are injected into the aluminum cathode and that a built-in field is present under open-circuit conditions. The less-than-optimum J-V characteristics are most probably due to a combination of factors linked to the morphological complexity of the device. First of all, the multilayer arrangement inside pores of such a small size, a premiere in this study, is likely responsible for the formation of voids, due to air being trapped between the layers during the deposition of the successive layers. This can be seen in Figure 7G. Similar voids would probably exist between the TiO 2 and MAPbI 3 layer, although not clearly characterized here. Second, the very thick layer of HTM, necessary to level the NAA layer, although not expected to compromise the function of the device to a large extent, is most probably responsible for significant losses. The high density of structural defects together with the thickness of the overall device are likely responsible for a great amount of charge trapping as well as fast rates of recombination dynamics with respect to charge percolation/sweep out to the respective electrodes. [62][63][64][65] Then, finally and quite importantly the very low photovoltaic effect can, to a very significant extent, be attributed to the strong absorbance of the anode. Indeed, a UV-Visible characterization of the pristine alternating WO 3 /Ag layers showed that the layer possesses a stronger absorbance than that expected from the literature (see Supporting Information for details). Further optimization would be necessary to find a better trade-off between high conductivity and low absorbance.

| CONCLUSION
This study is, to the best of our knowledge one of the first examples of devices in which NAA is used to contain the components of the device and the aluminum substrate is used as an electrode. We demonstrated that an ohmic contact can be established between the latter substrate and the material the NAA is filled with. Moreover, the application of the concept to PV devices of such complexity as PSC clearly demonstrate the applicability of NAA for the fabrication of advanced electronic devices, such as batteries, sensors, photovoltaics, or light-emitting devices in which the confinement properties of the NAA can be put to use to shape the materials at the nanoscopic level and where the aluminum substrate can be used directly as an electrical contact. Interestingly enough, and contrary to our expectations, only a subtle difference in the crystallization behavior of the MAPbI 3 was observed when confined in nanosized pores with respect to when the MAPbI 3 is deposited on a flat surface. This effect is likely linked to the very fast crystal growth kinetics of the MAPbI 3 with respect to the nucleation step. This is further confirmed by the fact that the presence of numerous nucleation sites resulting from the highly porous nature of the NAA does not have a significant impact on the size and crystalline volume of the infiltrated MAPbI 3 material with respect to a MAPbI 3 layer deposited on bare aluminum substrates. Future work is currently underway to gain a deeper understanding of the latter experimental findings. As such, infiltrations of MAPbI 3 precursors with different PbI 2 /MeNH 3 I ratios are currently been investigated as it has been shown that the formation of crystalline MAPbI 3 follows different paths depending on the latter ratio. 66 Additionally, future work will include reducing the thickness of the NAA layer, reducing the thickness of the HTM layer, and decreasing the absorption of the semi-transparent anode. Although the manufacture of PSC devices using such type of NAA scaffold has revealed challenging, this study holds great promise for future applications. Indeed, the versatile nature of the NAA pores, which can be etched selectively using mild acidic solutions at low temperature, makes it a very good candidate to fabricate electrodes with textured materials. Indeed, after the pores have been filled with a water-resistant material one could simply etch entirely or partially the walls of the container's pores to reveal a nearly perfect arrangement of pillars connected ohmically to the aluminum substrate with a very high surface area. This concept is not limited to the pore size described in this article and can be applied to pore sizes ranging from 10 nm to over 500 nm in diameter.

| Titanium dioxide infiltration
First, a titanium dioxide sol-gel precursor solution (TiO xsolution) was prepared in an N 2 -filled glovebox, via the sol-gel method by mixing 0.2 ml of titanium (IV) isopropoxide (Ti[OCH(CH 3 ) 2 ] 4 ) with 0.1 mL of 2-methoxy ethanol (CH 3 OCH 2 CH 2 OH) and 1 mL of ethanolamine (H 2 NCH 2 CH 2 OH). 67,68 Subsequently, the TiO x -solution was diluted in anhydrous methanol (CH 3 OH) with a volume ratio of 1:3 (TiO x -precursor solution: CH 3 OH). 69 Secondly, 100 μL of the diluted TiO x -solution were deposited on the top of different barrier layer-free NAA (f-NAA) substrates. The solution was left standing on top of the substrates for 30s, to let the solution diffuse through the pores, and the substrates were spun at 4000 rpm (2160 rpm s −1 ) for 90 seconds. The substrates were subsequently left in air at room temperature for 1 hour, heated at 150ºC for 10 minutes, and finally heated at 500ºC for 30 minutes in a furnace.

| Methylammonium lead iodide perovskite infiltration
Several MAPbI 3 perovskite precursor solutions with different concentrations (1.5, 2.0, 2.5 mol L −1 ) were prepared in an N 2 -filled glovebox from lead iodide (PbI 2 ) and methylammonium iodide (MeNH 3 I, MAI) using anhydrous N,N-dimethylformamide (DMF) as solvent. Each precursor was dissolved in 500 µL of DMF separately. To promote the dissolution of the lead salt, PbI 2 was heated in DMF at 150ºC for 10 minutes on a hot plate and left to cool down to room temperature after which MAI was added. The precursor solution was subsequently deposited on the NAA substrate as follows: 100 μL of MAPbI 3 perovskite precursor solution was cast on top of different f-NAA-TiO 2 substrates. To facilitate the displacement of the air trapped at the bottom of the nanopores, the substrates were placed under vacuum for a minute before being spun. To do so, a funnel connected to a vacuum pump was tightly fitted to the spin coater chuck using a silicone rubber seal after the precursor solution was cast on top of the substrate. The substrates were spin-coated at 4000 rpm (3996 rpm s −1 ) for 45 seconds, to obtain the desired nanostructured MAPbI 3 perovskite layer with a thickness ranging from 400 nm and 600 nm. Finally, the substrates were placed in a hotplate and were annealed at 100°C for 1 hour to obtain crystalline MAPbI 3 perovskite films.

| Deposition of the anode
After the spiro-OMeTAD had been deposited, the substrates were placed in a thermal evaporator chamber (vacuum pressure <3 × 10 −6 mbars, in an N 2 -filled glovebox), and the semi-transparent electrode was deposited using a shadow mask to pattern the electrode as described by Jun Hee Han et al 70 Tungsten trioxide (WO 3 ) and silver (Ag) were evaporated successively from a tungsten boat at a constant rate (1-2 A/s). 71 The final anode had the following structure: WO 3 (50 nm)/Ag (12 nm)/WO 3 (105 nm)/ Ag (12 nm)/WO 3 (50 nm).
The transmittance spectra in the UV-Visible wavelength range (300-800 nm) of the semi-transparent anode and the reference glass substrate is shown in Figure S4.