Optimized light-driven electrochemical water splitting with tandem polymer solar cells

Tandem polymer solar cells are used for light-driven electrochemical water splitting. To attain a high enough electrochemical potential a new wide band gap electron donor polymer (PTPTIBDT-OD) is developed and used in combination with [70]PCBM as an electron acceptor in a tandem device architecture with two identical photoactive layers. This homo-tandem device comprises an intermediate ZnO/PEDOT:PSS/MoO 3 charge recombination layer to connect the two subcells electrically and optically. The homo-tandem solar cell has an open-circuit voltage of 1.74 V and reaches a power conversion e ﬃ ciency (PCE) of 5.3%. In combination with RuO 2 as the electrocatalyst for oxygen evolution and RuO 2 or Pt catalysts for hydrogen evolution, sunlight-driven electrochemical water splitting occurs with a solar-to-hydrogen conversion e ﬃ ciency of h STH ¼ 4.3%. Owing to the very high ﬁ ll factor of the polymer tandem cell (0.73), water splitting takes place near the maximum power point of the homo-tandem solar cell. As a consequence, the di ﬀ erence between PCE and h STH is only due to the overpotential losses.


Introduction
Electrochemical water splitting occurs at a standard potential of 1.23 V but in practice substantially higher potentials (1.4 to 1.8 V) are required due to overpotentials for the hydrogen and oxygen evolution reactions. As a consequence, sunlight-driven electrochemical water splitting devices use multi-junction cell congurations in which two or more photovoltaic cells are stacked and connected in series to reach the high potential while still using an appreciable part of the solar spectrum. Triple junction stacks of amorphous silicon solar cells are oen used for light-driven electrochemical water splitting purposes, because the summed voltage of the three cells provides the required potential. [1][2][3][4] Recently, we demonstrated that triple junction polymer-fullerene solar cells can also be used successfully for this purpose. 5,6 If the water splitting potential is sufficiently reduced by using appropriate catalysts for hydrogen and oxygen evolution, tandem solar cells may also be used. 7,8 This is benecial in terms of device fabrication, because there is one less junction and photoactive layer, and also in terms of the overall solar-to-hydrogen conversion efficiency. In principle, a tandem solar cell can generate up to 50% higher current density than a triple junction cell at the same optical band gap, 7 because the photon ux is distributed over only two instead of three absorber layers. Provided that the required electrochemical potential is achieved, the solar to hydrogen conversion efficiency (h STH ) only depends on the current density of the solar cell during operation. Hence tandem cells offer an advantage. Of course, the electrochemical potential for sunlight-driven electrochemical water splitting poses a lower threshold for the optical band gap of the semiconductors that can be used. The minimum optical band gap is higher for tandem cells than that for triple junction cells, and this limits the use of the complete solar spectrum.
In our recent example of a water splitting device, a triple junction polymer solar cell operating at a voltage of 1.49 V was used in combination with RuO 2 electrocatalysts for hydrogen and oxygen evolution. 6 A tandem solar cell that is intended to split water at this potential requires an open-circuit voltage (V oc ) of about 1.75 V and only a few examples of polymer tandem solar cells with such high V oc exist, using absorber layers with a wide optical band gap. 5,[9][10][11] The higher V oc is generally achieved at the cost of lower current densities and wide band gap materials that can provide high current densities are scarce. 12,13 One approach towards a high voltage tandem cell is a device conguration in which both photoactive layers are composed of the same wide band gap material. These homo-tandem cells can provide higher efficiencies compared to the corresponding single junction devices because of improved absorption of light and enhanced charge collection. 9,14-16 When light is distributed over two identical but thinner layers, light absorption can be equally high as in one thick layer, but bimolecular recombination of charges decreases because the distance over which charges are collected is reduced, 9 resulting in an overall higher efficiency.
In this paper we demonstrate light-driven electrochemical water splitting using a series connected tandem polymer solar cell. In order to achieve a sufficiently high V oc , a new wide band gap material PTPTIBDT-OD (poly(4,10-(2-octyldodecyl)-4,10dihydrothieno[2 0 ,3 0 :5,6]-pyrido [3,4-g]thieno [3,2- Fig. 1) has been developed. The polymer comprises an electron decient pentacyclic lactam unit that was recently introduced in donoracceptor conjugated copolymers to afford optical band gaps higher than 1.7 eV in combination with a V oc of about 0.9 V, when copolymerized with different p-conjugated donors. [17][18][19][20] To ensure a loss free series connection of the two subcells, we introduce a ZnO/PEDOT:PSS/MoO 3 stack as the intermediate recombination layer. Light-driven electrochemical water splitting performed with RuO 2 -RuO 2 and RuO 2 -Pt electrodes in 1.0 M KOH electrolyte shows a solar to hydrogen conversion efficiency of $4.3%.

Synthesis and characterization
PTPTIBDT-OD was synthesized by a Stille polymerization of 2,8dibromo-4,10-di(2-octyldodecyl)-4,10-dihydrothieno[2 0 ,3 0 :5,6]pyrido [3,4-g]thieno [3,2-c]isoquinoline-5,11-dione with 2,6-bis-(trimethylstannyl)benzo[1,2-b:4,5-b 0 ]dithiophene as described in detail in the ESI. † The resulting polymer had a M n ¼ 35.4 kg mol À1 . The optical band gap (E g ) of PTPTIBDT-OD is 2.04 eV as determined from the onset of absorption in chloroform solution and thin lm (Fig. 1). PTPTIBDT-OD shows a weak uorescence at 642 nm (1.93 eV) with a vibronic progression towards lower energies. The oxidation and reduction potentials of PTPTIBDT-OD were measured using cyclic voltammetry on spin-coated thin lms on ITO covered glass substrates in acetonitrile containing 0.1 M TBAPF 6 as the electrolyte, resulting in estimates for the frontier orbital energy levels of E HOMO ¼ À5.75 eV and E LUMO ¼ À3.19 eV. The electrochemical band gap, E CV g ¼ 2.56 eV, is signicantly larger than the optical band gap.  Table 1). The J-V characteristics and the EQE measurement of this cell are shown in Fig. 2. The shortcircuit current density (J sc ) determined from the EQE aer integration with the AM1.5 G solar spectrum amounts to J sc ¼ 8.38 mA cm À2 and is slightly larger than the one determined from the J-V characteristics (J sc ¼ 8. However, reaching a sufficiently high voltage in a tandem conguration is not trivial. The reason is the nature of the recombination contact. In the frequently used recombination contact of a bilayer of ZnO and pH neutral PEDOT:PSS, 21 the pH neutral PEDOT:PSS layer, which serves as the hole extracting contact for the back cell, has a lower work function (4.65 eV) than that of the common, highly acidic, PEDOT:PSS (5.05 eV). 22 The reduced work function limits the maximum voltage of the second subcell. The use of pH neutral PEDOT:PSS is dictated by the high solubility of ZnO in water at a low pH, which prevents spin coating acidic PEDOT:PSS on top of a ZnO layer. 21 If the back cell uses a polymer with a deep HOMO level to reach a high V oc , the large difference between the low work function of pH neutral PEDOT:PSS and the deep HOMO of the polymer causes a non-ohmic contact and introduces losses in the V oc of the back cell. 5,22 A partial solution for increasing the work function of pH neutral PEDOT:PSS is spin coating a thin Naon layer on top. 5,22 However, a considerable portion of the lost voltage still remains to be recovered aer adding the Naon layer. 5,22 Another approach to improve the work function of pH neutral PEDOT:PSS is depositing a thin layer of molybdenum oxide (MoO 3 ) onto pH neutral PEDOT:PSS by thermal evaporation. Evaporated MoO 3 is a well-known hole collector that can replace PEDOT:PSS in normal conguration single junctions. 23 Because of its high work function (5.40 eV), MoO 3 is expected to prevent voltage losses in high voltage back cells. MoO 3 is also commonly used as a hole extracting contact in inverted single  16 Homo-tandem PTPTIBDT-OD:[70]PCBM cells were made with additional Naon or MoO 3 layers to modify the work function of pH neutral PEDOT:PSS. The tandem device layout is shown in Fig. 4a. In Fig. 4b and Table 1 we show the tandem device characteristics. In these tandem cells, both photoactive layers were around 110 nm and initially no optimization was made for front and back cell thicknesses. As expected, the tandem cell without any work function modication layer resulted in a low V oc of 1.53 V. With the addition of the Naon layer, the V oc improved slightly to 1.58 V, but MoO 3 is signicantly more effective in increasing the V oc up to 1.72 V. Aer optimizing the tandem cell by reducing the front cell thickness for a more balanced current generation in the front and back cells of the series connected tandem solar cell, 26 the resulting cell had a PCE of about 5.3% (Table 1).

Single and tandem junctions
To further characterize the tandem solar cell, the external quantum efficiency (EQE) was measured. Because the solar cell consists of two identical absorber layers, the conventional tandem EQE measurement method described by Gilot et al. 27 cannot be applied. 28 Due to the interference effects in the layer stack, a different spectral response is expected for the front and the back cell inside the tandem. 28 The data markers in Fig. 4c show two different responses recorded when the EQE of the homo-tandem cell was measured without (black squares) and with (red circles) bias light (at an intensity of 532 nm to give a J sc equivalent to that for AM1.5 G white light), both measured at zero electrical bias for the tandem cell. The two measurements clearly show different spectral responses. To gain more insight into the origin of this difference, optical modelling using the transfer matrix formalism was performed on the homo-tandem device. The simulations use the wavelength dependent refractive index (n) and extinction coefficient (k) for all layers in the stack. The n and k values determined for the PTPTIBDT-OD: [70] PCBM blend are shown in the ESI. † The simulations provide the fraction of light absorbed by the front and back cells inside the tandem cell. The fraction of absorbed light at each wavelength was then corrected by the internal quantum efficiency (IQE) to obtain an estimate for the EQE of each of the two subcells (solid lines in Fig. 4c). The IQE used in the simulations was determined from the ratio of a single junction EQE and the fraction of absorbed photons, and was assumed to be the same for both   subcells. When the experimental and simulated EQEs are compared, it is observed that the EQE measurement without optical bias resembles the expected response of the front cell, while the measurement with optical bias coincides fairly well with the predicted response of the back cell (Fig. 4c). The latter is expected, because the front cell absorption at 532 nm is signicantly higher than the back cell absorption (solid black line in Fig. 4c), such that under 532 nm illumination the back cell is current limiting and will be the one that is measured in an EQE experiment (red circles in Fig. 4c). The AM1.5 G integrated current density estimated from the light biased-EQE measurement is 4.00 mA cm À2 and corresponds (in a rst approximation) to the short-circuit current of the tandem cell. Consistently, the value corresponds to J sc ¼ 4.19 mA cm À2 , measured in the J-V characteristics. Without light bias, one would expect the EQE to always reect the current-limiting subcell, but this only holds when both subcells exhibit a negligible dark current. Using the highly conductive pH neutral PEDOT in the intermediate contact, this condition is not always met, and in such a case the unbiased EQE can represent the response of the cell with the lowest dark current as explained in detail by Colsmann et al. 29 This is what is seen in the un-biased EQE experiment. The magnitude of the measured EQE is then lower than that of the actual EQE and consequently the actual current generation of the subcell cannot be estimated in this case.
The V oc of the optimized tandem cell reached up to 1.74 V. Surprisingly, the ll factors of the homo-tandem cells with MoO 3 layers were high (0.73-0.75) likely as a consequence of the reduced light intensity in the subcells compared to the single junction. Thus, the V max of the optimized tandem cell was very high (1.48 V). In fact this V max enables water splitting to take place at the maximum power point of this solar cell.

Light-driven electrochemical water splitting
For a high solar-to-hydrogen energy conversion efficiency (h STH ) in light-driven electrochemical water splitting, an optimized tandem solar cell (0.0676 cm 2 ) was connected to two RuO 2 catalytic electrodes on Ti substrates ($1.2 cm 2 catalyst) for oxygen and hydrogen evolution in 1.0 M KOH. The reasons for  selecting RuO 2 catalysts for both reactions are their low overpotentials, the ease of preparation and the good stability of the catalyst. 6 The solar cell was illuminated with simulated AM1.5 G solar light for 20 min. The current density and voltage characteristics measured during a 20 min water splitting experiment are shown in Fig. 5. Aer 15 min of water splitting, the stabilized operating voltage (V op ) is 1.50 V, which is very close to the V max of the solar cell (1.48 V). Hence, the homo-tandem cell operates almost at its maximum power point, which is tailored to sunlight-driven electrochemical water splitting. Fig. 5a shows the J-V characteristics of the solar cell just before, during and aer the 20 min experiment. The J-V curves of the tandem cell before and aer the water splitting experiment demonstrate the stability of the tandem cell during the test. The current under operation, J op ¼ 3.46 mA cm À2 , shows a minimal decrease with time (Fig. 5b), reecting the stability of the solar cell and the catalysts. Assuming 100% faradaic efficiency, the h STH ¼ 1.23 Â 3.46/100 ¼ 4.3% aer 15 min of operation (Table 2).
In the experiment we use a 0.0676 cm 2 solar cell in combination with catalyst areas of 1.1 and 1.3 cm 2 . The relative areas of the solar cell and the catalyst are important for the overpotential. The Tafel plot in Fig. 6 shows the overpotentials for hydrogen and oxygen evolution for RuO 2 catalysts as a function of the current density. At the current densities reached for the solar cell and the catalyst, the expected total overpotential is about 0.25 V, resulting in a total potential of 1.23 + 0.25 ¼ 1.48 V, which is in very good agreement with the operating potential of 1.50 V (Fig. 5b). If the catalyst area would be reduced, the overpotential would increase and this would move the operating point away from the maximum power point of the solar cell. A reduction of the catalyst area by e.g. a factor of ten would increase the overpotential by 0.1 V and reduce h STH to 3.5%. Hence, balancing the nature and surface area of the catalysts with those of the materials used in the solar cell is crucial to achieve optimal performance.
Next to RuO 2 we also tested Pt as a catalyst for the hydrogen evolution reaction. 3,7,8 For this purpose, the preceding experiment was repeated under identical conditions and with the same homo-tandem solar cell, only replacing the RuO 2 catalyst for hydrogen evolution with a Pt plate ($1.5 cm 2 ). Fig. 7 shows the characterization of a 20 min water splitting experiment. The results are very similar to the previous experiment with two RuO 2 catalysts. Replacing RuO 2 with Pt for hydrogen evolution reduced the stabilized V op to 1.48 V. The resulting J op (3.49 mA cm À2 ) and h STH (1.23 Â 3.49/100 ¼ 4.3%) aer 15 min of operation were very similar to those with RuO 2 as the catalyst for hydrogen evolution. The stabilization of V op and J op in this case was faster. A reason can be that the charging of the double layer on Pt occurs faster than that on RuO 2 for hydrogen evolution. The stability of the homo-tandem cell is evidenced in Fig. 6a.   For future and long term application the stability of the lightdriven electrochemical water splitting stability is important. The 20 min experiments depicted in Fig. 5b and 7b indicate no fast degradation, but cannot be taken as evidence for stable operation. For stable light-driven water splitting the stability of the polymer solar cells and the catalysts is important. The stability of polymer solar cells has been discussed in several reviews [30][31][32] and is determined by the stability of the active layer, the electrodes, and the interface layers which can be inuenced by light, temperature, and the ingress of water and oxygen. The stability of RuO 2 for oxygen and hydrogen evolution reactions in alkaline electrolytes has been described in detail, 33 showing that dissolution of RuO 2 during oxygen evolution is limiting the stability but that it is stable during hydrogen evolution. In the ESI † we demonstrate the stability of the RuO 2 catalysts for 2 h, at current densities 20 times higher than in the water splitting experiments.

Conclusions
Light-driven electrochemical water splitting has been demonstrated using a new wide band gap semiconductor blend, PTPTIBDT-OD:[70]PCBM, in a homo-tandem cell conguration. To achieve a high enough voltage the homo-tandem cell uses an intermediate recombination contact layer that consists of a thin layer of MoO 3 deposited on a conventional ZnO/pH neutral PEDOT:PSS junction to increase the work function at the hole extracting side of the stack.
For light-driven electrochemical water splitting, the PTPTIBDT-OD:[70]PCBM homo-tandem cell was combined with a RuO 2 catalyst for oxygen evolution and RuO 2 or Pt catalysts for hydrogen evolution in 1.0 M KOH. The h STH is 4.3% and the device operated almost at the maximum power point of the solar cell due to its remarkably good ll factor (0.73). As a consequence, the difference between the PCE of the solar cell and the h STH of water splitting is solely due to overpotential losses. Further improvements in h STH will rely on improved polymer semiconductors to increase PCE and better electrocatalysts, rather than improving the design or optimization of the light-driven electrochemical water splitting device.

Materials
All commercial chemicals were used as received. Ruthenium(III) chloride (35-40% Ru) was obtained from Acros Organics. The platinum plate was obtained from Drijout. PTPTIBDT-OD was synthesized as described in the ESI. † [70]PCBM was obtained from Solenne BV. Water used in depositions and electrolytes was puried in a Millipore system and has a resistance of at least 18 MU.

Electrochemical measurements
Cyclic voltammetry (CV) was conducted under an inert atmosphere using an Autolab PGSTAT30 with a three electrode setup equipped with an ITO working electrode covered with a spin coated polymer lm, a silver counter electrode, and a silver electrode coated with silver chloride (Ag/AgCl) as the quasi reference electrode in combination with ferrocene/ferrocenium (Fc/Fc + ) as the internal standard. A 1 M solution of tetrabutylammonium hexauorophosphate (TBAPF 6 ) in acetonitrile was used as the electrolyte. HOMO and LUMO levels were calculated from the oxidation and reduction potentials E ox/red vs. Fc/Fc + using E HOMO/LUMO ¼ À5.23 À qE ox/red eV.
The Tafel plot was constructed from the cyclic voltammetry (CV) measurements performed in the stationary current mode as described in detail in ref. 6.

RuO 2 catalysts from thermal decomposition
Ti substrates (thickness 0.5 mm, 1.5 cm Â 2 cm) were sonicated in acetone and then treated with air plasma in a Femto PCCE low pressure plasma system (Diener Electronic). A 270 W plasma was applied for 2 min. The RuO 2 catalyst was prepared via thermal decomposition of RuCl 3 . For this purpose an aqueous RuCl 3 solution (200 mL of a 0.2 M) was placed on a titanium substrate to form a catalyst area of about 1.3 cm 2 . The substrate was rst dried at 90 C on a hot plate for 20 min and then oxidized at 350 C in air in an oven for 3 h.

Device preparation
Photovoltaic devices were prepared rst by spin casting a 40 nm thick PEDOT:PSS (Clevios® P VP AI 4083, H. C. Starck) onto precleaned glass substrates with indium tin oxide (ITO) patterns (Naranjo Substrates). Prior to spin casting, PEDOT:PSS was ltered through an Acrodisc® LC25mm syringe lter with a 0.45 mm PVDF membrane. Then, the active layer was spin cast. Single junction devices were completed by thermally evaporating a back contact of 1 nm LiF and 100 nm Al at 3 Â 10 À7 mbar. For the tandem devices with a Naon layer, the intermediate contact consisting of ZnO nanoparticles ($30 nm), pH neutral PEDOT:PSS ($15 nm), and Naon ($4 nm) was deposited sequentially by spin casting. For devices with the MoO 3 layer, rst ZnO nanoparticles ($30 nm) and pH neutral PEDOT:PSS ($15 nm) were spin cast. Aerwards, a 10 nm thick MoO 3 lm was deposited by thermal evaporation. For optimal performance, ZnO nanoparticles were deposited inside a glovebox with a nitrogen environment, while all other layers were spin cast in air. No heat treatment was applied to any of the layers. Layer thicknesses were measured with a Veeco Dektak 150 surface proler.
For PTPTIBDT-OD:[70]PCBM homo-tandems, both active layers were spin cast from a solution of PTPTIBDT-OD and [70] PCBM (1 : 1.5 w/w) in chloroform containing 10 vol% o-DCB at 7 mg mL À1 polymer concentration. ZnO nanoparticles 34,35 of $5 nm diameter were spin cast from a solution of 10 mg mL À1 ZnO in isopropanol (IPA). The pH neutral PEDOT:PSS was prepared by diluting Orgacon Neutral pH PEDOT:PSS from Agfa with ultra-pure water at a 1 : 1 volume ratio and adding 200 mL L À1 IPA. The solution was then ltered with a 5.0 mm Whatman Puradisc FP30 syringe lter. Naon® (Aldrich chemistry, peruorinated ion exchange resin 5 wt% in a mixture of lower aliphatic alcohols and H 2 O, containing 45% water) was rst diluted in ethanol with a 1 : 200 volume ratio. The resulting solution was spin cast directly onto pH neutral PEDOT for an optimized thickness of 2-5 nm.

Characterization
A Keithley 2400 source-measurement unit was used to measure current density to voltage (J-V) characteristics of the devices. The illumination was carried out with $100 mW cm À2 white light from a tungsten-halogen lamp ltered by a Schott GG385 UV lter and a Hoya LB120 daylight lter. No mismatch correction was performed. The measurements were performed inside a glovebox with a nitrogen atmosphere. The tandem devices were exposed to UV illumination (with a Spectroline EN-160L/F 365 nm lamp from Spectronics Corporation) for about 10 min to provide an ohmic contact between the ZnO and pH neutral PEDOT layers before being measured. To prevent parasitical charge collection due to the high lateral conductivity of pH neutral PEDOT, tandem and triple junction devices were measured with a mask that is slightly smaller than the device area, which is determined by the overlap of the ITO and Al electrodes. The actual device area was 0.09 cm 2 , while the corresponding mask size was 0.0676 cm 2 . For single junction devices, more accurate short circuit currents under AM1.5 G were calculated by convolution of the EQE measurements with the AM1.5 G solar spectrum.
EQE measurements were performed in a homebuilt set-up. Mechanically modulated (SR 540, Stanford Research) monochromatic (Oriel Cornerstone 130) light from a 50 W tungsten halogen lamp (Osram 64610) was used as a probe light together with continuous bias light from a solid state laser (B&W Tek Inc., l ¼ 532 nm, 30 mW) through an aperture of 2 mm diameter. The intensity of the bias laser was adjusted using a variable neutral density lter. The response was recorded using a lock-in amplier (Stanford Research Systems SR830), over a resistance of 50 U. For all the single junction devices, the measurement was carried out under a representative illumination intensity (AM1.5 G equivalent, provided by the 532 nm laser). A calibrated silicon solar cell was used as the reference. The devices were kept behind a quartz window in a nitrogen lled box during the measurements.
Solar to hydrogen conversion efficiencies were determined using a home-built setup. As the water splitting experiments took place in air, the solar cells were kept behind a quartz window in a nitrogen lled box and connected to the catalysts through external cables. The solar cell was illuminated with white-light from a tungsten-halogen lamp ($100 mW cm À2 ) ltered by using a Schott GG385 UV lter and a Hoya HMC 80A 72 mm daylight lter. The solar cell was positioned by ensuring that the short-circuit current in this setup corresponds to AM1.5 G power standards. A Keithley 2600 source-measurement unit was used for the simultaneous measurement of current and voltage during water splitting.

Optical simulations
The optical modelling was done in SetFos version 3.2 (Fluxim AG, Switzerland). The optical constants for PTPTIBDT-OD: [70] PCBM were determined from reection/transmission measurements on active layers on quartz.