Ge-doped Hematite for an Unassisted Water Splitting System with Enhanced Efficiency


 To boost the photoelectrochemical water oxidation performance of a hematite photoanode, high temperature annealing has been widely applied to enhance crystallinity and remove the physical interface between the hematite and the fluorine doped thin oxide (FTO) substrate. However, the high temperature also results in unintentional Sn-doping due to thermal diffusion from the bottom FTO substrate. Therefore, when using additional dopants and the subsequent high temperature annealing process to enhance performance, the procedure should more precisely be considered co-doping of the hematite photoanode. However, at present, the interaction between the unintentional Sn and intentional dopant is poorly understood. Here, using germanium (Ge), which has been proven a promising dopant in previously reported simulations, we investigated how Sn diffusion affects overall PEC performance in Sn:Ge co-doped systems. After revealing the negative interaction of Sn and Ge dopants, we developed a facile Ge-doping method which suppresses Sn diffusion from the FTO substrate, significantly improving hematite performance. The Sn:Ge-hematite photoanode showed a photocurrent density of 4.6 mA cm− 2 at 1.23 VRHE with an excellent low turn-on voltage. After combining with a perovskite solar cell, our tandem system achieved outstanding 4.8% solar-to-hydrogen conversion efficiency (3.9 mA cm− 2 in an unassisted water splitting system). Our work provides important insights on a promising diagnostic tool for future co-doping system design.

Therefore, seeking alternative dopants may provide a more straightforward way of overcoming hematite's low conversion efficiency. Among various dopant candidates [33][34][35] , germanium (Ge) may be the most promising alternative as an n-type dopant 33 . Ge can dramatically enhance donor density while maintaining the crystallinity of hematite, leading to an outstanding solubility in hematite 33,36 . Prezhdo et al. reported density functional theory (DFT) calculation results showing that Ge was more soluble in hematite than Si and Sn due to the balance between atomic radius and formation enthalpy 33 . Further, Ge has a guiding effect on the preferential growth of the (110) plane of the hematite crystal, which promotes high electrical conductivity 36,37 . 3 Despite theoretical results that Ge could provide the superior photoelectrochemical properties compared with the current representative dopants in various respects, the highest water splitting performance for Ge-doped hematite reported so far is still far lower than those for representative dopants-doped hematite 18,23,29,36 . The strong discrepancy between calculated results and experimental data for doped hematite may be attributed to some variables that were neglected in the calculation. We hypothesized that unintentional Sn-doping from the fluorine doped tin oxide (FTO) substrate, which inevitably occurs during the high temperature annealing process (above 700 o C) used to improve the crystallinity of hematite, is one of such variables.
It has already been clearly proved that thermal diffusion of Sn from the FTO substrate indeed occurs, and is one of key factors which boosts PEC performance 18,38 .
Therefore, for a more realistic experimental approach, the presence of the Sn dopant and its impact need to be carefully considered in relation to the desired dopant.
Here we report that the water splitting performance of Ge-doped porous hematite (Ge-PH) can bring the experimental data more closely in line with the superior theoretical results of Ge-doped hematite by preventing the unintentional Sn-doping.
The approach produces a remarkable performance improvement compared to previous Ge-doped hematite (Ge-H), as well as hematite prepared with the commonly used representative dopants (Ti, Sn and Si).
We confirmed by both experiment and DFT calculation that when the Ge and Sn dopants were co-present, the crystallinity of the hematite significantly deteriorated due to structural distortion. We also proved that Ge-doping by the thermal diffusion of Ge in the GeO2 overlayer, reported in this study, mitigated the negative interaction 4 between the two dopants (Sn and Ge) and created numerous OER active sites, while maintaining the crystallinity of the hematite surface. More Importantly, we first report that Ge-PH can lower the overpotential of OER than pure hematite, using both theoretical simulations and experimental data.
By coupling a perovskite solar cell to the back of our photoanode, we achieved ca.
4.8% SHT efficiency for an unbiased tandem PEC water splitting system. Our Ge-PH effectively maximized the efficiency of unassisted water splitting, supported by a low turn-on voltage system with high performance.
To the best of our knowledge, this work demonstrates the highest STH efficiency for a single hematite photoanode-based tandem device, which may be a stepping-stone for a breakthrough in stagnant hematite-based PEC performance.

5
Fabrication process and morphology effect.

-Fabrication process of H, Ge-H, and Ge-PH
Pristine Fe2O3 (H) and Ge-doped Fe2O3 (Ge-H) photoanodes were fabricated using conventional methods as reported (Fig. 1a) 36,37,39 . Briefly, FeOOH nanorods were grown on an FTO substrate using the common hydrothermal method and then rapidly annealed at 800 o C for 20 min to form H (top in Fig. 1a). Ge-H (bottom in Fig. 1a) refers to bulk Ge-doped Fe2O3 hydrothermally grown in a mixture solution of FeCl3 and GeO2 followed by a rapid annealing step at 800 o C for 20 min as reported previously 5,29 . To fabricate the Ge-PH (middle in Fig. 1a), as-fabricated FeOOH nanorods were immersed in a Ge solution for 30 min and rapidly annealed. The Ge solution for doping was made by dissolving GeO2 powers in deionized water. Since all of the samples were subjected to the high-temperature annealing step (800 o C for 20 m), which creates Sn-doped Fe2O3, we deliberately omit mentioning the Sn for simplicity in this study. ii) The GeO2 coated FeOOH nanorod undergoes insitu conversion into Ge-PH by subsequent high-temperature annealing. In this process, the Kirkendall effect, the motion of the interface between two materials due to different diffusion rates of each atom, is partly involved in the creation of pores, as reported by Gong's group 40 , as well as the mass evaporation of water in the hard template of the GeO2/FeOOH. This high-temperature dehydration creates Ge-PH with mesopores inside, via the previously reported gas entrapping mechanism 29 .

-Morphology of H, Ge-H, and Ge-PH
As shown in Figs 1e-i, the energy dispersive X-ray (EDX) mapping of Fe, O, Ge and Sn elements by STEM analysis shows spatially uniform distribution and the porous morphology of the Ge-PH.
Ge-PH with a nanoporous structure has two main advantages over H or Ge-H. First, the path distance for the generated holes to travel from inside to the surface of the hematite, where oxygen generation occurs, is shortened (10-15 nm), which helps address the critical issues of the short hole-diffusion length of hematite, as shown in Fig. 1d. Second, the occurrence of pores in the Ge-PH increases the number of 8 reaction sites for oxygen evolution, simply by increasing the surface area. As shown in the BET data for the surface area and pore distributions ( Supplementary Fig. 4), Ge-PH exhibited five-fold (10 m 3 /g) increased surface area compared to Fe2O3 or Ge-H (~2m 3 /g) with a mesopore morphology. Beside the structural differences between Ge-H and Ge-PH, the Ge in Ge-PH was doped in the final step by the thermal diffusion of Ge from the surface, whereas the Ge was uniformly doped in Ge-H at the beginning step, during the process of forming the FeOOH state.  Ge-H (~1.9 mA cm -2 at 1.23VRHE) photoanodes, respectively.

PEC water oxidation activity and characterization
To determine whether this remarkable improvement was simply due to the hematite porosity, we fabricated Fe2O3 with a similar porous morphology using other currently representative dopants (Sn, Ti, or Si). The results clearly showed that Ge was superior to Sn, Ti or Si dopants for hematite, as shown in Supplementary Fig. 5.
The reason can be explained by the advantage of Ge as a dopant in hematite, including the feasible atomic radius of Ge, and the low formation enthalpy of the secondary phase of GeO2 as previously reported 27 .
In particular, Ge-PH showed high performance at low voltage without an anodic shift in the onset potential, despite the doping. It has been reported that, in a typical doping system, the increase in defect sites produced by doping can enhance carrier mobility, and carrier density in the bulk, while simultaneously providing recombination sites on the surface, resulting in an anodic shift of the onset potential 41 . Furthermore, the Fe 2+ 10 formed by n-doping in hematite is also known to act as a recombination site on the surface, which consequently retards the water oxidation reaction in doped hematite. This is in accordance with the anodic shift of the onset potential for all doped hematite photoanodes (Sn, Ti, Si-doped hematite), including the Ge-H in this study, compared to that for H. Therefore, one of the critical issues has been to optimize the two components (onset potential vs current density), which operate in an opposite manner at 1.23VRHE.
Thus, the result here that the Ge-doping in Ge-PH did show a cathodic shift of the onset potential is certainly interesting ( Supplementary Fig. 6), and strongly suggests that there could be an important factor in our Ge-PH (Surface Ge-doped hematite).
To explore this phenomenon more systematically, we carried out various scientific analyses of Ge-H and Ge-PH. First, the XRD patterns showed similar hematite peaks to H without the new phase formation in Ge-H and Ge-PH, as shown in Fig. 2b. In the Raman spectra, the appearance of the forbidden longitudinal optical (LO) mode, corresponding to the peak at 660 cm -1 , is indicative of the symmetry breakdown induced by structural disorder, scattered LO phonons 4 . The LO peak was largely increased and broadened in the Ge-H, compared to H, whereas the much reduced LO peak was observed in Ge-PH, as shown in Fig. 2c. This implies the symmetry breaking by Ge-doping in Ge-H is much larger than in Ge-PH.
A correlation between the Ge and Sn dopants was confirmed by X-ray photoelectron spectroscopy (XPS) data. The observation of a Ge 3d peak at 31.6 eV from Ge-H and Ge-PH indicates that the Ge atoms were successfully doped in Fe2O3 by the high temperature annealing process, as shown in Fig. 2d 37,39 . We hypothesized that these probably suggest that Sn has a greater influence on structural distortion, due to the larger atomic size than Fe and the excess charge coming from the n-type dopants 42 .
To clearly pinpoint these assumptions, the chemical compositions of H, Ge-H and Ge-PH were examined by XPS depth profile. The results showed that the Sn-doping ratio of H and Ge-H were 4.5-9.5% in the whole region. However, the doping ratio of Sn in Ge-PH was much reduced, with maximum 0.7-0.8% in the whole region, as shown in Fig. 2g. This suggests that Sn-doping by thermal diffusion from the FTO substrate was greatly suppressed by the GeO2 overlayer in the long and thin nanorods compared to the short and thick nanorods without the overlayer. Since Ge-PH has an unfavourable and long Sn diffusion path from the bottom FTO substrate, it has less Sn content on the surface of the hematite where the OER reaction occur, resulting in less chances for Ge:Sn combination, as described in Fig. 3e.
The Ge-doping ratios of Ge-H and Ge-PH were measured to be 3.4-5.5% and 7.7-13.8% in the whole structure region, as shown in Fig. 2h. Although the total doping content (Sn+Ge) of Ge-H and Ge-PH was similar (around 8-14%, Supplementary Fig.   8), the doping ratio (Sn/Ge) of Ge-H was 14-19 times higher than in Ge-PH. Therefore, we can conclude that Ge and Sn will have a negative influence, causing structural distortion (as proven in Figs. 2c and 2f) when they are in close contact, which may be due to the negative repulsion between the two n-dopants of Sn and Ge.

-Negative interaction of Sn and Ge
To understand the interaction between the Ge and Sn dopants in Fe2O3, we calculated the formation energy for Ge-doped Fe2O3 and Ge:Sn co-doped Fe2O3, as shown in Fig. 3a. The results revealed that the formation energy for Ge:Sn co-doping (blue) was higher than that for Ge single doping (red). The high formation energy of Ge:Sn codoping in Fe2O3 indicates the low dopant solubility and low ionization in hematite 33 .
We explored the negative interaction caused by the co-presence of Ge and Sn by comparing the atomic structures of Ge-doped hematite and Ge:Sn co-doped hematite using DFT calculations. As can be seen in Fig. 3b, the Ge:Sn co-doped hematite experiences greater symmetry breaking after the re-positioning of the Fe atoms, while the substitutional single Ge-doping did not produce any noticeable distortion in the atomic arrangement.
Based on DFT calculations, we drew the atomic arrangement of hematite with the substitution of heteroatoms to clearly understand this phenomenon, as shown in Fig.   3c. Single-Ge-doped hematite did not show much distortion since the Ge dopant has a radius similar to Fe in hematite, and Ge becomes more soluble than other representative metal dopants. In the case of Sn-dopant hematite, relatively high structural distortion occurs since the Sn dopant has a larger radius than Fe. When Ge and Sn are co-present, additional strong electron repulsion between Fe atoms neighboring the Ge and Sn dopants is produced by the excess electron charges from the n-type metal dopants in Fe2O3.
-Observation of the structural disorder 15 Fig. 3d is the XPS spectra of Fe 2p, which shows that Ge-H has more Fe 2+ than Ge-PH. This is because that Ge-H has more than 6 times the amount of Sn than Ge-PH, as proven in Fig. 2g. Ge-H has a large amount of diffused Sn, where Ge is present throughout the hematite nanorod. In Ge-PH, however, the majority of the Ge and Sn dopants are positioned in different regions, and the amount of diffused Sn is relatively small in the surface region where the OER reaction occurs specifically, thus minimizing the adverse effect caused by the co-existence of the two n-dopants, as shown in Fig.   3e. Therefore, due to the lower content of Sn, Ge-PH was expected to experience a lesser distortion than Ge-PH.
These results well explain the XRD, Raman, EXAFS, XPS spectra and the PEC activity upon front or back illumination observed in Fig. 2 and Supplementary Fig. 9, which show that the structural distortion observed in Ge-H caused by co-doping of Sn and Ge in hematite was almost recovered in Ge-PH, which had a status similar to the original undoped hematite.
Recombination rate and surface activity. The negative effects of co-doping with the two n-type elements found in this study could also be confirmed by electrochemical analysis. The highest charge carrier density of Ge-PH, which is inversely proportional to the lowest slope of the curve in the Mott-Schottky plot (Fig. 4a), was in a good agreement with the simulation and experimental results. To support the extraordinarily reduced overpotential and excellent performance in Ge-PH observed in our experimental result, DFT calculations were performed to determine the theoretical overpotential. Fig. 4c shows the calculated free energy for each elementary step. It is known that the rate determining step for hematite is the reaction B (*OH → *O) where the deprotonation from *OH can make the charge state (*O) very unstable [43][44][45] . In undoped hematite, therefore, the reaction B corresponding to deprotonation from *OH has the highest free energy in the reaction pathway and 18 the reaction potential was determined to be 2.2372 eV. The calculated overpotential for undoped hematite is 1.007 eV, which is in reasonable agreement with previous theoretical studies for (0001) hematite 43,44 .
To lower the free energy for the reaction B, it is necessary to reduce the instability of *O. When Ge is doped in hematite, the charge state of *O can be more stable since an n-type dopant Ge provides the electron to oxygen 33,35,43 . Therefore, the free energy of the reaction B is significantly reduced by Ge doping. One the other hand, due to a trade-off relationship of the free energy between the reaction B (*OH → *O) and the reaction C (*O → *OOH) 43 , the rate determining step of Ge doped hematite is the reaction C, which has a 0.119 eV lower overpotential (0.888 eV) than undoped hematite, which is in consistent with our experimental J-V curve.
Charge separation efficiency was calculated based on the LSV curves under illumination in 1M NaOH and 1M NaOH containing 0.5 M hole scavenger, Na2SO3 as shown in Fig. 4d. Notably, Ge-PH showed a substantially higher charge transfer efficiency than H and Ge-H over the entire tested potential range, and approached 80% at potentials beyond 1.3VRHE as shown in Fig. 4e. The results of the electrochemical analysis and DFT calculations clearly support the reason for the low onset potential of Ge-PH. In order to confirm the feasibility of our photoanode for unbiased solar water splitting, we evaluated the performance of Ge-PH in a tandem configuration. We prepared a tandem device containing a single PSC and a hematite photoanode similar to the Zscheme in natural photosynthesis, in which two semiconductors with different absorption spectra are efficient over a broad part of the solar spectrum, and deliver a high STH efficiency for water splitting.
For this setup, we employed a perovskite solar cell (PSC) fabricated using a recently developed procedure (short-circuit current (Jsc)=21.60 mA cm -2 , open-circuit voltage (Voc)=1.16V, and fill factor (FF)=75.07%; power conversion efficiency (PCE)=18.85%, Supplementary Fig. 10) 46 . The PSC is unable to drive the reaction on its own (or with an efficient electrocatalyst) because its photovoltage is less than what is thermodynamically required to split water 47,48 .
A schematic of the tandem configuration, with the PSC connected electrically and optically in series with the hematite is shown in Fig. 5a. The NiFeOx OER catalyst was deposited on a Ge-PH photoanode, which further helps shift the onset potential with enhanced performance. When NiFeOx was applied to Ge-PH, the photocurrent density of the NiFeOx/Ge-PH reached 4.6 mA cm -2 at 1.23 VRHE as shown in Fig. 5b.
To estimate the operating current density, J-V curves of the PSC were measured by 21 placing the hematite photoanode before the solar cell to account for optical absorption by the hematite photoanode as shown in Fig. 5c. The operating current density in the tandem configuration was thus estimated to be around 3.9 mA cm -2 . The assembled tandem device was subsequently tested in 1M NaOH electrolyte without additional external bias in a two-electrode configuration, using the J-T curve under 1 SUN (AM 1.5G, 100mW cm -2 ). The current density closely matched the operating current extracted from Fig. 5c, with excellent stability, as shown in Fig. 5d and Supplementary   Fig. 11. The STH conversion efficiency was calculated to be 4.8% for the Ge-PH and PSC tandem system. To the best of our knowledge, this is the highest STH efficiency obtained for a single hematite-based photoanode with a tandem device, as shown in Supplementary Tables 1 and 2. Finally, we calculated the faradaic efficiency of the tandem device by measuring the H2 and O2 evolution under AM 1.5 illumination in 1 M NaOH electrolyte. As shown in Fig. 5e, the hydrogen gases produced on the Pt mesh and the oxygen gases on NiFeOx/Ge-PH were around 68.5 µmol and 34.0 µmol after 120 min, respectively, indicating a 2:1 ratio of the water splitting mechanism. The ratio between the measured and predicted gas evolution rates gives a faradaic efficiency of 87-95% throughout the measurements. Therefore, most of the photo-generated charges were consumed by water splitting (hydrogen/oxygen gas generation) in our tandem system.

Conclusion
In summary, we present a novel approach to achieve the theoretically potential results 22 in a water splitting system of co-doped hematite. We demonstrated that the morphology controlled Ge-doped hematite with the reduced content of unintentionally doped Sn, taking into account the negative interaction between guest Ge and predoped Sn, can be a stepping-stone to approach hematite's theoretical efficiency, including the high photocurrent density and the low turn-on voltage. Employing our findings and outstanding performance, an unassisted water splitting system delivered the excellent photocurrent density of ~ 3.9 mA cm -2 in 1 M NaOH electrolyte. Our insight and co-doping strategy for water splitting using hematite potentially provides a new paradigm for electrode design, and could be further extended to other heteroatoms-dopant systems utilized in numerous applications including solar conversion, sensing, and opto-ferroelectric device. The hole-conducting material was spin-coated at 3000 rpm for 30 s on the perovskite/mp-TiO2/bl-TiO2/FTO. Finally, a gold layer was deposited on the hole conducting layer using a thermal evaporator.

Preparation of the H (Fe2O3) and
DFT calculation details. All calculations were performed in the framework of the spinpolarized density functional theory with the projector augmented wave (PAW) method 50 using the Vienna ab-initio simulation package (VASP) code 51 . The exchange-correlation was considered using the generalized gradient approximation of Perdew, Burke and Ernzerhof (PBE) 52 . The cut-off energy for the plane wave basis set 27 was 500eV, and Monkhorst-Pack k-point mesh of 4x4x1 was used for all the slab structure of α-Fe2O3 (hematite). The ionic positions were relaxed until a force convergence of 0.01 eV A -1 was reached. Because of the strongly correlated 3d states in transition metal oxide systems, we used the GGA+U framework to modify the selfinteraction 53 . The values of U-J of all the 3d metals were set to 4.2 eV for good agreement with the experimental band-gap of α-Fe2O3 (2.2eV). The hexagonal unit cell of α-Fe2O3 was optimized with a layered anti-ferromagnetic (AFM) ordering. For pure α-Fe2O3 unit cell, the lattice parameters calculated within PBE+U were a=b=5.07 Å and c=13.88Å, and they were consistent with the experimental values of a=5.04Å and c=13.75Å 54 . Each fully-relaxed bulk structure of pristine and Ge doped α-Fe2O3 was used to determine the lattice parameters of each (1 x 1) slab structure. A vacuum layer at least 12Å was used to minimize the interaction between the periodic surface along z axis. We focused on the surface reaction on (0001) α-Fe2O3 surface because it is one of the natural growth faces of hematite 55 . Dopant substitutions were made at both outermost Fe layers to consider the maximum doping effect on surface reactions and to remove the polarization from broken symmetry 43 . Hydrogen passivation was used to prevent the transfer of hydrogen atoms from the active site to the other surface oxygen. We passivated only one of the three surface oxygen atoms to minimize the hydrogen bonding that affects the reaction.
We considered the following OER mechanism with four elementary steps 56 .
The * represents chemisorption with the reactive sites on the surface. According to Rossemiesl et al. 56 , at standard conditions(pH=0, p=1bar, T=298K), the reaction free energy(△G) of each step is calculated as follows: is the external potential. At the standard condition with Φ=0, the highest free energy ( ∆G max ) is equal to reaction potential for electrochemical reaction potential and (∆G max − 1.23) is equal to overpotential (η).

Data availability.
The data that support the findings of this study will be made available upon request.