Atomic layer deposition triggered Fe-In-S cluster and gradient energy band in ZnInS photoanode for improved oxygen evolution reaction

Vast bulk recombination of photo-generated carriers and sluggish surface oxygen evolution reaction (OER) kinetics severely hinder the development of photoelectrochemical water splitting. Herein, through constructing a vertically ordered ZnInS nanosheet array with an interior gradient energy band as photoanode, the bulk recombination of photogenerated carriers decreases greatly. We use the atomic layer deposition technology to introduce Fe-In-S clusters into the surface of photoanode. First-principles calculations and comprehensive characterizations indicate that these clusters effectively lower the electrochemical reaction barrier on the photoanode surface and promote the surface OER reaction kinetics through precisely affecting the second and third steps (forming processes of O* and OOH*) of the four-electron reaction. As a result, the optimal photoanode exhibits the high performance with a significantly enhanced photocurrent of 5.35 mA cm−2 at 1.23 VRHE and onset potential of 0.09 VRHE. Present results demonstrate a robust platform for controllable surface modification, nanofabrication, and carrier transport.

1. The authors claim that the Fe-In-S atomical structure formed, which is also the absolute proof for DFT calculation. However, this result is questionable, especially with ALD flow for generating Fe2O3. Besides, the peak position of XANES spectra reveals the Fe atoms are more likely ascribed to be those in Fe2O3. The peaks in Figure 2f assigned to Fe-S and Fe-In are also questionable.

[Author Reply]:
Many thanks for your kind support and efforts in reviewing our manuscript.
According to the data in this manuscript, there is no Fe2O3 on the surface of samples. Some detailed explanations and the further description of XANES spectra are provided: 1) The peak position of XANES spectra does not reveal the Fe atoms ascribed to be those in Fe2O3. In the area enclosed by the red dotted line in Fig. R1a, unlike Fe2O3 standard sample, there is no characteristic signal at the energy range between 7140-7165 eV of our samples.
Again, this can be used to confirm the difference in atomic structure around Fe ions in our samples compared with Fe2O3 standard sample.
2) The peaks in Fig. 2f assigned to Fe-S and Fe-In are reasonable. Due to the larger atomic radial distances of S and in comparison to those of O and Fe in Fe2O3, thus, the peak positions in R-space of Fe-S and Fe-In are located at the longer R distances (see the Fe K-edge r-space of Fe2O3 in Fig. R1b). These results provide strong evidences about the existence of Fe-In-S in our samples.
3) There is no Fe2O3 on the surface of the photoelectrode. Although the ALD deposition procedure involves ferrocene and ozone that used as iron and oxygen precursors, which are more likely to generate Fe2O3. Under normal condition, the ALD flow is used for generating Fe2O3, especially on various metal oxide materials. However, there are some difference in this manuscript when the substrate material is ZnInS. There is a new mechanism for this special case in our results. The relevant proof data are as follows: Firstly, the X-ray absorption spectroscopy (XAS) of the Fe L3 absorption edges (Fig. R1a-c) indicates there is no oxides formed. Fig. R1a and b shows the different characteristics of curves from standard FeO and Fe2O3 samples, manifesting that no oxides are formed. In addition, the peaks at 2.1 Å and 3.5 Å are assigned to Fe-S bond and Fe-In bond (Fig. R1c), respectively, and the content of Fe-S bond increases with the increased deposition cycles. These characterizations indicate that the Fe-In-S atomical structure is formed at the surface of photoanode and there is no Fe2O3.
Secondly, in order to confirm that Fe2O3 is not produced through the ALD deposition process based on ZnInS, the X-ray diffraction (XRD) pattern of ZISZ/Fe with 700 cycles is shown in Thirdly, to confirm that there exists no Fe2O3 on the surface of photoanode, the XPS of samples treated under different ALD cycles is shown in Supplementary Fig. 6. With the increased cycles, the peak intensity of In gradually decreases while that of Fe increases. The reduced intensity of In results from the replacement of Fe to build Fe-In-S group, indicating there is no Fe2O3 generation.
Finally, we also measured the XPS at different depths through Ar ion etching ( Supplementary   Fig. 8). As the etching depth increases, the peak intensity of Zn gradually increases and its peak position shifts toward higher energy ( Supplementary Fig. 8a). The peak intensity of In also increases while the peak position shifts toward lower energy ( Supplementary Fig. 8b). When the etching depth increases to 50 nm, the signal of Fe substantially disappears ( Supplementary   Fig. 8c), but the peak position of O shifts toward higher energy ( Supplementary Fig. 8d).
According to these data, we can get the following conclusions: 1) Fe is mainly deposited on the surface with a thickness of about 50 nm. 2) The shift of Zn peak toward higher energy and O peak toward lower energy are due to the formed Zn-O. The Zn-O bond is shorter than Zn-S bond, thus the binding energy of Zn-O is higher. 3) The opposite trend of the peak intensity of In and Fe also indicates the successful substitution. In a word, the lattice O appears inside the samples and Fe exists on the surface, respectively. Thus it is impossible that Fe2O3 can be formed on the surface of the samples.
In summary, through the above characterizations, we conclude that the Fe2O3 can not form during the ALD flow and there is no Fe2O3 in the samples. Unlike the normal ALD deposition process, there is a new mechanism for the present material system.  Figure S7 and S15 are unreasonable. The surface structural stability of the model in Figure S15 should be tested.

The models for DFT calculations in
[Author Reply]: According to your suggestion, we have added the equivalent circuit in the manuscript.
We thank the Reviewer for raising this issue. The ZnIn2S4 (ZIS) surface is represented by a ZIS (001) slab in our simulations, which was created using the ZnIn2S4 unit cell 1 widely used by previous DFT calculations 2 . The introduction of an 8-atom ZnS cluster on the ZIS (001) surface forms the ZISZ system, referring to the WO3-x/ZnIn2S4 system 3 . The replacement of an In atom or a Zn atom with an Fe atom in the ZISZ system leads to the ZISZ/Fe system. The substitution is energetically favorable because the calculated formation energy is negative. To further examine the stability of four systems, we carried out adiabatic molecular dynamic simulations at room temperature lasting for 8 picoseconds and plotted the evolution of total energy of each system in Supplementary Fig. 19  [Author Reply]: Many thanks for your suggestions in reviewing our manuscript. Both the articles you provided explain the Von in detail. According to these articles, we have made some changes and provided the original OCP data in this manuscript.
Refer to these works, we have redefined the potential of J=0.02 mA cm -2 as Von (defined as the potential at which 0.02 mA cm -2 was first measured). In addition, the influencing factors of 4. The self-oxidation decomposition potential of ZnInS is more negative than the OER potential, and there are many possibilities to exhibit photoanodic current. For example, the photo-generated holes cause the self-decomposition of photoanodes. The authors haven't excluded the self-oxidation of photoanodes in 0.5 M Na2SO4.

[Author Reply]:
In this manuscript, we also excluded the photoanode in Na2S/Na2SO3.
However, this electrolyte is used as hole scavengers, therefore we did not use it as the main test system. The reasons why we select 0.5 M Na2SO4 as electrolyte are as follows: 1) The main innovation in this paper is the construction of cocatalysts, promoting the rapid transfer of photogenerated holes. If other electrolytes are used for measuring the photoanode, such as 0.5 M Na2SO3 or Na2S/Na2SO3, etc., that is not conducive to this experiment to characterize the role of cocatalysts since these electrolytes are usually used as hole scavengers.
2) For most of articles, 0.5 M Na2SO3 or Na2S/Na2SO3 are used as hole scavengers to obtain the injection efficiency, proving the effectiveness of cocatalysts. Under normal conditions, it is not recommended to directly apply this electrolyte as a test system.
3) Regarding the self-oxidation process you mentioned, if the process exists in this electrolyte, the dark current should have an obvious oxidation peak, but the dark current is obviously normal (Fig. R2). In addition, many previous results were also based on this electrolyte for photoelectrochemical (PEC) testing as shown in Table R1.   In this manuscript, IMPS was performed to study the charge transfer and transit dynamics processes, including surface charge-transfer efficiency (ηtran), surface charge-transfer rate constants (Ktran), and surface charge recombination rate constants (Krec). The relationship between them can be described by the following formula: the calculation details as shown in Fig. R3, thus through two formula we can obtain the ηtran, Ktran, and Krec.
On the other hand, as you mentioned, the frequency at the minimum imaginary part of ZISZ is higher than ZISZ/Fe, which means the electron transfer time of ZISZ/Fe is longer than ZISZ.
This result can be interpreted by the following formula: τd is electron transfer time, fmin is the frequency of the lowest point of the imaginary part in IMPS, which is shown in Fig. R3.
The longer τd is due to the less interface recombination. The increased charge-transfer rate and the reduced charge recombination rate of holes cause the reduced interface recombination.
Thus the longer τd of ZISZ/Fe is not converse to the conclusion of that the surface chargetransfer efficiency for ZISZ/Fe has a huge improvement compared with ZIS and ZISZ. In the part of photoanode, the surface OER process is mainly dominated by photogenerated holes. The

Reviewer #2 (Remarks to the Author):
The work has reported an efficient ZnInS based photoanode with the modification of Fe-In-S group. It's encouraging to see that the Fe-In-S group has dramatically promoted the surface reaction performance of the ZnInS based photoelectrode. The manuscript is well written and should be able to attract the broad interest of readers in the relevant fields. For the authors to further improve the manuscript quality, I would suggest the authors to consider and clarify the following issues, as listed below: (1) The photoelectrodes exhibit high photocurrent dentistry, so what's the key evidence to confirm the current is generated by water oxidation process? The first principle investigation of the surface OER process on the surface may not represent the case in the experimental research.
Moreover, the stability investigation is also very important for claiming the efficiency of the prepared photoelectrode. Note that the photocorrosion can also produce high photocurrent, so the authors are encouraged to further check the key mechanism.
[Author Reply]: According to your suggestion, some further explanation has been added.
If the oxidation process occurs during the material rather than the water, this process should still exist without illumination. Since the theoretical water splitting voltage is above 1.23 V, the advantage of PEC water splitting is that solar energy can provide additional energy to achieve photoelectricity complementation. Therefore, water could be splitted under a low voltage (less than 1.23 V). Unless the measured current value is not from the oxidation of water, but from the oxidation reaction of the material itself, in the absence of light for supplementation, it is theoretically impossible to split water under low voltages, and there will be no large current value. Similarly, for the photoanode, if the photocurrent generated in PEC water splitting does not come from the oxidation process of water, then there should still be a higher dark current or oxidation peak in the dark state. According to the discussions above, if the current value obtained does not come from the oxidation process of water, a relatively high dark current or redox peak will appear in J-V curves without illumination. However, in this work, the LSV measurement without illumination in Fig. R4a does not show the abnormal curve, and the oxidation peak does not appear. Thus, the photoelectrode with a high J is generated by water oxidation process under illumination rather than the self-oxidation of photoanodes.
In addition, we conducted a stability test in Fig. R4b, which shows the stability has been improved to certain extent after loading Fe-In-S cocatalyst, but there are still some gaps compared with some metal oxides. The explanation about the stability of ZnInS probably has the following points: 1) Metal sulfides are inherently unstable. During the test, the material acts as photoanode and generates a large number of photo-generated electron-hole pairs under illumination. Owing to its poor surface OER reaction kinetics of photoanode, a large number of photo-generated holes accumulate on the surface of the photoanode, causing serious self-corrosion. According to IMPS data, we can find that the interface transfer constant has been greatly improved after loading Fe-In-S, which is the reason why the stability of ZISZ/Fe has been improved.
2) Another important reason about the stability of ZnInS is that the elements are unstable in the test process. We carried out the ICP test (Table R2) on the tested electrolyte and found that it contains a lot of Zn, In, and S elements, especially Zn and S elements, which shows that the photoanode instability is due to the dissolution of its own elements in the process of stability testing. Currently, there is no good solution for the stability of ZnInS and this is not the core of this article. Some works focus on the stability design such as adding some metal ions to the electrolyte to inhibit element dissolution, or loading a barrier layer of metal oxide. However, these schemes are all ready-made and can not reflect the innovation of this work. Thus, the stability design part is only to verify the Fe-In-S clusters to promote the transfer of photogenerated holes by detecting the improvement of stability.

[Author Reply]:
Many thanks for your kind support and efforts in reviewing our manuscript.
According to this article, ABPE does come from a two-electrode system, but according to the literature on PEC published in recent years, the ABPE should be measured via a threeelectrode system. The advantages of this are: 1) The advantage of PEC lies in the complementarity of optoelectronics. The decomposition voltage of water is 1.23 V vs. RHE. Therefore, when comparing the performance of PEC, the comparison is often performed at 1.23 V vs. RHE to highlight the advantages of optoelectronics.
However, the pH provided by different electrolytes is inconsistent. For unified comparison, we need to use a three-electrode system for unified comparison. The comparison of ABPE should be standardized. As a relatively uniform variable, using the three-electrode system to RHE to unify the standard to RHE is conducive to direct comparison between different electrolytes.
And in the three-electrode system, the bias voltage provided by the system also exists as an external bias voltage.
2) The ABPE calculation process via a three-electrode system is also utilized by some J refers to the photocurrent density (mA cm -2 ) produced by the photoelectrode, whereas Plight is the power density obtained at a specific wavelength (λ), respectively.
ηabs. is the efficiency of the light harvested, and it can be calculated from the obtained light absorbance curves: where the A is light absorbance measured by a UV-vis spectroscopy. The capability of light harvesting gradually increases with the decreased bandgap for ZIS, ZISZ, and ZISZ/Fe (Fig. 3b). The incident photo-to-current efficiency (IPCE) and absorbed photon-to-current efficiency (APCE) show the increased IPCE and APCE value over the whole wavelength range (Supplementary Fig. 13), indicating the outstanding PEC performance of ZISZ/Fe. The IPCE and APCE is in accordance with the variation of photocurrents shown in Fig. 3a. This superior performance confirms the excellent light absorption, ηsep, and injection efficiency (ηinj), which we will discuss as follows.
However, theoretical photocurrent density (Jabs.) is a unified standard for judging the light absorption of the ZIS, ZISZ, and ZISZ/Fe photoanodes. The unity converted photocurrent density (Jabs.) is calculated by the following formula: where λmax is the maximum light absorption edge of a photoelectrode, λ (nm) is the light wavelength, E (λ) is the power density (mW cm −2 ) at a specific wavelength (λ) of the standard solar spectrum., and ηabs. is the efficiency of the light harvested, and it can be calculated from the obtained light absorbance curves: where the A is light absorbance measured by a UV-vis spectroscopy.
According to these formula, we can calculate the theoretical Jabs. of the three electrodes, and the value is 4.47, 6.07 and 6.54 mA cm -2 , respectively.
In summary, we claimed "the improved performance is attributed to the increased light absorption, ηsep, and injection efficiency (ηinj)." 2) In the XPS measurements, the detection depth is less than 10 nm. In order to detect the composition of surface deeper than 10 nm, the Ar ions etching process is used to peel off embedded materials generally. In other words, since most of the samples are in the atmospheric environment before XPS analysis, it is easy to cause adsorption pollution. For measuring the true information of sample surface, the sample surface must be pre-cleaned. The commonly used surface cleaning technology is argon ion etching technology.
3) If Ar ions etching changes the surface states due to the reductive nature of Ar, the chemical states will be reduced. However, according the Supplementary Figure 8 (5) As for the Fe-In-S, the authors called it a "group". Is this more like a "cluster", but not a group with well-identified composition and stoichiometry. In addition, they authors should justify why the surface catalyst is Fe-In-S based materials, but not Fe-S based ones? The latter may be more likely for the ALD deposition process.

[Author Reply]:
According to your suggestion, we have called the Fe-In-S as a "cluster".
In addition, we justify the surface catalyst is Fe-In-S based materials, the reason as follows: 1) As you considered, the ALD deposition process is more likely for growing FeS, but the surface catalyst is Fe-In-S based materials rather than FeS. Firstly, the process of replacing In with Fe occurs on the surface. Secondly, we did not supplement the S source, so this deposition process is not an experiment for preparing FeS. Furthermore, when the replacement reaction of Fe occurs, the Fe-In-S exists as a whole in this area, it is not possible to isolate Fe-S alone.
Therefore, we believe that Fe-In-S has a good catalytic effect on the surface.
2) According to the XPS data, we can know that when Fe enters the ZIS material, the reaction of Fe replaces In occurs. Furthermore, according to the DFT theoretical calculation model, we find that substitution of In with Fe requires smaller formation energy than replacement of Zn with Fe, which suggests that the former substitution is energetically favorable. Furthermore, the replaced Fe atom is distant from the Zn atom and facilitates to form Fe-In-S bond between the Fe dopant and its surrounding In and S atoms. Therefore, the Fe-In-S formed in this area as a whole catalyzes the oxidation of water.
3) The most important evidence is that we have performed XAS fine spectrum analysis on the sample and found that it is not consistent with the standard spectrum of FeS (Fig. R5). It proves that there is no FeS on the surface, therefore, the surface catalyst is Fe-In-S based materials. Furthermore, according to the Fe K-edge extended X-ray absorption fine structure (EXAFS) as functions k2χ(k) and its Fourier-transformation (FT-EXAFS) (FT-κ2χ(κ)) for ZISZ/Fe (Fig. 2f), we can find that Fe-S and Fe-In appear on the surface of the sample, so we think that it cannot be simply named after FeS alone. According to the dark current provided in this manuscript, we can find that Fe-In-S cluster exists as a good photo-cocatalyst. According to the J-V curves, the Fe-In-S cluster does not generate a significant current in the dark state (Fig. R6), but provides a significant performance improvement in the light state (Fig. 3a).
We know that electrocatalysts can reduce the water oxidation reaction barrier in the process of water splitting, but not all electrocatalysts can be used to build cocatalysts in PEC water splitting. It is necessary to ensure that the dark current is constant for constructing the cocatalysts, so that the impact of electrocatalytic water splitting can be avoided.

Reviewer #1 (Remarks to the Author):
The authors have answered most of the questions that I concern. Therefore, I consider the manuscript in this version can be accepted.

Reviewer #2 (Remarks to the Author):
The revised manuscript has carefully considered all the comments of the reviewers and made further improvement. I would suggest the acceptance and only have one minor comment on the the IPCE and APCE data of the ZISZ/Fe photoanodes in the supplementary Figure S13, why the data show some drop in the wavelength range of 360 nm.

Reviewer #3 (Remarks to the Author):
This authors have synthesized ordered nanosheet arrays ZISZ/Fe photoanode material by ALD, which shows reasonably high OER activity. DFT calculation was used to study the OER mechanism to understand and verify the experimental results. However, I find some arguments are rather vague, and conclusions are not fully supported their results, and suggest that the authors carefully consider the following issues: 1. Hexagonal ZnIn2S4 is a layered material with the thickness of each layer about 10Å, and the layers are connected by van der Waals interaction. When constructing the ZIS model for DFT calculation, why is the one-layer model used instead of the multi-layer model? 2. The DFT model of ZISZ is stoichiometric, but is not without question. This model should be validated so as to be representative of the experimentally synthesized materials. 3. In the DFT model of ZISZ/Fe, how to determine the position of the replaced In atom when replacing In with Fe, and whether is the replacement of In in the middle layer considered? The authors have mentioned that the experimentally synthesized ZISZ/Fe has Zn-O bonds at the bottom, but the role of O is not clearly explained. Also, have the authors considered the influence of Zn-O bond in the DFT calculation? 4. In these DFT models, vacuum is added to eliminate the influence of periodic boundary conditions, but for these asymmetric models, is the dipole correction considered in the calculation?
In addition, what are the oxidation state and spin state of Fe introduced in the system? 5. The entire structure of the OER four-step reaction calculated by DFT should be given in detail in the supplementary information, rather than just some partial structures in Figure 3e and Figure 3f. Moreover, the original total energy given in Supplementary Figure 19 can be converted into a cumulative average over time, will it be clearer? 6. Figure 2d and Supplementary Figure 6b analyze Fe 2p through XPS spectroscopy. As the number of cycles increases, the peak position and chemical composition of Fe have changed. How do you explain the change in chemical composition here? 7. Figure 2f is obtained by Fourier transform of the Fe k-edge EXAFS. As the number of ALD cycles increases, no obvious peak is seen at the position corresponding to Fe-In at 3.5Å. This does not mean that Fe-In bonds are formed in the ZISZ/Fe system, does it? 8. Figure 4e  The authors have answered most of the questions that I concern. Therefore, I consider the manuscript in this version can be accepted.
We thank the Reviewer for the constructive comments that helped us to improve the quality of our manuscript.
Reviewer #2 (Remarks to the Author): The revised manuscript has carefully considered all the comments of the reviewers and made further improvement. I would suggest the acceptance and only have one minor comment on the the IPCE and APCE data of the ZISZ/Fe photoanodes in the supplementary Figure S13, why the data show some drop in the wavelength range of 360 nm. where the A is light absorbance measured by a UV-vis spectroscopy.
Thus, owing to the similar rule of the UV-visible absorption spectra, the APCE and IPCE show the same trend.
Reviewer #3 (Remarks to the Author): This authors have synthesized ordered nanosheet arrays ZISZ/Fe photoanode material by ALD, which shows reasonably high OER activity. DFT calculation was used to study the OER mechanism to understand and verify the experimental results. However, I find some arguments are rather vague, and conclusions are not fully supported their results, and suggest that the authors carefully consider the following issues: We thank the Reviewer for the critical and constructive comments that helped us to improve the quality of our manuscript.  The following part has been incorporated into the revised manuscript.
According to the previuos DFT calculations on the WO3-x/ZnIn2S4 53 and the Li2Sx/ZnIn2S4 systems, we add an 8-atom ZnS cluster into the ZIS (001) surface and leads to the stoichiometric ZISZ system, Supplementary Fig. 7a, consistent with present experimental conditions. The

[Author Reply]:
This is a good point, thank you. We calculated the formation energy of the replacement of In with Fe in the middle layer in the ZISZ system, Supplementary Fig. 7d, and found that this value (-4.92 eV) is 0.8 eV larger than the configuration when the Fe substitutes to one In atom at the bottom layer (-5.71 eV), Supplementary Fig. 7e. Therefore, we chose the geometry shown in Supplementary Fig. 7e to perform the OER calculations because the formation energies of the two configurations by replacing of Zn with Fe are also larger than the case shown in Supplementary Fig. 7e. As a result, the original discussion and results remain unchanged ( Fig. 3e and f).
In the DFT calculation, we not considered the influence of Zn-O bond, the reason as follows: 1) In the experimental part, in order to prove that the promotion of surface OER is not related to O, only O3 was introduced into the ALD chamber (samples are denoted as ZISZ/O) to avoid the formation of Fe-In-S. The J of ZISZ/O is lower (Supplementary Fig. 17a) and the surface resistance is higher ( Supplementary Fig. 17b)  The following part has been incorporated into the revised manuscript. added the corresponding discussion in the sections of "Computational Details" and in the 2 nd paragraph of section "Results and discussion" during the revision.

Supplementary
The following part has been incorporated into the revised manuscript.
The calculated Bader charge and magnetic moment on the Fe ion correspond to 1.687 and 2.662, indicating that the oxidation state and spin state of the Fe in the ZISZ/Fe system are +2 and 3.
5. The entire structure of the OER four-step reaction calculated by DFT should be given in detail in the supplementary information, rather than just some partial structures in Figure 3e and Figure 3f. Moreover, the original total energy given in Supplementary Figure 19 can be converted into a cumulative average over time, will it be clearer?

[Author Reply]:
As the Reviewer required, we have provided all structures of the OER fourstep reaction calculated by DFT in Supplementary Fig. 20. The original total energy given in Supplementary Fig. 19 has been converted into a cumulative average over time. The small energy fluctuations around the equilibrium positions demonstrate that all structures are stable in the present study. We have added the corresponding description in the section of "Computational Details" during the revision.
The following part has been incorporated into the revised manuscript.  In other words, O-doped composition appeared at the bottom of the nanosheet. According to some references, the O doping can cause the band gap to narrow [1][2][3] . Therefore, the bottom of the sample should be a material with a narrower band gap. According to this statement, the direction of carrier movement remains constant.

Supplementary
On the other hand, limited by current technical means, the area during the Ar etching process is too small to detected. Thus, the band gap of different etching depth cannot be detected. Some