Investigation on a dopant-free hole transport material for perovskite solar cells

Abstract

A tetraphenylethene derivative (TPE-4DPA) was synthesized and applied as the hole transporting material for planar perovskite solar cells. The use of TPE-4DPA for device fabrication requires no oxidization process and lower amount of additives, delivering an overall power conversion efficiency of 12.81%. Remarkably, even in the absence of the additives, the solar cell based on the pristine TPE-4DPA still generates a J_sc as high as 20.50 mA cm⁻², and moderate PCE of 9.12%. Using no additives is of critical importance, which simultaneously increases the stability of the devices and reduces the fabrication cost of perovskite solar cells, when compared to devices using Li-TFSI doped Spiro-OMeTAD.

---

Introduction

Solar cells based on organic–inorganic hybrid perovskites have aroused widespread interest owing to their high efficiency and low cost.^1–3 The organometal halide perovskites have advantages such as direct band gap, large absorption coefficient, high charge carrier mobility and long charge diffusion length.^4,5 Furthermore, their optoelectronic properties can be facilely tuned by modification of the halide constituents. The first attempt at using organolead halide perovskite as a light absorber in solar cells gained a power conversion efficiency (PCE) of 3.8% using liquid electrolyte in the sensitized device structure.⁶ Within six years, the overall conversion efficiencies have achieved tremendous improvement to 20.1%.⁷ To date, most of the efficient perovskite solar cells involve organic hole transport materials (HTMs).^8–11 Among these organic materials, the Spiro-OMeTAD (2,2-7,7-tetrakis(N,N′-diparamethoxy-phenylamine 9,9′-spirobifluorene)) is the most popular one and have attracted wide attention, due to its good solubility, high hole concentration and good hole mobility.^12,13 In spite of these merits, there are still some disadvantages of Spiro-OMeTAD. The high synthetic cost is one of the major obstacles to commercialize the perovskite solar cells. In addition, ionic additives or p-type dopants (Li-TFSI, lithium bis(trifluoromethylsulfonyl)-imide) are required for Spiro-OMeTAD to increase its carrier densities.^14,15 As in the pristine form, the Sprio-OMeTAD has low hole density and conductivity.^16,17 While a fact existed is that these additives are detrimental to the perovskite layers. It has been reported that these additives may lead to decomposition of the moisture-sensitive perovskite.^18,19 Additionally, the Spiro-OMeTAD on top of perovskite needs to be exposed to the ambient atmosphere for oxidation during the device fabrication, which might cause degradation of the perovskite active layer. In this regard, new HTMs based on small organic molecules, polymers and inorganic salts have been intensively explored in these years as potential alternatives for Spiro-OMeTADs.^20,21

Among the reported HTMs, small organic molecules are extensively investigated, since they are cheap in cost and easy to purify. Besides, they also show good infiltration into the nanostructure which induces reproducible device performances. Most recently, impressive photovoltaic performances have been obtained using small molecule based HTMs.^22,23 However, most of HTMs reported for solution-processed perovskite solar cells requires the addition of ion additives, e.g., lithium bis(trifluoromethylsulfonyl)imide (Li-TFSI) with 4-tert-butylpyridine (tBP) to get improved hole mobility and high device performances.^24,25 Few HTMs in their pristine form can compete with the additive doped Spiro-OMeTAD for perovskite solar cells.^26–29 P. Qin et al. reported a branched conjugated HTM working alone for meso-structured perovskite solar cells with a high PCE of 12.8%.²⁶ Ganesan et al. reported a new spiro-dioxepine HTM exhibiting high efficiency of 12.7% without the dopant.²⁷ Kazim and coworkers developed a linear acene derivative as HTM, and power conversion efficiency (PCE) of 11.8% is achieved with pristine TIPS-pentacene.²⁸ The use of Li-TFSI should be avoided, because it will not only increase the costs, but also seriously leads to decomposition of the perovskite, due to its high hydroscopicity.^{18,19,29–32} Therefore, it is still a great challenge to develop new HTMs with easy and convenient synthetic procedures, especially materials that could avoid complex fabrication and work without the dopants.

In this work, we report a dopant-free HTM based on tetraphenylethene derivative TPE-4DPA (Scheme 1) with very simple synthetic route and low cost. The TPE derivative has been demonstrated to be efficient HTM for perovskite solar cells quite recently.^33,34 In the reported cases, the TPE derivatives need to be doped with additives to achieve high power conversion efficiency in mesoscopic perovskite solar cells. While in our work, the TPE-4DPA is demonstrated to be an efficient dopant-free HTM. We fabricate the planar perovskite solar cells, which is easier and of lower cost to fabricate compared with the mesoscopic structure. And the doped TPE-4DPA shows comparable efficiency to the values reported in literature. Remarkably, we find that even without additives, the pristine TPE-4DPA still generates a J_sc as high as 20.50 mA cm⁻² and moderate PCE of 9.12% in planar perovskite solar cells, suggesting highly efficient hole collection and transporting ability of TPE-4DPA from the perovskite layer. In addition, the pristine TPE-4DPA covered perovskite film shows better stability compared to the Li-TFSI doped Spiro-OMeTAD.

Scheme 1 Molecular structures of Spiro-OMeTAD and TPE-4DPA.

Results and discussion

TPE-4DPA was prepared through a facile one-step reaction between 1,1,2,2-tetrakis(4-bromophenyl)ethane and bis(4-methoxyphenyl)amine with a high yield of 73% (Scheme S1†).³⁵ The product is fully characterized by ¹H NMR, ¹³C NMR and the HRMS spectra. Detailed methods and characterizations are provided in the ESI.†

Cyclic voltammetric measurements were carried out to analyze and compare the energy levels of the new HTM with Spiro-OMeTAD for the favorable hole injection process from the perovskite layer. For comparison, we measured CV of both materials under the same condition with ferrocene as an internal standard (Fig. S4b†). The HOMO and LUMO energy levels of TPE-4DPA were calculated to be −5.17 eV and −2.48 eV, respectively. The suitable energy level with the perovskite layer as shown in Fig. 1 indicates a good hole extraction and electron blocking ability of TPE-4DPA. It is noteworthy that the HOMO level of TPE-4DPA is about 100 mV lower than that of Spiro-OMeTAD, which is more favorable for the hole transfer from the perovskite to TPE-4DPA (Fig. 1a). The hole mobility of TPE-4DPA, according to literature report, was 5.92 × 10⁻⁵ cm² V⁻¹ s⁻¹ determined by the SCLC method.³³ The cross section viewed SEM image of a complete device is shown in Fig. 1b. A perovskite layer with the thickness about 350 nm sandwiched between the 50 nm-thick TiO₂ layer and the 150 nm-thick HTM layer is observed. The large crystallites of the perovskite layer expand across the whole thickness and maintain a high coverage on the substrate.

Fig. 1 (a) Energy diagram of each layer in perovskite solar cells; (b) SEM cross-sectional image of the perovskite solar cell made with TPE-4DPA as HTM.

The performance of the HTM was evaluated using the planar structure (FTO/TiO₂/CH₃NH₃PbI₃/HTM/Ag). Firstly, a thin TiO₂ compact layer was prepared by spin coating the acidic titanium isopropoxide solution on FTO substrate and calcining at 500 °C for 1 h. Then, the FTO/TiO₂ substrate was transferred into a N₂-filled glovebox to prepare the perovskite layer. In particular, the PbI₂ solution and CH₃NH₃I (MAI) solution were spin-coated on the substrate in sequence, followed by annealing at 100 °C for 1 h. The intercalation and reaction between PbI₂ and MAI produced a smooth CH₃NH₃PbI₃ film. Afterwards, the TPE-4DPA and Spiro-OMeTAD solution in the absence/presence of additives were spin-coated on the perovskite layer, separately. Finally, a 100 nm thick layer of Ag was thermally evaporated on top of the HTM layer. The overlap between the Ag electrode and the layers below defines the active area of the device as 0.12 cm². As for the Spiro-OMeTAD, the oxidation in air is required in order to get a high doping density. Correspondingly, we also measured the performance of TPE-4DPA with or without the oxidization processes.

The J–V responses of the perovskite solar cells incorporating Spiro-OMeTAD and TPE-4DPA as HTMs are presented in Fig. 2a with a scan rate of 50 mV/50 ms and a backward scan direction from open circuit condition to short circuit condition under AM 1.5G illumination. Obviously, the performance of the device using Spiro-OMeTAD as HTM is quite poor in the absence of the additives or without the oxidization process, showing a quite low PCE of 7.66% and 0.55%, respectively. Only when Spiro-OMeTAD is doped with additives and undergoes the oxidizing procedure, the performance of the device could be improved to 14.37%, with V_oc, J_sc and FF of 1.05 V, 20.59 mA cm⁻² and 0.665, respectively. Therefore, the additives and oxidization process are essential treatments for getting good performance of Spiro-OMeTAD.

Fig. 2 (a) Current–voltage characteristics of the TPE-4DPA and Spiro-OMeTAD based perovskite solar cells in different conditions; (b) IPCE of TPE-4DPA in the presence and absence of additives.

While for TPE-4DPA, even without oxidizing in the air, the doped device achieves PCE as high as 12.81% (in planar perovskite solar cells), which is comparable with the reported value of 13.09% (in mesoscopic perovskite solar cells).³⁴ Remarkably, the undoped device under the same condition, still generates a moderate PCE of 9.12%. It is obvious that the efficiency of the undoped device is of one order's improvement compared to the Spiro-OMeTAD. Moreover, the oxidization process applied to TPE-4DPA has no further improvement on PCE. Devices based on TPE-4DPA show a V_oc of 1.08 V, similar with that based on the Spiro-OMeTAD. This is in consistent with the energy level diagram got from the CV scan as shown in Fig. S4b.† The FF of TPE-4DPA-based device is 0.584, which is inferior to that of the doped Spiro-OMeTAD device. It possibly results from the higher charge transport resistance of the HTM layer due to a lower doping density. The J_sc of the solar cell using TPE-4DPA is 20.31 mA cm⁻², which is comparable to that measured from the Spiro-OMeTAD standard device. It is worth mentioning that even without additives and the oxidization process, the solar cell based on the pristine TPE-4DPA still presents a J_sc as high as 20.50 mA cm⁻², suggesting efficient hole collection and hole transporting ability of TPE-4DPA from the perovskite layer. The J–V curve hysteresis is frequently observed in the perovskite solar cells, especially in those planar-structured devices.³⁶ In our experiment, the device adopts Spiro-OMeTAD as HTM shows a hysteresis of 6.16% while the ones with TPE-4DPA shows smaller hysteresis of 3.88% and 3.00% for the doped and undoped devices, respectively. The IPCE spectra of solar cells based on the doped and undoped TPE-4DPA as HTMs is shown in Fig. 2b. The integrated short-circuit current density is 17.4 mA cm⁻² and 18.0 mA cm⁻² for the devices with undoped and doped TPE-4DPA as HTM, respectively. The slightly smaller integrated J_sc may result from the different test condition (The IPCE is tested in ambient environment while the J–V curve is scanned in a N₂-filled glovebox.). Similar spectra in the two conditions confirm that TPE-4DPA is an effective hole extraction material and hole transport material even without additives or oxidation process. We also note that, in the undoped devices, lower V_oc and inferior FF are observed (Table 1), since doping of additives could help to improve the photovoltaic performance, and this result is consistent with a recent work by Ahmad and coworkers.³⁷ However, in its pristine form, TPE-4DPA performed well and showed higher PCE compared to the standard Spiro-OMeTAD.

Table 1 Photovoltaic parameters of perovskite solar cells with TPE-4DPA and Spiro-OMeTAD as HTMs with and without additives or oxidation process

Devices	Voc (V)	Jsc (mA cm^−2)	FF	PCE (%)	Hysteresis (%)
TPE-4DPA doped	1.08	20.31	0.58	12.81	3.88
TPE-4DPA undoped	0.98	20.50	0.45	9.12	3.00
Spiro-OMeTAD undoped	0.97	20.04	0.39	7.66	1.22
Spiro-OMeTAD undoped + unoxidized	0.86	3.54	0.18	0.55	0.06
Spiro-OMeTAD standard	1.05	20.59	0.66	14.37	6.16

---

It is interesting that spiro-OMeTAD and TPE-4DPA with similar molecular structure show different properties. Considering the same subunits of the two compounds, the differentiation should come from the core structures, that is, the spiro-bifluorene and the tetraphenylethylene. In the film state, the tetraphenylethylene unit tends to undergo planarization, while the spiro-bifluorene is still in a twisted conformation in films. Therefore, the stability of the TPE-4DPA-based cell compared with the Spiro-OMeTAD based one may be attributable to an interfacial tight packing due to a planar and sterically bulky configuration.³⁴ And according to literature results, the hole mobility of TPE-4DPA is a little lower than that of the Spiro-OMeTAD,^33,34 which explains the lower power conversion efficiency of the TPE-4DPA based perovskite cell.

The stability of the perovskite layer covered by pristine TPE-4DPA and the Li-TFSI doped Spiro-OMeTAD is examined using the UV-vis spectra as shown in Fig. 3. Typically, the perovskite film with a layer of HTM covered on top is stored in ambient condition for 12 h with 60% RH exposed to ordinary lamp light. Initially, both films show similar absorption spectra with an edge at 780 nm, which matches well with those reported data. For the pristine TPE-4DPA covered perovskite film, it shows little degradation in the absorption spectrum even after storing in air for 12 h. While for the Li-TFSI doped Spiro-OMeTAD covered perovskite film, most area of the perovskite film turns from dark brown to transparency, possibly due to the formation of hydrated compounds (Fig. 3). According to the absorption spectrum, significant decrease in absorption intensity is found in the range from 450 to 800 nm for Spiro-OMeTAD. The superior stability improvement could be ascribed to no additives used in the TPE-4DPA layer, since it has been reported that the tert-butylpyridine additive can induce degradation of CH₃NH₃PbI₃, and the lithium salt is quite hydrolysis which is supposed to accelerate the hydration of perovskite.^18,19 The result clearly demonstrates that the doped additives will accelerate the degradation of the PSC devices. Furthermore, the device stability is also examined by storing the unpackaged device at ambient condition with 60% RH and room light illumination. The performance of the device is characterized every 6 h under inert atmosphere in order to get rid of the influence of moisture and O₂ during the test. The normalized performance evolution is shown in Fig. S7.† The device based on TPE-4DPA maintained about 66% of the initial performance while the Spiro-OMeTAD analogue decreased to less than 3% of the initial performance. Therefore, to maintain long-term stability, designing new HTMs with high hole mobility in their pristine form is of great importance. As the perovskite film is unstable towards moisture and O₂ in air, the elimination of the oxidization process and using less or none additives (when using TPE-4DPA) are of significant importance, which enables the whole fabrication process to be carried out under an inert atmosphere.

Fig. 3 Light absorption spectra of the perovskite films covered by the Li-TFSI doped Spiro-OMeTAD and pristine TPE-4DPA after storing in air with 60% RH for 12 h. Inset shows the film color changes in ambient condition with 60% RH and under room light illumination 12 h.

In general, TPE-4DPA based cells shows inferior conversion efficiency compared to that of the spiro-OMeTAD based standard device. Yet we believe that there is still room for improvement. As shown in Table 1, both J_sc and V_oc values are comparable for the two hole transporting small molecules, but the FF of TPE-4DPA based devices are obviously lower. If the fabrication of the planar device could be optimized, the TPE-4DPA will have better performance.

Experimental

Device fabrication

FTO-coated glass with sheet resistance of 14 Ω sq⁻¹ was washed by sonication with deionized water, ethanol and acetone, and then treated with oxygen plasma for two minutes. A compact layer of TiO₂ was deposited on the FTO substrate by spin-coating the titanium precursor (0.24 M titanium isopropoxide and 0.12 M HCl in ethanol) at 5000 rpm for 60 s following by calcination on a hotplate at 500 °C for 60 min. Subsequently, the PbI₂ solution (0.8 M) and MAI solution (0.25 M) were spin-coated on the substrate in sequence at 2000 rpm for 60 s, followed by annealing at 100 °C for 1 h in a nitrogen-filled glovebox. Then, the pristine TPE-4DPA solution (15 mg mL⁻¹ in chlorobenzene) or the Li-TFSI doped TPE-4DPA solution (15 mg TPE-4DPA, 5 μL 4-tert-butylpyridine (tBP) and 3 μL lithium-bis(trifluoromethanesulfonyl)imide (Li-TSFI) stock solution (520 mg mL⁻¹ in acetonitrile) in 1 mL chlorobenzene) was spin-coated on the perovskite layer at 1000 rpm for 60 s. As for the reference cells, Spiro-OMeTAD was deposited by spin coating a solution (72.5 mg Spiro-OMeTAD, 28.5 μL tBP and 17.5 μL Li-TSFI stock solution in 1 mL chlorobenzene) at 5000 rpm for 60 s. After oxidizing the HTM layer in air for 15 h (for devices without oxidizing, this step could be omitted), the cell was completed by thermally evaporating a 100 nm thick silver layer.

Conclusions

In conclusion, we have demonstrated a dopant-free HTM based on tetraphenylethene derivative in planar perovskite solar cells. The energy levels estimated from CV are well aligned with respect to the perovskite energy level. Device fabricated using doped TPE-4DPA exhibits an overall power conversion efficiency of 12.81% even without oxidation process, which is comparable to the literature report values for the TPE derivative. More excitingly, the pristine TPE-4DPA still presents a J_sc as high as 20.50 mA cm⁻² and PCE of 9.12% even in the absence of the additives, suggesting efficient hole collection ability of TPE-4DPA from the perovskite layer and excellent hole transporting ability. In addition, the pristine TPE-4DPA covered perovskite film exhibits superior stability compared to the Li-TFSI doped Spiro-OMeTAD covered film, as the doped additives are reported to accelerate the degradation of the PSC devices and thus are detrimental to the devices. We believe that the TPE-based HTMs have great potential to be an alternative to the expensive Spiro-OMeTAD due to their ease of fabrication, low cost, excellent solubility and stability.

Acknowledgements

The authors thank the National Natural Science Foundation of China (No.51203046) for financial support. T. C. acknowledges the grants from RGC Theme-based Research Scheme under project no. T23-407/13 N and the Thousand Young Talents Program.

Notes and references

1. D. B. Mitzi, Prog. Inorg. Chem., 2007, 55, 1.
2. J. H. Heo, S. H. Im, J. H. Noh, T. N. Mandal, C.-S. Lim, J. A. Chang, Y. H. Lee, H.-J. Kim, A. Sarkar, M. K. Nazeeruddin, M. Grätzel and S. II Seok, Nat. Photonics, 2013, 7, 486.
3. W. Chen, Y. Wu, Y. Yue, J. Liu, W. Zhang, X. Yang, H. Chen, E. Bi, I. Ashraful, M. Grätzel and L. Han, Science, 2015, 350, 944.
4. B. Wang, X. Xiao and T. Chen, Nanoscale, 2014, 6, 12287.
5. M. M. Lee, J. Teuscher, T. Miyasaka, T. N. Murakami and H. J. Snaith, Science, 2012, 338, 643.
6. A. Kojima, K. Teshima, Y. Shirai and T. Miyasaka, J. Am. Chem. Soc., 2009, 131, 6050.
7. National Renewable Energy Laboratory Efficiency Chart, http://www.nrel.gov/ncpv/images/efficiencychart.jpg, accessed January 2015.
8. Y. S. Kwon, J. Lim, H.-J. Yun, Y.-H. Kim and T. Park, Energy Environ. Sci., 2014, 7, 1454.
9. Z. Yu and L. Sun, Adv. Energy Mater., 2015, 5, 1500213.
10. N. J. Jeon, J. Lee, J. H. Noh, M. K. Nazeeruddin, M. Grätzel and S. II Seok, J. Am. Chem. Soc., 2013, 135, 19087.
11. B. Wang, K. Y. Wong, X. Xiao and T. Chen, Sci. Rep., 2015, 5, 10557.
12. J. Burschka, N. Pellet, S.-J. Moon, R. Humphry-Baker, P. Gao, M. K. Nazeeruddin and M. Grätzel, Nature, 2013, 499, 316.
13. M. Liu, M. B. Johnston and H. J. Snaith, Nature, 2013, 501, 395.
14. H. Li, K. Fu, A. Hagfeldt, M. Grätzel, S. G. Mhaisalkar and A. C. Grimsdale, Angew. Chem., Int. Ed., 2014, 53, 4085.
15. J. Liu, Y. Wu, C. Qin, X. Yang, T. Yasuda, A. Islam, K. Zhang, W. Peng, W. Chen and L. Han, Energy Environ. Sci., 2014, 7, 2963.
16. J. Burschka, A. Dualeh, F. Kessler, E. Baranoff, N. L. Cevey-Ha, C. Y. Yi, M. K. Nazeeruddin and M. Gratzel, J. Am. Chem. Soc., 2011, 133, 18042.
17. T. Leijtens, I. K. Ding, T. Giovenzana, J. T. Bloking, M. D. McGehee and A. Sellinger, ACS Nano, 2012, 6, 1455.
18. J. H. Noh, S. H. Im, J. H. Heo, T. N. Mandal and S. I. Seok, Nano Lett., 2013, 13, 1764.
19. L. Zheng, Y.-H. Chung, Y. Ma, L. Zhang, L. Xiao, Z. Chen, S. Wang, B. Qu and Q. Gong, Chem. Commun., 2014, 50, 11196.
20. S. Ma, H. Zhang, N. Zhao, Y. Cheng, M. Wang, Y. Shen and G. Tu, J. Mater. Chem. A, 2015, 3, 12139.
21. T. Swetha and S. P. Singh, J. Mater. Chem. A, 2015, 3, 18329.
22. P. Gratia, A. Magomedov, T. Malinauskas, M. Daskeviciene, A. Abate, S. Ahmad, M. Grätzel, V. Getautis and M. K. Nazeeruddin, Angew. Chem., Int. Ed, 2015, 54, 11409.
23. P. Qin, N. Tetreault, M. I. Dar, P. Gao, K. L. McCall, S. R. Rutter, S. D. Ogier, N. D. Forrest, J. S. Bissett, M. J. Simms, A. J. Page, R. Fisher, M. Grätzel and M. K. Nazeeruddin, Adv. Energy Mater., 2015, 5, 1400980.
24. M. Cheng, C. Chen, X. Yang, J. Huang, F. Zhang, B. Xu and L. Sun, Chem. Mater., 2015, 27, 1808.
25. H. Choi, J. W. Cho, M.-S. Kang and J. Ko, Chem. Commun., 2015, 51, 9305.
26. P. Qin, S. Paek, M. I. Dar, N. Pellet, J. Ko, M. Gräztel and M. K. Nazeeruddin, J. Am. Chem. Soc., 2014, 136, 8516.
27. P. Ganesan, K. Fu, P. Gao, I. Raabe, K. Schenk, R. Scopelliti, J. Luo, L. H. Wong, M. Grätzel and M. K. Nazeeruddin, Energy Environ. Sci., 2015, 8, 1986.
28. S. Kazim, F. J. Ramos, P. Gao, M. K. Nazeeruddin, M. Grätzel and S. Ahmad, Energy Environ. Sci., 2015, 8, 1816.
29. G. Niu, W. Li, F. Meng, L. Wang, H. Dong and Y. Qiu, J. Mater. Chem. A, 2014, 2, 705.
30. S. Kazim, M. K. Nazzeeruddin, M. Grätzel and S. Ahmad, Angew. Chem., Int. Ed, 2014, 53, 2812–2824.
31. C. Huang, W. Fu, C.-Z. Li, Z. Zhang, W. Qiu, M. Shi, P. Heremans, A. K.-Y. Jen and H. Chen, J. Am. Chem. Soc., 2016, 138, 2528–2531.
32. F. Zhang, C. Yi, P. Wei, X. Bi, J. Luo, G. Jacopin, S. Wng, X. Li, Y. Xiao, S. M. Zakeeruddin and M. Grätzel, Adv. Energy Mater., 2016, 1600401.
33. L. Cabau, I. Garcia-Benito, A. Molina-Ontoria, N. F. Montcada, N. Martin, A. Vidal-Ferran and E. Palomares, Chem. Commun., 2015, 51, 13980.
34. H. Choi, K. Do, S. Park, J.-S. Yu and J. Ko, Chem.–Eur. J., 2015, 21, 15919.
35. Y. Ma, H. Ma, Z. Yang, J. Ma, Y. Su, W. Li and Z. Lei, Langmuir, 2015, 31, 4916.
36. M. Deepa, F. J. Ramos, S. M. Shivaprasad and S. Ahmad, ChemPhysChem, 2016, 17, 913–920.
37. F. J. Ramos, K. Rakstys, S. Kazim, M. Grätzel, M. K. Nazeeruddin and S. Ahmad, RSC Adv., 2015, 5, 53426–53432.

---

Footnotes

† Electronic supplementary information (ESI) available. See DOI: 10.1039/c6ra07603c

‡ These authors contributed equally to this work.

---