**Transparent Platinum Counter Electrode Prepared by Polyol Reduction for Bifacial, Dye-Sensitized Solar Cells**

#### **Alvien Ghifari , Dang Xuan Long, Seonhyoung Kim, Brian Ma and Jongin Hong \***

Department of Chemistry, Chung-Ang University, 84 Heukseok-ro, Dongjak-gu, Seoul 06974, Korea; alvien.ghifari@gmail.com (A.G.); dxlong.bk@gmail.com (D.X.L.); kimsh9560@gmail.com (S.K.); kehgeeng@gmail.com (B.M.)

**\*** Correspondence: hongj@cau.ac.kr; Tel.: +82-2-820-5869

Received: 7 February 2020; Accepted: 3 March 2020; Published: 11 March 2020

**Abstract:** Pt catalytic nanoparticles on F-doped SnO2/glass substrates were prepared by polyol reduction below 200 ◦C. The polyol reduction resulted in better transparency of the counter electrode and high power-conversion efficiency (PCE) of the resultant dye-sensitized solar cells (DSSCs) compared to conventional thermal reduction. The PCEs of the DSSCs with 5 µm-thick TiO<sup>2</sup> photoanodes were 6.55% and 5.01% under front and back illumination conditions, respectively. The back/front efficiency ratio is very promising for efficient bifacial DSSCs.

**Keywords:** dye-sensitized solar cell; counter electrode; bifacial; platinum; ethylene glycol

#### **1. Introduction**

Dye-sensitized solar cells (DSSCs) have shown promise as low-cost photovoltaics compared to commercially available Si solar cells. They hold great potential for building-attached photovoltaics (BAPVs) and building-integrated photovoltaics (BIPVs), because of their adjustable color/transparency and superior performance in dim light [1–4]. Recently, bifacial DSSCs, which can convert incoming sunlight to electricity through both front and back sides, have been an attractive alternative for photovoltaic devices [5,6]. The standard DSSC consists of a substrate coated with transparent conducting oxides (TCOs), a dye-grafted mesoscopic TiO<sup>2</sup> photoanode, a platinized counter electrode (CE), and an electrolyte containing a redox couple. Among the key components, the CE plays a prominent role in maintaining a flow of current by regenerating the redox mediator. Unfortunately, Pt is a highly expensive and scarce metal, and thus alternative materials, including carbon-based materials [7,8], transition metal compounds [9], conducting polymers [10,11], and their composites [12,13] have been explored.

Nevertheless, Pt is still favored because of its superior electrocatalytic activity and good conductivity. The conventional methods for preparing Pt CEs are vacuum sputtering of a Pt target and thermal decomposition of a platinic acid (H2PtCl6) precursor [14,15]. However, these high-energy-consuming approaches increase the cost and energy payback time of DSSCs. In addition to logistical issues, the light absorption at the Pt CEs should be minimized for bifacial operation. Therefore, it is crucial to develop new preparation methods for highly transparent Pt CEs at low temperatures.

Polyol-based synthesis is a versatile and straightforward liquid-phase method that uses high-boiling and multivalent alcohols to synthesize nanomaterials without the requirements of high pressure and autoclaves [16,17]. Polyol can simultaneously act as a reducing agent and water-equivalent solvent. Its chelating ability or controlling the nucleation of nanomaterials. For example, Ouyang and coworkers reported nanostructured Pt CEs prepared by polyol reduction of H2PtCl<sup>6</sup> in ethylene glycol

(EG) [18,19]. They also utilized EG vapor for solventless chemical reduction of the platinum precursor below 200 ◦C [20]. Song et al. employed the hydrolysis of urea in a two-step EG solution reduction to achieve uniform dispersion and a smaller size of Pt nanoparticles on conducting glass substrates [21]. Unfortunately, the Pt CEs were prepared by drop-casting the EG-based Pt precursor solution. It has proven challenging to prepare a thin layer of Pt nanoparticles and thus provide highly transparent CEs for the bifacial DSSCs. Therefore, we prepared a highly transparent Pt counter electrode by spin-coating a platinic EG solution and further chemical reduction below 200 ◦C. We also investigate the photovoltaic performance of the DSSCs from the viewpoint of bifacial operation. ‐ ‐ ‐ ‐

#### **2. Materials and Methods**

#### *2.1. Fabrication of Pt Counter Electrodes*

F-doped SnO<sup>2</sup> (FTO) glass (TEC 8, Pilkington; sheet resistance = 8Ω/) substrates were ultrasonically cleaned using acetone, isopropyl alcohol, and deionized water. The substrates were baked at 150 ◦C for 10 min to completely remove residual water. Then 10 mM platinic acid in ethylene glycol was spin-coated on the FTO substrates. Subsequently, the sample was placed in a muffle furnace and heated to a certain temperature (e.g., 130, 150, 170, 190, and 210 ◦C). The annealing was maintained for 12 h, and then the sample was cooled down to room temperature. For comparison, 40 mM of the platinic acid solution in 2-propanol was prepared and then spin-coated on the same substrates, followed by heating at 425 ◦C for 1 h in the muffle furnace [22–24]. The schematic diagram of Pt CE preparation is shown in Figure 1. ‐ Ω □ ‐ ‐ ‐

**Figure 1.** Schematic diagram of counter electrode preparation: polyol reduction (PR) and thermal decomposition (TD).

#### *2.2. Characterization*

‐ ‐ ‐ ‐ α ‐ ‐ The surface images of the fabricated Pt CEs were obtained using a field-emission scanning electron microscope (FE-SEM; Sigma, Carl Zeiss AG, Oberkochen, Germany). X-ray photoelectron spectroscopy (XPS) spectra were acquired using a Thermo Scientific K-alpha XPS system (Waltham, MA, United States) with an Al Kα X-ray source monochromator (1486.6 eV). The optical transmittance spectra were recorded using UV/Vis spectroscopy (V-730, JASCO, Tokyo, Japan). An electrochemical measurement station (CompactStat, Ivium Technologies, Eindhoven, The Netherlands) was used to conduct all electrochemical characterization. Cyclic voltammetry (CV) was recorded at a scan rate of 50 mV/s with a three-electrode system that consists of a Pt CE as a working electrode, a Pt wire as a counter electrode, and Ag/AgCl as a reference electrode. A solution of 10.0 mM LiI, 1.0 mM I2, and 0.1 M LiClO<sup>4</sup> in CH3CN was used as the electrolyte to investigate the electrocatalytic properties of

the Pt CEs for redox reactions. Tafel and electrochemical impedance spectroscopy (EIS) measurements were performed on symmetric cells consisting of CE|electrolyte|CE. Tafel polarization measurements were conducted at a scan rate of 10 mV/s. In EIS measurements, a 10 mV amplitude sinusoidal potential perturbation was input over a frequency range from 1 MHz to 0.1 Hz at zero bias potential. The EIS spectra were analyzed using the equivalent circuit fitting routine in the ZView software (AMETEK, Leicester, UK).

#### *2.3. Device Fabrication and Characterization*

The FTO glass substrates were cleaned with O<sup>2</sup> plasma for 10 min, dipped in an aqueous solution of 40 mM TiCl<sup>4</sup> at 75 ◦C for 30 min, and then rinsed several times with deionized water. The TiO<sup>2</sup> paste (Transparent TiO2, ENBKOREA, Gumi, Korea) was screen-printed on the FTO glass, and the printed film was calcinated at 300 ◦C for 30 min and 575 ◦C for 1 h in the muffle furnace. The final areas of the TiO<sup>2</sup> photoanodes were 0.2025 cm<sup>2</sup> . The photoanodes were treated in the TiCl<sup>4</sup> solution and then heated at 500 ◦C for 30 min on a hot plate. After O<sup>2</sup> plasma treatment, the photoanodes were dipped into a 0.5 mM dye solution of cis-diisothiocyananoto-bis(2,2-bipyridyl-4,4′ -dicarboxylate) ruthenium(II) (N719, Merck KGaA, Darmstadt, Germany) in ethanol for 24 h. The dye-grafted photoanode and Pt CE were assembled with 25 µm-thick thermoplastic film (Surlyn, Solaronix, Aubonne, Switzerland) and sealed by heating. An iodide-based redox electrolyte (Iodolyte AN-50, Aubonne, Switzerland) was injected into the pre-drilled holes in the side of the counter electrode and then sealed. The photovoltaic characteristics were investigated under AM 1.5 global one sun illumination (100 mW/cm<sup>2</sup> ) using a solar cell I–V measurement system (K3000 LAB, McScience, Suwon, Korea). The photocurrent density (*Jsc*), open-circuit voltage (*Voc*), fill factor (*FF*), and power conversion efficiency (η) were recorded simultaneously. Monochromatic incident photon-to-current conversion efficiency (IPCE) was collected to evaluate the spectral response of the solar cells (K3100, McScience, Suwon, Korea). EIS measurement on the fabricated devices was carried out by the same protocol mentioned above.

#### **3. Results**

Polyol-based synthesis is a versatile technique to prepare Pt nanostructures. In this study, EG was chosen as the solvent and reducing agent of Pt precursors because of its low boiling point and viscosity. Also, byproducts from the EG reduction, such as glycoaldehyde and diacetyl, could be easily removed [25]. Figure 2a–f shows the surface of the FTO glass substrates decorated with Pt nanoparticles formed by polyol reduction at different temperatures (130, 150, 170, 190, and 210 ◦C) and thermal decomposition at 425 ◦C (hereafter, "polyol-reduced" is abbreviated as "PR" and "thermally decomposed" is abbreviated as "TD"). Tiny Pt nanoparticles (<10 nm) were formed and dispersed on the FTO surface for all reaction temperatures. The chemical reduction by EG allows for depositing Pt nanoparticles at low temperatures, and is thus feasible for plastic substrates. As the reduction temperature increased, the aggregation of the Pt nanoparticles diminished, and no aggregation could be observed at the temperature of 190 ◦C. This indicates that the slower evaporation of EG could result in the growth of larger nanoparticles, more agglomeration, and dendrites. However, the aggregated nanoparticles appeared again at a temperature higher than the boiling point of EG (i.e., 197 ◦C) because of the inability to control particle nucleation and growth at the elevated temperature [17]. Also, the pyrolysis of H2PtCl<sup>6</sup> at 425 ◦C directly led to the formation of large Pt nanoparticles prominently populating the FTO. Accordingly, we predict that their size and distribution will affect the transparency and catalytic activity of the resultant electrodes.

XPS was performed to determine the formation of metallic Pt during our chemical reduction. In Figure 3, the spectra of survey XPS indicates that the samples contained Sn, O, Pt, and Cl—no other elements except carbon were detected. Carbon can result from precursors or sample handling. Interestingly, as the reduction temperature increased, the signal for Cl 2p related to ionic Pt species decreased and then disappeared. Narrow XPS scans of the Pt-4f core level region are also provided in Figure 3. The C 1 s peak at 285.0 eV was used to calibrate all the XPS data. The Pt 4f signal is composed

‐

ܫ ି

ᇞ

‐ ‐

ᇞ

of three pairs of deconvoluted doublets. The first doublet (71.4 eV and 74.8 eV) corresponds to the platinum in the zero-valent state (i.e., Pt(0)), while the second doublet (72.3 eV and 75.6 eV) can be assigned to platinum in the 2+ valence state (i.e., Pt(II)) [26]. The third (73.7 eV and 77.0 eV) is caused by the Pt4<sup>+</sup> species, such as [PtCl6] <sup>2</sup>−, on the surface (i.e., Pt(IV)) [27]. Table 1 summarizes the binding energies and relative integrated peak areas of the deconvoluted peaks at the Pt-4f core level region. The increase in the reduction temperature resulted in the change in the valence state from Pt2<sup>+</sup> to Pt<sup>0</sup> . After the polyol reaction, the Pt species clustered together with different compositions. Accordingly, the reaction temperature is of significant importance in preparing metallic Pt nanoparticles.

‐ ‐ ‐ ‐ ‐ ‐ ‐ ‐ **Figure 2.** Field-emission scanning electron microscope (FE-SEM) images of Pt films on FTO substrates: (**a**) PR-130, (**b**) PR-150, (**c**) PR-170, (**d**) PR-190, (**e**) PR-210, and (**f**) TD-425.

‐ ‐ ‐ ‐ ‐ **Figure 3.** X-ray photoelectron spectroscopy (XPS) scans of polyol-reduced Pt on FTO substrates: (**a**) survey, (**b**) PR-130, (**c**) PR-190, and (**d**) PR-210.

ିܫ3 ⇌ 2݁ି ି

<sup>ଷ</sup>ܫ2 ⇌ 2݁ି <sup>ଶ</sup>ܫ3

‐ ‐ ‐ ‐

ି

ଷܫ ି

‐ ‐

 ‐ ‐ ‐

> ଷܫ ିܫି/

> > ଷܫ


**Table 1.** Binding energies (eV) and relative proportions (S, in %) of each component measured by XPS.

In the DSSC operation, an *I* − 3 /*I* <sup>−</sup> redox couple is commonly utilized as a redox mediator. The CE should collect electrons from the external circuit and effectively catalyze the reduction of *I* − 3 to *I* −. The electrocatalytic activity of the PR CEs was determined using cyclic voltammetry (CV), as shown in Figure 4a. All CV curves show two typical pairs of oxidation and reduction peaks (Ox-1/Re-1 and Ox-2/Re-2), which are described by Equations (1) and (2), respectively [28]:

$$I\_3^- + 2e^- \rightleftharpoons 3I^- \tag{1}$$

$$2\text{J}\_2 + 2e^- \rightleftharpoons 2\text{I}\_3^- \tag{2}$$

‐

ଷܫ ି

The catalytic reduction activity of the CE can be determined by the first oxidation and reduction peaks (Ox-1 and Re-1). Its electrochemical reversibility can be assessed from the peak-to-peak separation (∆*Ep*), which is the difference between the anodic and cathodic peak potentials. A smaller ∆*E<sup>p</sup>* reflects the higher electrocatalytic activity of the CE.

ଷܫ ିܫି/ **Figure 4.** (**a**) Cyclic voltammograms for the redox of *I* − 3 /*I* <sup>−</sup> species, (**b**) electrochemical impedance spectroscopy (EIS) Nyquist plots for the symmetric cells, and (**c**) Tafel polarization curves.

‐ EIS measurements were also conducted using the symmetrical cells constructed with two identical electrodes. The Nyquist plots (Figure 4b) consist of two apparent semicircles at a high frequency (first semicircle) and a low frequency (second semicircle), respectively. The Randles-type circuit (insert in Figure 4b) was used to simulate the plots. The corresponding parameters are summarized in Table 2. *Rs* is the series resistance, and can be derived by the intercept of the high-frequency semicircle on the real axis (*Z*' axis). *Rct* is the charge transfer resistance at the interface between CE and electrolyte, and can be determined from the radius of the first semicircle on the real axis. All the PR-CEs have nearly

‐

ᇞ **Ω Ω Ω**

‐

‐

‐ ‐ ‐ ‐ ‐ ‐

ଷܫ ିܫି/ ‐

the same *Rs* , and thus its effect on photovoltaic performance can be ignored. A smaller *Rct* accelerates triiodide reduction, and thus the PR-190 CE has the superior catalytic activity. The low-frequency semicircle results from Nernst diffusion impedance (*ZN*) of the redox species in the electrolyte, which is inversely proportional to the diffusion coefficient of *I* − 3 in the cells [28]. The decrease in *Z<sup>N</sup>* leads to an increase in the electrocatalytic activity of the CE.

**Table 2.** Electrochemical parameters of polyol-reduction Pt films at different reaction temperatures.


Figure 4c shows Tafel polarization plots of the symmetrical cells comprising the Pt CEs and *I* − 3 /*I* − electrolytes. The Tafel plot can be separated into three consecutive zones: polarization, Tafel, and limit diffusion zones. The Tafel and limit diffusion zones are valuable for obtaining both the limiting current density (*Jlim*) and current density (*J*0), which correlate with the electrocatalytic activity of the CEs [29]. The intersection of the cathodic branch and the equilibrium potential line can be considered *J*0, and thus the steep Tafel slope implies a large *J*0. Theoretically, *J*<sup>0</sup> can also be calculated using Equation (3):

$$J\_0 = \frac{RT}{nFR\_{ct}}\tag{3}$$

where *R* is the gas constant, *T* is the absolute temperature, *n* is the number of electrons participating in the electrochemical reduction of *I* − 3 , and *F* is Faraday's constant. In the limit diffusion zone, *Jlim* can be determined by the intersection of the cathodic branch and the *y*-axis. *Jlim* is directly proportional to a diffusion coefficient of *I* − 3 (*D*) at the same potential, and can be expressed as Equation (4):

$$J\_{\rm lim} = \frac{2neDCN\_A}{l} \tag{4}$$

where *e* is the elementary charge, *C* is the concentration of *I* − 3 , *N<sup>A</sup>* is the Avogadro constant, and *l* is the distance between two electrodes. Notably, the values of *J*<sup>0</sup> and *Jlim* followed the same trend observed in both CV and EIS analyses. Accordingly, the electrocatalytic activity of *I* − 3 /*I* <sup>−</sup> is as follows: PR-190 > PR-170 > PR-210 > PR-150 > PR-130.

The light absorption at the counter electrode should be minimized to improve light harvesting in the bifacial DSSCs. Figure 5a shows the transmittance spectra of the prepared Pt CEs and bare FTO glass. Polyol reduction manifested higher transparency compared to thermal decomposition in the whole visible light regime, which will be beneficial to bifacial applications. Unfortunately, the increase in reduction temperature resulted in the decrease in transmittance of PR CEs. We think that metallic Pt nanoparticles possibly have a negative effect on the light transparency. Figure 5b,c shows, respectively, the current density-voltage (J–V) characteristics and IPCE spectra of the DSSCs (5 µm-thick TiO2), which were illuminated from the TiO<sup>2</sup> photoanode side (i.e., front illumination). The photovoltaic parameters of the front- and rear-illuminated DSSCs are summarized in Table 3. The improved electrocatalytic activity of the PR–Pt CE resulted in better photovoltaic performance: power conversion efficiency (η) increased from 5.50% (PR-130) to 6.55% (PR-190). Also, PR-190 exhibited better photovoltaic performance than TD-425 because of the improvement of *R<sup>s</sup>* and *Z<sup>N</sup>* related to the electrocatalytic activity of the CE. It should be noted that all the DSSCs employing PR–Pt CEs maintained more than 76% of their front-illumination efficiency when lit from the Pt CE side (i.e., back illumination). The decrease of η in the back-illumination condition is related to the transmission losses due to the

Pt-based electrocatalyst and the *I* − 3 /*I* <sup>−</sup> electrolyte [6]. The ratio of the back-illumination efficiency to the front-illumination efficiency (η(*R*)) followed the trend of the transmittance of the PR–Pt CEs observed above. <sup>ଷ</sup>ܫ ‐ ‐ ିܫି/ ‐ *η*

*η* ‐ ‐ ‐

ൌ ܬ

ൌ ܬ

‐ ‐

‐

‐ *η* ‐

ଷܫ ି

ଷܫ ି

‐ ‐ ‐ ‐ ‐

ܴܶ ௧ܴܨ݊

ܰܥܦ2݊݁ ݈

> ଷܫ ି ܰ

‐

‐ μ ‐

ଷܫ ିܫି/

‐ μ ‐ ‐ ‐ **Figure 5.** (**a**) Transmittance spectra of the Pt counter electrodes, (**b**) J–V characteristic curves of the dye-sensitized solar cells (DSSCs) (5 µm-thick TiO<sup>2</sup> ) under one sun front illumination, and (**c**) their incident photon-to-current conversion efficiency (IPCE) spectra.


**Table 3.** Photovoltaic parameters of DSSCs (5 µm-thick TiO<sup>2</sup> ) with different Pt counter electrodes <sup>a</sup> .

<sup>a</sup> The parameters were obtained from at least five cells fabricated with each CE condition; each cell was measured five times.

To date, various transparent CEs have been developed for bifacial applications, such as photovoltaic windows and façades. Table 4 shows the photovoltaic performance of the bifacial DSSCs with platinum and non-platinum CE materials compared to our work. Although DSSC fabrication conditions (e.g., TiO<sup>2</sup> thickness, dyes, electrolytes) are not the same, our polyol reduction technique proved sufficient for fabricating bifacial DSSCs.


**Table 4.** Photovoltaic parameters of bifacial DSSCs with different CE materials.

<sup>a</sup> Transmittance was calculated from absorbance. <sup>b</sup> Our work.

#### **4. Conclusions**

In this work, highly transparent Pt CEs were prepared using the polyol reduction technique at low temperatures (<200 ◦C). Our facile and versatile technique provided better electrocatalytic activity and transparency than conventional thermal decomposition methods, and thus brought significant improvement to the photovoltaic performance of the bifacial DSSCs. In particular, the bifacial DSSC with PR-190 attained 6.55% for front illumination and 5.01% for back illumination, while that with TD had 6.42% for front illumination and 4.35% for back illumination. Because our process does not require an elevated temperature, we are currently exploring the fabrication of flexible bifacial DSSCs that employ polymeric substrates. In addition, further optimization (e.g., [Co(bpy)3] <sup>3</sup>+/2<sup>+</sup> electrolytes) should result in better bifacial DSSC performance.

**Author Contributions:** Conceptualization, A.G. and D.X.L.; methodology, A.G., S.K., and D.X.L.; formal analysis, A.G., S.K., B.M., and D.X.L.; investigation, A.G., D.X.L., and J.H., data curation, A.G. and D.X.L.; writing—original draft preparation, A.G., B.M., and J.H.; writing—review and editing, B.M. and J.H.; visualization, A.G. and J.H.; supervision, J.H.; project administration, J.H.; funding acquisition, J.H. All authors have read and agreed to the published version of the manuscript.

**Funding:** This research was supported by a National Research Foundation of Korea (NRF) grant funded by the Ministry of Science and ICT (MSIT) of Korea for the First-Mover Program for Accelerating Disruptive Technology Development (NF-2018M3C1B99088457), and the Energy Technology Development program through the Korea Institute of Energy Technology Evaluation and Planning (KETEP), funded by the Ministry of Trade, Industry and Energy (MOTIE) of Korea (No. 20193010014740 and No. 20193020010440).

**Acknowledgments:** We acknowledge the Chung-Ang University Research Scholarship in 2019.

**Conflicts of Interest:** The authors declare no conflict of interest.

#### **References**


© 2020 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

## *Article* **Arrays of Plasmonic Nanostructures for Absorption Enhancement in Perovskite Thin Films**

#### **Tianyi Shen , Qiwen Tan, Zhenghong Dai, Nitin P. Padture and Domenico Pacifici \***

School of Engineering, Brown University, 184 Hope Street, Providence, RI 02912, USA; tianyi\_shen@brown.edu (T.S.); qiwen\_tan@brown.edu (Q.T.); zhenghong\_dai@brown.edu (Z.D.); nitin\_padture@brown.edu (N.P.P.)

**\*** Correspondence: domenico\_pacifici@brown.edu

Received: 15 June 2020; Accepted: 7 July 2020; Published: 9 July 2020

**Abstract:** We report optical characterization and theoretical simulation of plasmon enhanced methylammonium lead iodide (MAPbI3) thin-film perovskite solar cells. Specifically, various nanohole (NH) and nanodisk (ND) arrays are fabricated on gold/MAPbI<sup>3</sup> interfaces. Significant absorption enhancement is observed experimentally in 75 nm and 110 nm-thick perovskite films. As a result of increased light scattering by plasmonic concentrators, the original Fabry–Pérot thin-film cavity effects are suppressed in specific structures. However, thanks to field enhancement caused by plasmonic resonances and in-plane interference of propagating surface plasmon polaritons, the calculated overall power conversion efficiency (PCE) of the solar cell is expected to increase by up to 45.5%, compared to its flat counterpart. The role of different geometry parameters of the nanostructure arrays is further investigated using three dimensional (3D) finite-difference time-domain (FDTD) simulations, which makes it possible to identify the physical origin of the absorption enhancement as a function of wavelength and design parameters. These findings demonstrate the potential of plasmonic nanostructures in further enhancing the performance of photovoltaic devices based on thin-film perovskites.

**Keywords:** perovskite solar cells; surface plasmon polaritons; plasmonic nanostructures; absorption enhancement; FDTD simulations

#### **1. Introduction**

Hybrid organic–inorganic perovskites have become one of the most popular photovoltaic materials due to their high absorption coefficient [1], long carrier diffusion length [2] as well as low-cost fabrication process [3–7], which make them excellent candidates for high-efficiency solar cells. As one of the most popular members of perovskite materials, methylammonium lead iodide or CH3NH3PbI<sup>3</sup> (MAPbI3) possesses such a high absorption coefficient that a 400 nm-thick film is sufficient to absorb most of the incident solar spectrum below its bandgap. Although the record power conversion efficiency (PCE) of a single-junction perovskite solar cell has reached values up to 25.2% [8], there is still much room to boost the PCE. Therefore, several methods have been proposed to improve the performance of perovskite solar cells, one of which involves the use of surface plasmons. Plasmonic effects could be employed to improve the performance of perovskite solar cells [9–15] by embedding nanoparticles [16–30] or plasmonic concentrators [31–35] that can increase absorption, especially near the optical band edge of the material. Previous theoretical research has shown that nanohole (NH) or nanodisk (ND) arrays embedded on gold electrodes with a thin layer of MAPbI<sup>3</sup> deposited on them can significantly improve the solar cell PCE by up to ∼10%. The reduced film thickness has the implied benefits of reducing the amount of toxic materials, improving electronic performance and enabling fabrication on flexible substrates [36].

In this work, we report the experimental fabrication, structural and optical characterization, as well as theoretical simulations of plasmon enhanced MAPbI<sup>3</sup> perovskite solar cell with NH and ND arrays on gold substrates, as a function of geometry parameters and varying thickness of MAPbI<sup>3</sup> deposited on top. Four geometry features, height *h*, diameter *D*, pitch *P* and MAPbI<sup>3</sup> thickness *t*, are systematically studied by three-dimensional finite-difference time-domain (3D FDTD) simulations to identify the physical effects responsible for the absorption and PCE enhancement.

#### **2. Materials and Methods**

An adhesion layer of 4 nm-thick titanium and a layer of 200 nm-thick gold was sequentially deposited on a 1 mm-thick quartz substrate using electron beam evaporation. Subsequently, NH or ND arrays of varying geometry parameters were fabricated on the thick gold film using focused ion beam (FIB) milling.

Following the FIB milling, the substrates were treated with UV-ozone for 45 min to enhance the wettability. The MAPbI<sup>3</sup> precursor solution was prepared by dissolving 159 mg of methylammonium iodide (Greatcell, Queanbeyan, Australia) and 461 mg of PbI<sup>2</sup> (Sigma-Aldrich, St. Louis, MO, USA) in 78 mg dimethyl sulfoxide (Sigma-Aldrich, St. Louis, MO, USA) and 1368 mg of N,N-dimethylformamide (Acros organics, NJ, USA) to obtain a 30 wt% solution. To deposit the MAPbI<sup>3</sup> layer, the solution was spin-coated at 4000 rpm for 30 s with an acceleration of 1300 rpm/s in a nitrogen-filled glove box. At the 10th second of spinning, 250 µL diethyl ether (Sigma-Aldrich, St. Louis, MO, USA) was dripped onto the substrate. The as-coated film was then annealed at 100 ◦C for 20 min to obtain the MAPbI<sup>3</sup> thin film. The film thickness can be controlled by adjusting the amount of N,N-dimethylformamide. Schematic illustrations of the fabricated NH and ND array coated with MAPbI<sup>3</sup> thin film are displayed in Figure 1a,b, respectively.

**Figure 1.** Schematic illustrations of methylammonium lead iodide (MAPbI3) thin-film over nanostructured gold (Au) surface with (**a**) NH or (**b**) ND array, respectively. *t* represents the perovskite film thickness on the top of gold surface. *P* represents the triangular array pitch, while *h* and *D* respectively represent the height and diameter of a plasmonic concentrator. A negative *h* indicated NH, while a positive *h* corresponded to ND.

The thickness of the synthesized MAPbI<sup>3</sup> films were characterized using a variable angle spectroscopic ellipsometer (J.M. Woollam, Lincoln, NE, USA, M-2000). The reflectance spectra were characterized using an inverted microscope (Nikon Instruments, Melville, NY, USA, Eclipse Ti) with a small numerical aperture objective (10×, NA = 0.3) to mimic normal incidence. A randomly polarized broadband light source was used to illuminate the sample. The reflected light from the samples was coupled into a spectrograph (Princeton Instruments, Acton, MA, USA, Acton SpectraPro SP-2300) and was detected using a charge-coupled device camera (Princeton Instruments, Acton, MA, USA, Pixis 100). A piece of flat silicon wafer was used as a calibration reference during the characterization.

Apart from experimentally characterizing the absorptance spectra, simulations are also performed using FDTD method. A commerical-grade 3D electromagnetic simulator from Lumerical Inc. (Vancouver, BC, Canada) was used [37]. Anti-symmetric and symmetric boundary conditions were used in the lateral directions while perfectly matched layer boundary conditions were used in the vertical directions. The real and imaginary parts of refractive indices for gold and MAPbI<sup>3</sup> were obtained from literature [38,39]. Considering the balance between computational cost and accuracy for different structures, the mesh size varied among 4 nm × 4 nm × 2 nm, 4 nm × 4 nm × 1 nm and 3 nm × 3 nm × 2 nm in the Cartesian coordinates.

#### **3. Results and Discussion**

Figure 2a,b,e,f show the tilted and top-view scanning electron microscope (SEM) micrographs of example NH and ND arrays prior to the MAPbI<sup>3</sup> synthesis. To investigate the absorption enhancement brought by the plasmonic nanostructure, the corresponding reflectance spectra *R*(*λ*) were measured, and the measured absorptance spectra *A*(*λ*) = 1 − *R*(*λ*) are reported in Figure 2c,g. Absorptance of nanostructured gold surface (black solid lines) shows a similar spectral shape as that of flat gold surface (black dashed lines) for both NH and ND arrays, while the NH array exhibits relatively higher absorption. This is because a portion of the scattered light by plasmonic concentrators is not collected by the finite NA of the objective lens, therefore a lower measured reflectance (i.e., higher absorption). By contrast, ND array in Figure 2e,f has smaller feature size and larger array pitch. Therefore, a smaller portion of light is scattered and a higher portion of reflected light is collected by the objective (i.e., lower absorption). The scattered light can be efficiently absorbed by adding a layer of MAPbI<sup>3</sup> on the top. After being coated with a 110 nm-thick MAPbI<sup>3</sup> film, the absorptance spectrum of the flat-interface structure (red dashed lines) shows a peak at wavelength *λ* = 526 nm. This results from the Fabry–Pérot cavity effect supported by the perovskite thin-film. The absorption decreases with increasing wavelength because the perovskite layer no longer supports any Fabry–Pérot cavity mode and the material has weaker absorbing property at longer wavelength. With NH or ND nanostructures (red solid lines), the absorption gets significantly enhanced at longer wavelengths, as is shown as red fillings in Figure 2d,h. For NH array, the spectral peak at *λ* = 526 nm gets weaker. This is because the structure disrupts the interference in the film [36]. In contrast, ND array still preserves that spectral peak due to its smaller feature size and lower nanostructure concentration. The absorptance spectra of perovskite/nanostructured gold are normalized to that of perovskite/flat gold and are displayed in Figure 2d,h. The blue filling at lower wavelengths represents the suppression of Fabry–Pérot cavity mode while the red filling represents the absorption enhancement brought by the plasmonic nanostructure. It is obvious that the example NH could not preserve the cavity effect as well as ND array (larger blue filling area), but the greater absorption enhancement by the plasmonic effects at longer wavelengths compensates for the loss at lower wavelengths (larger red filling area). As a comparison, for the example ND array, both blue and red filling areas are much smaller in Figure 2h, indicating negligible suppression at short wavelengths but also less significant enhancement at longer wavelengths.

A variety of nanostructured gold surfaces (NH or ND arrays) were fabricated based on optimal design considerations [36]. The top row of Figure 3 shows the SEM micrographs of the NH (left) and ND arrays (right). The measured and normalized absorptance spectra of these NH and ND arrays are shown in Figure 3a–d. For NH arrays (left column), as pitch *P* increases from 200 nm (circle symbols) to 250 nm (triangle symbols), the spectral peaks redshift from *λ* = 632 nm to *λ* = 666 nm. This results from the pitch-dependent collective surface plasmon polariton (SPP) resonances in the hexagonal lattice [36,40]. In addition, with hole depth increasing from 40 nm (solid symbols) to 70 nm (open symbols), the Fabry–Pérot cavity mode gets disrupted, resulting in a relatively lower absorption at *λ* = 526 nm, but there is more significant absorption enhancement at longer wavelengths. A similar behavior is also observed in ND arrays: structures suppressing the Fabry–Pérot thin-film interference effects show stronger absorption enhancement near the MAPbI<sup>3</sup> band edge.

**Figure 2.** (**a**,**b**,**e**,**f**) display the tilted-view (**a**,**e**) and top-view (**b**,**f**) SEM micrographs of the fabricated nanohole (NH) array with *P* = 250 nm, *D* = 200 nm, *h* = −70 nm (**a**,**b**) and nanodisk (ND) array with *P* = 400 nm, *D* = 100 nm, *h* = 40 nm (**e**,**f**). (**c**,**g**) The measured absorptance spectra *A* = 1 − *R* for nanostructured (solid lines) or flat (dashed lines) gold with (red lines) or without (black lines) 110nm-thick MAPbI<sup>3</sup> thin-film. (**d**,**h**) show the absorptance spectra for 110nm-thick MAPbI<sup>3</sup> thin-films on nanostructured gold normalized to that of equal-thickness MAPbI<sup>3</sup> on flat gold. The red filling (*A*/*Are f* > 1.0) represents absorption enhancement while the blue filling (*A*/*Are f* < 1.0) represents absorption suppression.

The absorption enhancement with different perovskite layer thicknesses is also investigated experimentally. MAPbI<sup>3</sup> layers with different thicknesses (*t* = 75, 110, 300 nm) were synthesized on the same nanostructure and the corresponding absorptance spectra were characterized and displayed in Figure 4. The spectral peaks and dips in the flat-interface cases (dashed lines) again result from the Fabry–Pérot cavity modes supported by different perovskite thicknesses at different wavelengths. The SEM micrographs of nanostructured surfaces are displayed in the top panels. Low-density and small-feature-size nanostructures and the corresponding absorptance spectra are shown in the left column. For different thicknesses *t*, these two structures can preserve the Fabry–Pérot cavity modes, as well as bring additional absorption enhancement. However, the absorption enhancement becomes much less significant as *t* increases to 300 nm, because most of the incident light is absorbed by the perovskite material before interacting with the plasmonic concentrators. For NH or ND arrays with larger feature sizes and higher concentration (right column), Fabry–Pérot cavity effects are significantly suppressed at the resonant wavelengths. In addition, both high-density and large-feature-size arrays exhibit stronger absorption enhancement at other wavelength ranges comparing to the low-density and small-feature-size arrays. Moreover, obvious absorption enhancement is observed even when *t* increases to 300 nm.

To further evaluate how the plasmonic effects and absorption enhancement will affect the overall performance of real photovoltaic devices, the PCE can be obtained using the measured absorptance spectra for different active layer thicknesses. With the detailed balance assumption, PCE can be expressed as

$$\text{PCE} = \frac{\int\_{\lambda < \lambda\_{\text{g}}} \text{AM1.5}(\lambda) \frac{2\pi\lambda}{\hbar c} A(\lambda) E\_{\text{\textdegree}} d\lambda}{1 \text{ sun}} \tag{1}$$

where *λ<sup>g</sup>* is the bandgap wavelength of MAPbI3, AM1.5(*λ*) is the wavelength-dependent air mass 1.5 solar radiation spectrum, *h*¯ is the reduced Planck constant and 1 *sun* is the 1000 W/m<sup>2</sup> incident solar power density [41]. The experimentally measured absorptance spectra includes the absorption in both gold and MAPbI3, but the portion absorbed by gold is relatively small according to previously reported study [36]. For NH and ND arrays in Figure 4 coated with multiple active layer thicknesses, the calculated PCE and PCE enhancement is shown in Figure 5.

**Figure 3.** Top-view SEM micrographs (top), experimental absorptance spectra (**a**,**c**) and normalized absorptance spectra (**b**,**d**) for 110nm-thick MAPbI<sup>3</sup> on flat (dashed lines) and nanostructured (solid lines with symbols) gold NH (left column) and ND (right column) arrays. The vertical dashed lines in (**a**) show the absorption spectral redshift as a function of pitch *P* (increasing from 200 nm to 250 nm).

**Figure 4.** Measured absorptance spectra of flat (dashed lines) and nanostructured (solid lines) gold surfaces coated with varying-thickness MAPbI<sup>3</sup> with *t* = 75 nm (**a**,**d**); 110 nm (**b**,**e**); 300 nm (**c**,**f**). The left (right) column corresponds to nanostructured gold surfaces with low (high) surface density of optical scatterers. The top panels show top-view SEM micrographs of gold surfaces with low (1–2) and high-concentration (3–4) ND or NH arrays.

The PCE of flat structure PCE*f lat* (t) (dashed line) is obtained using the multilayer interference model. For all three thicknesses (*t* = 75, 110 and 300 nm), all four nanostructure arrays exhibit PCE enhancement. Specifically, for active layer thickness *t* = 75 nm, high-density NH and ND arrays (red symbols) increase the PCE by 26.6% and 45.5%, respectively.

As a comparison, low-density nanostructure arrays (blue symbols) do not enhance the PCE as much. As the thickness increases to 300 nm, the PCE enhancement becomes much less significant due to the weaker plasmonic effect. The performance difference narrows between arrays of different concentration and feature sizes, enhancing the PCE by 3.7–7.0%.

As demonstrated in the experiment, plasmonic nanostructures with various geometry parameters exhibit different absorption enhancement behavior. Therefore, FDTD simulations were performed to further investigate the influence of different geometry parameters on the absorption enhancement.

**Figure 5.** (**a**) Power conversion efficiency (PCE) and (**b**) PCE enhancement estimated from experimental absorptance spectra for varying MAPbI<sup>3</sup> thickness values, *t* = 75, 110 and 300 nm with low (blue symbols) and high (red symbols) surface density of optical scatterers in NH (open symbols) and ND arrays (solid symbols). The dashed lines represent the PCE of flat-interface structures as a function of MAPbI<sup>3</sup> thickness. A strong PCE enhancement is clearly demonstrated for high-density nanostructures at *t* = 75 and 110 nm (red symbols).

Figure 6a–d display the simulated absorptance spectra of an example NH array (*P* = 250 nm, *D* = 200 nm, *h* = −70 nm) coated with 110 nm-thick MAPbI<sup>3</sup> film. In the left column, each of the four parameters are individually varied, meanwhile the other three parameters are kept fixed. When each parameter is varied, specific spectra with spectral peaks of interest are selected and respectively exhibited in Figure 6e–h and the insets show the electric field intensity distribution at the peak wavelengths.

First, the role of NH depth |*h*| is investigated. It is noticeable that the spectral peak at *λ* = 545 nm redshifts when |*h*| increases from 10 nm (black solid line) to 70 nm (yellow line). This peak results from the shallow hole, not disrupting the Fabry–Pérot effect (the bright horizontal band in inset (1)). When |*h*| increases to 50 nm, the peak redshifts due to a localized surface plasmon resonance (LSPR) arising inside the hole (inset (2) in Figure 6e). When |*h*| further increases to 150 nm, the constructive interference near the MAPbI3/gold interface is restored, but is not as strong (faint horizontal band in inset (3) in Figure 6e).

**Figure 6.** (**a**–**d**) Simulated absorptance spectra of representative NH arrays coated with a 110 nm-thick MAPbI<sup>3</sup> film. Each of the geometry parameters (i.e., height, *h*, diameter, *D*, pitch, *P* and MAPbI<sup>3</sup> thickness, *t*) is individually varied while keeping the others constant. (**e**–**h**) Representative absorptance spectra from (**a**–**d**) showing spectral peaks of interest. The color maps reported as insets show cross sections of electric field intensity simulated at each corresponding spectral peak wavelength by using 3D finite-difference time-domain (FDTD).

Figure 6b,c respectively demonstrate the absorptance spectra change with varying pitch *P* and diameter *D*. The black and purple dashed lines correspond to absorptance spectra of 110 nm and (110+|*h*|) nm-thick MAPbI<sup>3</sup> film on flat gold, respectively. With small feature sizes (small diameter *D*) or low NH concentration (large pitch *P*), the absorptance spectra are similar to that of 110 nm-thick MAPbI<sup>3</sup> on flat gold (black dashed lines). As the NH size or concentration increases, the absorptance spectral shape starts to shift towards that of 180 nm-thick MAPbI<sup>3</sup> on flat gold (purple dashed lines), which results from the Fabry–Pérot mode between the bottom of the holes and top of MAPbI<sup>3</sup> film. In addition, another peak near 630 nm shows up in some structures. Insets (6, 9) in Figure 6f,g show

that these spectral peaks attribute to the LSPR inside the NH structure. Moreover, an additional spectral peak is observed at *λ* = 736 nm. These peaks results from the LSPR between neighboring NH structures, as is shown in insets (7, 10) in Figure 6f,g. Figure 6d demonstrates the thickness dependence of the absorptance spectra. With increasing thickness, the multiple peaks corresponds to the LSPR inside the holes or between neighboring holes. In addition, the resonances gets weaker as *t* increases due to less light–matter interaction.

Likewise, the role of the ND geometry parameters in absorption enhancement is also investigated. Figure 7 shows the simulated absorptance spectra of an example ND array (*P* = 400 nm, *D* = 100 nm, *h* = 40 nm) coated with MAPbI<sup>3</sup> thin film of *t* = 110 nm.

**Figure 7.** (**a**–**d**) Simulated absorptance spectra of representative ND arrays coated with a 110 nm-thick MAPbI<sup>3</sup> film. The geometry parameters (i.e., height, *h*, diameter, *D*, pitch, *P* and MAPbI<sup>3</sup> thickness, *t*) are individually varied while the other three parameters are kept constant; (**e**–**h**) Representative absorptance spectra from (**a**–**d**) showing spectral peaks of interest. The color maps reported as insets show cross sections of electric field intensity simulated at each corresponding spectral peak wavelength by using 3D FDTD.

While varying *h* (Figure 7a), the absorptance spectra shape mainly follows that of flat-interface structure (dashed black line), showing peaks at *λ* = 530 nm, which corresponds to the Fabry–Pérot effect of 110 nm-thick MAPbI<sup>3</sup> on gold. Insets (1, 3) in Figure 7e show a discontinuous horizontal band, indicating that the existence of ND disrupts the cavity effect, but the spectral peaks still persists. Additionally, the ND also brings additional plasmonic resonances, as is shown in inset (3), enhancing the absorption with increasing height. Moreover, for a specific structure (*h* = 60 nm), another resonance is observed on the top of the disk structure, as is shown in inset (2) in Figure 7e.

The diameter *D* and array pitch *P* play a similar role in ND arrays: smaller feature sizes and low concentration preserve cavity effects, but also contribute to less plasmonic effect, as is shown in Figure 7b,c. For *D* = 320 nm (Figure 7f), specifically, the suppression of Fabry–Pérot effect is observed (inset (4) in Figure 7f). Similar to NH arrays, high-density (small *P*) ND arrays also suppress the cavity effects, while contributing to significant plasmonic effects (insets (7–9) in Figure 7g). In addition, multiple orders of localized standing-wave surface plasmonic resonances are observed on the top surface of the ND structure, as are reported in insets (5, 6) in Figure 7f.

Figure 7d shows the absorptance spectra of structures with varying MAPbI<sup>3</sup> thickness *t* coated on the ND arrays. Due to the small ND size, as well as the relatively large array pitch, the spectra peaks generally result from the Fabry–Pérot cavity effects (horizontal bands in insets (10–12) in Figure 7h). In spite of the existence of the ND structure, the modes still persist.

The PCE is calculated based on the previous simulation results, and is reported as black circle lines in Figure 8. The PCE of NH and ND structure are respectively displayed in the left and right column. As is expected, when coated with 110nm-thick MAPbI3, most of the structures with nanostructured surfaces exhibit significantly enhanced PCE (black circles) compared to their flat-interface counterparts (red dashed lines). It is worth noting that NH structures are adding extra perovskite material volume to the 110 nm-thick film, while ND structures have less perovskite material than the flat structure. To take this into account, two additional references are exhibited in Figure 8: (1) the blue dashed lines in Figure 8a–c and gray dashed lines in Figure 8e–g respectively show PCE*f lat*(t = 110±|*h*| nm) for NH and ND arrays; (2) the red triangle lines represent the PCE of flat structures with a combined thickness that accounts for the presence of two effective Fabry–Pérot cavities, which can be respectively expressed as

$$\text{PCE}\_{\text{combined}} = \frac{\pi D^2}{2\sqrt{3}P^2} \text{PCE}\_{flat} \text{(t=110 \pm |h|\text{nm}) + (1 - \frac{\pi D^2}{2\sqrt{3}P^2}) \text{PCE}\_{flat} \text{(t=110 \texttimes nm)} \tag{2}$$

where the plus (minus) sign refers to NH (ND) arrays. For most of the structures, the nanostructured-surface PCE is greater than PCE*combined*, which further proves that the plasmonic effects improve the solar cell performance. In addition, the difference between these two PCEs generally increases with increasing *D* and decreasing *P*, i.e., the cell performance benefits more from high-density and large-size nanostructures. In addition, with small thickness *t*, a larger portion of light penetrates the perovskite active layer, interacting strongly with the nanostructure array, and therefore also results in a larger difference.

**Figure 8.** Calculated PCE for NH (**a**–**d**) and ND (**e**–**h**) arrays as a function of varying geometry parameters (i.e., height, *h*, diameter, *D*, array pitch, *P* and MAPbI<sup>3</sup> thickness, *t*). Dashed lines show the PCE of flat structures with *t* = 110 nm or 110±|*h*| nm, while red triangle lines show the PCE of flat structures with a combined thickness that accounts for the presence of two effective Fabry–Pérot cavities in the nanostructured structure.

<sup>3</sup>

<sup>3</sup>

#### **4. Conclusions**

In conclusion, various plasmonic nanostructure arrays are experimentally fabricated on gold surfaces coated with perovskite films with varying thickness and their optical absorption properties are theoretically calculated by using 3D FDTD simulations. It is shown that, although thin-film interference effects can be negatively affected by the presence of nanoscatterers, overall, the plasmonic concentrators allow for significant absorption enhancement as the result of a combination of physical effects, i.e., (1) localized optical resonances (including surface plasmon resonances); (2) plasmonic modes in vertical cavities (such as the space within nanoholes); (3) in-plane constructive interference of surface plasmon polaritons that propagate along the metal/dielectric interface. Structures with higher surface density of nanoscattereres and with reduced film thickness generally show larger absorption and calculated PCE enhancements. The experimental results are supported by 3D FDTD simulations that help identify the individual physical effects and elucidate their interplay in redistributing the incident field intensity, which in turns determine the observed absorption and the calculated PCE enhancements.

**Author Contributions:** Data curation, T.S. and Q.T.; formal analysis, T.S. and Q.T.; resources, Z.D.; supervision, N.P.P. and D.P.; writing—original draft, T.S. and Q.T.; writing—review and editing, Z.D., N.P.P. and D.P. All authors have read and agreed to the published version of the manuscript.

**Funding:** This material is based upon work supported by the National Science Foundation under Grant No. OIA-1538893.

**Acknowledgments:** The authors thank Min Chen for fruitful discussions.

**Conflicts of Interest:** The authors declare no conflict of interest.

#### **References**


© 2020 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

#### *Article*
