Introduction of metallic nanoparticles that can generate the surface plasmon resonance (SPR) has been considered as a prominent option for enhancing the performance of polymer solar cells (PSCs), as the radiative scattering and field confinement by the SPR can extend the effective photon traveling length and manipulate the spatial absorption profile. Despite many successful efforts to favorably exploit metallic nanoparticles, further studies of their effects on the PSC performance have been demanded to achieve the full benefit from them. Here, we systematically investigate the optical and photovoltaic performances of PSCs with disorderly distributed silver nanoprisms embedded in the photoactive material. Due to the superior properties of the plasmonic scattering of this class of nanoparticles, a significant improvement of photon absorption is gained from the devices with silver nanoprisms, particularly in the wavelength range of substandard absorption property including the band-edge wavelengths. While such absorption improvement can be obviously reinforced as an increase in the particle density, its level becomes saturated and decayed eventually because of the concurrently promoted photon loss by plasmonic absorption. At the optimal configurations of silver nanoprisms for the productive light trapping effect, the incorporated PSC devices present a photocurrent of ∼17.76 mA/cm2 and a power conversion efficiency of ∼9.68%, where their net increase ratios are ∼10% and ∼13% compared to the reference PSC devices, respectively. Details of numerical modeling and experiments for both metal nanoprisms and PSC devices offer an optimum route to tailoring metallic nanoparticles for high-performance PSC systems.
Polymer solar cells (PSCs) have been a competent option for emerging photovoltaics due to the favorable properties including straightforward fabrication process, thin device structure, light module weight, and operation capability in the flexible configuration.1–4 Despite all these advantages, their insufficient performance for the solar spectrum absorption stemming from ultrathin (<200 nm) polymeric photoactive layers with a remarkably short exciton diffusion length (<100 nm) and a low carrier mobility (<10−3 cm2/Vs) has limited the power conversion efficiency (PCE) and hence the practical applications.5–7 Numerous studies have thus been conducted so far to overcome this hurdle by introducing photoactive polymers with high quantum yields (e.g., ternary donor/acceptor),8–11 material combinations with high open-circuit voltages (e.g., non-fullerene acceptors),8,12–14 and nanophotonic materials with improved absorption properties (e.g., plasmonic and cavity resonances).15–17 Among these approaches, exploiting nanophotonic materials has been particularly of interest, as they can not only improve the PSC efficiency with the pre-existing materials but also synergistically reinforce the PSCs with alternatively developed materials.17,18 Meanwhile, the use of nanoscale metallic materials that can excite the surface plasmon resonance (SPR) has been a popular choice for the nanophotonic management of PSCs. The radiative scattering and field confinement at SPR wavelengths could augment the photon traveling length (i.e., light trapping effect) and control the spatial profile of photon absorption, thereby boosting the photocurrent of PSCs.19–21 Due to highly compatible incorporation processes, the randomly distributed metal nanoparticles that support the localized SPRs have been widely employed in forms of those embedded in photoactive and/or interface layers of PSCs.20,22–24 Although abundant reports have consistently demonstrated possible routes to handle metal nanoparticles (e.g., constituent material classes, particle shapes, and distributions) for enhancing the PSC performance,25–29 room for further enhancement still remains as they have not been fully studied yet, especially the effects of resonance wavelengths and distribution densities of particles embedded in PSCs. In this regard, herein we systematically investigate the optical and photovoltaic performances of PSCs incorporated with metal nanoparticles to obtain the maximum performance enhancement. With silver (Ag)-based nanoparticles that possess the superior efficiency of radiative scattering than other metal particle options,30–32 a particle of triangular prism shape is chosen to produce the localized SPRs at wavelengths near the band-edge of photoactive materials, considering significant degradation of the absorption property in this wavelength range.33 Through the numerical optical modeling of PSCs with the disorderly distributed Ag nanoprisms or nanospheres in the photoactive layer, the effects of geometric parameters and densities for the embedded nanoparticles are quantitatively studied to identify the optimal SPR configurations such as resonance wavelengths and scattering/absorption properties. The experimental characterizations for PSC devices with the embedded Ag nanoprisms advocate the underlying physics of the proposed design systems.
II. RESULTS AND DISCUSSION
A. Optical characteristics of synthesized silver nanoprisms
To achieve favorable effects from metal nanoparticles embedded in PSCs, the plasmonic resonance of nanoparticles preferentially needs to have the optical properties of high scattering coefficient for an effective increase in the photon traveling length and of a low absorption coefficient for suppression of photon absorption loss. From this point of view, a triangular Ag prism is employed as a nanoparticle configuration in this device system due to its remarkable scattering property.31 To prepare Ag nanoprisms (NPMs) that would be buried in the photoactive layer, we synthesized them through the method described in the literature.34 Briefly, the synthesis process began with the formation of Ag seeds. A solution of Ag seeds with a sphere shape and a diameter of 10 nm was produced by dropping 0.5 mM silver nitrate (AgNO3, 5 ml) aqueous solution into an aqueous mixture solution of 2.5 mM trisodium citrate (TSC, 5 ml), 10 mM sodium borohydride (NaBH4, 0.3 ml), and poly(sodium 4-styrenesulfonate) (PSSS, 500 mg/l, 0.25 ml). The growth of Ag NPMs from Ag sphere seeds was implemented by dropping another 0.5 mM AgNO3 (3 ml) aqueous solution into the Ag seed solution additionally mixed with 10 mM ascorbic acid (0.075 ml) and deionized (DI) water (5 ml). The size of grown Ag NPMs was determined by the amount of the Ag seed solution under a fixed amount of the ascorbic acid solution, where NPM side lengths (Ls) of ∼20, ∼60, and ∼80 nm were observed when using 0.3, 0.12, and 0.04 ml Ag seed solutions, respectively. It is noteworthy that the thickness (t) of the grown Ag NPMs rather followed the size of the Ag sphere seeds (∼10 nm), regardless of their side length. For blending the as-grown Ag NPMs with the photoactive material dissolved in a nonpolar solvent, the water-based polar solvent of Ag NPMs had to be replaced with an identical nonpolar solvent. To implement the solvent substitution without severe aggregation, the Ag NPMs were capped with a low-surface-energy material of polyvinylpyrrolidone (PVP) before extracting the precipitated Ag NPMs in the centrifuged solution. Capping of Ag NPMs with PVP was simply carried out by mixing the as-grown Ag NPM solution with the PVP-dissolved ethanol.35 Figure 1(a) presents a scanning transmission electron microscope (STEM) image of a drop-casted sample of the PVP-capped Ag NPMs dispersed in a photoactive material solvent of chlorobenzene and 1,8-diiodoctane mixture (97:3 by volume) [inset of Fig. 1(a)] in which adequately dispersed Ag NPMs could be observed.
As the Ag NPMs are disorderly distributed in the photoactive layer, the localized SPRs excited along three-different axes of NPMs can collectively affect the device performance. Figure 1(b) shows the calculated scattering cross section (σsca) spectra of the Ag NPM (Ls = 20 nm and t = 10 nm) in a homogeneous dielectric medium (n = 1.34) for incident light polarizations parallel to the NPM side (x-polarization), vertical to the NPM side (y-polarization), and orthogonal to the NPM plate (z-polarization). Here, the calculation was performed with a commercial software using the finite-difference time domain (FDTD) method (FDTD Solutions, Lumerial). Depending on the polarization directions, the σsca peak shifted from ∼400 nm to ∼615 nm due to the change in the maximum oscillation length. The σsca peaks at ∼595 (x-polarization) and ∼615 nm (y-polarization) can be regarded as the main resonances of the NPM because the localized SPRs for these polarizations were generated due to the unique NPM shape, while the σsca peak at ∼400 nm (z-polarization) indicates the rather parasitic resonance arising from the finite thickness of the NPM. The main resonances are unquestionably red-shifted if the side length of the Ag NPM is increased, as described in Fig. 1(c), where the normalized extinction (σext,n) spectra from the experiments and the normalized scattering (σsca,n) spectra obtained from averaging the calculated σsca spectra along three-different axes are provided for Ag NPMs with various side lengths (Ls = 20 nm, 61 nm, and 77 nm) and a fixed thickness (t = 10 nm) as well as for Ag nanosphere (NS) with a diameter of 10 nm in water/ethanol surroundings (n = ∼1.34). It can be recognized that a wavelength of the parasitic resonance (i.e., from the z-polarization) of the Ag NPM is similar to the wavelength of the localized SPR of the Ag NS due to their analogous oscillation lengths, and the experimental σext,n spectra well correspond to the calculated σsca,n spectra in terms of peak positions, albeit discrepancies of peak broadness provoked by the non-uniform size and shape distribution. Meanwhile, since the Ag NPMs would be embedded in the photoactive material with a high refractive index (n = ∼1.9) when applied to PSCs, the resonance manner of Ag NPMs in such a medium needs to be identified. Figure 1(d) indicates the σsca,n spectra for Ag NPMs and NS appeared in Fig. 1(c) but existed in a lossless higher-index medium with n of 1.9. As expected from the literature,36,37 an increase in the refractive index of the surrounding medium led to a red-shift of the localized SPRs for all Ag particle cases. In this high-index medium, the Ag NPM with a Ls value of 61 and 77 nm exhibited the peak resonance at ∼1000 and ∼1100 nm, respectively, which can hardly offer positive effects on the PSC device property due to the resonances below the bandgap when selecting poly[4,8-bis[5-(2-ethylhexyl)thiophen-2-yl]benzo[1,2-b:4,5-b′]dithiophene-2,6-diyl-alt-3-fluoro-2-[(2-ethylhexyl)carbonyl]thieno[3,4-b]thiophene-4,6-diyl] (PTB7-Th):(6,6)-phenyl C71 butyric acid methyl ester (PC71BM) as a photoactive material combination (bandgap energy = ∼1.55 eV).6 On the other hand, the peak resonances of the AgNS with a diameter of 10 nm and the Ag NPM with a Ls value of 20 nm emerged at the mid-band (∼480 nm) and band-edge (∼800 nm) wavelengths of these photoactive materials, respectively, indicating that the meaningful contribution to the PSCs can be expected from these particles. It needs to be mentioned that low levels of scattering-to-absorption cross section ratio (σsca/σabs) at small Ag NPMs [inset of Fig. 1(d)] become a handicap when applied to the PSCs due to the limited scattering property.38
B. Polymer solar cells with embedded silver nanoprisms
To understand the optical effects of Ag NPMs embedded in the photoactive layer, the FDTD-based numerical simulation was conducted for the PSCs with Ag NPMs over manageable NPM parameters such as the side length and particle density (the number of particles per unit volume). Figure 2(a) schematically depicts a modeled PSC structure in this simulation, where the fully disordered Ag NPMs are inserted into a type of an inverted PSC comprising indium tin oxide (ITO) (150 nm), poly[(9,9-bis(3′-(N,N-dimethylamino)propyl)-2,7-fluorene)-alt-2,7-(9,9-dioctylfluorene)] (PFN) (10 nm), PTB7-Th:PC71BM (100 nm), molybdenum trioxide (MoO3) (10 nm), and Ag (100 nm). The Ag NPM contribution to the photon absorption of the PSC device was quantitatively assessed by introducing the integrated solar flux absorption (Sabs) in the photoactive layer (i.e., PTB7-Th:PC71BM), as it directly specifies how large an amount of the photocurrent can be generated against the Shockley–Queisser current limit. The Sabs is calculated as39–42
where λ, h, c, I1.5G, and Aact are wavelength, Planck’s constant, speed of light, AM 1.5G standard solar irradiance, and photon absorption in the photoactive layer, respectively. The integral limit values of λo and λf were determined as 350 nm and 850 nm, respectively, in consideration of the absorption wavelength band of the photoactive layer. Here, Aact was yielded by integrating the spatial profile of the absorbed power density (pabs) normalized by the source power (Ps) over the photoactive layer space in the simulation,40,42
where ε and are material permittivity and electric field intensity, respectively. Figure 2(b) presents the results of the calculated Sabs values for devices with various Ag NPMs (Ls = 20–50 nm) and Ag NSs (diameter = 10 nm) as an increase in the particle density (ρNPM for the Ag NPM density and ρNS for the Ag NS density) until its modeling limit of preventing the overlap distribution. The Sabs values for all Ag NPM cases went up rapidly in the low ρNPM range owing to the expanded photon traveling length by the SPR scattering but became saturated afterward as a consequence of simultaneously increased photon loss by the SPR absorption of Ag NPMs. When comparing the maximum Sabs values achieved at high ρNPM values, the case of Ag NPMs with a Ls of 20 nm shows a superior absorption property than the other Ag NPM cases of which Sabs maximum is even considerably higher than that of the Ag NS case. From the spectra of photon absorption in the photoactive material (Aact), in Ag NPMs (ANPM, calculated from an equation similar to Eq. (2) but with integration over the Ag NPM space), and in both the photoactive material and Ag NPMs (Aact+NPM = Aact + ANPM) for the case of Ag NPMs with 20 nm Ls at various ρNPM values (5–25 µm−2 nm−1) [Fig. 2(c)], it can be identified that its increased Sabs value at high ρNPM values was provoked by the enhanced Aact in the mid- and long-wavelength ranges. The markedly raised Aact levels at high ρNPM values near the band-edge wavelength were attributed to the effectively boosted photon traveling path by the strong SPRs under the extremely low absorption coefficient regime (∼0.25 µm−1 at ∼800 nm), while noticeably increased Aact levels around the 600 nm mid-band wavelength were from the tail of band-edge SPRs. The ANPM levels that signify the photon loss were also augmented monotonically, thereby enabling the Aact+NPM levels to reach the upper limit of absorption. In the meantime, the tendency of the Sabs improvement along an increase in the particle density can be reciprocally changed at density levels over the optimum because the photon loss by the SPR absorption becomes dominant. This behavior is clearly observed in the Ag NS case [Fig. 2(b)]. As shown in the similar absorption spectra for the Ag NS cases (i.e., Aact, absorption in Ag NSs; ANS, Aact+NS = Aact + ANS) in Fig. 2(d), the Aact level was marginally promoted up to a limited value as increasing the ρNS by the SPR scattering at wavelengths (∼550 nm, ∼800 nm) in the comparatively low absorption coefficient regime; however, it was diminished gradually elsewhere due to the substantial photon loss by the SPR absorption (i.e., ANS), albeit an increase in the total absorption Aact+NS. Figure 2(e) represents the absorbed solar flux spectra in the photoactive layer (Aflux,act) for PSC devices with the optimal Ag NPMs (Ls = 20 nm, ρNPM = 25 µm−2 nm−1) and the Ag NSs (ρNS = 60 µm−2 nm−1) and without any particles under the AM 1.5G illumination. It can be consistently figured out that a relatively large amount of incident solar photons are absorbed at the Ag NPM devices. As described in contour plots of the normalized absorbed power density (pabs/Ps) of PSC devices with and without the optimal Ag NPMs in Fig. 2(f), the notably intensified pabs/Ps occurs in the space adjacent to the Ag NPMs, which is valuable, in particular, for the photon absorption in the band-edge wavelength (∼800 nm).
After these optical simulation analyses, the photovoltaic performance of the fabricated PSC devices with the optimally designed Ag NPMs (Ls = 20 nm) was characterized to experimentally investigate their effects on devices. The experimental samples of inverted, bottom-illuminated PSCs consisted of glass (1 mm, substrate)/ITO (150 nm, transparent cathode)/PFN (10 nm, electron injection layer)/PTB7-Th:PC71BM (100 nm, 1:1.5 in weight, photoactive layer)/MoO3 (10 nm, hole injection layer)/Ag (100 nm, opaque anode). The insertion of Ag NPMs into the PTB7-Th:PC71BM photoactive layer was executed by mixing co-solvent (mixture of chlorobenzene/1,8-diiodoctane) solutions of the Ag NPM-dispersed solution and the PTB7-Th/PC71BM blend solution by 8% in volume [Fig. 3(a)], where the density of the Ag NPMs in the PTB7-Th:PC71BM was controlled by changing the Ag NPM content (4–12 wt. %) in the Ag NPM-dispersed solution before mixing it with the PTB7-Th/PC71BM blend solution. Figures 3(b) and 3(c) and Table I provide the representative current density (J)–voltage (V) curves and corresponding photovoltaic performance parameters, respectively, for the PSC devices with and without the Ag NPMs, measured under the simulated AM1.5G illumination (100 mW/cm2) at room temperature. The improved short-circuit current density (Jsc) of devices was definitely achieved by adding the Ag NPMs. As expected from the optical simulation results, an increase in the Ag NPM density led to a more significant improvement of the Jsc, as examined such that the Ag NPM-embedded devices using the Ag NPM solutions of 4 wt. % and 8 wt. % exhibited Jsc levels of ∼16.87 mA/cm2 and ∼17.76 mA/cm2 and hence PCE levels of ∼9.04% and ∼9.68%, respectively, while the reference devices without Ag NPMs showed a Jsc level of ∼16.16 mA/cm2 (PCE = ∼8.56%). However, the device performance became degraded (Jsc = ∼17.28 mA/cm2, PCE = ∼9.12%) if the Ag NPM density by using the 12 wt. % Ag NPM solution was further increased. This deterioration of devices at the high Ag NPM density seems to be evoked from the consequentially strengthened effects of the exciton quenching appearing in the vicinity of Ag NPMs by non-radiative decay through the near-field coupling between the exciton and the surface plasmon mode of Ag NPMs (i.e., a decrease in the Jsc)43–45 [Fig. 3(d)] and/or the incomplete formation of the bulk heterojunction layer (i.e., reduction in the fill factor).46,47 Table I summarizes the photovoltaic performance of the present device in comparison with the similar configurations of plasmonic PSC devices reported in the literature.20,26,48–50 Although the aid of the metal nanoparticle treatment (e.g., capping with dielectric, clustering) exhibited a considerable performance enhancement, the most favorable performance at devices only with the inserted metal nanoparticles was observed in the present device in terms of both PCE and its enhancement levels. Meanwhile, it is noteworthy that a format of the Ag NPM-incorporated photosynthetic protein interlayer can be another route to augment the performance of PTB7-Th:PC71BM devices due to their synergistic effect of photon absorption enhancement.50
|Photoactive material .||Particle configuration .||PCE (%) .||Enhancement (%)a .||Reference .|
|Au prism clusters||9.48||14.4|
|PTB7-Th:PC71BM||Ag prisms||9.68||13.1||This work|
|PTB7-Th:PC71BM||Ag prism/LHCII hybridsb||10.57||17.1||50|
|Photoactive material .||Particle configuration .||PCE (%) .||Enhancement (%)a .||Reference .|
|Au prism clusters||9.48||14.4|
|PTB7-Th:PC71BM||Ag prisms||9.68||13.1||This work|
|PTB7-Th:PC71BM||Ag prism/LHCII hybridsb||10.57||17.1||50|
The enhancement of the PCE compared to the control device without any particles.
Interlayer between the photoactive layer and the electron transport layer.
To implement further characterizations of the experimentally optimal PSCs with Ag NPMs, we measured the total reflection (R) and external quantum efficiency (EQE) spectra for the PSC devices with and without the Ag NPMs (Ls = 20 nm, 8 wt. % Ag NPM solution) and hence derived their total absorption (At = 1 − R) and internal quantum efficiency (IQE = EQE/At) spectra. As shown in Fig. 4(a), inserting the Ag NPMs into the devices led to the improvement of the At levels in the mid- and long-wavelength ranges (i.e., λ > ∼470 nm), which was reasonably consistent with the optical simulation results described earlier. An amount of the At improvement at the band-edge wavelengths was not significant than expected from the simulation; this discrepancy was highly likely to arise from the insufficient density of the Ag NPMs and a slight blue-shift of the SPR by the limited accuracy of the modeled refractive index of the PTB7-Th/PC71BM. Similarly to the behavior of the At improvement, the EQE levels were also improved by adding the Ag NPMs at wavelengths between ∼470 and ∼750 nm, as presented in Fig. 4(b). Such analogous behaviors of the At and EQE improvements can translate into the almost identical IQE spectra, regardless of the existence of the Ag NPMs, implying that the remarkable Jsc enhancement from the embedded Ag NPMs was a result of the advanced optical properties, not of any changes in the electrical properties.
In summary, we systematically studied the optical and photovoltaic performances of PSCs with the randomly distributed Ag NPMs that were embedded in the PTB7-Th:PC71BM photoactive layer. Owing to the exceptional scattering properties of the localized SPRs at Ag NPMs, the considerable effect of the photon traveling length extension could be attained from the inserted Ag NPMs. As such, the effect from the plasmonic scattering was particularly beneficial at wavelengths near the band-edge due to the inferior absorption properties of the photoactive material by its comparatively low absorption coefficient, adding the Ag NPMs that were designed for producing the band-edge resonances offered a significant improvement of the photon absorption performance to the PSC devices. An increase in the Ag NPM density provoked further advancement of the photon absorption until saturated by the simultaneously increased SPR absorption loss. From the optical gain of Ag NPMs, Jsc of ∼17.76 mA/cm2 and a PCE of ∼9.68% were achieved at devices with optimal Ag NPMs, which were ∼10% and ∼13% higher compared to the reference devices without any particles, respectively. We anticipate that the outcome of this study can contribute to breaking the barrier of the PSC performance.
A. Preparation of silver nanoprisms
Ag nanoparticles with the shape of triangular prism were grown through the seed-mediated growth method.34 To prepare the Ag sphere seed solution, an AgNO3 aqueous solution (0.5 mM, 5 ml) was dropped into a mixed aqueous solution of TSC (2.5 mM, 5 ml), NaBH4 (10 mM, 0.3 ml), and PSSS (500 mg/l, 0.25 ml) at the rate of 2 ml/min using a burette, followed by stirring for 20 min at room temperature. Subsequently, another AgNO3 aqueous solution was added into a mixture of the pre-synthesized Ag seed solution (0.3 ml, 0.12 ml, and 0.04 ml), aqueous ascorbic acid solution (10 mM, 0.075 ml), and DI water (5 ml) at the rate of 1 ml/min to grow triangular prisms (side length = 20 nm, 61 nm, 77 nm). The as-grown Ag NPM solution was mixed with the PVP-dissolved ethanol to perform capping of Ag NPMs with PVP. After three times of centrifugation, the ethanol/DI water solvent of the resultant solution was evaporated in a vacuum oven at 120 °C for 1 h. The precipitated Ag NPMs were then dispersed in an added solvent of chlorobenzene and 1,8-diiodoctane (97:3 by volume) to produce the PVP-capped Ag NPM solutions (4, 8, and 12 wt. %).
B. Fabrication of PSCs with embedded silver nanoprisms
On a cleaned ITO (150 nm)/glass (1 mm) substrate with acetone, isopropyl alcohol, and DI water, a blend solution of PFN (2 mg, 1-Material) dissolved in a mixture of methanol (10 ml) and acetic acid (20 µl) was spin-coated (2000 rpm, 60 s) and dried at room temperature in a nitrogen-filled glove box. Subsequently, the prepared mixture solution by blending a photoactive solution of PTB7-Th (8 mg/ml, Lumtec)/PC71BM (12 mg/ml, Nano-C) dissolved in a solvent of chlorobenzene/1,8-diiodoctane (97:3 by volume, Sigma Aldrich) with the PVP-capped Ag NPMs dispersed in the identical solvent (4, 8, and 12 wt. %) was spin-coated again (1000 rpm, 120 s) and dried at room temperature for 1 h. MoO3 (10 nm) and Ag (100 nm) were successively deposited by a thermal evaporation system as a final step.
C. Numerical FDTD modeling of PSCs with silver nanoprisms
The optical characteristics of Ag nanoparticles and PSCs were simulated by a commercial software on the basis of the FDTD method (FDTD Solutions, Lumerial). The σsca spectra of Ag nanoparticles were computed for a single particle defined in a 3D simulation volume, where a source function of “total-field scattered-field” and the surrounding “power” monitors were employed for capturing the scattering field intensity. To calculate the absorption in a given material at specific wavelengths, the E-field profiles under a light source of normally incident, continuous plane-wave with a broad Gaussian frequency spectrum (230–1000 THz) were simulated with a modeled structure in which randomly distributed (e.g., position, rotation angle) Ag NPMs with a minimum particle-to-particle distance of their side length were employed. The measured values of the refractive indices for most polymer materials (e.g., PTB7-Th/PC71BM and PFN) were used, as described in our previous report,42 while a refractive index of Ag was referred to the literature.51
D. Optical and photovoltaic characterizations
The current–voltage characteristics were measured using a source meter (Keithley 2400, Tektronix) and a solar simulator (K201 LAB55, McScience) under a calibrated AM 1.5G solar spectrum (100 mW/cm2). Measurements of the reflection and EQE spectra for PSC samples were conducted with a spectral measurement system (K3100 IQX, McScience), where a silver reflector was applied as a 100% reflection standard.
H.-E.C. and S.H.C. contributed equally to this work.
This work was supported by the National Research Foundation of Korea (NRF) funded by the Ministry of Science, ICT, under Grant Nos. NRF-2017M1A3A3A02016782 and NRF-2019R1C1C1008201.