In Depth Analysis of Photovoltaic Performance of Chlorophyll Derivative-Based "All Solid-State" Dye-Sensitized Solar Cells.

Chlorophyll a derivatives were integrated in "all solid-state" dye sensitized solar cells (DSSCs) with a mesoporous TiO2 electrode and 2',2',7,7'-tetrakis[N,N-di(4-methoxyphenyl)amino]-9,9'-spirobifluorene as the hole-transport material. Despite modest power conversion efficiencies (PCEs) between 0.26% and 0.55% achieved for these chlorin dyes, a systematic investigation was carried out in order to elucidate their main limitations. To provide a comprehensive understanding of the parameters (structure, nature of the anchoring group, adsorption …) and their relationship with the PCEs, density functional theory (DFT) calculations, optical and photovoltaic studies and electron paramagnetic resonance analysis exploiting the 4-carboxy-TEMPO spin probe were combined. The recombination kinetics, the frontier molecular orbitals of these DSSCs and the adsorption efficiency onto the TiO2 surface were found to be the key parameters that govern their photovoltaic response.


Introduction
Fossil fuel resource exhaustion, increasing energy demands and environmental problems have triggered an intensification of research for sustainable energy sources, which lead to the development of photovoltaic devices, fuel cells, wind turbines, etc. Photovoltaic conversion of solar energy is considered to be one of the most significant ways of addressing the growing global energy crisis [1]. In this context, utilizing biomass resources for the conversion of solar energy has been receiving increasing attention from industry, academia and governments and will become a major challenge of our future [2][3][4]. Biomass has a great potential as a renewable feedstock for producing solar energy [5,6]. Indeed, protein-pigment complexes in plants, green algae, and cyanobacteria convert solar energy into chemical energy through oxygenic photosynthesis with high efficiency [7][8][9]. Calvin showed through a model based on a photosynthetic electron transport system containing membranes, carotenoids and pigments that electric energy can be obtained by conversion of sunlight [10]. A similar approach based on an artificial photosynthesis led in 1991 to dye-sensitized solar cells (DSSCs, also known as Grätzel cells) [11]. A DSSC consists of a mesoporous nanocrystalline n-type semiconductor (typically TiO 2 ) sensitized with a dye, deposited onto an anode and immersed in redox active electrolyte (generally triiodide/iodide), completed by a counter-electrode (cathode) [12]. Upon illumination, dye molecules anchored onto the TiO 2 surface absorb incoming photons, allowing the photoinduced electron-transfer from their excited state into the TiO 2 conduction band. Then, the electrons are transferred to the counter electrode, thus creating a current. The redox electrolyte reduces the oxidized dye molecules back to their ground state to enable continuous electron production. Even if power conversion efficiencies (PCEs) above 13% have been reached for liquid-electrolyte DSSCs [13,14], improvements are still required in order to be commercially viable.
First, there is a crucial need for solid-state approaches since the use of liquid electrolytes is a major hurdle for commercialization by industry due to the corrosive nature of the commonly used I 3 − /I − redox shuttle [15,16]. DSSCs incorporating solid-state hole transport materials (HTMs) have been actively investigated [17][18][19], and decent efficiencies close to 12% have been already achieved by molecular and inorganic HTMs combined with state-of-the-art organic dyes [20,21]. Additionally, solid-state DSSCs (ssDSSCs) led to a major breakthrough by using hybrid lead halide perovskite sensitizers such as CH 3 NH 3 PbI x Cl 3-x , enabling the advent of perovskite solar cells with efficiencies now above 25% [22]. However, important issues concerning the device stability and the toxicity of the components still remain. Secondly, it is highly required to use efficient and cost-effective dyes in place of expensive and scarce ruthenium complexes [23,24] or elaborated organic dyes requiring numerous synthetic steps [15,25]. In contrast, natural organic dyes and their derivatives are promising candidates for developing DSSCs due to their low cost, abundance and non-toxicity [5]. An important biomass resource developed in the south of France is spirulina, a green algae mainly used as food supplement. One of the most valuable compound that can be extracted from spirulina is chlorophyll a, (Chl a-12 g/kg) [26] a photo-and electroactive chlorin-type macrocycle involved in photosynthesis, because it efficiently absorbs light in the 400-450 nm and the 600-700 nm regions and up to 800 nm depending of its aggregation state [27]. The use of chlorophyll derivatives as sensitizers has been investigated in liquid electrolyte-based DSSCs leading to energy-to-electricity conversion efficiency (η) up to around 8% [28][29][30]. Concerning solid-state DSSC, there are only a few reports in the literature describing the use of chlorophyll a derivatives as dye [31][32][33].
Herein, we report the preparation of all-solid state DSSCs using a series of Chl a derivatives ( Figure 1) as sensitizers with different anchoring groups to graft to the TiO 2 surface and 2 ,2 ,7,7 -tetrakis[N,N-di(4-methoxyphenyl)amino]-9,9 -spirobifluorene (spiro-OMeTAD) as HTM. The choice of spiro-OMeTAD HTM was motivated by its facile implementation and high performance rendering it particularly suitable for testing DSSCs. To elucidate the relationship between the molecular structure of the dye and the photovoltaic efficiency, UV-Vis absorption spectroscopy, density functional theory (DFT) calculations, incident photon-to-current efficiency (IPCE), transient photovoltage decay measurements and EPR analysis exploiting the spin probe 4-carboxy-2,2,6,6-tetramethylpiperidine 1-oxyl (4-carboxy-TEMPO, indicated as 4C-T) have been carried out.

Synthesis of the Dyes Derived from Chlorophyll a
The chemical structures of chlorophyll a-based dye sensitizers depicted in Scheme 1 were obtained from methyl pyropheophorbide-a (MPPa). MPPa was prepared in four steps: (i) extraction of Chl a from Molecules 2020, 25,198 3 of 20 spirulina biomass with ethanol, (ii) demetallation of Chl a, (iii) acid-catalyzed transesterification of the phytyl chain in methanol, and (iv) decarboxylation of the ester moiety located on the 13-position with 2,4,6-collidine [34,35]. Three families of corresponding free-base and zinc chlorins can be distinguished depending on the localization of the anchoring group. All the chlorins have one carboxylic acid for grafting onto the TiO 2 surface. Chlorins M1-M2 exhibit a non-conjugated alkyl anchoring group while the others possess either a conjugated vinyl anchoring group (M3-M4) or a cyanoacrylic group along the Q y axis of the chlorin skeleton (M5-M6). M1 was synthesized through the acid hydrolysis of MPPa with hydrochloric acid in acetone. Methyl pyropheophorbide-d (MPPd) was prepared by cleavage of the alkene in the presence of osmium tetroxide (OsO4) and sodium periodate (NaIO4) [35]. Trans-3 2 -carboxy-pyropheophorbide-a (M3) was synthesized in two steps: (a) a Wittig reaction with MPPd and (b) the hydrolysis with TFA of the corresponding Chlorin-1. A Knoevenagel reaction of MPPd with cyanoacetic acid in piperidine/THF yielded chlorin M5. Finally, metalation reactions of the free base chlorins M1, M3, and M5 with Zn(OAc)2 in MeOH/THF afforded the corresponding zinc complexes M2, M4, and M6. M1 was synthesized through the acid hydrolysis of MPPa with hydrochloric acid in acetone. Methyl pyropheophorbide-d (MPPd) was prepared by cleavage of the alkene in the presence of osmium tetroxide (OsO 4 ) and sodium periodate (NaIO 4 ) [35]. Trans-3 2 -carboxy-pyropheophorbide-a (M3) was synthesized in two steps: (a) a Wittig reaction with MPPd and (b) the hydrolysis with TFA of the corresponding Chlorin-1. A Knoevenagel reaction of MPPd with cyanoacetic acid in piperidine/THF yielded chlorin M5. Finally, metalation reactions of the free base chlorins M1, M3, and M5 with Zn(OAc) 2 in MeOH/THF afforded the corresponding zinc complexes M2, M4, and M6.

Optical Properties of Chlorin Derivatives
The UV-Visible absorption spectra of chlorins M1-M6 both in THF and grafted on the TiO 2 surface through carboxylate bonding are presented in Figure 1. Table 1 summarizes their characteristics in THF and on TiO 2 . The shape of the UV-Visible absorption spectra in THF are relatively similar, with five observed bands when going from blue to red: Soret, Q x (0,1), Q x (0,0), Q y (0,1), and Q y (0,0). The Q y peaks position of the vinyl-substituted chlorins M3-M6 are redshifted by about 10-20 nm in comparison with chlorins M1-M2 due to the introduction of electron-withdrawing groups on the vinyl moiety at the C3 position. The bathochromic shifts of the Q y peaks of chlorins M5-M6 are larger than those observed for M3 (7-15 nm) and M4 (1-4 nm) because of the higher electron-withdrawing ability of the cyano group. Upon zinc metalation, the Soret absorption bands are redshifted (chlorins M2, M4, and M6), implying an increase of the conjugated π-electron delocalization along X axis while the hypsochromic shift of the Q y indicates a decrease of the extent of conjugated π-electron delocalization along Y axis [36]. The UV-Visible absorption spectra of chlorins M1-M6 both in THF and grafted on the TiO2 surface through carboxylate bonding are presented in Figure 1. Table 1 summarizes their characteristics in THF and on TiO2. The shape of the UV-Visible absorption spectra in THF are relatively similar, with five observed bands when going from blue to red: Soret, Qx (0,1), Qx (0,0), Qy(0,1), and Qy(0,0). The Qy peaks position of the vinyl-substituted chlorins M3-M6 are redshifted by about 10-20 nm in comparison with chlorins M1-M2 due to the introduction of electron-withdrawing groups on the vinyl moiety at the C3 position. The bathochromic shifts of the Qy peaks of chlorins M5-M6 are larger than those observed for M3 (7-15 nm) and M4 (1-4 nm) because of the higher electron-withdrawing ability of the cyano group. Upon zinc metalation, the Soret absorption bands are redshifted (chlorins M2, M4, and M6), implying an increase of the conjugated π-electron delocalization along X axis while the hypsochromic shift of the Qy indicates a decrease of the extent of conjugated π-electron delocalization along Y axis [36].   The binding of the chlorin dyes onto the TiO 2 surface leads to a broadening of the absorption bands, probably due to π stacking interactions between chlorin dye molecules, as extensively studied in previous investigations [37]. The comparison of the chlorin absorption spectra in THF solutions as the monomeric species ( Figure 1a) and on TiO 2 as aggregates also reveal that the Q Y bands shift to longer wavelengths suggesting the formation of J aggregates [31]. The π stacking of dyes when grafted onto the TiO 2 surface was evidenced from the UV-Visible absorption spectra, as discussed above. Such aggregation is known to affect the photoelectron injection efficiency and thus, the overall conversion efficiency [38,39]. A typical way to lower the dye aggregation at the semiconductor interface consists in using co-adsorbents such as chenodeoxycholic acid (CDCA), the most commonly used additive in DSSC preparation [40][41][42]. Indeed, its strong binding to the TiO 2 surface partly displaces dye molecules, reducing their self-aggregation. Nevertheless, the addition of co-adsorbent molecules also reduces the dye loading and thus, photon harvesting. It is therefore crucial to judiciously adjust the amount of co-adsorbent used during the impregnation. To optimize the loading of the co-adsorbent in the dye solution, the impregnation was performed for 15 h using different concentrations of CDCA (from 0 to 3 mM) in a 0.3 mM THF solution of the dye. The choice of THF was motivated by the good solubility of the free-base and zinc-metalated chlorin in that solvent.
To determine the amount of dye M2 grafted onto the TiO 2 surface as a function of the concentration of co-adsorbent, UV-Visible absorption measurements were performed ( Figure 2). Dye M2 grafted from a 0.3 mM solution shows an absorbance of 0.55 for the Soret absorption band at around 420 nm and 0.25 for the lower energy Q band at around 665 nm. This value dramatically decreased when the concentration of co-adsorbent was increased up to 3 mM and this spectral evolution is in good agreement with previous studies reported in the literature [43,44]. Indeed, the co-adsorbent tends to decrease the adsorption of dye M2 via a competitive anchoring process, thereby decreasing the dye concentration and thus, the dye aggregation onto the TiO 2 surface.
The current density-voltage (J-V) characteristics of the corresponding devices were then measured and are depicted in Figure 3. The extracted photovoltaic parameters are presented in Table 2. To evaluate the efficiency of our fabrication process and thus, the accuracy of the collected data, a solar cell based on the well-known organic dye, D102, was used as reference [45,46].  The current density-voltage (J-V) characteristics of the corresponding devices were then measured and are depicted in Figure 3. The extracted photovoltaic parameters are presented in Table  2. To evaluate the efficiency of our fabrication process and thus, the accuracy of the collected data, a solar cell based on the well-known organic dye, D102, was used as reference [45,46].   The current density-voltage (J-V) characteristics of the corresponding devices were then measured and are depicted in Figure 3. The extracted photovoltaic parameters are presented in Table  2. To evaluate the efficiency of our fabrication process and thus, the accuracy of the collected data, a solar cell based on the well-known organic dye, D102, was used as reference [45,46].  The performance obtained for the reference cell is in agreement with those previously reported in the literature [45,46]. Modest photovoltaic performances are observed for dye M2 compared to liquid DSSCs, where the short-circuit current density (J SC ) was found to be reduced by a factor of about 10. This finding could mainly be attributed to the reduction in the electrode thickness for solid-state devices, leading to a lower light harvesting efficiency of the active layer. The devices where the TiO 2 electrode was sensitized with the dye M2 alone, gave photovoltaic efficiencies of 0.19%. When M2 was combined with CDCA, the PCE increased up to 0.56% for the 1:5 molar ratio. Further increase in Molecules 2020, 25,198 7 of 20 co-adsorbent to the 1:10 molar ratio caused a decrease in efficiency (η = 0.42%), attributed to the dye loading reduction. Table 2. Photovoltaic parameters of ssDSSCs sensitized with M2 or D102 under simulated solar emission (100 mW·cm −2 , AM1.5G) as a function of the co-adsorbent concentration: short-circuit current density (J SC ), open-circuit voltage (V OC ); fill factor (FF), and power conversion efficiency (η). The incident photon-to-current efficiency (IPCE) recorded for the above mentioned ssDSSCs indicates that the photocurrent response of M2-based DSSC impregnated with 1.5 mM of co-adsorbent exceeds 4% for the last Q band at 675 nm, which is higher than the response without co-adsorbent (2% at the same wavelength; Figure 4). The estimated photocurrent, obtained by integrating the IPCE spectra over the AM1.5G solar spectrum, was consistent with the photocurrent measured under solar simulator and reported in Table 2. Adding 1.5 mM of co-adsorbent for the impregnation enabled to double the current density value from 0.60 to 1.30 mA·cm −2 (0.73 vs. 1.73 mA·cm −2 in Table 2, respectively). The increased quantum yield was ascribed to the reduction of dye aggregation in the presence of CDCA as co-adsorbent, which minimized the intermolecular energy transfer and increased the electron injection efficiency. This effect compensated the loss of light harvesting caused by the lower dye content of the TiO 2 photoanode with CDCA as co-adsorbent and led to the improvement of the current density. Based on this result, a concentration of 1.5 mM of co-adsorbent was chosen for the preparation of ssDSSCs since it offered the best compromise between light harvesting and power conversion efficiency.  The performance obtained for the reference cell is in agreement with those previously reported in the literature [45,46]. Modest photovoltaic performances are observed for dye M2 compared to liquid DSSCs, where the short-circuit current density (JSC) was found to be reduced by a factor of about 10. This finding could mainly be attributed to the reduction in the electrode thickness for solidstate devices, leading to a lower light harvesting efficiency of the active layer. The devices where the TiO2 electrode was sensitized with the dye M2 alone, gave photovoltaic efficiencies of 0.19%. When M2 was combined with CDCA, the PCE increased up to 0.56% for the 1:5 molar ratio. Further increase in co-adsorbent to the 1:10 molar ratio caused a decrease in efficiency (η = 0.42%), attributed to the dye loading reduction. The incident photon-to-current efficiency (IPCE) recorded for the above mentioned ssDSSCs indicates that the photocurrent response of M2-based DSSC impregnated with 1.5 mM of coadsorbent exceeds 4% for the last Q band at 675 nm, which is higher than the response without coadsorbent (2% at the same wavelength; Figure 4). The estimated photocurrent, obtained by integrating the IPCE spectra over the AM1.5G solar spectrum, was consistent with the photocurrent measured under solar simulator and reported in Table 2. Adding 1.5 mM of co-adsorbent for the    Figure 5 shows the J-V characteristics of the ssDSSCs prepared from dyes M1 to M6 (0.3 mM in THF) and CDCA co-adsorbent (1.5 mM in THF). The corresponding photovoltaic data are listed in Table 3.
Molecules 2019, 24, x; doi: www.mdpi.com/journal/molecules THF) and CDCA co-adsorbent (1.5 mM in THF). The corresponding photovoltaic data are listed in Table 3.  Compared to the D102 performances, those obtained with chlorin dyes were low (Table 3). Nevertheless, only a few publications have described the fabrication of ssDSSCs with chlorophyll derivatives and the values obtained here are in line with those reported [31]. They were even better than the performances obtained in some liquid-DSSCs with other chlorophyll derivatives [47][48][49][50]. ssDSSCs based on dye M3 gave the highest PCE (0.67%) due to the highest short-circuit current (JSC), while the open-circuit photovoltage (VOC) and fill factors (FF) were found to be similar to other chlorin dyes, except for M2, which showed the highest VOC value. In terms of grafting groups, the PCE values for ssDSSCs based on chlorin dyes exceed 0.26% in the following order: acrylic acid > alkyl acid > cyanoacrylic acid. The introduction of cyano electron-withdrawing groups in chlorins M5-M6 indeed   Compared to the D102 performances, those obtained with chlorin dyes were low (Table 3). Nevertheless, only a few publications have described the fabrication of ssDSSCs with chlorophyll derivatives and the values obtained here are in line with those reported [31]. They were even better than the performances obtained in some liquid-DSSCs with other chlorophyll derivatives [47][48][49][50]. ssDSSCs based on dye M3 gave the highest PCE (0.67%) due to the highest short-circuit current (J SC ), while the open-circuit photovoltage (V OC ) and fill factors (FF) were found to be similar to other chlorin dyes, except for M2, which showed the highest V OC value. In terms of grafting groups, the PCE values for ssDSSCs based on chlorin dyes exceed 0.26% in the following order: acrylic acid > alkyl acid > cyanoacrylic acid. The introduction of cyano electron-withdrawing groups in chlorins M5-M6 indeed strongly decreased the J SC values, resulting in the ssDSSCs with the lower performances. This trend has been already observed when such chlorin dyes were used in liquid DSSCs [51]. Nevertheless, the difference observed in ssDSSCs was not as dramatic as in liquid DSSCs. For instance, the presence of cyano groups on the free base chlorin M3 drastically decreased the PCE in liquid DSSCs by around 75% while the decrease represents only 37% in ssDSSCs.

Recombination Kinetics
To examine the recombination kinetics of chlorin-based ssDSSCs, transient photovoltage decay measurements (TPV) measurements were carried out and compared to D102. Since the porous electrode and the anchoring groups of the dyes are quite similar (i.e., same fabrication process and only one carboxylic anchoring function on each dye), TPV measurements reflect the direct influence of the dye on the TiO 2 /dye/spiro-OMeTAD interface. The observed differences in recombination kinetics could mainly be attributed to the chemical nature and/or steric conformation of the dye at the TiO 2 surface as well as the contribution from the dye/spiro-OMeTAD interface. The observed difference in slope between D102 and the chlorin dye sensitizers indicates that different recombination mechanisms took place in ssDSSCs sensitized with these compounds (Figure 6). The recombinations for chlorin-sensitized solar cells appeared to be faster than that with D102, thus leading to a V OC loss. This result was coherent with the photovoltaic parameters obtained under illumination. Indeed, chlorin-based ssDSSCs exhibited V OC values around 550 mV, except for M2 whereas that based on the D102 dye sensitizer led to a V OC of 760 mV.
Molecules 2019, 24, x 9 of 21 Molecules 2019, 24, x; doi: www.mdpi.com/journal/molecules strongly decreased the JSC values, resulting in the ssDSSCs with the lower performances. This trend has been already observed when such chlorin dyes were used in liquid DSSCs [51]. Nevertheless, the difference observed in ssDSSCs was not as dramatic as in liquid DSSCs. For instance, the presence of cyano groups on the free base chlorin M3 drastically decreased the PCE in liquid DSSCs by around 75% while the decrease represents only 37% in ssDSSCs.

Recombination Kinetics
To examine the recombination kinetics of chlorin-based ssDSSCs, transient photovoltage decay measurements (TPV) measurements were carried out and compared to D102. Since the porous electrode and the anchoring groups of the dyes are quite similar (i.e., same fabrication process and only one carboxylic anchoring function on each dye), TPV measurements reflect the direct influence of the dye on the TiO2/dye/spiro-OMeTAD interface. The observed differences in recombination kinetics could mainly be attributed to the chemical nature and/or steric conformation of the dye at the TiO2 surface as well as the contribution from the dye/spiro-OMeTAD interface. The observed difference in slope between D102 and the chlorin dye sensitizers indicates that different recombination mechanisms took place in ssDSSCs sensitized with these compounds (Figure 6). The recombinations for chlorin-sensitized solar cells appeared to be faster than that with D102, thus leading to a VOC loss. This result was coherent with the photovoltaic parameters obtained under illumination. Indeed, chlorin-based ssDSSCs exhibited VOC values around 550 mV, except for M2 whereas that based on the D102 dye sensitizer led to a VOC of 760 mV. Besides, chlorin dyes M3 and M5 show an enhanced electron lifetime in comparison with the M1 analog. Electron lifetime associated with dye M3 was also slightly larger than that of dye M5, which suggests acrylic acid anchoring group was more favorable than cyanoacrylic acid anchoring group to decrease the charge recombination at the TiO2/electrolyte interface. However, the performances of ssDSSCs based on M5 were lower than those based on M1. Since the dye M1 contains a non-conjugated anchoring group, M1 may lie rather flat on the TiO2 surface due the larger flexibility of the alkyl spacer, whereas the more rigid conjugated spacer of M3 and M5 probably favors tilted or up-right adsorption. Finally, the influence of the zinc metal center was more difficult to interpret: on one hand, M2 presented a significant increase in electron lifetime compared to M1; on the other hand, M3 and M4 were in the same range. These findings were coherent with the photovoltaic Besides, chlorin dyes M3 and M5 show an enhanced electron lifetime in comparison with the M1 analog. Electron lifetime associated with dye M3 was also slightly larger than that of dye M5, which suggests acrylic acid anchoring group was more favorable than cyanoacrylic acid anchoring group to decrease the charge recombination at the TiO 2 /electrolyte interface. However, the performances of ssDSSCs based on M5 were lower than those based on M1. Since the dye M1 contains a non-conjugated anchoring group, M1 may lie rather flat on the TiO 2 surface due the larger flexibility of the alkyl spacer, whereas the more rigid conjugated spacer of M3 and M5 probably favors tilted or up-right adsorption. Finally, the influence of the zinc metal center was more difficult to interpret: on one hand, M2 presented a significant increase in electron lifetime compared to M1; on the other hand, M3 and M4 were in the same range. These findings were coherent with the photovoltaic parameters obtained under illumination. As discussed above, M2 exhibited a higher Voc than M1 while the Voc was approximately the same for M3 and M4.

DFT Calculations
To elucidate the relationship between the nature of the anchoring group and the photovoltaic performances observed, density functional theory (DFT) calculations were performed. The calculations show that all the LUMO + n orbitals had a significant contribution from the acrylic group for M4 and from the cyanoacrylic acid group for M6 while only the LUMO + 3 orbital covered the alkyl carboxylic group for M2 (Figure 7). This is critically important since the orbital overlap between the LUMO + n of the dye and the TiO 2 conduction band edge should be as large as possible in order to maximize the electron transfer probability. M4 and M6 dyes exhibited similar π-electron systems leading to similar electron density on the π-conjugated anchoring group for LUMO, LUMO + 1, and LUMO + 2 orbitals.

DFT Calculations
To elucidate the relationship between the nature of the anchoring group and the photovoltaic performances observed, density functional theory (DFT) calculations were performed. The calculations show that all the LUMO + n orbitals had a significant contribution from the acrylic group for M4 and from the cyanoacrylic acid group for M6 while only the LUMO + 3 orbital covered the alkyl carboxylic group for M2 (Figure 7). This is critically important since the orbital overlap between the LUMO + n of the dye and the TiO2 conduction band edge should be as large as possible in order to maximize the electron transfer probability. M4 and M6 dyes exhibited similar π-electron systems leading to similar electron density on the π-conjugated anchoring group for LUMO, LUMO + 1, and LUMO + 2 orbitals.

Dye Adsorption
To get insight into the relationship between the observed performances and the dye structure, the dye adsorption of free-base chlorins M1, M3, and M5 was investigated by the use of electron paramagnetic resonance spectroscopy (EPR). This experimental technique allowed us to compare the cells performance with the dye adsorption on the TiO2 thanks to a paramagnetic radical probe 4carboxy TEMPO (coded as 4C-T). This molecule grafts onto the TiO2 surface without perturbing the system and is able to monitor its molecular surrounding through spin-spin interactions. As described more in depth elsewhere [53], the main parameters extracted from the spectra are as follows: (i) Aii, which is the component of the hyperfine coupling tensor between the electronic and nuclear spin that provides a measurement of the probe environmental polarity; (ii) the correlation time for the radical rotational diffusion motion, τ, that is related to the radical environmental microviscosity, which, in turn, increases due to interactions between the radical and its environment; (iii) the intrinsic line width, its increment in slow motion conditions being due to dipole−dipole spin−spin interactions (through this parameter, interactive close sites can be identified); and (iv) the Heisenberg exchange frequency, Wex, which increases when radicals swap each other the unpaired electrons due to collisions in a fluid medium (this parameter is related to a high radical local concentration on the TiO2 surface). In many cases, the EPR spectra are constituted by two well-identifiable components characterized by probes at different polarity and mobility. In order to extract the two components and to evaluate the relative percentages, experimental spectra were subtracted from each other and then the components were separately simulated and doubly integrated to evaluate the relative percentages. If subtraction of experimental spectra cannot be performed, one component is simulated and the other one is determined by subtraction of the simulated spectrum from the total one. Afterwards, the second component is also simulated. The accuracy for all the parameters in the EPR analysis was between 2% and 3%, based on the fitting between the experimental and the computed line shapes, slightly increasing if line broadening occurs. Figure 9 shows the EPR spectra obtained for DSSCs built with the different dyes M1, M3, and M5 and containing the spin probe 4C-T. These spectra were considered good if reproducible at least for three different cells built in the same experimental conditions.

Dye Adsorption
To get insight into the relationship between the observed performances and the dye structure, the dye adsorption of free-base chlorins M1, M3, and M5 was investigated by the use of electron paramagnetic resonance spectroscopy (EPR). This experimental technique allowed us to compare the cells performance with the dye adsorption on the TiO 2 thanks to a paramagnetic radical probe 4-carboxy TEMPO (coded as 4C-T). This molecule grafts onto the TiO 2 surface without perturbing the system and is able to monitor its molecular surrounding through spin-spin interactions. As described more in depth elsewhere [53], the main parameters extracted from the spectra are as follows: (i) A ii , which is the component of the hyperfine coupling tensor between the electronic and nuclear spin that provides a measurement of the probe environmental polarity; (ii) the correlation time for the radical rotational diffusion motion, τ, that is related to the radical environmental microviscosity, which, in turn, increases due to interactions between the radical and its environment; (iii) the intrinsic line width, its increment in slow motion conditions being due to dipole−dipole spin−spin interactions (through this parameter, interactive close sites can be identified); and (iv) the Heisenberg exchange frequency, Wex, which increases when radicals swap each other the unpaired electrons due to collisions in a fluid medium (this parameter is related to a high radical local concentration on the TiO 2 surface). In many cases, the EPR spectra are constituted by two well-identifiable components characterized by probes at different polarity and mobility. In order to extract the two components and to evaluate the relative percentages, experimental spectra were subtracted from each other and then the components were separately simulated and doubly integrated to evaluate the relative percentages. If subtraction of experimental spectra cannot be performed, one component is simulated and the other one is determined by subtraction of the simulated spectrum from the total one. Afterwards, the second component is also simulated. The accuracy for all the parameters in the EPR analysis was between 2% and 3%, based on the fitting between the experimental and the computed line shapes, slightly increasing if line broadening occurs. Figure 9 shows the EPR spectra obtained for DSSCs built with the different dyes M1, M3, and M5 and containing the spin probe 4C-T. These spectra were considered good if reproducible at least for three different cells built in the same experimental conditions. From Figure 9, we could determine that: (i) the intensity significantly decreased for DSSC-M5 with respect to the other DSSC-dyes. The intensity was evaluated by double integrating the spectra (Table 1). We also noticed that the intensity largely decreased when compared to devices containing only the spin probe (results not shown) and (ii) the spectra were constituted by two components whose main features are indicated in the figure. These two components were extracted and quantified (Table 4). Each component was then computed as shown in Figure 10 for the different cells. The figure also shows the main parameters used for the computation, which are the <A> value (=(Axx + Ayy + Azz)/3) measuring the polarity, the correlation time for the rotational motion, τ, measuring the interactions on the basis of the slowing down of mobility, and the line width, measuring the local concentration of the radicals. Based on the correlation time for the rotational motion, τ, obtained from the computations, the two components were termed "fast" and "slow", since they originated from spin probes quite free to move at the interface and interacting with the TiO2 surface, respectively. From Figure 9, we could determine that: (i) the intensity significantly decreased for DSSC-M5 with respect to the other DSSC-dyes. The intensity was evaluated by double integrating the spectra (Table 1). We also noticed that the intensity largely decreased when compared to devices containing only the spin probe (results not shown) and (ii) the spectra were constituted by two components whose main features are indicated in the figure. These two components were extracted and quantified (Table 4). Each component was then computed as shown in Figure 10 for the different cells. The figure also shows the main parameters used for the computation, which are the <A> value (=(A xx + A yy + A zz )/3) measuring the polarity, the correlation time for the rotational motion, τ, measuring the interactions on the basis of the slowing down of mobility, and the line width, measuring the local concentration of the radicals. Based on the correlation time for the rotational motion, τ, obtained from the computations, the two components were termed "fast" and "slow", since they originated from spin probes quite free to move at the interface and interacting with the TiO 2 surface, respectively. The low intensity of the spectrum of DSSC-M5 if compared to the other DSSCs clarified the behavior of these dyes: M5 was poorly adsorbed, but the probes and M5 preferentially interacted with the surface (slow component) and only a small amount of dyes remained at the interface (free component). In other words, the total amount of adsorbed dye M5 as well as the amount of interacting probes + M5-dyes were small. That amount was obviously smaller than the amount of probes and dye M3 adsorbed at the TiO2 surface. Considering the slow component, it was the same for M5 and M3 corresponding to strongly interacting dyes at polar, quite close sites. Conversely, the fast component shows a high mobility and the environment was less polar than that of the slow component. These results also explained why dye M3 exhibited better performances than M5. M1 was the less interacting dye, not only marked by the lowest percentage of slow (interacting) component compared to the other dyes, but also because the slow component is slower (lower τ) in comparison with the other dyes. It is also interesting to note that the fast component for M1 shows a slightly lower mobility, lower polarity and higher line width compared to the fast component from the other dyes. This indicates a higher local concentration of M1 dyes in solution, in line with a lower adsorption at the TiO2 surface. The low intensity of the spectrum of DSSC-M5 if compared to the other DSSCs clarified the behavior of these dyes: M5 was poorly adsorbed, but the probes and M5 preferentially interacted with the surface (slow component) and only a small amount of dyes remained at the interface (free component). In other words, the total amount of adsorbed dye M5 as well as the amount of interacting probes + M5-dyes were small. That amount was obviously smaller than the amount of probes and dye M3 adsorbed at the TiO 2 surface. Considering the slow component, it was the same for M5 and M3 corresponding to strongly interacting dyes at polar, quite close sites. Conversely, the fast component shows a high mobility and the environment was less polar than that of the slow component. These results also explained why dye M3 exhibited better performances than M5. M1 was the less interacting dye, not only marked by the lowest percentage of slow (interacting) component compared to the other dyes, but also because the slow component is slower (lower τ) in comparison with the other dyes. It is also interesting to note that the fast component for M1 shows a slightly lower mobility, lower polarity and higher line width compared to the fast component from the other dyes. This indicates a higher local concentration of M1 dyes in solution, in line with a lower adsorption at the TiO 2 surface.

Instrumentation and Methods
Spirulina powder was provided from Phyco-Biotech (Montpellier, France). Titania paste (99% anatase), titanium(IV) chloride (99,9%), chenodeoxycholic acid (96%), bis(trifluoromethane)sulfonimide lithium salt (99%), 4-tert-butylpyridine (98%) were purchased from Sigma-Aldrich Chemie S.a.r.l. (Saint-Quentin Fallavie, France). Spiro-OMeTAD and FTO-coated conducting glass substrates were purchased from Solaronix SA (Aubonne, Switzerland). Reactions were performed under argon using oven-dried glassware and Schlenk techniques. Dry solvents were obtained by using a solvent purification system, PureSolve MD5 purchased from Inert Technology (Amesbury, MA, USA). 1 H and 13 C{ 1 H} NMR spectra were recorded either on a Bruker Avance-300, 400, or 600 spectrometers (Bruker, Billerica, MA, USA) for each synthetic intermediates and were calibrated to tetramethylsilane (TMS) on the basis of the relative chemical shift of the solvent as an internal standard. Chemical shifts (δ) are expressed in parts per million (ppm) and coupling constants (J) are expressed in Hertz. Abbreviations used for 1 H-NMR spectra are as follow: s singlet, d doublet, and m multiplet. Mass spectra (MS) were recorded on MicroTOF QII t and RAPIFLEX instruments (Bruker France SAS, Marne la Vallée, France) in the positive mode for ESI and MALDI-ToF, respectively. UV-Visible absorption spectra were obtained at room temperature on a Shimadzu UV2401 PC (Shimadzu Corporation, Kyoto, Japan) and on a SAFAS D.E.S. PC UV/Vis scanning spectrometers (Safas, Monaco).

DFT Calculations
The ground-state geometry and electronic structure of all dyes and a large TiO 2 cluster (representative of the bulk) were obtained at the density functional theory (DFT) level using the B3LYP hybrid functional and the 6-311G(d,p) basis set. All calculations were performed using Gaussian09 [57]. Solvent effects were taken into account by employing the polarizable continuum model (PCM) using the chloroform solvent as a proxy for the dielectric effects in the DSSC device.

Fabrication of Solar Cells
FTO-coated conducting glass substrates were cleaned by an ultrasonic treatment with water, acetone and isopropanol for 10 min each before being treated for 5 min by UV-ozone. A compact layer of TiO 2 was deposited by spray-pyrolysis at 450 • C from a solution of titanium tetra-isopropoxide and acetylacetone in ethanol following by annealing at 450 • C for 20 min (Film thickness 200 nm).
A nanoporous TiO 2 layer was applied by spin-coating from a solution of a commercial TiO 2 paste (DSL 18NRT, Dyesol) in ethanol containing nano-sized anatase particles (Film thickness 1.6 µm). The layers were then gradually annealed from 250 to 500 • C over 45 min. The substrates were then treated in a 0.02 M TiCl 4 aqueous solution for 2 h, rinsed with water and annealed at 450 • C for 45 min. The electrodes were finally immersed in dye solutions that may hold chenodeoxycholic acid as co-adsorbent for 15 h in the dark. After rinsing the electrodes, the hole conductor layer was deposited by spin-coating from a spiro-OMeTAD (Solaronix) solution in chlorobenzene (180 mg.mL −1 ) containing tert-butylpyridine and LiTFSI as additives, following a reported recipe [45]. Gold top electrodes were finally evaporated under vacuum (10 −6 mbar) using shadow masks that define two active areas per substrate (around 0.15 cm 2 each).

Characterization of Solar Cells
The J-V measurements, in the dark and under illumination, were performed in air using a Keithley model 2400 digital source meter by applying independent external voltage to the cell and by measuring the photogenerated current out from the cell. The spectral mismatch between the emission solar simulator (Newport 1600W) and the global AM1.5G solar spectrum was corrected using a mismatch factor and the solar simulator irradiance was adjusted accordingly using a certified silicon reference cell in order to achieve an equivalent AM1.5G irradiance of one sun (100 mW·cm −2 ) on the test cells.
The incident photon to current efficiency (IPCE) was estimated using a monochromated 75 W xenon lamp (Newport). The photocurrent, measured in static regime by a calibrated picoammeter (Keithley 485), was compared to the calibration current response recorded using a certified calibrated photodiode (Newport). All data shown in this work were the average values of at least four parallel tests.

Fabrication of Solar Cells for EPR Experiments
FTO slides were washed and sonicated five times for 15 min each in water, ethanol, twice in isopropanol and finally, analytical grade ethanol. FTO-TiO 2 (18 NRT Solaronix) films were prepared by "doctor blading" (uniformly spreading TiO 2 paste on a pre-defined area of 1.25 cm 2 on FTO glass with a glass bar). The FTO electrode was then left for 30 min on an oven at 450 • C in order to sinter the anatase crystals. The dye adsorption on FTO-TiO 2 film electrodes was carried out by heating the electrodes for 10 min at 75 • C on a hot plate, followed by immersion into a single dye solution in 50% acetonitrile 50% tBuOH for a given time of 16 h. The DSSCs were electrically characterized by the use of a solar simulator (Newport model 96,000 Oriel), equipped with a 150 W ozone free xenon lamp, and a digital multimeter (Keithley 2600A) connected to a computer that used dedicated software for test running.

EPR Characterization of Solar Cells
Before performing the EPR analysis of the complete DSSC, it was necessary to measure the solar cell electrical performance in the absence and the presence of the spin probe. Indeed, only cells showing well-measurable and reproducible electrical performances (not reported) were tested by EPR. After measuring their electrical functionality, the cells were placed on the EPR cavity where a static magnetic field is applied while a source of microwaves is also applied to the sample. This allows the resonance condition that is necessary to promote the probe (4C-T) signal. EPR spectra were recorded with an EMX-Bruker spectrometer operating at the X band (9.5 GHz) and interfaced with a PC (software from Bruker for handling and recording the EPR spectra). The EPR spectra were simulated with the NLSL software [58].

Transient Photovoltage Decay Mesurements
Transient photovoltage measurements were done on the cells illuminated by a bias light provided by white LED of variable intensity [59]. An additional light pulse was superimposed using a 550 nm green LED, and the transient electrical response of the cell was recorded in open-circuit conditions using the high impedance input of a digital oscilloscope (Tektronix DPO4032, Tektronix, Bracknell, UK) Transient photovoltage curves were fitted using mono-exponential functions, ensuring a perturbing regime where the transient photogenerated charge density was kept below 1% of the stead-state photogenerated charge density.

Conclusions
A series of chlorophyll derivatives M1-M6 equipped with either a non-conjugated carboxy or a conjugated acrylic or cyanoacrylic anchoring group was prepared for their use as dye component in all-solid state DSSCs. The use of a co-adsorbent was found to be crucial to improve the device performance by avoiding the chlorin dye aggregation. The solar energy conversion efficiencies of DSSCs bearing the acrylic group (M3-M4) were larger than those bearing the cyanoacrylic group (M5-M6). The weakest performances of M5-M6 compared to M3-M4 were attributed to several factors: (i) a faster recombination kinetic at the titania/electrolyte interface, (ii) the position of the LUMO below the TiO 2 conduction band is not suitable for electron injection, and (iii) a lower amount adsorbed onto the TiO 2 surface. Nevertheless, although M1-M2 exhibited not suitable features in terms of recombination kinetics and dye adsorption, higher PCEs were obtained which may suggest that M1-M2 molecules have a suitable orientation on the TiO 2 surface due the larger flexibility of the alkyl spacer, which may lead to improved charge transfer. This question will be addressed in future investigations as well as their long-term stability.