An atomistic picture of the regeneration process in dye sensitized solar cells

Abstract

A highly efficient mechanism for the regeneration of the cis-bis(isothiocyanato)bis(2,2′-bipyridyl-4,4′-dicarboxylato)-ruthenium(II) sensitizing dye (N3) by I^- in acetonitrile has been identified by using molecular dynamics simulation based on density functional theory. Barrier–free complex formation of the oxidized dye with both I^- and , and facile dissociation of and from the reduced dye are key steps in this process. In situ vibrational spectroscopy confirms the reversible binding of I₂ to the thiocyanate group. Additionally, simulations of the electrolyte near the interface suggest that acetonitrile is able to cover the (101) surface of anatase with a passivating layer that inhibits direct contact of the redox mediator with the oxide, and that the solvent structure specifically enhances the concentration of I^- at a distance which further favors rapid dye regeneration.

The basic design of today’s high performance dye sensitized solar cells (DSSC) was developed in the early 1990’s by Grätzel et al. (1, 2). The photoactive part of these devices consists of a wide band gap semiconductor covered by a monolayer of sensitizing dye. The semiconductor is directly supported by a transparent electrode on one side, while the dye is connected to the back electrode via a liquid electrolyte or a solid hole conducting material. The initial step of the photovoltaic process is a light induced electron injection from the dye into the semiconductor material. This process yields an oxidized dye and an energetic electron. Rapid regeneration (reduction) of the dye by the electrolyte prevents back transfer of the electron or degradation of the photo-oxidized dye (3). Meanwhile, the energetic electron diffuses away from the dye, passing through the electrode and an external load, finally reaching the counter electrode where it regenerates the electrolyte.

The class of devices with the highest light to current conversion efficiency (above 11%) (4, 5) is based on sintered nano–crystalline anatase as the semiconducting oxide, Ruthenium polypyridyl dyes as sensitizer, and the iodide/triidiode redox couple dissolved in a nitrile containing organic solvent as electrolyte. DSSC using organic dyes, solid state electrolytes, different semiconductors, or redox couples do not match this performance but expose other desirable properties for the commercial use of this technology. Optimizing the various components of the oxide/dye/electrolyte interface, the efficiency and stability of the systems can be improved and its cost reduced. A detailed knowledge of the interfacial structure and key reaction mechanisms in high performance DSSC is, therefore, essential to guide rational design of improved devices. Numerous experimental and theoretical studies have already led to a deeper understanding of this system, but several basic questions are still the subject of active research. In this study, we propose a consistent regeneration mechanism of the dye and discuss the role of several structural aspects of the solid/liquid interface. The system we study consists of anatase, the N3 or cis-diisothiocyanato-bis(2,2′-bipyridyl-4,4′-dicarboxylato) ruthenium(II) bis(tetrabutylammonium) sensitizer (N719) and iodide/triiodide in acetonitrile. In the first part of this article, we present computational results for the regeneration mechanism and the experimental validation of an intermediate complex predicted by these calculations. The second part discusses the impact of the solvent/electrolyte structure at the semiconductor interface on the operation of DSSC.

In an effort to elucidate the regeneration mechanism of the dye, previous spectroscopic studies have identified a short-lived -radical intermediate (6). Additionally, spectroscopic evidence for the existence of an intermediate dye-iodine complex has been presented (6–8), and recent DFT studies on the gas phase molecule verified that iodide can indeed bind to the oxidized dye (9). Here, employing DFT based molecular dynamics simulations in the condensed phase that probe not only iodide but also the and the formation of , we propose a complete regeneration mechanism, which we partially verify with experiments. The concentration of is much lower than that of I^-, but plays a crucial role in the full mechanism. We qualitatively confirm the earlier gas phase calculations that predict binding of the iodide ion and radical to the gas phase dye. (9) Most importantly, we extend these studies to explicitly include the solvent, so that we can provide thermodynamical data, and insight in the association/dissociation kinetics. In the gas phase, both I^- and can bind the oxidized dye either covalently at the sulfur atom of the SCN group, or more weakly via bipyridyl iodine interactions (9). Interestingly, in the sulfur bound complexes the central metal is reduced, these complexes, thus, involve a neutral I(0) or I₂, while the charge is not transferred for bipyridyl the bound complex. Simulations of the solvated complexes show that only the sulfur bound complexes are stable in acetonitrile, while the bipyridyl bound structures dissociate within 1 ps of simulation without electron transfer to the dye. Consequently, we focus on the stable sulfur bound [dye(II):I(0)] and [dye(II):I₂] complexes in the remainder of this work. To quantify the stability of these complexes, and to get insight in their association kinetics, we have computed the free energy profiles of dissociation by using thermodynamic integration along a specified reaction coordinate. For this calculation, it is useful to take as a reaction coordinate the sum of the distances between the iodine–ions and the sulfur atoms of both SCN groups. This guarantees that dissociation is a smooth process, i.e., the recombination of the ion with the dye is prevented as the reaction coordinate increases. Nevertheless, computing free energies from DFT based molecular dynamics is computationally demanding, and a statistical error of 1–2 kcal/mol can not be excluded. Free energy profiles, shown in Fig. 1, confirm the stability of both complexes, and predict a binding free energy of about 2–3 kcal/mol and 6–7 kcal/mol for the [dye:I] and [dye:I₂] complex, respectively. Additionally, we find that the free energy profile is barrierless, which indicates that the reaction rate for association will be diffusion limited. In unconstrained simulation, formation of the [dye:I] complex was observed on the ps time scale for all points along the reaction coordinate.

Fig. 1.

Shown are the free energy profiles for association of the [dye:I] complex (Lower Curve) and [dye:I₂] complex (Upper Curve) in acetonitrile solution. The reaction coordinate employed (x-axis) is the sum of the two S-I distances (see text for details). Both complexes are, thus, weakly stable and association is essentially barrierless. The Inserts show snapshots of the stable complexes in solution, the solvent is not shown. The average S-I bond length is 3.12 Å and 2.83 Å for [dye:I] and [dye:I2], resp.

As shown in Fig. 2, we have been able to detect the [dye:I₂] complex in situ by a sequence of ATR-IR experiments on the reduced N3 dye attached to anatase nano–particles in contact with acetonitrile and molecular I₂. The spectrum of the neat dye attached to the surface shows a single high intensity peak at 2098 cm^-1 corresponding to the C-N stretching mode of the SCN-ligand. After the addition of I₂, the peak splits into two signals, one remaining at 2098 cm^-1 with lower intensity, and a new one at 2130 cm^-1. The reduced intensity of the SCN-ligand signal, together with the new signal at 2130 cm^-1, can be explained by the formation of an I₂SCN ligand. The assignment of the peaks was verified by computing the gas phase spectrum of the [dye:I₂] complex, in which we found the signal of the I₂SCN ligand to be about 40 cm^-1 higher than the neat SCN-ligand. The computed weak binding free energies are in good agreement with the experimental observation that the binding of I₂ is easily reversible, i.e., flowing neat acetonitrile through the cell restores the original signal at 2098 cm^-1 and eliminates the peak at 2130 cm^-1.

Fig. 2.

Experimental IR-spectra of the N3 dye attached to TiO₂ in acetonitrile, before adding iodine (Black Line), with I₂ added (Red Line), and after washing iodine away (Blue Line). The pronounced peaks at 2098 cm^-1 and 2130 cm^-1 correspond to vibrations of the SCN and I₂SCN ligands, and support our finding of reversible complex formation.

To investigate the remainder of the regeneration process, we have added additional I^- and ions to our simulations of the [dye:I] and [dye:I₂] complexes. An interaction of I^- with the [dye:I] complex has been suggested before (6–9). In all cases, this led to an immediate dissociation of the S–I bond as soon as the additional iodide reached the binding site of the halide, yielding solvated or and a reduced dye. Because the SCN- group can be considered a pseudo–halide in the reduced complex, such a spontaneous dissociation is expected. These results combined suggest an efficient pathway for dye regeneration under working conditions in DSSC, which is shown in Fig. 3. The first part of the reaction is the interaction of I^- with the oxidized dye, yielding a stably bound [dye:I] complex. This complex interacts with I^- to form a reduced dye and free . Contrary to the assumption that is formed in the electrolyte by disproportionation of , we propose as an alternative pathway that instead interacts with another oxidized dye and forms a [dye:I₂] complex that can further react with I^- to yield , or that interacts with a [dye:I] complex to directly form . Even though competes with I^- in these reactions, this mechanism has the advantage that it does not require two radicals that are only present in low concentration, to meet in bulk solution. Instead, the radical anion can react in place of formation with a photo-oxidized dye close by. Remarkably, the full pathway proceeds in a barrier free way. This is in agreement with the fast regeneration rates observed in experiments and the unrivaled efficiency of this combination of dye and redox pair.

Fig. 3.

Proposed mechanism of the dye regeneration. In a barrier free way, two oxidized dyes are reduced by I^-, and I^3- is produced.

The proposed formation of various dye–iodide complexes during dye regeneration raises the question of the availability of iodide in this active region. Furthermore, the influence of the solvent, and in particular the excellent performance of acetonitrile, has not been fully explained yet. To address this question, we employ atomistic models of the interface based on the (101) surface of anatase, various ions, and acetonitrile. First, employing force field based molecular dynamics, we compute the probability of finding an ion at a given distance from the surface by using thermodynamic integration (TI). The probability profiles can directly be interpreted as concentration profiles, and are shown in Fig. 4 for a number of monovalent anions (F^-, Cl^-, and I^-), cations (Li⁺,Na⁺,K⁺), and . All curves exhibit a non-monotonic behavior, with several distinct minima and maxima, contrary to the smooth distribution that a Poisson–Boltzmann description of this system would predict. Calculations of the distance dependent solvent dielectric constant (Fig. S1), indeed, suggest that local linear response is not any longer valid in this region and that the solvent at the interface can not be described with a single dielectric constant (10, 11). This indicates the limitations of continuum theory in this important region, and suggests that explicit atomistic models might also enhance theoretical predictions of interfacial electron transfer based on continuum models (12). Interestingly, the concentration of I^- strongly peaks at about 10 Å from the surface, and exceeds the bulk concentration, while the concentration of all other monoatomic ions studied decreases towards the surface. A high concentration of I^- near the surface, in particular near the SCN groups, which can be expected to be at about 10 Å, will contribute to a rapid reduction of oxidized dyes, enhancing the performance of DSSCs. In the case of the radical, the structure is less pronounced, but there is certainly no strong thermodynamic gradient that would favor diffusion away from the interface. This supports our proposed reaction mechanism, which suggests that further reacts with dye-iodide complexes to produce near the interface. We note that bromide has previously been used in X-ray reflectivity studies of the counter ion distribution near a lipid monolayer (13). In that study, a minor second maximum could be discerned in the ion distribution, but the experimental resolution did not allow quantification of the small feature. We believe that the pronounced effect we observe in this system, for a heavy ion near a potentially flat surface, could be verified experimentally.

Fig. 4.

The concentration, relative to the bulk value, of various monovalent ions and as a function of the distance to the anatase (101) surface. Data obtained from thermodynamic integration in the neat, otherwise salt-free solvent. I^- is present in higher concentration near the interface, while the other monovalent ions are being repelled from the interface.

The ionic profiles discussed so far have been obtained by simulations that mimic very low ionic strength. The observed structure can therefore not be attributed to a correlation between ions, but must be induced by the solvent. In previous work, we have observed a strong layering of the acetonitrile near the anatase(101) surface, due to a favorable interaction of the nitrile group with the titanium atoms of the surface, and the pronounced dipolar nature of the solvent (14). Here, we observe that the structure of the liquid, as illustrated with a density profile in Fig. 5, is indeed reflected in the distributions of the ions. The fact that iodide is preferentially found in the layered solvent close to the interface, whereas the other ions are rather expelled from that region, must be related to the size of the ion. Indeed, in the model employed, the van der Waals parameters are the only differences between the ions. Iodide, having roughly the size of an acetonitrile molecule, can be better accommodated in the layers than the other, smaller ions that disrupt the layered solvent structure more. For other solvents, the match between solvent and iodide might be less favorable and the solvent layering less pronounced. The enhanced iodide concentration at the interface might thus be a feature that is particular to acetonitrile and might contribute to the efficiency of this particular electrolyte.

Fig. 5.

The Left Graph shows the relative concentration profile of iodide compared to the probability density of the acetonitrile central carbon atom. The layering of the solvent (Black Line) near the interface clearly correlates with the peaks in the ion distribution (Green Line). The image on the Right shows iodide near the interface, approximately at the location of the first maximum in the probability density profile.

The effect we observe persists at experimentally relevant concentrations of the ions that we illustrate in Fig. 6 with ion profiles obtained by direct MD simulation at a 0.5 M concentration of I^- and tetrapropylammonium (TPA⁺). The I^- concentration profile obtained in this way agrees very well with the profile obtained from TI. Furthermore, an alternation between layers of high anion and high cation concentration can clearly be observed. Inspection of the molecular dynamics simulations also revealed that diffusion of I^- is facile parallel to the surface, while perpendicular motion is clearly hindered by free energy barriers. Our molecular dynamics data also shows that the first solvent layer that features strong interactions of the nitrile groups with the oxide is never penetrated by ions. This suggests that such a solvent layer can effectively passivate the surface, reducing the back electron transfer to solvent species such as or . A solvent which eliminates these losses improves device efficiency.

Fig. 6.

The concentration, relative to the bulk value, of iodide and tetrapropylammonium (TPA⁺) as a function of the distance to the anatase (101) surface. The binned data is obtained from molecular dynamics simulation at a typical salt concentration (0.5 M), while the Contiguous Line is as in Fig. 4.

To summarize our results, we have proposed a complete mechanism for the N3 dye regeneration, which leads via a barrier free pathway from the oxidized dye and I^- to the reduced dye and . Key in this process are stable [dye:I] and [dye:I₂] intermediates that dissociate spontaneously upon addition of further I^- or . Furthermore, we find that atomistic properties of the electrolyte, both solvent and ion, play an important role in shaping the distribution of ions near the solid/liquid interface. We find that iodide in acetonitrile is a combination that favors dye regeneration, displaying a marked concentration peak in the region of the dye. At the same time, this solvent passivates the surfaces, and reduces back electron transfer, by preventing direct ion–oxide encounters. This insight contributes to the understanding of the most efficient dye sensitized solar cells constructed to date, and is expected to guide rational design of improved devices.

Materials and Methods

All density functional theory (DFT) and force field (FF) based molecular dynamics (MD) simulations were performed using the CP2K program package (15). DFT calculations are based on the hybrid Gaussian and plane wave scheme with the Perdew-Burke-Enzerhof (16) exchange correlation functional and corresponding pseudo potentials (17, 18). The solvent has been described with Gaussian basis sets of double ζ quality with one polarization function, while a more accurate triple-ζ basis with double polarization has been employed for the dye (19). The exception are Ruthenium and Iodine, for which highly accurate molecularly optimized (20) basis sets have been used. The plane wave density cutoff was set to 280 Ry. The FF for acetonitrile is based on point charges, van der Waals interactions and harmonic bonded terms as found in literature (21), TiO₂ was modeled using point charges, and a Buckingham potential from literature (22), while the interaction between both species has been derived and validated against ab initio data in our earlier work (14). Parameters for the cations have been taken from literature (23, 24), for fluoride we employed σ = 2.90 Å and ϵ = 0.105 kJ/mol, for chloride σ = 4.04 Å and ϵ = 0.157 kJ/mol, for iodide σ = 4.60 Å and ϵ = 0.438 kJ/mol similar to ref. 25, the van der Waals interactions between ions and solvent are obtained by the usual combination rules, while only electrostatic interactions are retained between oxide and ions. For classical pre-equilibration of solvent-dye geometries the dye has been parameterized based on the generalized Amber force field. In the first part of this work, constrained ab initio MD is used to calculate the binding free energy of I^- and to the oxidized dye solvated in acetonitrile. These simulations have been performed in the microcanonical ensemble (NVE), using a timestep of 0.5 fs, with an initial temperature of 300 K. The equilibrium geometry has been equilibrated for 5 ps using first principles MD. For each value of the reaction coordinate, simulations in the range 7–10 ps have been performed, and the first 1 ps discarded for equilibration. The initial configurations have been obtained by immersing optimized gas phase structures of the dye in 58–60 molecules of acetonitrile, and equilibrating the solvent molecules only with FF based MD for 1 ns in the constant pressure ensemble (NPT). The equilibrated unit cell contains approx. 400 atoms and its volume is 20 × 20 × 20 Å³, which limits the dye-ion to about 6 Å. The classical simulations in the second part of this work are NVE, employ a 0.5 fs timestep and are at approx. 300 K. For the TI simulation, systems contain approx. 3,700 atoms of which one ion (22.7 × 20.5 × 113.0 Å³) and 800 ps of dynamics has been used for averaging the constraint force in each point of the thermodynamic integration (73 points per profile). The explicit simulation of the electrolyte at a finite concentration, starting from an initially random ion pattern, is an average over eight runs, each approx. 3.7 ns, systems contain approx. 21,700 atoms of which 56 I^- (155.2 × 40.9 × 45.4 Å³). Infrared spectra were measured on a Bruker IFS 66/S FT-IR spectrometer equipped with a dedicated attenuated total reflection (ATR)-IR attachment (Optispec) and a liquid nitrogen cooled mercury-cadmium-telluride detector. All spectra were recorded at a resolution of 2 cm^-1. The details of the ATR-IR setup is described elsewhere. (26) The film consisting of a titania nanoparticle (Ti-Nanoxide D, Solaronix) was prepared by homogeneously depositing the particles on a Ge internal reflection element that was in advance coated with a 10 nm TiO₂ layer by physical vapor deposition. Subsequently, the film was dried and calcined at 733 K in air for 1 h and transferred to the ATR-IR setup. 0.2 mM of the N719 dye (Solaronix) in ethanol (> 99.9%, Merck) was slowly admitted until the signals due to dye adsorption were stabilized. The solution in contact with the dye-adsorbed film was first exchanged with acetonitrile (> 99.9%, Sigma–Aldrich) and then further exchanged with a 0.05 M iodine (> 99.8%, Fluka) solution in acetonitrile to study the interaction of the adsorbed dye with iodine. The reversibility of the dye-iodine interacting complex was further examined by washing the iodine solution with neat acetonitrile. All the measurements were carried out at 313 K.