Report

Ultrafast Long-Range Charge Separation in Organic Semiconductor Photovoltaic Diodes

See allHide authors and affiliations

Science  31 Jan 2014:
Vol. 343, Issue 6170, pp. 512-516
DOI: 10.1126/science.1246249

Early Separation

In photovoltaic devices, electrons excited by the absorption of light must travel across a junction, while the positively charged “holes” they leave behind effectively migrate in the opposite direction. If the electrons and holes do not separate efficiently, they can recombine and fail to produce any appreciable current. Gélinas et al. (p. 512, published online 12 December; see the Perspective by Bredas) studied this separation process by ultrafast optical absorption spectroscopy in thiophene-derived donor-fullerene acceptor systems common in organic photovoltaics and report a rate significantly faster than simple charge diffusion would suggest. The results implicate a coherent charge delocalization process, likely to involve fullerene π-electron states.

Abstract

Understanding the charge-separation mechanism in organic photovoltaic cells (OPVs) could facilitate optimization of their overall efficiency. Here we report the time dependence of the separation of photogenerated electron hole pairs across the donor-acceptor heterojunction in OPV model systems. By tracking the modulation of the optical absorption due to the electric field generated between the charges, we measure ~200 millielectron volts of electrostatic energy arising from electron-hole separation within 40 femtoseconds of excitation, corresponding to a charge separation distance of at least 4 nanometers. At this separation, the residual Coulomb attraction between charges is at or below thermal energies, so that electron and hole separate freely. This early time behavior is consistent with charge separation through access to delocalized π-electron states in ordered regions of the fullerene acceptor material.

Organic photovoltaic cells (OPVs) consist of a nanostructured blend of donor (D) and acceptor (A) semiconductors (1, 2). Photons absorbed in either material create molecular excitons, which can dissociate at the D-A heterojunction into holes on D and electrons on A (3, 4). The hole and electron are still subject to their mutual Coulomb interaction and can self-trap at the heterojunction (58), giving rise to charge transfer (CT) excitons. However, in efficient OPV blends that use fullerenes as the acceptor, electron and hole escape from the heterojunction and long-range charge separation is efficient (9, 10) (Fig. 1A). The motion of charges away from the heterojunction had been generally attributed to diffusion (11, 12); however, recent results have suggested that delocalized states may play a role in this process (9). Here we directly measure the electron-hole separation process at the heterojunction and find that the Coulomb barrier is surmounted at times as short as 40 fs, suggesting that rapid charge motion away from the interface through delocalized band states is necessary for long-range charge separation.

Fig. 1 Schematics of interfacial photophysical processes in OPVs (and molecules studied).

(A) Overview of charge photogeneration at a heterojunction. Light absorption generates excitons in the bulk (1a) and at interfaces (1b). When next to an interface, excitons undergo rapid charge transfer into an electron-hole pair (2), generating a dipole-like electric field in its surroundings (Embedded Image). The electron and hole then separate further and form free charges (3). (B) Stark shift of the absorption spectrum (S) due to an electric field Embedded Image (dark blue) and the resulting electro-absorption signature (red) calculated from the difference between the shifted and unshifted spectra. The EA amplitude is proportional to Embedded Image. (C) Chemical structure of PC61BM (gray) and PCDTBT (red). (D) Chemical structure of PC71BM (gray) and p-DTS(FBTTh2)2 (blue). (E) Absorption spectra of the molecules presented in (D).

To temporally resolve the electron-hole separation process, we require a probe that is sensitive to the distance between these charges. The electric field generated as the charges separate (Fig. 1A) (1315) serves this purpose by shifting the energy levels of neighboring molecules, causing a change in their electronic transition energies and an associated optical electro-absorption (EA) (16) signature, represented schematically in Fig. 1B (17). We measured these EA signals with sub–30 fs precision, using transient absorption (TA) spectroscopy. This allows us to calculate the energy stored in the electric field as the charges separate and hence the mean electron-hole distance as a function of time, as we can calibrate the time-resolved data against steady-state EA measurements.

We studied two high-efficiency model systems. The first consists of blends of a solution-processable small molecule (18), p-DTS(FBTTh2)2, ((7,7'-(4,4-bis(2-ethylhexyl)-4H-silolo[3,2-b:4,5-b']dithiophene-2,6-diyl)bis(6-fluoro-4-(5′-hexyl-[2,2'-bithiophen]-5-yl)benzo[c][1,2,5] thiadiazole))), as electron donor with PC71BM, (phenyl-C71-butyric acid methyl ester), as acceptor (19). The second system consists of blends of the polymer PCDTBT (20), (poly[N-11"–henicosanyl-2,7-carbazole-alt-5,5-(40, 70- di-2-thienyl-20, 10, 30-benzothiadiazole)]), as electron donor with PC61BM, (phenyl-C61-butyric acid methyl ester), as acceptor. Figure 1, C to E, shows their molecular structure and absorption spectra.

The small-molecule fullerene blend system [p-DTS(FBTTh2)2:PC71BM] was chosen as it exhibits sharper optical transitions than literature-standard polymer-fullerene blends. As we develop below, this property leads to a strong EA response that enables us to separate this feature from the other excited-state absorption features. We investigated two blends with different donor/acceptor composition, 60:40 and 90:10. Blends containing a 60:40 weight ratio of p-DTS(FBTTh2)2:PC71BM, processed from chlorobenzene with 0.4% diiodooctane (DIO) as a solvent additive, achieve very high internal quantum efficiency (IQE) and power-conversion efficiencies (PCEs) above 7% (19, 21). They have been shown to comprise crystalline regions of p-DTS(FBTTh2)2 and PC71BM aggregates (21). We contrast this system with the 90:10 blend, processed from chloroform, which shows much lower quantum efficiency and for which the low-fullerene fraction precludes aggregation of the fullerene (see supplementary materials) (2224).

PCDTBT is a widely studied noncrystalline polymer, which when blended with PCBM gives IQEs close to 100% and PCEs as high as 7% (25, 26). As has been discussed previously, PCBM effectively intercalates into the polymer side chains. Thus, at low fullerene loading the PCBM is mainly unaggregated, forming a mixed phase with the polymer (23). As the fullerene content is increased, the excess fullerene begins to form aggregates (27). Here we contrast a low PCBM loading, 20% (4:1), which does not show fullerene aggregation, to the optimized 80% (1:4) loading that shows high quantum efficiencies and gives rise to aggregated fullerene domains.

Figure 2, A to C, shows TA data for the 90:10 and 60:40 p-DTS(FBTTh2)2:PC71BM blends covering the visible and near-infrared spectrum from 100 fs to 2 ns. The positive features between 600 and 700 nm are due to bleaching of the ground-state absorption features. The negative signal peaking at 1230 nm (extending up to 1500 nm) is due to the photoinduced absorption (PIA) of singlet excitons, in agreement with measurements on pristine films of p-DTS(FBTTh2)2 (fig. S2). The negative signal that peaks at 800 nm is due to the PIA from positive charges (polarons) on the donor. Its amplitude is similar for both blends, indicating that the density of photogenerated holes is comparable. However, the subsequent behavior is very different. In comparison to the 90:10 blend, the 60:40 blend shows a more complex behavior around 710 nm. The strength of the bleach is reduced at early times, and a negative response is observed beyond 10 ps. This is clearly seen (Fig. 2D) in the spectra at 100-ps delay of the two blends, of a neat film of p-DTS(FBTTh2)2 (100:0), and of the steady-state EA spectra of p-DTS(FBTTh2)2 (measured independently; see fig. S7). When compared to the neat film, the 90:10 blend shows a reduction in stimulated emission at wavelengths >720 nm, indicating efficient exciton quenching. In addition, the 60:40 blend also shows a large negative signal peaking at 710 nm, which matches the measured steady-state EA spectrum in this region.

Fig. 2 Transient absorption spectra of excited states in p-DTS(FBTTh2)2:PC71BM films.

(A and B) Visible and near-infrared measurements of a 60:40 film prepared identically to the active layer of an efficient OPV device. The samples were excited with 700-nm laser pulses at a fluence of 2 μJ/cm2. (C) Measurement of a 90:10 film, where PC71BM aggregation is prevented and charge separation is inefficient. (D) TA time slices of the 100:0, 90:10, and 60:40 films normalized at 640 nm. The steady-state EA signature, obtained from a device using a steady-state electro-absorption measurement (CW), is proportional to the first derivative of the absorption spectrum.

From the data presented in Fig. 2, we identify two time scales for charge generation. We consider that the early subpicosecond dynamics are due to the dissociation of excitons photogenerated at the heterojunction, whereas the later evolution is due to excitons photogenerated in bulk p-DTS(FBTTh2)2 domains that diffuse before reaching heterojunctions where they dissociate. We use a global analysis to separate spectral features that have different time evolutions (see supplementary materials). Figure 3A shows the separation of the different features in the 60:40 blend over the time range 30 to 2500 fs. The evolution of the singlet exciton and hole absorption spectra is described by two time constants of 82 ± 5 fs and 22 ± 0.1 ps (with 70% weight to the 82-fs component); the exciton component decreases and the charge component grows with time (Fig. 3B). Our global analysis reveals a third feature with distinct temporal dynamics. This feature closely resembles the steady-state EA spectrum (Fig. 3A), confirming the presence of an EA signature in the 60:40 blend. The singlet spectrum in Fig. 3A is in good agreement with that of neat p-DTS(FBTTh2)2 films in Fig. 2D, where the signal is due only to singlet excitons. The signature of the hole polaron matches the spectrum of the 90:10 blend (Fig. 2D) but differs substantially from the 60:40 blend, where features due to both holes and EA are present. This comparison indicates that charge generation occurs in the 90:10 blend without inducing an EA signature, in contrast to the 60:40 blend.

Fig. 3 Excited-state and EA dynamics of p-DTS(FBTTh2)2:PC71BM and PCDTBT:PC61BM blends.

(A) Spectral signatures present in the p-DTS(FBTTh2)2:PC71BM (60:40) blend. The signals are obtained from a global fit spanning the entire visible range (see text for details). The data set spans 30 to 2500 fs, and the sample was excited with broadband (525 to 625 nm) laser pulses compressed to 30 fs at a fluence of 8 μJ/cm2. The reference EA is obtained independently (see Fig. 2D). (B) Dynamics of the S1, hole, and EA amplitudes obtained by fitting the data set with a linear combination of the excited-state signatures shown in (A). The yellow zone indicates the region over which artefacts prevent accurate measurement of the EA. (C and D) Spectral signatures and their associated dynamics for three different compositions of PCDTBT:PC61BM blends (see supplementary materials for details of the analysis). The samples were excited at a fluence of 2 μJ/cm2, and the reference EA spectrum was measured on a device. (E) Total energy stored in the electric field per electron-hole pair (at room temperature and 4 K). The EA is converted to a spatially integrated electric field square using a calibration from steady-state measurements (Embedded Image, where CEA is the calibration constant for a given system (see supplementary materials for details). The time-resolved energy per hole (Embedded Image) is obtained assuming that half of the field is in the donor phase. The low temperature (4 K) measurement grows continuously up to 1 ns, where it saturates at ~200 meV (see supplementary materials).

Turning to the dynamics of the EA feature (Fig. 3B), we note that some oscillations are present within the first picosecond, which are due to a nonresonant artefact, and this limits our reconstruction of the EA kinetics to delays ≥300 fs. At 300 fs, we already observe a large signal, more than half of the value at 2 ps, beyond which the signal saturates. At low temperatures (4 K), the early time response is the same, but the later growth is delayed (fig. S4). As noted above, the EA feature is absent from the 90:10 blends where charges are generated efficiently but are unable to separate away from the heterojunction. This implies that the EA of a barely separated charge pair is negligible, and hence the large EA signal occurs due to longer-range charge separation. The EA signature in the 60:40 blend thus allows us to resolve the separation of electron-hole pairs.

Figure 3D shows the spectral signatures of singlet and charge features in 4:1 and 1:4 blends of PCDTBT:PCBM. The spectral signatures of charges in the two blends differ substantially, with the 1:4 blend having a more pronounced negative feature at 640 nm. This is qualitatively similar to the behavior of the p-DTS(FBTTh2)2:PC71BM blends where the 60:40 blend, which shows EA, has a more pronounced negative feature at 710 nm than the 90:10 blend, which does not show EA. Using the difference between the 4:1 and 1:4 PCDTBT:PCBM blends allows us to extract a third spectral component for the 1:4 blend (purple line in Fig. 3D; see supplementary materials for details) that matches the steady-state EA spectrum. The dynamics of the blends are shown in Fig. 3C: The singlet exciton is rapidly quenched (86 fs) by charge transfer in both the 4:1 and 1:4 blends, with corresponding growth in the charge population. There is no EA response for the 4:1 blend (Fig. 3D), but the 1:4 blend shows a large EA signal at 40 fs. For this system, the nonresonant artefact is not as pronounced as for the p-DTS(FBTTh2)2:PC71BM, allowing accurate EA detection at delays as short as 40 fs.

By performing quasi–steady-state electro-absorption measurements, which measure the EA signature induced by an oscillating electric field on p-DTS(FBTTh2)2:PC71BM and PCDTBT:PCBM diode devices, we can calibrate the EA amplitude to the electric field present in the semiconductor layer (see supplementary materials for details). Using this calibration, we convert the transient EA amplitude of Fig. 3 to a spatially averaged value of the square of the electric field. We calculate the root mean square field strength, and hence the energy stored in the electric field, as Embedded Image.

Figure 3E shows the energy stored in the electric field per charge pair as a function of time for the various blends studied (taking a value of 3.5 for ε and modeling half the field to be present in the donor phase). For the 60:40 p-DTS(FBTTh2)2:PC71BM, the energy per charge pair reaches more than half its maximum value by 300 fs, and for the 1:4 PCDTBT:PCBM blends, it attains its saturation value by 40 fs. The energies involved are substantial, well above kBT (where kB is the Boltzmann constant and T is temperature), and provide direct evidence that considerable work must be done to overcome the Coulomb attraction between the separating electron hole pairs in these low–dielectric constant materials. Crucially, most of this work against the Coulomb interaction is done on ultrafast time scales. This requires that charges undergo rapid spatial separation on these time scales.

With simple electrostatic modeling and the assumption that the initial step of photoinduced electron transfer just across the heterojunction causes an average separation of electron and hole by 1.5 nm in the absence of fullerene aggregation, we find that charges reach a separation of 4 to 5 nm in both systems at the earliest times measured (see supplementary materials). The further growth for the 60:40 p-DTS(FBTTh2)2:PC71BM blend on time scales greater than 300 fs is most likely due to the diffusive motion of the hole within the crystalline p-DTS(FBTTh2)2 domains. For PCDTBT, this is not observed, as the larger fullerene domains lead to an increased initial separation, resulting in the saturation of the EA signal. (This occurs when the separation reaches 5 to 6 nm; see supplementary materials for details.)

The length scales being considered here for the separation of charges, ≈4 to 5 nm, are consistent with known OPV morphology in efficient systems, where the presence of pure domains of fullerene, 5 nm in length (28, 29), is strongly correlated with suppressed geminate recombination and improved device performance. In our measurements, the presence of fullerene aggregates leads to large EA signals and electrostatic energies at short times, strongly suggesting that they are key to enabling ultrafast long-range charge separation.

Our measurements are particularly sensitive to the early-time separation of charge carriers at heterojunctions, due to our direct measurement of |E2|, which is insensitive to the initial direction of charge separation with respect to charge collection electrodes (which we expect to be randomly directed, given the bulk heterojunction morphology of these materials). This initial process is not dependent on the presence of an external electric field. By contrast, other time-resolved techniques, such as electric field–induced second harmonic (EFISH) (30), are sensitive to a change in |E|. In such experiments, this change in |E| arises when carriers drift in the field set up by the electrodes and the external circuit. This necessarily builds up at later times. For example, Vithanage et al. (30) used EFISH to measure net charge motion due to drift toward the electrodes of 2 nm at 1 ps and 5 nm at 100 ps. This provides complementary information to our measurements of the first steps in charge separation (see supplementary materials for details).

Based on the above results, we propose a simple phenomenological model of the heterojunction interface. Films without aggregated fullerene domains do not exhibit a substantial EA response, which implies that the mobile component of the charge pair is the electron at early times. We model the electron motion on a nanoscale face-centered cubic lattice of acceptor sites, consisting of localized single-electron energy levels that are coherently coupled to their nearest neighbors (a tight binding model), giving rise to a band of delocalized eigenstates with bandwidth B. We include a Gaussian distribution of acceptor-site energies to introduce disorder. As expected, we find that when the static disorder of the on-site energies is within the bandwidth, disorder does not localize the states. The initial excitation is a single exciton on an adjacent donor site.

We include a Coulomb well, of depth W, surrounding the donor to model the presence of the hole, which we assume does not move during the first 200 fs of charge separation. The well lowers the energies of electron states in the local vicinity of the donor (Fig. 4A), introducing a set of bound states and narrowing the energetic width of delocalized states to ≈B-W. Typical structures of the emergent eigenstates are illustrated in Fig. 4A1-2), with corresponding examples of the actual states given in fig. S11. We emphasize that our model is only valid for a few hundred femtoseconds after exciton dissociation when delocalized states are briefly accessed; after this, polaron formation will localize the electronic states and holes and electrons will move via incoherent hopping with comparable mobilities.

Fig. 4 Model of initially accessible electronic states in fullerene derivatives and calculated electron-hole separation distance.

(A) Excited states before and immediately after charge transfer. When excitons (S1) dissociate at interfaces with aggregated PC61/71BM, the isoenergetic charge transfer places the electrons in delocalized band states, where they undergo wave-like propagation within the PC61/71BM aggregate. In this model, the hole at the interface induces a well of depth W, reducing the width of the band from its bulk value B to ~B-W. This system can sustain two typical electron wave functions represented as ϕ1,2. The electron is either trapped at the interface (ϕ1) or propagating through the band (ϕ2). (B) Calculation of electron-hole separation dynamics per charge pair for (i) injection of a fully coherent electron wave packet; (ii) tunneling of the electron into delocalized states (Fermi golden rule, FGR); and (iii) tunneling of the electron into localized states (i.e., Marcus-type electron transfer). The multiple lines represent different values of disorder and couplings, spanning 100 to 200 meV and 30 to 50 meV, respectively (31, 32). r0 is the separation distance of the initial next-neighbor charge-transfer state [typically 1.5 nm (9)], and l is the length of the PC61/71BM aggregate over which the wave function is delocalized (5.25 nm for the calculation shown).

If the energy of the electron on the donor site lies within the spread of fullerene eigenstates at the interface, the electron can undergo resonant transfer. When this energy lies below the conduction edge (CE, Fig. 4A), the initial electron wave function can only mix with bound states localized near the hole (ϕ1), forming a bound charge-transfer state. However, if the initial electron energy lies within the conduction channel above this edge, then the initial state is mixed with states that are fully delocalized across the fullerene lattice (ϕ2), enabling long-range charge separation to occur.

The model described above assumes spatial coherence between acceptor sites (delocalization). To describe the dynamics of the system, we consider two cases: (i) incoherent transitions between a localized donor site and the delocalized acceptor eigenstates, or (ii) a fully coherent case where the initial electron is described as phase-coherent superposition of these delocalized eigenstates. We perform simulations for both these models on a 5.25 nm by 5.25 nm by 5.25 nm lattice of acceptor sites and for a range of couplings and disorder strengths between 30 and 50 meV and 0.1 and 0.2 eV, respectively (31, 32); all parameters are described in full in the supplementary materials.

Incoherent transitions arise from a perturbative treatment of the D-A coupling [case (i), see supplementary materials]. This generates a set of Fermi golden rule transition rates from the localized donor site into the delocalized states described earlier. Results are shown in Fig. 4B, yellow curves. We find that charge separation occurs via a single hop of ~4 nm. For case (ii), we evolve the coherent superposition under the Schrödinger equation, which generates a wave packet that rapidly propagates across the Coulomb well and into the acceptor crystallite (see supplementary materials for details). Results are shown in Fig. 4B, red curves. Again, for all parameter values considered, electron and hole separate by 3 to 5 nm within 300 fs. For both cases, the separation distance is determined primarily by the size of the acceptor aggregate in our model. Currently, our experiments cannot differentiate between these two separation mechanisms.

We also consider charge dynamics under fully incoherent localized dynamics (Marcus theory)—i.e., in the absence of delocalization—for the same parameters (see supplementary materials for details). This leads to rapid exciton dissociation, with hole and electron on nearest-neighbor sites (Fig. 4B, blue curve), to form tightly bound CT states that are not expected to dissociate rapidly.

Our demonstration and model of short-time charge delocalization and coherent motion is very different from models for exciton coherence and delocalization that have been suggested as being key to efficient OPV performance (33). Within our study, we find no need to invoke such excitonic processes (see supplementary materials for details).

Our results rationalize the apparent asymmetry between efficient electron-hole capture in organic light-emitting diodes (34) and near-unity photoconversion quantum efficiencies in OPVs (25) by revealing that ultrafast charge separation through delocalized band-like states in fullerene aggregates is key to efficient charge separation. Moreover, the fast time scale for this process indicates that efficient charge separation requires no excess energy beyond that needed to overcome the Coulomb interaction. This is in contrast to Onsager-like models that require excess energy in hot states. Our results suggest that the real energy loss during charge separation lies elsewhere—for instance, in later energetic relaxation of charges through polaron formation or the presence of defect-mediated gap states—and that such energy losses are not fundamental for efficient charge separation.

Supplementary Materials

www.sciencemag.org/content/343/6170/512/suppl/DC1

Supplementary Text

Figs. S1 to S14

References (3544)

Reference and Notes

  1. Acknowledgments: We thank the Engineering and Physical Sciences Research Council, and the Winton Programme (Cambridge) for the Physics of Sustainability for funding. S.G. acknowledges funding from the Fonds québécois de recherche sur la nature et les technologies; A.R. thanks Corpus Christi College for a Research Fellowship; A.K. thanks National Research Foundation Singapore for a scholarship; J.C. thanks the Royal Society for a Dorothy Hodgkin Fellowship; and T.S.v.d.P. acknowledges funding from the Center for Energy Efficient Materials, an Energy Frontier Research Center funded by the U.S. Department of Energy, Office of Science, Basic Energy Sciences under Award DC0001009.
View Abstract

Navigate This Article