JOURNAL OF APPLIED PHYSICS

VOLUME 96, NUMBER 12

15 DECEMBER 2004

Conjugated polymer/molten salt blends: The relationship between morphology and electrical aging F. Habrard, T. Ouisse,a) O. Stephan, and M. Armand Laboratoire de Spectrométrie Physique, Centre National de la Recherche Scientifique-Unité Mixte de Recherche C5588, Université Joseph Fourier, Grenoble 1, BP 87, 38402 Saint-Martin d’Hères Cedex, France

M. Stark Laboratoire de Spectrométrie Physique, Centre National de la Recherche Scientifique-Unité Mixte de Recherche C5588, Université Joseph Fourier, Grenoble 1, BP 87, 38402 Saint-Martin d’Hères Cedex, France Centre National de la Recherche Scientifique, Laboratoire d’Etudes des Propriétés Electroniques des Solides, BP 166, 38042 Grenoble Cedex 9, France, and European Synchrotron Radiation Facility, BP 220, 38043 Grenoble Cedex 9, France

S. Huant Laboratoire de Spectrométrie Physique, Centre National de la Recherche Scientifique-Unité Mixte de Recherche C5588, Université Joseph Fourier, Grenoble 1, BP 87, 38402 Saint-Martin d’Hères Cedex, France

E. Dubard European Synchrotron Radiation Facility, BP 220, 38043 Grenoble Cedex 9, France

J. Chevrier Centre National de la Recherche Scientifique, Laboratoire d’Etudes des Propriétés Electroniques des Solides, BP 166, 38042 Grenoble Cedex 9, France and European Synchrotron Radiation Facility, BP 220, 38043 Grenoble Cedex 9, France

(Received 22 July 2004; accepted 27 August 2004) Mixing molten salts with luminescent conjugated polymers provides the ability to lower the current threshold of organic electroluminescence (EL) devices. However, this process generally results in unwanted premature electrical aging. We used dynamic atomic force microscopy (AFM) in the electric force detection regime to study the phase microseparation occurring in the salt-polymer blend. We found that in the spin-coated layers, the molten salt most often forms discotic, roughly self-organized sub-microdomains. Their size and density strongly depend on the polymer side chains and overall molecular weight, on the molten salt nature, and on their respective concentrations in solution. We show that the diode current is injected into the vicinity of the interface between the salt and polymer domains, since it is proportional to the perimeter of the salt domains per surface unit, as estimated from the AFM images. The electrical aging is attributed to the degradation of the interface between the polymer and salt domains. This is further corroborated by a number of electrical data. © 2004 American Institute of Physics. [DOI: 10.1063/1.1808230] doping.2–8 Recently, we proposed to use a molten salt as the ionic liquid, replacing the conventionally chosen lithium triflate salt (LiTf).5–8 The use of a molten salt compatible with organic polymers allows us to avoid the introduction of poly(ethylene oxide) (PEO) into the blend (this latter product is usually required for complexing the ionic species and making them mobile under the application of an electric field.2–4 Even for thick layers, adding ionic liquid results in a considerable lowering of the current threshold voltage. Notably, it is possible to obtain a very low threshold with “any” kind of electrode, e.g., Al and even ITO cathode.7 Unfortunately, such devices are known for switching on more slowly and degrading faster than their conventional OLED counterparts.2–8 We treated the slow switching problem in a previous paper and showed that in our devices, it was mostly due to the interface roughness.6 In addition to conventional oxidation problems,8 such devices suffer from a dramatic decrease of their current drive capability after a few hours of continuous operation. To briefly illustrate the particular aging kinetics of the LEC’s, we have plotted in Fig. 1 the variation

I. INTRODUCTION

“Plastic-based” electronics has not yet reached industrial maturity, but research carried out on such devices has greatly progressed during the last decade. In particular, the most advanced topic deals with the fabrication of efficient light emitters for producing low-cost displays.1 Organic lightemitting diodes (OLED’s) made from small organic molecules are already commercialized in small area displays, and much effort is being put into the task of producing conjugated polymer-based OLED’s exhibiting satisfying operating lifetimes. In this paper, we focus on the case of organic light-emitting electrochemical cells (LEC). The operating principle of such devices consists in mixing a conducting polymer with an ionic salt so as to lower the effective energy barrier acting against the carrier injection. This is allowed by the creation of charged electrolytic double layers at the electrode-active layer interfaces and bulk electrochemical a)

Author to whom correspondence should be addressed; FAX: 33-4-76635495; electronic mail: [email protected]

0021-8979/2004/96(12)/7219/6/$22.00

7219

© 2004 American Institute of Physics

Downloaded 03 Aug 2006 to 193.48.255.141. Redistribution subject to AIP license or copyright, see http://jap.aip.org/jap/copyright.jsp

7220

J. Appl. Phys., Vol. 96, No. 12, 15 December 2004

FIG. 1. Variation of the EL signal vs time for LEC’s made from PF14 and THA-TFSI. The parameter is the stress voltage.

of the electroluminescence (EL) signal versus time with the stress voltage as a parameter. The data are for PF14 and THA-TFSI (see their definition in Sec. II below). In Fig. 1, the initial EL rising step is due to the formation of the electrical double layers at both interfaces, and depends on the surface roughness and blend quality near the electrodes.6 However, after a relatively short time, one systematically observes a dramatic fall of the electroluminescence, which is accompanied by a similar decrease of the current drive capability. The higher the stress voltage, the steeper the fall. Our best devices “last” only for a couple of hours with a bias at a few volts above the LEC threshold. Since this premature device failure is not observed with conventional OLED’s made from the same polymer, it must be ascribed to the LEC structure itself. It occurs independently of the nature of the polymer and salt. The aim of this paper is to elucidate the physical cause of this premature degradation and to provide unambiguous experimental data so as to clearly identify it. We will show that it is neither due to a complete electrochemical degradation of the salt nor to a complete oxidation of the conjugated polymer but to a degradation of the interface between both kinds of domains. Both the degradation mechanism and the initial device performance can be straightforwardly related to the microscopic morphology of the blend. We will present a detailed investigation of our active layers through atomic force microscopy (AFM) observations. AFM measurements conducted on polymer-salt blends have been recently reported by Wenzl et al. in the topographic mode.9 They showed that introducing LiTf in ladder-type poly(p-phenylene) results in drastic changes in the surface topology, from which it is possible to infer and quantify the microphase separation. Here, we used our AFM in a somewhat different fashion, since we made our observations not only in the dynamic mode but also in a noncontact mode involving electric force detection [electric force microscopy (EFM) (Refs. 10–12)]. To do so, we control the electrical bias applied between the conducting tip and the sample. As long as the electric field is kept to moderate values, the conjugated polymer domains behave as an insulator, whereas the molten salt domains are highly polarizable, leading to a strong and long-range electric interaction with the AFM tip. Hence, it is possible to accurately map the boundaries be-

Habrard et al.

tween the salt and the polymer microdomains. We conducted such experiments on a wide range of samples based on polyfluorene (PF) as the conjugated polymer, varying either the nature of the side chains, the overall molecular weight, the molten salt, or the solution concentrations. In all cases, we observed a micro- or sub-microphase separation. The molten salt usually forms small discotic domains, whose distribution is roughly self-organized. We related the extent of this phase separation to the physical parameters listed above. We fabricated LEC’s keeping the same parameter values and observed that the current drive capability is indeed proportional to the perimeter of the salt domains per unit surface, as deduced from the EFM imaging. This clearly indicates that most of the current injection takes place at the boundary between both kinds of domains. To corroborate our analysis, we also present in the last section a number of experimental data, which all indicate that the premature failure of our device is due to the degradation of the interface between the molten salt and the polymer domains. We also show that the best blend quality is obtained when PEO side chains are grafted onto the polyfluorene units. II. DEVICE FABRICATION AND AFM IMAGING

We synthesized four different PF materials following protocols already described elsewhere.5–8,13 The polymers are, respectively, poly[9,9-dinonyl-fluorene2,7-diyl] (denoted as PF9), poly[9,9-ditetradecan-fluorene2,7-diyl] (denoted PF14), a random co-polymer poly[(9,9ditetradecan-fluorene-2,7-diyl)-co-(9,9-dioctyl-fluorene-2,7diyl)] (denoted as COPF14-8), and a polyfluorene with one PEO side chain, poly[9-(7,10,13,16-tetraoxaheptadecyl), 9⬘-(tetradecane)-fluorene-2,7-diyl] (denoted as PEOPF) (Ref. 14). The molten salts are all based on the use of alkylated cations in order to make them compatible with organic polymers and organic solvents. We used tetrahexylammoniumbis(trifluromethyl-sulfonyl-imide) THA-TFSI, tetraoctylammoniumbis(trifluoromethyl-sulfonyl-imide) (TOA-TFSI), trihexyl(tetradecyl)-phosphoniumbis(trifluoromethyl-sulfonyl-imide) (THTDP-TFSI), and (2,4,4-trimethylpentyl)phosphinate (THTDP-TMPP). The ammonium-based salts were synthesized in our laboratory, while the two phosphonium-based salts were bought from Strem Chemicals, Inc. Further, general information about the applications and interest of polyfluorenes can be found, for instance, in Ref. 15. Polymer-molten salt mixtures were mostly prepared in chloroform solutions, but as well in toluene or dichloromethane without significantly different results. The layers were formed by spin coating at a rotating speed equal to 1000 rpm, resulting in a layer thickness around 100 nm. Diode fabrication has already been extensively detailed, and we refer the reader to Refs. 5–8 for additional information. We prepared either spin-coated layers of various blends without metallization on top or diodes with the same active layers and a top Al metallization. The bottom layer was ITO in all cases. In all devices, this ITO layer was not etched, and thus formed a metalliclike conducting plane covering the whole substrate. For the LEC’s, the cathode was defined by evapo-

Downloaded 03 Aug 2006 to 193.48.255.141. Redistribution subject to AIP license or copyright, see http://jap.aip.org/jap/copyright.jsp

J. Appl. Phys., Vol. 96, No. 12, 15 December 2004

rating Al through a shadow mask, forming circular contacts with a 2-mm diameter. The diodes were electrically probed with a relatively wide, polished, and highly flexible metallic strip to prevent polymer damage during electrical monitoring. The tests were carried out inside a dry glove box under argon ambient. The AFM experiments were carried out on a commercial AFM Nanoscope III Dimension 3100 from Veeco Instruments. Here, we note that both the polymer and the molten salt are hydrophobic (but PEOPF). Thus, it was possible to conduct the image scans under air without being appreciably perturbed by humidity problems. The silicon tips were coated with tungsten carbide, with a spring constant around 3 N / m, a resonance frequency around 60 kHz, and a quality factor around 300. We use the AFM in the dynamic mode, driving the cantilever mechanically close to its natural resonance frequency. Imaging the topography of the sample without accounting for its electrical properties results in smooth topographical images, which do not permit us to identify different microdomains. The overall rms surface roughness is in the 2-nm range, whatever is the molten salt-polymer blend used. In order to detect the molten salt microdomains, it is necessary to use the specificity of electric force microscopy and to apply a potential between the sample and the conducting tip. For reasonable voltage values (i.e., a few volts) imposed between the ITO underlying electrode and the AFM tip, the conjugated polymer behaves as a dielectric material, whereas the dissociated salt formed from free anions and cations is highly polarizable. The resulting capacitive coupling between the salt domains and the tip results in long-range forces modifying the tip oscillations whenever it is positioned above a molten salt domain, even in the noncontact mode. Indeed, the difference between the potential of zero charge of the molten salt and the work function of the tip metal is not zero, so that applying zero voltage between the ITO and the tip is already sufficient for creating charged interfacial layers at the bottom and at the top of the salt domains. Hence, in the following, all the images have been obtained by simply grounding both the ITO underlying layer and the conducting tip. First, an approach-retract curve is completed in a polymer area (which indeed constitutes most of the sample surface so that in most cases, the point can be randomly chosen). Then, a feedback amplitude is selected very close to the unperturbed amplitude, so that the cantilever is rather far from the surface (from 50 to 100 nm) and is only weakly affected by the sample topography. A surface scan is eventually proceeded. III. RESULTS

Typical AFM phase images are shown in Fig. 2. Figure 2(a) illustrates the case of COPF14-8 associated to TOATFSI. It is clearly seen that the molten salt forms submicrodomains of roughly circular shape. It is also clearly seen in Figs.2(a)–(e) that the distance between the tip and the sample is large enough for the cantilever not to be sensitive to the polymer areas but to the electric force induced by the

Habrard et al.

7221

FIG. 2. EFM phase imaging of spin-coated layers made from a blend of polyfluorene and molten salt in the dynamic and noncontact modes.

polarized molten salt only. The disk diameter is, in general, larger than the layer thickness, so that it is reasonable to consider that the molten salt domains extend from the top to the bottom of the active layer. This will be indirectly confirmed by the electrical data (Sec. IV). It is worth noticing that at tip-sample distances corresponding to the onset of electrical influence, one observes a small hysteresis in the approach-retract curves. This instability may induce oscillations in the phase and amplitude, depending on the scan speed. These oscillations are, for instance, clearly visible in the molten salt domains appearing in Fig. 2(c). For small salt concentrations, which correspond to the values actually used in our LEC’s, the domains are circular, the smallest sizes being around 100 nm. However, increasing this concentration to higher values leads to a progressive deformation of their shape and even to a coalescence between some domains [see Fig. 2(b)]. Although not shown here, a further increase (ratio of 50% weight weight between the polymer and salt) leads to the formation of macrodomains with irregular shapes and sizes above 10 ␮m, visible with an optical microscope. Only one salt behaves in a qualitatively different manner: in Fig. 2(f), it can be seen that using a phosphinate anion leads to an undesired macrophase separation (this picture was acquired from an optical micro-

Downloaded 03 Aug 2006 to 193.48.255.141. Redistribution subject to AIP license or copyright, see http://jap.aip.org/jap/copyright.jsp

7222

J. Appl. Phys., Vol. 96, No. 12, 15 December 2004

scope). Accordingly, LEC’s fabricated with THTDP-TMPP produced neither current nor electroluminescence at reasonable voltage values. Depending on the polymer and salt nature, we often observed that the size distribution was double peaked around two characteristic size values [see, e.g., Figs. 2(c) and 2(d)]. In such cases, the disks tend to roughly self-organize: one can see in Figs. 2(c) and 2(d) that the smaller domains are roughly aligned along circular lines located at a given distance from the bigger ones. We think that when cations and anions begin to aggregate to form a “large” disk, this domain progressively pumps all the molten salt from its surroundings and expels a corresponding polymer volume, so that ultimately, the disk size gets limited around a given diameter, encircled by a polymer ring, whose width scales accordingly. Then, no other domain can form in the ring area due to a complete lack of ionic liquid, and smaller disks can only form around those rings, hence, drawing circular patterns around the large diameter salt domains. We note that this argument can explain as well why the diameters in the samples with a single-peaked distribution also present a reasonable regularity in size, since they must be limited by the formation of a polymer ring around the salt domains [see, e.g., Fig. 2(e)]. From the AFM images of Fig. 2, it is clear that the blend morphology is eminently dependent on the polymer and salt nature. We summarize below a number of qualitative observations related to that point. The overall molecular weight of the polymer influences the quality of the blend. Everything otherwise fixed, the longer are the polymeric chains, the worse is the blend; from the gel permeation chromatography, we found that increasing the chain molecular weight of PF14 from M n ⬇ 15 000 to M n ⬇ 30 000 results in a degraded blend quality, which leads in turn to a substantial degradation of both the EL and current drive capabilities. THA-TFSI gives better blends than TOA-TFSI for some polymers, but for others, TOA-TFSI is better. Nevertheless, both THA- and TOA-TFSI are better than THTDP-TFSI, whatever the polymer used is. The best blend is obtained by grafting the PEO segments onto the PF chain [see Fig. 2(e)]. This is not surprising, since PEO is known for complexing ionic species and is indeed used as a component of “conventional” LEC’s. Here, it is important to link the quality of the blend to the electrical properties of the LEC’s. Since in our thin layers, the phase microseparation occurs at the boundary of molten salt disks, in LEC’s, we expect the electrostatic influence of the salt to prevail only at the neighborhood of those structures. We thus expect the current to flow mostly close to the molten salt domain-polymer interface. Indeed, confirmation of that hypothesis is found in analyzing the electrical data, as shown in the next section. IV. CORRELATION BETWEEN BLEND MORPHOLOGY AND LEC CURRENT DRIVE CAPABILITY AND AGING

To assess the relationship between the phase microseparation and the LEC performance, we have processed a number of diodes with active layers similar to those produced for AFM imaging. The AFM images allow us to estimate the

Habrard et al.

FIG. 3. Variation of the LEC current at 5 V vs the salt domain perimeter per surface unit as extracted from EFM imaging. The LEC’s and the spin-coated layers used for EFM were processed with the same parameters (polymer = 2 g / L and salt= 1 g / L or 0.5 g / L).

overall perimeter of the molten salt domains per unit surface. In Fig. 3, we plotted the variation of the LEC current measured at 5 V as a function of the perimeter per unit surface of the salt domains. Remarkably, the current level is in good approximation proportional to the salt domain perimeter rather than its area. It is thus much probable that most of the LEC current is injected into the neighborhood of the salt domains. This naturally stems from the fact that the cations and anions, which form the electrochemical double layers, are able to electrostatically lower the effective energy barrier experienced by the carriers only close to the salt-polymer interface. We also note that this confirms our hypothesis that the salt domains identified from AFM imaging extend vertically throughout the whole layer. The electrochemical stability window of the salts, as measured from cyclic voltammetry experiments, exceeds 6 V for THA-TFSI and TOA-TFSI. However, we cannot know how this window is adjusted with respect to the internal potential inside the active layer, so that one could suspect that electrochemical degradation plays a prominent role in device failure. We can show that this is indeed not the case for reasonable stress voltage values. Figure 4 depicts the experimental variation of the imaginary part of the diode complex impedance versus frequency, in the case of PF14 associated to THTDP-TFSI. The two curves were measured before stress and after a 1-h, 5-V biasing, respectively. This electrical stress led to a current and EL reduction of about one order of magnitude at the stress voltage. However, it is seen in Fig. 4 that the impedance is not appreciably modified. Since this value reflects that of the electrochemical double-layer capacitance,7 we can infer from Fig. 4 that the current degradation cannot be ascribed to an electrochemical degradation of the salt. This feature was observed independently of the salt nature. Hence, now we can proceed onto showing that the bulk of the conducting polymer itself is not degraded, except in the vicinity of the molten salt domains. Figure 5 shows the current–voltage characteristics mea-

Downloaded 03 Aug 2006 to 193.48.255.141. Redistribution subject to AIP license or copyright, see http://jap.aip.org/jap/copyright.jsp

Habrard et al.

J. Appl. Phys., Vol. 96, No. 12, 15 December 2004

FIG. 4. Experimental dependence of the imaginary part of the LEC complex impedance on frequency. Parameters: PF14= 1.5 g / L and THTDP-TFSI = 0.6 g / L.

sured for a virgin device and after several voltage stress steps at 5.5 V, in the case of PEOPF associated to THA-TFSI. Empirically, defining a LEC threshold voltage VT as the voltage for which the current exceeds a given value I0 = 20 ␮A, the threshold voltage shift ⌬VT is plotted versus time in the inset of Fig. 5. It can be seen that the polymer is not fully destroyed but that the LEC threshold voltage is simply repelled to higher values after each step. Hence, the polymerconducting capabilities are not totally suppressed as a first analysis of the current drop might suggest, and it is always possible to recover the initial current drive by an increase in voltage. The threshold voltage shift seems to proceed logarithmically in time. These data are quite in agreement with the hypothesis that current injection takes place in the vicinity of the polymer-salt domain interface. At a given voltage, the near-interfacial current path is subject to irreversible degradation, but then increasing the voltage and, hence, the electrochemical double-layer charge, enhances the spreading of the polymer region, which is subject to the electrostatic influence of the accumulated cations and anions, so that it is possible to inject electrons and holes into polymer parts not yet affected by the electrical aging. Even the light-emission level can be partly recovered by increasing the bias voltage. We thus consider these experimental data as a good confirmation that the polymer current flow is confined close to the

7223

salt domain interface. And the fast degradation process may reasonably be attributed to the fact that in such confined domains, the current density gets high enough for the polymer to degrade at an accelerated rate. It may be premature to attribute the same physical origin to the degradation process occurring in conventional LEC’s formed with LiTf and PEO, since the morphology of such devices appear to be somewhat different from ours.2,9 Nevertheless, it is clear that these devices also suffer from phase microseparation, as demonstrated by the TEM measurements2 and detailed topographical AFM data.9 As these devices are also affected by relatively fast electrical aging, it seems natural to invoke the same kind of explanation. An improvement in reliability should come from an improvement in the blend morphology so as to reduce the local current density in the conducting paths through an increase in their overall number. V. CONCLUSION

EFM imaging turns out to be an interesting mean of visualizing the phase separation in polymer-molten salt blends, since both kinds of domains behave very differently when subjected to a weak electric field. It is thus possible to map the molten salt domains in an easy and effective way. We observed that in the case of organic-compatible salts based on ammonium or phosphonium cations, and for spincoated layers, the molten salt forms discotic domains with diameters in the deep sub- to micrometer range. The combination of EFM imaging with electrical data enabled us to formulate an explanation of the accelerated LEC degradation rate. We think that most of the hole and electron flow occurs through high current-density paths located in the vicinity of the polymer-molten salt domain interface. Hence, the final current drop is due to the irreversible degradation of the conjugated polymer-molten salt interface. An improvement in device lifetime should thus come from an improvement in the quality of the blend. In this respect, grafting the PEO segments onto the polymer backbone results in a better incorporation of the molten salt into the polymer matrix. Unfortunately, this results in a somewhat more tedious synthesis and does not give yet a significant performance improvement. Eventually, we suspect that the fast degradation already observed by other authors in the case of more conventional LEC’s also results from microphase separation. The task of determining the more appropriate functional groups to be grafted onto the polyfluorene units to enhance the blend is now in progress at our laboratory. ACKNOWLEDGMENTS

This work was financially supported by the CNRS program “Matériaux Nouveaux/Fonctionnalités Nouvelles,” France. One of the authors (M.S.) gratefully acknowledges the funding (Feodor Lynen Fellowship) by the Alexander von Humboldt Foundation, Germany. 1

FIG. 5. Variation of the LEC current vs applied voltage for different stress times with a stress voltage equal to 5.5 V. The inset shows the variation of the LEC threshold voltage as a function of stress time. Parameters: PEOPF= 2 g / L and THA-TFSI= 1 g / L.

R. H. Friend et al., Nature (London) 397, 121 (1999). Q. Pei, G. Yu, C. Zhang, Y. Yang, and A. J. Heeger, Science 269, 1086 (1995). 3 Y. Cao, G. Yu, A. J. Heeger, and C. Y. Yang, Appl. Phys. Lett. 68, 3218 (1996). 2

Downloaded 03 Aug 2006 to 193.48.255.141. Redistribution subject to AIP license or copyright, see http://jap.aip.org/jap/copyright.jsp

7224 4

F. P. Wenzl et al., Synth. Met. 102, 1138 (1999). S. Panozzo, M. Armand, and O. Stéphan, Appl. Phys. Lett. 80, 679 (2002). 6 T. Ouisse, O. Stéphan, M. Armand, and J. C. Leprêtre, J. Appl. Phys. 92, 2795 (2002). 7 T. Ouisse, M. Armand, Y. Kervella, and O. Stéphan, Appl. Phys. Lett. 81, 3131 (2002). 8 T. Ouisse, O. Stéphan, and M. Armand, Eur. Phys. J.: Appl. Phys. 24, 195 (2003). 9 F. P. Wenzl, C. Suess, A. Haase, P. Poelt, D. Somitsch, P. Knoll, U. Scherf, and G. Leising, Thin Solid Films 433, 263 (2003). 5

Habrard et al.

J. Appl. Phys., Vol. 96, No. 12, 15 December 2004 10

B. D. Terris, J. E. Stern, D. Rugar, and H. J. Mamin, Phys. Rev. Lett. 63, 2669 (1989). 11 C. Guillemot et al., Europhys. Lett. 59, 566 (2002). 12 T. Mélin, D. Deresmes, and D. Stiévenard, Appl. Phys. Lett. 81, 5054 (2002). 13 O. Stéphan, V. Collomb, J. C. Vial, and M. Armand, Synth. Met. 113, 257 (2000). 14 O. Stéphan, M. Armand, and J. C. Vial, US Patent 6,699,654 B2 (2 March 2004). 15 U. Scherf and E. J. W. List, Adv. Mater. (Weinheim, Ger.) 14, 477 (2002).

Downloaded 03 Aug 2006 to 193.48.255.141. Redistribution subject to AIP license or copyright, see http://jap.aip.org/jap/copyright.jsp