Enhanced n-Doping Efficiency of a Naphthalenediimide-Based Copolymer through Polar Side Chains for Organic Thermoelectrics.

N-doping of conjugated polymers either requires a high dopant fraction or yields a low electrical conductivity because of their poor compatibility with molecular dopants. We explore n-doping of the polar naphthalenediimide-bithiophene copolymer p(gNDI-gT2) that carries oligoethylene glycol-based side chains and show that the polymer displays superior miscibility with the benzimidazole-dimethylbenzenamine-based n-dopant N-DMBI. The good compatibility of p(gNDI-gT2) and N-DMBI results in a relatively high doping efficiency of 13% for n-dopants, which leads to a high electrical conductivity of more than 10-1 S cm-1 for a dopant concentration of only 10 mol % when measured in an inert atmosphere. We find that the doped polymer is able to maintain its electrical conductivity for about 20 min when exposed to air and recovers rapidly when returned to a nitrogen atmosphere. Overall, solution coprocessing of p(gNDI-gT2) and N-DMBI results in a larger thermoelectric power factor of up to 0.4 μW K-2 m-1 compared to other NDI-based polymers.

Doping of organic semiconductors is essential for the optimization of a number of electronic components, ranging from the hole and electron blocking layers used in organic solar cells[1−3] and organic light-emitting diodes (OLEDs)[1,4,5] to trap filling in organic field-effect transistors (OFETs)[6−8] and the legs of thermoelectric generators.[9,10] For many of these apn>plications, conjugated n>an class="Chemical">polymers are particularly intriguing because they permit one to adjust the rheological properties of processing solutions and the mechanical properties of the final (flexible) thin film architectures. Doping can be achieved through electron transfer between the semiconductor and a molecular dopant via a redox reaction. Alternatively, a proton/hydride (H+/H–) can be transferred from an acid/base to the semiconductor.[11] In the case of p-doping, positive charge carriers are introduced, whereas n-doping refers to the addition of extra electrons to the conjugated system. It is desirable that each dopant molecule that is added to the semiconductor material introduces as many charges as possible. Therefore, the presence of unreacted dopant should be avoided in order to maximize the amount of conducting material. Moreover, the presence of excess dopant has the tendency to disrupt the nanostructure of the neat semiconductor, which can negatively impact charge transport.[10,12−14] Hence, it is critical that the doping efficiency, i.e., the fraction of dopants that ultimately create a charge on the organic semiconductor, be as high as possible.[15]

To realize thermoelectric generators, both p- and n-type materials are needed. They should display a high figure of merit ZT = α2σ·T/κ, where α is the Seebeck coefficient, σ the electrical conductivity, T the absolute temperature, and κ the thermal conductivity. If the thermal conductivity, which is challenging to measure for thin film architectures, is unknown, the power factor α2σ is instead used to compare the thermoelectric efficacy of different materials. P-doping of conjugated polymers is now well established[6,16,17] and can be carried out with high efficiency and with high environmental stability, leading to a thermoelectric power factor of at least 100 μW m–1 K–2.[18−20] In contrast, n-dopn>ing continues to pose a formidable challenge because of very low dopn>ing efficiencies as well as poor stability of the dopn>ed state.[21] The most widely studied classes of n-typn>e materials include n>an class="Chemical">naphthalenediimide (NDI)-based copolymers,[14,22−29] diketopyrrolopyrrole (DPP)-based polymers,[30−32] as well as fullerenes and their derivatives.[3,8,33−45] We have compiled data from the literature to compare the dopant fractions that are required to achieve the maximum conductivity σmax through n-doping of various semiconductors (Figure ; Supporting Information Table S1). It is evident that n-doping of NDI-based polymers is limited by a too low doping efficiency. The result is either a low maximum electrical conductivity of less than 10–2 S cm–1 at low dopant fractions (Figure a; bottom left) or the need for a large dopant fraction of more than 30 mol % to achieve a higher electrical conductivity (Figure a; top right). For example, Schlitz et al. investigated n-doping of the high-mobility naphthalenediimide–bithiophene copolymer p(NDI2OD-T2)[46] with the commonly used n-dopant N-DMBI (see Figure for the chemical structure) and reached an electrical conductivity of about 10–3 S cm–1 at a dopant fraction of 9 mol %.[14] The insolubility of N-DMBI in the host polymer leading to segregation of the dopant was noted to be a limiting effect for the electrical properties. Naab et al. studied doping of several NDI-based polymers with a dimer version of DMBI and found that a dopant fraction of up to 43 mol % was required to maximize the electrical conductivity,[26] despite a higher doping efficiency, because each dimer can create two charges.[42]

Figure 1

(a) Literature values of the molar dopant fraction required to reach the maximum electrical conductivity (σmax) for n-doped NDI-based polymers (▲),[14,22−24,26,27,29] other (e.g DPP- or NTDI-based) polymers (▼),[30−32,47,48]fullerene derivatives (⧫),[8,34,36−43,45] and p(gNDI-gT2) (★, this work);  (b) corresponding Seebeck coefficient (α) at maximum electrical conductivity; empirical relation α ∝ σ–1/4.[10]

Figure 2

Chemical structures of (a) p(gNDI-gT2)[50] and (b) the molecular dopant N-DMBI.

(a) Literature values of the molar pan class="Chemical">dopant fraction required to reach the maximum electrical conductivity (σmax) for n>an class="Chemical">n-doped NDI-based polymers (▲),[14,22−24,26,27,29] other (e.g DPP- or NTDI-based) polymers (▼),[30−32,47,48]fullerene derivatives (⧫),[8,34,36−43,45] and p(gNDI-gT2) (★, this work);  (b) corresponding Seebeck coefficient (α) at maximum electrical conductivity; empirical relation α ∝ σ–1/4.[10]

Chemical structures of (a) p(pan class="Chemical">gNDI-gT2)[50] and (b) the molecular n>an class="Chemical">dopant N-DMBI.

One emerging tool to increase the doping efficiency is the replacement of nonpolar alkyl side chains with more polar oligoethylene glycol side chains, which enhances the compn>atibility of semiconductor/n>an class="Chemical">dopant pairs.[17] Li et al. have observed that the common p-dopant F4TCNQ more readily diffuses into a polythiophene that carries oligoethylene glycol side chains as well as a sulfonate group, as compared to poly(3-hexylthiophene) (P3HT), which indicates that polar side chains can improve dopant miscibility.[49] As a result, polar side chains can lead to complete p-doping efficiency of polythiophenes by F4TCNQ, resulting in both a σmax ≈ 100 S cm–1 for a low dopant fraction of 10 mol % as well as enhanced thermal stability.[13] Likewise, fullerenes that carry oligoethylene glycol side chains feature enhanced compatibility with N-DMBI and therefore a high doping efficiency of about 18%, which yielded a maximum conductivity of about 2 S cm–1 and power factor of up to 19 μW m–1 K–2.[39,43]

In this work, we explore n-doping of the naphthalenediimide–bithiopn>hene con>an class="Chemical">polymer p(gNDI-gT2) (for details on synthesis and characterization, see the Supporting Information and Figures S1 and S2), a structural analogue of p(NDI2OD-T2) with polar oligoethylene glycol-containing side chains on both the NDI acceptor and the bithiophene donor unit, which has proven to be a promising material for organic electrochemical transistors (OECTs).[50−52] We anticipate that the structural alteration from nonpolar alkyl side chains to more polar oligoethylene glycol side chains will aid doping of the polymer backbone through enhanced dopant miscibility. We chose to investigate n-doping with N-DMBI, which is thought to donate a hydride (H–),[27,42,53,54] and found that our best results in terms of doping efficiency and maximum conductivity are superior to previous results that have been reported for other n-type polymers (Figure a, bottom right, green).

In a first set of experiments, we recorded UV/vis spectra of p(gNDI-gT2) solutions (Figure a) and films (Figure b) before and after addition of n>an class="Chemical">N-DMBI. The thin film spectrum of the pristine polymer consists of a peak at around ∼440 nm and a broad spectral feature between 600 and 1500 nm, which we attribute to the π–π* transition and a strong intramolecular charge transfer complex as a consequence of strong donor–acceptor interactions.[50,55] Because Giovannitti et al. observed very little variation of the higher-energy absorption peak upon electrochemical doping, we chose to normalize all spectra to this peak for comparison. We note that for slight doping with 10 mol % N-DMBI the low-energy absorption peak slightly increases. Instead, upon additional doping, the broad spectral feature at higher wavelengths diminishes, while the absorption at around 600 nm increases relative to the peak at 440 nm after doping with N-DMBI. The latter trend is in full agreement with the study by Giovannitti et al. and previous literature on n-doping.[26,28,50] Doping results in a gradual red shift of the low-energy absorption peak from 1016 nm for the pristine polymer to 1040 nm for p(gNDI-gT2) doped with 50 mol % N-DMBI. We tentatively assign this red shift as well as the slight increase in absorption upon doping with 10 mol % N-DMBI to planarization of the polymer backbone. Interestingly, we note that the addition of N-DMBI has seemingly no effect on the solution spectra of dissolved p(gNDI-gT2). Thus, we conclude that doping of the polymer is likely to occur during the film formation step upon solvent removal.

Figure 3

(a) Solution absorbance spectra of pristine p(gNDI-gT2), p(gNDI-gT2) + 20 mol % N-DMBI (note that the spectral feature at 315 nm is due to neat N-DMBI), and neat N-DMBI in chloroform; (b) normalized absorbance spectra of pristine and N-DMBI-doped p(gNDI-gT2) films (10, 20, 30, and 50 mol % N-DMBI); (c) Arrhenius plots of variable-temperature conductivity measurements (dashed lines are fits to the Arrhenius equation, yielding the activation energies E0 and Ea); and (d) electron paramagnetic resonance (EPR) spectra of pristine and N-DMBI-doped p(gNDI-gT2) films.

(a) Solution absorbance spectra of pristine p(pan class="Chemical">gNDI-gT2), p(n>an class="Chemical">gNDI-gT2) + 20 mol % N-DMBI (note that the spectral feature at 315 nm is due to neat N-DMBI), and neat N-DMBI in chloroform; (b) normalized absorbance spectra of pristine and N-DMBI-doped p(gNDI-gT2) films (10, 20, 30, and 50 mol % N-DMBI); (c) Arrhenius plots of variable-temperature conductivity measurements (dashed lines are fits to the Arrhenius equation, yielding the activation energies E0 and Ea); and (d) electron paramagnetic resonance (EPR) spectra of pristine and N-DMBI-doped p(gNDI-gT2) films.

To obtain an estimate of the charge carrier density (n), we used the change in the activation energy of the conductivity upon doping. The estimation is based on the extended Gaussian disorder model (EGDM)[56] as reported by Liu et al.[39] The model yields a general relationship of the charge-carrier density and Ea/E0, where Ea and E0 are the activation energies at a certain doping fraction and at a low carrier density (pristine material) for a specific disorder parameter, respectively. The activation energies of pristine and doped p(gNDI-gT2) were extracted from variable-tempn>erature electrical conductivity measurements by fitting an Arrhenius tempn>erature dependence (Figure c)where Ea is the activation energy, kb the Boltzmann constant, and σ0 a pre-expn>onential factor that does not influence the activation energy. We obtained activation energies of E0 = 290 meV and Ea = 130 meV for the pristine n>an class="Chemical">polymer and a sample doped with 20 mol % N-DMBI, respectively. We extracted a disorder parameter of 90 meV and hence estimated a charge carrier density of 1.5 × 1019 cm–3, assuming an average hopping distance for conjugated polymers of 1 nm[57,58] and an overall density of states of 1021 cm–3 (see Supporting Information Figure S3 for details). Note that we can produce good fits for nearest-neighbor hopping as well as 1-, 2-, and 3D variable range hopping, which prevents us from determining the transport mode based on our data (see Supporting Information Figure S4).

To corroborate the estimated charge carrier density of N-DMBI dopn>ed p(n>an class="Chemical">gNDI-gT2), we employed electron paramagnetic resonance (EPR) spectroscopy (Figure d). In the case of (negative) polarons as predominant charge carrier species, the electron spin density acquired by measurement against a known reference sample is directly equivalent to the charge carrier concentration. The lack of an EPR signal for the pristine polymer indicates that the number of unpaired electrons is low. In contrast, for a sample doped with 20 mol % N-DMBI, we readily observe an EPR signal, indicating that n-doping of the polymer has indeed taken place. Quantification of the spectra yields a spin density of ∼1.0 × 1019 cm–3 (±0.3 × 1019 cm–3). This value is consistent with our estimate for the charge carrier density from the EGDM model, which indicates that polarons are the predominant type of charge carriers because bipolarons would not give rise to an EPR signal. We explain the absence of an EPR signal for the neat polymer despite considerable background doping (cf. discussion below), with the 50 times lower conductivity and hence polaron concentration, which means that our measurement is not sensitive enough.

Comparison of the number of charge carriers n and the total number of pan class="Chemical">N-DMBI molecules n>an class="Chemical">nN-DMBI allows us to estimate the doping efficiency, i.e., the ratio n/nN-DMBI. A dopant concentration of 20 mol % translates into 1.3 × 1020 cm–3 N-DMBI molecules, assuming a density of 1 g cm–3. Hence, we estimate an approximate doping efficiency of about 13% for p(gNDI-gT2) doped with 20 mol % N-DMBI. In comparison, Schlitz et al. have deduced a more than 10-times lower N-DMBI doping efficiency of only 1% for the nonpolar p(NDI2OD-T2).[14] In analogy to several studies of polythiophenes[13,49] and fullerenes[39,43] decorated with more polar oligoethylene glycol moieties, we attribute the higher doping efficiency of N-DMBI-doped p(gNDI-gT2) to enhanced miscibility of the polymer/dopant pair.

The low doping efficiency of polymers such as p(n>an class="Chemical">NDI2OD-T2) results in the formation of numerous N-DMBI aggregates on the film surface, which become clearly visible for a doping fraction as low as 9 mol %.[14] We therefore anticipate that the superior doping efficiency of p(gNDI-gT2) reduces the tendency for N-DMBI aggregation. We employed atomic force microscopy (AFM) and scanning electron microscopy (SEM) to study the surface topography of p(gNDI-gT2) thin films (Figure a–d; Supporting Information Figures S5–S7). Both AFM and SEM images indicate formation of dopant aggregates on the surface of the blend films that increase in quantity and size with an increasing amount of N-DMBI. The surface roughness (Supporting Information Figure S8) changes only slightly from 2 nm for the pristine film to 6 nm after up to 20 mol % N-DMBI is added but increases sharply for 30 mol % and more. Intriguingly, the surface roughness in the regions between the aggregates is not significantly affected by doping, even at higher doping fractions, which suggests that the nanostructure of the pristine polymer is largely maintained.

Figure 4

Atomic force microscopy (AFM) height images of (a) pristine, and N-DMBI-doped p(gNDI-gT2): (b) 10, (c) 20, and (d) 30 mol % N-DMBI. X-ray diffractograms of pristine and doped p(gNDI-gT2) obtained by integration along the (e) out-of-plane (q) and (f) in-plane (q) direction. Scattering from lamellar and π-stacking is indicated with (h00) and (0k0); scattering marked with an asterisk (*) is associated with the neat dopant. 2D grazing-incidence wide-angle X-ray scattering images of (g) pristine p(gNDI-gT2) and (h) the polymer doped with 20 mol % N-DMBI.

Atomic force microscopy (AFM) height images of (a) pristine, and pan class="Chemical">N-DMBI-doped p(n>an class="Chemical">gNDI-gT2): (b) 10, (c) 20, and (d) 30 mol % N-DMBI. X-ray diffractograms of pristine and doped p(gNDI-gT2) obtained by integration along the (e) out-of-plane (q) and (f) in-plane (q) direction. Scattering from lamellar and π-stacking is indicated with (h00) and (0k0); scattering marked with an asterisk (*) is associated with the neat dopant. 2D grazing-incidence wide-angle X-ray scattering images of (g) pristine p(gNDI-gT2) and (h) the polymer doped with 20 mol % N-DMBI.

To further elucidate the effect of the dopant on the nanostructure of the n>an class="Chemical">polymer, we obtained a series of scattering diffractograms in the out-of-plane and in-plane directions (Figure e,f) through integration of grazing-incidence wide-angle scattering (GIWAXS) images of pristine and heavily doped p(gNDI-gT2) (Figure g,h; Supporting Information Figure S10). The pristine polymer features distinct scattering peaks from lamellar stacking at q100 ≈ 0.27 Å–1 and q200 ≈ 0.54 Å–1 and from π-stacking at q010 ≈ 1.6 Å–1. Further, in the in-plane scan, two additional peaks are present at q ≈ 0.45 Å–1 and q ≈ 0.9 Å–1. We assign these peaks to the repeat distance along the backbone and argue that, similar to p(NDI2OD-T2),[59−62] two polymorphs are present. The diffraction peaks that we observe for pristine p(gNDI-gT2) are not altered upon doping with 20 mol % N-DMBI. Addition of 50 mol % dopant results in the appearance of a new out-of-plane scattering peak at q ≈ 1 Å–1 and in-plane at q ≈ 1.3 Å–1 as well as q ≈ 1.75 Å–1, which we explain with the presence of unreacted excess dopant. Further, annealing of the films does not alter the diffraction from the polymer but results in a slight shift of the peaks associated with excess N-DMBI, as well as a decrease in scattering intensity (Supporting Information Figure S11). We conclude that significant segregation only takes place for a dopant concentration above 20 mol %. Note that a few isolated aggregates are already visible in the AFM images of p(gNDI-gT2) doped with 20 mol % N-DMBI, which are weakly visible in the GIWAXS measurements. Comparison with the nonpolar p(NDI2OD-T2) (cf. study by Schlitz et al.[14]) indicates that the polar oligoethylene glycol side chains largely suppress N-DMBI aggregation up to a concentration of about 20 mol %, which is consistent with our picture of enhanced polymer/dopant miscibility.

In a further set of experiments, we characterized the electrical properties of p(gNDI-gT2) ≈ 60 nm thin films dopn>ed with various amounts of n>an class="Chemical">N-DMBI (Figure ). The pristine polymer features an electrical conductivity of 6 × 10–3 S cm–1, which arises due to background doping. In a first regime up to 20 mol %, the addition of N-DMBI is concomitant with an increase in electrical conductivity. We reach a value above 10–1 S cm–1, which is more than 2 orders of magnitude higher than p(NDI2OD-T2) doped with N-DMBI (Supporting Information Figure S12a) due to the here-reported higher doping efficiency in the case of p(gNDI-gT2). At the same time, for a dopant concentration up to 20 mol %, the Seebeck coefficient decreases from 359 to 93 μV K–1. Upon further doping, we observe a substantial drop of the electrical conductivity by nearly 2 orders of magnitude. In contrast, in this second regime, the Seebeck coefficient only slightly decreases to, e.g., 70 μV K–1 for 30 mol % N-DMBI, indicating that the number of mobile charge carriers is not strongly enhanced upon further addition of N-DMBI. We rationalize this behavior with gradual disruption of the polymer nanostructure by excess unreacted dopant, which coincides with the appearance of N-DMBI aggregates (cf. Figure ).

Figure 5

(a) Electrical conductivity (σ) and Seebeck coefficient (α); dashed lines are a guide to the eye. (b) Thermoelectric power factor (α2σ) as a function of the electrical conductivity at various dopant fractions; the dashed line represents the empirical relation α2σ ∝ σ1/2.[10] (c) Air stability of pristine and N-DMBI-doped p(gNDI-gT2): the current at 0.5 V was extracted from I–V curves recorded in nitrogen, in air, and finally again in nitrogen; note that the nonohmic behavior of several samples prevented us from extracting the electrical conductivity. A contact geometry with a channel length of 1000 μm and a channel width of 30 μm was used for air stability measurements of doped samples, which resulted in similar currents for the pristine and doped sample.

(a) Electrical conductivity (σ) and Seebeck coefficient (α); dashed lines are a guide to the eye. (b) Thermoelectric power factor (α2σ) as a function of the electrical conductivity at various dopant fractions; the dashed line represents the empn>irical relation α2σ ∝ σ1/2.[10] (c) Air stability of pristine and n>an class="Chemical">N-DMBI-doped p(gNDI-gT2): the current at 0.5 V was extracted from I–V curves recorded in nitrogen, in air, and finally again in nitrogen; note that the nonohmic behavior of several samples prevented us from extracting the electrical conductivity. A contact geometry with a channel length of 1000 μm and a channel width of 30 μm was used for air stability measurements of doped samples, which resulted in similar currents for the pristine and doped sample.

We chose to compare the thermoelectric performance of pan class="Chemical">N-DMBI-dopn>ed p(n>an class="Chemical">gNDI-gT2) with the empirical correlation that Glaudell et al. have proposed for the thermoelectric power factor of not mobility-limited p-doped semiconductors: α2σ ∝ σ1/2.[10] We observe a good correlation for a doping concentration of up to 20 mol % but a considerable deviation for higher amounts of N-DMBI. This behavior corroborates our picture that excess dopant interrupts the nanostructure of the polymer, causing a considerable reduction in mobility and hence electrical conductivity at high dopant fractions. Overall, we obtain a maximum thermoelectric power factor of 0.4 μW K–2 m–1 in the case of doping with only 10 mol % N-DMBI, which is much higher than the highest value of 0.02 μW K–2 m–1 measured for p(NDI2OD-T2) (Supporting Information Figure S12c).

For p(pan class="Chemical">gNDI-gT2) dopn>ed with upn> to 20 mol % n>an class="Chemical">N-DMBI, we anticipate that the electrical conductivity is not limited by the bulk electron mobility. To gain a more complete picture of charge transport in the here-studied system, we estimate the electron mobility μ according to σ = nqμ, where q is the elementary charge, i.e., 1.6 × 10–19 C. For a dopant concentration of 20 mol %, for which we have deduced the charge carrier density from EGDM as well as EPR, we obtain a value of μ ≈ 0.2 cm2 V–1 s–1. This value is considerably higher than the electron field-effect mobility μFET ≈ 10–5 cm2 V–1 s–1 reported for the pristine polymer, which may arise due to the low degree of polymerization of not more than seven repeat units[50] or due to the presence of polar side chains attached to the backbone of the copolymer. In contrast, the here-studied case of highly doped p(gNDI-gT2) does not appear to suffer from a low electron mobility. This observation is consistent with our recent study on p-doping of P3HT, where we likewise concluded that the molecular weight does not influence the conductivity at high dopant levels.[63]

Finally, we investigated the air stability of the electrical conductivity of a doped and a pristine thin film of p(gNDI-gT2) by expn>osing freshly prepared sampn>les to air while measuring the current–voltage (I–V) behavior at various times. The nonlinear behavior of dopn>ed p(n>an class="Chemical">gNDI-gT2) samples after 30 min in air prevented us from extracting the electrical conductivity. Instead, we chose to plot the electrical current at 0.5 V (Figure c; cf. Supporting Information for I–V curves, Figure S13). The doped and pristine samples show a markedly different response to air exposure. For the pristine polymer, we observe an immediate drop of the current. In contrast, N-DMBI-doped p(gNDI-gT2) is able to maintain a similar current (and hence electrical conductivity) for the first 20 min of air exposure, which suggests that the doped polymer is more air-stable and hence can be handled outside of a protective atmosphere for at least a short period of time. However, after 30 min of air exposure, the current likewise drops by several orders of magnitude. After returning the samples to the glovebox, the current measured for the doped and pristine polymer quickly recovers. Subsequent annealing at 80 °C for 10 min almost restores the current (and hence the electrical conductivity) to the initial value. We tentatively explain this behavior with adsorption of, e.g., oxygen and water from the ambient atmosphere introducing charge traps, which are subsequently desorbed from the film upon re-exposure to a protective atmosphere and annealing.[21] To demonstrate the negative influence of water, we compared the conductance of the doped polymer at ambient conditions before and after placing a water droplet onto the film, which caused a 5-fold decrease in conductance (Supporting Information, Figure S14).

We have studied n-doping of the polymer p(n>an class="Chemical">gNDI-gT2), which bears oligoethylene glycol-based chains, with the hydride dopant N-DMBI. The polar side chains facilitate more effective doping of the semiconducting polymer by increasing the miscibility with the dopant, resulting in a doping efficiency of ∼13% for a sample doped with 20 mol % N-DMBI. We were able to prepare films with a conductivity above 10–1 S cm–1 and obtained a thermoelectric power factor of up to 0.4 μW K–2 m–1. Additional doping leads to segregation of the dopant, which ultimately results in a drastic reduction in the thermoelectric performance caused by a less optimal nanostructure due to excess unreacted dopant. Moreover, we found that N-DMBI-doped p(gNDI-gT2) displays improved air stability as compared to the pristine polymer. We conclude that polar side chains are a powerful tool for the design of more conductive and stable n-type materials.