ZT > 0.1 Electron‐Carrying Polymer Thermoelectric Composites with In Situ SnCl2 Microstructure Growth

An n‐type pyromellitic diimide polymer composite with in situ microstructure growth of the common element compound SnCl2 reaches power factor of 50–100 μW m−1 K−2, the highest purely n‐type polymer composite power factor yet reported. The composite has a gigantic Seebeck coefficient between −4000 and −5000 μV K−1, many times higher than other polymer composites.

Solution-deposited polymer-based composites are interesting for ambient temperature cooling and subwatt power generation where suffi cient power is more critical than high effi ciency. Such composites combine solution processing, mechanical fl exibility, and low thermal conductivity with suffi cient power factor (PF), contributing to high fi gures of merit, ZT = S 2 σT / κ , where S is Seebeck coeffi cient, σ is electronic conductivity, S 2 σ is PF, T is absolute temperature, and κ is thermal conductivity. [ 1,2 ] Thermoelectric performance of hole-carrying (p-type) polymers has been enhanced so ZT exceeds 0.1. [ 3 -9 ] ZT of 0.1 would be an important milestone for a new class of materials, or for medical or mobile applications.
Krebs and co-workers demonstrated that printing is the most rational choice for scalable manufacturing of large-area energy saving devices. [ 10 ] Yazawa and Shakouri and a few others compared polymer materials with conventional semiconductor materials in terms of power per unit mass and unit cost. [11][12][13] Polymers were found valuable for lightweight remote power generation for sensors in conjunction with passive heat sinks (utilizing low heat fl ux).
We added SnCl 2 from 0 to 90 wt% to P1 and P2 , separately. Incorporation of SnCl 2 and subsequent drop-casting or spincasting on glass resulted in spontaneous formation of tin-based microcrystals within the P1 polymer matrix during solvent evaporation. Figure 2 shows laser optical images of SnCl 2 in PyDI-5FPE for concentrations 20-90 wt%, and for pure SnCl 2 fi lms. The pure SnCl 2 fi lms were processed exactly the same way as composites, meaning that the powder was dispersed in organic solvent but not mixed with polymer, just drop-cast or spin-coated onto the substrate utilizing the same conditions. P1 is particularly effective at promoting, or allowing, SnCl 2 micro/nanostructure growth. Elongated microcrystals can be observed for starting SnCl 2 concentrations above 30 wt%, having aspect ratios >10. By contrast, only well-dispersed globular clusters are obtained in P2 . The particles obtained from initial concentrations of 60-80 wt% SnCl 2 blended with P1 resulted in the most prominent shapes (sharper interfaces) and greater monodispersity, reaching widths of 10-40 µm and lengths of 100-400 µm. Spin-casting resulted in very welldispersed particulates, having signifi cantly reduced dimensions, but increased homogeneity and still showing rod-and barlike structures. Optical images for 40, 60, and 80 wt% SnCl 2 samples obtained by spin-casting are shown in Figure S1 (Supporting Information).
Clusters and elongated crystals grown in the polymer matrix both appear to be mostly SnCl 2 , possibly with some metallic tin and tin oxide, according to energy dispersive spectroscopy (EDS) utilizing a scanning electron microscope (SEM), X-ray photoemission spectroscopy (XPS), and X-ray diffraction (XRD). EDS shows that crystals have high SnCl 2 content but also have some oxide presence, and that there is essentially no difference in the composition of different structures/shapes (clusters vs crystals vs pure fi lm). Elemental maps of particles in polymers obtained by EDS can be found in Figures S2 and S3 (Supporting Information). We observe that the SnCl 2 additive is dispersed homogeneously within the polymer at a molecular level in addition to the secondary phase particulates at micrometer scale that are more easily observed. XPS suggests SnCl 2 -inpolymer has decreased work function, consistent with preliminary scanning Kelvin probe microscopy (KPM) and ultraviolet photoelectron spectroscopy (UPS) results shown in Figures S4 and S5 (Supporting Information). KPM shows that the poly mer chemical potential is pulled toward vacuum and SnCl 2 is pushed away from vacuum at the interface, which suggests that electrons accumulate in the polymer at their interface, with negative end of the interfacial dipole within the polymer. UPS shows that charge transfer occurs by evidence of an extra broad peak at lowest binding energy ( Figure S5, Supporting Information). This is a direct observation of the interfacial charge transfer and new states that lie within the polymer bandgap, where electrons generated at the polymer-inorganic interface can reach the surface and contribute to the UPS signal.
Small XRD peaks can be indexed to tin metal and tin oxide, demonstrating that the obtained product is composed of multiple hybrid structures and interfaces ( Figure S6, Supporting Information). The polymer-grown tin structures formed in P1 exhibit very similar phases ( Figure S7, Supporting Information).
Thermal conductivity κ of polymers and composites was measured using the femtosecond laser-based frequencydependent transient-thermorefl ectance method in conjunction with a multiple location and multiple spot-size sampling technique (see the Supporting Information). [ 40 ] The κ we measured is 0.14 W m −1 K −1 for both P1 and P2 with an uncertainty of 0.02 W m −1 K −1 while the heat capacity is 1.40 J cm −3 K −1 with an uncertainty of 0.2 J cm −3 K −1 . The measurements on dropcast SnCl 2 composites indicate that values of κ are only slightly increased relative to pristine polymers for both P1 and P2 , converging on 0.2 W m −1 K −1 for all compositions as the area sampled is increased to cover representative contributions from both polymer and SnCl 2 components (Figure 1 b and Figure S10, Supporting Information). The thermal conductivity of composites depends on the volume fraction, morphology, the thermal conductivity of the polymer and the inorganic crystals, and the interfacial thermal conductance. [41][42][43] Although the thermal conductivity of SnCl 2 is not available in the literature, we expect the value to be on the order of 0.3-0.6 W m −1 K −1 , which is consistent with many similar crystalline dichlorides. [ 44,45 ] In addition, we observed maxima of 0.5 W m −1 K −1 in Figure S10 (Supporting Information), which would be close to the thermal conductivity of SnCl 2 itself. Considering that the microstructure is relatively large, on the order of a few micrometers, based on the models cited, [41][42][43] the contribution of the intrinsic interface resistance might be negligible in these composites. The comparison of the measured data with the Nielson model is presented at the conclusion of the Supporting Information. While this model appears to fi t reasonably for the lower SnCl 2 volume fractions, it is likely not suitable for the highly anisotropic crystallites formed at higher SnCl 2 concentrations.  The Seebeck coeffi cient S was evaluated using standard thinfi lm thermoelectric measurement techniques, taken on a homemade setup, utilizing preferred electrode geometry that reduces measurement error. [ 46 ] The sample is mounted between a pair of Peltier heater-cooler tiles, with one electrode of the sample over each tile. Thermal EMF (Δ V ) and the temperature difference (Δ T ) were measured simultaneously by probing the pair of electrodes with a source meter and thermocouples. We used an optimal thin-fi lm electrode geometry (being a set of narrow, parallel strips, and no semiconductor material outside of the interelectrode region or space between electrodes), which minimizes external conduction pathways. In order to minimize signal-tonoise and overcome the increased device impedance associated with this otherwise preferred electrode arrangement, gallium indium eutectic was used on the voltage probe contacts to minimize contact resistance. Thermal paste is used to minimize thermal contact resistance between the bottom of our substrates and Peltier devices ( Figure S8, Supporting Information). Total electrode length is >10 mm and distance between electrodes is ≈2 mm, where the effective width of the electrodes is the area covered by semiconductor, which is also ≈2 mm. The resulting ratio of aspect ratio for the channel to the aspect ratio of the electrodes is close to or less than unity, ensuring small relative errors and minimal overestimation of S . [ 46 ] More precise measurements become diffi cult because the aspect ratio of the electrodes should be greater than the channel aspect ratio for accurate S measurement, although thin channel widths are required for low signal-to-noise and for accurately measuring resistance of low-conductivity materials due to device impedance. Sheet resistance was measured by four-probe technique with gallium indium eutectic applied to the probe tips to further minimize contact resistance and to mimic S measurements, and the electrical conductivity was calculated using poly mer fi lm thickness of each device. All devices are tested under ambient conditions in air, but protected from light and thermal convection. Figure 1 c shows S , σ , and PF as function of SnCl 2 concentration in polymer matrices P1 and P2 prepared by drop-casting. The values are averages over at least ten samples, taken from 3 to 5 repeated experiments, balancing statistical signifi cance with the consideration that it takes about 1 h for each run. P1 shows modest σ in its undoped form (0.00057 S cm −1 ), and a significant S (−380 µV K −1 ). P2 shows greater σ (0.00124 S cm −1 ), as expected due to higher intermolecular overlaps of diimide cores, resulting in much greater PF despite its lower intrinsic S (−280 µV K −1 ). PFs are 0.00014 and 0.0095 µW m −1 K −2 for P1 and P2, respectively. Our PF value for the pristine P2 is almost one-tenth the value recently reported in a doped P2 system. [ 21 ] SnCl 2 evaporated onto glass, nominally 50 nm thick, shows PF an order of magnitude greater than does drop-cast SnCl 2 , about 0.16 and 0.038 µW m −1 K −2 , respectively. Evaporated and drop-cast pure SnCl 2 both show σ around 4-5 × 10 −3 S cm −1 , but the S is apparently increased by a factor of fi ve when sublimed (which presumably forms more pure and densely connected crystalline solids), reaching about −530 µV K −1 . The sublimed and drop-casted materials both form fi lms comprising multiple structures ( Figure S9, Supporting Information).
PFs of our hybrid composites are greater than those of any of the individual components, reaching 50-100 µW m −1 K −2 for SnCl 2 precursor concentrations above 50 wt% blended with P1 .
Thus, estimating PF of 80 µW m −1 K −2 for 50-80 wt% SnCl 2 , T of 300 K, and κ of 0.2 W m −1 K −1 , we project ZT of 0.12. PF and κ are constant within experimental error over this range, and thus ZT is as well. PF, and thus ZT, is about one-tenth as large for P2 composites. It appears that in situ microstructure growth of SnCl 2 enormously increases S , and possibly σ of the polymer. S increases signifi cantly for P1 with increasing concentration of SnCl 2 , saturating around −4500 µV K −1 above 50 wt% SnCl 2 . σ increases only slightly for P1 hybrids with low concentrations of SnCl 2 . σ increases sharply between 30 and 50 wt% and remains constant up to 90 wt% SnCl 2 , leveling at about 0.05 S cm −1 . A strong increase in electrical conductivity appears to be correlated with overlapping of the elongated Sn-containing crystals, as they reach suffi cient size to create a percolated network above 40 wt% SnCl 2 . Utilizing P2 as the polymer host matrix showed a similar trend as P1 blended with SnCl 2 , but lower PFs closer to 10 µW m −1 K −2 were obtained. Though with a much lower σ , the PFs are on par with highperformance p-type hybrid composites.
Contributions of static electricity on an electrically isolated probe or highly resistive contacts to the very large Seebeck coeffi cients we measured are ruled out by the electrode and probing arrangements. As long as probes are small enough to be at constant temperature over their whole area, then the largest temperature difference on the sample must be across or within the interelectrode region, and we meet these criteria. The fact that we obtain obvious trends of S as a function of material composition is further evidence that the data refl ect material rather than contact properties.
Our results on n-polymers with inorganic additives prompt similar experiments using hole-carrying polymers PQT12 ( P3 ) and PBTTT-C14 ( P4 ). The weak dopant Co(acac) 3 will form large crystals readily in p-type polymers such as P3 and P4 . S > 2000 µV K −1 and power factors of 10 µW m −1 K −2 were obtained, with slight increase in hole conductivity (from 0.0013 to 0.021 S cm −1 ). Different morphologies, including elongated uniaxial microcrystals, were demonstrated using 20, 40, and 60 wt% additives within P3 polymer matrix, as seen in Figure 3 . Similar features were obtained for Co((acac) 3 ) blended with P4 , which are included in Figure S11 (Supporting Information). The results of S , σ , and PF plotted for P3 blended with Co((acac) 3 ) as a function of dopant concentration are also shown in Figure S12 (Supporting Information). We anticipate considerably higher power factors as doping capacity of the inorganic phase is increased, by either using stronger dopants or obtaining more homogeneous dispersion by arresting particle growth at nanoscale dimensions. Further detailed exploration of metal acacp -polymer systems will be published separately.
In situ fabrication of inorganic particles within the polymer matrix is advantageous because of simpler preparation and improved interfacial interactions, [ 25 ] and better morphology. [ 4 ] There are several competing arguments for either macrocrystals or nanosized inclusions being more desirable. First, nanostructuring increases surface area to volume ratio greatly at a cost of increasing contact resistances. [ 9 ] Although microcomposites are generally less homogeneous compared to nanocomposites, they may have greater conductivity because larger inclusions may have greater carrier mobilities due to longer grain size and thus potentially longer mean-free paths. Smaller nanoscale dopants can also have high tendency to diffuse in organic systems, leading to detrimental instability. [ 47 ] However, phononic thermal conductivity is greatly reduced due to the abundance of interfaces achieved by nanostructuring. The control of particle growth (i.e., growth rates, crystal structure and orientation, and particle alignment) leading to optimized power factors and reduced phononic thermal conductivity would be essential for further increases in composite ZT.
The platform described above offers numerous opportunities to increase ZT further, especially for n-type polymer composites. S and σ can increase simultaneously in our composites because of the inorganic compound introducing some new fi lled states to the polymer besides having some conductivity itself. Simple additions of n-doping elements such as sulfur and aluminum to SnCl 2 should increase σ . This should not result in accompanying increase in κ ( Figure S18, Supporting Information) because we would remain in the phonon thermal conductivity-dominated regime. Growth of precise nanostructures will increase S and σ while decreasing κ further if it allows maximum interface phonon scattering while maximizing doping and transport pathways.
Our use of a novel unipolar n-type polymer and inorganic additives has led to the fi rst demonstration of ZT > 0.1 in electron-carrying polymer composites that contain neither highly toxic nor rare elements. Leveraging recent advances in n-poly mer doping, we anticipate equivalent power factors from both composite polarities to be achievable in the near future.

Experimental Section
Microscope slides (Corning Inc.) were cleaned by sonication in deionized water, acetone, and 2-propanol for 10 min each, and blowing dry with N 2 . Electrode contacts were predeposited by thermally evaporating Au or Al from alumina crucibles using an Edwards thermal evaporation system (50 nm thick, pressure below 5 × 10 −6 Torr, deposition rate 0.5 Å s −1 ). Deposition rates and thicknesses were monitored by quartz crystal microbalance.
Polymer and inorganics were weighed in separate glass vials and made up to 10 mg mL −1 . The polymer was placed on a hot plate (80 °C) while solutions of inorganic were sonicated (30 min). The polymer was blended in the desired concentration with inorganic solutions using a pipette, and vigorously stirred (10 min) before drop-casting. The fi nal blend was dropped by pipette into 2D wells, which were fabricated by laying a pattern of Novec fl uorinated polymer, on the glass substrates having predeposited metal electrodes. The result was a square cm fi lm with 1-2 µm thickness, lying over two metal electrodes with a channel gap (each about 0.2 cm across and 1 cm long). The solvent was allowed to evaporate overnight while samples were kept in a petri dish in the fume hood. Next, samples were placed in vacuum oven overnight (100 °C under low vacuum) to remove residual solvent. Just before taking electrical measurements, the fi lms were heated under nitrogen atmosphere in the oven (150 °C for 10-30 min).
Compositional and structural analysis of all P1 -based composites was carried out by X-ray photoemission spectroscopy, X-ray diffraction, and by energy dispersive spectroscopy utilizing the backscatter detector of a JEOL scanning electron microscope. Scanning Kelvin probe microscopy, a form of atomic force microscopy, was utilized to observe energy-band alignment and vacuum level offset between material domains.
Regarding Seebeck coeffi cient measurements, numerous measurements of Δ V , typically 500, were made for each value of Δ T , with a standard deviation of 1-5% (0.01-0.05 mV) per data set. These Δ V were averaged to eliminate the noise signal induced by the environment and obtain just the consistent steady-state value. Six incremental Δ T were imposed on the sample, so the slopes of Δ V versus Δ T give values of the Seebeck coeffi cient. The linearity of the data of Δ V and Δ T was used as a key criterion to ensure valid measurements. The setup was checked using Ni metal ingot or Te thin fi lms deposited from vapor, for which the values we obtained (−19.8 ± 0.8 µV K −1 for Ni, and ≈400 µV K −1 for Te) agreed well with those reported. Resistivity measurements were employed using four-probe measurement method with an Agilent 4155C Semiconductor Parameter Analyzer using low-resistance probes (Micromanipulators). Thermal conductivity was measured with a timedomain thermorefl ectance (TDTR) system using femtosecond lasers (details are described in the Supporting Information).

Supporting Information
Supporting Information is available from the Wiley Online Library or from the author.