Thermoelectric transport in surface-and antimony-doped bismuth telluride nanoplates

Thermoelectric transport in surfaceand antimony-doped bismuth telluride nanoplates Michael Thompson Pettes,1,2 Jaehyun Kim,1 Wei Wu,2 Karen C. Bustillo,3 and Li Shi1,a 1Department of Mechanical Engineering, The University of Texas at Austin, Austin, Texas 78712, USA 2Department of Mechanical Engineering and Institute of Materials Science, University of Connecticut, Storrs, Connecticut 06269, USA 3National Center for Electron Microscopy, Molecular Foundry, Lawrence Berkeley National Laboratory, Berkeley, California 94720, USA

Bismuth telluride (Bi 2 Te 3 ) is one of the most commonly used thermoelectric materials. 1,2It was predicted over two decades ago that quantum confinement of electrons in two-dimensional Bi 2 Te 3 layers can result in enhanced thermoelectric power factor S 2 σ, where S is the Seebeck coefficient and σ is the electrical conductivity. 3Recent experimental observations of surface electronic states protected by time-reversal symmetry in bismuth chalcogenides, including Bi 2 Te 3 , 4 Bi 2 Se 3 , 5 and BiSbTeSe 2 , 6 have also led to theoretical predictions of enhancement of the thermoelectric performance in ultrathin layers of these so-called topological insulators. 7,8The enhancement is predicted in ultrathin bismuth chalcogenides with a precise control of the Fermi level and carrier concentration.However, bismuth chalcogenides synthesized by different methods often suffer from chalcogen vacancies in the bulk crystal 9,10 and surface doping from exposure to the air environment, 11,12 both of which make these materials degenerate n-type semiconductors with reduced thermoelectric power factors.In ultrathin materials, the extrinsic n-type environmental surface doping can dominate the intrinsic p-type surface dipole arising from topological surface state occupation. 13everal methods to modulate the carrier concentration and optimize the Fermi energy in these ultrathin samples have been reported.These methods involve substitutional doping of Ca 14 and Sb 15,16 on the Bi atomic sites in order to optimize the Fermi energy and maximize the contribution from the protected surface states.Two especially relevant advances to tune the Fermi energy in Bi 2 Te 3 involve doping with antimony 15,16 to create (Bi 1−x Sb x ) 2 Te 3 in the range between x = 0 (Bi 2 Te 3 ) and x = 1 (Sb 2 Te 3 ).Trends in both studies are similar: antimony incorporation lowers the Fermi energy, although the studies differ on the exact composition that leads to the so-called bulk-insulating electron transport, x = 0.50, in one work 15 and x = 0.96 in the other. 16t is noted that these x values were defined by the source material composition.This trend has the effect of changing the majority carrier type from electrons in Bi-rich samples to holes in Sb-rich samples, which has been confirmed by Hall measurements. 15Additionally, to offset the effect of extrinsic environmental doping, highly electronegative hole-injecting molecules, such as NO 2 14 and tetrafluoro-tetracyanoquinodimethane (F 4 -TCNQ), 17 have been employed to control the surface potential of two-dimensional (2D) graphene, 18,19 ultrathin Bi 2 Se 3 , 17 and WSe 2 20 but not yet for Bi 2 Te 3 .Although there are several compatible molecules for use as surface dopants, 21 F 4 -TCNQ has often been used for proof-of-concept investigations as it provides a very strong surface p-type doping effect compared to other hole injecting molecules due to its high electron affinity of ∼5.2 eV.This surface-effect has been experimentally observed in F 4 -TCNQ-coated graphene by angle-resolved photoemission spectroscopy (ARPES). 19Although these compositional and surface doping methods appear to be promising, their effects on thermoelectric transport in ultrathin topological insulator materials such as Bi 2 Te 3 remain elusive.
Here we report measurements of the three in-plane transport properties-thermal conductivity (κ), electrical conductivity (σ ≡ ρ −1 , where ρ is the electrical resistivity), and Seebeck coefficient (S)-that enter the expression for thermoelectric figure of merit, zT = (S 2 σ/κ)T, on thin (Bi 1−x Sb x ) 2 Te 3 nanoplates with x ranging from 0.07 to 0.95.The carrier concentration in the nanoplates is varied by antimony substitution, while the surface potential is tuned by direct deposition of a hole-injecting molecule, F 4 -TCNQ, onto all surfaces of individually measured suspended nanoplates.We observe a maximum p-type Seebeck coefficient and corresponding maximum zT at x = 0.5 and a general p-type doping effect upon exposure to F 4 -TCNQ.This surface effect is efficient enough to allow observation of a remarkable change of the majority carrier type from electrons to holes in a lightly doped (x = 0.07), 9 nm-thick sample.
We have used a chemical vapor deposition (CVD) method to synthesize the (Bi 1−x Sb x ) 2 Te 3 nanoplates (NPs) on SiO 2 -coated Si substrates similar to that demonstrated in previous reports. 12,15,22ncorporation of Sb is achieved by varying the molar ratio of Bi 2 Te 3 and Sb 2 Te 3 source materials. 15The mole fractions of Sb dopants reported here as x in (Bi 1−x Sb x ) 2 Te 3 correspond to that in the precursors.Random incorporation of Sb on Bi atomic sites is facilitated by the comparable lattice parameters and crystal symmetry of these materials (Figure 1(a)) as has been verified in detail by a previous work. 15e have measured the composition of three samples (x = 0.07, 0.25, and 0.50) using two different electron microscopes with two different high sensitivity energy dispersive x-ray spectroscopy (EDS) detectors (FEI Nova NanoSEM 450 with an Oxford X-Max 80 XMX1105 silicon drift detector; FEI 80-300 TitanX with a FEI Super-X Quad windowless silicon drift detector system).Variation between instruments is large, with up to 2933% variation in the calculated Sb content and a 595% variation in the value of Te deficiency.A large variation is observed even for the same instrument, which can be attributed to two mechanisms: (1) the irradiated sample volume is very small due to the very thin samples, and (2) even for large sample volumes, EDS is seen as a qualitative tool where uncertainty can be on the order of several percent. 23For this reason we have chosen to list Sb content x in our samples as the nominal molar ratio of Bi:Sb in the source materials as well as to allow fair comparison with previous reports, which have also reported x using the molar ratio of source precursors. 15,16We note that accurate determination of molecular composition is a critical challenge and while techniques such as inductively coupled plasma mass spectrometry (ICP-MS) offer better than part per billion accuracy, the technique is designed for large volume samples with more than several mg in mass. 24fter growth, the samples were transferred to a suspended micro-thermometry test platform (Figure 1(b)) using an electrochemically sharpened probe.Transfer was performed in ambient conditions under an optical microscope.Samples were then placed into a variable-temperature cryostat, within roughly 2-10 h after growth.The electrical, thermal, and thermoelectric properties of the nanoplates have been obtained with the same four-probe thermoelectric measurement method as reported previously. 12,25The sample was first held at the maximum temperature (∼450 K) for several hours, after which the sample stage temperature was gradually lowered to each measurement temperature.Electrical contact to the nanoplate sample was obtained by electrical current annealing at ∼450 K. Transmission electron microscopy (TEM) analysis was performed after transport measurements were conducted (Figure 1(c)).
Additionally, we have characterized the compositional stability of the x = 0.5 and x = 0.07 samples after storage in an inert atmosphere for two years (Figure 2).EDS was used to qualitatively map the compositional homogeneity of the nanoplates in an uncorrected FEI Titan 80-300 TEM operating at 120 kV using a FEI Super-X Quad windowless detector based on silicon drift technology with a solid angle of 0.7 sr.We found a uniform distribution of Sb atoms within the structures in the x = 0.5 sample, as opposed to segregated Bi 2 Te 3 and Sb 2 Te 3 phases.
The measured thermoelectric properties are shown in Figures 3-5.The thicknesses, geometry, and room-temperature properties for the five (Bi 1−x Sb x ) 2 Te 3 samples in this work are listed in Table I, where the Sb content, x = 0.07, 0.25, 0.5, 0.7, and 0.95, are nominally based on the precursor composition.
It has been suggested by a recent study that the measured Bi 2 Te 3 nanoplates synthesized by a similar method were unintentionally doped. 12Additionally, it was suggested that exposure to air resulted in n-type surface band-bending on the order of 230 meV, 12 which is in qualitative agreement with that observed byARPES. 11In contrast to the degenerate behavior observed in the Bi 2 Te 3 samples measured in the previous work 12 and the current (Bi 1−x Sb x ) 2 Te 3 samples with x = 0.5 and higher, where the measured electrical conductivity decreases with increasing temperature, opposite temperature dependence expected for non-degenerate semiconductors was observed in the two samples with x = 0.07 and 0.25 (Figure 3(a)).In comparison, a similar non-degenerate or so-called bulk insulating behavior was observed at high Sb-doping levels in the x range of 0.75-0.98(determined by MBE precursor flux composition) in another experiment with (Bi 1−x Sb x ) 2 Te 3 thin layers, 16 likely due to differences in sample thickness, synthesis conditions, and chemical compositions in the samples synthesized from source materials with similar x.The Seebeck coefficient of the (Bi 1−x Sb x ) 2 Te 3 NPs exhibits a clear dependence on the amount of Sb substitution (Figures 3(b) and 4(b)).At a low Sb substitution level of x = 0.07, the Seebeck coefficient is negative and the lowest in magnitude of all samples.The low Seebeck coefficient and non-degenerate electrical conductivity of this sample suggests that it is in the intrinsic regime.Sb-doping at x near 0.25 and higher resulted in positive Seebeck coefficients at near room temperature, with a maximum of 115 µV K −1 at x = 0.5.As with Bi 2 Te 3 nanoplates, the magnitude of S for the Sb-substituted nanoplates was reduced considerably compared to the maximum reported bulk Bi 2 Te 3 values, 225 and −250 µV K −1 for p-and n-type, respectively, at room temperature. 26,27he positive and negative Seebeck coefficient values suggest that the majority carriers are holes and electrons, respectively.It is noted that the measurement device used here is not able to obtain Hall coefficient data on the same suspended sample for which the thermoelectric properties are measured.In principle, the sign of the Hall coefficient and the Seebeck coefficient can disagree for a mixed carrier sample with high minority-to-majority carrier mobility ratio, for example in multiband conductors such as single crystalline LaAgSb 2 . 28For Bi 2 Te 3 the electron and hole mobilities are similar. 26,27Therefore, we do not expect the sign of the Seebeck coefficient in this system to differ from that expected for its Hall coefficient.0][31] While the measured Seebeck coefficient suggests p-type transport for the samples with x = 0.25 or higher, p-type transport was observed only for x ≥ 0.94 in an earlier work, 16 where x is defined by the precursor flux composition in molecular beam epitaxy (MBE).In contrast, p-type behavior was only observed for samples with x < 0.5 in another work, 15 where x is defined by the source composition.These different results in the majority carrier type trend reflect large variations in the actual composition or extrinsic surface doping in these thin nanoplates synthesized in different setups.In samples with thicknesses below 18 nm (x = 0.07, 0.25, 0.5, 0.95), the total thermal conductivity of the ultrathin (Bi 1−x Sb x ) 2 Te 3 NPs is lower than that reported previously for ultrathin undoped Bi 2 Te 3 with comparable thicknesses. 12The room-temperature total κ for these samples is on the order of 0.4-0.8W m −1 K −1 (Figures 3(c) and 4(c)) and is well below the basal-plane values reported for bulk n-and p-type Bi 2 Te 3 , 27 although differences in sample thickness and electronic contribution to κ make extraction of the alloy scattering effects on the lattice thermal conductivity non-trivial.For the sample with x = 0.7, which is ∼42 nm thick, κ is roughly double that of the other samples and decreases with increasing temperature.We also note that these values for total κ are also below the lattice thermal conductivity of similarly doped bulk (Bi 1−x Sb x ) 2 Te 3 . 2espite the reduced lattice thermal conductivity, the reduced Seebeck coefficient has limited zT in the NPs (Figures 3(d) and 4(d)).The zT value reached a maximum in the as-grown samples of 0.30 for x = 0.5 at ∼350 K.This value is an improvement over the maximum zT of 0.24 measured for undoped, ∼9-25 nm-thick Bi 2 Te 3 NPs in a previous work. 12ince the charge carrier concentration affects both σ and κ, we compare the properties of the (Bi 1−x Sb x ) 2 Te 3 NPs with those of un-intentionally doped Bi 2 Te 3 NPs 12 and bulk Bi 2 Te 3 27 by plotting κ versus σ (Figure 5(a)) and S versus σ (Figure 5(b)).We observe a general trend of decreasing κ with decreasing σ for the measured samples (Figure 5(a)), with both lower values of σ and κ achieved than for undoped Bi 2 Te 3 for all samples except x = 0.7, which exhibits high electrical conductivity but still exhibits lower κ than both Bi 2 Te 3 NPs and bulk Bi 2 Te 3 at similar σ.We also observe an unusual trend of increasing S with σ (Figure 5(b)) for all samples except for x = 0.7, which exhibits a relatively low S and high σ.The transport properties in these nanoplates are a function of thickness, therefore the observation that the x = 0.7 sample does not follow compositional trends shown in Figure 5(b) likely arises from its thickness of ∼42 nm, more than twice that of the other samples, which is expected to result in reduced diffuse surface-electron and surface-phonon scattering rates.To gain further insight into thermal transport mechanisms in the ultrathin (Bi 1−x Sb x ) 2 Te 3 NPs, we determine the lattice contribution to the thermal conductivity using the Wiedemann-Franz law as κ − σLT where L is the Lorenz number and T is the absolute temperature.To obtain the Lorenz number, we use our measured thermoelectric properties together with an ab initio model reported previously. 12Although this model reports L for the case of flat electronic bands and neglects band bending, this treatment is more appropriate than estimation based on the metallic limit of L 0 .As shown in Figures 5(c) and 5(d), the room temperature values of κ − σLT are on the order of 0.2-0.7 W m −1 K −1 for all samples, and the trend of increasing κ lattice with increasing temperature suggests that surface and defect scattering of phonons dominate for all the samples measured here.The result also implies that the high total κ for x = 0.7 is due to its high electrical conductivity likely arising from a higher carrier concentration and possibly a lower electron-boundary scattering rate in the thicker sample.In comparison, undoped Bi 2 Te 3 nanoplates with thicknesses from ∼9 to 25 nm were found to possess lattice thermal conductivities on the order of 0.5-0.8W m −1 K −1 , 12 which is 1.2-3.1 times higher than the doped samples of this work.The difference can be attributed to additional alloy scattering in the (Bi 1−x Sb x ) 2 Te 3 samples studied here.Additionally, the room temperature κ lattice for bulk compounds

FIG. 1 . 3 Pettes
FIG. 1.(a) Crystal structure of the 2D layered material (Bi 1−x Sb x ) 2 Te 3 , where each quintuple layer is 10.17 Å thick. 32(b) Scanning electron and (c) transmission electron micrographs of the sample with x = 0.5 transferred onto a micro-bridge device, taken after deposition and removal of excess organic molecule F 4 -TCNQ.The boundaries of the nanoplate and amorphous coating are depicted by dashed lines.Scale bars are 10 µm for (b) and 10 nm for (c).

FIG. 2 .FIG. 3 .
FIG. 2. Energy dispersive x-ray spectroscopy (EDS) analysis of aged (Bi 1−x Sb x ) 2 Te 3 nanoplates for x = 0.50 shows uniform antimony distribution across the nanoplate after two years of storage in an inert atmosphere, scale bar is 500 nm.(a) High-angle annular dark-field image, contrast is from carbon thickness; (b) Bi L-edge; (c) Sb K-edge; (d) Te K-edge.(e) Background-corrected spectrum shows the signal to noise ratio of respective edges.

104810- 6 Pettes
FIG. 5. (a) Measured total thermal conductivity, κ, plotted versus electrical conductivity, σ, for (Bi 1−x Sb x ) 2 Te 3 nanoplates (NPs) shown in comparison with values reported for bulk Bi 2 Te 3 in the basal plane 27 (asterisks) and values for undoped Bi 2 Te 3 NPs 12 (stars).(b) Measured Seebeck coefficient, S, versus σ for the (Bi 1−x Sb x ) 2 Te 3 nanoplates (NPs) shown in comparison with values reported for bulk Bi 2 Te 3 in the basal plane 26 (asterisks), values for undoped Bi 2 Te 3 NPs 12 (stars), and the theoretical value for bulk Bi 2 Te 3 in the basal plane 12 (dotted line).(c) Temperature-and (d) thickness-dependent lattice contribution to the thermal conductivity for the (Bi 1−x Sb x ) 2 Te 3 NPs calculated by the Wiedemann-Franz law, κ − σLT, where the Lorenz number, L, has been obtained from a previously reported model 12 and the measured thermoelectric properties.The κ − σLT for undoped Bi 2 Te 3 NPs 12 (stars), κ lattice calculated for Bi 2 Te 3 from a phonon transport model in the diffuse boundary scattering regime 12 (line), and κ lattice for bulk Bi 2 Te 3 26 (asterisks) and bulk (Bi 0.5 Sb 0.5 ) 2 Te 3 2 (unfilled stars) in the basal plane are shown in (d) for comparison.Results for as-grown samples (filled symbols) and after deposition and removal of excess organic molecule F 4 -TCNQ (open symbols) indicate the surface doping procedure and remaining organic monolayers did not result in observable damage of the suspended NPs or contribute significantly to the measured total thermal conductance.The value of κ -σLT for the sample with x = 0.5 (green circle) is nearly identical before and after F 4 -TCNQ deposition.The legend shown in (a) applies to panels ((a), (b), and (d)), the legend shown in (c) applies to panels ((a)-(d)), nanoplate thicknesses are listed in parentheses.

104810- 7 Pettes
et al.APL Mater.4, 104810 (2016) TABLE I. Summary of sample dimensions and room temperature thermoelectric properties for as-grown and F 4 -TCNQ-surface doped (Bi 1−x Sb x ) 2 Te 3 nanoplates.The specified Sb content is nominally based on the composition of precursor materials.The lattice contribution to the thermal conductivity has been calculated as κ − σLT, where L is obtained from a previous model 12 based on the measured thermoelectric properties.σ (10 4 Ω/m) S (10 −6 V/K) κ (W/(m K)) κ − σLT (W/(m K))