Simultaneous enhancement of conductivity and Seebeck coefficient in an organic MotttransistorYoshitaka Kawasugi, Kazuhiro Seki, Yusuke Edagawa, Yoshiaki Sato, Jiang Pu, Taishi Takenobu, Seiji Yunoki,Hiroshi M. Yamamoto, and Reizo Kato Citation: Applied Physics Letters 109, 233301 (2016); doi: 10.1063/1.4971310 View online: http://dx.doi.org/10.1063/1.4971310 View Table of Contents: http://scitation.aip.org/content/aip/journal/apl/109/23?ver=pdfcov Published by the AIP Publishing Articles you may be interested in Apparatus for measuring Seebeck coefficient and electrical resistivity of small dimension samples using infraredmicroscope as temperature sensor Rev. Sci. Instrum. 84, 054903 (2013); 10.1063/1.4805016 Simultaneous measurement of the Seebeck coefficient and thermal conductivity in the cross-sectional directionof thermoelectric thick film J. Appl. Phys. 112, 104511 (2012); 10.1063/1.4766911 First principles study of Seebeck coefficients of doped semiconductors ZnTe1−xFx and ZnTe1−yNy J. Appl. Phys. 111, 033701 (2012); 10.1063/1.3679569 Huge Seebeck coefficients in nonaqueous electrolytes J. Chem. Phys. 134, 114513 (2011); 10.1063/1.3561735 Controlled doping of phthalocyanine layers by cosublimation with acceptor molecules: A systematic Seebeck andconductivity study Appl. Phys. Lett. 73, 3202 (1998); 10.1063/1.122718

Reuse of AIP Publishing content is subject to the terms at: https://publishing.aip.org/authors/rights-and-permissions. Download to IP: 134.160.214.55 On: Tue, 06 Dec 2016

04:47:08

Simultaneous enhancement of conductivity and Seebeck coefficientin an organic Mott transistor

Yoshitaka Kawasugi,1,a) Kazuhiro Seki,2,3 Yusuke Edagawa,4 Yoshiaki Sato,1 Jiang Pu,4

Taishi Takenobu,5 Seiji Yunoki,2,3,6 Hiroshi M. Yamamoto,1,7 and Reizo Kato1

1Condensed Molecular Materials Laboratory, RIKEN, Wako, Saitama 351-0198, Japan2Computational Condensed Matter Physics Laboratory, RIKEN, Wako, Saitama 351-0198, Japan3Computational Materials Science Research Team, RIKEN Advanced Institute for Computational Science(AICS), Kobe, Hyogo 650-0047, Japan4Department of Applied Physics, Waseda University, Tokyo 169-8555, Japan5Department of Applied Physics, Nagoya University Furo-cho, Chikusa-ku, Nagoya, 464-8601, Japan6Computational Quantum Matter Research Team, RIKEN, Center for Emergent Matter Science (CEMS),Wako, Saitama 351-0198, Japan7Research Center of Integrative Molecular Systems (CIMoS), Institute for Molecular Science, Okazaki,Aichi 444-8585, Japan

(Received 30 August 2016; accepted 17 November 2016; published online 5 December 2016)

We report on the electrical conductivity and Seebeck coefficient of an electric-double-layer

transistor based on an organic Mott insulator. The measurements were performed along the two

in-plane crystallographic axes (a and c) of the same device. While the Seebeck coefficient along the

a-axis was decreased by electron or hole doping, the value along the c-axis was increased by hole

doping. This is in contrast to the general trade-off relation between the conductivity and the Seebeck

coefficient. The simultaneous enhancement of the conductivity and the Seebeck coefficient is attrib-

uted to pseudogap formation in the hole-doped state, where a steep slope of the density of states

emerges at the chemical potential because of the electron interaction. Published by AIP Publishing.[http://dx.doi.org/10.1063/1.4971310]

Thermoelectric devices, which directly convert heat into

electricity, have recently attracted considerable attention as

energy-harvesting applications. Thermoelectric efficiency is

given by the dimensionless figure of merit ZT, which is

defined as

ZT ¼ rS2T

j; (1)

where r is the electrical conductivity, S is the Seebeck coef-

ficient, and j denotes the thermal conductivity at tempera-

ture T. All these parameters depend on the carrier density

and generally have trade-off relations, for example, the

Seebeck coefficient decreases when the conductivity is

enhanced by carrier doping. However, such a trade-off rela-

tion can be violated and a high ZT is expected under specific

conditions. According to the Mott formula,1 S in metals or

degenerate semiconductors is expressed as

S ¼ p2

3

k2BT

�eð Þ@ ln r Eð Þ@E

� �E¼EF

; (2)

where EF is the Fermi energy and �e is the electron charge.

Provided that r(E) is proportional to the density of states

N(E), S is proportional to 1/N(E) and @N(E)/@E at EF.

Therefore, the trade-off relation between r and S may be

overcome if @N(E)/@E is sufficiently large. Indeed, the

simultaneous enhancement of r and S resulted in a high

power factor rS2 in the Heusler alloys Fe2VAl1�xGex, which

have a dip of N(E) near EF.2 Such a dip structure of N(E) is

called a pseudogap.

We recently reported on the pseudogap in an electric-

double-layer transistor (EDLT) based on an organic Mott

insulator j-(BEDT-TTF)2Cu[N(CN)2]Cl (abbreviated to

j-Cl).3 j-Cl is a quasi-two-dimensional organic Mott insula-

tor, which comprises the alternating layers of conducting

BEDT-TTFþ0.5 radical cations and insulating Cu[N(CN)2]Cl�

counteranions (Fig. 1(a)). According to band calculations,

j-Cl is expected to be a metal because the highest occupied

molecular orbital (HOMO) band of BEDT-TTF is three-

quarters filled. However, the strong dimerization of BEDT-

TTF makes it effectively half-filled. The strong onsite (on-

dimer) Coulomb repulsion in the half-filled band makes the

electrons be aligned commensurately with the lattice potential,

thereby resulting in the Mott insulating state. When electrons

or holes were doped into the organic Mott insulator, the com-

mensurability was reduced, and the effect of the electron inter-

action was weakened, resulting in a marked decrease in the

resistivity. However, the pseudogap remained at specific k-

points under hole doping. The pseudogap emerged as a result

of interactions between carriers at EF. Therefore, a small N(E)

and large @N(E)/@E around EF are expected in the pseudogap

state. In this letter, we demonstrate that r and S can be

increased simultaneously in an organic Mott insulator by car-

rier doping. The simultaneous increase is probably because of

the large enhancement of @N(E)/@E due to pseudogap

formation.

Thin single crystals of the organic Mott insulator j-Cl

were electrochemically synthesized. A crystal was laminated

on a polyethylene naphthalate (PEN) substrate, on whicha)Electronic mail: kawasugi@riken.jp

0003-6951/2016/109(23)/233301/4/$30.00 Published by AIP Publishing.109, 233301-1

APPLIED PHYSICS LETTERS 109, 233301 (2016)

Reuse of AIP Publishing content is subject to the terms at: https://publishing.aip.org/authors/rights-and-permissions. Download to IP: 134.160.214.55 On: Tue, 06 Dec 2016

04:47:08

18-nm-thick Au electrodes and heaters had previously been

evaporated. The j-Cl crystal was shaped into a cross along

the crystallographic a- and c-axes, which are usually parallel

to the diagonals of the rhombic crystal, using a pulsed laser

beam with a wavelength of 532 nm (Fig. 1(b)). The crystal

axes were determined by its diamond shape before laser-

cutting, where the longer diagonal is a-axis and shorter

diagonal is c-axis. This crystal orientation is also confirmed

by the signs of Seebeck coefficients themselves.4 The

EDLT device was fabricated by mounting an ion gel on the

cross-shaped crystal and the Au side gate electrode. We

employed poly(vinylidene fluoride-co-hexafluoropropylene)

[PVDF-HFP] with 58% w/w 1-butyl-3-methylimidazolium

tetrafluoroborate [BMIM-BF4] as the ion gel. Details of the

electrolysis and crystal lamination are described in our previ-

ous papers. The temperature was controlled using a Physical

Property Measurement System (Quantum Design) and the

thermal electromotive force was measured with a nanovolt-

meter (Agilent 34420A). The measurements were performed

at temperatures between 60 and 160 K where the resistance

was moderate (R< 105 X) and the ion gel was frozen, with

steps of 20 K (cooling and warming rates: 2 K/min). At lower

temperatures where the resistance was higher, the DV vs DTplots tended to deviate from the linear relation, namely, the

reliability of the Seebeck measurements deteriorated. The

gate voltage Vg was applied in the following order: �1.2,

�0.9, �0.6, �0.3, 0, 0.3, 0.6, 0.9, 1.2, 1.3, and 1.4 V. While

the gate voltage was varied, the sample was warmed to

220 K. For the Seebeck measurements, the thin Au lines

(width: 2 lm, thickness: 18 nm) patterned on the substrate

were employed as heaters to generate a temperature gradient.

For details of the Seebeck measurements, see supplementary

material. The thermovoltage DV linearly increased with the

temperature difference DT, giving the Seebeck coefficient Sas the slope as shown in Fig. 1(c).

First, we show the experimental results for r and S. The

gate-induced carriers were confined at the surface of the

j-Cl crystal. However, the bulk (thickness �40 nm) was also

conducting at the temperatures studied here (Fig. S2 of

supplementary material). To focus on the gate-induced sur-

face states, the surface conductivity rs is shown here. On the

other hand, we show the bare Seebeck coefficient to

overview its gate dependence because the estimation of the

surface value is more ambiguous than that of the conductiv-

ity. We attempt to estimate the surface Seebeck coefficient

in the latter part of this paper. Figures 1(d) and 1(e) show the

gate voltage and temperature dependencies of the surface

conductivity along the a- and c-axes (rsa and rsc), respec-

tively. Without a gate voltage, the resistivity was semicon-

ducting (Fig. S2 of supplementary material). Both a positive

and negative gate voltage enhanced the surface conductivity

(ambipolar transistor), where the minimum conductivity was

observed at Vg¼þ0.3 V. The gate-induced conducting sur-

face was metallic down to approximately 100 K. Comparing

the two crystallographic axes, rsc was more asymmetric

upon doping. On the assumption that the gate-induced car-

riers are confined in a single BEDT-TTF layer (1.5 nm), rs

was approximately 200 S/cm (a-axis, Vg¼�1.2 V, and

T¼ 100 K).

Figure 2 shows the gate voltage and temperature depen-

dencies of S for the heat flow along the a (Sa)- and c (Sc)-

axes. As shown in Fig. 2(a), the sign of Sa (Sc) was positive

(negative) in accordance with the bulk j-Cl.4 The effect of

doping on S strongly depended on the crystallographic direc-

tion. While jSaj was reduced monotonically by both electron

and hole doping, as in the case of typical semiconductors,

jScj was enhanced by hole doping. The contour plots in Figs.

2(b) and 2(c) clearly show this tendency at all the tempera-

tures studied here and also the slight increase of Sc by a small

amount of electron doping. The original data used to obtain

the contour plots are shown in Fig. S3 (supplementary

material).

We discuss these results by considering the electronic

state of j-Cl and its doping dependence. According to the

Boltzmann equation approach,5 ra and Sa (a¼ a, c) are given

as

ra ¼ e2 K0½ �aa; Sa ¼1

�eð ÞTK�1

0 K1

� �aa; (3)

where Kn (n¼ 0, 1) is defined as

FIG. 1. (a) Molecular arrangement of the conducting BEDT-TTF layer of

j-Cl (top view). Each conducting layer is separated by the closed-shell insu-

lating anion Cu[N(CN)2]Cl� along the b-axis (not shown). (b) Optical image

of a sample. The crystallographic axes were assigned from the sign of the

Seebeck coefficient. The side gate electrode (area: �600� 600 lm2) was

evaporated 300 lm away from the j-Cl crystal. (c) Thermovoltage DV vs.

the temperature difference DT at 160 K and Vg¼ 0 V. (d), (e) Contour plots

of the gate-induced conductivity rs along the a- and c-axes, respectively.

Here, rs is defined as rðVgÞ � rðVg ¼ 0:3VÞ, where Vg¼ 0.3 V is the charge

neutrality point.

233301-2 Kawasugi et al. Appl. Phys. Lett. 109, 233301 (2016)

Reuse of AIP Publishing content is subject to the terms at: https://publishing.aip.org/authors/rights-and-permissions. Download to IP: 134.160.214.55 On: Tue, 06 Dec 2016

04:47:08

Kn¼ 2Xm;k

sm kð Þvm kð Þvm kð ÞT �@f Eð Þ@E

� �E¼Em kð Þ

Em kð Þn: (4)

Here, Em(k) is the electron band dispersion of the band index

m at momentum k measured from the chemical potential,

vmðkÞ ¼ ðvamðkÞ; vc

mðkÞÞT

is the velocity vector with vamðkÞ

¼ @EmðkÞ=@ð�hkaÞ being the group velocity, smðkÞ is the qua-

siparticle lifetime, and f(E) is the Fermi distribution function.

The Mott formula Eq. (2) is based on the Sommerfeld expan-

sion of Eq. (3) which is valid when EF � kBT. These equa-

tions indicate that both r and S strongly depend on the Fermi

velocity vm(kF), where kF is the Fermi momentum defined by

Em(kF)¼ 0. Figure 3(a) shows the Fermi surface of the sim-

plest tight-binding model for undoped j-Cl without the elec-

tron interaction. When the electron interaction is considered,

no Fermi surface is expected because the ground state is the

Mott insulating state.3 However, the conducting states under

doping and/or at high temperatures are essentially deter-

mined by the original non-interacting band structure, which

can be modified by the electron correlation effect. The

electron-like Fermi sheet (blue part in Fig. 3(a)) has a moder-

ate energy dispersion along the a-axis, resulting in a small

group velocity along the a-axis. Therefore, it dominantly

contributes to rc and Sc. Likewise, the hole-like Fermi

pocket (red part in Fig. 3(a)) contributes more to ra and Sa

than to rc and Sc. In-plane anisotropy in S with a positive Sa

and a negative Sc has indeed been observed in various

j-BEDT-TTF salts, and the sign and the temperature depen-

dence of S were explained by the Boltzmann equation

approach based on the tight-binding band calculations.6,7

Therefore, to understand qualitatively the experimental

results, we assume that rc and Sc (ra and Sa) are governed

mostly by the electron-like Fermi sheet (hole-like Fermi

pocket).

In our previous report, the calculations based on the

cluster perturbation theory, which takes into account the

electron correlation effect, revealed that the Mott insulating

state became metallic and N(E) at the chemical potential was

increased by either electron or hole doping.3 However, the

evolution of N(E) with doping was quite asymmetric. The

pseudogap emerged at the center of the electron-like Fermi

sheets under hole doping but was almost absent in electron

doping. Figure 3(b) shows N(E) for 17% hole doping (which

roughly corresponds to a gate voltage of �1 V [Ref. 3]) in

the Brillouin zone enclosed by lines C-M-Z (blue solid line)

and C-M-X (red solid line), which roughly correspond to the

electron-like Fermi sheet and the hole-like Fermi pocket,

respectively. Both 1/N(E) and @N(E)/@E are larger in the for-

mer region than in the latter region. The parameter set for

the calculations (t0=t ¼ �0:44, U/t¼ 5.5, and t¼ 65 meV,

where t and t0 are the transfer integrals between the neighbor-

ing sites and the next-neighboring sites along the c-axis,

respectively, and U is the onsite Coulomb repulsion) is

quoted from the first principles calculations for j-Cl.8

However, we have also confirmed that the pseudogap

occurred by hole doping even with a different parameter set

(t0=t ¼ �0:8, U/t¼ 7, and t¼ 55 meV from semi-empirical

calculations) and different hole doping levels.3 The marked

doping asymmetry in rc observed experimentally may be

attributed to the smaller N(E) in the hole-doped region than

in the electron-doped region, which stems from the pseudo-

gap formation on the electron-like Fermi sheet. The slight

increase in Sc for low electron doping (Vg¼þ0.6 V in Fig.

2(c)) can also be attributed to a pseudogap because the calcu-

lations predicted the presence of minor pseudogaps under

electron doping (at a different portion of the electron-like

Fermi sheet).

The simultaneous enhancement of r and S has thus been

qualitatively understood from the aspect of pseudogap for-

mation. However, the measured S shown in Fig. 2 is strongly

suppressed by the bulk crystal because the bulk is also

FIG. 2. (a) Temperature dependence of the Seebeck coefficient at Vg¼ 0,

�1.2, and þ1.2 V. The error bars denote the sum of the standard deviation

of the DV vs DT plots and the error derived from the polynomial fitting of

DT=I2heater (see Fig. S1 in supplementary material). (b), (c) Contour plots of

the absolute value of the Seebeck coefficient along the a- and c-axes,

respectively.

FIG. 3. (a) Non-interacting Fermi surface of the simplest tight-binding

model for j-Cl (t0=t ¼ �0:44) at half-filling. (b) Density of states at 30 K for

17% hole doping (t0=t ¼ �0:44, U/t¼ 5.5, and t¼ 65 meV). The blue and

red solid lines show the density of states enclosed by lines C-M-Z and

C-M-X, respectively. The dashed line denotes the total density of states in

the first Brillouin zone. The Fermi energy is located at zero energy.

233301-3 Kawasugi et al. Appl. Phys. Lett. 109, 233301 (2016)

Reuse of AIP Publishing content is subject to the terms at: https://publishing.aip.org/authors/rights-and-permissions. Download to IP: 134.160.214.55 On: Tue, 06 Dec 2016

04:47:08

conducting. Finally, we estimate the surface Seebeck coeffi-

cient Ss and the surface power factor rsS2s from the measured

Seebeck coefficient of two parallel connected layers, which

is described as9,10

S ¼ rbSb þ rsSs

rb þ rs

; (5)

where the suffixes b and s denote the bulk and surface. Here

we employed r and S at the maximum resistivity point

(charge neutrality point) at Vg¼þ0.3 V as rb and Sb, respec-

tively. Figure 4 shows the gate voltage dependencies of rs,

Ss, and rsS2s at 100 K along the a- and c-axes. Note that the

values of Ss at Vg¼�0.3, 0, and þ0.3 V are not shown

because of the large ambiguity due to the small rs. Instead,

the corresponding values of S at Vg¼�0.3, 0, and þ0.3 V

are shown for reference. The maximum value of Ss reached

136 6 8 lV/K (c-axis, Vg¼�1.2 V), which is comparable

with that of the materials with pudding-mold-type band

structures such as NaxCoO2 (Ref. 11) and s-type organic

conductors.12 Therefore, we obtained a relatively large rsS2s

for an organic material (95 6 7 lWm�1K�2) in the pseudo-

gap state.

In summary, we measured r and S along the two in-

plane crystallographic axes (a and c) of the same organic

Mott EDLT based on j-Cl. Owing to the collapse of the

Mott-Hubbard gap, ra and rc were enhanced by both elec-

tron and hole doping. However, Sc was enhanced by hole

doping, probably because of the large enhancement of

@N(E)/@E at the chemical potential caused by the electron-

interaction-driven pseudogap formation, while no such effect

was observed in the a-axis direction. As a result, rc and Sc

were simultaneously increased by hole doping, giving a rela-

tively large rsS2s . These results indicate that the pseudogap in

strongly correlated electron systems, which has attracted

considerable attention in the field of condensed matter

physics, can be applied to improve the thermoelectric prop-

erties of a material.

See supplementary material for details of the Seebeck

measurements, temperature dependence of the device con-

ductance without gate voltage, and the original data used to

obtain the contour plots in Fig. 2.

We would like to acknowledge Teijin DuPont Films

Japan Limited for providing the PEN films. Computations

have been done using HOKUSAI facility of Advanced

Center for Computing and Communication at RIKEN. This

work was supported by MEXT and JSPS KAKENHI (Grant

Nos. JP16H06346, JP16H04140, JP15K17714, JP26102012,

and JP25000003), JST ERATO, and MEXT Nanotechnology

Platform Program (Molecule and Material Synthesis).

1N. F. Mott and H. Jones, The Theory of the Properties of Metals(Clarendon Press, Oxford, 1936).

2Y. Nishino, S. Deguchi, and U. Mizutani, Phys. Rev. B 74, 115115 (2006).3Y. Kawasugi, K. Seki, Y. Edagawa, Y. Sato, J. Pu, T. Takenobu, S.

Yunoki, H. M. Yamamoto, and R. Kato, Nat. Commun. 7, 12356 (2016).4M. A. Tanatar, S. Kagoshima, T. Ishiguro, H. Ito, V. S. Yefanov, V. A.

Bondarenko, N. D. Kushch, and E. B. Yagubskii, Phys. Rev. B 62, 15561

(2000).5J. M. Ziman, Principles of the Theory of Solids, 2nd ed. (Cambridge

University Press, Cambridge, England, 1972).6T. Mori and H. Inokuchi, J. Phys. Soc. Jpn. 57, 3674 (1988).7R. C. Yu, J. M. Williams, H. H. Wang, J. E. Thompson, A. M. Kini, K. D.

Carlson, J. Ren, M.-H. Whangbo, and P. M. Chaikin, Phys. Rev. B 44,

6932 (1991).8H. C. Kandpal, I. Opahle, Y.-Z. Zhang, H. O. Jeschke, and R. Valent�ı,Phys. Rev. Lett. 103, 067004 (2009).

9P. Pichanusakorn and P. Bandaru, Mater. Sci. Eng. R 67, 19 (2010).10T. A. Cain, S.-B. Lee, P. Moetakef, L. Balents, S. Stemmer, and S. J.

Allen, Appl. Phys. Lett. 100, 161601 (2012).11K. Kuroki and R. Arita, J. Phys. Soc. Jpn. 76, 083707 (2007).12H. Yoshino, H. Aizawa, K. Kuroki, G. C. Anyfantis, G. C. Papavassiliou,

and K. Murata, Phys. B 405, S79 (2010).

FIG. 4. (a) and (b) Gate voltage Vg dependencies of the surface conductivity rs and surface Seebeck coefficient Ss along the a- and c-axes at 100 K. Note that

the values of Ss at Vg¼�0.3, 0, and þ0.3 V are not shown because of the large ambiguity due to the small rs. Instead, the corresponding values S at

Vg¼�0.3, 0, and þ0.3 V (open symbols) are shown for reference. (c) Gate voltage dependence of the surface power factor rsS2s at 100 K. Here, we assumed

that the gate-induced carriers are confined within a single BEDT-TTF layer (1.5 nm), and we estimated the values in three spatial dimensions for comparison

with other materials.

233301-4 Kawasugi et al. Appl. Phys. Lett. 109, 233301 (2016)

Reuse of AIP Publishing content is subject to the terms at: https://publishing.aip.org/authors/rights-and-permissions. Download to IP: 134.160.214.55 On: Tue, 06 Dec 2016

04:47:08