Ionic thermoelectric supercapacitors
Dan Zhao, Hui Wang, Zia Ullah Khan, J. C. Chen, Roger Gabrielsson, Magnus
Jonsson, Magnus Berggren and Xavier Crispin
The self-archived postprint version of this journal article is available at Linköping
University Institutional Repository (DiVA):
http://urn.kb.se/resolve?urn=urn:nbn:se:liu:diva-128769
N.B.: When citing this work, cite the original publication.
Zhao, D., Wang, H., Ullah Khan, Z., Chen, J. C., Gabrielsson, R., Jonsson, M., Berggren, M., Crispin,
X., (2016), Ionic thermoelectric supercapacitors, Energy & Environmental Science, 9(4), 1450-1457.
https://doi.org/10.1039/c6ee00121a
Original publication available at:
https://doi.org/10.1039/c6ee00121a
Copyright: Royal Society of Chemistry
http://www.rsc.org/
Ionic Thermoelectric Supercapacitors
Dan Zhao
(a)
, Hui Wang
(a)
, Zia Ulla Khan
(a)
, Jincan Chen
(b)
, Roger Gabrielsson
(a)
, Magnus P. Jonsson
(a)
, Magnus
Berggren
(a)
and Xavier Crispin
(a)
*
(a)
Department of Science and Technology,Campus Norrköping, Linköping University,S-60174
Norrköping, Sweden,
(b)
Department of Physics, Xiamen University, People’s Republic of China
* Corresponding author: Xavier.crispin@liu.se
Temperature gradients are generated by the sun and a vast array of technologies and can
induce molecular concentration gradients in solutions via thermodiffusion (Soret effect). For ions,
this leads to a thermovoltage that is determined by the thermal gradient ΔT across the electrolyte,
together with the ionic Seebeck coefficient α
i
. So far, redox-free electrolytes have not been explored
in thermoelectric applications due to a lack of strategy to harvest the energy from the Soret effect.
Here, we demonstrate the conversion of heat into stored charge via the ionic Soret effect in an Ionic
Thermoelectric Supercapacitor (ITESC), thus providing a new means to harvest energy from
intermittent heat sources. We show that the stored electrical energy of the ITESC is proportional to
(ΔTα
i
)
2
and that its α
i
reaches beyond 10 mV/K. The resulting ITESC can convert and store several
thousand times more energy as compared to a traditional thermoelectric generator connected in
series with a supercapacitor.
INTRODUCTION
Various thermoelectric concepts are currently under investigation for conversion of thermal
energy into electrical energy, with the goal to provide efficient thermoelectric systems. First, electronic
charge carriers in a conductor thermodiffuse when subjected to a temperature gradient, which leads
to a thermovoltage known as the Seebeck voltage. Thermoelectric generators (TEGs) that utilize the
Seebeck effect are typically composed of semi-metals [1, 2], inorganic semiconductors [3, 4], and
electronically conducting polymers have also recently been explored [5]. Secondly, thermovoltages can
originate from the thermogalvanic effect, which results from temperature-dependent entropy changes
during electron transfer between a redox molecule and an electrode [6]. Hence, thermogalvanic cells
are based on electrolytes with redox couples, such as ferricyanide/ferrocyanide. The Soret effect [7]
of redox free electrolyte, i.e. from ionic charge carriers constitute yet a third thermoelectric concept
that, to the best of our knowledge, has not previously been considered for energy harvesting.
Analogous to the electronic Seebeck effect, the Soret effect is a result of thermo-diffusion of ions in an
ionic solid [8, 9] or electrolyte [10]. This produces an ionic concentration gradient and a corresponding
thermo-voltage that is governed by the temperature difference across the material and the ionic
Seebeck coefficient α
i
.
For a traditional thermoelectric leg, composed of a semiconductor and two metal contacts, a
constant electrical power can be provided to an external load by imposing a temperature gradient
along the metal-semiconductor-metal stack. The same harvesting principle is, however, not directly
applicable if the semiconductor is replaced by an electrolyte solution with ions as charge carriers. The
reason for this is that the thermo-diffused ions are blocked at the surface of the metal electrode and
cannot pass through the external circuit. Instead, the ions will be accumulated in excess at the metal
surface where they form an electric double layer capacitor (EDLC) [11]. Our novel approach utilizes this
EDLC principle and harvest energy from the ionic thermoelectric effect by charging and discharging
carbon nanotube super-capacitor electrodes. The resulting device is here referred to as an ionic
thermoelectric supercapacitor (ITESC). In addition to demonstrating and characterizing this novel
concept, we compare the characteristics of the ITESC with those obtained for a traditional electronic
thermoelectric generator (TEG) coupled to a supercapacitor (SC).
RESULTS
1. Thermoelectric characterization of PEO-NaOH.
We choose a low molecular weight PEO ended with alcohol groups (Mw=400g/mol) as the
electrolyte, because it is non-volatile and can withstand relatively high temperatures (<120 C). The
facilitated thermoelectric effect is stable since there is no solvent evaporation and degradation for
large ∆T over long period of times. We added NaOH to the liquid PEO (C
(NaOH)
=0.75mM) to transform
the terminating alcohol groups (-C-OH) into anionic alkoxide end-groups (-C-O
-
Na
+
). [12] Nuclear
magnetic resonance (1H-NMR) and Fourier Transform Infrared Spectroscopy (FTIR) confirm that the
reaction has occurred (fig. S1 and S2 in supplementary information 1). The resulting electrolyte is
composed of polymeric anions that are very weakly mobile and mobile Na
+
cations. The ionic
conductivity measured by impedance spectroscopy is 8.13×10
-5
S/cm at room temperature and the
activation energy for the ionic transport is 476meV (see supplementary information 2). We investigate
the thermoelectric properties of the PEO-NaOH solution. The thermoelectric liquid is injected into a
small cylindrical chamber (1mm thick, diameter of 10mm). Planar Au electrodes are evaporated on
both sides of the chamber and are in direct contact with the polymer electrolyte (fig.1a). In order to
monitor the temperature difference between the two electrodes, we fabricated thermistors under the
electrodes (details are provided in supplementary information 3). At any given ∆T, the corresponding
Soret-induced open voltage can be obtained. This is exemplified in Fig. 1b during heating one side of
the device while cooling of the other side. The thermovoltage induced by different ∆T was investigated
in the temperature range between 25 C and 35 C. For each ∆T, the voltage was measured 5min after
stabilization. The results are presented in Fig. 1c and demonstrate that V
thermo
varies linearly with ∆T.
The slope of the linear fit gives an ionic Seebeck coefficient of +11.1 mV/K for the PEO-NaOH electrolyte.
The thermovoltage originates from non-compensated thermodiffusion of cations and anions, with an
overall sodium cation accumulation at the cold electrode, as expected since alkoxylate and carboxylate
anions are mostly immobile. The magnitude of the ionic Seebeck coefficient is remarkably high. Indeed
for Na
+
in a diluted aqueous solution, the heat of transport Q* is about 3 kJ/mol, [13] giving an expected
Seebeck coefficient of α
Na+
=Q* /N
a
|e|T=0.1mV/K, where N
a
is the Avogadro number and |e| is the
charge of an electron. This is 100 times smaller than what is found for the PEO-NaOH electrolyte. The
Born model [14] allows a first estimate of the heat of transport [15]
and the thermopower. [16]
In this
model, the square of the dielectric constant of the medium is inversely proportional to the Seebeck
coefficient. Although the static dielectric constant of PEO (ε
0
=36.1 at 20°C) is lower than that of water
(ε
0
=80.6 at 20°C), its temperature variation (dε/dT=0.02) is negligible compared to water (dε/dT=0.49).
As a result of those compensating effects, we conclude that the large difference in the thermovoltage
of Na
+
in water and in PEO cannot be attributed to the difference in dielectric constant (see
supplementary information 4). Thermodiffusion is a complex process that also involves the interaction
and entropy change between the solvent and the ions.[17] The complexity is not reduced with the
presence of polymer chains. For instance, ions in polyelectrolyte and salt solution are expected to have
a Soret coefficient 100 times larger than a salt solution without polyelectrolyte. [18] This could be one
origin for the large ionic Seebeck coefficient measured for PEO-NaOH. Other reports also indicate large
thermoelectric coefficients for polymer electrolyte membranes, from few mV/K to around 500 mV/K
[19]. Aside from the major ionic Seebeck voltage, there may be another minor contribution due to a
thermogalvanic voltage associated with the Faradic process involved in the leakage current [20].
Regardless of reason, it is clear that ionic polymer solutions can provide remarkably high Seebeck
coefficients, and relatively low thermal conductivity of 0.216 wm
-1
k
-1
, as measured by the 3-omega
method (see supplementary information 5), which may be found useful in various applications.
The Au electrodes of the device can be functionalized with self-assembled multi-wall carbon
nanotubes to form a supercapacitor (fig. 1a) (CNT, detailed preparation protocols are provided in the
supplementary information 6). CNTs self-assembled onto the Au-contact provide intercalation of
electrolyte components, thus providing a relatively much higher capacitance value over a projected
area unit. Both Au and CNT electrodes are electrochemically stable over a potential window much
wider than what is measured from the induced thermovoltage, which means that no Faradic processes
related to electrode deterioration are involved in the ITESC measurements. Similar saturation of V
thermo
vs. time (inset fig. 1b) and evolution of V
thermo
vs. ∆T (fig.1c) are obtained using both Au and CNT
electrodes, which indicates that the nature of the electrodes does not play a role and that the
thermovoltage is an intrinsic property of the polymer electrolyte.
2. Capacitance properties characterization
Carbon nanotube films have previously shown to be suitable as electrodes in
supercapacitors,[21] primarily due to their large surface area [22]. Application of a voltage over the
PEO-NaOH capacitor induces migration of the cations towards the negatively charged CNT electrode,
at which an electric double layer (EDL) is formed (illustrated in fig 2a). We compare the electrical
characteristics of three devices that are based on the same polymer electrolyte PEO-NaOH, but with
different electrodes: two different amounts of CNT assembled onto Au electrodes (we define them as
thick and thin CNT electrodes) and the bare gold electrode. Before exploring the effect of a thermal
gradient, we characterize the capacitors electrically (at 22°C) by impedance spectroscopy and cyclic-
voltammetry.
With impedance spectroscopy, an alternating voltage V
ac
(f)= 5 mV is applied across the
electrolyte-based supercapacitors, the polarization characteristics of the electrolyte strongly depend
on the frequency (f)
[23, 24] and the total complex impedance Z
Tot
(f) is measured. Fig. 2b shows Nyquist
plots, i.e. the real Z
Re
(f) vs. the imaginary Z
Im
(f) part of the impedance of the device with Au, thin and
thick CNT electrodes, in the frequency ranging from 0.001Hz to 100kHz (without any DC bias). As shown,
all the Nyquist plots consist of two distinct parts including a distorted semicircle at the high frequency
region and a sloped line in the low frequency region. The x-intercept in the Nyquist plots represents
the equivalent series resistance (R
s
), which corresponds to the electrolyte solution resistance. R
s
is
almost identical for the different devices and is about 1.5kΩ. Also, the presence of a semicircle
indicates that an additional resistive element (R
p
) acts in parallel and it is associated to a low-frequency
leakage resistance (see equivalent circuit in the inset of fig. 2). The equivalent circuit parameters (table
1) have been obtained by fitting the experimental Nyquist spectra. For the Au electrodes, the
capacitance is around 5μFcm
-2
, which is typical of EDLCs [25]. Because of the extremely high effective
surface area of the CNT electrode networks (≈120-430m
2
g
-1
) [26], the specific capacitance is typically
larger than for a planar metal electrode [19] and reaches in our case 1.03mFcm
-2
for the thick CNT
electrodes and 0.48mF cm
-2
for the thin electrodes.
Typical cyclic voltammetry (CV) characteristics of the different devices are presented in Fig 2c
for the same scan rate (dV/dt=10mV/s). The devices exhibited a near-ideal rectangular shape,
indicating purely capacitive behavior. The amplitude of the current is proportional to the capacitance
of the electrodes (I=-CdV/dt): small for Au and large for CNT electrodes. The capacitance of the thick
CNT electrode is estimated to about 0.95mFcm
-2
; while for the thin CNT electrodes it is about
0.48mFcm
-2
. This is consistent with the result from the impedance measurements.
The charge and discharge characteristics of the SCs are studied by electric potential static
charge-discharge measurements (Fig. 2d). For the charging procedure, a power supply, R
load
and SCs
are connected in series, and the voltage across R
load
is recorded. Around 5 min after the recorded
voltage stabilizes, the power supply is replaced by a short at those two electrodes, and the discharging
current is obtained by recording the voltage drop over R
load
. The device is charged with a constant
potential of 50mV. Two measurements with different charging times are shown in the graph, indicating
that the discharging curve does not change for prolonged charging times. This means that the stable
leakage current, which is most likely a Faradic current between the electrode and the electrolyte, does
not contribute to the accumulated charge on the CNT electrodes. In other words, the amount of charge
transferred to and stored in the capacitor will always be less than the calculated charge using Q=CV
load
,
since a part of the charging voltage is lost across R
load
even at full charging. We infer that this leakage
current is related to the parallel resistance (R
p
) in the equivalent circuit. After charging, the stored
charge may gradually vanish due to self-discharging. We investigate this by measuring the decay of the
open voltage (without connecting R
load
) in Fig. S7 (see supplementary information 7). The device with
thick CNT electrodes provides significantly better charge retention than the device with thin CNT
electrodes. For this reason, we choose to focus the rest of the investigation on the PEO-NaOH based
supercapacitor with thick CNT electrodes.
