pubs.acs.

org/NanoLett

Harvesting Waste Thermal Energy Using a

Carbon-Nanotube-Based Thermo-
Electrochemical Cell
Renchong Hu,† Baratunde A. Cola,§ Nanda Haram,| Joseph N. Barisci,† Sergey Lee,†,¶
Stephanie Stoughton,† Gordon Wallace,⊥ Chee Too,⊥ Michael Thomas,⊥ Adrian Gestos,⊥
Marilou E. dela Cruz,‡ John P. Ferraris,†,‡ Anvar A. Zakhidov,† and Ray H. Baughman*,†,‡
†
Alan G. MacDiarmid NanoTech Institute, ‡ Department of Chemistry, University of Texas at Dallas,
Richardson, Texas 75083, § George W. Woodruff School of Mechanical Engineering, Georgia Institute of Technology,
Atlanta, Georgia 30332, | Department of Engineering Sciences, Maharashtra Academy of Engineering, Alandi (D),
Pune 412 105, India, and ⊥ Intelligent Polymer Research Institute ARC Centre of Excellence for Electromaterials
Science, University of Wollongong, Northfields Avenue, Wollongong, NSW 2522, Australia

ABSTRACT Low efficiencies and costly electrode materials have limited harvesting of thermal energy as electrical energy using thermo-
electrochemical cells (or “thermocells”). We demonstrate thermocells, in practical configurations (from coin cells to cells that can be
wrapped around exhaust pipes), that harvest low-grade thermal energy using relatively inexpensive carbon multiwalled nanotube
(MWNT) electrodes. These electrodes provide high electrochemically accessible surface areas and fast redox-mediated electron transfer,
which significantly enhances thermocell current generation capacity and overall efficiency. Thermocell efficiency is further improved
by directly synthesizing MWNTs as vertical forests that reduce electrical and thermal resistance at electrode/substrate junctions. The
efficiency of thermocells with MWNT electrodes is shown to be as high as 1.4% of Carnot efficiency, which is 3-fold higher than for
previously demonstrated thermocells. With the cost of MWNTs decreasing, MWNT-based thermocells may become commercially
viable for harvesting low-grade thermal energy.

KEYWORDS Thermo-electrochemical cells, thermogalvanic cells, thermocells, carbon nanotubes, thermal energy harvesting,
low-grade waste heat

L ow-grade heat (temperature below 130 °C), from such The electrochemical Seebeck effect was first demon-
sources as industrial waste streams, geothermal activ- strated over 100 years ago.8,9 For a hypothetical redox
ity, and solar heating, is projected to be a major system B S ne- + A, the Seebeck coefficient, S, can be
sustainable energy source.1,2 For many years the research expressed as
community has focused on using solid-state thermoelectrics
and Stirling engines to efficiently harvest low-grade heat as ∂V ∆SB,A
S) ) (1)
electrical energy. However, despite much progress over the ∂T nF
past decades, current thermoelectric energy conversion
technology is not very cost-effective and is constrained by where V is the electrode potential, T is temperature, n is the
physical and material limitations,3 and Stirling engine tech- number of electrons involved in the reaction, F is Faraday’s
nology is disadvantaged by high initial cost and problems constant, and ∆SB,A is the reaction entropy for the redox
with long-term reliability.4 Thermo-electrochemical cells couple.10,11 Thermocells using aqueous potassium ferrocya-
(otherwise known as thermogalvanic cells or thermocells) nide/ferricyanide redox solution have been studied by many
that utilize the temperature dependence of electrochemical groups because this redox system reversibly exchanges one
redox potentials (i.e., the Seebeck effect) to produce electri- electron per iron atom and produces a large reaction en-
cal power may become an attractive alternative for harvest- tropy, yielding Seebeck coefficient (>1 mV/K) and high
ing low-grade heat, given their simple design, direct thermal- exchange current.11-16 However, to obtain efficiencies of
to-electric energy conversion, continuous operation, low reasonable interest noble metals such as Pt are usually
expected maintenance, and zero carbon emission.5-7 required as electrode materials in thermocells, and this
restricts commercial viability.14-18 Also, the best prior-art
thermocells typically have efficiencies of ∼0.40% of Carnot
* To whom correspondence should be addressed. E-mail: ray.baughman@ efficiency (when efficiency is correctly evaluated, as dis-
utdallas.edu.
¶
cussed below).15,16 In fact, it was previously predicted that
Current address: Plextronics, Inc. 2180 William Pitt Way, Pittsburgh, PA 15238.
Received for review: 09/30/2009 a power conversion efficiency of 1.2% of the Carnot ef-
Published on Web: 00/00/0000 ficiency would be difficult to achieve.15

© XXXX American Chemical Society A DOI: 10.1021/nl903267n | Nano Lett. XXXX, xxx, 000–000
FIGURE 1. (a) Schematic of a thermocell with nanostructured electrodes showing concentration gradients of the ferri-ferrocyanide redox
ions during power generation. (b) SEM micrograph of MWNT buckypaper. The average MWNT diameter is 10 nm. (c) SEM micrograph of vertical
MWNT forest. The average MWNT diameter is approximately 20 nm. The inset shows coinlike stainless steel substrates fully coated with
vertical MWNTs. Each substrate has a diameter of 2 cm. (d) Gerischer-Marcus model for the thermocell. The Van Hove singularities for MWNTs
are shown close to the Fermi level. E0 is the standard potential of Fe(CN)63-/Fe(CN)64-, λ is the reorganization energy of Fe(CN)63-/Fe(CN)64-
redox reactions.46

Carbon nanotubes (CNTs) have been of considerable in several practical scenarios. Our experiments revealed that
interest for electrochemical applications, due to their excep- the performance of MWNT-based thermocells is scalable,
tional electronic, mechanical, and chemical properties.19-24 and that reduced contact resistance at MWNT electrode/
Advanced electrodes that incorporate CNTs have been substrate junctions can significantly enhance efficiency.
shown to provide attractive performance for electrochemical The power conversion efficiency of a thermocell is ex-
applications (e.g., batteries, supercapacitors, and fuel pressed as
cells25-32), due to their high surface area, ability to carry
large current densities, and fast electron transfer kinetics.33-35 1/4VOCISC
We here describe CNT-based thermocells that utilize the η) (2)
Ak(∆T/d)
ferri/ferrocyanide redox couple and electrodes made from
carbon-multiwalled nanotubes (MWNT) buckypaper and
vertically aligned MWNT arrays (see Figure 1). This bucky- and the relative power conversion efficiency is ηr ) η/(∆T/
paper is made by a filtration process that is analogous to that Th), where VOC is the open circuit voltage, ISC is the short
used for making ordinary paper. The performance of MWNTs circuit current, A is the cross-sectional area of the cell, d is
as thermocell electrodes supersedes that of conventional the electrode separation distance, ∆T is the temperature
electrode materials, including Pt foil and graphite sheet. With difference between the two electrodes, k is the thermal
a hot-side temperature of 65 °C and a temperature differ- conductivity of the electrolyte (or the effective thermal
ence of 60 °C, a maximum output power of 1.8 W/m2 was conductivity of the electrolyte and separator), and Th is the
achieved in a stagnant cell, corresponding to an efficiency temperature at the hot side. The maximum output power,
relative to the Carnot cycle efficiency of 1.4%. Different PMAX, is 1/4VOCISC and Ak(∆T/d) is the input thermal energy
designs were developed to demonstrate energy harvesting needed to maintain ∆T. ∆T/Th is the Carnot efficiency. As

© XXXX American Chemical Society B DOI: 10.1021/nl903267n | Nano Lett. XXXX, xxx, 000-–000
previously described, erroneous assumptions have been surfaces of these coin cells. The MWNT forests were grown
often made in evaluating energy conversion efficiencies.15 using a trilayer catalyst (30 nm Ti, 10 nm Al, and 2 nm Fe),
Efficiencies are also overestimated if the A in eq 1 is taken plasma-enhanced CVD, and previously reported process
as the cross-sectional area of an electrode, and the effective conditions.41 The MWNT forests were approximately 100
area for ionic conduction and heat flow is much larger due µm tall. Vertical alignment of MWNTs and good coverage of
to a larger cross-sectional area of the cell. MWNTs on the surfaces of the packaging substrates is
The efficiency of the thermocell can be more fundamen- illustrated in Figure 1c. The thermal resistance at MWNT
tally expressed by combining eq 1 and eq 2 to give forest/substrate junctions was measured in previous work
to be less than 0.01 cm2K/W, and the resistance of a solder
1/4∆SB,A jSCRT joint, the buckypaper electrodes are effectively soldered to
η) (3) the packaging substrates in the coin cell configuration, is
nF
5-fold higher than this value.41 The electrolyte used for all
thermocells in this work was 0.4 M K3Fe(CN)6/K4Fe(CN)6
where jSC ()ISC/A) is the short circuit current density and RT aqueous solution, as this was demonstrated to be the
()d/k) is the thermal resistance of the cell. The ferri/ferro- optimum redox couple concentration.
cyanide redox couple, where n ) 1, is known to produce a The performance of MWNT buckypaper, Pt foil (Sigma-
relatively high reaction entropy,36 so we here focus on Aldrich Co.), and graphite sheet (98% graphite, GrafTech,
increasing the parameters that critically limit thermocell Inc.) electrodes were directly compared using a U-shaped
efficiency, jSC and RT. The intrinsically large surface area of electrochemical cell. The hot side temperature was a con-
MWNTs and fast electron transfer between MWNT elec- stant 65 °C while the cold side temperature was 5 °C, the
trodes and electrolyte directly enhance jSC.37-40 Thermal two electrodes were 5.0 cm apart, and each electrode sheet
conduction losses are mitigated (i.e., RT is increased) using was 1.0 cm × 0.5 cm. Conductive Ag paste was used to
two primary approaches that also enhance jSC, (1) we took promote good electrical contact between buckypaper elec-
advantage of the highly porous three-dimensional structure trodes and Pt wire leads. The contacts were then covered
of MWNT buckypaper by using scroll-like electrodes, which by insulating paint to prevent possible artifacts due to
are commonly used in batteries and capacitors, to reduce interaction between the Ag paste and the electrolyte. Every
cross-sectional area normal to the direction of heat flow data point of VOC and ISC was acquired after allowing suf-
while allowing a high degree of electrolyte exposure to the
ficient time for the values to stabilize. Similar Seebeck
electrode surface for redox reactions to occur, and (2) we
coefficients of 1.4 mV/K were measured for all the electrode
directly synthesized vertically aligned MWNT arrays (i.e.,
materials, which is expected because the Seebeck coefficient
nanotube forests) to establish electrodes that are in good
is governed by the thermodynamics of the redox couple, not
thermal and electrical contact with the thermocell pack-
the electrode material, as shown in eq 1.
aging.41,42
The jSC and power (PMAX) generated by the MWNT bucky-
The MWNT buckypaper was prepared according to pro-
cedures described in previous work.43,44 The MWNTs are of paper electrodes for an optimized electrical load were the
high purity (containing less than 1% catalyst) and hundreds highest among the tested electrode materials, 64.6 A/m2 and
of micrometers long, with diameters of ∼10 nm. The MWNT 1.36 W/m2, respectively. The normalized current density, jSC/
buckypaper is typically ∼35 µm thick and has an electrical ∆T, was 1.08 A/(m2K) and the normalized area power
conductivity of 100 S/cm44 and an internal surface area of density, PMAX/∆T2, was 3.8 × 10-4 W/(m2K2). The Pt elec-
278 m2/g by the Brunauer-Emmett-Teller method. Figure trodes generated a jSC of 48.4 A/m2 and a PMAX of 1.02 W/m2,
1b shows a scanning electron micrograph (SEM) of the corresponding to a jSC/∆T of 0.81 A/(m2K) and a PMAX /∆T2
MWNT buckypaper. Cyclic voltammograms (CV) show quasi- of 2.8 × 10-4 W/(m2K2). The graphite sheet electrodes
reversible electron transfer reactions between MWNT buck- generated a jSC of 36.6 A/m2 and a PMAX of 0.76 W/m2,
ypaper electrodes and Fe(CN)63-/4- in aqueous electrolyte. corresponding to a jSC/∆T of 0.61 A/(m2K) and a PMAX/∆T2 of
The MWNT buckypaper also shows three times higher 2.1 × 10-4 W/(m2K2). The MWNT buckypaper electrodes
charging current density during CV scans than the same thus produced 33 and 77% higher jSC/∆T than Pt and
physical area Pt foil. graphite sheet electrodes, respectively. Previous work re-
The electrical energy output of thermal electrochemical ported jSC/∆T ) 0.5 A/(m2K) and PMAX/∆T2 ) 2.0 × 10-4
cells can be optimized by ensuring that the temperature drop W/(m2K2) using Pt electrodes with the same interelectrode
is predominantly between the electrodes, rather than be- distance and only one side of each electrode exposed to
tween the heat source and heat sink and the corresponding electrolyte.16 When we blocked one side of the anode and
proximate electrodes. To reduce temperature drop from cathode with Scotch tape, jSC/∆T decreased 26% for the
thermal contact resistance for thermal cells based on con- MWNT electrodes and decreased 28% for the Pt electrodes,
ventional coin cell battery containers, MWNT forests were which brings the present jSC/∆T for Pt electrodes into close
directly synthesized on the internal stainless steel electrode agreement with this previous work.

© XXXX American Chemical Society C DOI: 10.1021/nl903267n | Nano Lett. XXXX, xxx, 000-–000
FIGURE 2. (a) The Teflon cell with adjustable interelectrode distance. The cell consists of a HOTWATT heater, a Cu rod sealed in Teflon tape
with cold liquid circulating inside, a glass frit separator, and two OMEGA K-type thermocouples wrapped with Teflon tape. (b) VOC and jSC
versus temperature difference between electrodes. The distance between electrodes, d, was 2.6 cm. Each electrode was 1.8 mg and 1.5 cm
× 0.8 cm × 30 µm. (c) ISC and PMAX versus MWNT electrode area with ∆T ) 47 °C, d ) 2.6 cm, MWNT buckypaper thickness 30 µm. (d) jSC and
PMAX versus buckypaper electrode thickness with the area of each electrode fixed at 1.0 cm × 0.5 cm, ∆T ) 60 °C, d ) 2.6 cm.

Upon reaching equilibrium, jSC in the Nernst-Plank equa- MWNT/electrolyte interface is enhanced as a result of fast
tion consists of diffusion current density jd ) FΣ(znDnDCn/Dx) kinetics.39 The improved kinetics over other carbon materi-
and migration current density jm ) F2/RT(DV/Dx)Σ(zn2DnCn), als such as graphite sheet is likely the result of the quasi-
where zn is the charge, Dn is the diffusion coefficient, Cn is one-dimensional crystalline structure of MWNTs and their
the ion concentration, R is the gas constant, and x is the resulting highly localized electron density of states near the
distance from one electrode to the point of interest between Fermi level (i.e., Van Hove singularities).45 The Gerischer-
the two electrodes (the other parameters are the same as Marcus model in Figure 1d illustrates this kinetic effect.
defined above). Here jm is similar in all cases because the The Teflon cell shown in Figure 2a was used to evaluate
potential differences across the electrodes are similar given performance scaling with ∆T. The distance between two
the same ∆T. Therefore, jd, or more specifically the concen- equivalent size MWNT buckypaper electrodes was set at 2.6
tration gradient, DCn/Dx, must be different for the Pt, MWNT, cm. The hot side was maintained at 70 °C using a resistive
and graphite electrodes studied here. One reason for the heater connected to a temperature controller and the cold
relatively large jSC (or jd) generated by MWNT electrodes is side temperature was adjusted by circulating coolant from
the large number of redox reaction sites established by the a thermostat bath. As shown in Figure 2b, both VOC and jSC
high internal surface area of such electrodes. This surface increase linearly with ∆T, indicating the cell operates within
area (or concentration gradient) advantage is likely the cause a linear region. Variation of external load resistance resulted
of MWNT electrodes producing higher current densities than in linear variation of cell current on the voltage drop across
Pt foil electrodes because the kinetics of the two materials the load resistor, so output power was optimized when
appears to be similar within the resolution of our experi- effective internal cell resistance equaled the external load
ments (see Figure S1 of Supporting Information). However, resistance. The cell in Figure 2a was also used to test the
high-surface-area activated carbon fabric (Spectracarb 2225 feasibility of making larger MWNT-based thermocell devices
from Engineered Fibers Technology, LLC with an internal (these tests were conducted with ∆T fixed). As shown in
surface area of 2500 m2/g) yielded a low jSC/∆T of 0.05 Figure 2c, increasing the electrode area from 0.4 to 1.2 cm2
A/(m2K) (evaluated using geometric area) when tested in the increased the short circuit current from 0.8 to 2.7 mA and
U-shaped cell, which implies that surface area enhancements increased the maximum power output from 16 to 55 µW.
alone are not sufficient to generate large jSC. The compari- For the given range of electrode sizes, a linear relationship
sons of the cyclic voltammetry (CV) scans for MWNT and between thermocell performance and electrode area is
graphite sheets in Figure S1 of Supporting Information clearly shown, indicating scalability toward larger devices.
clearly indicate slower electron transfer on graphite elec- When the thickness of buckypaper electrodes with fixed area
trodes. It has been suggested that electron transfer at the was increased from 40 to 280 µm, jSC increased from 47 to

© XXXX American Chemical Society D DOI: 10.1021/nl903267n | Nano Lett. XXXX, xxx, 000-–000
 efficiency is 1.4%, which is 17% higher than a previously
estimated upper bound for thermocells using the ferri/
ferrocyanide redox couple.15
Thin coin-type thermocells with MWNT buckypaper and
vertically aligned forest electrodes, which could potentially
be powered by body heat, were used to evaluate the feasibil-
ity of stable long-term performance and scalable device
fabrication (see Figure 4a). The thermocells were constantly
discharged through a resistor load of 10 Ω. For a nominal
∆T of 45 °C, voltage was 51.2 mV and current density
stabilized at 30.4 A/m2 in the coinlike thermocell with
FIGURE 3. Mark II cell with MWNT scrolls as electrodes. Two
thermocouples (not shown) were used to measure temperature in buckypaper electrodes (see Figure 4b). For a nominal ∆T of
the cell near the hot and cold electrodes. 60 °C, VOC was 70.4 mV and jSC stabilized at 55.7 A/m2 (jSC/
∆T ) 0.93 A/(m2K)) in the coinlike thermocell with MWNT
forest electrodes (see Figure 4c). Because of vertical nano-
67 A/m2 and PMAX increased from 0.9 to 1.45 W/m2 (see tube alignment that ostensibly improves electrolyte access
Figure 2d). The dependencies of jSC and PMAX on buckypaper to MWNT surfaces, and because of improved thermal and
sheet thickness were nonlinear and a saturation thickness electrical contact to the packaging substrates,41,42 the ef-
was evident during testing, likely because diffusion of the ficiency of coinlike thermocells with MWNT forest electrodes
redox mediator through increasing thickness sheets be- was approximately 30% higher than the efficiency of coin
comes a limiting factor. To demonstrate voltage and power cells with buckypaper electrodes (we estimate that ap-
scaling, two identical thermocells (as shown in Figure S4 of proximately 4 °C of temperature was lost across the less
Supporting Information) were connected in series. The open intimate buckypaper/substrate junctions).
circuit voltage for the combined cells was 140 mV, which is Concerns about the stability of potassium ferrocyanide/
equal to the sum of the open circuit voltages of the individual ferricyanide, especially upon illumination, have been repor-
cells. The maximum output power of the combined cells was ted.47,48 Previous work also reported the possibility of mass
101.5 µW, which approximately equals the sum of the transport overpotential leading to detrimental buildup of
maximum output powers of the two cells when they are concentration gradient in the cell, in particular the establish-
individually operated (see Table S2 of Supporting Informa- ment of higher electrolyte concentration at the cold side.
tion). Such claims suggest a decline in current if thermocells are
The electrode spacing is adjustable in the cell shown in subject to long-term discharge. However, as shown in Figure
Figure 2a. When electrode separation distance was de- 4, the coinlike thermocell with buckypaper electrodes was
creased, both jSC and PMAX increased because ions could successfully continuously discharged over a 10 Ω load for
diffuse quickly over shorter distances, that is, ionic mass more than three months, and the coinlike thermocell with
transport was enhanced; yet, efficiency dropped significantly MWNT forest electrodes was continuously discharged for a
because a higher thermal energy input was required to week with no change in performance. For each test the
maintain the same ∆T. Similar observations were made in circuits were disconnected only when measuring VOC, which
a previous study.16 We here addressed this trade-off be- takes about 5 min per data point. Both voltage and the
tween current density and heat loss by forming MWNT discharged current were stable during long-term testing,
buckypaper into scrolls and aligning them along the rolling which demonstrates the stability of the aqueous equimolar
axis to minimize the heat loss channel and increase jSC. Two redox solution in closed systems. We posit that this level of
buckypaper scroll electrodes, each having a diameter of 0.3 stability occurs because the electrolyte reaches equilibrium
cm and weighing 0.5 mg, were positioned in a glass tube and the MWNTs are very stable in the solution within the
with a diameter of 0.3 cm so that the scroll ends faced each temperature range. The demonstrated stability of MWNT-
other (this configuration is shown as Mark II in Figure 3). The based thermocells during continuous use suggests that if the
distance between the electrodes was d ) 5.0 cm. The hot heat source is stable the power generation will always
side was held at 65 °C by a temperature controller and the operate at near 100% capacity.
cold side of the cell was immersed in a mixture of ice/water A thermocell produces its maximum VOC once the differ-
and had a temperature of 5 °C. A jSC of 85 A/m2, or jSC/∆T of ence in redox ion concentration at the electrodes reaches
1.42 A/(m2K), and a PMAX of 1.8 W/m2, or PMAX/∆T2 of 5 × its highest value and the electron transfer process at the
10-4 W/(m2K2), were measured based on the cross-section electrode/electrolyte interfaces reach equilibrium. It can take
area of the cell. The power conversion efficiency for the Mark as much as 10 min for this to occur in a thermocell where
II thermocell is 0.24%, which is approximately an order of the electrolyte is stagnant, as shown in Figure 5c. The actual
magnitude higher than values obtained in prior work using equilibrium time depends on partially the amount of elec-
Pt electrodes under similar test conditions.16 The relative trolyte around the electrode with more electrolyte requiring

© XXXX American Chemical Society E DOI: 10.1021/nl903267n | Nano Lett. XXXX, xxx, 000-–000
FIGURE 4. (a) Schematic assembly of the coinlike thermocell. Each buckypaper electrode is 0.5 cm × 0.5 cm and attached to the package with
Ag paste. The MWNT forest electrodes are grown directly on the caps as shown in Figure 1c. The electrodes are separated by five layers of
Nomex HT4848 soaked with electrolyte. The cells are sealed by pneumatic crimping. The cell diameter is 2.0 cm and height is 0.2 cm. (b)
Voltage and current versus operation time for a coin cell with buckypaper electrodes discharging over a 10 Ω load. The nominal ∆T was 45
°C (variation <3%). (c) VOC and jSC versus time for coin cell with MWNT forest electrodes. The nominal ∆T was 60 °C.

longer time. After the maximum VOC was reached, shorting testing, which suggests that stable performance can be
the circuit for the “stagnant” thermocell in Figure 5c de- achieved in long-term operation.
creased jSC from its peak value of 105 A/m2 to approximately Because numerous pipes that carry hot or cold fluids are
half its peak value, 49 A/m2. To sustain jSC at its maximum installed in facilities such as power and chemical processing
value consumed redox ions must be replenished fast enough plants, a thermocell wrapped around a pipe is demonstrated
at the electrode/electrolyte interfaces to eliminate the limit- as a candidate technology for harvesting these rich supplies
ing effects of mass transport via sluggish ion diffusion. of energy. Such pipes can serve as hot or cold sources for
Stirring or circulating the electrolyte to quickly deliver fresh thermocells with the ambient regulating the temperature of
electrolyte to the electrodes can achieve this, and such the electrode not in contact with the pipe (imagine steam
approaches have been demonstrated to increase jSC in flowing through a pipe located in Alaska as a potential
thermocells with Pt electrodes.37 A MWNT-based thermocell scenario for this energy harvesting concept). Here we dem-
with circulated electrolyte (i.e., a “flow” cell) is shown in onstrate a thin thermocell configuration wrapped around a
Figures 5a,b. A pump located between each compartment stainless steel pipe with cool liquid flowing within (ammonia
and an electrolyte reservoir associated with each compart- flowing through a pipe in a chemical plant in Florida is a
ment circulated the electrolyte. The temperatures of the hot candidate scenario). The pipe assembly consisted of a first
and cold reservoirs were held constant by a temperature- layer of MWNT buckypaper electrode, then two layers of
controlled heater and heat sink, respectively. Each compart- Nomex HT4848 as separator, and an outer layer of MWNT
ment contained one MWNT buckypaper electrode that was buckypaper electrode. A piece of stainless steel sheet was
1.2 cm × 0.5 cm. Stagnant and flow MWNT-based thermo- wrapped around the outer layer buckypaper and two clamps
cells driven by the same ∆T are compared in Figure 5c. and glue were used to seal the entire assembly. A resistive
Notably, VOC and jSC were maintained in the flow thermocell heater was wrapped around the stainless steel sheet. Elec-
at the stagnant cell peak values because fresh electrolyte was trolyte was injected using a syringe, and afterward the
constantly delivered to the electrodes as discussed above. syringe hole was sealed with glue. The schematic layout and
For the flow cell with MWNT electrodes, jSC/∆T reached 1.12 actual assembly of the pipe-wrapped thermocell is shown
A/(m2K), while the stagnant cell provided only 0.6 A/(m2K). in Figure 6. This cell produced a VOC of 21 mV, which
For comparison, the flow cell using Pt electrodes provided corresponds to a ∆T of 15 °C across the thin layer structure,
a jSC/∆T of 0.89 A/(m2K), which is close to previously and a jSC and PMAX of 7.5 A/m2 (jSC/∆T ) 0.5 A/(m2K)) and
reported maximum value for a similar Pt electrode flow 0.039 W/m2 (PMAX/∆T2 ) 1.8 × 10-4 W/(m2K2)), respectively.
cell.37 The PMAX/∆T2 of the MWNT electrode flow cell reached The cost of renewable energy technologies per unit watt
4 × 10-4 W/(m2K2), as compared with 2 × 10-4 W/(m2K2) of generated power is a metric that greatly affects com-
for the stagnant cell. As shown in Figure 5d, the flow mercial viability assuming similar product lifetimes. The
thermocell produced stable VOC and jSC during 8 days of material cost of coinlike thermocells includes the cost of the

© XXXX American Chemical Society F DOI: 10.1021/nl903267n | Nano Lett. XXXX, xxx, 000-–000
FIGURE 5. (a) Schematic assembly of the Teflon flow thermocell. The two compartments are separated by a cellular ester membrane (Whatman)
and two layers of Nomex fabric. (b) Photograph of the flow thermocell setup. (c) Transitional VOC and jSC in a stagnant and flow cell. (d)
Performance of flow cell with MWNT buckypaper electrodes versus time. Each electrode was 1.2 × 0.5 cm. The nominal ∆T approximately
50 °C.

FIGURE 6. (a) Schematic assembly of the thin thermocell wrapped around a pipe carrying cool, flowing liquid. Two layers of Nomex are used
to separate the two electrodes. The distance between electrodes is approximately 0.2 cm. (b) Photograph of a thermocell wrapped around a
stainless steel pipe to generate power.

electrochemically active potassium ferri-ferrocyanide and these cells operating at a ∆T of 60 °C is estimated to be

MWNTs. The current cost per watt of active materials for $5.14/W, which is slightly higher than the cost per watt of

© XXXX American Chemical Society G DOI: 10.1021/nl903267n | Nano Lett. XXXX, xxx, 000-–000
photovoltaic modules.49 If current predictions are realized Supporting Information Available. CV scans of Pt, graph-
and the price of MWNTs decreases to $45/kg from today’s ite sheet, and MWNT buckypaper electrodes in Fe(CN)63-/4-
price of ∼$150/kg, we estimate that the material cost of aqueous electrolyte; U-shaped thermocell; Mark I thermocell
active materials for coinlike thermocells will reach $2.76/ configuration and performance; performance of Mark I
thermocells in series; performance versus electrolyte con-
W. Since the output power of these thermocells increases
centration; effect of interelectrode distance; detailed cost
quadratically with ∆T, development of electrolytes that can analysis. This material is available free of charge via the
operate to higher temperatures can substantially increase Internet at http://pubs.acs.org.
applications potential. Though the present continuously
operated thermal cells provided essentially constant output REFERENCES AND NOTES
for the maximum evaluated time period (90 days), opera- (1) Innovative Energy Systems Challenge; Funding Opportunity
Number DE-PS36-06GO96009; U.S. Department of Energy: Wash-
tional lifetimes of years or decades will be desirable for some
ington, DC, 10/20/2005.
applications and this might require improvements in the (2) The Future of Geothermal Energy; MIT-led interdisciplinary panel;
choice of electrolyte and redox mediator. Also, we are here Massachusetts Institute of Technology, 2006.
(3) Vining, C. B. Nat. Mater. 2009, 8, 83.
assuming that the waste energy is essentially free and (4) Mancini, T.; Heller, P.; Butler, B. J. Sol. Energy Eng. 2003, 125,
unlimited, since the relatively low conversion efficiencies of 135.
even the presently improved thermocells is a problem if cost- (5) Agar, J. N. In Advances in Electrochemistry and Electrochemical
Engineering; Delahay, P., Tobias, C. W., Eds.; J. Wiley & Sons:
effective technologies having higher efficiencies become New York, 1961; Vol. 3, Chapter 4.
available. (6) Haase, R. Thermodynamics of Irreversible Processes; Addison-
Wesley: Reading, MA, 1962.
In summary, we have shown that MWNT buckypaper and (7) Ratkje, S. K.; Ikeshoji, T. J. Electrochem. Soc. 1990, 137, 2088.
MWNT forest are efficient electrode materials for thermo- (8) Coggeshall, G. W. Z. Phys. Chem. 1895, 24, 62.
cells because they offer fast kinetics and large amounts of (9) Richards, T. W. Z. Phys. Chem. 1897, 24, 39.
(10) Eastman, E. D. J. Am. Chem. Soc. 1928, 50, 292.
electrochemically accessible internal surface area. The ef- (11) Quickenden, T. I.; Vernon, C. F. Solar Energy 1986, 36, 63.
ficiency of thermocells with MWNT electrodes was demon- (12) Ikeshoji, T. Bull. Chem. Soc. Jpn. 1987, 60, 1505.
strated to be as high as 1.4% relative to the Carnot Cycle, (13) Ikeshoji, T. F.; Nahui, N. B.; S. Kimura, S.; Yoneya, M. J. Electroa-
nal. Chem. 1991, 312, 43.
which is 3 times higher than the efficiency of conventional (14) Burrows, B. J. Electrochem. Soc. 1976, 123, 154.
thermocell devices with Pt electrodes. With improvements (15) Quickenden, T. I.; Mua, Y. J. Electrochem. Soc. 1995, 142, 3985.
(16) Mua, Y.; Quickenden, T. I. J. Electrochem. Soc. 1996, 143, 2558.
in cell design and optimization of MWNT properties and (17) Shindo, K.; Arakawa, M.; Hirai, T. J. Power Sources 1998, 70, 228.
electrode structure, thermocell efficiency is likely to increase. (18) deBethune, A. J.; Licht, T. S.; Swendeman, N. J. Electrochem. Soc.
Thin coinlike thermocells were fabricated and operated for 1959, 106, 616.
(19) Wang, J. Trends Anal. Chem. 2002, 21, 226.
three months to provide essentially constant power output. (20) Wang, J. Acc. Chem. Res. 2002, 35, 811.
In such configurations, direct synthesis of MWNT forest (21) Britto, P. J.; Santhanam, K. S. V.; Rubio, A.; Alonso, J. A.; Ajayan,
electrodes was shown to provide improved thermal contact P. M. Adv. Mater. 1999, 11, 154.
(22) Gooding, J. J. Electrochim. Acta 2005, 50, 3049.
that contributed to a 30% increase in efficiency as compared (23) Merkoci, A.; Pumera, M.; Llopis, X.; Perez, B.; del Valle, M.;
to buckypaper electrodes that required secondary attach- Alegret, S. Trends Anal. Chem. 2005, 24, 826.
(24) Gong, K.; Yan, Y.; Zhang, M.; Su, L.; Xiong, S.; Mao, L. Anal. Sci.
ment to the package substrates. The performance of MWNT- 2005, 21, 1383.
based thermocells was shown to be scalable and amenable (25) Rajesh, B.; Thampi, K. R.; Bonard, J. M.; Xanthopoulos, N.;
to complex systems. With the cost of MWNTs decreasing, Mathieu, H. J.; Viswanathan, B. J. Phys. Chem. B 2003, 107, 2701.
(26) Li, W. Z.; Liang, C. H.; Zhou, W. J.; Qiu, J. S.; Li, H. Q.; Sun, G. Q.;
thermocells with the performance reported here may de- Xin, Q. Carbon 2004, 42, 436.
velop into an economical solution for harvesting untapped (27) Matsumoto, T.; Komatsu, T.; Nakano, H.; Arai, K.; Nagashima,
supplies of low-grade heat. Moreover, the enhanced ther- Y.; Yoo, E.; Yamazaki, T.; Kijima, M.; Shimizu, H.; Takasawa, Y.;
Nakamura, J. Catal. Today 2004, 90, 277.
mocell performance demonstrated in this study using MWNT (28) He, Z. B.; Chen, J. H.; Liu, D. Y.; Tang, H.; Deng, W.; Kuang, W. F.
electrodes suggests that other nanostructured electrode Mater. Chem. Phys. 2004, 85, 396.
(29) Che, G.; Lakshmi, B. B.; Fisher, E. R.; Martin, C. R. Nature 1998,
materials might also be applied to significantly enhance the 393, 346.
efficiency of thermocell devices. (30) Frackowiak, E.; Beguin, F. Carbon 2001, 39, 937.
(31) Niu, C.; Sichel, E. K.; Hoch, R.; Roy, D.; Tennent, H. Appl. Phys.
Lett. 1997, 70, 1480.
Acknowledgment. The authors thank late Professor (32) Baughman, R. H.; Zakhidov, A. A.; de Heer, W. A. Science 2002,
Alan G. MacDiarmid for discussions of ferri-ferro cyanide 297, 787.
and other insights, Mei Zhang for synthesizing MWNTs, (33) Campbell, J. K.; Sun, L.; Crooks, R. M. J. Am. Chem. Soc. 1999,
121, 3779.
and Robert Haazer for additional tests of Seebeck coef- (34) Cai, H.; Cao, X.; Jiang, Y.; He, P.; Fang, Y. Anal. Bioanal. Chem.
ficient of ferri-ferro cyanide redox couple. We also thank 2003, 375, 287.
Office of Naval Research MURI Grant N00014-08-1-0654, (35) Zhang, N.; Xie, J.; Varadan, V. K. Smart Mater. Struct. 2006, 15,
123.
NSF Grant DMI-0609115, Robert A. Welch Foundation (36) Hupp, J. T.; Weaver, M. J. Inorg. Chem. 1984, 23, 3639.
Grant AT-0029, and the Australian Research Council for (37) Josserand, J.; Devaud, V.; Lagger, G.; Jensen, H.; Girault, H. H. J.
financial support. Electroanal. Chem. 2004, 565, 65.

© XXXX American Chemical Society H DOI: 10.1021/nl903267n | Nano Lett. XXXX, xxx, 000-–000
(38) Baughman, R. H.; Cui, C.; Zakhidov, A. A.; Iqbal, Z.; Barisci, (43) Zhang, M.; Atkinson, K. R.; Baughman, R. H. Science 2004, 306,
J. N.; Spinks, G. M.; Wallace, G. G.; Mazzoldi, A.; De Rossi, D.; 1358.
Rinzler, A. G.; Jaschinski, O.; Roth, S.; Kertesz, M. Science 1999, (44) Hall, L. J.; Coluci, V. R.; Galvão, D. S.; Kozlov, M. E.; Zhang, M.;
284, 1340. Dantas, S. O.; Baughman, R. H. Science 2008, 320, 504.
(39) Nugent, J. M.; Santhanam, K. S. V.; Rubio, A.; Ajayan, P. M. Nano (45) Schonenberger, C.; Bachtold, A.; Strunk, C.; Salvetat, J.-P.; Forro,
L. Appl. Phys. A: Mater. Sci. Process. 1999, 69, 283.
Lett. 2001, 1, 87.
(46) Morisaki, H.; Ono, H.; Yazawa, K. J. Electrochem. Soc. 1989, 136,
(40) Andrews, R.; Jacques, D.; Qian, D.; Rantell, T. Acc. Chem. Res.
1710.
2002, 35, 1008. (47) MacDiarmid, A. G.; Hall, N. F. J. Am. Chem. Soc. 1953, 75, 5204.
(41) Cola, B. A.; Xu, J.; Cheng, C.; Xu, X.; Hu, H.; Fisher, T. S. J. Appl. (48) MacDiarmid, A. G.; Hall, N. F. J. Am. Chem. Soc. 1954, 76, 4222.
Phys. 2007, 101, No. 054313. (49) As of Dec. 29, 2009, the solar module price in the United States
(42) Park, M.; Cola, B. A.; Siegmund, T.; Xu, J.; Maschmann, M. R.; market is $4.31 per Watt of peak power according to http://
Fisher, T. S.; Kim, H. Nanotechnology 2006, 17, 2294. solarbuzz.com/moduleprices.htm.

© XXXX American Chemical Society I DOI: 10.1021/nl903267n | Nano Lett. XXXX, xxx, 000-–000