Thermoelectric cloths using carbon nanotube yarn for wearable electronics

The need for energy harvesting technology as a power source for isolated small electronic devices is increasing. Especially in wearable applications, body heat is one of the promising energy sources, and therefore thermoelectric technology is attracting attention. For such applications, the ease of installation and the user’s comfortableness should be emphasized against coldness, stiffness, or stickiness. It is also essential to measure whether the required power can be generated at an acceptable cost without increasing or decreasing the naturally occurring heat flow. In this paper, we review the progress of thermoelectric cloths using carbon nanotube yarns, which have been studied by the authors with a consistent policy, including experimental and technical aspects, and propose a direction in which wearable thermoelectric generators should be developed. We hope this paper will also serve as a hint for those conducting similar research.


Introduction
There already exist many small, isolated electronic devices around our lives.The trend of increase in the number of electronic equipment scattered in various locations, such as healthcare devices and various IoT sensors, will be intensified in the future.Energy harvesting technology, which generates electricity from ambient energy at each use point, will become increasingly important. 1)he four major types of ambient energies available in indoor environments are light, radio waves, mechanical energy such as vibration and pressure, and heat (Table I).Among these, thermal energy is not small in density.Furthermore, it should be noted that thermal energy differs from others in that it does not have a minimum value of zero, especially when the human body is assumed.When used on the human body or living organisms, including animals, the ability to steadily generate power is a major advantage.In contrast to these advantages, the drawback is that the energy conversion efficiency is remarkably low compared to others.To compensate for this, a strategy of capturing a large amount of energy with large-area elements that emphasize ease of installation and "usability"-that is, the user will not feel cold, stiff, sticky, or otherwise uncomfortable-will be necessary.Besides, from the viewpoint of using waste energy, it will also be important to evaluate the materials and devices not on the energy conversion efficiency but on whether the necessary power can be generated without increasing or decreasing the steady-state heat flow at an allowable cost.Therefore, a strategy that goes beyond conventional wisdom, where efforts have been made to increase energy conversion efficiency, is required for thermoelectric materials/devices for energy harvesting.
In this paper, we leave the role of introducing related studies in the world, which may become obsolete within a few years after publication, to other review papers 2) and give an overview of our research on "thermoelectric cloths" which we have been studying with a coherent policy based on the background mentioned above.We would like to suggest one possible direction for the thermoelectric generators (TEGs) as a usable energy harvester for wearable electronics.We would also like to include some experimental and technical aspects so that those who are doing similar research will not be led astray by the detour we have taken.

Performance requirements for thermoelectric materials and devices
First, let us start with the general requirements for thermoelectric materials.The performance of thermoelectric materials is generally expressed as a dimensionless figure of merit: ( ) a s k = and the maximum expected energy conversion efficiency is calculated by: where α is the Seebeck coefficient (V•K −1 ), σ is the electrical conductivity (S•m −1 ), κ is the thermal conductivity (W•m −1 •K −1 ), T is the average temperature of the module (K), and T H and T L are the temperatures of the high-and lowtemperature sides of the module (K), respectively.Equation ( 2) is a monotonically increasing function with increasing ZT.Therefore, in the development of thermoelectric materials, the search for a material with a larger ZT is the right way.In other words, a material with a large Seebeck coefficient and electrical conductivity and a small thermal conductivity is an excellent thermoelectric material.Organic materials are inherently promising in terms of thermal conductivity.Heat transport in solids is carried out by the propagation of lattice vibrations (phonons), the rate of which is generally higher in materials with larger elastic moduli.In organic solids, lattices (or aggregates) are formed mainly by van der Waals interactions, resulting in a low elastic modulus and low group velocity of phonons propagating through the molecules.In addition, the skeletons of  PROGRESS REVIEW some organic molecules and proteins are highly flexible, and the intramolecular transmission of vibrations is extremely slow.In fact, the κ of most organic solids, regardless of whether they are small molecules or polymers, falls within the range of approximately 0.1-0.5 W•m −1 •K −1 .In contrast, silicon, a typical example of an inorganic semiconductor, has a κ of about 100 W•m −1 •K −1 , and even thermoelectric materials, which are selected as inorganic semiconductors with small κ, have a typical κ range of 1-10 W•m −1 •K −1 .In contrast, carbon nanotubes (CNTs), which are used as the primary material in this study, are known to have a longitudinal thermal conductivity exceeding 1000 W•m −1 •K −1 at the single-molecule level 3) and a high macroscopic thermal conductivity of several-ten to several-hundred W•m −1 •K −1 in the spun yarn state as shown later in this paper.Therefore, the issue is how to suppress their high thermal conductivity to increase ZT.
Furthermore, for flexible thermoelectric conversion devices used in wearable or living environments, a small κ is a more important factor than a large ZT value.In such applications, water-cooling mechanisms and large cooling fins cannot be used on the low-temperature side, and heat dissipation is solely dependent on natural air cooling from a flat surface.Because of this severe rate-limiting process for heat flow, generating sufficient temperature difference in the device is difficult.Figure 1 shows the results of calculations of conversion efficiency as a function of thermal conductivity and thickness of the module under actual wearable use, assuming natural air cooling from a flat module. 4)To fully utilize the capability of thermoelectric materials, a thickness of several millimeters is necessary, even with a thermal conductivity of about 0.1 W•m −1 •K −1 , which is a low value even among organic materials.
This calculation was performed assuming a constant skin temperature regardless of the thermal conductivity of the module.Still, the skin temperature should decrease in reality as the device induces more heat flow because the skin and underlying fat layer are just a thermal resistance with low thermal conductivities, of which values are 0.5-2.8 and 0.1-0.4W•m −1 •K −1 , respectively. 5)To ascertain the extent of this impact, we examined the difference in skin temperature when wearing a wristband, as a model of the thermoelectric cloth, and a commercially available thermoelectric smartwatch (Powerwatch, MATRIX) incorporating an inorganic thermoelectric module.The results are shown in Fig. 2, which indicates that the skin temperature of all three subjects was approximately 2 °C lower when wearing the thermoelectric smartwatch than when wearing the wristband.Therefore, it can be assumed that the high thermal conductance of the TEG causes a decrease in power generation efficiency beyond that calculated in Fig. 1 due to this decrease in the skin temperature.
The above considerations clearly indicate the need for a TEG with low thermal conductivity close to typical cloths, a thickness of a few millimeters, and high flexibility and stretchability suitable for wearable applications.The material that satisfies these conditions is nothing but "cloth."Therefore, the authors have been conducting research to comprehensively develop materials, device structures, and fabrication methods to realize thermoelectric cloths.

Material design for low thermal conductivity
3.1.Basic concept and its demonstration Metals or polymers are usually considered materials that can be easily processed into threads.However, metals are generally unsuitable for thermoelectric materials due to their low Seebeck coefficient and high thermal conductivity.Therefore, conductive polymers or CNTs, regarded as inorganic polymers, are promising options.Based on the results of our previous survey for organic materials, 6,7) the authors concluded that CNT composite materials might be promising.
CNT composites often exhibit relatively large Seebeck coefficients and extremely high power factors (PF = α 2 σ).As a means of reducing the thermal conductivity of such CNT composites and possibly increasing the Seebeck coefficient, one of the authors (MN) has devised the use of a structure where a core-shell molecule is bonded between adjacent CNTs, as shown in Fig. 3.Most of the phonon modes in a CNT have relatively high vibrational frequencies, reflecting  the rigid two-dimensional lattice of the sp 2 carbon.In contrast, if a molecule with a soft structure, such as an organic compound, as a shell is bridging two CNTs, it should be difficult to transmit vibrations through the molecular junction.On the other hand, if the shell is about 1 nm thick and the core is a semiconductor, electrons can be transmitted through the junction by the resonant tunneling effect.In other words, we expected the core-shell molecule to act as an ideal phonon-blocking and electron-tunneling junction for thermoelectric materials.Furthermore, if the junction effectively blocks heat transport, a local temperature difference will be created on both sides of the junction.This is expected to enhance the thermopower due to the Seebeck effect of the junction in series with the CNT as a current path.
The Seebeck coefficient when the junction determines the charge transport is expressed by putting the transmission function of the junction at the spectral conductivity in the general Seebeck coefficient theoretical equation, 9) as follows:  3) is larger than that in the denominator, indicating that the Seebeck coefficient of the junction increases as the band edge moves away from μ e , but the electrical conductance of the junction decreases, as in the inverse relationship observed in the bulk semiconductors.Therefore, to obtain the maximum ZT with the materials having such molecular junctions, it is essential to optimize the design of the electronic states at the junction as well as the design as a phonon blocker.
Based on the above considerations, it is desirable for the junction molecules to satisfy the following conditions: (1) the core should be inorganic particles that can be selected independently from the shell to freely change the electronic level and obtain sufficient transmittance in tunneling; (2) the shell should have a soft structure with the resonance wavenumber as low as possible to suppress thermal transport effectively; (3) the thickness of the shell should be about 1 nm at the junction to ensure a certain level of tunneling probability while suppressing thermal transport; and (4) the surface can be chemically modified to control adsorption probability on CNTs.While considering these conditions, the first author (MN) realized that a cage-shaped protein encapsulating inorganic particles, which had been studied by the other author (IY), would be optimal.
a)].Hereafter, the modified Dps with the ability to selectively adsorb on CNTs will be referred to as C-Dps.When we studied the charge transport properties of the monolayers of cage-shaped proteins having different sizes, we found that the smaller the size is and the more rigid the structure is, the higher the charge transport capability is, and that Dps has the highest charge transport capability among those we tested.13) Even without Figure 4(b) shows a schematic of the junction where two CNTs are bridged by C-Dps molecules encapsulating inorganic particles.This is a hybrid material that combines three different material fields -nanocarbon, protein, and inorganic nanoparticle-at the junction.For the first-stage demonstration, the C-Dps-adsorbed CNTs were purified by centrifugation, and a thin film of the CNT/C-Dps complex was formed from the aqueous dispersion by the casting method.8) It was, then, confirmed that heat conduction was suppressed to about 1/100 in the largest case compared to that of the CNT-only agglomerate without decreasing the PF.This dramatic decrease in thermal conductivity is undoubtedly due to the formation of molecular junctions as depicted in Fig. 4(b).However, such a decrease in thermal conductivity cannot be explained by a simple series connection of low thermal conductivity materials because the ratio of protein against the total heat flow path is small.Therefore, the thermal boundary resistance at the CNT/C-Dps interface must be important.Since the origin of the thermal boundary resistance at such a molecular junction has not been yet well understood, the progress of scientific study is desired.

Maximizing thermoelectric performance by carrier doping
Most thermoelectric materials are categorized into semiconductors, and the PF exhibits a peak at a certain carrier density because, in general, the Seebeck coefficient decreases but conductivity increases with carrier density.Therefore, carrier density control is important in thermoelectric materials to obtain the maximum power generation capability of the material.A detailed study on the dependence of the Seebeck coefficient and conductivity on carrier density in both metallic and semiconducting CNTs has been reported using FET structures which can modulate a wide range of the electron chemical potential. 14)Experimental results with purified semiconducting CNTs at RT are roughly the same as the already-known theoretical variations in three-dimensional semiconductors.
In the case of CNT-based TEGs, the cost of CNT raw materials accounts for a large proportion of the total cost of the device, so we use as-grown metal-semiconductor mixed CNTs to reduce the cost.Furthermore, since the diameter distribution and defect density of CNTs differ material by material, the dependence of Seebeck coefficient and conductivity on chemical potential cannot be uniquely determined.Understanding this uncertainty, the general tendency of the Seebeck coefficient and electrical conductivity of the mixed CNT against the electron chemical potential is depicted in Fig. 5.When the chemical potential is at the neutral point, the Seebeck coefficients of electrons and holes are balanced, resulting in a very small Seebeck coefficient and low conductivity.Shifting the chemical potential toward the VB (to the left in Fig. 5) increases the p-type (positive) Seebeck coefficient and shifting it toward the conduction band (to the right) increases the n-type (negative) Seebeck coefficient and conductivity.Further shifting of the chemical potential causes the Seebeck coefficient to peak at a certain point and then decrease again, while the conductivity continues to increase.The PF also peaks at slightly outside positions than the Seebeck coefficient does.In chemical doping, the range in which the chemical potential can be changed is often narrower than in FETs, so the highly doped regions where the Seebeck coefficient decreases at both ends of the graph may not be observed.
In general, p-type thermoelectric properties are obtained when thin films or yarns of CNT are fabricated under conditions of exposure to air or measured in air.This has been explained as the adsorbed oxygen molecules pinning the Fermi level of CNTs near the VB. 15)CNTs for thermoelectric applications, which are often used with various additives, may include other effects as well to determine the "undoped" state.As a result, "undoped" CNT thin films and yarns are often located near the arrows in Fig. 5, i.e. weak p-type.Therefore, n-type doping is more difficult because the p-type dopants must first be compensated to go into the n-type region and, even after that, the p-type doping effect automatically proceeds in the atmosphere.
Chemical doping methods for CNTs can be divided into two types.One is the charge transfer type [Fig.6(a)], in which neutral molecules with significantly different electronegativity from CNTs are added and they act as donors or acceptors by direct charge transfer from CNTs.This method is commonly used for organic semiconductors.In the other case, a salt consisting of a cation and an anion with unbalanced size and vapor pressure is added, one of which becomes neutralized through a certain process and evaporates into the surrounding gas phase or diffuses into the surrounding liquid phase, thereby disrupting the charge balance

PROGRESS REVIEW
of the salt adsorbed on the CNT, and the CNT becomes its counter ion.In this case, the charge transfer between the anion or cation to be neutralized and the CNT does not necessarily have to occur directly.
The advantage of the charge transfer type is that the doping can be carried out in the gas phase using sublimable small molecules as dopants, and for example, p-type and n-type regions can be pattern printed as in sublimation printing.For example, our group has confirmed that p-type thermoelectric properties can be optimized by gas-phase doping of CNT yarn with TCNQ and F 4 TCNQ. 16)There are many stable compounds which work as strong acceptors for p-type doping.However, finding strong and stable donors for ntype doping is relatively difficult.For example, alkali metals, which have been classically used for many molecular semiconductors, are very reactive and diffuse easily.Strong molecular donors are also unstable because they are extremely sensitive to the atmosphere due to their low ionization energy. 17)he advantage of the charge stabilization type is that various combinations of stable salts can be used, and salts composed of organic cations and halogen anions, which are commonplace materials, are particularly suitable for n-type doping.Another advantage is that a large amount of electrically neutral salts adsorbed on CNT thin films and yarns can prevent CNTs from direct exposure to the atmosphere, and when the doping effect is lost over time, the large amount of neutral salts can be expected to cause a spontaneous doping reaction again to suppress the decrease of doped carriers.On the other hand, disadvantages include the difficulty of gas-phase doping due to the difference in vapor pressure between anions and cations, and the possibility that some host materials, such as proteins in our case, may degenerate because the salts often contain highly reactive molecules or atoms.
Here, an example of liquid phase doping of CNT yarn with a quaternary ammonium salt as an n-type doping agent is presented. 18)CNT (eDIPS, EC1.5, Meijo Nano Carbon) was spun with polyoxyethylene stearyl ether (Emulgen 350, Kao Chemical) as a dispersing agent and heat-treated to a weak ntype.This can be a nice starting material for n-type doping because the p-type doping effect by air is suppressed.The ntype doping treatment was performed by immersing it in a dimethylformamide (DMF) solution (0.54 M) of tetrabutylammonium halide (TBAX, X is a halogen element) for 90 min in inert gas and then drying it.To investigate the influence of the anion halogen element, Cl, Br, and I (abbreviated as TBACl, TBABr, and TBAI, respectively) were compared.
The thermoelectric properties of the undoped and doped yarns are summarized in Fig. 7. TBAX doping hardly changes the Seebeck coefficient but increases the electrical conductivity and decreases the thermal conductivity.As a result, the ZT is about three times higher than that of the undoped yarn.The reason for the little change in the Seebeck coefficient is assumed to be that both samples are near the the equilibrium state is more biased to the right side for bromine than chlorine and iodine than bromine, and the more produced neutral halogen molecules on the right side of the equation diffuse into the solution or volatilize during drying and leave the CNT yarn, thereby stabilizing the excess TBA + cations and the electron-doped CNTs as counter anions.This result suggests an important guideline upon selecting a dopant agent for charge stabilization type carrier doping.
Although it has been confirmed that larger cations are more effective and atmospherically stable in n-type doping with salts, 19) the neutralizability of anions is also an important parameter.The stability of the TBAI-doped CNT yarn was also evaluated in the air without passivation coating, and the PF increased at first and then gradually decreased, reaching almost the same value after 35 d.The initial increase of PF may be due to the slow doping reaction during storage as predicted.Further study is needed to properly understand this phenomenon.
Note that this doping technique is tested for CNTs and is not guaranteed to work effectively for core semiconductors in C-Dps molecules, another important factor that determines thermoelectric performance.Even so, this result is valuable because the contribution of CNTs to the PF is significant even in CNT/C-Dps complex yarn.For tuning the electronic levels of the junction including the core's band structure, we believe that it is more effective to control the composition of core at the time of introducing the core elements into Dps, and this research is currently in progress.

Passivation with polymer coating
Even for n-type doped CNTs, which are considered unstable in air as mentioned above, stabilization for a month is possible by optimizing the doping agent.However, to ensure stable operation of the device for a more extended period in more severe atmospheres, including laundry, and to prevent direct contact of CNTs with skin and other surfaces, it is inevitable to coat the yarn with a tough thin film that has gas barrier function.Therefore, considering the affinity with protein molecules, we investigated a passivation coating by an aqueous solution process.
CNT raw material (e-DIPS, EC1.5, Meijo Nanocarbon) with sodium dodecylbenzenesulfonate (SDBS) added as a dispersing agent were used as weak p-type yarns, and those with polyoxyethylene stearyl ether (Emulgen 350, Kao Chemical) as a weak n-type yarn.A 10 wt% aqueous solution of modified polyvinyl alcohol (PVA) (GOHSENX Z-200, Mitsubishi Chemical) was used for coating, and (NH 4 ) 2 Zr(OH) 2 (CO 3 ) 2 (Zircosol AC-7, Daiichi Kigenso Kagaku Kogyo) was added as a cross-linking agent at 10% to the PVA.The PVA was cross-linked by immersing the spun yarn in this mixed aqueous solution for 1 min, followed by heat treatment at 70 °C for 10 min.
Figure 8 shows the dependence of the thermoelectric properties of p-and n-type CNT yarns on the measured atmosphere, whether in air or in a high vacuum.In particular, the uncoated n-type yarn shows a significant change in its thermoelectric property, with p-type properties in air and

PROGRESS REVIEW
n-type in vacuum, whereas the difference between p and ntype properties in air and vacuum was confirmed to be sufficiently small for the coated ones.
Next, as a more challenging test, we verified the stability of the CNT yarn with triphenylphosphine (TPP), which is an effective n-type dopant but tends to be unstable in air.The results are shown in Fig. 9.While the uncoated one loses the effect of dopant and becomes p-type after about 3 h of retention in air, the coated one shows stable n-type characteristics for at least seven days.From these results, it can be said that the cross-linked PVA coating has a sufficient passivation effect.The materials for passivation coatings have yet to be studied extensively enough and will need to be explored extensively for consumer use of thermoelectric cloths.

Highly accurate thermal conductivity measurement of yarn
We emphasize here that although accurate thermal conductivity measurement is important to correctly evaluate the performance of thermoelectric materials, it is not so easy when the samples are small in quantity.For example, the 3ω method, which we used to evaluate the thermal conductivity of yarns in the early stages of our study, has the advantage that it is easy to measure with a small amount of yarn samples and allows direct measurement of thermal conductivity using a common sample with Seebeck coefficient and electrical conductivity measurements.However, it has been found that this method possibly causes substantial measurement errors due to various factors, and what is worse is that it is difficult to judge the correctness of each measurement. 20)Therefore, in collaboration with several thermal conductivity measurement experts, a so-called round-robin test was conducted to measure the thermal conductivity of the same sample using various measurement methods without prior sharing of results, and it was determined that the DC heating T-type method and the spot periodic heating radiation thermometry method would be able to measure thermal conductivity and thermal diffusivity, respectively, with relatively high accuracy. 21)Here we have to consider that the value required in evaluating thermoelectric materials is the thermal conductivity.As shown in Table II, the density and specific heat capacity of CNT composite materials vary greatly depending on the fabrication conditions and composition.So, the density and specific heat capacity must also be accurately measured for each sample if the thermal diffusivity is measured.This is difficult to carry out when only a small amount of sample is available, which leads to errors in thermal conductivity.Therefore, the variation of the DC heating T-type method, namely a DC heating cross-junction method, is used to obtain highly accurate thermal conductivity in this study.This method is a modification of the DC heating T-type method to facilitate the mounting of thread-shaped samples.Figure 10 shows a photograph of the electrode substrate for the measurement and the sample fixture used to attach the sample to it.In the DC heating T-type method, the temperature dependence of the resistance of the Pt wire is measured beforehand so that the temperature of the Pt wire can be determined from its resistance.In the next step, the relationship between the power input and the average temperature of the Pt wire alone is measured, then the sample is bridged from the center of the Pt wire to the heat bath, and the same measurement is performed to calculate the thermal conductance of the sample from the decrease in the average temperature of the Pt wire due to the heat escaping through the sample.For a more detailed explanation of the principle and analytical equation, please refer to Ref. 21 and references therein.In the improved cross-junction method, the sample wires, bridging the upper and lower heat baths, contact the Pt wire at the midpoint of each other to easily attach the hard-tocut thin CNT yarn to the electrode substrate.The heat generated by the Pt wire is allowed to escape through the parallel connection of the two sample wires above and below the contact.Therefore, the sample thermal resistance, R s , in the analytical formula of the T-type method is replaced by the parallel connection of the upper and lower two sample thermal resistances, R s1 and R s2 , and calculated as follows: R R R To suppress the effect of heat loss due to residual gas in the chamber, 20) the measurement is performed in a high vacuum chamber, and the length/thickness ratio is adjusted so that the thermal conductance of the Pt wire is approximately equal to that of the sample to increase the measurement sensitivity, and the sample length must be increased for high thermal conductivity samples where the contact resistance between the Pt wire and the sample becomes a problem.Such careful measurements are expected to provide accurate thermal conductivity values with an accuracy of a few percent.

Device design for thermoelectric cloths
Flexible thermoelectric conversion elements are not immediately completed only by developing a high-performance CNT composite.In general, composites containing a high proportion of CNT are stiff and brittle, and cannot satisfy both a thickness of several millimeters and flexibility.][24] Even in such a case, it is challenging to make it follow various body movements like clothes.We have, therefore, proposed a method of making flexible TEGs by spinning CNT composite and sewing it into a cloth-like substrate of desired thickness which satisfies the requirements discussed in Sect. 2. 4) Figure 11 shows a schematic diagram of the wet spinning process we used.CNT dispersion is injected into a rotating vessel filled with methanol as condensate to form a gel-like yarn.CNT yarn with a diameter of 20-60 μm is fabricated by drying it while slowly pulling it up in the air.In the initial study, sodium dodecyl sulfate (SDS) was used as a dispersing agent, and polyethylene glycol (PEG) as a binder. 4)The CNT raw material used in this study was SW-CNTs produced by the eDIPS method and provided by the Saito Group of AIST, 25) and the dispersion method, which is described in detail later, was the simplest Method A.
Figure 12(a) shows the structure of the thermoelectric cloth fabricated for the first time using the CNT yarn.p/n stripe-doped CNT yarn was sewn into a felt with matching doping and sawing pitch, forming a series-connected structure of the π-type cells as shown in the cross-sectional view.The stripe-doping pattern was formed by wrapping the CNT yarn, which is originally p-type due to the effect of oxygen, around a small piece of plastic and applying the ionic liquid (IL), 1-butyl-3methylimidazolium hexafluorophosphate ([BMIM]PF 6 ), with 10 wt% dimethyl sulfoxide (DMSO) on one side as an n-type doping agent.The high viscosity of IL prevents the p/n boundary from becoming ambiguous as a dopant dissolved in solution penetrates through the CNT yarn for a long distance.10.Photograph showing the electrode-substrate for the DC heating cross-junction method and the sample fixture used to fix the Pt and sample wires to the proper position.The height of the two heat baths for the sample wire is designed to be lower than the electrodes for the Pt wire to maintain pressure for a better sample-Pt wire contact.This electrode-substrate can also be used for the thermopower measurement when a sample wire is set to the position of the Pt wire.Fig. 11.Fabrication of a CNT yarn by wet spinning method. 4)y this method, the doping pitch can be freely adjusted by the width of the plastic peace and can be matched to the thickness of the substrate cloth.
Figure 13(a) shows the results of the bending test for the fabricated thermoelectric cloth.Because the inner radius when bent was not measurable, this is not exactly a bending test but rather a folding test.The folding operation was repeated 160 times, and the change in the resistance of yarn was less than 2%.This is a sufficiently high bending resistance compared to the results of other flexible thermoelectric devices 26,29) that had been reported until the paper's publication.This high bending resistance was achieved because the active material, CNT yarn, is not firmly fixed to the substrate, making the active material less susceptible to bending stress.This is a significant advantage for wearable applications.Depending on the sewing method, it is possible to make the CNT spun yarn stress-free against expansion and compression of the substrate fabric, making it resistant to the same usage as ordinary clothing.
The prototype device was tested by lightly touching one side with a finger, and the other side was naturally cooled by stationary air (24 °C), as shown in Fig. 13(b).Kapton film was inserted between the finger and module to prevent electricity leakage by fingers.Thermoelectromotive force was generated at the moment of touch, and after 4 s, a stable voltage of approximately 2.3 mV was obtained.From this output voltage, the temperature difference generated in the module was estimated to be approximately 5 K. On the other hand, a control experiment using a commercially available TEG (Bi 2 Te 3 , 31 cells, TEC1-03104, Thermal Electronics) resulted in a temperature difference of only about 0.6 K.This difference indicates the importance of using TEGs with low thermal conductivity for wearable applications.
The device structure in Fig. 12(a) is easy to fabricate when the stitch pitch is large, but when the pitch is small, there is a problem of misalignment in the p/n boundary structure and the problem that a certain width of low conductivity region exists at the p/n boundary, increasing internal resistance that does not contribute to power generation.Furthermore, the disconnection of yarn is a significant issue for wearable devices.Since many π-type cells are connected in series, a single-point disconnection will make all the cells in the series connection unused for power generation.Therefore, a new device structure was proposed: after fabricating p-type and n-type yarns separately, they are alternately sewn into the substrate while weaving adjacent yarns together, forming a mesh structure as shown in Fig. 12(b). 28,30)ince the current can be easily rerouted by the π-type elements in a series and parallel mesh structure, the drop of output power by the breakage of yarn should be less.To investigate the disconnection resistance of the new device structure, a Monte Carlo simulation using LTspice 31) was performed with a smallscale device structure and an equivalent circuit, as shown in Fig. 14(a).The module's output power was calculated by randomly disconnecting each thread connecting the nodal points with a probability of 0%-50%, and the mean and standard deviation are shown in Fig. 14(b).One can see that the expected output power exceeds 40% of the initial value even if the threads are cut at a probability as high as 30%.and (b) oblique projection of mesh-type braided structure. 27,28)) (b) The Japan Society of Applied Physics by IOP Publishing Ltd PROGRESS REVIEW

Integration of the material design and the device design
The results presented in the previous section were obtained using CNT/polymer composite for ease of device prototyping.We must, therefore, integrate the established device design with the protein molecular junction.To stably spin CNT/C-Dps complexes, we have tested various dispersion methods, and part of the try-and-error results were summarized in a paper. 30)In this experiment, single-wall CNTs (Tuball, OCSiAl) were used as raw material, and four dispersion methods shown in Fig. 15 were compared.Details are given below: Method A: 3 mg•ml −1 of CNT/DI water solution was prepared, with the addition of buffer solution adjusted to pH 6.0 using sodium dihydrogen phosphate dihydrate and disodium hydrogen phosphate 12-water.Then, C-Dps solution was added to make 12 mg•ml −1 concentration, followed by ultrasonication for 3 min.Finally, the CNT/C-Dps complex was collected by membrane filtration (cellulose acetate, 200 nm).
Method B: 9 mg of CNT and 4 ml of IL, [BMIM]PF 6 , were mixed using agate mortar for 1 h.Then, the CNT/IL mixture was transferred to a beaker with methanol 500 ml and DI water 500 ml for the dilution of the IL.Finally, the collected CNT dispersed again with biomolecules in the same manner as Method A.
Method C: 9 mg of CNT and 4 ml of IL were mixed using agate mortar for 1 h.Then, the CNT/ IL mixture was transferred to the dialysis tube (cellulose) in a methanol bath.Finally, the collected CNT dispersed again with biomolecules in the same manner as Method A.
Method D: 4 ml of IL with surfactant (2.8 wt%) was stirred for 1 h at 500 rpm.Then 10 mg of CNT was added for the mixing process in the agate mortar for 1 h.After the addition of 4 ml DI water, the solution was stirred and centrifuged for 10 min at 15 000 rpm. 5 ml of surfactant solution (in DI water, 3 wt%) was added and shaken for 1 min, then it was centrifuged again.The addition of surfactant solution and centrifugation were repeated three times.Afterward, the precipitated CNT with 20 ml of methanol was transferred to the dialysis tube in the methanol bath.Finally, the CNT dispersed again with biomolecules in the same manner as Method A.
The dispersions prepared by these four methods were dropped onto a TEM grid, and the morphology of the CNT/ C-Dps complex was observed.CNT/C-Dps complexes were formed in all cases.Among the four methods, Method D had the highest coverage of C-Dps and the smallest average diameter of CNT bundles (i.e.CNTs were better dispersed).Figure 16 shows the TEM photograph of the CNT/C-Dps complex prepared by Method D. Here, the C-Dps does not contain the core, but the Dps shell appears white due to the phosphotungstic acid staining.Four CNT bundles with an average diameter of 4 nm are covered with C-Dps molecules in this image, of which area coverage is approximately 30%.If C-Dps is adsorbed at this level of coverage, C-Dps molecular junctions will likely be formed with a high probability between adjacent bundles in the yarn.
In Fig. 17, we compare the thermoelectric properties of CNT/C-Dps-complex yarns fabricated with the four dispersion methods.While Method A, used in the early stage of this study, had difficulty in continuously spinning, CNT/C-Dpscomplex yarns were successfully spun with Methods B-D.In addition to CNT and C-Dps, the other components in the spun yarn were ILs or surfactants used for dispersion, both of which have been confirmed to act as weak n-type dopants in previous experiments.However, since positive Seebeck coefficients were obtained in all samples, it is assumed that C-Dps either acts as a weak p-type dopant or has the effect of eliminating ILs and surfactants.It is also speculated that the C-Dps without cores are mainly hole conductive, which may enhance the p-type Seebeck effect.Comparing the three successful cases in spinning, the Seebeck coefficient and electrical conductivity are the highest, and thermal conductivity is the lowest in Method D, resulting in the highest ZT.Now, the CNT/C-Dps complex yarn can be stably fabricated by improving the dispersion method.Let us, then, check whether the effect of reducing thermal conductivity, which was confirmed in the direction of the film thickness in

PROGRESS REVIEW
the initial stage of the research, is also obtained in the longitudinal direction of the yarn.Figure 18 compares the thermal conductivity of yarns fabricated from CNT dispersions without and with C-Dps prepared by Methods B-D.As described in the first half of this paper, CNTs themselves are high thermal conductivity materials, and the high thermal conductivity of CNT is utilized in the yarn due to their uniaxial orientation, resulting in a maximum thermal conductivity of approximately 180 W•m −1 •K −1 .Such a high value is unsuitable for thermoelectric applications.In contrast, it was confirmed that the thermal conductivity was reduced to about 1/8 by forming a C-Dps molecular junction.Although it is still far from the requirements shown in Fig. 1, it is essential to note that the thermal conductivity of the actual device is diluted by the percentage of CNT yarn in the device area (which is determined by the required power generation and allowable cost, but is roughly estimated to be about 10%), Considering that the heat dissipation rate can be increased by taking advantage of the high emissivity and high specific surface area of CNT yarn, the thermal conductivity obtained with CNT/C-Dps complex will be sufficiently practical even at the present state.
Finally, we would like to show a picture showing the demonstration of the power generation by the thermoelectric cloth with CNT/C-Dps-complex yarn without touching the device surface in Fig. 19.The output voltage indicated on the

PROGRESS REVIEW
smartphone is just rising by the heat radiation from the hand with a latex glove.The fact that CNTs efficiently absorb IR radiation from heat sources 32) is thought to work as an advantage for such non-contact power generation.

Conclusions
In this paper, we explained the importance of a certain degree of thickness and low thermal conductivity in flexible TEGs aiming at wearable applications and energy harvesters for indoor environments to generate sufficient power even under conditions of insufficient thermal coupling.We also proposed that thermoelectric cloths are ideal as TEGs that can fit any three-dimensional shape and have elasticity to follow the deformation of the heat source.Some of the elemental technologies from the research results conducted by the authors over many years were explained, which may guide the readers to understand what is necessary for the total technology to realize such an ideal thermoelectric cloth.
The flexible and stretchable TEGs with yarn as the functional unit would be suitable not only for the CNT composites presented here but also for other nanocarbon composites and conductive polymers.We hope this article will provide hints for the increasing research on thermoelectric materials/devices as wearable energy harvesters.

Fig. 1 .
Fig.1.Dependence of the conversion efficiency on the thickness and thermal conductivity of the thermoelectric module having constant ZT.4) Natural air cooling from the flat surface of the module is assumed and the energy conversion efficiency is normalized by the maximum value.

Fig. 2 .
Fig. 2. Comparison of skin temperatures when wearing a cotton wristband that mimics a thermoelectric cloth and a commercial thermoelectric smartwatch: (a) the wristband and (b) the thermoelectric smartwatch used in the experiment.(c) Skin temperatures of three subjects measured for 10 min.The subjects were three males in their 20 s, and the solid, dotted, and singledotted lines are the results of the same individuals.
where e is the elementary charge, μ e is the chemical potential of the electrons, D CNT is the density of states function of the CNT, τ j is the junction transmission function, and f FD is the Fermi-Dirac function.From this, if the bottom of the core's conduction band is closer to μ e , as shown in Fig.3, the Seebeck coefficient becomes n-type, and if the top of the VB is closer, the Seebeck coefficient becomes p-type.The exponent of e e m in the numerator of Eq. (

Fig. 3 .Fig. 4 . 4 ©
Fig. 3. Proposed idea of the molecular junction with the function of suppressing heat transport and promoting thermoelectric effect: 8) (top) A schematic diagram of the junction structure, (middle) an energy band diagram of the junction, and (bottom) the temperature distribution in the junction.

Fig. 5 .
Fig. 5. Schematic graph showing the change in electrical conductivity and Seebeck coefficient of (thin dotted line) metallic, (thin solid line) semiconducting, and (thick solid line) mixed CNT aggregates when the chemical potential of electrons is modulated by doping.

Fig. 6 . 6 ©
Fig. 6.Two types of typical chemical doping mechanisms: (a) "charge transfer" between host and dopant by adding neutral dopant molecules having sufficiently different electronegativity and (b) "charge stabilization" by adding unbalanced salt as a guest resulting in the counter-ionization of the conducting host to maintain charge neutrality.

Fig. 7 .
Fig. 7. Thermoelectric properties of pristine and doped CNT yarns.17)The error bar indicates the standard deviation of the measured values for three samples.

17
Fig. 7. Thermoelectric properties of pristine and doped CNT yarns.17)The error bar indicates the standard deviation of the measured values for three samples.

7 ©
Fig. 7. Thermoelectric properties of pristine and doped CNT yarns.17)The error bar indicates the standard deviation of the measured values for three samples.

Fig. 8 .
Fig. 8. Atmosphere dependence of thermoelectric properties in (a) p-type and (b) n-type CNT yarns.

Fig. 9 .
Fig. 9. Variation of the Seebeck coefficient and conductivity of CNT yarn over storage time in the air: (red) with and (black) without polymer coating.

Fig.
Fig.10.Photograph showing the electrode-substrate for the DC heating cross-junction method and the sample fixture used to fix the Pt and sample wires to the proper position.The height of the two heat baths for the sample wire is designed to be lower than the electrodes for the Pt wire to maintain pressure for a better sample-Pt wire contact.This electrode-substrate can also be used for the thermopower measurement when a sample wire is set to the position of the Pt wire.

Fig. 12 .
Fig. 12. Schematic illustration of thermoelectric cloth fabricated by sewing CNT yarn into fabric: (a) cross-sectional view of a series-type structure by straight stitching4,26) and (b) oblique projection of mesh-type braided structure.27,28)

Fig. 13 . 10 ©
Fig. 13.(a) Folding test of a thermoelectric cloth using CNT yarn, and (b) demonstration of power generation by body heat.4)

Fig. 14 .
Fig.14.Disconnection resistance simulation results of a thermoelectric cloth with mesh-type braided structure: (a) the device structure tested and the equivalent circuit for a mesh, and (b) the variation of output power with disconnection probability.

Fig. 15 .
Fig.15.Flow chart showing the difference between dispersion methods of CNT/C-Dps complex, and materials used in this work.28)

Fig. 16 .Fig. 17 . 12 ©
Fig.16.TEM image of the CNT/C-Dps complex in the dispersion prepared with Method D.28) Four CNT bundles are covered with C-Dps molecules appearing white dots, of which area coverage is approximately 30%.

Table I .
Major environmental energies available for indoor use.

Table II .
Statistics of the density and specific heat capacity values of CNT thin films and yarns without protein junctions measured in our group (87 samples for density and 37 samples for specific heat).
©2023The Author(s).Published on behalf of The Japan Society of Applied Physics by IOP Publishing Ltd