Macroscopic weavable fibers of carbon nanotubes with giant thermoelectric power factor

Low-dimensional materials have recently attracted much interest as thermoelectric materials because of their charge carrier confinement leading to thermoelectric performance enhancement. Carbon nanotubes are promising candidates because of their one-dimensionality in addition to their unique advantages such as flexibility and light weight. However, preserving the large power factor of individual carbon nanotubes in macroscopic assemblies has been challenging, primarily due to poor sample morphology and a lack of proper Fermi energy tuning. Here, we report an ultrahigh value of power factor (14 ± 5 mW m−1 K−2) for macroscopic weavable fibers of aligned carbon nanotubes with ultrahigh electrical and thermal conductivity. The observed giant power factor originates from the ultrahigh electrical conductivity achieved through excellent sample morphology, combined with an enhanced Seebeck coefficient through Fermi energy tuning. We fabricate a textile thermoelectric generator based on these carbon nanotube fibers, which demonstrates high thermoelectric performance, weavability, and scalability. The giant power factor we observe make these fibers strong candidates for the emerging field of thermoelectric active cooling, which requires a large thermoelectric power factor and a large thermal conductivity at the same time.

1. The author should have used various chirality carbon nanotubes, which means that their band gaps are different. It is unclear how to tune the Fermi levels of all the carbon nanotubes whose electronic band structures are dissimilar. Furthermore, it is necessary to state how to control electronic transport properties at the contact between carbon nanotubes. 2. One of the main reasons for the high power factor is the extremely high electrical conductivity. I recommend measuring it in another lab, if possible, to confirm the result. 3. The Seebeck coefficient is strongly dependent on the contact. For example, the contact may create Schottky barriers, which will affect the measurement results. It is recommended to use different contact materials in the measurements to see if the result is reliable. 4. It would be useful to provide more information such as SEM, XRD, and TEM images regarding the CNT fiber after the annealing processes. 5. It is necessary to provide more quantitative descriptions about the reasons for changes of the Seebeck coefficient and electrical conductivity of CNT fibers due to dedoping process at the annealing temperature of 350 K and 500 K. 6. It is necessary to provide more information regarding the thermoelectric properties of the thermoelectric generator (TEG). Make sure the flexibility by measuring the resistance and output performance of the TEG when they are bent. In addition, it is necessary to provide the charging time of the capacitor to turn on the LED bulb as well as the stability and the reproducibility of test results of the TEG. We are grateful to the reviewer for carefully reading our manuscript and making detailed comments. In the following, we respond to each of his/her comments in detail:

Major Comment:
Organic thermoelectric materials are of great interest in the fields of waste heat recovery and active cooling. Carbon nanotubes have unique mechanical and electrical properties, which have been widely studied for thermoelectric applications. The authors reported that the carbon nanotube fibers exhibited giant thermoelectric power factor, which was due to the one-dimensional quantum confinement of charge carriers (shifting the Fermi energy close to the van Hove singularity in the electronic density of states). The power factor obtained in this paper is impressive. But after a careful reading, I am not convinced by the argument approaching van Hove singularity leads to the high power factor.
Response to comment the major comment: We thank the reviewer for this insightful comment. We did not intend to argue that the high power factor solely arose from approaching a van Hove singularity in the electronic density of states, but we believe that having an ultrahigh electrical conductivity is also important. However, we agree with the reviewer that it was not  Figure 17b), suggesting that the ultrahigh conductivity was key to the achieved giant PF. Although σ for individual metallic SWCNTs has been reported to be ~100 MSm -135, 36 , σ for CNT assemblies used in TE studies has been lower than 1 MSm -1 . In this study, the excellent morphology of the aligned CNT fibers led to the well-preserved σ of >10 MSm -1 even in a macroscopic assembly, resulting in the ultrahigh value of PF. PF values for CNT assemblies can be further enhanced in the future by improving the sample morphology, better preserving the σ of individual CNTs.
In addition to improving σ, tuning EF is also crucial for maximizing PF.
Supplementary Figures 17c and d show not only the maximum values but also other experimental values from EF tuning measurements for this study and for samples from Ref. Supplementary Figure 17d, within the same sample, increasing σ does not always result in improving PF; in this study, increasing σ decreased both S and PF. Furthermore, the decrease or increase of S and PF with respect to σ is not monotonic; S as well as PF show a peak when EF is near a VHS in the DOS, as demonstrated in Figure 2.

As shown in
Therefore, it is important to understand the EF dependence of TE properties, and tune EF accordingly to maximize PF.
These results highlight the promising properties of CNTs as a TE material.
Conventionally, materials with high σ were considered undesirable for TE device applications because of the well-known trade-off between S and σ. However, as shown in this study, CNTs can provide relatively large S if EF is tuned to the vicinity of a VHS in the DOS, in spite of having an ultrahigh σ. This is because S can be enhanced through a narrow carrier distribution 13 , which is achieved by 1D quantum confinement of charge carriers in the case of CNTs.

Comment #1:
The authors claim that the carbon nanotube fibers could be centimeter-long. However, it is hard to understand why the length of the carbon nanotube fiber need to be emphasized.
Carbon nanotube fibers or yarns at meters or tens of meters scale have been often reported.
Response to comment #1: The reviewer is absolutely correct that CNT fibers and yarns with meters and tens of meters of length is typical. The CNT fiber production method used in this study can produce hundreds of meters of continuous fiber in our lab at Rice and considerably longer fibers in a startup from Rice (DexMat). In an earlier version of our manuscript, we had used the adjective "large-scale" to describe these fibers, but a few co-authors felt that such a term was too generic. We thus switched to "centimeter-scale" because we used pieces of fibers that were a few centimeters long in the current thermoelectric study, which should be contrasted to earlier reports on high PF values that used microscale materials. In fact, 2D materials used in prior thermoelectric studies such as graphene, which achieved the highest reported PF at room temperature, were only micrometer-scale, not desirable for practical large-scale applications. On the other hand, we observed a giant PF in a truly macroscopic sample, and thus, we wanted to emphasize it as one of the strengths of our sample. From a materials production viewpoint, continuous fibers were made with lengths >100 m and then cut into centimeter-scale pieces for the TE measurements and thermoelectric generator demonstration. However, we now realize that readers with a background in material production might get confused, and we thank the reviewer for this comment. Therefore, we have changed the expression "centimeter-scale" to "macroscopic" in the title, and added a few sentences to clarify this point in Page 8, line

129-135, as shown below.
Note that the giant PF in this work was observed in macroscopic samples (~1 cm length), whereas the highest PF reported for 2D materials has been measured in microscopic samples (typically with dimensions of a few μm). CNT fibers were produced in continuous runs (over 100 m in total length, as discussed above) and cut into centimeter-scale pieces to conduct measurements. We further demonstrate this strength, i.e., scalability, in a later section by using CNT threads that were produced by plying multiple fibers via a continuous plying machine. mWm-1K-2, respectively. These values are all over two times higher than the PF of carbon nanotube fibers obtained in this paper. "comparable" may not be an accurate word to describe the large differences between them.

We have also added Supplementary
Response to comment #3: We have changed the adjective "comparable" to "approaching"  Figure 1). The viscosity-averaged aspect ratio was 6.7 (± 0.1) x 10 3 , measured using a capillary thinning extensional rheometer 24 . A solution spinning method 15,16 was used to spin CNTs into a continuous fiber. CNTs were first dissolved in chlorosulfonic acid (CSA) to create a spin dope 25 . The dope was then filtered and extruded into a coagulant. Finally, the coagulated fiber was collected onto a rotating drum. This method produced meters (>100 m) of fiber with densely packed and highly aligned CNTs, as shown in Figure 1a. The average diameter of the fiber was determined to be 8. CNTs inside the solution spun fiber have a high aspect ratio as well as a low impurity density, and are highly crystalline, leading to exceptional mechanical (a tensile strength of 4.2±0.2 GPa) and electrical (σ > 10 MSm -1 ) properties while retaining flexibility 15,17 .

Comment #4:
S1 and S2 were used to represent both the equations and the CNTs, which makes it hard to understand what the authors were trying to say in the supporting information.

Comment #5:
The thermoelectric properties of DWCNT can be directly calculated with ATK. Why the authors use a simplified model to approximate the thermoelectric properties of DWCNT?
Response to comment #5: DWCNTs generally have an incommensurate crystal structure (except rare cases), which leads to a quasi-periodic potential energy landscape without "weavablity" should be changed with "weavability". Please check the typos.
Response to comment #7: We thank the reviewer for pointing out this typo. It has been corrected.
We first estimated the EF of four chemically treated fibers discussed in Figure  - We are grateful to the reviewer for carefully reading our manuscript and making detailed comments. In the following, we respond to each of his/her comments in detail: The DWCNTs used in this study were not chirality-sorted, but they had a diameter distribution; the average outer-(inner-) wall diameter was 1.8±0.2 nm (0.9±0.1 nm), determined by high resolution TEM analysis. The inner-wall diameter distribution was confirmed by an absorbance spectrum for the raw CNT material (Supplementary Figure 6), and the outer-wall diameter distribution is shown in Figure   1c of Ref. 6 , where the same raw CNT material was used. In short, a diameter histogram for our sample would show two distinct peaks, at around 0.9 nm and 1.8 nm, instead of a single-peak distribution typical of a SWCNT ensemble. Because the bandgap energy of a CNT is determined by the diameter 5 , a narrow distribution in diameter results in a narrow distribution in bandgap. This assumption was further confirmed by the optical study ( Supplementary Figure 7a), where the absorbance spectra can be interpreted as a superposition of spectra for semiconducting and metallic SWCNTs for outer wall and inner wall.
2. Electronic transport in assemblies of CNTs occurs through two distinct mechanisms: intertube transport (transport between CNTs) and intratube transport (for which the conductivity is usually limited by electron-phonon scattering at high temperatures) 7 .
One way to determine which of these two mechanisms is dominant under given conditions is via a study of the temperature (T) dependence of σ. "Semiconducting" behavior (σ rising with increasing T) can be attributed to intertube transport, whereas "metallic" behavior (σ decreasing with increasing T) can be attributed to intratube transport. In typical macroscopic CNT samples, the T dependence of σ shows semiconducting behavior up to room temperature 8 . However, as shown in Figure 4a of Ref. 7 , the excellent sample morphology of our CNT fibers (a high degree of CNT alignment, a high density, a high CNT aspect ratio, and a low density of impurities) led to the metallic behavior above 200 K. Therefore, at room temperature (300 K), where our measurements were conducted, our data is not significantly affected by intertube transport.

Comment #2:
One of the main reasons for the high power factor is the extremely high electrical conductivity. I recommend measuring it in another lab, if possible, to confirm the result.
Response to comment #2: We measured the electrical conductivity of the as-produced fiber (reported as 11±2 MS/m in the manuscript) in three different labs to confirm the reproducibility.
The results are summarized in Table R1. The as-produced continuous fiber (> 100 m) was cut into pieces and distributed among three labs. All of them used a four-probe method so that the contact resistance was negligible, and the measurements were conducted at room temperature.
As shown in Table R1, all three labs confirmed the reported value in the manuscript. The Seebeck coefficient is strongly dependent on the contact. For example, the contact may create Schottky barriers, which will affect the measurement results. It is recommended to use different contact materials in the measurements to see if the result is reliable.

Response to comment #3:
We thank the reviewer for this thoughtful recommendation. In order to address this concern, we have conducted additional measurements as follows. We measured the Seebeck coefficient of the as-produced CNT fiber using different contact materials. This CNT fiber was identical to the "as produced" sample in the paper. Here we checked the Seebeck coefficients of the fiber using three different contacts: silver paste, carbon paste, and gold paste.
The device structure and the measurement method were the same as described in the Methods section. The results are summarized in Table R2. As shown in Table R2, the Seebeck coefficients were the same within the experimental errors, and thus, we conclude that any contact effect is negligible in this study.  It would be useful to provide more information such as SEM, XRD, and TEM images regarding the CNT fiber after the annealing processes.
Response to comment #4: We thank the reviewer for this suggestion. To address this question, we have added SEM and Raman characterization data for the CNT fibers after the annealing process to the Supplementary Information (Supplementary Figure 3).
CNT fibers that we produce using the solution spinning method typically have diameters of 10 μm and cannot be imaged via TEM. Therefore, we use SEM to characterize diameter and surface morphology. We conduct TEM only on raw CNT materials, before spinning them into fibers, which can be seen in Supplementary Figure 1.  It is necessary to provide more quantitative descriptions about the reasons for changes of the Seebeck coefficient and electrical conductivity of CNT fibers due to dedoping process at the annealing temperature of 350 K and 500 K.
Response to comment #5: The as-produced fiber from the solution spinning method is heavily doped by chlorosulfonic acid. Annealing the as-produced fiber desorbs the chlorosulfonic acid with an activation energy that is proportional to B , where B is the Boltzmann constant and is the temperature 10 . Therefore, annealing at a higher temperature leads to desorbing more dopant, leading to stronger dedoping (moving the Fermi energy closer to the charge neutrality point). We chose 350 °C as the first annealing temperature to ensure the annealing temperature is well above the boiling point of chlorosulfonic acid (~152 °C).
Quantitative determination of the Fermi energy shift difference between the sample annealed at 350 °C and the one annealed at 500 °C is more challenging. In this study, it was limited to a quantitative estimation based on the optical absorption measurements (Supplementary Figure 7). Furthermore, because the electrical conductivity monotonically increases with doping in carbon nanotubes, the decrease of the electrical conductivity after annealing is usually explained as a dedoping effect.
Another way to estimate the Fermi energy shift is to use Raman spectroscopy 11 . Figure   R1 shows

Comment #6:
It is necessary to provide more information regarding the thermoelectric properties of the thermoelectric generator (TEG). Make sure the flexibility by measuring the resistance and output performance of the TEG when they are bent. In addition, it is necessary to provide the charging time of the capacitor to turn on the LED bulb as well as the stability and the reproducibility of test results of the TEG.

Response to comment #6:
We thank the reviewer for these insightful suggestions. We have conducted a series of additional measurements to address the reviewer's points: 1. Flexibility. In order to ensure the performance of the thermoelectric generator (TEG) device while they are bent, we bent one TEG unit (consisting of fifteen CNT threads) with a bending radius of 3.18 mm (see Supplementary Error! Reference source not found.c and d), and applied a temperature difference while maintaining the bending.
Supplementary Error! Reference source not found.e shows a generated voltage as a function of the temperature difference without bending (black) and with bending (red).
The slope of the two fitting curves differs only by 3.8 %, verifying that there is essentially no degradation in device performance due to bending. This information has been added to the Supplementary Information.  3. Reproducibility. The demonstration of lighting an LED by four TEG units was reproduced multiple times. We will attach three videos of the demonstration filmed on 09/23/2020, 10/05/2020, and 05/11/2021 (entitled as "IMG_1250", "IMG_1364", and "IMG_2371", respectively). We did not edit the videos so that the reviewer can check the original data of them easily, thus so please do not bother background voices. Figures   R3a, b, and c show the LED that was driven by four TEG units on 09/23/2020, 10/05/2020, and 05/11/2021, respectively. Note that the TEG successfully turned on the LED after more than half year on 05/11/2021, demonstrating indisputable reproducibility and stability. We investigated the flexibility of this device by performing bending tests. The device was wound around cylinders with specific diameters, as shown in the inset of Figure   3c, and the electrical conductivity (σ) at each bending radius was compared to that of the original state without bending (σ0). Figure 3c shows that no significant change (less than 2 %) occurred up to a bending radius of 0.1 mm. We repeated the bending for 200 times with a fixed bending radius of 0.1 mm. As shown in Figure 3d, the conductivity did not change more than 1.9%. Furthermore, we applied a temperature difference to the device while it was bent. Supplementary Figure 18e shows a generated voltage as a function of the temperature difference during bending, verifying that there was essentially no degradation in device performance due to bending. We are grateful to the reviewer for carefully reading our revised manuscript and making valuable comments. In the following, we respond to each of his/her comments in detail:

Comment #1:
The Seebeck coefficient is pretty large despite the extremely high electrical conductivity. The van Hove singularity of CNTs whose chirality is different are dissimilar so it is not easy to improve the Seebeck coefficient by changing the doping level. It is challenging to uniformly control the doping level of the different chirality CNTs, which would influence the Seebeck and electrical conductivity. It is necessary to address these issues.

Response to comment #1:
We thank the reviewer for making this helpful comment. This comment actually deals with two separate issues, which we individually address below:  Figure   R1a, the Seebeck coefficient (Sind) becomes maximum for (22,0) when EF = -0.05 eV, while (13,13)'s maximum occurs at EF = -0.57 eV. However, we can model our sample as an assembly of nanotubes that are electrically connected in parallel. In this combined system, as shown in Figure R1b, Stot becomes maximum when EF = -0.16 eV, which is distinct from both -0.05 eV and -0.57 eV. We emphasize that our model is based on what has previously been used successfully for simulating CNT samples containing multiple chiralities. 1-3 This comment helped us realize that we did not clarify the difference between the combined model and models based on individual chiralities in our previous manuscript.
Hence, we have added subscripts "ind" to thermoelectric quantities (e.g., Sind) obtained from the models based on individual chiralities, and subscripts "tot" (e.g., Stot) to quantities obtained from the combined model. Furthermore, we have modified the main text (Page 12, line 226-227) to avoid any confusion.
However, as shown in this study, CNTs can provide relatively large S if EF is properly tuned, in spite of having an ultrahigh σ.

(B) "It is challenging to uniformly control the doping level of the different chirality
CNTs". First, while we understand the reviewer's concern, we point out the important fact that the work function of CNTs is essentially independent of the chirality. [4][5][6] Therefore, the relationship between the EF shift and the doping level is common to all chiralities. Second, we experimentally assessed the EF position by monitoring the suppression of optical absorption peaks caused by Pauli blocking, as summarized in Figure R2. Our TEM analysis showed two distinct diameter peaks, at around 0.9 nm and 1.8 nm, corresponding to the inner and outer tubes. Then the empirical Kataura plot 7 allowed us to identify different optical transition energies. Figure R2a shows an example for the outer-wall diameter, where a red shaded area indicates the diameter range from TEM. Note that we did not observe the E11 S peak in the as-produced film ( Figure R2d) due to Pauli blocking; doping shifted EF into the valence band ( Figure R2e). See Supplementary Note 1 for more details.
It is important to point out that Figure R2b (top) involves many chiralities: it contains 28 chiralities ranging from (15,8) (1.606 nm in diameter) to (15,14) (1.994 nm). Therefore, the peak in Figure R2b (bottom) is inhomogeneously broadened due to the presence of multiple chiralities. However, there is no visible peak in Figure R2d (bottom), suggesting that EF was effectively shifted into the valence band by doping for all semiconducting chiralities within this diameter range. Therefore, although we agree with the reviewer that there would in principle be small differences between different CNTs (such as outer-wall vs. inner-wall CNTs), Figure R2d convincingly suggests that in practice they are not large enough to change the conclusion of our manuscript.   Figure 7b).

Comment #2:
ZT needs to be plotted in the main manuscript so that readers can have the information without additional calculation.
Response to comment #2: We agree that ZT is an important quantity, and thus, this information should be included in the main text. While we did not measure the thermal conductivity of different samples under different conditions in this work (and thus we are unable to plot ZT), we estimated ZT for the CNT fiber with the largest power factor (annealed at 500 ˚C) by using the previously reported thermal conductivity value for a similar sample. 8 We have added the following sentence to the main text (Page 7, line 122-124).
Comment #3: 1. LED has been turned on with 83 mV but I wonder if there is a LED that can be operated with such a low voltage. It is necessary to describe the specification of the LED.
2. Also the LED was turned on for a very short period of time while TE is supposed to generate electricity continuously. This would be possible presumably with the circuitry. More detailed information is necessary for readers to have a better understanding about this work.
Response to comment #3: 1. The reviewer is correct that an LED cannot be driven by a voltage as small as 83 mV.
Indeed, there was nothing special about the LED (TEG-DMO, Custom Thermoelectric, LLC). We utilized a bootstrap converter (VB0410-1, TXL Group, Inc.), as mentioned in Page 13 line 249 in our previous manuscript. This DC/DC converter operates when the input voltage is 40 mV or higher, and the output voltage is 1 to 10 V, depending on the input voltage and load.
2. The LED was driven by a voltage that was discharged from capacitors. As we discussed and then discharged them to turn on the LED. In an RC discharging circuit, the voltage across the capacitor (Vc) at any instant in time during the discharging period is Assuming the thermal conductivity for similar annealed CNT fibers 16 , 580 Wm -1 K -1 , the average (maximum) ZT is estimated to be 6×10 -3 (7×10 -3 ) at 300 K.

= exp (− ),
where Vs is the initial voltage across the capacitor, t is the time, and τ is the time constant, defined as = , where R is the resistance and C is the capacitance. This equation indicates that Vc is 98.2 % discharged when = 4 . In our case, the time constant was 0.61 seconds. The converter requires a minimum voltage of 40 mV to operate; otherwise, it outputs 0 V. Therefore, even when the initial voltage Vs is 80 mV, Vc is less than 40 mV after 0.43 seconds, which explains why the LED was on for a short period of time.
We agree that we did not make these points clear in our previous manuscript, since details were only discussed in the Methods section and in the Supplementary Movie 1.