Suppressing Thermal Conductivity of Suspended Tri-layer Graphene by Gold Deposition

Transcription

1 Suppressing Thermal Conductivity of Suspended Tri-layer Graphene by Gold Deposition Jiayi Wang, Liyan Zhu, Jie Chen, Baowen Li,* and John T. L. Thong * The thermal properties of graphene are subject of considerable interest because of graphene s unusually high thermal conductivity ( κ ), and its potential in thermal management applications. Despite interesting predictions of very high κ for free-standing graphene near room temperature, experimental measurements to determine the thermal conductivity of such ultrathin graphene films comprising a single layer or a few atomic layers present a great challenge. A micro-raman spectroscopy approach, based on either the red shift of the Raman G- or D-band [ 1, ] or from the intensity ratio of Stokes/anti- Stokes scattering [ 3 ], was introduced recently for the thermal conductivity measurement of suspended single-layer graphene (SLG) and few-layer graphene (FLG). However, the reported values of κ for suspended SLG at 300 K show great variation ranging from 600 to 5800 W/K-m, [ 1 4 ] largely due to differences in the assumed absorbed laser power. In addition, the temperature resolution of the micro-raman technique is often poorer than 50 K due to the limited temperature sensitivity, and further degrades the accuracy of κ obtained. In contrast, the conventional thermal-bridge configuration offers direct measurements of the heating power and precise temperature ( T ) readout in vacuum, from which κ can be accurately extracted. This technique was previously only employed to study thermal transport in supported graphene [ 5 7 ] due to the challenge of preparing samples of suspended graphene straddling two suspended micro-thermometers. In more recent work, the as-measured κ of suspended SL-, [ 8 ] bilayer (BL-) [ 9 ] or 5-layer [ 7 ] graphene is unexpectedly low, which could be due to the detrimental effect of polymeric residues [ 9 ] or possible Dr. J. Wang, Prof. B. Li, Prof. J. T. L. Thong NUS Graduate School for Integrative Sciences and Engineering Singapore, , Republic of Singapore phylibw@nus.edu.sg ; elettl@nus.edu.sg Dr. J. Wang, Prof. J. T. L. Thong Department of Electrical and Computer Engineering Singapore, , Republic of Singapore Dr. J. Wang, Dr. L. Zhu, Dr. J. Chen, Prof. B. Li Department of Physics and Centre for Computational Science and Engineering (CCSE) Singapore, 11754, Republic of Singapore Prof. B. Li Center for Phononics and Thermal Energy Science School of Physical Science and Engineering Tongji University Shanghai, 0009, P. R. China DOI: /adma fluorination by XeF treatment [ 7 ]. Chemical functionalization has been shown theoretically to reduce κ of graphene significantly for small coverage of hydrogen, [ 10,11 ] fluorine [ 1 ] and hydrocarbon groups. [ 13 ] Accurate thermal transport measurement on suspended clean and pristine graphene remains as a great challenge, while systematic experimental thermal transport studies on the effect of substrate interaction are still lacking. In this work we report thermal conductance measurements on suspended tri-layer graphene (TLG) using a suspended microelectrothermal system (METS) in vacuum (< mbar). κ at 300 K was determined to be 1400 ± 140 and 1495 ± 150 W/K-m for two 5 μ m long samples with width of 5.04 and 1.8 μ m, respectively. Although these values are lower than the theoretically calculated value, [ 4,14 ] they are still more than twice that of the measured value of suspended BLG reported by Pettes et al. [ 9 ] This enhancement in the measured thermal conductivity is attributed to the better surface cleanness and high quality of TLG prepared by a dry-transfer method we have developed for sample fabrication. The descending trend of room temperature κ with respect to Au coverage agrees well with the results of molecular dynamics (MD) simulations we carried out for free-standing TLG. From the simulations, a continuous reduction in the density of states (DOS) of flexural acoustic phonon (ZA) modes is obtained, suggesting that the increasing suppression of ZA modes and the reduction of phonon lifetime by phonon leakage between TLG and the adjacent Au are primarily responsible for the reduction in κ at 300 K. Phonon scattering at the C Au interface and C C interface between suspended and supported (by Au nanoparticles) regions would reduce the mean free path (MFP) of the phonons, and further decreases κ. All aforementioned factors are the underlying physical mechanisms for the significant reduction in κ observed experimentally after Au deposition. The METS device used for the thermal bridge measurements ( Figure 1 a) is similar to the design reported by Shi et al., [ 5 ] who also described the measurement methodology in detail. Our design has a through hole for imaging in the transmission electron microscope (TEM), and provision for 4-point electrical measurements of the sample. The heater and sensor labeled in Figure 1 a each comprises a Pt loop as a resistance thermometer suspended by six long beams for thermal isolation from the substrate bulk. To facilitate transfer of extremely fragile TLG, the nominally suspended heater and sensor structures are linked together by two 10 μ m long nitride beams, and are also strapped sideways to the substrate by another two 71 μm long nitride beams this arrangement is to minimize relative displacements of the heater and sensor during sample preparation and mounting that could otherwise tear the fragile graphene. The structure is released from the supporting Si substrate by wet etching prior to transfer of the graphene onto the device. 6884

2 Figure 1. Optical images of released METS (a) without and (b) with suspended TLG (transmission image); scale bars: 5 μm. In view of the detrimental effect of polymeric residues on the thermal transport property of graphene, [ 9 ] we have developed a dry-transfer process (Supporting Information I) to ensure that the sample mounted on the METS is clean and free of such residues. After sample transfer (Figure 1 b), the 4 nitride straps were cut using a focused-ion beam (FIB) to disconnect the heater and sensor from each other, and from the sides of the substrate, taking great care to avoid both direct and indirect exposure of the sample to the ion beam. The whole device was then thermally annealed in H /Ar ambient at 30 C for hours [ 15 ] prior to thermal measurement. We fabricated two suspended TLG samples (DGS1 of μm and DGS of μm, L W ), where the two longitudinal edges were not patterned for preservation of the natural edge to minimize edge roughness and damage induced by plasma treatment. We followed the same approach described by Wang et al. [ 7 ] for the thermal measurements on suspended TLG. The device was loaded in the vacuum chamber ( 10 6 mbar) for in-situ annealing at 600 K for 6 hours to remove residues and to desorb physisorbed gas molecules on the TLG. The total thermal conductance of the six connecting beams G b to the substrate is given by: Q G b = T h + T s = 1.114I h R h T h + T s (1) while the thermal conductance from heater to sensor G TLG is T s G TLG = G b = 1.114I h R h T s T h T s Th T () where Δ T h,s = T h,s T sub, and Q = 1.114I h R h is the total thermal power generated by I h based on the equivalent thermal circuit and calculation in Ref. [7 ]. Shi et al. [ 5 ] previously assumed a uniform temperature distribution over both the heater and sensor platforms, and hence the temperature rise Δ T h,s obtained from the four terminal electrical resistance R h,s could then be used to represent the temperature rise at the edges of the heater (sensor). However, in reality R h,s only represents the average temperature over the heater (sensor) platform. For a sample of high thermal conductance, the finite element simulations we carried out show that the indicative temperature could differ significantly from the actual temperature at the edge of the heater (sensor). As such, the internal thermal resistance of the heater and the sensor are determined from the finite element simulation, and the sample thermal resistance is obtained by subtracting this internal thermal resistance from the measured total thermal resistance based on the equivalent thermal circuit shown in Figure Sa, Supporting Information. The details of the finite element simulation and the method to calculate the thermal conductance of suspended TLG can be found in Supporting Information II. An initial thermal measurement was first carried out on the pristine sample to obtain its original thermal conductivity ( κ o ) at 300 K. Thermal evaporation of Au was then carried out on the backside of this METS-mounted sample (to avoid shorting of the Pt loops on the topside) along with 4 6 TLG reference samples prepared using the same dry transfer technique for characterization using a TEM and an atomic force microscope (AFM). Following a cycle of evaporation, one of each type of reference sample was taken out to analyze the Au particle size, s Figure. TEM images of suspended TLG samples at different Au area coverage; scale bars: 0 nm. Inset in (c): The selected area electron diffraction of the sample, the perfect hexagon matches the diffraction pattern of TLG. 6885

3 Figure 3. (a) Normalized room temperature thermal conductivity with respect to the Au area coverage ratio of gold nanoparticles on the bottom side of TLG. (b) Normalized room temperature thermal conductivity by MD simulation with respect to the Au atomic coverage for a suspended TLG ( L = 50 nm). height, and coverage (Supporting Information III), while the METS-mounted sample was measured again to obtain the new κ. Figure shows the TEM images of the TLG samples at different Au area coverage. Figure 3 a shows the normalized κ at T = 300 K with respect to the Au area coverage percentage. The thermal conductivity is calculated from κ = G s L /Wt, where t is the thickness of TLG. κ o for the pristine sample is measured to be 1400 ± 140 and 1495 ± 150 W/K-m for DGS1 and DGS, respectively. Such room temperature values are much higher than that reported for BLG with size of μm (L W ) using a thermal bridge method, [ 9 ] even though the thermal conductivity of suspended graphene is expected to decrease with respect to the increasing thickness. [ 4,14 ] We propose that this is due to the better surface cleanness and higher quality of our samples. Figures 4a and 4 b show a TEM image of the TLG sample and an AFM image of graphene flakes of different thickness without any apparent traces of impurities and residues, while the Raman spectrum (Figure 4 c) has no observable D peak. Nevertheless, the measured room temperature thermal conductivities are lower than the theoretical prediction [ 4 ] for freestanding TLG ( 330 W/K-m, κ was normalized to the length L = 5 μ m). This could be attributed to the thermal boundary resistance (TBR) for phonon transport from the heater to suspended TLG (and from TLG to the sensor) through the C C interface at the platform edge. The presence of TBR was explained by Wang et al. [ 7 ] and it was experimentally measured to be 10 5 K/W for the two junctions in total in the case of TLG with a width of 5 μm. This TBR is an additional thermal resistance in series with sample thermal resistance in the equivalent thermal circuit shown in Figure Sa, and it is reasonable to scale the TBR in inverse proportion to the width of TLG based on the assumption of constant thermal boundary resistivity for the same type of interface. This gives us a TBR of and K/W for DGS1 and DGS, respectively. By taking the TBR into consideration, the corrected values of κ o are 1860 and 80 W/K-m, respectively, which are very close to the theoretical value. This demonstrates that the quality of the samples measured is comparable to that of pristine graphene. The presence of the TBR also accounts for the much lower κ of pristine DGS1 at low temperatures (see Figure S6) compared to that of high quality graphite, as this extrinsic scattering effect can dominate the thermal conductivity of the sample at low temperatures instead of intrinsic Umklapp scattering. We observe a clear descending trend of normalized κ with respect to Au area coverage percentage up to 73% with a maximum reduction of 8% in Figure 3 a. A similar descending trend is also observed in a MD simulation with respect to Au atomic coverage up to 50% as shown in Figure 3 b. The coverage percentage in MD simulation is defined as the ratio of Au atoms to carbon atoms in the adjacent layer. In MD simulation, a TLG with a dimension of 50 5 nm ( L W, Figure S7) is modeled. Detailed parameters for MD simulations are provided in Supporting Information V. For free-standing pristine TLG, κ o is estimated to be 116 ± 14 W/K-m. This value is lower than the theoretical prediction and our experimental results, as graphene's κ depends on the length of the flakes along which the heat transports, and the length L ( 50 nm) of TLG used in our simulations is orders of magnitude smaller. As Au atoms are randomly loaded on graphene, Figure 4. (a) TEM image of suspended TLG prepared by dry transfer technique; scale bar: 100 nm. Inset: Diffraction pattern of the sample; scale bar: 5 1/nm. (b) AFM image of graphene flakes on Si/SiO substrate that had undergone the same transfer process; scale bar: μ m. (c) Raman spectrum of suspended TLG prepared by dry transfer process. Inset: Raman spectrum in the range of 100 to 1400 cm 1 indicating the absence of a detectable D peak. 6886

4 Figure 5. Density of states of phonon modes for (a) out-of-plane and (b) in-plane phonons of top layer at different Au atomic coverage with respect to phonon frequency. κ decreases remarkably for increasing coverage of Au atoms. To understand the physical mechanism responsible for the reduction in κ, we investigate the vibrational phonon density of states (DOS) of the top layer of graphene as shown in Figure 5. It clearly indicates negligible influence on the in-plane DOS by the presence of loaded Au atoms (Figure 5 b). On the other hand, the out-of-plane DOS is apparently reduced compared to that of free-standing TLG (Figure 5 a). The two major peaks in Figure 5 a correspond to acoustic and optical out-of-plane modes (ZA and ZO), respectively. [ 16 ] The height of ZA peak monotonically decreases as Au coverage increases, indicating that the loaded Au atoms inhibit the out-of-plane motion of graphene. Previously it has been reported that the ZA phonon modes contribute 77% of heat transfer in suspended SLG at 300 K. [ 6 ] Klemens [ 17 ] also suggested that phonons will leak from graphene into a substrate of lower phonon velocity, which could reduce the thermal conductivity by 0% to 50% depending on the phonon velocity difference. Although the net phonon leakage in steady state would reach zero, such a leakage process still greatly reduces the phonon lifetime in graphene, [ 6 ] and hence lowers the thermal conductivity of graphene. Moreover, the presence of Au atoms causes strong phonon scattering at the C Au boundary that reduces the phonon MFP. All the aforementioned factors will significantly weaken the phonon transport properties of graphene. Therefore, κ of TLG decreases as the Au coverage increases. The experimentally-observed reduction in κ is much larger than the simulated reduction a 5% reduction in κ was obtained for DGS at 36% Au coverage, while only 4% reduction in κ was calculated for 40% coverage in the simulations. The difference is due to (a) the inability to describe covalent bonding interaction between the small fraction of Au and defects in graphene and formation of multi-layered Au islands in MD simulations and (b) the size effect. Detailed discussion can be found in Supporting Information VII. However, the experimental reduction of 8% in κ is larger than the 77% reduction calculated for the ZA-mode contribution towards heat conduction in free-standing SLG. The additional reduction in κ can be attributed to the effect on the inplane phonon modes (LA and TA) by the boundary scattering at the C C interface between the suspended and supported TLG similar to the bulk case discussed previously. [ 7 ] At intermediate temperatures such as 300 K, transmission at the boundary is mainly determined by the mismatch in the DOS of phonon modes, since the phonon wavelength is short and the scattering is purely diffusive. [ 18 ] The scattering lowers the MFP of the in-plane phonon modes, further reducing κ. Strain induced in TLG by Au deposition could be another source of modification to κ. The effect of strain on the κ of graphene and graphene nanoribbons has been studied theoretically for both ballistic [ 19 ] and non-ballistic [ 0,1 ] thermal transport, and was found to be an enhancement effect for ballistic transport under tensile strain and a reduction effect for non-ballistic transport. The thermal transport in our sample is clearly non-ballistic due to the large sample dimensions and impurity (Au) attachment. The two C C bonds (bonds A and B in Figure 1 a of Ref. [ 0 ] ) in the hexagonal lattice of graphene are shown [ 0,1 ] to have deviated bond lengths for both uniaxial and biaxial tensile strain, which would result in imperfect lattice structure of graphene. The inset in Figure c shows the electron diffraction pattern of DGS after final Au deposition. The diffraction pattern is in good agreement with the perfect graphitic hexagon drawn. In the worst case, only a small amount of uniaxial or biaxial compressive strain might be present in the lattice. Since less than 10% reduction in κ was found to result from 10% compressive (very strong) strain in graphene, it is reasonable to neglect the contribution of strain to the reduction of thermal conductivity observed in our mass loading experiment. The slight increase in the normalized κ after 7% coverage for the experimental results is due to the bridging of Au nanoislands into a giant Au network, which provides a parallel thermal path for heat conduction from the heater to the sensor, as the derived κ is based on heat conduction taking place solely through the TLG. Figure c shows the TEM image of DGS for 99.3% Au coverage, at which the normalized κ shows the most significant increase. It is clearly seen that the Au nano-islands have merged together to form a percolation network, with a total Au thickness more than 10 nm. Though κ of such a thin film of Au is much poorer than the bulk value for Au (14% of bulk κ for nm Au thin film [ ] ), the thickness of Au deposited is still 10 times that of TLG, and results in significant contribution to the total thermal conductance. This additional thermal conduction through the Au thin film is responsible for the increase in κ. In conclusion, we have experimentally studied the thermal conductivity of two suspended TLG samples using a thermalbridge configuration. The room temperature values of κ were found to be 1400 ± 140 and 1495 ± 150 W/K-m without the correction for in-plane TBR in the calculation. These values are considerably high for thermal bridge measurement as we have improved the sample cleanness and quality of graphene to be comparable to pristine samples. This represents a significant step forward for the investigation of fundamental physics of thermal conduction in D materials as low temperature measurements are easily carried out using the thermal bridge configuration. We experimentally monitored 6887

5 and theoretically calculated the reduction of κ in suspended TLG as the result of deposition of Au at T = 300 K, both showing a trend of decreasing normalized κ with respect to Au coverage. The reduction in κ is mainly attributed to the suppression effect of ZA phonon modes in TLG by Au which is reflected in a progressive reduction of the DOS of ZA phonon modes with respect to increasing Au coverage, and the reduction in phonon lifetime in TLG due to phonon leakage between TLG and Au. Furthermore, the boundary scattering at the C Au and C C interfaces also increases with respect to Au coverage, and contributes significantly to the reduction in κ. Our results are the first reported experimental work which quantitatively studies how deposited impurity atoms affect thermal conductivity of graphene, and demonstrates the potential to practically suppress the thermal conductivity in suspended graphene by deposition of impurity particles, and thereby facilitate the practical development of graphene-based devices with tunable thermal conductivity for thermal management. Supporting Information Supporting Information is available from the Wiley Online Library or from the author. Acknowledgements This work is supported by the Ministry of Education (MOE), Singapore, by Grant MOE01-T Received: July 0, 013 Published online: September 18, 013 [1] A. A. Balandin, S. Ghosh, W. Bao, I. Calizo, D. Teweldebrhan, F. Miao, C. N. Lau, Nano Lett., 008, 8, 90. [] W. Cai, A. L. Moore, Y. Zhu, X. Li, S. Chen, L. Shi, R. S. Ruoff, Nano Lett. 010, 10, [3] C. Faugeras, B. Faugeras, M. Orlita, M. Potemski, R. R. Nair, A. K. Geim, ACS Nano 010, 4, [4] S. Ghosh, W. Bao, D. L. Nika, S. Subrina, E. P. Pokatilov, C. N. Lau, A. A. Balandin, Nat. Mater. 010, 9, 555. [5] L. Shi, D. Li, C. Yu, W. Jang, D. Kim, Z. Yao, P. Kim, A. Majumdar, J. Heat Transf. 003, 15, 881. [6] J. H. Seol, I. Jo, A. L. Moore, L. Lindsay, Z. H. Aitken, M. T. Pettes, X. Li, Z. Yao, R. Huang, D. Broido, N. Mingo, R. S. Ruoff, L. Shi, Science 010, 38, 13. [7] Z. Q. Wang, R. G. Xie, C. T. Bui, D. Liu, X. X. Ni, B. W. Li, J. T. L. Thong, Nano Lett. 011, 11, 113. [8] X. F. Xu, Y. Wang, K. Zhang, X. Zhao, S. Bae, M. Heinrich, C. T. Bui, R. G. Xie, J. T. L. Thong, B. H. Hong, K. P. Loh, B. W. Li, B. Oezyilmaz, arxiv: v [9] M. T. Pettes, I. Jo, Z. Yao, L. Shi, Nano Lett. 011, 11, [10] Q. X. Pei, Z. D. Sha, Y. W. Zhang, Carbon 011, 49, 475. [11] S. K. Chien, Y. Z. Yang, C. K. Chen, Appl. Phys. Lett. 011, 98, [1] W. X. Huang, Q. X. Pei, Z. S. Liu, Y. W. Zhang, Chem. Phys. Lett. 01, 55, 97. [13] S. K. Chien, Y. Z. Yang, C. K. Chen, Carbon 01, 50, 41. [14] L. Lindsay, D. A. Broido, N. Mingo, Phys. Rev. B 011, 83, [15] M. Ishigami, J. H. Chen, W. G. Cullen, M. S. Fuhrer, E. D. Williams, Nano Lett. 007, 7, [16] N. Mounet, N. Marzari, Phys. Rev. B 005, 71, [17] P. G. Klemens, Int. J. Thermophys. 001,, 65. [18] F. X. Alvarez, J. Alvarez-Quintana, D. Jou, J. R. Viejo, J. Appl. Phys. 010, 107, [19] X. C. Zhai, G. J. Jin, Europhys. Lett. 011, 96, [0] N. Wei, L.Q. Xu, H. Q. Wang, J. C. Zheng, Nanotech. 011,, [1] F. Ma, H. B. Zheng, Y. J. Sun, D. Yang, K. W. Xu, P. K. Chu, Appl. Phys. Lett. 01, 101, [] G. Chen, P. Hui, Appl. Phys. Lett. 1999, 74,