Influence of optically active defects on thermal conductivity of polycrystalline diamond

Transcription

1 Eur. Phys. J. Appl. Phys. 80, (2017) EDP Sciences, 2017 DOI: /epjap/ Regular Article THE EUROPEAN PHYSICAL JOURNAL APPLIED PHYSICS Influence of optically active defects on thermal conductivity of polycrystalline diamond Qinyu Kong 1, Alvarado Tarun 2, Chuan Ming Yap 2, Siwei Xiao 2, Kun Liang 1, Beng Kang Tay 1,*, and Devi Shanker Misra 2 1 NOVITAS, Nanoelectronics Center of Excellence, School of Electrical and Electronic Engineering, Nanyang Technological University, Singapore , Singapore 2 IIa Technologies Pte Ltd., 17 Tukang Innovation Drive, Singapore , Singapore Received: 21 June 2017 / Received in final form: 14 September 2017 / Accepted: 21 September 2017 Abstract. We systematically studied the influence of optically active defects on thermal conductivity for polycrystalline diamonds (PCDs) with different colour, crystalline quality and impurity concentrations. The thermal conductivities of PCDs on the growth (top) and nucleation (bottom) surfaces were characterized with 3v technique. It is found that the bottom surface shows lower thermal conductivity as compared to the top surface. This could be due to the higher defect density in the bottom surface. Defects analyzed includes nondiamond carbon phase, C H stretching vibration, Si vacancy, and substitutional nitrogen (Ns 0 ). Our results suggest that, for the top surface, the heat transport is mainly controlled by the concentration of Ns 0. For the bottom surface, non-diamond carbon phase, Si vacancy, C H stretch and Ns 0 defects all lead to an obvious reduction in the thermal conductivity. Most importantly, we derived a well fitted equation that estimates the thermal conductivity by optical transmittance, and the equation was demonstrated to be valid at any wavelength in visible region. 1 Introduction Diamond has the highest thermal conductivity (2200 W/ mk) among all known materials [1]. Growing diamond by microwave plasma assisted chemical vapor deposition (MPCVD) has largely reduced the cost and make it commercially available for thermal applications [2]. Among all diamond materials, polycrystalline diamond (PCD) stands out because it is more economical for production and able to be grown into large wafer sizes on various substrates [3]. Besides, the thermal conductivity of high quality PCDs can reach up to 2200 W/mK [4], which is nearly identical to the value of pure single crystalline diamond (SCD). All these advantages make PCD a competitive candidate for advanced thermal management material. However, the thermal properties of PCD is highly process dependent [5]. Due to the columnar growth mechanism, the properties of the top surfaces (growth surfaces) and the bottom surfaces (nucleation surfaces) can be different [3]. So far, the studies discussing the diamond thermal conductivity difference on top and bottom surfaces are explained by the grain size difference [6,7]. However, when grain size is comparable or larger than the phonon mean free path within diamond, grain size effect is no longer significant [7]. For PCDs with large grain sizes in * EBKTay@ntu.edu.sg micrometers, a systematic study is still lacking on the thermal conductivity difference between top and bottom surfaces due to other kinds of defect scattering, such as impurities. Besides, since both thermal conductivity and optical transmission are highly affected by defect concentration [8], it is interesting to study the correlation between these two properties. In this work, thermal conductivities and various defect densities for PCD are well characterized on both top and bottom surfaces. The degree of crystallinity and the presence of non-diamond carbon phase within PCDs are evaluated. The defects of diamond in the form of single substitutional nitrogen (Ns 0 ), C H stretch bonds, the neutral and charge nitrogen (N-V 0/ ) vacancies as well as silicon vacancy (Si-V) are analyzed in details with regards to their impacts on thermal conductivities. A strong dependence of thermal conductivity on optical transmission in visible region is observed. This enables a fast and reasonable estimation of thermal conductivity by visual inspection. 2 Experimental 2.1 Sample preparation PCD samples were prepared at IIa Technologies Pte Ltd. using MPCVD system. There are 21 PCDs grown on 1 cm 1 cm Si substrates under different growth parameters namely: temperature, pressure and C/H/O/N ratio p1

2 2 Q. Kong et al.: Eur. Phys. J. Appl. Phys. 80, (2017) Fig. 1. SEM images for one freestanding PCD sample after removal of Si substrate: (a) cross-section; (b) top surface (growth side); and (c) bottom surface (nucleation side). Fig. 2. (a) PL and Raman spectra (inset) for the top and bottom surfaces of one PCD sample. The PL spectrum for one high quality SCD sample shown as a dashed line for reference. (b) Thermal conductivity versus the integrated intensity of non-diamond carbon phase band from 1350 to 1650 cm 1. (c) Thermal conductivity versus the integrated intensity of the background luminescence from 560 to 900 nm. (d) Thermal conductivity versus the intensity of Si-V peak. After growth, Si substrates were dissolved using a mixture of nitric acid (HNO3) and hydrofluoric acid (HF). Figure 1 shows the SEM images for one freestanding PCD sample after removal of Si substrate. PCDs were then mechanically polished to surface roughness less than 10 nm on both sides. The thicknesses of PCD samples range from 0.25 to 1 mm after polishing. The grain sizes were further estimated after H2/O2 etching of the surfaces, and the range is from 2 to 200 mm. 2.2 Thermal measurements The thermal conductivities of both the top and bottom surfaces for PCDs were measured at room temperature (25 C) using 3v technique [9]. A metal wire composed of 20 nm Ti and 100 nm Au is deposited on the PCD surface, which function as both a heater and thermometer. The metal wire is 1 mm in length and 8 mm in width. An alternating current (AC) of v is applied to the metal wire p2

3 Q. Kong et al.: Eur. Phys. J. Appl. Phys. 80, (2017) 3 Fig. 3. (a) FTIR absorbance spectrum from 400 to 5000 cm 1 for one PCD sample. (b) FTIR single phonon absorption region. (c) FTIR CH x absorption band. This AC current input causes a temperature wave at a frequency of 2v to diffuse into the PCD sample due to the Joule heating effect, which further result in a 3v voltage harmonics U 3v as [10]: U 3v ¼ U3 TCR 4plkR 0 lnð2vþþln b2 D 2j ; ð1þ where U is the voltage across the metal wire; R 0 is the electrical resistance of the metal wire; TCR is the temperature coefficient of resistance for the metal wire, which is measured to be K 1 ; l is the length of the metal wire; b is the half width of the metal wire; k is the thermal conductivity of measured sample; D is the thermal diffusivity of the measured sample; and j is a constant. According to equation (1), the thermal conductivity k is deduced from the slope of the U 3v with a function of ln(2v). The thermal conductivity obtained for each sample is the averaged value measured from 5 different positions of the sample surface, and the standard deviation is less than 10%. The thermal conductivity of a high quality SCD with no detectable defects is measured as K 0 =2300W/mK as reference. The direction of thermal conductivity measured by 3v technique is determined by the ratio of the width of heater to the thickness of sample underneath [11]. In our measurement, the width of heater is only 8 mm, which is much smaller than the average thickness of PCD samples (500 mm). Thus the heater can be treated as a line heater, and the measurement is more sensitive to the in-plane thermal conductivity. 2.3 Optical measurements Micro-Raman and photoluminescence (PL) spectra were measured using Ar + laser (514.5 nm) excitation of the Renishaw InVia Raman microscope. The laser spot size was approximately 1 mm using a 50 microscope objective with a numerical aperture of Fourier transform infrared spectra (FTIR) were obtained by a Nicolet 6700 spectrometer (Thermo Scientific) using a KBr beam splitter and a DTGS KBr detector. Ultraviolet-visible spectra were collected by a UV 2600 spectrophotometer (Shimadzu) with the ISR-2600 integrating sphere attachment. 3 Results and discussion 3.1 Photoluminescence (PL) and Raman analysis Figure 2a shows the PL and Raman spectra (inset) for one PCD sample. The intensities have been normalized to the diamond first-order Raman peak around 552 nm (1332 cm 1 ). The PL spectrum for a high-quality SCD with no detectable neutral and charge nitrogen (N-V 0/ ) vacancies and Si-V defect is shown as dashed line in Figure 2a for reference. For bottom surfaces, a strong Si-V peak appears at 738 nm, which is a result of the interaction between the growth gases and the Si substrate. For top surfaces, the zero phonon lines for NV 0 and NV show up at 575 and 637 nm, and the Si-V peak is absent. A background luminescence from 560 to 900 nm is observed on both the top and bottom surfaces, from which we can estimate the p3

4 4 Q. Kong et al.: Eur. Phys. J. Appl. Phys. 80, (2017) Fig. 4. (a) Integrated intensity for CH x absorption from 2760 to 3030 cm 1 versus thermal conductivity on the bottom surface. (b) Integrated intensity for CH x absorption from 2760 to 3030 cm 1 versus thermal conductivity on the top surface. Fig. 5. (a) The correlation between through-plane CH x absorption and estimated surface CH x content. (b) Thermal conductivity difference DK (K top K bottom ) versus CH x absorption. defect concentration, since for high quality and pure singlecrystalline diamond, the PL background is almost zero [12]. Other researchers have used the background luminescence in the visible wavelength as a semi-quantitative method of approximating purity [13 15]. A non-diamond phase band is observed from 1350 to 1650 cm 1 on the bottom surface. From bottom to top surface, the full-width at halfmaximum (FWHM) of first order Raman peak of diamond drops from 5.17 to 2.43 cm 1, which indicates a higher degree of crystallinity of the top surface. As shown in Figure 2b, bottom surfaces show a higher intensity of non-diamond carbon phase (I ND ) compared with top surfaces. As I ND increases, the thermal conductivity on both top and bottom surfaces decreases. However, when I ND is very small (see the grey area in Fig. 2b), the thermal conductivities on top surfaces distributes randomly. This indicates the non-diamond carbon phase is not the main contributor and the influences from other defects could be more dominant on top surfaces. For bottom surfaces, the influence from non-diamond carbon phase is more obvious because of the higher defect concentration. As shown in Figure 2c, top surfaces show a lower PL background intensity I PL than bottom surfaces. This indicates that the defect densities decrease as diamond grows. As I PL increases, the thermal conductivity on top and bottom surfaces decreases. For all PCD samples measured, the Si-V peak only appeared on the bottom surface. Similarly, as the intensity of Si-V peak on bottom surfaces increases, the thermal conductivity decreases (see Fig. 2d), which is in good agreement with the molecular dynamic simulation work [16]. 3.2 Fourier transform infrared spectroscopy (FTIR) analysis The FTIR absorption spectrum for one PCD sample, as well as a defect-free SCD, is shown in Figure 3, which is normalized by scaling the sample thickness to be 1 cm. As shown in Figure 3a, the intrinsic diamond absorption is observed in two-phonon ( cm 1 ) and threephonon ( cm 1 ) regions. Beyond 4000 cm 1, there is no detectable absorption. For perfect diamonds without any lattice distortion, no single phonon absorption is expected. In Figure 3b, although TO(L) and O(G) modes are observed in single phonon region, the absorption is very low (less than 1 cm 1 ). No nitrogen related absorption band is detectable in single phonon region, thus our PCD samples are mostly type IIa diamonds. The CH x stretching band is observed in the region from 2760 to 3030 cm 1 (see Fig. 3c), which is superposition of several CH x vibration modes [17] p4

5 Q. Kong et al.: Eur. Phys. J. Appl. Phys. 80, (2017) 5 Fig. 6. UV Vis transmission spectra for PCDs with the highest, medium, lowest transmission at 800 nm are shown for example. The transmission spectra for a high quality single crystalline diamond and transmission limit are shown for reference. Fig. 8. Fitting value of A under wavelengths from 450 to 800 nm. Fig. 7. (a) Ns 0 absorption coefficient a Ns 0 thermal conductivity. at 270 nm versus thermal conductivity. (b) Optical transmission (%) at 633 nm versus The CH x absorption a CHx was integrated from 2760 to 3030 cm 1. As shown in Figure 4a, for the bottom surface, the thermal conductivity decreases as a CHx increases. The data fits well with the model reported in reference [4]. For the top surface, the dependence of thermal conductivity on a CHx is not as obvious as that on the bottom surface (see Fig. 4b). One possible reason is that FTIR is a throughplane characterization, thus a CHx collected by FTIR do not directly reflect the CH x concentration on the top or the bottom surface, but the CH x concentration through the whole sample. As reported [18], the CH x concentration shows a linear correlation with the concentration of nondiamond carbon phase for CVD diamonds. Thus it is reasonable to use the surface intensity of non-diamond carbon phase characterized by Raman to estimate the surface CH x concentration. Figure 5a shows the correlation between the through-plane CH x absorption and estimated surface CH x intensities. Linear correlations are observed for both top and bottom surfaces. The slope for the bottom surface is much larger than that for top surface, which indicates the concentration of CH x on the bottom surface is much higher than that on the top surface. The standard fitting error for the slope of the top surface (±84%) is much higher than that of the bottom surface (± 11%). This implies that the CH x absorption collected by FTIR is not sensitive enough to reflect the CH x concentration on the top surface. As shown Figure 5b, the thermal conductivity difference DK increases as a CHx increases. It demonstrates that hydrogen impurities affect the bottom surface more than the top surface. 3.3 Ultraviolet visible spectroscopy (UV VIS) analysis PCD samples studied in this work are in a wide range of optical transmission (see Fig. 6). The UV Vis spectrum for the high quality SCD with no detectable defect is also shown in Figure 6 as a reference, which is almost overlapped with the transmission limit. According to the UV Vis spectra, once the wavelength exceeds the fundamental cut-off at 220 nm, the optical transmission rises quickly in UV region. Then it smoothly reaches to a limit in visible region. Electron spin resonance studies on MPCVD diamonds [19] suggests that nitrogen is incorporated within diamond lattice as single substitutional nitrogen atoms Ns 0. The concentration of Ns 0 absorption can be extracted at 270 nm p5

6 6 Q. Kong et al.: Eur. Phys. J. Appl. Phys. 80, (2017) Fig. 9. Thermal conductivity versus grain size for top and bottom surfaces. absorption from UV Vis spectra using the method described by Nistor [20]. Figure 7a shows the correlations between the Ns 0 absorption coefficients a Ns 0 at 270 nm versus thermal conductivities on top and bottom surfaces. Despite the scattered data points, we observe an overall increase in the Ns 0 concentration with a decrease in thermal conductivity. It is because of more phonon scattering by the nitrogen occurs. In this work, we were investigating type IIa samples with subppm concentration of nitrogen. Operating at such a low level and range of nitrogen inevitably increases the scatter we see in Figure 7a. Unlike CH x defects, DK between top and bottom surfaces showsno dependence onns 0 concentration. The reason could be that Ns 0 distributes evenly on the top and bottom surfaces because the nitrogen gas concentration is kept constant during the entire growth of PCDs along with other growth parameters. Thus the impact of Ns 0 is not significant on the observed difference in the thermal conductivity between the top and bottom surfaces. No obvious absorption features are observed in visible region(seefig.6),thusitismorereasonabletousethe optical transmission at visible region for thermal conductivity estimation, instead of the UV region. Figure 7b shows a strong dependence of thermal conductivity on optical transmission at 633 nm. The optical transmission is normalized by scaling the sample thickness to be 0.5 mm. The pictures of PCD samples with optical transmission of 22.5%, 45.2% and 65.2% at 633 nm are also shown in Figure 7b for reference. The correlation between thermal conductivity K and optical transmission T is fitted using [21]: K ¼ 1 þ Aln T ; ð2þ r T where A is a constant and K 0 is thermal conductivity of pure and defect-free diamond. In this work, we used the measured thermal conductivity for the high quality SCD with no K 0 detectable defect as K 0 = 2300 W/mK. After fitting, A top = ; A bottom = were obtained for the optical transmission at 633 nm. As the optical transmission decreases, the thermal conductivity of the bottom surface drops faster than that of the top surface, which is due to the presence of more defects (such as CH x, Si-V and smaller grains) on the bottom surface than the top surface. The trendlines of thermal conductivity on the top and bottom surface intersect at 2300 W/mK when the optical transmission reaches up to the transmission limit of 70.75%, which is the value for the high quality SCD with no detectable defects. Comparing with the specific defects discussed before (such as non-diamond carbon phase, Si-V, CH x,orns 0 ), the R-square of thermal conductivity versus optical transmission is much closer to 1 (see Tab. 1), which indicates a much stronger dependence. This is because optical transmission not only gives information about defects densities but also the lattice dislocation and grain boundaries, which influence thermal conductivity as well. Therefore it is more reasonable to use optical transmission to estimate thermal conductivity for type IIa diamonds. In fact, a strong dependence of thermal conductivity on optical transmission is observed at any wavelength in visible region, not only at 633 nm. We also found that equation (2) fitted well on any wavelength in visible to near IR ( nm) region. A linear increase in the fitting value of A is observed for both top and bottom surfaces as the wavelength increases (see Fig. 8). This is somewhat expected because the optical absorption coefficient a is wavelength dependent. The linear behavior is due to the almost linear change of the refractive index for PCDs from 450 to 800 nm. The fitting value of A and the slope (DA/Dl) on the top surface is smaller than that on the bottom surface, because the top surface has lower defect densities than the bottom surface. For pure diamonds (T T r ) with no detectable defect and uniform throughout its thickness, the thermal conductivity K reaches up to K 0, and the fitting value of A approaches zero p6

7 Q. Kong et al.: Eur. Phys. J. Appl. Phys. 80, (2017) 7 Table 1. R-square values for the correlations between thermal conductivities and different optical characterizations. Dependence of thermal conductivity R-square for top surface (W/mK) R-square for bottom surface (W/mK) Optical transmission at 633 nm (T) Non-diamond carbon phase (I ND ) PL background intensity (I PL ) Si-V intensity (I SiV ) CH x absorption (a CHx ) Ns 0 absorption (a Ns 0) Grain size analysis Because of the polycrystalline structure, the properties of PCDs are dependent on the grain sizes. For all the PCDs measured, top surfaces always show larger grain sizes than the bottom surfaces. It is in good agreement with the results described in Figure 2c, where top surfaces shows lower defect densities than bottom surfaces, since defects tends to be concentrated at grain boundaries [22]. As a result, for one PCD sample, the top surface always shows higher thermal conductivity than the bottom surface. However, if we compare different PCD samples together, no correlation is observed between thermal conductivity and grain size (see Fig. 9). As reported [7], at room temperature, the phonon mean free path within diamond is less than 10 mm, which implies the thermal conductivity reduction due to grain boundary scattering is not significant unless grain size is comparable to or less than 10 mm. It gives a good explanation to the random distribution of thermal conductivity when grain sizes are from tens to hundreds. However, as shown in the inset of Figure 9, even for the range of small grain size (from 2 to 18 mm), no clear dependence of thermal conductivity on grain size is observed. It is because the defect scattering difference caused by varying growth conditions is more significant comparing with the grain size effect. After all, at room temperature, the rate of grain boundary scattering is much smaller than other types of phonon scattering rate, such as impurities [23]. 4 Conclusions The influence of optically active defects on thermal conductivities has been studied systematically for PCDs. The top surface shows lower defect densities and higher thermal conductivity than the bottom surface. For top surfaces, Ns 0 is the main factor limiting the heat transport. The influence from non-diamond carbon phase and C H stretch is not significant because of the low concentrations of related defects. For bottom surfaces, non-diamond carbon phase, Si vacancy, C H stretching bonds and Ns 0 defects all lead to an obvious decrease in thermal conductivity. No clear dependence of thermal conductivity on grain sizes is observed for a grain size range of mm. For PCD with large grain sizes in micrometers, since the grain size is comparable to the phonon mean free path within diamond, the thermal conductivity reduction due to grain boundary scattering is no longer significant. A strong dependence of thermal conductivity on optical transmission in visible region is observed. A simple fitting equation has been established to estimate thermal conductivity using any wavelength in visible region. This enables a fast and reasonable estimation of thermal conductivity by visual inspection. The authors would like to acknowledge the funding by Ministry of Education (grant number: MOE2014-T ). References 1. J. Asmussen, D. Reinhard, Diamond films handbook (CRC Press, New York, United States, 2002) 2. S.T. Lee, Z. Lin, X. Jiang, CVD diamond films: nucleation and growth, Mater. Sci. Eng. R: Rep. 25, 123 (1999) 3. Mark A. Prelas, Galina Popovici, Louis K. Bigelow, Handbook of Industrial Diamonds and Diamond Films (CRC Press, New York, United States, 1997) 4. D. Twitchen et al., Thermal conductivity measurements on CVD diamond, Diam. Relat. Mater. 10, 731 (2001) 5. N. Govindaraju et al., Comparative study of textured diamond films by thermal conductivity measurements, Appl. Phys. A 85, 331 (2006) 6. J. Graebner et al., Thermal conductivity and the microstructure of state-of-the-art chemical-vapor-deposited (CVD) diamond, Diam. Relat. Mater. 2, 1059 (1993) 7. R.B. Simon et al., Effect of grain size of polycrystalline diamond on its heat spreading properties, Appl. Phys. Express 9, (2016) 8. A.M. Zaitsev, Optical properties of diamond: a data handbook (Springer Science & Business Media, Berlin, Germany, 2013) 9. D.G. Cahill, Thermal conductivity measurement from 30 to 750 K: the 3v method, Rev. Sci. Instrum. 61, 802 (1990) 10. S.-M. Lee, D.G. Cahill, Heat transport in thin dielectric films, J. Appl. Phys. 81, 2590 (1997) 11. Y.S. Ju, K. Kurabayashi, K.E. Goodson, Thermal characterization of anisotropic thin dielectric films using harmonic Joule heating, Thin Solid Films 339, 160 (1999) 12. A. Tarun et al., Impact of impurities and crystal defects on the performance of CVD diamond detectors, Diam. Relat. Mater. 63, 169 (2016) 13. S. Zhao, Characterization of the electrical properties of polycrystalline diamond films (The Ohio State University, Columbus, Ohio, USA, 1994) 14. M. Roy, V. George, A. Dua, In-depth analysis of impurities in HFCVD diamond thin films grown from acetone/hydrogen mixtures, Diam. Relat. Mater. 12, 1569 (2003) p7

8 8 Q. Kong et al.: Eur. Phys. J. Appl. Phys. 80, (2017) 15. K. Bogdanov et al., Size-dependent Raman and SiV-center luminescence in polycrystalline nanodiamonds produced by shock wave synthesis, RSC Adv. 6, (2016) 16. J. Che et al., Thermal conductivity of diamond and related materials from molecular dynamics simulations, J. Chem. Phys. 113, 6888 (2000) 17. S. Al-Riyami, S. Ohmagari, T. Yoshitake, Fourier transform infrared spectroscopic study of nitrogen-doped ultrananocrystalline diamond/hydrogenated amorphous carbon composite films prepared by pulsed laser deposition, Diam. Relat. Mater. 20, 1072 (2011) 18. H. Windischmann et al., Intrinsic stress in diamond films prepared by microwave plasma CVD, J. Appl. Phys. 69, 2231 (1991) 19. M. Hoinkis et al., Paramagnetic nitrogen in chemical vapor deposition diamond thin films, Appl. Phys. Lett. 59, 1870 (1991) 20. S. Nistor et al., Nitrogen and hydrogen in thick diamond films grown by microwave plasma enhanced chemical vapor deposition at variable H2 flow rates, J. Appl. Phys. 87, 8741 (2000) 21. J.E. Graebner, Simple correlation between optical absorption and thermal conductivity of CVD diamond, Diam. Relat. Mater. 4, 1196 (1995) 22. A.V. Hetherington, C.J.H. Wort, P. Southworth, Crystalline perfection of chemical vapor deposited diamond films, J. Mater. Res. 5, 1591 (1990) 23. K.E. Goodson, Thermal conduction in nonhomogeneous CVD diamond layers in electronic microstructures, J. Heat Transfer 118, 279 (1996) Cite this article as: Qinyu Kong, Alvarado Tarun, Chuan Ming Yap, Siwei Xiao, Kun Liang, Beng Kang Tay, Devi Shanker Misra, Influence of optically active defects on thermal conductivity of polycrystalline diamond, Eur. Phys. J. Appl. Phys. 80, (2017) p8