APPLIED PHYSICS LETTERS **, 171***-****

Thermal conduction in nanocrystalline diamond lms: Effects of the grain

boundary scattering and nitrogen doping

W. L. Liu, M. Shamsa, I. Calizo, and A. A. Balandina

Nano-Device Laboratory, Department of Electrical Engineering, University of California Riverside,

Riverside, California 92521

V. Ralchenko, A. Popovich, and A. Saveliev

General Physics Institute of the Russian Academy of Sciences, 38 Vavilov Street, Moscow 119991, Russia

Received 25 July 2006; accepted 7 September 2006; published online 26 October 2006

The authors investigated thermal conductivity K in nanocrystalline diamond NCD lms on

silicon using the 3 and laser ash techniques. The K temperature dependence has been studied for

the undoped and nitrogen-doped NCD lms for T = 80 400 K and compared with that in

microcrystalline diamond MCD lms. The effects of phonon scattering from the grain boundaries

and lm interfaces on thermal conduction have been studied using three different models. For NCD

the room temperature K is 0.1 0.16 W / cm K and decreases with nitrogen doping. The K

temperature dependence in NCD is different from that in MCD lms and can be adequately

described by the phonon-hopping model. 2006 American Institute of Physics.

DOI: 10.1063/1.2364130

Diamond materials, owing to their unusual properties por deposition CVD -grown MCD lm. The measurements

have been carried out using 3 technique for T = 80 400 K.

such as extreme hardness, chemical inertness, and high ther-

In addition, the laser ash technique LFT has been used to

mal conductivity, have been used for wear-resistive coatings,

study K dependence on N2-doping concentration at RT. To

optical windows, surface acoustic-wave devices, and heat

elucidate the role of the phonon-grain-boundary and phonon-

spreaders. To optimize the material properties, different

lm interface scattering, we evoked three different models. It

growth technologies and types of diamond materials have

is intriguing that K temperature dependence in NCD lms is

been developed. Recently introduced nanocrystalline

diamond1 NCD attracted much attention for potential ap- intermediate between that of crystalline and disordered ma-

plications in electronics.2 Synthesized by argon-rich micro- terials and described very well by the phonon-hopping

model. The latter suggests that K in NCD is limited by the

wave plasma assisted chemical vapor deposition process,

properties of the grain boundaries in a wide T range.

NCD lm can be grown with the grain diameter as small as

The samples were grown on Si substrates in a micro-

d 2 5 nm. Compared with faceted microcrystalline dia-

wave plasma CVD reactor DF-100 model, 2.45 GHz . The

mond MCD, NCD lms are smooth pinhole-free even for

MCD lm of 3.4 m thickness and surface roughness Ra

small 1 m thicknesses. It has also been demonstrated

= 133 nm, referred to as Poly, has been grown for 2 h with

that the electrical conductivity of NCD lms can be changed

gas mixture 4 % CH4 / 96% H2 at substrate temperature of

by nitrogen N2 doping to form n-type material.3 These

800 C, pressure of 90 Torr, and microwave power of

properties make NCD suitable for the proposed carbon-based

4.3 kW. The NCD samples NCD 0 lm thickness of

electronics,4 in particular, with components made of carbon

2.2 m, NCD 15 8.9 m, and NCD 25 9.5 m were

nanotubes. It may also be used in future downscaled comple-

grown in the same CVD reactor using Ar-rich gas mixtures

mentary metal-oxide semiconductor technology.

Ar/ 5 % H2 / 2 % CH4 / N2 with N2 concentrations of 0%,

Understanding thermal conduction in NCD lms is im-

15%, and 25%, respectively Ar was varied to balance at

portant for applications in electronics and coatings. In its

following conditions: microwave power of 2.4 kW, pressure

bulk form, diamond has the highest thermal conductivity K

of 90 Torr, and substrate temperature of 800 C. The aver-

of all materials, which at room temperature RT is K

age surface roughness Ra for the NCD is below 40 nm as

= 10 22 W / cm K depending on the quality.5 8 At the same

measured by the atomic force microscopy AFM see Fig.

time, only limited research has been done to understand ther-

1 a ; no columnar growth features were seen in the lm

mal transport in NCD.9 Vlasov et al.10 investigated the ther- cross section Fig. 1 b . The grain size d is 22 26 nm and

mal properties of the diamond-carbon composites d 2 m for NCD x-ray diffraction XRD data and MCD

6 nm and obtained K = 0.003 0.017 W / cm K at RT. lms scanning electron microscopy SEM data, respec-

Ahmed et al.11 reported RT K 0.26 W / cm K for their tively.

NCD. The previous studies did not establish temperature de- The sample postgrowth composition has been inspected

pendence, which is essential for understanding the dominant using Renishaw Raman spectrometer. Figure 2 shows Raman

scattering mechanisms limiting K in NCD lms. spectra under 488 nm excitation. In MCD lm the most

In this letter we report the temperature dependent study prominent peak is at 1332 cm 1, which corresponds to opti-

of thermal conduction in the undoped and N2-doped NCD cal vibrations in sp3-bonded carbon atoms in the diamond

lms. For comparison, we have measured K in chemical va- crystal structure while a weak peak at 1500 cm 1 is associ-

ated with sp2-dominant disordered carbon.12 14 The peaks

observable in our NCD samples, i.e., 1140 and 1330 cm 1,

a

Author to whom correspondence should be addressed; electronic mail:

and the overlapping bulge from two peaks, 1450 and

abqbq7@r.postjobfree.com; http://ndl.ee.ucr.edu

0003-6951/2006/89 17 /171915/3/$23.00 89, 171***-*-**** American Institute of Physics

Downloaded 07 Dec 2006 to 138.23.213.224. Redistribution subject to AIP license or copyright, see http://apl.aip.org/apl/copyright.jsp

171915-2 Liu et al. Appl. Phys. Lett. 89, 171***-****

FIG. 3. Thermal conductivity of the nanocrystalline diamond lms as a

function of the nitrogen doping concentration measured by the laser ash

technique squares and 3 method stars .

e-beam evaporation and lift-off. The measurements were

conducted inside a vacuum cryostat. A numerical program

based on the solution of the heat diffusion equations was

developed to extract K values.16 LFT, used for RT study, is

FIG. 1. a AFM image of the surface topography and b SEM image of the

based on measuring traveling time of the thermal wave ex-

cross section of a typical nanocrystalline diamond lm sample with 25% N2

cited by a laser pulse, which determines the thermal diffusiv-

doping . Surface roughness is Ra = 38.5 nm.

ity D. In these measurements a pulsed neodymium-doped

yttrium aluminum garnet laser was utilized for heating the

1560 cm 1, are characteristic for NCD. The peak of sample surface, while the temperature kinetics was moni-

tored with the HgCdTe detector.10 The thermal conductivity

1140 cm 1 is due to trans-polyacetylene at grain boundaries,

while the peak at 1450 cm 1 represents the mixture of a-C was found as K = D C, where and C are the mass density

with graphite content. The peaks at 1330 and 1560 cm 1 and speci c heat of the material, respectively. Figure 3

are D and G bands, respectively. With the addition of N2, the shows K of NCD with different doping levels obtained using

intensity of D band decreases while the intensity of the G two different methods at RT. In processing LFT data, we

assumed = 3.51 g / cm3 and C = 0.511 J / g K, which corre-

band increases accompanied by a shift to lower frequencies,

which is in line with the theory13 and reported experimental spond to crystalline diamond. 3 method determines K di-

data.14 rectly and does not require and C inputs. One can see that

The 3 technique is based on driving ac through the there is good agreement in K values obtained by these two

metal heater line at frequency 1, which results in heating, techniques. The measured K = 0.1 0.16 W / cm K for un-

measurable as a resistance change at the frequency of doped NCD is much smaller than the values for crystalline

3 .15,16 Since doped NCD lms were electrically conductive, diamond. K reduces further with increasing doping due to

stronger phonon scattering on point defects,17 reaching

we deposited 90-nm-thick SiN insulation layer on top sur-

faces. The Cr 10 nm / Au 100 nm metallic heater- 0.06 W / cm K in 25% N2-doped lm.

Figure 4 shows K as a function of temperature T for

thermometer wires with widths of 10 and 30 m were pat-

MCD lm Poly and two NCD lms NCD 0 undoped and

terned on top of the insulation layer and fabricated by the

NCD 25 doped with 25% N2 . Unlike bulk crystals, MCD

K-T behavior does not have a pronounced low-T peak and

rolls off slower than 1 / T around RT. These indicate that the

phonon scattering from the lm interfaces and grain bound-

aries play an important role at low T. The measured RT value

for MCD sample is K = 5.51 W / cm K. On the other hand, K

temperature dependence of NCD samples is rather different,

which suggests stronger grain boundary scattering. The

monotonic K increase with temperature is similar to that in

disordered materials. RT thermal conductivities for NCD

samples is 0.16 and 0.08 W / cm K for the undoped d

= 22 nm and doped d = 26 nm lms, respectively, i.e., one

to two orders of magnitude lower than those in MCD.

To explain K-T dependence for NCD and MCD lms,

we evoked three different theoretical models. The results of

calculations are shown in Fig. 4 with solid lines. The curve

for bulk diamond is calculated using Callaway s model,18

FIG. 2. Raman spectra from microcrystalline and three nanocrystalline dia-

which is commonly used to describe phonon transport in

mond lms.

Downloaded 07 Dec 2006 to 138.23.213.224. Redistribution subject to AIP license or copyright, see http://apl.aip.org/apl/copyright.jsp

171915-3 Liu et al. Appl. Phys. Lett. 89, 171***-****

that the effect of dopant clustering on K is somewhat offset

by the fact that N2 addition leads to increase in the grain size,

which was observed earlier.13

Notable discrepancy of PHM predictions for MCD at

low T can be explained by the Kapitza resistance at the lm

interfaces, which is not included in the model. Indeed, t pa-

rameter in PHM takes care of the thermal resistance at the

grain boundaries but does not account for the resistance at

the MCD/Si and MCD/insulator interfaces. In NCD lms,

the omission of the Kapitza resistance at the lm interfaces

does not show up because the total grain boundary scattering

in NCD is much stronger than in MCD since d is two orders

of magnitude smaller . To sum up, the thermal properties of

NCD are to the large degree determined by the grain bound-

aries and size over the large temperature range. In MCD

lms characterized by much larger grain size, the thermal

Kapitza resistance at the lm interfaces also plays an impor-

tant role at low temperature.

FIG. 4. Thermal conductivities as a function of temperature measured for

The authors are thankful to E. Loubnin for XRD analysis

microcrystalline Poly diamond and undoped NCD 0 and nitrogen-doped

NCD 25 nanocrystalline diamond lms. The solid lines are simulation of the NCD samples. The work at UCR was supported, in

results obtained using the Callaway, phonon-hopping, and minimum thermal

part, by MARCO Center on Functional Engineered Nano

conductivity models.

Architectonics FENA, and NSF through an award to

A.A.B. The work at GPI was supported by the Russian Min-

istry of Science and Education through Contract No.

bulk crystals. In our calculations, we used bulk scattering

02.445.11.7353.

parameters from Ref. 5. This curve presents the upper bound

for K, which can be obtained in the crystalline diamond with 1

D. M. Gruen, Annu. Rev. Mater. Sci. 29, 211 1999 .

given parameters without boundary effects on phonons. Sub- 2

W. S. Yang, O. Auciello, J. E. Butler, W. Cai, J. A. Carlisle, J. E. Gerbi, D.

stantial deviation of the Callaway-model curve from the M. Gruen, T. Knickerbocker, T. L. Lasseter, J. N. Russell, L. M. Smith,

measured data for MCD at low T suggests strong phonon and R. J. Hamers, Nat. Mater. 1, 253 2002 .

3

scattering from the grain boundaries and substantial Kapitza O. A. Williams, S. Curat, J. E. Gerbi, D. M. Gruen, and R. B. Jackman,

thermal boundary resistance19 at the MCD lm interfaces.20 Appl. Phys. Lett. 85, 1680 2004 .

4

T. Zimmerman, M. Kubovich, A. Denisenko, K. Janioschovsky, O. A.

The low bound for K in carbon-based material is pro- Williams, D. M. Gruen, and E. Kohn, Diamond Relat. Mater. 14, 416

vided by the minimum thermal conductivity MTC ap- 2005 .

5

proach based on Einstein s model for the speci c heat and R. Berman, P. R. W. Hudson, and M. Martinez, J. Phys. C 8, L430 1975 .

6

A. V. Sukhadolau, E. V. Ivakin, V. G. Ralchenko, A. V. Khomich, A. V.

the assumption of the random walk between localized exci-

tations in the disordered material.21 MTC model is conven- Vlasov, and A. F. Popovich, Diamond Relat. Mater. 14, 589 2005 .

7

E. Woerner, C. Wild, W. Mtiller-Sebert, R. Locher, and P. Koidl, Diamond

tionally used for disordered materials. One can see that the Relat. Mater. 5, 688 1996 .

MTC gives K values, which are an order of magnitude below 8

J. E. Graebner, J. A. Mucha, L. Seibles, and G. W. Kammlott, J. Appl.

the measured data, which is reasonable since the grains in Phys. 71, 3143 1992 .

9

D. M. Gruen, S. Liu, A. R. Krauss, and X. Pan, J. Appl. Phys. 75, 1758

NCD are crystalline inside and the lm is only partially dis-

1994 .

ordered. 10

A. Vlasov, V. Ralchenko, S. Gordeev, D. Zakharov, I. Vlasov, and P.

The theoretical approach, which gives the best agree- Belobrov, Diamond Relat. Mater. 9, 1104 2000 .

ment with the measured temperature dependence of thermal 11

S. Ahmed, R. Liske, T. Wunderer, M. Leonhardt, R. Ziervogel, C. Fansler,

conductivity in NCD lms, is the phonon-hopping model T. Grotjohn, J. Asmussen, and T. Schuelke, Diamond Relat. Mater. 15,

PHM .22 PHM has been proposed for polycrystalline mate- 389 2006 .

12

W. L. Hsu, D. M. Tung, E. A. Fuchs, K. F. McCarty, A. Joshi, and R.

rials. It assumes that the phonon transport inside the grain Nimmagadda, Appl. Phys. Lett. 55, 2739 1989 .

follows bulk rules while the phonon transition from one 13

S. Bhattacharyya, O. Auciello, J. Birrell, J. A. Carlisle, L. A. Curtiss, A.

grain to another, i.e., hopping, is characterized by the trans- N. Goyette, D. M. Gruen, A. R. Krauss, J. Schlueter, A. Sumant, and P.

parency parameter t. In general, t is obtained from tting to Zapol, Appl. Phys. Lett. 79, 1441 2001 .

14

G. Z. Wang, F. Ye, C. Chang, Y. Liao, and R. C. Fang, Diamond Relat.

experimental data, although in simplest cases it can be cal-

Mater. 9, 1712 2000 .

culated from rst principles.22 The best- t PHM curves for 15

D. G. Cahill, Rev. Sci. Instrum. 61, 802 1990 .

the undoped NCD d = 22 nm and MCD d = 2 m were 16

W. L. Liu and A. A. Balandin, Appl. Phys. Lett. 85, 5230 2004 ; J. Appl.

obtained with t = 0.32 and t = 0.9, respectively, while the best- Phys. 97, 073***-**** ; M. Shamsa, W. L. Liu, A. A. Balandin, and J. L.

Liu, Appl. Phys. Lett. 87, 202***-**** .

t curve for doped NCD d = 26 nm was calculated with t 17

J. Zou, D. Kotchetkov, A. A. Balandin, D. I. Florescu, and F. H. Pollak, J.

= 0.2. One can see an excellent agreement between the theory Appl. Phys. 92, 2534 2002 ; Appl. Phys. Lett. 79, 4316 2001 .

and experiment for NCD lms. The smaller t value for the 18

J. Callaway, Phys. Rev. 113, 1046 1959 .

19

doped NCD can be related to the fact that nitrogen dopants P. L. Kapitza, J. Phys. Moscow 4, 181 1941 .

20

E. T. Swartz and R. O. Pohl, Appl. Phys. Lett. 51, 2200 1987 .

cluster on the grain boundaries, which is in line with obser- 21

D. G. Cahill and R. O. Pohl, Solid State Commun. 70, 927 1989 .

vations reported in Ref. 13. Accumulation of dopant atoms 22

L. Braginsky, N. Lukzen, V. Shklover, and H. Hofmann, Phys. Rev. B 66,

on the boundary and thicker grain boundaries composed of 134***-**** .

nondiamond carbon23 reduces the boundary transparency and 23

J. Birrell, J. A. Carlisle, O. Auciello, D. M. Gruen, and J. M. Gibson,

increases thermal resistance of NCD. It is interesting to note Appl. Phys. Lett. 81, 2235 2002 .

Downloaded 07 Dec 2006 to 138.23.213.224. Redistribution subject to AIP license or copyright, see http://apl.aip.org/apl/copyright.jsp

to 138.23.213.224. Redistribution subject to AIP license or copyright, see http://apl.aip.org/apl/copyright.jsp

Contact this candidate