Materials and Design 63 (2014) 719–728

Contents lists available at ScienceDirect

Materials and Design

journal homepage: www.elsevier .com/locate /matdes

Fabrication, interface characterization and modeling of oriented graphiteflakes/Si/Al composites for thermal management applications

http://dx.doi.org/10.1016/j.matdes.2014.07.0090261-3069/� 2014 Elsevier Ltd. All rights reserved.

⇑ Corresponding author. Tel./fax: +86 21 5474 7597.E-mail address: zhe.chen@sjtu.edu.cn (Z. Chen).

Cong Zhou a, Gang Ji b,c, Zhe Chen a,⇑, Mingliang Wang a, Ahmed Addad b, Dominique Schryvers c,Haowei Wang a

a State Key Laboratory of Metal Matrix Composites, Shanghai Jiao Tong University, Shanghai 200240, Chinab Unité Matériaux et Transformations, UMR CNRS 8207, Bâtiment C6, Université Lille 1, 59655 Villeneuve d’Ascq, Francec Electron Microscopy for Materials Science (EMAT), University of Antwerp, Groenenborgerlaan 171, 2020 Antwerp, Belgium

a r t i c l e i n f o

Article history:Received 15 May 2014Accepted 5 July 2014Available online 17 July 2014

Keywords:CompositesGraphite flakeThermal conductivityInterface structureAmorphous

a b s t r a c t

Highly thermally conductive graphite flakes (Gf)/Si/Al composites have been fabricated using Gf, Si pow-der and an AlSi7Mg0.3 alloy by an optimized pressure infiltration process for thermal management appli-cations. In the composites, the layers of Gf were spaced apart by Si particles and oriented perpendicular tothe pressing direction, which offered the opportunity to tailor the thermal conductivity (TC) and coeffi-cient of thermal expansion (CTE) of the composites. Microstructural characterization revealed that theformation of a clean and tightly-adhered interface at the nanoscale between the side surface of the Gf

and Al matrix, devoid of a detrimental Al4C3 phase and a reacted amorphous Al–Si–O–C layer, contributedto excellent thermal performance along the alignment direction. With increasing volume fraction of Gf

from 13.7 to 71.1 vol.%, the longitudinal (i.e. parallel to the graphite layers) TC of the compositesincreased from 179 to 526 W/m K, while the longitudinal CTE decreased from 12.1 to 7.3 ppm/K (match-ing the values of electronic components). Furthermore, the modified layers-in-parallel model better fittedthe longitudinal TC data than the layers-in-parallel model and confirmed that the clean and tightly-adhered interface is favorable for the enhanced longitudinal TC.

� 2014 Elsevier Ltd. All rights reserved.

1. Introduction

In the microelectronics industry, the ever-increasing powerdensity and continuing miniaturization of semiconductor andoptoelectronic devices lead to mounting thermal managementchallenges [1,2]. In order to realize high performance, long lifeand high reliability, it is imperative to develop very efficient elec-tronic packaging materials with high thermal conductivity (TC)and low coefficient of thermal expansion (CTE). The TC should beas high as possible to effectively dissipate heat generated fromelectronic components, while the CTE needs to be lower than10 ppm/K in order to minimize thermal stress due to the CTE mis-match between electronic components (4–7 ppm/K) and packagingmaterials [3–10]. Towards this goal, both single phase and compos-ite systems have been investigated. Pure metals including Cu andAl have relatively high TCs, but their CTEs of 17 and 23 ppm/K,respectively, are much higher than those of the electronic compo-nents [2,10]. Mo and W reinforced Cu matrix composite candi-

dates, despite their low CTEs of 6–7 ppm/K being compatiblewith those of electronic components, have relatively low TCs ofaround 200 W/m K [10–13]. Two other composites have been con-sidered, but they also failed to meet expectations. One is SiC/Alcomposites where the main drawback is again its insufficient TCof 170–250 W/m K [2,8,14]. The other one is diamond/metal (Al,Cu or Ag) composites which exhibits excellent thermal propertieswith TCs approaching 600 W/m K [12,15,16]. However, applicationof the latter has been limited to niche markets due to the high priceof diamond particles; moreover, poor machinability of the dia-mond/metal composites has been an additional limitation [8].

To date, several further carbon-based materials (e.g., carbonfibers [17,18], carbon nanotubes (CNT) [19–22], graphite particles[18,23,24], porous graphite performs [25,26], graphite foam[18,27], graphene [28,29] etc.) have been considered and inte-grated in metal matrices via various fabrication techniques. Graph-ite flakes (Gf), in particular, have attracted significant recentattention for thermal management applications due to their supe-rior thermal properties, low cost and ease of machining[1,18,30,31]. However, stacking of Gf leaves almost no spacebetween the flakes, which makes liquid metal infiltration underpressure an almost unfeasible technique for fabrication of, for

720 C. Zhou et al. / Materials and Design 63 (2014) 719–728

example, Gf/Al composites [18]. To address this issue, Prieto et al.[31] introduced a small proportion of SiC particles which actedas spacers between the layers of Gf. As a result, the produced Gf/SiC/AlSi12 composites exhibited alternating layers of Gf and hadTC in the range of 294–390 W/m K.

Highly crystalline graphite has extremely high TC and near-zeroCTE along the basal plane of its crystal lattice [17,32]. Consideringthis high anisotropy in thermal properties parallel and perpendic-ular to the basal plane of Gf, control of the preferred orientation ofGf is crucial since it is one of the main factors enhancing the TC ofGf/Al composites [1]. A further concern is the graphite/Al interfaceconfigurations, which should facilitate thermal exchange acrossthe interface by providing good adhesion and minimum thermalboundary resistance. However, the influence of interface formationand diffusion on the interface conductance is still not fully under-stood. It has been well documented that the excessive formation ofaluminum carbide (Al4C3) phase at the interface hinders thermaltransfer and degrades mechanical properties [26]. In this work,we report fabrication and interface characterization of orientedGf/Si/Al composites by room-temperature mixing and pressing fol-lowed by an optimized pressure infiltration process. The micro-structure of the composites is characterized at length scales fromthe macro to the nanoscale to investigate presence of Al4C3 in par-ticular and more generally the detailed graphite/Al interfaceassembly configuration. The effect of the volume fractions of Gf

on thermal properties (TC and CTE) of the composites is also inves-tigated and discussed. Furthermore, efforts are devoted to presentpredictive models for fitting experimental TC values of the orientedGf/Si/Al composites by considering the interfacial features revealedat the nanoscale.

2. Experimental details

2.1. Fabrication of the oriented Gf/Si/Al composites

The raw materials consisted of Gf, Si particles and Al alloy (AlSi7-

Mg0.3) ingots. The Gf was acquired from Shangdong Graphite Co.,Ltd., China. The Si particles (99.9% purity) were supplied by JiangsuMaoyuan Co., Ltd., China. The thermo-physical properties of theseraw materials are given in Table 1. The main reasons for using Siinstead of SiC particles are (i) Si particles are more easily machin-able and have a lower density of 2.33 g/cm3 than SiC (3.2 g/cm3)and (ii) SiC suffers from its severe reactivity with molten Al [33].AlSi7Mg0.3 alloy was used for infiltration due to its good fluidityas compared to pure Al.

Fig. 1 illustrates the typical fabrication process of the Gf/Si/Alcomposites used in this work. The three key steps were: (1) mixingof Gf with Si particles; (2) pressing the mixture into a block and (3)pressure infiltration. In particular, Gf and Si particles with volumeratios of 1:4, 1:2, 1:1, 2:1 and 4:1 were mixed by mechanical stir-ring for 30 min. The mixtures were then loaded in a steel mouldand uniaxially pressed to obtain a block in a pressing apparatususing a pressure of 50 MPa. As will be illustrated by Optical Micros-copy (OM) images below, this step can space and orient the layersof Gf. The block was then heated up to 400 �C. The AlSi7Mg0.3 alloy

Table 1Thermo-physical properties of the raw materials used in this work.

Raw material q (g/cm3) Cp (J/g/K) TC (W/m K) CTE (ppm/K)

AlSi7Mg0.3 2.685 0.89 151 23.5Si 2.33 0.70 150 4.1Gf 2.26 0.71 1000a [17] �1.5a [32]

38b [31] 25b [34]

a Parallel to the graphite layers.b Perpendicular to the graphite layers.

(melting temperature of around 581 �C) was heated to 760 �C,degassed and cleaned in a graphite crucible. Subsequently, themolten Al alloy was poured onto the block, and a pressure of100 MPa was applied for 60 s to force the molten Al alloy to fillthe pores inside the block.

It has been reported that a relatively low infiltration tempera-ture and a high cooling rate can suppress the formation of theinterfacial Al4C3 phase [25,26,35]. This is what led us to use themoderate 760 �C as an optimal temperature preventing Al4C3 for-mation on the one hand, yet facilitating infiltration of the AlSi7-

Mg0.3 alloy on the other hand. At this temperature, better fluidityof the AlSi7Mg0.3 alloy with respect to pure Al (melting tempera-ture of around 660 �C versus 581 �C for the alloy) can be correlatedwith the lower melting temperature of the former. It is well knownthat an external pressure is usually applied to facilitate infiltrationof molten Al alloy into the pores inside the block [23]. Too high apressure however can hinder infiltration by making the infiltrationchannels too small, while too low a pressure requires longer infil-tration times at high temperature which may favor Al4C3 forma-tion. Thus, in order to maximize infiltration and prevent Al4C3

formation, an optimized infiltration pressure of 100 MPa wasemployed based on our experience with our homemade facility.

2.2. Microstructural characterization

The microstructure was examined using OM and Scanning Elec-tron Microscopy (SEM, JEOL JSM-7600F). The formed Gf/Si/Al com-posites were cut and mechanically polished. The volume fraction ofGf (VGf

) was determined with image analysis software (Image-ProPlus 6.0) based on a grey-scale threshold criterion. Table 2 displaysthermo-physical properties of the Gf/Si/Al composites with differ-ent VGf

. The volume fraction of Si particles (VSi) in the Si/Al matrixis also listed in Table 2.

X-Ray Diffraction (XRD) measurements were performed on a D8ADVANCE X-Ray diffractometer, operating at 40 kV/40 mA andusing Cu Ka radiation (k = 0.15406 nm). 2h scans were performedbetween 20� and 70� with a scan speed of 5�/min. Particular carewas taken in checking for the presence of Al4C3 (JCPDS file 65-9731). For this, careful scanning was performed in the angle rangesof 30.5–33.0� and 39.5–41.0� with a step size of 0.005�. The scantime was 5 s per increment. Degree of graphitization of initial Gf

was determined by the JSPS method [36] where standard silicon(200 Mesh, 99.99%, Soekawa Chemical Co., Japan) was used forcalibration.

The Gf/Si/Al composite (55.3% in VGf, see Table 2) was further

analyzed by (High-Resolution) Transmission Electron Microscopy((HR)TEM) in order to examine the graphite/Al interfaces, beingparallel and perpendicular to the alignment direction along thexy-plane of Gf. TEM thin foils were cut parallel to the alignmentdirection, mechanically polished and finally thinned using a GatanModel 691 precision ion polishing system. A FEI Tecnai G2-20 Twinmicroscope operated at 200 kV and equipped with a Gatan Energy-Dispersive X-ray Spectroscopy (EDX) analyzer was used for(HR)TEM examination. Selected Area Electron Diffraction (SAED)was performed for phase identification. Scanning TEM mode witha probe size of a few nanometers was used for EDX analyses. ForHRTEM, significant sample tilting was avoided in order not to over-lap the lattice contrasts from Gf and Al matrix at the interface.

2.3. Measurements of the thermo-physical properties

The TC (k) of the composites was calculated from the thermaldiffusivity (a), density (q) and specific heat capacity (Cp) usingEq. (1) [1]:

k ¼ a � Cp � q ð1Þ

Fig. 1. Schematic of the fabrication process for preparing the Gf/Si/Al composites.

Table 2Thermo-physical properties of the obtained Gf/Si/Al composites.

Sample VGf(%) VSi

1�VGf(%) q (g/cm3) Cp (J/g/K) a (mm2/s) TC (W/m K) CTE (ppm/K)

1 13.7 66.1 2.43 0.767 96a 179a 12.1a

49b 90b 11.5 b

2 24.0 65.7 2.41 0.761 145 a 266 a 11.4 a

45 b 83 b 9.7 b

3 38.5 65.2 2.38 0.753 195 a 350 a 10.0 a

40 b 72 b 7.5 b

4 55.3 64.6 2.35 0.742 272 a 475 a 9.1 a

26 b 45 b –5 71.1 64.3 2.32 0.731 310 a 526 a 7.3 a

24 b 41 b –

a Parallel to the graphite layers.b Perpendicular to the graphite layers.

C. Zhou et al. / Materials and Design 63 (2014) 719–728 721

The laser-flash diffusivity instrument (Nano-Flash-Apparatus,LFA 447, NETZSCH) was used to determine the thermal diffusivityof the cylindrical specimen (£ 12.6 � 2 mm) at room temperature.The specific heat capacity of the composites was calculated by thelinear rule of mixtures (ROM) using the specific heat capacity val-ues of the components (i.e. Gf, Si particles, and Al alloy) [31]. TheCTE values of the specimens (£ 5 � 25 mm) were characterizedwith a dilatometer (DIL 402C, NETZSCH) from room temperatureup to 300 �C. As shown in Table 2, the CTE values perpendicularto the xy-plane of the prepared samples No. 4 and No. 5 werenot measured due to sample size limitations.

2.4. Modeling approaches

Layers-in-parallel [37] and layers-in-series models [38] wereused to calculate longitudinal (i.e. along the xy-plane, see Fig. 1)and transversal (i.e. along the z-axis) TC of the oriented Gf/Si/Alcomposites, respectively. The layers-in-parallel model considersheat-transfer characteristics parallel to the major heat flow direc-tion of layered composites, while the layers-in-series model con-siders those perpendicular to the major heat flow direction. Themodels are described as follows:

Layers-in-parallel model:

KcL ¼ fKL þ ð1� f ÞKm ð2Þ

Layers-in-series model:

1Kc

T

¼ fKTþ 1� f

Kmð3Þ

where Kc , K and Km refer to the TC of composites, Gf and matrix,respectively; the subscripts L, T denote the longitudinal and trans-versal directions, and f is the volume fraction of Gf. It should benoted that in reality both simple models could yield a certaindegree of overestimation as the important effect of interfacial ther-mal resistance on TC is not taken into account. Thus, as will bedetailed below, we modified the models in order to improve theirpredictive capacity.

3. Results and discussion

3.1. Microstructures of the starting materials and the oriented Gf/Si/Alcomposites

Fig. 2 illustrates the morphologies of Gf and Si particles used inthis work. The Gf displays neat platelet morphology with an aver-age diameter of a plate of about 500 lm (see Fig. 2a) and a meanthickness of about 20 lm (see inset in Fig. 2a). The Si particles haveirregular shape with a mean particle size of 10 lm (see Fig. 2b). Thedegree of graphitization of the initial Gf is estimated to be higherthan 99% by the JSPS method from Fig. 3a [36]. As is well known,a well arrayed graphite structure is especially important for a highTC of the graphite materials [30]. Such a high degree of graphitiza-tion indicates that the Gf are highly crystalline and have a very highTC along the basal plane of the crystal lattice.

As shown by the OM images of the obtained Gf/Si/Al compositesin Fig. 4a–e, the elongated dark regions represent the Gf while thelight ones are the Si/Al matrix. No pores are visible, indicating a

Fig. 2. SEM micrographs showing the morphologies of (a) as-received Gf and (b) Si particles used in this work.

Fig. 3. XRD patterns of (a) the initial Gf (Si peaks originate from a small amount of pure Si powder intentionally added as a calibration standard) and (b) the Gf/Si/Al compositewith a Gf volume fraction of 55.3%; the inserts in (b) are careful scanning XRD traces in the angle ranges 30.5–33.0� and 39.5–41.0� to highlight possible but not observed(101), (012) and (107) diffraction peaks of the Al4C3 phase.

722 C. Zhou et al. / Materials and Design 63 (2014) 719–728

fully densified assembly. The layers of Gf are stacked mostly paral-lel to each other and oriented perpendicular to the pressing direc-tion (i.e. the z-axis, see Fig. 1). Bending of the Gf is morepronounced in the sample with high Gf volume (Fig. 4e). Fig. 4f isa SEM micrograph showing the Al alloy, the Si particles and thesharp interface between the graphite and the Si/Al matrix.

The formation of Al4C3 at the graphite/Al interface is thermody-namically favorable (e.g. �168 kJ/mol at 750 �C [39]) and is oftenobserved in graphite/Al composites produced by gas pressure infil-tration [26]. Thus, in order to further test for its presence as well asthat of other interfacial reaction products, the composites werealso examined at the macroscopic scale by XRD (see Fig. 3b) andat the nanoscale by (HR)TEM (see Fig. 5 and 6). The XRD patternsof all the obtained composites look similar. A representative onein Fig. 3b shows only the diffraction peaks of Al, Si and graphite.No evidence of Al4C3 indicates that this phase was not formed toany significant extent at the interface. What is revealed by(HR)TEM (Fig. 5) is an interfacial layer homogeneously distributedalong the Gf/Al interface, evidently binding the graphite and the Almatrix. This layer has a thickness of 40–50 nm (Fig. 5a) and is richin the elements C, O, Al and Si (Fig. 5f). An SAED pattern from thelayer is shown in Fig. 5b. The central diffuse ring provides evidenceof an amorphous phase, with as of yet unindexed crystals (evi-denced by the two accompanying sharp rings shown in Fig. 5b)embedded within the amorphous phase. The 020 reflection (d0 2 0

spacing of around 0.2 nm) of the indexed [105] Zone Axis Pattern(ZAP) of Al is located on one of the arrowed rings. This ring canhence be attributed to graphite since the d1 0 1 spacing of graphiteis also around 0.2 nm. The HRTEM image of graphite in Fig. 5c

shows the (002) basal planes (d0 0 2 spacing of around 0.34 nm),which are buckled with clearly visible dislocation-like defects.The morphology of the interfacial layer is shown with higher reso-lution in Fig. 5d. The nanocrystals are less than 5 nm in nominalsize and tend to distribute themselves adjacent to the interfaceswith the Al matrix and the Gf. One examined nanocrystal (pin-pointed by an arrow in Fig. 5e) exhibits lattice fringes with a d-spacing of roughly 0.19 nm (the resolution limit of themicroscope).

The interface between the side surface of Gf and Al matrix (i.e.side surface of Gf/Al interface) is represented in Fig. 6. As Gf onlypresents very high TC along its basal plane, this interface configu-ration is thus the key to determine the overall thermal propertiesof the oriented composites. Two interfaces nearly located at thetip of a graphite layer, as indicated in Fig. 6a, were characterizedby HRTEM shown in Fig. 6b–e. Completely different from theGf/Al interface previously revealed in Fig. 5, no reaction productsare detected along the whole interfaces. Only the amorphous car-bon layers of Gf, confirmed by the corresponding FFT (inset inFig. 6c) and with a thickness of around 5–8 nm, are observed totightly adhere to the Al matrix (highlighted by dotted lines inFig. 6b and c). Structural evolution is also visible in the layer ofgraphite, generally from the crystalline areas slightly away fromthe interfaces (Fig. 6d and e) to the poorly crystalline areas adja-cent to the amorphous layers (low parts of Fig. 6b and c). Anentirely turbostratic characteristic shown in Fig. 6d and e isproduced by bending, buckling and breaking of the (002) basalplanes. Comparatively, only a few basal planes can be very locallyidentified in Fig. 6b and c.

Fig. 4. OM micrographs (magnification of 100�) of the Gf/Si/Al composites with different Gf volume fractions: (a) 13.7%, (b) 24%, (c) 38.5%, (d) 55.3%, (e) 71.1%; and (f) atypical SEM micrograph of the graphite/Al interface.

C. Zhou et al. / Materials and Design 63 (2014) 719–728 723

3.2. Thermal properties and modeling

As shown in Table 2, the measured longitudinal TC of the ori-ented composites increases from 179 to 526 W/m K with anincrease of the VGf

from 13.7 to 71.1 vol.%, while the transversalTC decreases from 90 to 41 W/m K. Theoretical predictions of thelongitudinal and transversal TC, by means of the layers-in-paralleland the layers-in-series models, are plotted versus the experimen-tal data in Fig. 7a and b, respectively. The layers-in-parallel modelincreasingly overestimates the TC of the Gf/Si/Al composites withincreasing volume fraction of Gf (see Fig. 7a). In order to take intoaccount the effect of interfacial thermal resistance, the Gf can beconsidered to be coated with a very thin interfacial thermal barrierlayer [40]. Analytically, it can be solved by replacing the inclusionwith a non-ideal interface by an ‘‘effective’’ inclusion having aneffective TC, Keff , and can be expressed as

KeffL ¼

KL

1þ 2KLhD

ð4Þ

KeffT ¼

KT

1þ 2KTht

ð5Þ

where h is the interface thermal conductance (i.e. the reciprocal ofthe interfacial thermal resistance), t and D denote the thickness anddiameter of the inclusion. The value of h can be calculated by theacoustic mismatch model (AMM) [41], which treats the interfaceheat transfer in terms of continuum mechanics by calculating theprobability of an incident phonon traversing the interface [42]. his calculated to be:

h ffi 12qmCm

v3m

v2i

qmvmqiv i

ðqmvm þ qiv iÞ2ð6Þ

where q, C, v are the density, the specific heat capacity, and thephonon velocity, respectively; subscripts m, i denote the matrixand the inclusion, respectively.

Considering the fact that no reactions occurred at the side sur-face of Gf/Al interface, we assume that the influence of the amor-phous carbon layer on the TC of the composites is negligible

Fig. 5. Structural, chemical and morphological analyses of the Gf/Al interface by (HR)TEM: (a) an overview TEM bright-field (BF) image of the Gf/Al interface (interfacialboundaries are demarcated by dotted lines), (b) SAED pattern of the interfacial area showing the overlapped [105] ZAP of Al, a diffuse amorphous and a well-defined(arrowed) ring pattern, (c) HRTEM lattice image of graphite with corresponding Fast Fourier Transform (FFT) as inset, (d) highlighted BF image of the selected area in (a)showing the interfacial layer, (e) HRTEM lattice image showing nanosized crystals in the interfacial layer; (f), (g) and (h) are EDX spectra of the interfacial layer, Al matrix andgraphite, respectively. Cu peaks in (f) originate from the TEM sample holder.

724 C. Zhou et al. / Materials and Design 63 (2014) 719–728

since the layer thickness is much smaller (5–8 nm, see Fig. 6b andc) than the diameter of Gf (500 lm, see Fig. 2a). Thus, the side sur-face of Gf/Al interface can be regarded as the simple Gf/AlSi7Mg0.3

one and the interface thermal conductance can be calculated usingEq. (6). Given the lack of data for AlSi7Mg0.3, pure Al is considered

here for theoretical calculations. The material parameters for thecalculation are given in Table 3. We obtain h = 4.5 � 107 W/m2 K,which is very close to the experimental data (�5 � 107 W/m2 K)measured by Chen et al. [45]. The effective TCs of the Gf

parallel and perpendicular to the basal plane are calculated from

Fig. 6. Structural and morphological analyses of the side surface of Gf/Al interface by (HR)TEM: (a) a montage of TEM low-magnification BF images showing the side surface ofGf/Al interface, (b) and (c) HRTEM lattice images of the interfaces in the areas numbered 1 and 2 in (a), (d) and (e) HRTEM lattice images of the corresponding graphite layersaway from the interfaces shown in (b) and (c), respectively. Insets in (b), (c) and (d) are corresponding FFT patterns.

C. Zhou et al. / Materials and Design 63 (2014) 719–728 725

Eqs. (4) and (5) to be 918 and 35 W/m K, respectively. Hence, wereplace K in layers-in-parallel model and layers-in-series modelby Keff . The modified layers-in-parallel model and modified lay-ers-in-series model are expressed as:

KcL ¼ fKeff

L þ ð1� f ÞKm ð7Þ

1Kc

T

¼ f

KeffT

þ 1� fKm

ð8Þ

Fig. 7. Comparison of experimental data with theoretical predictions: (a) the longitudinal TC and (b) the transversal TC of the composites with different Gf volume fractions.

Table 3Material parameters for theoretical calculation.

Material Density(kg/m3)

Specific heat(J/kg/K)

Phononvelocity (m/s)

Ref.

Graphite 2260 710 14,800 [1,43]Aluminum 2700 895 3595 [41,44]

Fig. 8. CTE values of the composites with different Gf volume fractions.

726 C. Zhou et al. / Materials and Design 63 (2014) 719–728

Km can be obtained by experimentally measuring the TC of the Si/Alcomposite materials. As shown in Table 2, the volume fractions of Siparticles in the matrix of the different composites vary in a narrowrange between 64.2 and 66.1 vol.%. As a result, the average Km ofaround 104 W/m K measured in the Si/Al composite material con-taining 65 vol.% of Si particles was used for the calculations. Thelongitudinal TC data predicted by the modified layers-in-parallelmodel are given in Fig. 7a. The modified model clearly has a betterpredictive capacity as compared to the original one. The remainingdeviation may be attributed to the bending effect of the Gf since theGf itself has weak mechanical strength and is easily bent (see Fig. 4),which could deflect flow of heat [32]. The amount of curved flakesincreases with increasing volume fraction of Gf, leading to a rela-tively significant discrepancy at very high graphite contents asshown in Fig. 7a.

In the same way, the predictions of the transversal TC by meansof the layers-in-series and modified layers-in-series models aregiven in Fig. 7b. The transversal TC of the composites decreaseswith increasing volume fraction of Gf since the transversal TC ofGf (38 W/m K) is much lower than the matrix TC (104 W/m K). Inaddition, it is well known that TC of the amorphous phase is lowerthan that of its crystalline counterpart. Thus, the interfacial ther-mal resistance along this direction should be more pronouncedthan that along the longitudinal direction due to the presence ofthe reacted amorphous Al–Si–O–C layer, embedded with nanocrys-tals and with a thickness of 50 nm, at the Gf/Al interface (see Fig. 5).However, the fact that the predictive capacity is not improved byusing the modified model shows that the interfacial thermal resis-tance has only limited effect on the transversal TC of the compos-ites (see Fig. 7b).

The CTE values of the composites parallel and perpendicular tothe basal plane of Gf are summarized in Table 2 and plotted inFig. 8. Generally, the CTE values decrease with an increase of theGf content. The results also reveal the anisotropic CTE of the com-posites, as evidenced by the CTE values parallel to the xy-planealways being higher than those perpendicular to the xy-plane. Con-sidering the intrinsic CTE values of the Gf (�1.5 ppm/K parallel,versus 25 ppm/K perpendicular to the basal plane, see Table 1),the results seem counterintuitive. All efforts to fit the data using,for example, the ROM [32], Turner model [46], Kerner model

[47], and Schapery model [48] have failed. In the literature, Piretoet al. [31] also reported the same difficulties in modeling the exper-imental CTE values of the Gf/SiC/AlSi12 composites. The underlyingreason for the unusual thermal expansion behaviour of the com-posites requires further investigation but is out of the scope of thispaper.

3.3. Discussion of thermal properties and interface configurations

Based on the microstructures (see Figs. 2–6), the obtainedenhanced longitudinal TC of the Gf/Si/Al composites can originatefrom the following reasons. On the one hand, the oriented arrange-ment along the basal plane of highly crystalline Gf is very favorablefor global thermal transfer along the alignment direction. Addition-ally, we find that Gf stack, upon packing, with their basal planesparallel to each other and simple mixing with Si particles sufficesfor inducing requisite separations between the flakes (see Fig. 1).On the other hand, at the nanoscale the clean and tightly-adheredinterface (see Fig. 6) between the side surface of Gf and Al matrix,instead of the reacted amorphous layer (see Fig. 5) and detrimentalAl4C3 [26] is also favorable. As described above, the optimizedpressure infiltration processing conditions of moderate tempera-ture and pressure used in this work effectively suppress the forma-tion of Al4C3. The segregation of Si along the graphite/Al interface(see Fig. 5f), originating from the Al alloy with 7 wt% Si, is benefi-cial since small quantities of Si hinder C solubility in the Al matrixhence limiting Al4C3 formation by interfacial reaction [26,49]. Useof Gf with a high degree of graphitization (Fig. 3a) is an additionalfactor for hindering the nucleation of Al4C3 due to a lack of surfacedefects [26]. Interfacial amorphous layers have been previouslycharacterized by (HR)TEM in squeeze cast diamond/Al [50], spark

C. Zhou et al. / Materials and Design 63 (2014) 719–728 727

plasma sintered CNT/Cu [51,52] and hot-pressed vapor growncarbon fiber/Al composites [53]. In our case, decomposition andtransformation of graphite surfaces into amorphous carbon layerscan occur at the high temperature of our experiments. Incorpora-tion of O, Si and Al, most probably due to oxidation and interfacialreaction, tend to promote amorphous aluminosilicon oxycarbide(Al–Si–O–C) formation [54]. It was reported that the formed amor-phous phase may also act as the nucleation site for Al4C3 [50].Essentially, energy transfer between electrons in aluminum andphonons in graphite dominates heat conduction across the graph-ite/Al interface. The presence of the reacted interfacial amorphouslayer cannot facilitate this energy exchange due to strong phononscattering in the amorphous phase. This process may further bedissipated by the presence of interfacial oxygen atoms leading todegradation of the TC of the composites [55]. Nevertheless, as thistype of interface is parallel to the alignment direction of the com-posites, it only contributes to the transversal TC. Theoretical calcu-lations by means of the (modified) layers-in-series models alsoindicate that the effect of the amorphous layer on the transversalTC is very limited.

Furthermore, the side surface of Gf/Al interface configuration –a clean and well adhered interface – is beneficial for the improvedTC along the alignment direction of the composites. The consis-tency of the experimental TC data and the predictive calculationby the modified layers-in-parallel model also confirms that thepresence of an extremely thin carbon amorphous layer does notsubstantially degrade the longitudinal TC. Although it was reportedthat the TC of the amorphous carbon film is as low as about 0.2 W/m K [56], our results suggest that the negative effect of strong pho-non scattering in the amorphous carbon layer on interfacial ther-mal resistance can be marginal when the layer is only of a fewnanometers in thickness.

Finally, as revealed by OM and TEM in Figs. 4 and 6, the graphiteflakes incorporated in the composites are not uniform in thickness.This thickness variation as well as the aspect ratio of the graphiteflakes can have an effect on thermal conductivity of the compositesand deserves a systematic study.

4. Conclusions

Gf/Si/Al composites with high TC and low CTE have been fabri-cated by a pressure infiltration process using optimized conditions.Microstructure and in particular the graphite/Al interface configu-rations, were characterized using XRD, OM, SEM and (HR)TEM. Theevolution of the thermal properties of the composites as a functionof the volume fraction of Gf was studied. The results allow us todraw the following conclusions:

(1) A successful processing route involved (i) mixing Gf with theSi particles and alignment of Gf under pressure at room tem-perature and (ii) infiltration of melted AlSi7Mg0.3 alloy at amoderate temperature of 760 �C and pressure of 100 MPafor 60 s.

(2) Enhanced longitudinal TC resulted from the followingmicrostructural characteristics: (i) the Gf being stackedmostly parallel to each other and oriented parallel to thexy-plane and (ii) the clean and tightly-adhered side surfaceof Gf/Al interface, instead of the reacted C–O–Al–Si amor-phous layer and detrimental Al4C3.

(3) The global TC increased from 179 to 526 W/m K was corre-lated with an increase of the volume fraction of Gf from13.7 to 71.1 vol.%, while the CTE in the directions paralleland perpendicular to the xy-plane decreased from 12.1 to7.3 ppm/K and from 11.5 to 7.5 ppm/K, respectively. The fab-ricated Gf/Si/Al composites showed a TC in excess of 500 W/m K and a CTE compatible with those of electronic

components (4–7 ppm/K). Thus this composite and its devel-opment show promise for low-cost thermal managementapplications.

(4) The modified layers-in-parallel model, which considered theinterfacial thermal resistance introduced by the clean andtightly-adhered side surface Gf/Al interface, better predictedthe longitudinal TC data than the original model. Never-theless, based on the (modified) layers-in-series models,the Al–Si–O–C amorphous layer at the Gf/Al interface had alimited effect on the transversal TC.

Acknowledgements

This work is financially supported by the National NaturalScience Foundation of China (Grant No. 51201099), China Postdoc-toral Science Foundation (Grant No. 2012M520891), the founda-tion from the State Key Laboratory of Metal Matrix Composites,Shanghai Jiao Tong University (China) and the FWO project of Bel-gium (No. U2 FA 070100/3506). Many thanks are also due to Dr.George Serghiou, School of Engineering and Centre for MaterialsScience, University of Edinburgh, UK for revisions and discussions.

References

[1] Yuan GM, Li XK, Dong ZJ, Westwood A, Cui ZW, Cong Y, et al. Graphite blockswith preferred orientation and high thermal conductivity. Carbon2012;50(1):175–82.

[2] Mallik S, Ekere N, Best C, Bhatti R. Investigation of thermal managementmaterials for automotive electronic control units. Appl Therm Eng2011;31:355–62.

[3] Song JL, Guo QG, Gao XQ, Shi JL, Liu L. Microstructer and thermophysicalproperties of graphite foam/glass composites. Carbon 2010;49(4):1479–83.

[4] Zweben C. Advanced composites and other advanced materials for electronicpackaging thermal management. In: IMAPS: processes, prop and interfacesproc. Braselton (Georgia, USA); 2001. p. 360–65.

[5] Chung DDL, Zweben C. Composites for electronic packaging and thermalmanagement. In: Kelly A, Zweben C, editors. Comprehensive compositematerials, vol. 6. Amsterdam: Elsevier; 2000. p. 701–25.

[6] Zweben C. Ultrahigh thermal conductivity packaging materials. In: 21st IEEESEMI-THERM symp proc; San Jose (California, USA); 2005. p. 168–74.

[7] Schubert Th, Trindade B, Weißgärber T, Kieback B. Interfacial design of Cu-based composites prepared by powder metallurgy for heat sink applications.Mater Sci Eng, A 2008;475:39–44.

[8] Molina JM, Rhême M, Carron J, Weber L. Thermal conductivity of aluminummatrix composites reinforced with mixtures of diamond and SiC particles.Scripta Mater 2008;58(5):393–6.

[9] Vetterli M, Tavangar R, Weber L, Kelly A. Influence of the elastic properties ofthe phases on the coefficient of thermal expansion of a metal matrixcomposite. Scripta Mater 2011;64(2):153–6.

[10] Zweben C. Advanced thermal management materials for electronics andphotonics. Adv Microelectron 2010;37(4):14–9.

[11] Harper C. Electronic materials and processes handbook. 3rd ed. NewYork: McGraw-Hill Companies; 2004. 404–32.

[12] Ruch PW, Beffort O, Kleiner S, Weber L, Uggowitzer PJ. Selective interfacialbonding Al(Si)-diamond composites and its effect on thermal conductivity.Compos Sci Technol 2006;66(15):2677–85.

[13] Ellsworth MJ. Chip power density and module cooling technology projectionsfor the current decade. In: 9th Inter soc conf on thermal and thermomechphenom in electron syst (I-THERM). Las Vegas (Nevada, USA): 2004. p. 707–8.

[14] Molina JM, Piñero E, Narciso J, Garcia-Cordovilla C, Louis E. Liquid metalinfiltration into ceramic particle preforms with bimodal size distributions.Curr Opin Solid State Mater 2005;9:202–10.

[15] Tan ZQ, Li ZQ, Fan GL, Kai XZ, Ji G, Zhang LT, et al. Enhanced thermalconductivity in diamond/aluminum composites with a tungsten interfacenanolayer. Mater Design 2013;47:160–6.

[16] Beffort O, Khalid FA, Weber L, Ruch P, Klotz UE, Meier S, et al. Interfaceformation in infiltrated Al(Si)/diamond composites. Diam Relat Mater2006;15(9):1250–60.

[17] Ueno T, Yoshioka T, Ogawa JI, Ozoe N, Sato K, Yoshino K. Highly thermalconductive metal/carbon composites by pulsed electric current sintering.Synth Met 2009;159:2170–2.

[18] Prieto R, Molina JM, Narciso J, Louis E. Fabrication and properties of graphiteflakes/metal composites for thermal management applications. Scripta Mater2008;59(1):11–4.

[19] Chu K, Wu QY, Jia CC, Liang XB, Nie JH, Tian WH, et al. Fabrication and effectivethermal conductivity of multi-walled carbon nanotubes reinforced Cu matrix

728 C. Zhou et al. / Materials and Design 63 (2014) 719–728

composites for heat sink applications. Compos Sci Technol2010;70(2):298–304.

[20] Jiang L, Fan GL, Li ZQ, Kai XZ, Zhang D, Chen ZX, et al. An approach to theuniform dispersion of a high volume fraction of carbon nanotubes inaluminum powder. Carbon 2011;49(6):1965–71.

[21] Jiang L, Li ZQ, Fan GL, Cao LL, Zhang D. The use of flake powder metallurgy toproduce carbon nanotube (CNT)/aluminum composites with a homogenousCNT distribution. Carbon 2012;50(5):1993–8.

[22] Deng CF, Ma YX, Zhang P, Zhang XX, Wang DZ. Thermal expansion behaviors ofaluminum composite reinforced with carbon nanotubes. Mater Lett2008;62(15):2301–3.

[23] Rodríguez-Guerrero A, Sánchez SA, Narciso J, Louis E, Rodríguez-Reinoso F.Pressure infiltration of Al-12 wt.% Si–X (X = Cu, Ti, Mg) alloys into graphiteparticle performs. Acta Mater 2006;54(7):1821–31.

[24] Rodríguez-Guerrero A, Molina JM, Rodríguez-Reinoso F, Narciso J, Louis E. Porefilling in graphite particle compacts infiltrated with Al–12 wt.%Si and Al–12wt.%Si–1 wt.%Cu alloys. Mater Sci Eng, A 2008;495:276–81.

[25] Etter T, Kuebler J, Frey T, Schulz P, Löffler JF, Uggowitzer PJ. Strength andfracture toughness of interpenetrating graphite/aluminium compositesproduced by the indirect squeeze casting process. Mater Sci Eng, A2004;386:61–7.

[26] Etter T, Schulz P, Weber M, Metz J, Wimmler M, Löffler JF, et al. Aluminumcarbide formation in interpenetrating graphite/aluminium composites. MaterSci Eng, A 2007;448:1–6.

[27] Zhong YJ, Guo QQ, Li SZ, Shi JL, Liu L. Thermal and mechanical properties ofgraphite foam/Wood’s alloy composite for thermal energy storage. Carbon2010;48(5):1670–92.

[28] Liu XW, Mao JJ, Liu PD, Wei XW. Fabrication of metal-graphene hybridmaterials by electroless deposition. Carbon 2011;49(2):477–83.

[29] Bartolucci SF, Paras J, Rafiee MA, Rafiee J, Lee S, Kapoor D, et al. Graphene-aluminum nanocomposites. Mater Sci Eng, A 2011;528(27):7933–7.

[30] Liu ZJ, Guo QG, Shi JL, Zhai GT, Liu L. Graphite blocks with high thermalconductivity derived from natural graphite flakes. Carbon 2008;46(3):414–21.

[31] Prieto R, Molina JM, Narciso J, Louis E. Thermal conductivity of graphite flakes-SiC particles/metal composites. Composites A 2011;42(12):1970–7.

[32] Chen JK, Huang IS. Thermal properties of aluminum–graphite composites bypowder metallurgy. Composites B 2013;44(1):698–703.

[33] Chung DDL. Materials for thermal conduction. Appl Therm Eng2001;21(16):1593–605.

[34] Tsang DKL, Marsden BJ, Fok SL, Hall G. Graphite thermal expansionrelationship for different temperature ranges. Carbon 2005;43:2902–6.

[35] Leng JF, Wu GH, Zhou QB, Dou ZY, Huang XL. Mechanical properties of SiC/Gr/Al composites fabricated by squeeze casting technology. Scripta Mater2008;59(6):619–22.

[36] Kim T, Ham C, Rhee CK, Yoon SH, Tsuji M, Mochida I. Effects of oxidation andheat treatment of acetylene blacks on their electrochemical double layercapacitances. Carbon 2009;47:226–33.

[37] Truong HV, Zinsmeister GE. Experimental study of heat transfer in layeredcomposites. Int J Heat Mass Trans 1978;21(7):905–9.

[38] Progelhof RC, Throne JL, Ruetsch RR. Methods for predicting the thermalconductivity of composite systems: a review. Polym Eng Sci1976;16(9):615–25.

[39] Lin RY. Interface evolution in aluminum matrix composites during fabrication.Key Eng Mater 1995;104–107:507–22.

[40] Nan CW, Birringer R, Clarke DR, Gleiter H. Effective thermal conductivity ofparticulate composites with interfacial thermal resistance. J Appl Phys1997;81(10):6692–9.

[41] Swartz ET, Pohl RO. Thermal boundary resistance. Rev Mod Phys1989;61(3):605–68.

[42] Molina JM, Prieto R, Narciso J, Louis E. The effect of porosity on the thermalconductivity of Al–12 wt.% Si/SiC composites. Scripta Mater 2009;60(7):582–5.

[43] Prasher R. Thermal boundary resistance and thermal conductivity ofmultiwalled carbon nanotubes. Phys Rew B 2008;77(7):075424.

[44] Tan Z, Li Z, Xiong DB, Fan G, Ji G, Zhang D. A predictive model for interfacialthermal conductance in surface metallized diamond aluminum matrixcomposites. Mater Design 2014;55:257–62.

[45] Schmidt AJ, Collins KC, Minnich AJ, Chen G. Thermal conductance and phonontransmissivity of metal-graphite interfaces. J Appl Phys 2010;107(10):104907.

[46] Turner PS. Thermal-expansion stresses in reinforced plastics. J Res NBS1946;37(4):239–50.

[47] Kerner EH. The elastic and thermo-elastic properties of composite media. ProcPhys Soc B 1956;69:808–13.

[48] Schapery RA. Thermal expansion coefficients of composite materials based onenergy principles. J Compos Mater 1968;2(3):380–404.

[49] Joshua P, Ashkenazi D, Ganor M. The influence of a third element on theinterface reactions in metal-matrix composites (MMCs): Al-graphite system.Mater Sci Eng, A 2000;281(1):239–47.

[50] Khalid FA, Beffort O, Klotz UE, Keller BA, Gasser P. Microstructure andinterfacial characteristics of aluminium–diamond composite materials. DiamRelat Mater 2004;13(3):393–400.

[51] Kim KT, Eckert J, Liu G, Park JM, Lim BK, Hong SH. Influence of embedded-carbon nanotubes on the thermal properties of copper matrix nanocompositesprocessed by molecular-level. Scripta Mater 2011;64(2):181–4.

[52] Cho S, Kikuchi K, Kawasaki A. On the role of amorphous intergranular andinterfacial layers in the thermal conductivity of a multi-walled carbonnanotube-copper matrix composite. Acta Mater 2012;60(2):726–36.

[53] Chang KC, Xu ZF, Matsugi K, Sasaki G. Influence of fiber surface structure oninterfacial structure between fiber and matrix in vapor grown carbon fiberreinforced aluminum matrix composites. Mater Trans, JIM 2009;50(6):1510–8.

[54] Wootton AM, Rappensberger M, Lewis MH, Kitchin S, Howes AP, Dupree R.Structural properties of multi-component silicon oxycarbide glasses derivedfrom metal alkoxide precursors. J Non-Cryst Solids 1996;204:217–27.

[55] Kim KT, Cha SI, Gemming T, Eckert J, Hong SH. The role of interfacial oxygenatoms in the enhanced mechanical properties of carbon-nanotube-reinforcedmetal matrix nanocomposites. Small 2008;4:1936–40.

[56] Bullen AJ, O’Hara KE, Cahill DG, Monteiro O, von Keudell A. Thermalconductivity of amorphous carbon thin films. J Appl Phys2000;88(11):6317–20.