Thermal conductivity and thermal boundary resistance of nanostructures
 Konstantinos Termentzidis^{1, 2, 3}Email author,
 Jayalakshmi Parasuraman^{4},
 Carolina Abs Da Cruz^{1, 2, 3},
 Samy Merabia^{5},
 Dan Angelescu^{4},
 Frédéric Marty^{4},
 Tarik Bourouina^{4},
 Xavier Kleber^{6},
 Patrice Chantrenne^{1, 2, 3} and
 Philippe Basset^{4}
DOI: 10.1186/1556276X6288
© Termentzidis et al; licensee Springer. 2011
Received: 9 December 2010
Accepted: 4 April 2011
Published: 4 April 2011
Abstract
Abstract
We present a fabrication process of lowcost superlattices and simulations related with the heat dissipation on them. The influence of the interfacial roughness on the thermal conductivity of semiconductor/semiconductor superlattices was studied by equilibrium and nonequilibrium molecular dynamics and on the Kapitza resistance of superlattice's interfaces by equilibrium molecular dynamics. The nonequilibrium method was the tool used for the prediction of the Kapitza resistance for a binary semiconductor/metal system. Physical explanations are provided for rationalizing the simulation results.
PACS
68.65.Cd, 66.70.Df, 81.16.c, 65.80.g, 31.12.xv
Introduction
Understanding and controlling the thermal properties of nanostructures and nanostructured materials are of great interest in a broad scope of contexts and applications. Indeed, nanostructures and nanomaterials are getting more and more commonly used in various industrial sectors like cosmetics, aerospace, communication and computer electronics. In addition to the associated technological problems, there are plenty of unresolved scientific issues that need to be properly addressed. As a matter of fact, the behaviour and reliability of these devices strongly depend on the way the system evacuates heat, as excessive temperatures or temperature gradients result in the failure of the system. This issue is crucial for thermoelectric energyharvesting devices. Energy transport in micro and nanostructures generally differs significantly from the one in macrostructures, because the energy carriers are subjected to ballistic heat transfer instead of the classical Fourier's law, and quantum effects have to be taken into account. In particular, the correlation between grain boundaries, interfaces and surfaces and the thermal transport properties is a key point to design materials with preferred thermal properties and systems with a controlled behaviour.
Fabrication process of superlattices
We aim to fabricate optimized vertical nanosuperlattices (with layers ranging <100 nm each) with high thermoelectric efficiency. High thermoelectric efficiency occurs for high electrical conductivity and low thermal conductivity. The electronic conductivity will be controlled though the Si doping and the use of metal to fill in the trenches. The film thickness needs to be decreased, to decrease the individual layer thermal conductivity and increase the influence of the interfacial thermal resistance.
To obtain such dimension on a large area at low cost, we are developing a process based on the transfer by DRIE of 30nm line patterns made of diblock copolymers [4]. For this purpose, it is required to characterize them to the best possible degree of accuracy. Measurements at this scale will possibly be plagued by quantum effects [5, 6]. That is the reason why we fabricated first microscale superlattices, to make thermoelectric measurements free from quantum effects and then applied the method to characterize the final nanosuperlattice thermoelectric devices.
Simulations: thermal conductivity of superlattices
When the layer thickness of the superlattices is comparable to the phonon mean free path (PMFP), the heat transport remains no longer diffusive, but ballistic within the layers. Furthermore, decreasing the dimensions of a structure increases the effects of strong inhomogeneity of the interfaces. Interfaces, atomically flat or rough, impact the selection rules, the phonon density of states and consequently the hierarchy or relative strengths of their interactions with phonons and electrons. Thus, it is important to study and predict the heat transfer and especially the influence of the height of superlattice's interfaces on the cross and inplane thermal conductivities. This is a formidable task, from a theoretical point of view, as one needs to account for the ballistic motion of the phonons and their scattering at interfaces. Molecular dynamics is a relatively simple tool which accounts for these phenomena, and it has been applied successfully to predict heattransfer properties of superlattices.
Two routes can be adopted to compute the thermal conductivity, namely, the nonequilibrium (NEMD) [7] and the equilibrium molecular dynamics (EMD) [8]. In this article, we have considered both methods to characterize the thermal anisotropy of the superlattices. In the widely used direct method (NEMD), the structure is coupled to a heat source and a heat sink, and the resulting heat flux is measured to obtain the thermal conductivity of the material [9, 10]. Simulations are held for several systems of increasing size and finally thermal conductivity is extrapolated for a system of infinite size [11, 12]. The NEMD method is often the method of choice for studies of nanomaterials, while for bulk thermal conductivity, particularly that of high conductivity materials, the equilibrium method is typically preferred because of less severe size effects. Comparisons between the two methods have been done previously, concluding that the two methods can give consistent results [13, 14]. GreenKubo method for nanostructures is proven to have greater uncertainties than those of NEMD, but a correct description of thermal conductivity with EMD is achieved by establishing statistics from several results, starting from different initial conditions.
The superlattice system under study is made of superposition of LennardJones crystals and fcc structures, oriented along the [001] direction. The molecular dynamics code LAMMPS [15–17] is used in all the NEMD and EMD simulations. The mass ratio of the two materials of the superlattice is taken as equal to 2, and this ratio reproduces approximately the same acoustic impedance difference as that between Si and Ge. Periodic boundary conditions are used in all the three directions. Superlattices with period of 40a _{0} are discussed, where a _{0} is the lattice constant. The shape of the roughness is chosen as a right isosceles triangle. The roughness height was varied from one atomic layer (1 ML = 1/2a _{0}) to 24a _{0}. For each roughness, heat transfer simulations with NEMD were performed for several system sizes in the heat flux direction to extrapolate the thermal conductivity for a system of infinite size [11]. For EMD simulations, the size of the system is smaller than with NEMD simulations and only one size is considered 20a _{0} × 10a _{0} × 40a _{0}, where the last dimension is perpendicular to interfaces.
In the diffusemismatch model, on the other hand, phonons are diffusively scattered at interfaces, and their energy is redistributed in all the directions [20]. In practice, the acoustic model describes the physics of interfacial heat transfer at low temperatures, for phonons having large wavelengths, while the diffuse model is relevant for small wavelengths phonons. At the considered temperature in the current study, we are most probably in an intermediate situation where the physics is not captured by one single model. Nevertheless, both models predict that a moderate amount of interfacial roughness will tend to decrease the inplane TC, because rough interfaces will increase specular reflection and diffusive scattering of phonons travelling in the in plane direction. However, if the roughness is large enough, then locally, the phonons encounter smoothlike interfaces, and the partial group of phonons that are diffusely scattered in all space direction decreases. This might explain the further increase of the thermal conductivity when the roughness is large enough.
The behaviour of the crossplane thermal conductivity is different: it increases monotonously with the interfacial roughness. For smooth interfaces, the crossplane thermal conductivity is 50% lower than the inplane thermal conductivity. This anisotropy has to be taken into account for thermal behaviour of systems made of submicronic solid layers. Invoking again the acoustic mismatch model, we conclude that the transmission coefficient of the solid/solid interface is smaller than the reflection coefficient, which is not surprising if we consider the acoustic impedance ratio of the two materials. Roughness increases the transmission coefficient as it increases the diffused scattering at the interface [12].
The same qualitative trend regarding the influence of the roughness on the thermal conductivity of superlattices has been reported previously for materials with diffusive behaviour, without thermal contact resistance [21]. In this case, the variation of the inplane and crossplane conductivities with the interfacial roughness is due to the heat flux line deviation that minimizes the heat flux path in the material that has the lower thermal conductivity. This tends to increase the crossplane thermal conductivity. On the other hand, the increase of the roughness leads to the heat flux constrictions that decrease the inplane thermal conductivity. The qualitative interpretation of the results shows that the thermal contact resistance of the interface has a strong influence on the superlattice thermal conductivity.
Simulations: Kapitza resistance
Superlattices with rough interfaces
where S is the interface area. The latter formula expresses the fact that the resistance is controlled by the transmission of all the phonons travelling across the interface.
In the situation of interest to us here, the transmission of phonons is expected to be strongly anisotropic, and thus the resistance developed by an interface should depend on the main direction of the heat flux. To measure this anisotropy, we have generalised the previous equation and introduced the concept of directional resistance, by considering the heat flux q _{θ}(t) in the direction θ in (0,π/2) with the normal of the interface.
This angular Kapitza resistance quantifies the transmission of the heat flux in the direction making an angle θ with the normal of the interface.
Silver/silicon interfaces
The Cu and Ag films on Sioriented substrates are the principal combinations in largescale integrate circuits. Furthermore, with the fabrication process of verticalbuilt superlattices described in previous section, we are interested in the heat transfer phenomena related to the metal/semiconductor interfaces. The prediction of heat transfer in these systems becomes challenging when the thickness of the layers reaches the same order of magnitude as the PMFP. For heat transfer studies, MD is well suited for dielectrics since only phonons carry heat. For metals, coupling between phonons and electrons can be modelled with the twotemperature model [22]. For the above systems, it has been proven that the Kapitza resistance is mainly due to phonon energy transmission through the interfaces [23, 24]. The interfacial thermal resistance, known as the Kapitza resistance [25, 26] is important to be studied as it might become of the same order of magnitude than the film thermal resistance. In this section, interatomic potentials for Ag and Si are discussed. Using NEMD simulations, for an average temperature of 300 K, the Kapitza resistance of Si/Ag systems is determined.
Modified embeddedatom method (MEAM) is the only appropriate potential that can be used for metal/semiconductor systems. The first nearestneighbour MEAM (1NN MEAM) potential by Baskes et al. [27] and the second nearestneighbour MEAM (2NN MEAM) by Lee [28] are examined in the current study. The general MEAM potential is a good candidate for simulating the dynamics of a binary system with a single type of potential. For example, it can be applied for both fcc and bcc structures. Furthermore, this potential includes directional bonding, and thus can be applied for Si systems.
The 2NN MEAM potential allows recovering the expansion coefficient for Ag quite accurately while the 1NN MEAM potential significantly underestimates it. For Ag, the two potentials provide a good description for the more basic properties, such as cohesive energy, lattice parameters and bulk modulus [31]. Even if the 1NN MEAM potential gives results closer to the experimental values for dispersion curves, the values obtained for the linear thermal expansion are not reasonable. Therefore, the 1NN MEAM potential cannot be considered appropriate for simulating heat transfer for silver. Regarding the investigation of heattransfer temperature, the 2NN MEAM gives the best results for harmonic and anharmonic properties for silver and for silicon using the previous results of the literature [32]. Kapitza resistance is predicted for the 2NN MEAM Si/Ag potential. The interface thermal resistance, also known as Kapitza resistance, R _{K}, creates a barrier to heat flux and leads to a discontinuous temperature, ΔT, drop across the interfaces.
The Kapitza resistance obtained with NEMD is 4.9 × 10^{9} m^{2}K/W. The temperature profile for Si is almost flat due its high thermal conductivity. With MD simulations, it is not possible to simulate heat transfer due to the electrons, and thus the steep slope of Ag is due to its low lattice thermal conductivity. The value R _{KT} is in the range 1.4125 × 10^{9} m^{2}K/W which also includes the Kapitza conductance for dielectric/metal systems [33, 34].
Conclusions  Discussion
A new fabrication method for superlattices is used, reducing the time and fabrication costs. With the fabrication of vertical superlattices, several questions a rose for the influence of the roughness' height of the superlattices and the quality of interface on the thermal transport. When the length of the superlattice's period is comparable to the phononfree mean path, the heat transfer becomes ballistic.
The crossplane and inplane thermal conductivities of a dielectric/dielectric (representing SiGe systems) superlattice are predicted using EMD and NEMD simulations. Both methods give the same tendencies for the anisotropic heat transfer at superlattices with rough interfaces. The inplane thermal conductivity exhibits a minimum for a certain interfacial width, while the crossplane thermal conductivity increases modestly in increasing the width of the interfaces. The Kapitza resistance of these interfaces is also studied, with a proposed methodology in this article, introducing the concept of directional thermal resistance. Values presented here are coherent with the difference between the inplane and crossplane thermal conductivities.
Molecular dynamics simulations are also used to study the metal/semiconductor interfaces. Among all the interatomic potentials that are available, the MEAM potential is a good alternative to work with since it can be used for different materials. At 300 K, the 2NN MEAM potential gives the best results for the fundamental properties associated with the heat transfer of silicon and silver. Previous results [23, 24, 32] suggest that interfacial thermal conductance depends predominantly on the phonon coupling between silicon and metal lattices so that Si/Ag can be simulated without considering the contribution of electron heat transfer. The value of magnitude of the Kapitza resistance for a Si/Ag system is within the range of Kapitza resistance proposed in the literature.
This study proves that making rough instead of smooth interfaces in superlattices is a useful way to decrease the thermal conductivity and finally to design materials with desired thermal properties. Furthermore, when more interfaces are added (rough or smooth), i.e. when the superlattice's period decreases, the interfacial thermal resistance becomes comparable to the superlattice's layers thermal conductivity. With these two parameters, namely, the introduction of rough interfaces and the decrease of the superlattice's period, we can create systems with controlled values of the thermal conductivity.
Abbreviations
 DRIE:

deep reactive ion etching
 SEM:

scanning electron microscope
 PMFP:

phonon mean free path
 NEMD:

nonequilibrium molecular dynamics
 EMD:

equilibrium molecular dynamics
 LJU:

LennardJones units
Declarations
Acknowledgements
This study has been conducted within the framework of the projects ANRCOFISIS (ANR07NANO04703). COFISIS (Collective Fabrication of Inexpensive Superlattices in Silicon) is a project with collaboration between theoretical and experimental groups in ESIEE Paris, CETHIL and MATEIS at INSa of Lyon. The project COFISIS intends to develop integrated siliconbased and lowcost superlattices.
Authors’ Affiliations
References
 Marty F, Rousseau L, Saadany B, Mercier B, Francais O, Mita Y, Bourouina T: Advanced etching of silicon based on deep reactive ion etching for silicon high aspect ratio microstructures and threedimensional micro and nanostructures. Microelectronics Journal 2005, 36(Issue 7):673–677. 10.1016/j.mejo.2005.04.039View ArticleGoogle Scholar
 Mita I, Kubota M, Sugiyama M, Marty F, Bourouina T, Shibata T: Aspect Ratio Dependent Scalloping Attenuation in DRIE and an Application to LowLoss FiberOptical Switch. In Proc. of IEEE International Conference on MicroElectroMechanical Systems (MEMS 2006). Istanbul, Turkey; 2006:114–117.Google Scholar
 Kapitza PL: J Phys. Volume 4. (Moscow); 1941:181.Google Scholar
 Register RA, Angelescu D, Pelletier V, Asakawa K, Wu MW, Adamson DH, Chaikin PM: ShearAligned Block Copolymer Thin Films as Nanofabrication Templates. Journal of Photopolymer Science and Technology 2007, 20: 493. 10.2494/photopolymer.20.493View ArticleGoogle Scholar
 Hannay NB: Semiconductors. Reinhold: New York; 1959.Google Scholar
 Radkowski P III, Sands PD: Quantum Effects in Nanoscale Transport: Simulating Coupled Electron and Phonon Systems in Quantum Wires and Superlattices. Thermoelectrics 1999.Google Scholar
 Kotake S, Wakuri S: Molecular dynamics study of heat conduction in solid materials. JSME International Journal, Series B 1994, 37: 103.View ArticleGoogle Scholar
 Frenkel D, Smit B: Understanding Molecular Simulation: From Algorithms to Applications. San Diego: Academic Press Inc; 1996.Google Scholar
 Chantrenne P, Barrat JL: Analytical model for the thermal conductivity of nanostructures. Superlattices and Microstructures 2004, 35: 173. 10.1016/j.spmi.2003.11.011View ArticleGoogle Scholar
 Chantrenne P, Barrat JL: Finite size effects in determination of thermal conductivities: Comparing molecular dynamics results with simple models. J Heat Transfer  Transactions ASME 2004, 126: 577. 10.1115/1.1777582View ArticleGoogle Scholar
 Schelling PK, Phillpot SR, Keblinski P: Comparison of atomiclevel simulation methods for computing thermal conductivity. Physical Review B 2002, 65: 144306. 10.1103/PhysRevB.65.144306View ArticleGoogle Scholar
 Termentzidis K, Chantrenne P, Keblinski P: Nonequilibrium molecular dynamics simulation of the inplane thermal conductivity of superlattices with rough interfaces. Physical Review B 2009, 79: 214307. 10.1103/PhysRevB.79.214307View ArticleGoogle Scholar
 Poetzsch R, Böttger H: Interplay of disorder and anharmonicity in heat conduction: Moleculardynamics study. Physical Review B 1994, 50: 15757. 10.1103/PhysRevB.50.15757View ArticleGoogle Scholar
 Landry ES, McGaughey AJH, Hussein MI: Molecular dynamics prediction of the thermal conductivity of Si/Ge superlattices. Proc. ASME/JSME Thermal Engineering summer Heat Transfer Conf 2007, 2: 779. 2007 2007Google Scholar
 LAMMPS Molecular Dynamics Simulator[http://lammps.sandia.gov]
 Plimpton S: Fast Parallel Algorithms for Shortrange Molecular Dynamics. J Computational Physics 1995, 117: 1. 10.1006/jcph.1995.1039View ArticleGoogle Scholar
 Plimpton S, Pollock P, Stevens M: ParticleMesh Ewald and rRESPA for Parallel Molecular Dynamics Simulations. In Proc. 8th SIAM Conf. on Parallel Processing for Scientific Computing. Minneapolis, MN; 1997.Google Scholar
 Khalitnikov IM: Zh Eksp Teor Fiz. 1952, 22: 687.Google Scholar
 Swartz ET, Pohl RO: Thermal boundary resistance. Reviews of Modern Physics 1989, 61: 605. 10.1103/RevModPhys.61.605View ArticleGoogle Scholar
 Reddy P, Castelino K, Majumdar A: Diffuse mismatch model of thermal boundary conductance using exact phonon dispersion. Applied Physics Letters 2005, 87: 211908. 10.1063/1.2133890View ArticleGoogle Scholar
 Ladd AJC, Moran B, Hoover WG: Lattice thermal conductivity  a comparison of molecular dynamics and anharmonic lattice dynamics. Physical Review B 1986, 34: 5058. 10.1103/PhysRevB.34.5058View ArticleGoogle Scholar
 Rutherford AM, Duffy DM: The effect of electronion interactions on radiation damage simulations. Journal of Physics  Condensed Matter 2007, 19: 496201. 10.1088/09538984/19/49/496201View ArticleGoogle Scholar
 Mahan GD: Kapitza thermal resistance between a metal and a nonmetal. Physical Review B 2009, 79: 075408. 10.1103/PhysRevB.79.075408View ArticleGoogle Scholar
 Lyeo HK, Cahill DG: Thermal conductance of interfaces between highly dissimilar materials. Physical Review B 2006, 73: 144301. 10.1103/PhysRevB.73.144301View ArticleGoogle Scholar
 Hu M, Keblinski P, Schelling PK: Kapitza conductance of siliconamorphous polyethylene interfaces by molecular dynamics simulations. Physical Review B 2009, 79: 104305. 10.1103/PhysRevB.79.104305View ArticleGoogle Scholar
 Luo TF, Lloyd JR: Nonequilibrium molecular dynamics study of thermal energy transport in AuSAMAu junctions. J Heat and Mass Trasfer 2010, 53: 1. 10.1016/j.ijheatmasstransfer.2009.10.033View ArticleGoogle Scholar
 Baskes MI, Nelson JS, Wright AF: Semiempirical modified embeddedatom potentials for silicon and germanium. Physical Review B 1989, 40: 6085. 10.1103/PhysRevB.40.6085View ArticleGoogle Scholar
 Lee BJ, Baskes MI: Second nearestneighbor modified embeddedatommethod potential. Physical Review B 2000, 62: 8564. 10.1103/PhysRevB.62.8564View ArticleGoogle Scholar
 Lynn JW, Smith HG, Nicklow RM: Lattice Dynamics of Gold. Physical Review B 1973, 8: 3493. 10.1103/PhysRevB.8.3493View ArticleGoogle Scholar
 Touloukian YS, Taylor RE, Desai PD: Thermal ExpansionMetallic Elements and Alloys. Volume 12. New York: Plenum; 1975.View ArticleGoogle Scholar
 Lee BJ, Shim JH, Baskes MI: Semiempirical atomic potentials for the fcc metals Cu, Ag, Au, Ni, Pd, Pt, Al, and Pb based on first and second nearestneighbor modified embedded atom method. Physical Review B 2003, 68: 144112. 10.1103/PhysRevB.68.144112View ArticleGoogle Scholar
 Da Cruz CA, Chantrenne P, Kleber X: Molecular Dynamics simulations and Kapitza conductance prediction of Si/Au systems using the new full 2NN MEAM Si/Au crosspotential. In Proc ASME/JSME. Honolulu, Hawaii, USA; 2011. 8th Thermal Engineering Joint Conference AJTEC2011, March 13–17, 2011 8th Thermal Engineering Joint Conference AJTEC2011, March 1317, 2011Google Scholar
 Smith AN, Hostetler JL, Norris PM: Thermal boundary resistance measurements using a transient thermoreflectance technique. Microscale Thermophysical Engineering 2000, 4: 51. 10.1080/108939500199637View ArticleGoogle Scholar
 Stoner RJ, Maris HJ: Kapitza conductance and heat flow between solids at temperatures from 50 to 300 K. Physical Review B 1993, 48: 16373. 10.1103/PhysRevB.48.16373View ArticleGoogle Scholar
Copyright
This article is published under license to BioMed Central Ltd. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.