Riccardo
Rurali
*a,
Luciano
Colombo
abc,
Xavier
Cartoixà
d,
Øivind
Wilhelmsen
ef,
Thuat T.
Trinh
e,
Dick
Bedeaux
e and
Signe
Kjelstrup
*e
aInstitut de Ciència de Materials de Barcelona (ICMAB–CSIC) Campus de Bellaterra, 08193 Bellaterra, Barcelona, Spain. E-mail: rrurali@icmab.es
bDipartimento di Fisica, Università di Cagliari, Cittadella Universitaria, I-09042 Monserrato (Ca), Italy
cCatalan Institute of Nanoscience and Nanotechnology (ICN2), CSIC and The Barcelona Institute of Science and Technology, Campus UAB, Bellaterra, 08193 Barcelona, Spain
dDepartament d'Enginyeria Electrònica, Universitat Autònoma de Barcelona, 08193 Bellaterra, Barcelona, Spain
eDepartment of Chemistry, Norwegian University of Science and Technology, NO-7491 Trondheim, Norway. E-mail: signe.kjelstrup@ntnu.no
fSINTEF Energy Research, NO-7465 Trondheim, Norway
First published on 26th April 2016
We perform computational experiments using nonequilibrium molecular dynamics simulations, showing that the interface between two solid materials can be described as an autonomous thermodynamic system. We verify the local equilibrium and give support to the Gibbs description of the interface also away from the global equilibrium. In doing so, we reconcile the common formulation of the thermal boundary resistance as the ratio between the temperature discontinuity at the interface and the heat flux with a more rigorous derivation from nonequilibrium thermodynamics. We also show that thermal boundary resistance of a junction between two pure solid materials can be regarded as an interface property, depending solely on the interface temperature, as implicitly assumed in some widely used continuum models, such as the acoustic mismatch model. Thermal rectification can be understood on the basis of different interface temperatures for the two flow directions.
The discussion originates in the formulation first made by Gibbs.7 He proposed that the interface is an autonomous thermodynamic system, when described by excess densities, and that thermodynamic relationships can be written for these variables. The interface is autonomous in the sense that all its properties are univocally determined by such local variables. The description was developed under equilibrium conditions, but later it has been used successfully out of global equilibrium as well8 to model, in particular, the liquid–vapor phase transitions.5,6,9 In nonequilibrium, these assumptions imply that an interface can sustain a temperature, which is both compatible with its thermodynamic definition and is a local (i.e. different from the surroundings) property that follows from its autonomous character. All excess densities of an autonomous interface will depend on this temperature and not on the temperatures in the adjacent phases. The autonomous nature of a solid–solid interface provides a rigorous justification for tabulating the Kapitza resistance as a junction property, which is independent of the applied thermal bias and where the relevant variable is the interface temperature.
The reluctance in adopting this picture, rather than conceptual, was mostly due to the difficulty of measuring such an interface temperature. Accordingly, under nonequilibrium conditions it was natural to assign to the interface an average temperature, hiding its autonomous nature and hinting that its properties depend on the overall thermodynamic conditions, e.g. the thermal bias, rather than on its own thermodynamic variables.
Numerical simulations have supported the formulation of Gibbs, which implies a local equilibrium at the interface.5,6,9 Support has also been obtained from diffuse interface theories.10–12 However, all these results were obtained for the liquid–vapor interface. In this paper we provide evidence that this property also holds for solid–solid interfaces by performing a controlled set with computational experiments of the Si/Ge interface, namely the prototypical semiconductor heterojunction in many nanotechnology applications of current interest. In doing so, we also give a rigorous theoretical foundation to the common formulation of the Kapitza resistance.
This procedure gives an estimate of the interface thickness of 16.6 Å, i.e. 12 layers of the diamond lattice. It also leads to the interesting result that the interface region lies entirely in the Si half of the system, i.e. the last Ge bilayer, right before the heterojunction, has the same structural features as bulk Ge and relaxation effects all take place in Si. Therefore, the chemical interface (where the chemical identity of the atoms that occupy the zinc-blende lattice changes, dashed line in Fig. 1b) and the thermodynamic interface (defined through the variation of a suitable property P(x), shaded area in Fig. 1b) do not match. Notice that this conclusion is not general and depends on specific conditions (choice of lattice parameter, constant volume), which nevertheless reflect a possible experimental situation, of these calculations. Yet, these results show that such a decoupling is at least in principle possible.
〈Us〉 = 〈Esk〉 + 〈Esp〉, | (1) |
![]() | (2) |
In Fig. 2 we plot the internal energy of the interface as a function of the interface temperature for both the equilibrium and nonequilibrium conditions, which result in a qualitative and quantitative agreement, within the accuracy of the calculation. This result strongly supports the view of the solid–solid interface as an autonomous thermodynamic system. Our calculations show that no matter what the overall thermodynamic conditions of the system are, and we tested this hypothesis under conditions as different as equilibrium and nonequilibrium, the internal energy of the interface only depends on its temperature and not on the overall thermal bias conditions. These results also support that the local equilibrium, one of the underlying assumptions of nonequilibrium thermodynamics and thermodynamic modeling at large, holds. In other words, one can define a small enough piece of material which can be considered in equilibrium and assign to it a temperature, obeying T = (∂U/∂S)V,N, even under considerably out-of-equilibrium conditions. In what follows, to further test the hypothesis of local equilibrium, we calculate the thermal boundary resistance under different nonequilibrium conditions. Notice that the derivative of the internal energy with respect to the temperature at constant volume is the heat capacity, i.e. the amount of heat required to change the temperature of a given system by one degree. We can therefore define and calculate the heat capacity of the interface as
![]() | (3) |
![]() | ||
Fig. 2 Internal energy of the interface as a function of the interface temperature under equilibrium and non-equilibrium conditions. |
We found that the heat capacity of the interface was the same at equilibrium and nonequilibrium, and for the system studied, we estimated CsV to be 29 J K−1 mol−1. We performed the same calculation, but restricting this time to a region of Si sufficiently far from the interface and the cell boundary, obtaining a value of 33 J K−1 mol−1. We understand that both values of the heat capacity calculated at the interface and far away from it are not accurate since the present simulations do not have any quantum features, unlike that included in the more precise prediction in ref. 19 for bulk-like Si. Furthermore, the structure investigated here has, by construction, a pseudomorphic lattice meaning that both Si and Ge slabs are in fact under strain, so that the actual value of their heat capacity is expected to differ from the bulk-like one. Nevertheless, we remark that their relative difference (as large as 15%) is in fact meaningful, carrying important qualitative information, namely, the additional proof of the thermodynamic autonomy of the interface with respect to the neighboring bulk-like regions.
If the interface is an autonomous thermodynamic system, its thermal resistance can be treated as a system variable that depends solely on the interface temperature. To calculate the interface thermal resistance we need the temperature discontinuity across the interface. We then extrapolate the linear dependence of T(x) in the Si and Ge regions to the interface boundaries and obtain Ti and To (see Fig. 1c); the linear fits are performed conveniently far from the thermostats and from the interface. The ratio of the temperature jump
ΔTs = To − Ti, | (4) |
The present nonequilibrium thermodynamics approach proceeds along a different path. At the interface, the entropy production associated with the transport of heat9 is
![]() | (5) |
![]() | (6) |
![]() | (7) |
![]() | (8) |
We use eqn (8) to calculate rs for different nonequilibrium conditions. We apply ΔT of 200 and 400 K and a reverse bias of ΔT = −400 K. In each case we consider several average temperatures (TH + TC)/2 in order to sample many interface temperatures, Ts. In Fig. 3 we plot rs as a function of Ts. This plot shows clearly that rs indeed depends only on Ts: irrespective of the overall thermodynamic conditions, each value of Ts is associated with a corresponding rs. We make the same plot for the more common Kapitza resistance, rK = ΔTs/J (bottom panel), and obtain very similar conclusions. Indeed, rK can be obtained, to the lowest order in the temperature difference, from rs by multiplying it by (Ts)2, as shown in Table 1.
![]() | ||
Fig. 3 (Top) Interface thermal resistance calculated from eqn (8). (Bottom) Kapitza resistance calculated as ΔTs/J. The need for better statistics at low temperature results in minor discrepancies for Ts > 200 K (not appreciable in the log-scale of the upper panel). |
T s | r K | r s(Ts)2 |
---|---|---|
47.7 | 1.2 × 10−8 | 3.0 × 10−8 |
101.4 | 8.5 × 10−9 | 8.7 × 10−9 |
169.8 | 5.6 × 10−9 | 5.4 × 10−9 |
236.3 | 4.5 × 10−9 | 4.3 × 10−9 |
287.5 | 3.8 × 10−9 | 3.6 × 10−9 |
341.4 | 3.5 × 10−9 | 3.3 × 10−9 |
392.7 | 3.2 × 10−9 | 3.1 × 10−9 |
443.7 | 3.0 × 10−9 | 2.9 × 10−9 |
494.6 | 2.8 × 10−9 | 2.7 × 10−9 |
546.7 | 2.6 × 10−9 | 2.5 × 10−9 |
597.8 | 2.5 × 10−9 | 2.4 × 10−9 |
647.7 | 2.5 × 10−9 | 2.4 × 10−9 |
700.2 | 2.4 × 10−9 | 2.4 × 10−9 |
If one writes Fourier's law, in its integral form, for the entire system
ΔT = −![]() | (9) |
![]() | (10) |
We conclude our study with a final remark on thermal rectification, i.e. the preferential flow of heat in one direction.24 Previous studies have demonstrated that the different temperature dependences of the thermal conductivity of two materials result in some degree of rectification when they are brought together and form a junction.15,25–28 The role of the interface itself in the rectification, however, has thus far been neglected. Here we have shown that rs depends univocally on Ts. The latter, nonetheless, depends on how the overall thermal bias TH − TC is split between the two materials: the more their thermal resistances differ, the farther Ts will be from the mean temperature (TH + TC)/2 (see ref. 28 for a simple model). Consequently, a different temperature dependence of the thermal conductivity of the two materials results also in a different Ts upon forward or reverse bias. This is clearly seen in Fig. 4, where we have plotted Ts also in the case of a reverse bias ΔT = −400 K: the interface temperatures are different, when compared with the case of forward bias ΔT = 400 K, thus the interface resistances rs(Ts) will also be different and will contribute to the thermal rectification.
This journal is © the Owner Societies 2016 |