Abstract
The value measured in the amorphous structure with the same chemical composition is often considered as a lower bound for the thermal conductivity of any material: the heat carriers are strongly scattered by disorder and their lifetimes reach the minimum time scale of thermal vibrations. An appropriate design at the nanoscale, however, may allow one to reduce the thermal conductivity even below the amorphous limit. In the present contribution, using moleculardynamics simulation and the GreenKubo formulation, we study systematically the thermal conductivity of layered phononic materials (superlattices), by tuning different parameters that can characterize such structures. We have discovered that the key to reach a lowerthanamorphous thermal conductivity is to block almost completely the propagation of the heat carriers, the superlattice phonons. We demonstrate that a large mass difference in the two intercalated layers, or weakened interactions across the interface between layers result in materials with very low thermal conductivity, below the values of the corresponding amorphous counterparts.
Introduction
Materials with low thermal conductivity, κ, are employed in many modern technologies, such as thermal management in electronic devices or thermoelectric energy conversion^{1,2,3}. In general, low values of κ are observed in disordered solids^{4}, including topologically disordered systems and crystalline solids with size or mass disorder^{5,6,7,8}. We refer to the former as glasses, to the latter as disordered alloys. While in glasses disorder originates both from molecules size or mass heterogeneity and the topologically amorphous structure, in disordered alloys molecules characterized by heterogeneous attributes are located at ordered lattice sites. The low κ in disordered solids can be rationalized by considering the phenomenological kinetic theory expression^{9} κ = (1/3)Cv^{2}τ, which relates average velocity, v and lifetime, τ, (and therefore the mean free path ) of phonons to κ (C is the specific heat per unit volume). In good crystals, phonons lifetime is primarily controlled by anharmonic interactions. In contrast, in disordered solids, the disorder (or the elastic heterogeneity^{10}) reduces τ (or ) and, as a result, κ.
In early experimental investigations^{5,6}, Cahill et al. have studied the disordered alloys, e.g., (KBr)_{1−x}(KCN)_{x}, (NaCl)_{1−x}(NaCN)_{x} and shown that κ can be reduced to the glass value by controlling the relative composition x. In our work^{8} we in turn demonstrated that, in sizedisordered crystal (alloy), κ progressively decreases with increasing size mismatch, eventually converging to the corresponding glass value. When this limit is reached, τ is comparable to the time scale of thermal vibrations ( to the particle size), i.e., to the minimum time (length) scale^{8}. Heat propagation can therefore be described as a random walk of vibrational energies^{5,6}, or in terms of nonpropagating delocalized modes, the diffusons^{7}. For this reason, the value in the glass is generally considered as a lower bound for κ of materials with homogeneous chemical composition^{5,6}.
A crucial issue^{4} is whether thermal conductivity can be lowered below the glass limit through nanoscale phononic design^{3,11}. This possibility would allow to devise (meta)materials which are excellent thermal insulators while preserving good electronic properties, as needed in many applications^{1,2,3}. The most popular design to reach this goal is that of a lamellar superlattice^{12,13,14,15}, often composed of two chemically different intercalated layers, e.g., SiGe^{12,13} or GaAsAlAs^{14,15} (see also Fig. 1). In a superlattice, the thermal conductivity tensor is anisotropic, with the crossplane component, κ_{CP}, usually lower than the inplane value, κ_{IP}^{16,17}.
In recent experiments^{18,19,20}, ultralow values of κ_{CP}, suggested to be smaller than the amorphous limit, were measured. In particular, Costescu et al.^{18} demonstrated that the presence of a highdensity of interfaces decreases κ_{CP} of WAl_{2}O_{3} nanolaminates, below that of the amorphous Al_{2}O_{3}. An experiment by Chiritescu et al.^{19} achieved ultralow thermal conductivity in layered WSe_{2} crystals, by disordering the crystalline WSe_{2} sheets. Finally, Pernot et al.^{20} also observed very low values of κ_{CP}, below that of amorphous Si, in Ge nanodots multilayers separated by Si crystals.
Although the above experimental works have demonstrated very low values of κ in superlattice systems, we note that these values have not been systematically compared to those assumed in the glasses with exactly the same chemical composition. Since different chemical species are expected to produce different effects on κ, it is therefore still not completely clear whether superlattice structuration alone can lower κ below the amorphous limit. Note, for instance, that very recent numerical work^{21}, has shown that a superlattice composed by layers with randomized thicknesses can indeed show a κ below the value pertaining to the disorderedalloy with the same composition. This limit, however, is generally higher than that in the corresponding glass^{5,6,7,8}, which should therefore be considered the true amorphous limit to be beaten.
In addition, a general framework to rationalize in a coherent single picture the previous observations of very low κ is, to the best of our knowledge, still lacking.
In this work, we address these two issues. Building on the comparison of the superlattice with the corresponding amorphous structure, we clarify the mechanisms allowing for ultralow thermal conductivity in the former. We have studied by computer simulation a numerical model that allows one to exactly compare ordered and disordered systems with identical chemical composition and access detailed information on the entire normal modes spectrum, providing, as a consequence, a complete understanding of the heat transfer process. As the lifetime of heat carriers is already minimum in glasses^{8}, we demonstrate that the key to even lower thermal conductivities is to suppress their propagation across the interfaces between the constituent layers.
More in details, we have focused on three distinct design principles for superlattices, mimicking similar configurations actually employed in experiments. These are based on the facecenteredcubic (FCC) lattice structure and are composed of: (S1) two intercalated crystalline layers formed by point particles with different masses; (S2) ordered crystalline layers intercalated to massdisordered alloy layers; and (S3) identical crystalline layers with modified (weakened) interactions across the interfaces (see the Methods section and Table 1). We show that a large mass difference between layers (S1) and weakened interactions between layers (S3) efficiently obstruct the propagation of phonons, resulting in a very large reduction of the superlattice thermal conductivity, even below the values pertaining to the glass phases with identical composition. Based on our results, we conclude with a discussion of the optimal strategy to follow towards very low thermal conductivity materials.
In Fig. 1 we show a schematic illustration of a superlattice composed by two intercalated layers, A and B, both of thickness W/2. The competition between two length scales, the repetition period of the superlattice, W and the mean free path of the superlattice phonons, , determines the coherent or incoherent character of phonon transport, as described in^{22,23,24} and demonstrated by numerical simulations^{25,26,27} and recent experiments^{28}.
For , the incoherent phonon transport is independent in the different layers and phonons can be effectively treated as particles. In this case, the Boltzmann transport equation applies^{29,30} and the particlelike phonons are scattered within the layers (internal resistance) and at the interfaces (interfacial resistance or Kapitza resistance^{31,32,33}). The thermal conductivity in the crossplane direction can be written as^{22}
where
Here, κ_{A} and κ_{B} are the thermal conductivities of materials A and B and is the Kapitza length^{34}. R is the interfacial resistance, which exists even at a perfect interface and depends on the nature of the contacting materials (e.g., crystalcrystal, crystalglass)^{31,32}. For (), the interfacial resistance is relatively large (small) compared to the internal resistance. Both κ_{CP} and κ_{IP} (the inplane thermal conductivity) increase with W, due to the decrease of the interfacial resistance density^{29,30}. In the diffuse limit W → ∞, where the interfacial resistance can be neglected, κ_{CP} and κ_{IP} have the upper bounds and , respectively.
When , phonon transport is coherent^{22,23,24,25,26,27,28} and the wave nature of phonons cannot be neglected. In this regime, κ_{CP} decreases with increasing W, in contrast with the incoherent case. The reduction of κ_{CP} is explained with the emergence of a band gap at the Brillouin zone boundary, due to bandfolding^{35}: increasing W augments the frequency gap in the dispersion relation. This, in turn, decreases the average group velocity v of phonons, finally reducing κ_{CP}. Miniumklapp processes^{36}, occurring at the miniBrillouin zone, also contribute to the reduction of κ_{CP}. At the crossover length , between the incoherent and the coherent transport regimes, κ_{CP} assumes a minimum value when plotted against W^{22,23,24,25,26,27,28}. We have encountered this situation in the case of superlattice S1, as we will see below.
Details of the structure of the interface between layers are also known to significantly affect phonon transport^{37,38,39,40,41,42,43,44,45,46}. It has been reported that interfacial roughness^{37,38,39} or mixing^{40,41} reduce both κ_{CP} and κ_{IP} and can even suppress the coherent nature of phonons, with κ_{CP(IP)} increasing monotonously at any W. The interface topology is also an important factor to determine the phonon transport^{42,43}. While we will not address precisely this situation in detail here, the superlattice S2 of our study bears some similarities with it.
Finally, the stiffness of interfacial bondings, which can be controlled by applying pressure^{44,45} or tuning chemical bonding^{46}, has significant effects on heat transport features, which will be demonstrated by the study of the S3 superlattice.
Results
In Table 1, we present the details of the three superlattice systems studied in this work, with values of the important quantities: κ_{A} and κ_{B} are the thermal conductivities of layers A and B, respectively; and are the cross and inplane diffuse limits of κ_{CP} and κ_{IP}; R is the interfacial resistance, the Kapitza length; κ_{glass} and κ_{alloy} are the thermal conductivities of the glass and disordered alloy with exactly the same composition as the indicated superlattice.
The number density of all systems was fixed at , with a corresponding crystal lattice constant a = 1.58. In order to minimize anharmonic couplings and focus primarily on the contribution arising from the details of the nanostructuration, we considered a low temperature value T = 10^{−2}.
Thermal conductivities have been estimated by the GreenKubo formulation^{47,48}. We have calculated both components of the superlattices thermal conductivities, κ_{CP} and κ_{IP}, by varying the pattern repetition period W. Note that W indicates the number of monolayers of the lattice structure, where the distance between adjacent monolayers is a/2 for the perfect FCC structure (see also Fig. 1). In the following, we systematically compare the value of κ_{CP} to , κ_{glass}, κ_{alloy}, to evaluate the efficiency of the superlattice structures in minimizing heat transfer in the direction of the patterning. The inplane behaviour has been similarly quantified by comparing κ_{IP} to . In addition, we have also characterized the vibrational states by using a standard normalmodes analysis^{8,49}.
All important information about the system models and methods used for the simulation production runs and analysis are given in the Methods section. Additional details about specific points are included in the Supplementary Materials.
S1. Superlattice composed of two intercalated crystalline layers with different masses
In Fig. 2 we show the thermal conductivities, κ_{CP} and κ_{IP} (symbols), as functions of the replication period, W, for the layers mass ratios m_{B}/m_{A} = 2, 4 and 8. Note that we have chosen to fix the average mass, 〈m〉 = (m_{A} + m_{B})/2 = 1, rather than fixing a reference value m_{A} = 1 and varying m_{B}. This latter protocol has indeed an obvious drawback: the average mass would increase when considering different values of the ratio m_{B}/m_{A}, implying a trivial effect on the thermal conductivity which scales as 〈m〉^{−1/2}. This would therefore hinder the possibility to isolate the contribution to lowering the thermal conductivity which originates from the mass mismatch alone.
The values of the diffuse limits and as well as those of the glass and the disordered alloy constituted by the same species (see Table 1) are also shown as lines in Fig. 2. As expected, the relation holds for the pure materials. In the studied Wrange, W = 2 to 40 (monolayers), the inplane value κ_{IP} shows a very weak dependence on W, as was observed for superlattices with perfect interfaces in Refs 26,39. The value of κ_{IP} is close to, although lower than, , indicating that slight inplane phonon scattering at the interface is still active.
More interestingly, as W increases, the crossplane value κ_{CP} decreases steeply, reaches a minimum value at and next increases mildly at larger W. This Wdependence is consistent with previous predictions^{22,23,24,25,26,27,28} and corresponds to the crossover at W^{*} from coherent to incoherent phonon transport. In the incoherent regime, W > 20, from Eq. (1) and the data of κ_{CP} (dashed line in Fig. 2) we can extract the values of the interfacial resistance, R and the Kapitza length, , which are presented in Table 1.
Note that for m_{B}/m_{A} = 8 (Fig. 2(c)), we do not observe a clear thermal conductivity minimum. More precisely, even at the largest value W = 40, κ_{CP} is still orders of magnitude lower than , indicating that the interfacial resistance R results in a strong reduction of κ_{CP} in this range of W. Equivalently, the Kapitza length is significantly larger than the maximum period W = 40. Also, if , Eq. (1) can be approximated as , i. e., κ_{CP} increases linearly with W, with a slope, , which decreases by increasing the ratio m_{B}/m_{A}. For instance, it is for m_{B}/m_{A} = 2 and for m_{B}/m_{A} = 4. For m_{B}/m_{A} = 8, the slope is so small that κ_{CP} appears almost flat when plotted against W, for 20 ≤ W ≤ 40 (Fig. 2(c)).
The data shown in Fig. 2 demonstrate that κ_{CP} can be indeed lowered below the disordered alloy limit for m_{B}/m_{A} = 2 and even below the glass limit for higher mass heterogeneities, m_{B}/m_{A} = 4 and 8. These results are consistent with the experimental work of Ref. 18 and demonstrate that the interface formed between dissimilar materials effectively reduces κ_{CP}. It is also worth noting that the thermal conductivity tensor is very strongly anisotropic in this case, with .
The vibrational modes of the structure, i.e., the superlattice phonons, are key to understand the above behaviour of thermal conductivity. In Fig. 3 we show the vibrational density of states (vDOS), g(ω), for m_{B}/m_{A} = 4 and W = 2 to 80. g_{A}(ω) and g_{B}(ω) of the bulk crystals of type A and B as well as the vDOS of the glass and of the disordered alloy are also shown for comparison. Note that . At small W = 2, g(ω) of the superlattice roughly follows that of the disordered alloy, implying that the vibrational states in the two layers are strongly mixed. In this situation, phonons are able to propagate in both the cross and inplane directions. On the other hand, as W increases, g(ω) generates features increasingly similar to those identifying g_{A}(ω) and g_{B}(ω), separately. In particular, in the lowω region g(ω) follows g_{B}(ω) (the heavy crystal B), whereas g_{A}(ω) (the light crystal A) controls g(ω) in the highω region. This result indicates that different parts of the vibrational spectrum are active in the two layers, with high (low)ω modes preferentially excited in the light (heavy) layer A (B). In this situation, phonon propagation is largely obstructed in the crossplane direction, leading to the observed large reduction of κ_{CP}. We remark that phonons propagate in the inplane direction with small constraints, as shown by the large value of κ_{IP} close to . This implies that phonons, whose propagations are blocked in the crossplane direction, are actually specularly reflected at the interface and confined in the inplane direction.
The separation of the vibrational states found in the g(ω) becomes more clear when considering the vibrational amplitudes associated with the eigenstates k. In Fig. 4 we show the vibrational amplitudes, and (Eq. (6)), in the two layers A and B for each mode k, together with the binned average values (solid lines). Based on the relations and , we can define a relative degree of excitation of particles in the two layers, by the threshold value 0.5: large excitations correspond to , small excitations to . If , particle vibrations in both layers are of the same degree and correlated.
At small W = 2,4 we find, particularly in the lowω region, a large fraction of vibrational states with . As W increases, in the high frequency region , where is the highfrequency boundary in g_{B}(ω), only particles in the light layer A vibrate (), whereas those in the heavy layer B are almost immobile, as indicated by . In this ωregion, phonon propagation in the crossplane direction is therefore almost completely suppressed. On the other hand, for , particles pertaining to the heavy layer B show large vibrational amplitudes (), while vibrations in layer A tend to be small (). More in details, for W ≥ 20, we see that the averaged amplitudes are much larger in the B layer () than in the A layer () in the 2 < ω < 7.5 range. Contrary to the case of , however, a significant number of modes are excited in both layers A and B, even with . We therefore conclude that, for , some phonons still propagate in the crossplane direction, contributing to κ_{CP}.
We note that our observation of the vibrational separation in both the vDOS and vibrational amplitudes is consistent with results reported previously^{27,41,50}. Indeed, the simulation work of Ref. 27 reported a separation in the vDOS of the Si isotopicsuperlattice (^{28}Si^{42}Si superlattice). A recent simulation work^{41} focused on partial inverse participation ratios in a superlattice similar to the one considered here, reporting vibrational modes separation between layers. Ref. 50 attributed the reduction of thermal conductivity to a mechanism described as phonon localization, which we consider to be essentially the same phenomenon as the vibrational separation described here.
We believe that this concept of vibrational separation is a simple and accurate framework to rationalize the behaviour of thermal conductivity in superlattices. In particular, it provides a complete characterization of the minimum in the Wdependence of κ_{CP}. Indeed, in the range W = 2 to 20 identifying the coherent regime, the vibrational separation hinders the coherent phonon propagation in the crossplane direction, leading to the large reduction of κ_{CP}. In contrast, inplane phonon propagation is very mildly affected by the vibrational separation and, therefore, κ_{IP} keeps high values. Also, by considering and (solid lines), we conclude that the separation saturates to its maximum level at . Upon further increase W > 20, although averaged values show no significant changes, we recognize an increasing fraction of modes with and for (panels (e) W = 40 and (f) W = 80 in Fig. 4). This observation indicates that the separation tendency for modes with and becomes weaker, i.e., the correlation of vibrational features in the two layers decreases, which corresponds exactly to the incoherent transport picture and leads to the increase of κ_{CP}. Although transport becomes completely incoherent only for values of W of the order of the Kapitza length (note that for m_{B}/m_{A} = 4), this feature appears as soon as the vibrational separation is saturated, at the crossover point . Thus, the saturation point of the vibrational separation identifies the minimum value of κ_{CP}, which can be indeed below the glass limit.
S2. Superlattice composed of intercalated ordered crystalline layers and mass disordered alloy layers
This system consists of three components, with masses m_{A} = 1 in the crystalline layer A and m_{B1} and m_{B2} in the disordered alloy layer B. Note that also in this case, on the basis of the same arguments discussed above for S1, the average mass in layer B and that of the entire system are fixed as and , respectively, in order to eliminate the trivial contribution associated with different average values in the different cases.
In Fig. 5, we plot κ_{CP} and κ_{IP} for the mass ratios of the layer B, m_{B2}/m_{B1} = 2, 4 and 8. At small W ≤ 4, the values of both κ_{CP} and κ_{IP} are very close to those of the disordered bulk alloy formed by the same particles. As W increases, κ_{IP} increases gradually toward . This increase is controlled by the development of inplane phonon propagation in the ordered crystalline layer A. Indeed, the g(ω) of the superlattice, shown in Fig. 6, roughly follows that of the disordered bulk alloy at small W = 4, whereas at large W = 20,40 it is dominated by g_{A}(ω). In particular, the longitudinal peak around becomes clear, corresponding to that of the crystalline layer A.
The crossplane value κ_{CP} also increases with W, but reaches the limit value already at Since κ_{B} of the disordered alloy layer B is low (see Table 1), remains low, typically less than twice the disordered alloy value. As a result, the variation of κ_{CP} with W is small. This result indicates that scattering in the disordered alloy layer B dominates the thermal conduction in the crossplane direction. Both experimental work^{51} on Si(crystal)SiGe(disordered alloy) nanowires and numerical simulations^{40} have reported similar observations. We also note that the coherent nature of the superlattice phonons in the crossplane direction, which we observed in the S1 system, breaks down in S2. This is essentially equivalent to previous findings that disorder in interfacial roughness^{37,38,39}, or interfacial species mixing^{40,41} destroy the coherent features of vibrational excitations present in the investigated superlattices. As a consequence of these features, in superlattices of type S2 the variability of the crossplane heat transport is strongly bounded and the minimum limit of κ_{CP} just corresponds to the disordered alloy limit, i.e., κ_{CP} cannot be reduced below the glass limit.
Finally it is worth to mention that the thermal conductivity tensor becomes increasingly anisotropic at larger W due to the increase of κ_{IP}, showing a behaviour different than that observed in S1 where the anisotropy reaches the maximum at the crossover point .
S3. Superlattice composed by identical crystalline layers separated by weakly interacting interfaces
In Fig. 7 we show the Wdependences of κ_{CP} and κ_{IP} for the case where the energy scale associated with particles interactions across the interfaces (ε_{AB}) are lowered compared to those intralayers, with ε_{AB} = 0.5 and 0.1 in the two panels. In the figure, we also plot as lines the data for the corresponding onecomponent crystal and glass with unmodified interactions. Taking the one component system as reference is justified on the basis of earlier work^{8}. There, it was shown that when a system of soft spheres is frozen in a disordered state, slightly modifying even a substantial fraction of the interactions does not have any appreciable influence on the vibrational properties, including thermal transport. As a consequence, in the present case we do not expect any relevant modification to κ originating from a limited fraction (scaling with the surface of the interfaces) of modified interactions in the glass sample.
The inplane value κ_{IP} is almost independent of W and is very close to the value pertaining to the crystal. In contrast, κ_{CP} decreases monotonically by increasing W and especially in the weaker case ε_{AB} = 0.1, the observed reduction of κ_{CP} is dramatic. At W = 10, κ_{CP} equals the value obtained for the glassy sample and it is almost two orders of magnitude lower than this value at W = 20. This extremely low κ_{CP} is consistent with previous experimental work^{19}.
Some insight about the origin of this observation comes from the data shown in Fig. 8, where we display the average crossplane distance δz between adjacent crystalline planes (monolayers), normalized to the value in the perfect lattice, a/2. For ε_{AB} = 0.5 and W = 4, the system keeps the perfect lattice structure, with δz ≡ a/2 for all monolayers. In contrast, as ε_{AB} decreases and for a large value W = 20, δz becomes substantially larger than a/2 at the interfaces, which therefore assumes a local density lower than the average. At the same time, slightly reduced δz are also observed for the other intramonolayers, leading to an increase of the local density compared to the average. This heterogeneity hinders energy propagation across the interface and, as a result, phonons are specularly reflected and confined in the inplane direction. We remark that in the cases with W = 20, the values of δz at the interfaces located at W/2 = 10 and W = 20 are different, with a large discrepancy for ε_{AB} = 0.1. We rationalize this behaviour by observing that, during the preparation stage of the sample, the applied selective weakening of the interactions destabilizes the global equilibrium of the superlattice, with a concentration of mechanical stress close to the interfaces. Lattice planes far from the boundaries easily recover mechanical equilibrium by coherently reducing their mutual distance. In contrast, particles in monolayers adjacent to the interfaces move both outofplane and inplane, to optimize the local effective spring constants. The optimal solution found depends in general on the details of the local environment, explaining the observed discrepancy in δz at different interfaces.
The behaviour of κ_{CP} can be further elucidated by inspection of the main features of the vibrational spectrum. In Fig. 9 we plot the g(ω) of superlattice S3, together with the vibrational amplitudes and . The transverse and longitudinal phonon branches are observed in g(ω) (top panels) for all cases, similar to the homogeneous bulk crystal. For ε_{AB} = 0.5, W = 4, g(ω) shows an excess of lowerω modes compared to those present in the onecomponent crystal, simply due to the weakened interactions at the interfaces. As ε_{AB} decreases and W increases, g(ω) deforms, following the appearance of an increasing fraction of modes at increasing higher frequencies. This behaviour is certainly correlated to the observation made above (see Fig. 8) for the cases of ε_{AB} = 0.5, W = 20 and ε_{AB} = 0.1, W = 20, that the distance between monolayers far from the interfaces becomes smaller than a/2. The consequent larger mass density makes higher the frequency of phonon modes of given wavelength, leading to the shift of g(ω) towards higher frequencies.
We now focus on the vibrational amplitudes, and (Fig. 9, bottom panels). In the cases with ε_{AB} = 0.5 and W = 4 and 20, the particles in the two layers A and B show completely equivalent and correlated vibrations for the vast majority of the modes, as indicated by . This result implies that phonons indeed propagate across the weakened interfaces in the crossplane direction, but they are also partially reflected at the interface, causing the observed reduction of κ_{CP}. The situation changes drastically in the case ε_{AB} = 0.1 and W = 20, where the ultralow value of κ_{CP} can be reached. Except for the lowω modes, and are symmetrically randomly distributed around the average values , indicating that particles in layers A and B vibrate independently, in an uncorrelated manner. As a consequence, a very large fraction of vibrational modes do not cross at all the interfaces, but rather undergo a perfect specular reflection. In this situation, heat is not transferred between two adjacent layers A and B, leading to extremely low value of κ_{CP}, while keeping a high κ_{IP}.
We conclude by noticing that although specular reflection was also observed in the system S1, the physical mechanism behind this phenomenon is different in the two cases: vibrational separation causes reflection in the former, whereas weakened interactions across the interface, with the resulting augmented spacing between the layers, completely block crossplane phonon propagation in the latter.
Discussion
We have provided numerically, for the first time to our knowledge, a clear demonstration of very low thermal conductivities in superlattices, below the glassy limit of the corresponding amorphous structures. Blocking phonon propagation in ordered structures via interfaces design is the key principle. We have identified two possible strategies to achieve this goal: imposing a large mass heterogeneity in the intercalated layers (as in system S1) or degrading interlayers interactions compared to those intralayers (as in S3). We have found that in both cases phonons are specularly reflected at the interface and confined in the inplane direction. This reduces the crossplane thermal conductivity κ_{CP} below the corresponding glass limit, while keeping the inplane contribution κ_{IP} close to the pure crystalline value.
More specifically, in the case of mass mismatch (S1), propagation of phonons with high frequencies () is almost completely suppressed, whereas a fraction of lowfrequency phonons () are still able to propagate across the interfaces, contributing to κ_{CP} (Fig. 4(d–f)). Also, the minimum in thermal conductivity as a function of the repetition period W (Fig. 2) corresponds to a maximum in the vibrational separation between the layers of type A and B. These therefore act as true filters in complementary regions of the vibrational spectrum, suppressing significantly phonons transport in the direction of the replication pattern. On the other hand, attenuated interactions across the interfaces (S3) are able to block phonons at a very wide range of frequencies (see Fig. 9(c)), which results into extremely low values of κ_{CP}, even orders of magnitude lower than the corresponding glass limit (Fig. 7(b)). In this sense, directly modifying the interfaces seems to be the most effective strategy to obtain very low heat transfer. Note that this is a practically feasible route, since attenuated interfaces can be designed by exploiting materials with weak van der Waals forces among adjacent crystalline planes, as demonstrated in the case of WSe_{2} sheets in Ref. 19. Interfaces stiffness modification by controlling pressure^{44,45} or chemical bonding^{46} are additional possible routes to directly tune the strength of interfaces.
Our data also suggest that intercalating disordered alloy layers in ordered crystalline layers (S2) is not effective in lowering κ_{CP}. Indeed, we have demonstrated that in this case disorder is not sufficient to block the propagation of vibrational excitations, even though it makes phonons lifetimes short. The intercalated disordered alloy layer dominates phonon transport in the entire superlattice, notwithstanding the presence of the crystalline layers. As a result, thermal conductivity is very similar to the one of the disordered alloy and is only marginally modified by modulation of the period W (see Fig. 5). Also, as suggested in previous works, disorder in the interfacial roughness^{37,38,39} or interfacial mixing^{40,41} seems to already dominate over phonon transport and destroy the coherent nature of phonons.
In addition, as we understand from our analysis of vibrational amplitudes (Figs 4 and 9), it is much more problematic to block lowω (long wavelength, λ) phonons propagation, than those with highω (short λ). This situation is similar to what has been observed in bulk glasses, where the longλ acoustic waves are not scattered by the disorder and can propagate over long distances by carrying heat energy^{49,52}. Therefore, blocking or efficiently scattering the longλ phonons is also a key factor to achieve very low thermal conductivities, as was pointed out in Ref. 53. A possibility to realize this task is embedding in the targeted material objects featuring larger typical sizes, including nanoparticles^{54,55} or nano(quantum)dots^{20,56}. Based on this strategy, very low thermal conductivity was achieved experimentally in a SiGe quantumdot superlattice^{20}, even below the amorphous Si value. The additional possibility of introducing large size defects by the porous structuring of materials has also been explored in a recent numerical work^{57}. Here, values of thermal conductivity 10^{4} times smaller than that of bulk Si were reached in Si phononic crystals with spherical pores.
An other important point must be underlined. In the present work we have investigated the different systems at low temperature, to both focus on plainly structural effects and keep contact with the well controlled harmonic limit. Anharmonicities, however, are expected to play a crucial role at temperatures higher than the Debye temperature, T_{D}, which are those relevant for technological applications. This is a crucial aspect to be explored in extended future work.
In conclusion, we note that the three superlattice structures studied in the present work show totally different Wdependences of cross and inplane thermal conductivities. Our results therefore not only contribute to a deeper understanding of the physical mechanisms behind verylow thermal conductivity, they also provide insight for developing new design concepts for materials with controlled heat conduction behaviour.
Methods
System description
In this Section we provide details on the numerical models we have used for the superlattices. The corresponding amorphous structures (glasses) and disordered alloys with exactly the same composition were also prepared, for the sake of comparison with superlattice phases. We have considered in all cases a 3dimensional cubic box, of volume V = L^{3} (L being the linear box size), with periodic boundary conditions in all directions. In the superlattice and disordered alloy cases, particles were distributed on the FCC lattice sites. In the glass phases, they were frozen in topologically random positions following a rapid quench from the normal liquid phase below the glass transition temperature T_{g}, avoiding crystallization (see, for instance, Ref. 49 for details on the preparation of glasses). Particles, i and j, interact via softsphere (SS) or LennardJones (LJ) potentials:
where r is the distance between those two particles and σ^{ij} and ε^{ij} are the interparticle diameter and interaction energy scale, respectively. The potential is cutoff and shifted at r_{c} = 2.5σ^{ij}. Particle i has mass m^{i} and we have used σ, ε/k_{B} (k_{B} is the Boltzmann constant) and m as units of length, temperature and mass. As a reference, for Argon σ = 3.4 Å, ε/k_{B} = 120 K and m = 39.96 a.m.u. We considered the number density , corresponding to a lattice constant .
We prepared three superlattices, composed of intercalated FCC lattice layers, A and B, both of thickness W/2, as schematically illustrated in Fig. 1. The first superlattice (S1) consists of two crystalline layers formed by point particles with different masses, m_{A} and m_{B}. We have considered mass ratios m_{B}/m_{A} > 1, while keeping a constant average mass (m_{A} + m_{B})/2 = 1. As an example, the case m_{B}/m_{A} = 4 corresponds to m_{A} = 0.4 and m_{B} = 1.6. We have dubbed A and B as the light and heavy layers, respectively. Note that a mass ratio of m_{B}/m_{A} = 2.5 corresponds to the case of the realistic SiGe superlattice. Except for the above mass difference in the different layers, all particles are characterized by the same properties. In particular, they interact via the SS potential , with σ^{ij} = ε^{ij} = 1.
The second superlattice (S2) is composed of an ordered crystalline layer A intercalated to a disordered alloy layer B. m_{A} = 1 in A, whereas in B half of the particles have mass m_{B1}, m_{B2} the others and are randomly distributed on the lattice sites. Again, m_{B1} and m_{B2} are determined by the mass ratio m_{B2}/m_{B1}> 1, keeping a constant average value (m_{B1} + m_{B2})/2 = 1. All particles in both layers interact via the SS potential with σ^{ij} = ε^{ij} = 1.
The third superlattice (S3) is composed of identical crystalline layers A and B, but the interactions among particles in different layers (i.e., across the interfaces) are modified (weakened) compared to those intralayers. All particles have mass m_{A} = m_{B} = 1 and interact via the LJ potential , with σ^{ij} = ε^{ij} = 1. The energy scale of interactions between particles pertaining to different layers are, however, reduced to ε^{ij} = ε_{AB} < 1.
MD simulation and the GreenKubo method for the calculation of thermal conductivity
In the present study, all simulations have been realized by using the largescale, massively parallel molecular dynamics simulation tool LAMMPS^{58}. The systems were first equilibrated at relatively low temperature T = 10^{−2} by MD simulation in the NVTensemble. This choice was dictated by the need to reduce anharmonic effects, in order to primarily focus on the contribution of the structural features of the superlattices on thermal conductivity. We must note that our approach is classical and does not take into account the quantum mechanisms active in the lowT regime^{9}. These effects have important implications, increasing the contribution to the thermal conductivity coming from lowω vibrational excitations. At present, however, it is not obvious and still under debate how to effectively include quantum effects into a classical system^{59,60} and we have therefore chosen to stay within a fully classical approach.
Following the equilibration stage, we performed the production runs in the NVEensemble. The GreenKubo formulation^{47,48} was next applied to calculate the thermal conductivities, in the crossplane and inplane directions, respectively:
Here, J_{x,y} and J_{z} are the heat currents in the inplane (x,y) and crossplane (z) directions and 〈〉 denotes the ensemble average. Landry et al.^{48} have carefully confirmed the validity of the GreenKubo method for the calculation of superlattices thermal conductivity, by comparison with the direct method based on nonequilibrium simulation. In the bulk glasses and disordered alloys, , i.e., heat conduction is isotropic, whereas in the superlattices, they are expected to assume different values^{16,17}.
More in details, the equations of motions were integrated numerically with a time step, δt = 5 × 10^{−3}, for a total runtime t_{run} = N_{run}δt = 10^{5} (N_{run} = 2 × 10^{7} steps) for S1 and t_{run} = 10^{4} (N_{run} = 2 × 10^{6} steps) for both S2 and S3. The systems snapshots extracted from the trajectory have been used to calculate the correlation functions, of Eq. (4), which have been subsequently integrated numerically over a finite time window Δt. We have observed a clear convergence of the integrals for κ_{CP,IP}, to the limits of Eq. (4) for Δt = 10^{4} for S1 and Δt = 10^{3} for both S2 and S3. For each considered superlattice and all repetition periods W, we have performed 10 independent calculations starting from different initial system configurations. These has allowed us to obtain 10 independent sample values for κ_{CP} and 2 × 10 = 20 (we have considered both the x and y components) for κ_{IP}. These values were used to calculate averages and sampletosample fluctuations (standard deviations), which are shown as error bars in Figs. 2,5 and 7. More details on these calculations are reported in the Supplementary Materials.
Analysis of finite system size effect on thermal conductivity
In the GreenKubo calculations of thermal conductivities, one must be attentive to finite system size effects^{47,48}. Indeed, longwavelength phonons with λ > L are excluded from the simulation box due to the finite value L of the box size, which imposes important size effects on the numerical determination of κ. The box size therefore needs to be large enough to include a vibrational spectrum sufficient to establish an accurate description of anharmonic coupling (scattering) processes^{47}. We note that the considered T = 10^{−2} is low enough to substantially reduce anharmonic effects, but anharmonic couplings are still present. We can take care of finite size effects by increasing L to values where κ_{CP} and κ_{IP} become Lindependent. For the glass and disordered alloy thermal conductivities, we have confirmed that a system size L = 10a (N = 4,000) is sufficiently large to obtain correct values of , without any size effect^{8}.
In the superlattice cases, the appropriate L depends on the considered structure and the periodic repetition length W^{48}. More in details, we paid particular attention to the number P of repetitions, defined from L = PW, necessary to produce sufficient anharmonic couplings of phonons in the crossplane direction. We have therefore investigated the presence of finite size effects by analyzing different systems with sizes ranging from L = 10a (20 monolayers, N = 4,000) to 24a (48 monolayers, N = 55,296). We provide details of our analysis of the finitesize effects in the Supplementary Materials. In Figs. 2,5 and 7, we plot the values obtained by using the largest systems (the exact system size depends on both the superlattice type and W), which show the smallest finitesize effects.
For the S1 superlattice, we confirmed that the required number P of repetitions becomes larger for smaller W^{48}: one period (L = W) only is adequate for W ≥ 20, whereas four periods or more (L ≥ 4W) are required for W ≤ 8 (see the Supplementary Materials). We have therefore employed four pattern repetitions (L = 4W) for 10 ≤ W ≤ 12 and two (L = 2W) for 14 ≤ W ≤ 18. This behaviour is simple to rationalize by inspecting the data in Fig. 2, where the crossover between incoherent and coherent phonon transport occurs around . In the coherent regime W < 20, the wave character of the phonons becomes important and therefore a larger number of repetitions is necessary to produce the coherent wave interference processes correctly. In contrast, smaller values of P are needed (even P = 1) in the incoherent regime W > 20, where the incoherent particle nature of the phonons appears.
For the S2 and S3 superlattices the system size effects issue is much less pronounced than in the S1 case. We have confirmed that P = 1 or 2 (L = W or 2W) are sufficient for W ≥ 20, while two or more repetitions (L ≥ 2W) are appropriate for W < 20, for both S2 and S3 (see the Supplementary Materials). We can understand this behaviour by noticing that phonon transport is mainly determined by the scattering processes in the disordered alloy layer in S2 and the blocking at the weak interface for S3. In both cases the missing long wavelength phonons, with λ > L, play very little role in phonon transport and finite system size effects are consequently negligible.
Normal modes analysis
We have characterized the superlattice vibrational states (superlattice phonons) by performing a standard normalmode analysis^{8,49}. We have diagonalized the dynamical (Hessian) matrix calculated at local minima of the potential energy landscape and obtained eigenvalues λ^{k} and eigenvectors (polarization vectors) . Here, j is the particle index and k = 1, 2,…, 3N − 3 is the eigenmode number, where we have disregarded the three vanishing Goldstone modes. The eigenvectors are normalized such that , where δ_{kl} is the Kronecker delta function. The eigenfrequencies are next calculated as and the associated probability distribution (normalized histogram) directly provides the vDOS:
In addition, from the eigenvector e^{k} we have defined the vibrational amplitudes of mode k for layers A and B:
Note that for each k and, therefore, . Based on the values of and , one can determine in which layer particles are more displaced (excited) according to the associated eigenvector e^{k}. In particular, if (), particles in layer A (B) contribute more to mode k than those in layer B (A). In the case , particles in both layers contribute equivalently and in a correlated manner. Note that the normal mode analysis provides us with the system vibrational states in the harmonic limit T → 0 which, we believe, is an appropriate approximation for our case T = 10^{−2}, where anharmonicities are weak.
Additional Information
How to cite this article: Mizuno, H. et al. Beating the amorphous limit in thermal conductivity by superlattices design. Sci. Rep. 5, 14116; doi: 10.1038/srep14116 (2015).
References
Venkatasubramanian, R., Siivola, E., Colpitts, T. & O’Quinn, B. Thinfilm thermoelectric devices with high roomtemperature figures of merit. Nature 413, 597–602 (2001).
Minnich, A. J., Dresselhaus, M. S., Ren, Z. F. & Chen, G. Bulk nanostructured thermoelectric materials: current research and future prospects. Energy Environ. Sci. 2, 466–479 (2009).
Maldovan, M. Sound and heat revolutions in phononics. Nature 503, 209–217 (2013).
Goodson, K. E. Ordering up the minimum thermal conductivity of solids. Science 315, 342–343 (2007).
Cahill, D. G. & Pohl, R. O. Lattice vibrations and heat transport in crystals and glasses. Ann. Rev. Phys. Chem. 39, 93–121 (1988).
Cahill, D. G., Watson, S. K. & Pohl, R. O. Lower limit to the thermal conductivity of disordered crystals. Phys. Rev. B 46, 6131–6140 (1992).
Allen, P. B. & Feldman, J. L. Thermal conductivity of disordered harmonic solids. Phys. Rev. B 48, 12581–12588 (1993).
Mizuno, H., Mossa, S. & Barrat, J.L. Elastic heterogeneity, vibrational states and thermal conductivity across an amorphisation transition. EPL 104, 56001 (2013).
Kittel, C. Introduction to Solid State Physics (John Wiley and Sons, New York, 1996), 7th edn.
Mizuno, H., Mossa, S. & Barrat, J.L. Measuring spatial distribution of the local elastic modulus in glasses. Phys. Rev. E 87, 042306 (2013).
Hopkins, P. E. et al. Reduction in the thermal conductivity of single crystalline silicon by phononic crystal patterning. Nano Lett. 11, 107–112 (2011).
Lee, S.M., Cahill, D. G. & Venkatasubramanian, R. Thermal conductivity of SiGe superlattices. Appl. Phys. Lett. 70, 2957–2959 (1997).
Volz, S., Saulnier, J. B., Chen, G. & Beauchamp, P. Computation of thermal conductivity of Si/Ge superlattices by molecular dynamics techniques. Microelectronics Journal 31, 815–819 (2000).
Capinski, W. S. et al. Thermalconductivity measurements of GaAs/AlAs superlattices using a picosecond optical pumpandprobe technique. Phys. Rev. B 59, 8105–8113 (1999).
Daly, B. C. & Maris, H. J. Calculation of the thermal conductivity of superlattices by molecular dynamics simulation. Phys. B: Condens. Matter 316317, 247–249 (2002).
Yang, B., Liu, W. L., Liu, J. L., Wang, K. L. & Chen, G. Measurements of anisotropic thermoelectric properties in superlattices. Appl. Phys. Lett. 81, 3588–3590 (2002).
Mavrokefalos, A., Nguyen, N. T., Pettes, M. T., Johnson, D. C. & Shi, L. Inplane thermal conductivity of disordered layered Wse2 and (W)x(Wse2)y superlattice films. Appl. Phys. Lett. 91, 171912 (2007).
Costescu, R. M., Cahill, D. G., Fabreguette, F. H., Sechrist, Z. A. & George, S. M. Ultralow thermal conductivity in W/Al2O3 nanolaminates. Science 303, 989–990 (2004).
Chiritescu, C. et al. Ultralow thermal conductivity in disordered, layered Wse2 crystals. Science 315, 351–353 (2007).
Pernot, G. et al. Precise control of thermal conductivity at the nanoscale through individual phononscattering barriers. Nature Mater. 9, 491–495 (2010).
Wang, Y., Gu, C., Ruan, X. Optimization of the random multilayer structure to break the randomalloy limit of thermal conductivity. Appl. Phys. Lett. 106, 073104 (2015).
Simkin, M. V. & Mahan, G. D. Minimum thermal conductivity of superlattices. Phys. Rev. Lett. 84, 927–930 (2000).
Yang, B. & Chen, G. Partially coherent phonon heat conduction in superlattices. Phys. Rev. B 67, 195311 (2003).
Garg, J. & Chen, G. Minimum thermal conductivity in superlattices: A firstprinciples formalism. Phys. Rev. B 87, 140302 (2013).
Chen, Y., Li, D., Lukes, J. R., Ni, Z. & Chen, M. Minimum superlattice thermal conductivity from molecular dynamics. Phys. Rev. B 72, 174302 (2005).
Kawamura, T., Kangawa, Y. & Kakimoto, K. An investigation of thermal conductivity of nitridesemiconductor nanostructures by molecular dynamics simulation. J. Crystal Growth 298, 251–253 (2007).
Yang, N., Zhang, G. & Li, B. Ultralow thermal conductivity of isotopedoped silicon nanowires. Nano Lett. 8, 276–280 (2008).
Ravichandran, J. et al. Crossover from incoherent to coherent phonon scattering in epitaxial oxide superlattices. Nature Mater. 13, 168–172 (2014).
Chen, G. & Neagu, M. Thermal conductivity and heat transfer in superlattices. Appl. Phys. Lett. 71, 2761–2763 (1997).
Chen, G. Thermal conductivity and ballisticphonon transport in the crossplane direction of superlattices. Phys. Rev. B 57, 14958–14973 (1998).
Kim, E.K., Kwun, S.I., Lee, S.M., Seo, H. & Yoon, J.G. Thermal boundary resistance at Ge2Sb2Te5/ZnS:SiO2 interface. Appl. Phys. Lett. 76, 3864–3866 (2000).
Lampin, E., Nguyen, Q.H., Francioso, P. A. & Cleri, F. Thermal boundary resistance at siliconsilica interfaces by molecular dynamics simulations. Appl. Phys. Lett. 100, 131906 (2012).
Chen, J., Walther, J. H. & Koumoutsakos, P. Strain Engineering of Kapitza Resistance in FewLayer Graphene. Nano Lett. 14, 819–825 (2014).
Barrat, J.L., & Chiaruttini, F. Kapitza resistance at the liquidsolid interface. Mol. Phys. 101, 1605–1610 (2003).
Tamura, S., Hurley, D. C. & Wolfe, J. P. Acousticphonon propagation in superlattices. Phys. Rev. B 38, 1427–1449 (1988).
Ren, S. Y. & Dow, J. D. Thermal conductivity of superlattices. Phys. Rev. B 25, 3750–3755 (1982).
Daly, B. C., Maris, H. J., Imamura, K. & Tamura, S. Molecular dynamics calculation of the thermal conductivity of superlattices. Phys. Rev. B 66, 024301 (2002).
Imamura, K., Tanaka, Y., Nishiguchi, N., Tamura, S. & Maris, H. J. Lattice thermal conductivity in superlattices: molecular dynamics calculations with a heat reservoir method. J. Phys.: Condens. Matter 15, 8679–8690 (2003).
Daly, B. C., Maris, H. J., Tanaka, Y. & Tamura, S. Molecular dynamics calculation of the inplane thermal conductivity of GaAs/AlAs superlattices. Phys. Rev. B 67, 033308 (2003).
Landry, E. S. & McGaughey, A. J. H. Effect of interfacial species mixing on phonon transport in semiconductor superlattices. Phys. Rev. B 79, 075316 (2009).
Huberman, S. C., Larkin, J. M., McGaughey, A. J. H. & Amon, C. H. Disruption of superlattice phonons by interfacial mixing. Phys. Rev. B 88, 155311 (2013).
Termentzidis, K., Chantrenne, P. & Keblinski, P. Nonequilibrium molecular dynamics simulation of the inplane thermal conductivity of superlattices with rough interfaces. Phys. Rev. B 79, 214307 (2009).
Termentzidis, K., Merabia, S., Chantrenne, P. & Keblinski, P. Crossplane thermal conductivity of superlattices with rough interfaces using equilibrium and nonequilibrium molecular dynamics. International Journal of Heat and Mass Transfer 54, 2014–2020 (2011).
Hsieh, W.P., Lyons, A. S., Pop, E., Keblinski, P. & Cahill, D. G. Pressure tuning of the thermal conductance of weak interfaces. Phys. Rev. B 84, 184107 (2011).
Shen, M., Evans, W. J., Cahill, D. & Keblinski, P. Bonding and pressuretunable interfacial thermal conductance. Phys. Rev. B 84, 195432 (2011).
Losego, M. D., Grady, M. E., Sottos, N. R., Cahill, D. G. & Braun, P. V. Effects of chemical bonding on heat transport across interfaces. Nature Mater. 11, 502–506 (2012).
McGaughey, A. J. H. & Kaviany, M. Phonon transport in molecular dynamics simulations: formulation and thermal conductivity prediction. Advances in Heat Transfer 39, 169–255 (2006).
Landry, E. S., Hussein, M. I. & McGaughey, A. J. H. Complex superlattice unit cell designs for reduced thermal conductivity. Phys. Rev. B 77, 184302 (2008).
Monaco, G. & Mossa, S. Anomalous properties of the acoustic excitations in glasses on the mesoscopic length scale. Proc. Natl. Acad. Sci. USA 106, 16907–16912 (2009).
Yang, L., Yang, N. & Li, B. Reduction of thermal conductivity by nanoscale 3d phononic crystal. Sci. Rep. 3, 1143 (2012).
Li, D., Wu, Y., Fan, R., Yang, P. & Majumdar, A. Thermal conductivity of Si/SiGe superlattice nanowires. Appl. Phys. Lett. 83, 3186–3188 (2003).
Mizuno, H., Mossa, S. & Barrat, J.L. Acoustic excitations and elastic heterogeneities in disordered solids. Proc. Natl. Acad. Sci. USA 111, 11949–11954 (2014).
Luckyanova, M. N. et al. Coherent phonon heat conduction in superlattices. Science 338, 936–939 (2012).
Kim, W. et al. Thermal conductivity reduction and thermoelectric figure of merit increase by embedding nanoparticles in crystalline semiconductors. Phys. Rev. Lett. 96, 045901 (2006).
Zhang, H. & Minnich, A. J. The best nanoparticle size distribution for minimum thermal conductivity. Sci. Rep. 5, 8995 (2015).
Nika, D. L. et al. Reduction of lattice thermal conductivity in onedimensional quantumdot superlattices due to phonon filtering. Phys. Rev. B 84, 165415 (2011).
Yang, L., Yang, N. & Li, B. Extreme low thermal conductivity in nanoscale 3D Si phononic crystal with spherical pores. Nano Lett. 14, 1734–1738 (2014).
Plimpton, S. Fast parallel algorithms for shortrange molecular dynamics. J. Comput. Phys. 117, 1–19 (1995). Http://lammps.sandia.gov.
Turney, J. E., McGaughey, A. J. H. & Amon, C. H. Assessing the applicability of quantum corrections to classical thermal conductivity predictions. Phys. Rev. B 79, 224305 (2009).
BedoyaMartinez, O. N., Barrat, J.L. & Rodney, D. Computation of the thermal conductivity using methods based on classical and quantum molecular dynamics. Phys. Rev. B 89, 014303 (2014).
Acknowledgements
We thank P. Keblinski for helpful correspondence. This work was supported by the Nanosciences Foundation of Grenoble. J.L. B is supported by the Institut Universitaire de France. Most of the computations presented in this work were performed using the Froggy platform of the CIMENT infrastructure (https://ciment.ujfgrenoble.fr), which is supported by the RhôneAlpes region (GRANT CPER07_13 CIRA) and the Equip@Meso project (reference ANR10EQPX2901) of the programme Investissements d’Avenir supervised by the Agence Nationale pour la Recherche.
Author information
Authors and Affiliations
Contributions
H.M., S.M. and J.L.B. designed research, performed research, analyzed data and wrote the paper.
Ethics declarations
Competing interests
The authors declare no competing financial interests.
Electronic supplementary material
Rights and permissions
This work is licensed under a Creative Commons Attribution 4.0 International License. The images or other third party material in this article are included in the article’s Creative Commons license, unless indicated otherwise in the credit line; if the material is not included under the Creative Commons license, users will need to obtain permission from the license holder to reproduce the material. To view a copy of this license, visit http://creativecommons.org/licenses/by/4.0/
About this article
Cite this article
Mizuno, H., Mossa, S. & Barrat, JL. Beating the amorphous limit in thermal conductivity by superlattices design. Sci Rep 5, 14116 (2015). https://doi.org/10.1038/srep14116
Received:
Accepted:
Published:
DOI: https://doi.org/10.1038/srep14116
This article is cited by

The Quest for HighEfficiency Thermoelectric Generators for Extracting Electricity from Waste Heat
JOM (2021)

Size effect in thermoelectric materials
npj Quantum Materials (2016)
Comments
By submitting a comment you agree to abide by our Terms and Community Guidelines. If you find something abusive or that does not comply with our terms or guidelines please flag it as inappropriate.