Delaram
Rashadfar
a,
Brandi L.
Wooten
*b and
Joseph P.
Heremans
*abc
aDepartment of Mechanical and Aerospace Engineering, Ohio State University, Columbus, OH 43210, USA. E-mail: brandi.l.wooten.civ@army.mil
bDepartment of Materials Science and Engineering, Ohio State University, Columbus, OH 43210, USA
cDepartment of Physics, Ohio State University, Columbus, OH 43210, USA
First published on 14th March 2025
In ferroelectric materials, an electric field has been shown to change the phonon dispersion sufficiently to alter the lattice thermal conductivity, opening the possibility that a heat gradient could drive a polarization flux, and technologically, also opening a pathway towards voltage-driven, all solid-state heat switching. In this report, we confirm the validity of the theory originally developed for Pb(Zr,Ti)O3 (PZT) on the ferroelectric relaxor 0.67Pb[Mg1/3Nb2/3]O3–0.33PbTiO3 (PMN–33PT). In theory, the change in sound velocity and thermal conductivity with an electric field relates to the piezoelectric coefficients and the Grüneisen parameter. It predicts that in PMN–33PT the effect should be an order of magnitude larger and of opposite sign as in PZT; this is confirmed here experimentally. The effects are measured on samples never poled before and on samples that underwent multiple field sweep cycles and passed through two phase transitions with change in temperature. The thermal conductivity changes are linked to variations in the piezoelectric coefficients and can be as large as 8–11% at T ≥ 300 K. To date, this has been the only means of heat conduction modulation that utilizes changes in the phonon spectrum. While this technology is in its infancy, it offers another path to future active thermal conduction control.
New conceptsThis is the second experimental paper in the new field of polarization caloritronics describing the transport of the thermal fluctuations (labeled ferrons) of the polarization responsible for the decrease of the polarization with temperature in ferroelectrics. Ferrons are a subset of phonons that involve the motion of atoms whose charges are responsible for the polarization in ferroelectrics; they are optical phonons strain-coupled to acoustic phonons. They carry heat and polarization fluxes much like magnons do in ferromagnets. The present article reports the electric field dependence of the thermal conductivity of a relaxor ferroelectric, PMN–PT. It shows how a theory developed for ferron transport in PZT applies equally to the relaxor ferroelectric PMN–PT. The novelty and importance of the present paper are that it shows the theory to be predictive: PMN–PT was selected to maximize the effect, which proved indeed up to 2 orders of magnitude larger than in PZT. This is a new approach to the realization of electrically driven all-solid-state heat switches, based on electric-field driven modifications of the phonon spectrum in ferroelectrics. Furthermore, polarization fluxes could potentially lead to a cornucopia of new applications of voltage-driven phenomena including thermal transistors, new thermoelectric concepts, and phonon-based logic and memory devices. |
−jP = σ∇E − Sσ∇T | (1) |
−jQ = σΠ∇E − κ∇T | (2) |
For FMs, the thermal fluctuations of magnetization involve spin waves or magnons, while in FEs, it relates to a specific type of phonon called ferrons, as identified by Bauer et al.1 Like how MS decreases with rising temperature due to an increase in magnon population, PS follows suit due to a higher density of ferrons. Intuitively, in displacement ferroelectrics, ferrons include optical phonons that involve the displacement of the ions that give rise to atomic dipoles. However, the distinction between acoustic and optical modes remains well-defined only at zero wavevector (k); at non-zero k, the optical and acoustic phonons couple due to strain. Ferrons are thus optical and acoustic modes strain-coupled by the piezoelectric coefficients. In principle, one then expects an electric field dependence of the phonon dispersion and hence the lattice thermal conductivity. The former effect is observed in 0.70Pb[Mg1/3Nb2/3]O3–0.30PbTiO3 (PMN–30PT)2 and the latter in Pb(Zr,Ti)O3 (PZT)3 where the field is shown to affect the lattice thermal conductivity by about 2%. The present article strongly amplifies the earlier work of Wooten et al.3 by showing effects one and two orders of magnitude larger in the field-dependent thermal conductivity of PMN–33PT.
A quantitative theory3 for the dependence of sound velocity, lattice thermal conductivity and diffusivity with no adjustable parameters relates changes in the sound velocity v to the piezoelectric coefficients d33 and d31 and the Grüneisen parameter γ. Fundamentally, the theory is based on the change of the phonon dispersions with the change in the volume of the sample, which is characterized by the Grüneisen parameter. The applied electric field then changes the volume of the sample as characterized by the piezoelectric coefficients. Putting both together will then predict the change in sound velocity and thus in the lattice thermal conductivity. Quantitatively, piezoelectric coefficients are defined as and
, where e33(11) is the strain tensor component that gives the compression or expansion along (perpendicular to) the direction of ferroelectric order, which is parallel to the applied field E = E3. The mode (i) and wavevector (k)-dependent Grüneisen parameter is the logarithmic derivative of that phonon mode's frequency vis-à-vis the volume V of the sample,
, by definition. For longitudinal or transverse acoustic phonons at low k (i = LA or TA, respectively),
. The basis of the theory is to first describe how the piezoelectric coefficients relate the change in volume of the sample with the applied field to (d33 + 2d31) and then relate that to the logarithmic derivative of the sound velocity for each mode:
![]() | (3) |
Here, is the derivative of the sound velocity with respect to the applied electric field and γ is the specific heat-weighted mode and wavevector average of γ(i,k). From the sound velocity we move to the thermal conductivity using the equation
, where C represents the volumetric specific heat, l the mean free path, and τ the relaxation time; here, v is the average velocity of the LA and 2 TA modes. At temperatures above the Debye temperature (ΘD ∼150–180 K for PMN–PT4; this is the Debye temperature for the monoclinic (M) phase, and since the differences between the MA and MC phases are only in the direction of the polarization vector, no significant difference is expected), the specific heat is constant with E being a function of only the number of atoms per unit volume n, C = 3nkB (kB is the Boltzmann constant).
In a first approximation, above 200 K, we hypothesize that the scattering time is unaffected by the electric field, based on temperature-dependent measurements by Mante and Folger on BaTiO3.5 There, below 10 K, the phonon mean free path is on the order of the domain size and since this property is field sensitive, the thermal conductivity also depends on the electric field. This effect vanishes at higher temperatures where phonon–phonon scattering dominates. Assuming that the relaxation time is independent of the electric field (T ≫ 10 K), we can relate the electric field dependency of both the sound velocity and thermal conductivity κ above the Debye temperature as:
![]() | (4) |
While this approximation holds well for PZT,3 it is more questionable for PMN–PT. Indeed, neutron scattering data on PMN-30PT2 show both strong softening of the lower section of the TA phonon branch and a marked decrease in phonon lifetime. In this case,
![]() | (5) |
If eqn (3) and (4) have predictive values, they allow us to sort amongst known ferroelectrics for the compound that has the largest field dependence of thermal conductivity and is thus most useful in thermal switching applications. The primary application for modulating the heat conduction in FE materials lies in the implementation of voltage-driven heat switches. Heat switches, or thermal transistors, are a particularly useful element in heat management,6 which is key to the continued scaling up of the complexity in advanced semiconductor integrated circuits. They are also the key elements in all-solid-state power generation and refrigeration cycles based on electrocaloric and magnetocaloric effects, where no fluid is circulated but only heat. Many heat switching technologies exist,7 including one that is based on domain wall motion in ferroelectrics8 and another on the electrophononic effect based on the lowering of symmetry in ferroelectrics;9–11 however, the mechanism exploited here, the phonon spectrum shifts via an electric field, promises to give reasonable effects over a large temperature range if maximized. Ultimately, the goal is to create an electrically modulated, voltage-driven thermal switch with the largest switching ratio.
Eqn (3) sheds light on the material parameters needed to maximize the thermal modulation – very anharmonic phonons denoted by a large Grüneisen parameter γ and maximization of the summation (d33 + 2d31). The relaxor ferroelectric solid solution system Pb[Mg1/3Nb2/3]O3–PbTiO3 (PMN–PT) is known to exhibit enhanced piezoelectricity. In this work, we investigate the thermal conductivity modulation of PMN–PT using an electric field with the specific composition of (1 − x)Pb[Mg1/3Nb2/3]O3–xPbTiO3 with x estimated to be between 0.3 and 0.33 (further abbreviated as PMN–33PT throughout the text). The main practical and commercial advantages of PMN–PT over PZT are its much larger piezoelectric coefficient. This is the reason we picked it for this study, since our theory predicted that this should lead to a much stronger dependence of the thermal conductivity on the electric field. In the uncycled sample, we prove this is indeed the case, and the theory is followed quantitatively. To our surprise, this sensitivity increases further to 10% in the cycled material. Our model did not predict that, and we will discuss its possible origin. The disadvantage of PMN–PT is the fact that it undergoes various phase transitions with temperature and an applied electric field, which complicated the sample-to-sample reproducibility of the properties of the cycled samples, a difficulty already described by the wide variation in the reported values of d33 of commercially cycled PMN–33PT (d33 ranging from 1300 to 2500 pC/N in the literature).
The absolute value error of thermal conductivity near room temperature in the static heater and sink method is on the order of 10%. It is dominated by the radiative heat losses and by geometrical uncertainties. The latter is typically dominated by the distance between the thermocouples, where the error is given by the ratio between the size of the thermocouple contacts (which gives the uncertainty) and the distance between them. Here, this uncertainty is calculated to be 1.4–1.8%. In the cryostat, the radiative heat losses correspond to the thermal conductance that has been measured using a standard iron bar of known thermal conductivity; they are 0.10 mW K−1 at 200 K, 0.44 mW K−1 at 300 K and 0.85 mW K−1 at 360 K. Given the thermal conductivity and the geometry of the PMN–33PT samples, the losses represent <7% at 200 K, <29% at 300 K but <55% at 400 K. These losses have been subtracted from the measured conductance, but that procedure carries an uncertainty due to the difference in emissivity between the sample and the iron standard, which itself has an error of the order of 30%. Taken together, the radiative losses represent an uncertainty in the absolute values of 2% at 200 K, 10% at 300 K, but up to 18% at 360 K. The plots shown in Fig. 4 and Fig. S2 and S3 (ESI†) represent the field-dependence of the conductivity, which is only affected by the relative error. This is much smaller and dominated by the noise on the thermocouples, equivalent to about 10 mK or 1% of the temperature difference and thus the measured conductivity.
A parallel experiment was carried out using the cycled sample. The objective of this second experiment was to elucidate the impact of cycling the sample through several sweeps of the electric field. The mounting of the sample for this experiment was slightly different from that of the first measurement. The laser beam hit the strain gauge placed on the top of the sample to measure the longitudinal displacement while applying the electric field. The d31 coefficient was calculated by taking the slope of the strain vs. electric field.
![]() | ||
Fig. 1 Temperature dependent phase transitions observed in thermal conductivity measurements at zero field taken during field sweeps. In the legend, ‘1st zero’ refers to the thermal conductivity value at zero field as the swept field is decreasing, and ‘2nd zero’ refers to the value at zero field as the swept field is increasing (see later, Fig. 4). MA, MC, and T refer to the two monoclinic and tetragonal phases that PMN–33PT passes through as the temperature changes. |
These phases are also dependent on the electric field. For example, if the sample is slightly above room temperature, it can transition from MA to MC with an increasing field. These transitions are obvious in the strain vs. electric field curves found in Davis et al.12 Ultimately, these transitions are a function of the temperature, electric field, synthesis protocols, composition, and poling conditions.12,13 For example, PMN–33PT has a large piezoelectric coefficient ∼d33 = 1500 pC N−1;14 a recent paper13 related the poling conditions to d33 and found that d33 can change drastically with the poling electric field strength and AC or DC poling conditions. Tachibana et al.4 reported a lower thermal conductivity for PMN-32PT, suggesting that thermal conductivity is highly sensitive to sample conditions and thermal stresses induced by cycling, including the growth method; a higher thermal conductivity is indicative of a more pristine sample. These materials are complex in structure that leads to a considerable spread in reported absolute values of κ.
The polarization vs. electric field along the (001) direction results at room temperature are shown in Fig. 2. The coercive field is about 3 × 105 V m−1 and the saturation field is about 6 × 105 V m−1 with a polarization of about 0.06 C m−2.
![]() | ||
Fig. 2 Polarization of PMN–33PT at room temperature. Arrows indicate the direction of the swept field. |
Fig. 3 shows the relative change with the electric field of the resonant frequency, f, that corresponds to the longitudinal compressive wave of the whole sample in the RUS. Because both f and the sound velocity of the LA phonon depend on the square root of the elastic constant c11, this relative change equals the relative change in the sound velocity of the longitudinal acoustic phonon, i.e.. These measurements could not be taken on the cycled sample, as it had been instrumented for thermal conductivity measurements and its instrumentation perturbed the natural resonant frequencies of the sample.
Fig. 4a and b illustrate the behavior of the thermal conductivity of the fresh and cycled PMN–33PT samples with the field. The first run gives a of almost 4 × 10−8 m V−1. The thermal conductivity's field dependency changes during subsequent cycles of the positive and negative fields applied to the sample as shown in Fig. S2 (ESI†), which illustrates the first four cycles. The second sample cycled 20 times near room temperature (Fig. 4b) gives
. This condition is stable and reproducible (the curve after 10 cycles is shown in the inset and differs very little from the one cycled 20 times). The electric field dependence of the thermal conductivity shows a sign flip and is an order of magnitude larger vis-à-vis the fresh sample. The value is now two orders of magnitude larger than that reported in PZT
.3 Considering the minimum and maximum thermal conductivity values, the change in the thermal conductivity of the cycled PMN–33PT sample with a field at room temperature is close to 8%. This increases to 11% at 364 K (see Fig. S3, ESI†).
![]() | ||
Fig. 4 (a) and (b) Room temperature thermal conductivity measurements on (a) a fresh sample on its first run and (b) a cycled sample on its 20th run. The (b) inset gives the data at the 10th run. This is essentially the same as at the 20th run, showing that while a transition occurs during the first 10 runs (see Fig. S2, ESI†), the sample is stable after about 10 cycles with electric field. The arrows indicate direction of sweep while the green line represents where the slope was taken to calculate the ![]() |
Fig. 5 illustrates d31 measurements on two distinct samples: one pristine, untouched by an electric field, and the other, the cycled sample previously employed for thermal conductivity measurements. To measure the transverse piezoelectric constant, we measured the change of the length of the sample along the (100) direction as a function of the electric field along the (001) direction. This change in the length of the sample corresponds to changes in the strain tensor; the derivative of these strain tensor components gives the piezoelectric coefficient.
We calculated a d31 of −1327 pC N−1 for the fresh sample and −476 pC N−1 for the cycled sample, a reduction of nearly three times.
We also investigated the longitudinal piezoelectric coefficient −d33, corresponding to the displacement along the (001) direction parallel to the applied field. We measured d33 on a fresh sample by alternating between 0 and +100 V, back to zero to ensure that we remained in the linear regime and then to −100 V. This gave us a d33 of 1550 pC N−1 for the fresh PMN–33PT. We then cycled this sample roughly 20 times electrically. Fig. 6 shows the results for the cycled sample in which we obtained d33 = 2550 pC N−1 by averaging the slope in the positive and negative fields. Note that the zero-field value was zeroed out to account for hysteretic effects.
This treatment can be extended to the stable trace reported on the cycled sample in Fig. 4b, using the d31 value of −476 pC N−1 calculated from the measurement shown in Fig. 5b and the d33 value of 2550 pC N−1. We were unfortunately unable to measure on the cycled sample, as we prioritized the thermal conductivity measurements. The sample shattered into pieces too small for sound velocity measurements during dismount. The sign inversion between the cycled (Fig. 4b) and non-cycled (Fig. 4a) PMN–33PT is also predicted by the change in piezoelectric coefficients but not the additional increase in sensitivity by an order of magnitude.
We attribute this order of magnitude discrepancy between the predicted and measured to the dependence of τ on the electric field (eqn (5)) observed in PMN-30PT.2 A detailed study of the effect of scattering on the thermal conductivity using phonon lifetimes from inelastic neutron scattering is left for future work, as it has the potential to further enhance
and reach more practical thermal switching ratios.
To be complete, we must consider competing theories. Negi et al.13 argue that a change in the thermal conductivity in PMN–PT can come from domain wall scattering. In principle, there must be a contribution of this mechanism to the electric field dependence of κ, but we argue that the principal mechanism arises through a change in the sound velocity and phonon spectrum. The first reason for this argument is the RUS measurements shown in Fig. 3, where we show unequivocally that the sound velocity depends on the electric field; sound velocity is a property at thermodynamic equilibrium and not a transport property. The second reason is that the domain sizes in Negi's work are of the order of 1 μm, much larger than the average phonon mean free path. While domain wall scattering has been shown unequivocally to cause an electric field dependence on the thermal conductivity at cryogenic temperatures,5 this effect vanishes as the temperature increases to values where the phonon thermal mean free path becomes much smaller than the domain wall size, typically above 20 K.
Footnote |
† Electronic supplementary information (ESI) available. See DOI: https://doi.org/10.1039/d4mh01845a |
This journal is © The Royal Society of Chemistry 2025 |