betaFeedback

WorksPhysics and Astronomy

Second Harmonic Flexoelectricity

Abstract

Flexoelectricity (polarization induced by strain gradients) is assumed to be insensitive to space inversion in all materials: flexoelectric polarization should not depend on whether a crystal is bent upwards or downwards. In this work, we show that this assumption is wrong for ferroelectrics of all thickness and at all curvatures, owing to an intrinsic quadratic contribution that persists down to strain gradients of 1 m-1 or less, which is millions of times smaller than conventional expectations for nonlinear flexoelectricity. As a consequence, ferroelectric crystals can generate second-harmonic voltage at twice the oscillation frequency of the bending.

Flexoelectricity (electric polarization in response to strain gradients) is a universal property of materials, having been observed in insulators [1,2], semiconductors [3-5], metals [6], and even bones [7] or ice [8,9]. The universality of flexoelectricity is guaranteed by the fourth rank nature of the flexoelectric tensor that couples strain gradient to polarization: being even-parity, it is compatible with materials of any symmetry [10]. Another consequence of even-parity is that flexoelectric polarization should be insensitive to space inversion in all materials, including those with that already have spontaneous polarization and thus a broken inversion symmetry. In other words: flipping a ferroelectric crystal upside down should have no effect on its flexoelectric coefficient.

Mathematics being an exact science, this expectation is of course mathematically correct. However, it rests on a physically incomplete picture -and is therefore ultimately wrong. As we will show, the bending-induced polarization of poled ferroelectrics cannot be adequately described by the linear flexoelectric tensor and includes a nonlinear term at all curvatures -we have observed nonlinearity at curvatures as small as 10−2 m−110^{-2}\ \mathrm{m}^{-1}, the smallest we can measure in our laboratory, in stark contrast with the millions of times higher gradients (of the order of 107 m−110^{7}\ \mathrm{m}^{-1}) previously thought to be required to reach a nonlinear regime [11][12][13]. As a consequence, the flexoelectric response of poled ferroelectrics becomes asymmetric with respect to their polar orientation. This unexpected result is shown in Figure 1 for poled ceramics of ferroelectric BaTiO3\mathrm{BaTiO}_{3} (BTO, Figure 1).

Flexoelectric polarization schematic and bending-induced polarization results

Figure 1. Flexoelectric polarization schematic (A, B) and results showing asymmetric flexoelectricity, with the magnitude and sign of the flexoelectric polarization depending on the relative orientation of the strain gradient with respect to the ferroelectric polarization. (C) Bending-induced polarization as a function of strain gradient GG for poled ceramics of ferroelectric BaTiO3_3 of different thickness. The curves are constructed from the values of the linear (μeff\mu_{\mathrm{eff}}) and quadratic (weffw_{\mathrm{eff}}) coefficients measured by harmonic analysis of the charge generated by oscillatory bending. The result are thickness-dependent instead of the thickness-independent straight line expected from standard linear flexoelectricity, and both the magnitude and sign of the average polarization as a function of strain gradient (dashed orange and red lines) depend on whether the strain gradient is positive or negative, meaning that the total effective flexoelectric coefficient may be positive, negative or even zero depending on the magnitude and sign of the strain gradient with respect to the ferroelectric polarity.

To understand this effect, let us consider the simplest case of a monodomain ferroelectric beam with polarization normal to the surface (Figure 1A). The beam, of thickness hh normal to the zz axis, is bent around the yy axis to produce a strain εxx=Gz\varepsilon_{xx}=Gz, where GG is the curvature of the beam, and −h/2<z<h/2-h/2<z<h/2 (Figure 1B). The direction of the strain gradient is parallel or antiparallel to the ferroelectric polarization depending on whether the curvature is upward or downward, respectively. Being ferroelectric, the beam exhibits piezoelectricity in addition to flexoelectricity, and both effects can be included in a generalized Landau-based free energy:

F=FP(P)+C2ε2−qP2ε−fPdεdz−σε(1)F = F_P(P) + \frac{C}{2}\varepsilon^2 - qP^2\varepsilon- fP\frac{d\varepsilon}{dz} - \sigma\varepsilon \tag*{(1)}

where P=PzP = P_z is the out-of-plane polarization, FP(P)F_P(P) is the double-well potential (in the simplest case, F(P)=αP2/2+βP4/4F(P) = \alpha P^2/2 + \beta P^4/4 with α<0\alpha< 0 and β>0\beta> 0), ε\varepsilon is the strain, CC is the elastic constant, qq is the electrostrictive coupling, ff is the flexocoupling coefficient, and σ=CGz\sigma= CGz is the externally applied stress that produces elastic in-plane strain ε=σ/C=Gz\varepsilon= \sigma/C = Gz and strain gradient ∇ε=G\nabla\varepsilon= G. A self-consistent minimization of (1) results in the following polar response to bending, taken as the average across the crystal up to the second power of the induced strain gradient GG (see Methods I):

ΔPeff=μeffG+weffG2(2)\Delta P_{\mathrm{eff}} = \mu_{\mathrm{eff}}G + w_{\mathrm{eff}}G^2 \tag*{(2)}

where μeff\mu_{\mathrm{eff}} is the conventional (linear) flexoelectric coefficient and weffw_{\mathrm{eff}} is the quadratic one. The analytically calculated flexoelectric coefficients (Methods I) are:

μeff=χfandweff=−χ2(3f22P0+q2P0h26)=−3χ2f22P0−Ah2(3)\mu_{\mathrm{eff}} = \chi f \quad\text{and} \quad w_{\mathrm{eff}} = -\chi^2\left(\frac{3f^2}{2P_0} + \frac{q^2P_0h^2}{6}\right) = -\frac{3\chi^2f^2}{2P_0} - Ah^2 \tag*{(3)}

Where χ\chi is the dielectric susceptibility, ff is the flexocoupling coefficient, and

A=q2χ2P06=C2d224P0A = \frac{q^2\chi^2P_0}{6} = \frac{C^2d^2}{24P_0}

is a new term that we label as quadratic flexo-thickness coefficient.

The polarization as a function of strain gradient is thus not a straight line but a parabola (Figure 1C).

The nonlinear flexoelectric coefficient has not one but two contributions. The first, proportional to f2f^2, is essentially a nonlinear susceptibility: it represents the higher-order response of the ferroelectric polarization to the flexoelectric field Efl∝fGE_{\mathrm{fl}} \propto fG. As such, it is a perturbative effect that requires large strain gradients and may be seen as the analytical embodiment of models based on second-order corrections [12],[13]. Its key feature is its explicit odd-parity dependence on P0P_0, which implies that the sign changes if we invert the polarization.

The second term (flexo-thickness coefficient) is related to surface piezoelectricity: as we bend the crystal, the tensile strain on the convex side reduces the out-of-plane ferroelectric polarization, thus decreasing the local piezoelectric coefficient, which is proportional to PP. The reverse happens at the concave side: local compressive strain increases polarization and so too the local piezoelectric coefficient. Thus, as we bend the crystal, the initially-homogeneous piezoelectric becomes inhomogeneous and starts acting as a bimorph actuator [14]. Since surface strain grows with thickness and the piezoelectric coefficients also change with thickness, the surface polarization induced by bending changes as thickness squared. It is this thickness-squared dependence that makes the effect dominant at macroscopic scales. It is also worth noting that the nonlinear piezoelectricity that accounts for this effect is unrelated to high-order coefficients [16]; the value of AA is purely given by first-order parameters of the ferroelectric phase: the elastic constant CC, the linear piezoelectric coefficient dd, and the spontaneous polarization P0P_0. We also note that this term is susceptible to depolarizing effects, which we have neglected to simplify the analysis. Such simplification will not alter qualitatively the results but may reduce them quantitatively. We nevertheless find that the depolarization-free expression for AA (Methods I) reproduces well the experimental results, thus validating the simplification.

One consequence of Eq. (2) is that the effective flexoelectric coefficient (derivative of the polarization with respect to the strain gradient) must become zero when G′=−μeff/(2weff)G'=-\mu_{\mathrm{eff}}/(2w_{\mathrm{eff}}). Therefore, an empirically measured flexoelectric coefficient may be positive, negative, or zero, depending on thickness, polarity, and bending magnitude, which can account for enormous discrepancies between measurements of flexoelectricity for any given ferroelectric. We can also use this prediction of a “neutral gradient” G′G' to test the accuracy of the model. First, we note that G′G' is different for very thin and very thick samples. According to Eq. (3), in the thin limit h≪3f/(qP0)h \ll3f/(qP_0), quadratic flexoelectricity is determined by the first (thickness-independent) flexocoupling term, while in the thick limit h≫3f/(qP0)h \gg3f/(qP_0) it is dominated by quadratic flexo-thickness. The threshold value is ht≡3f/(qP0)∼1h_t \equiv3f/(qP_0) \sim1 to 10 μm10\,\mu\mathrm{m}, using standard values for BTO (Methods II).

According to Eq. (3), the total bending-induced polarization of BTO thin films (h≪hth\ll h_t) will be zero when

G′=−μeff2weff=P03χf∼105∼106 m−1(4)G'=-\frac{\mu_{\mathrm{eff}}}{2w_{\mathrm{eff}}}=\frac{P_0}{3\chi f}\sim10^5\sim10^6\ \mathrm{m}^{-1} \tag*{(4)}

Such strain gradients are equivalent to curvature radii of ∼\sim1-10 microns, quite common in wrinkled oxide membranes [17],[18]. Thus, even if large flexoelectric effects may a priori be expected in ferroelectric membranes owing to their large achievable curvatures, nonlinear flexoelectricity can get in the way and frustrate such expectations.

We have verified the predicted critical curvature of films in two independent ways: a first-principles DFT simulation and actual measurements of BTO thin films. For DFT, we perform a finite-size simulation with a thickness of 6 unit-cells with closed-circuit boundary conditions to eliminate depolarization and initial polarization set as positive (upwards) (Figure 2A). The strains of Ba ions were set as εxx=εmax⁡cos⁡(2πz/h)\varepsilon_{xx}=\varepsilon_{\max}\cos\left(2\pi z/h\right), where hh is the thickness of the model. This arrangement produces strain gradient ∂εxx/∂z\partial\varepsilon_{xx}/\partial z that is maximal in the regions highlighted with a square in Figure 2B (see also Methods IV). We then allowed the remaining ions to relax and, based on their displacements in the highlighted regions, we extracted the induced local polarization as a function of strain gradient G=2πεmax⁡/hG=2\pi\varepsilon_{\max}/h (Figure 2C). The tangent slope at G=0G=0 yields the linear coefficient μeff=0.12\mu_{\mathrm{eff}}=0.12 μC/m, in good agreement with prior DFT calculations [19] (experimental values are higher due to additional piezoelectric contributions [20]). Crucially, the results confirm the parabolic dependence of P on strain gradient, with a maximum at G′=−μeff/2weff∼3×105 m−1G'=-\mu_{\mathrm{eff}}/2w_{\mathrm{eff}}\sim3\times10^{5}\ m^{-1}, in excellent agreement with our predicted range.

DFT simulations and experimental measurements of strain-gradient-induced polarization in BaTiO3

Figure 2. (A, B, C) DFT of BTO. (A) Initial atomic arrangement model, free from strain gradients. The supercell comprises 5×1×65\times1\times6-unit cells. The atoms in green, red, and silver represent barium, oxygen, and titanium, respectively. (B) Atomic arrangement emulating the strain profile depicted in the right-hand-side inset. The blue lines enclose the region within which the tensile strain is applied. The unit cells selected for the calculation of flexoelectric polarization in the zz direction are highlighted with a black trapezoidal frame where, by geometry, the local strain is zero and the strain gradient is maximal. (C) Polarization as a function of strain gradient calculated from (B). (D) Scheme of an experiment where stress is applied to the BaTiO3 film using an AFM tip to induce a strain gradient. (E) Polarization reversal measurements are conducted using PFM, and the coercive voltage corresponds to the switching barrier, which is given by the sum of the bulk coercive field and the flexoelectric field. (F) Measured coercive voltages as functions of strain gradient. The transition voltage Vc1V_{c1} continuously decreases, while ∣Vc2∣\lvert V_{c2}\rvert first increases and then decreases, following a parabola whose minimum matches our expectation value.

For the experimental verification, we prepared BaTiO3_3(111)/SrRuO3_3/SrTiO3_3/mica (BTO/SRO/STO/mica, see Methods V). Since film and substrate are bonded only by weak van der Waals forces, the film can slide over the mica and thus the film can bend quite freely [21]. The BTO film was about 8 nm-thick and of good epitaxial quality (Supplementary Material S3-S5). Stress was applied to the BTO film using an atomic force microscope (AFM) probe (Figure 2D) and we used COMSOL to correlate the tip force with the strain gradient maximum under the tip, simplified as a sphere (Figure S6). For BTO with (111) orientation out-of-plane, the linear flexoelectric coefficient is negative (μeff<0\mu_{\mathrm{eff}}<0) [22]. By analyzing the switching hysteresis of the piezoresponse as a function of tip force (Figure 2E shows the typical curve for 100 nN, other forces are in Supplementary Material S7), we track the coercive voltage as a function of strain gradient. As shown in Figure 2F, the positive coercive voltage decreases with increasing pressure, as expected – a large enough strain gradient can switch polarization even without applied voltage [23]. The negative one, however, initially increases but then decreases; the turning point marks the gradient G′G' at which linear flexoelectricity is compensated by quadratic, and the experimental value G′∼1.2×106 m−1G'\sim1.2\times10^{6}\ m^{-1} is within the expectation range given by Eq. (4).

Three-panel COMSOL simulation showing the strain, strain gradient, and relationship between AFM tip force and strain gradient in a BaTiO3 film

Figure S6. Through COMSOL, a force simulation analysis was conducted on the BTO film, and the relationship between the strain gradient and the force applied by the AFM tip was obtained.

A consequence of quadratic flexoelectricity is that an oscillatory deformation can now generate a second-harmonic signal, that is, an alternate current (ac) output at twice the oscillation frequency of the cantilever. We have verified this prediction in a set of poled ceramic BaTiO3_3 samples (h=0.5h=0.5, 0.70.7, 11 and 1.5 mm1.5\ \mathrm{mm}; Figure 3A is a sketch of the experiment). Being much thicker than hth_t, the nonlinear flexoelectric coefficient (Eq. (3)) can be simplified as weff=−P0(χqh)2/6=−Ah2w_{\mathrm{eff}}=-P_0(\chi qh)^2/6=-Ah^2, and the bulk critical curvature is G′=χf/(Ah2)=3f/(q2χP0h2)G'=\chi f/(Ah^2)=3f/(q^2\chi P_0h^2). By substituting the relevant physical parameters of BTO into the equation, we get A∼50 C/m2A\sim50\ \mathrm{C/m^2}, and G′=0.28 m−1G'=0.28\ \mathrm{m}^{-1} (results for h=1 mmh=1\ \mathrm{mm}, all other samples are in Supplementary Material S9). The results indicate that, for low strain gradients (G<G′G<G') the linear (first harmonic) signal dominates (Figure 3B), but bigger strain gradients (G∼G′G\sim G') generate a comparable electrical signal at double the vibration frequency (Figure 3C). From the Fourier transform of the oscillatory polarization, we extract the first and second order amplitudes (see Figure S10). The first-harmonic as a function of strain gradient is plotted in Figure 3D and shows the same trend for all the samples, with μeff∼35 μC/m\mu_{\mathrm{eff}} \sim35\,\mu\mathrm{C}/\mathrm{m}. Meanwhile, the second harmonic (Figure 3E, see also Figure S11 and values in Table S1) is proportional to t2t^2. Fitting the parabola with Eq. (3), we obtain A≈60 C/m2A \approx60\,\mathrm{C}/\mathrm{m}^2, in excellent agreement with the theoretical prediction.

First and second harmonic flexoelectricity in BTO

Figure 3. First and second harmonic flexoelectricity in BTO. (A) Schematic of the experiment, whereby an alternate strain gradient is applied to a pre-poled sample (Supplementary Material S7 for actual measurements). (B) first harmonic (linear) flexoelectric response at low gradients (0.14 m−10.14\ \mathrm{m}^{-1}), (C) flexoelectric response at bigger applied strain gradients (1.4 m−11.4\ \mathrm{m}^{-1}) with a significant second-harmonic component, (D) dependence of the linear flexoelectric polarization on strain gradient for samples of different thickness: the flexoelectric coefficient (slope of the linear least-squares fit to the data) is independent of thickness, as expected; (E) dependence of the quadratic flexoelectric coefficient on sample thickness, showing the predicted parabolic dependence (F) temperature dependence of the quadratic flexoelectric coefficient (heating rate =3 K/min=3\ \mathrm{K/min}), vanishing above TcT_c.

Fourier analysis plots for BTO ceramic showing displacement and polarization signals, frequency spectrum, and first- and second-order signals

Figure S10. For BTO ceramic with thicknesses of 1 mm. (A) The frequency-doubling response observed under large strain gradients is demonstrated. (B) The By performing Fourier analysis, it can be determined that the curve is composed of frequency signals at 1 Hz and 2 Hz for polarization curve in (Fig. A). When the input vibrational signal is at 1 Hz, the polarization curve can be decomposed into (C) a first-order signal at 1 Hz, and (D) a second-order signal at 2 Hz.

Plot showing the relationship between second-order $\Delta P$ and strain gradient for BaTiO$_3$ at $h=0.5$ mm, $h=0.7$ mm, $h=1$ mm, and $h=1.5$ mm

Figure S11. The relationship between the second-order polarization and the strain gradient for different thicknesses of BTO.

Thickness/mmweffw_{\mathrm{eff}} / μC\mu\mathrm{C}G′/m−1G^{\prime}/\mathrm{m}^{-1}
0.5161.1
0.7300.58
1620.28
1.51400.12

Table S1. The second order flexoelectric coefficients and G′G^{\prime} for BTO slabs of different thickness.

The flexoelectric coefficient is thus never fully linear: neither in the thin film regime, nor in thick samples. Interestingly, while one might have expected nonlinearity to be more prominent for very thin samples (where larger gradients are easier to achieve), the contribution from surface piezoelectricity is such that the opposite is true: thicker samples show a stronger second-harmonic effect at much smaller gradients. This quadratic effect only exists because of ferroelectricity: the dependence on temperature (Figure 3F) shows that, above the Curie temperature of BTO, weffw_{\mathrm{eff}} vanishes (in Supplementary Material S12 we provide the output waveforms of BTO at 122 °C and 130 °C). We have tested the generality of these results by exploring other materials: relaxor ferroelectric PMN-PT ceramic, and PMN-PT single crystal (Supplementary Materials S13 and S14). Both showed the same trends, so quadratic flexoelectricity does not arise from grain boundaries [24] nor require specific compositions.

The similarity between the magnitudes of the first and second harmonic signals can be exploited to make a device which, for the same oscillatory input, selectively generates equivalent voltage outputs at the first or the second harmonic. We achieve such harmonic distillation by topological engineering of electrode connectivity and ferroelectric poling in the scheme shown in Figure 4A, where two identical sets of parallel-plate electrodes are deposited in the same ferroelectric slab and labelled as a-b (capacitor 1) and c-d (capacitor 2).

Harmonic distillation through connection topology

Figure 4. Harmonic distillation through connection topology. (A) Diagram of the initial polarization and of the electrodes positioned on a ferroelectric slab, with two capacitors (electrodes

If the a and c, b and d electrodes are joined (Abbreviation: ac-bd), the two capacitors are in parallel (Figure 4B) and the total charge induced by bending is the sum of their flexoelectric contributions (Figure S15). Because the nonlinear term has an odd dependence on polarity, the antiparallel nonlinear components mutually cancel and the measured polarization (charge divided by the area of the capacitors) is just Pf=μGP_f=\mu G (the analysis is shown in Methods VIII). Being linear with the gradient, the output is purely first harmonic. We confirm this prediction experimentally (Figure 4C) using a BTO ceramic slab with the electrode configuration depicted in Figure 4A (the individual output of each pair of electrodes is shown in Figure S16). The observation of purely linear flexoelectricity in capacitors with antiparallel polarization explains why the relationship between flexoelectric polarization and strain gradient appears to be linear in measurements of unpoled ferroelectrics [20][22]: random (unpoled) domain orientations act de facto as capacitors of opposite polarization.

Relationship between the flexoelectric polarization and the strain gradient for the topological electrode structures

Figure S15. (A) Due to the inverse symmetry of the structure, when the initial prepolarization direction is opposite, the flexo-polarization induced by bending has opposite higher order term coefficients, while the linear term coefficients are the same. (B) When the ac-bd and ad-bc electrodes are connected in series, the linear effects and high-order of flexo-polarization on the strain gradient are realized respectively.

Real-time electrical signals and vibration displacements for $\mathrm{BaTiO}_{3}$ with topological electrode structures

Figure S16. For BaTiO3\mathrm{BaTiO}_{3} with topological electrode structures, the real-time output of electrical signals and vibration displacements when connecting a single-sided electrode.

In contrast, if the a and d, b and c electrodes are joined, the electrode configuration is topologically equivalent to bending two capacitors of identical polarity in opposite directions (Figure 4D), which cancels the linear flexoelectric components because the gradients point in opposite directions, while at the same time adds the term that depends on the square of the gradient (Figure S15B), so Pf=wG2P_f=wG^2 (further analysis is shown in Methods VIII). The output is thus a perfect second-harmonic (Figure 4E). Such biharmonic generation was replicated in PMN-PT (Supplementary Material S17).

a-b and c-d) poled in opposite directions. (B) Connecting the electrodes ac and bd results in equivalent parallel capacitors where (C) the flexoelectric output signal has the same frequency as the displacement (pure first harmonic). (D) Connecting the electrodes ad-bc is equivalent to flipping one of the capacitors around, resulting in (E) flexoelectric output at double the frequency of the displacement (pure second harmonic).

The frequency-doubling output is somewhat reminiscent of electrostriction, but in reverse: while electrostriction generates a deformation proportional to the square of voltage [26], second harmonic flexoelectricity generates a voltage perfectly proportional to the square of deformation, even under very small deformations. The results thus fill a functional void, in addition to upending (literally) the relationship between flexoelectricity and symmetry in ferroelectrics.

Acknowledgment

This work was supported by the National Natural Science Foundation of China (Nos. 52472124, 12404106, 52325205 and 12404101) and the Natural Science Foundation of Jiangxi Province (No. 20252BAC200609). ZG Wang and LL Shu acknowledge the support from Nanchang University. ICN2 is supported by the Severo Ochoa program from Spanish MCIN / AEI (Grant No. CEX2021-001214-S. GC acknowledges support from the EU through HORIZON-MSCA-SE-3D-TOPO action (Grant agreement ID: 101236483) and from Spain’s MICINN through grant PID2023-148673NB-I00 (Project GRIPHO2).

Methods

Relationship between flexoelectric polarization and strain gradients from Landau theory

To obtain the response of polarization P(z)P(z) and strain ε(z)\varepsilon(z) within the cantilever to an applied stress σ(z)\sigma(z) and/or electric field EE, we minimize the Landau energy [Eq. (1)] with respect to P(z)P(z) and ε(z)\varepsilon(z), using the following full expression for the energy:

F=FP(P)+C2ε2−qP2ε−fPdεdz+δ2(dPdz)2−σε−PEF = F_P(P) + \frac{C}{2}\varepsilon^2 - qP^2\varepsilon- fP\frac{d\varepsilon}{dz} + \frac{\delta}{2}\left(\frac{dP}{dz}\right)^2 - \sigma\varepsilon- PE

where we have added the polarization correlation term controlled by δ\delta, and the explicit coupling between polarization and electric field −PE-PE. To obtain the energy minimum, we solve the system of Euler-Lagrange equations:

∂F∂P−ddz∂F∂(dP/dz)=dFPdP−δd2Pdz2−2qPε−fdεdz−E=0(5)\frac{\partial F}{\partial P} - \frac{d}{dz}\frac{\partial F}{\partial(dP/dz)} = \frac{dF_P}{dP} - \delta\frac{d^2P}{dz^2} - 2qP\varepsilon- f\frac{d\varepsilon}{dz} - E = 0 \tag*{(5)}
∂F∂ε−ddz∂F∂(dε/dz)=Cε−qP2+fdPdz−σ=0(6)\frac{\partial F}{\partial\varepsilon} - \frac{d}{dz}\frac{\partial F}{\partial(d\varepsilon/dz)} = C\varepsilon- qP^2 + f\frac{dP}{dz} - \sigma= 0 \tag*{(6)}

Solving the equation for the strain (M1.2) results in

ε=qCP2−fCdPdz+σC(7)\varepsilon= \frac{q}{C}P^2 - \frac{f}{C}\frac{dP}{dz} + \frac{\sigma}{C} \tag*{(7)}

As discussed in the main text, in the following we consider the specific case

σ(z)=CGz,−h2<z<h2,\sigma(z) = CGz,\quad-\frac{h}{2}<z<\frac{h}{2},

which describes the stresses produced by bending of a cantilever of thickness hh. The third term in Eq. (7), describing the elastic response to the applied stress, thus naturally gives contribution to strain εelast=Gz\varepsilon_{\mathrm{elast}}=Gz and to strain gradient dεelastdz=G\frac{d\varepsilon_{\mathrm{elast}}}{dz}=G.

Substituting Eq. (7) into the Euler–Lagrange equation for polarization Eq. (5) then results in the following:

∂∂P(FP(P)−qGzP2−q22CP4)−(δ−f2C)d2Pdz2−fG−E=0,\frac{\partial}{\partial P}\left(F_P(P)-qGzP^2-\frac{q^2}{2C}P^4\right)-\left(\delta-\frac{f^2}{C}\right)\frac{d^2P}{dz^2}-fG-E=0,

from where it is apparent that the strain gradient leads to a zz-dependent renormalization of the P2P^2 coefficient in the double-well potential expression through the electrostrictive coupling, producing an asymmetric polar response to the strain gradient at the top and at the bottom surfaces of the cantilever. For the following use, we rewrite the equation for polarization as follows:

∂∂P(FP∗(P)−qGzP2)−fG−E=0,(8)\frac{\partial}{\partial P}\left(F_P^*(P)-qGzP^2\right)-fG-E=0, \tag*{(8)}
FP∗(P)=FP(P)−q22CP4,F_P^*(P)=F_P(P)-\frac{q^2}{2C}P^4,

and we neglect the effect of the term (δ−f2C)d2Pdz2\left(\delta-\frac{f^2}{C}\right)\frac{d^2P}{dz^2} under the assumption that the thickness of the cantilever hh far exceeds the characteristic scale ξ∼((δ−f2C)χ)12\xi\sim\left(\left(\delta-\frac{f^2}{C}\right)\chi\right)^{\frac{1}{2}} of variations of polarization (typically ξ∼1\xi\sim1 nm in ferroelectrics).

In the absence of strain gradient and electric field, the spontaneous polarization is described by setting G=0G=0, E=0E=0 in Eq. (8):

dFP∗dP∣P=P0=0.\left.\frac{dF_P^*}{dP}\right|_{P=P_0}=0.

For instance, the classic double-well potential F∗(P)=α2P2+β4P4F^*(P)=\frac{\alpha}{2}P^2+\frac{\beta}{4}P^4 leads to the equation for spontaneous polarization αP0+βP03=0\alpha P_0+\beta P_0^3=0 and P0=−αβP_0=\sqrt{-\frac{\alpha}{\beta}}.

In the following, we will write the linear and quadratic responses to the strain gradients in terms of the linear and quadratic permittivity, as well as the linear piezoelectric coupling constants. Let us first list the explicit expressions for the three latter parameters. At first, we describe the effect of applying a uniform electric field EE on polarization, P(E)=P0+χE+κE2P(E)=P_0+\chi E+\kappa E^2, in a stress-free sample (G=0G=0). Equation Eq. (8) then rewrites as:

dFP∗dP−E=0.(9)\frac{dF_P^*}{dP}-E=0. \tag*{(9)}

The linear and quadratic responses to the applied electric field χ=dPdE∣E=0\chi=\left.\frac{dP}{dE}\right|_{E=0} and κ=12d2PdE2∣E=0\kappa=\left.\frac{1}{2}\frac{d^2P}{dE^2}\right|_{E=0} can be obtained by differentiating Eq. (9) with respect to EE:

ddE(dFP∗dP−E)=d2FP∗dP2dPdE−1=0,\frac{d}{dE}\left(\frac{dF_P^*}{dP}-E\right)=\frac{d^2F_P^*}{dP^2}\frac{dP}{dE}-1=0,
χ=dPdE∣E=0=(d2FP∗dP2)−1∣E=0,\chi=\left.\frac{dP}{dE}\right|_{E=0}=\left.\left(\frac{d^2F_P^*}{dP^2}\right)^{-1}\right|_{E=0},
d2dE2(dFP∗dP−E)=d3FP∗dP3(dPdE)2+d2FP∗dP2d2PdE2=0,\frac{d^2}{dE^2}\left(\frac{dF_P^*}{dP}-E\right)=\frac{d^3F_P^*}{dP^3}\left(\frac{dP}{dE}\right)^2+\frac{d^2F_P^*}{dP^2}\frac{d^2P}{dE^2}=0,
κ=12d2PdE2∣E=0=−12d3FP∗dP3(dPdE)(d2FP∗dP2)−1∣E=0=−χ32d3FP∗dP3∣E=0.\kappa=\left.\frac{1}{2}\frac{d^2P}{dE^2}\right|_{E=0}=-\left.\frac{1}{2}\frac{d^3F_P^*}{dP^3}\left(\frac{dP}{dE}\right)\left(\frac{d^2F_P^*}{dP^2}\right)^{-1}\right|_{E=0}=-\left.\frac{\chi^3}{2}\frac{d^3F_P^*}{dP^3}\right|_{E=0}.

For example, using the classic double-well potential F∗(P)=α2P2+β4P4F^*(P)=\frac{\alpha}{2}P^2+\frac{\beta}{4}P^4 leads to χ=−12α>0\chi=-\frac{1}{2\alpha}>0 and κ=−3χ22P0<0\kappa=-\frac{3\chi^2}{2P_0}<0 at P=+P0P=+P_0.

Next, we consider linear and quadratic piezoelectricity, i.e. the uniform responses of polarization to an applied uniform stress σ\sigma in the absence of electric field. Let us write polarization as P(σ)=P0+dσ+bσ2P(\sigma)=P_0+d\sigma+b\sigma^2, where dd and bb are linear and quadratic direct piezoelectric coefficients (while we do not use bb in the main text, it is instrumental to obtain the quadratic piezoelectricity explicitly). The total strain produced by an applied stress is given by Eq. (7), ε=qCP2+σc\varepsilon=\frac{q}{C}P^2+\frac{\sigma}{c}. Then linear and quadratic piezoelectric coefficients d=dPdσ∣σ=0d=\left.\frac{dP}{d\sigma}\right|_{\sigma=0} and b=12d2Pdσ2∣σ=0b=\left.\frac{1}{2}\frac{d^2P}{d\sigma^2}\right|_{\sigma=0} can be found by substituting this expression for strain into Eq. (5),

dFPdP−2q2P3C−2qPσC=dFP∗dP−2qPσC=0,\frac{dF_P}{dP}-\frac{2q^2P^3}{C}-\frac{2qP\sigma}{C}=\frac{dF_P^*}{dP}-\frac{2qP\sigma}{C}=0,

and differentiating it with respect to σ\sigma:

[ddσ(dFP∗dP−2qPσC)]σ=0=[(d2FP∗dP2−2qσC)dPdσ−2qPC]σ=0=dχ−2qP0C=0,\left[\frac{d}{d\sigma}\left(\frac{dF_P^*}{dP}-\frac{2qP\sigma}{C}\right)\right]_{\sigma=0}=\left[\left(\frac{d^2F_P^*}{dP^2}-\frac{2q\sigma}{C}\right)\frac{dP}{d\sigma}-\frac{2qP}{C}\right]_{\sigma=0}=\frac{d}{\chi}-\frac{2qP_0}{C}=0,
d=2qχP0C,d=\frac{2q\chi P_0}{C},
[d2dσ2(dFP∗dP−2qPσC)]σ=0=[d3FP∗dP3(dPdσ)2−4qCdPdσ+(d2FP∗dP2−2qσC)d2Pdσ2]σ=0=−2κχ3d2−2d2χP0+2bχ=0,\begin{aligned} \left[\frac{d^2}{d\sigma^2}\left(\frac{dF_P^*}{dP}-\frac{2qP\sigma}{C}\right)\right]_{\sigma=0}=\left[\frac{d^3F_P^*}{dP^3}\left(\frac{dP}{d\sigma}\right)^2-\frac{4q}{C}\frac{dP}{d\sigma}+\left(\frac{d^2F_P^*}{dP^2}-\frac{2q\sigma}{C}\right)\frac{d^2P}{d\sigma^2}\right]_{\sigma=0} \\ &=-\frac{2\kappa}{\chi^3}d^2-\frac{2d^2}{\chi P_0}+\frac{2b}{\chi}=0, \end{aligned}
b=d2P0(1+κP0χ2),b = \frac{d^2}{P_0}\left(1+\frac{\kappa P_0}{\chi^2}\right),
P(σ)=P0+dσ+d2P0(1+κP0χ2)σ2,P(\sigma)=P_0+d\sigma+\frac{d^2}{P_0}\left(1+\frac{\kappa P_0}{\chi^2}\right)\sigma^2,

where we have expressed electrostriction qq in terms of the piezoelectric constant dd, q=Cd2χP0q=\frac{Cd}{2\chi P_0}, and used d2FP∗dP2=1χ\frac{d^2F_P^*}{dP^2}=\frac{1}{\chi}, d3FP∗dP3=−2κχ3\frac{d^3F_P^*}{dP^3}=-\frac{2\kappa}{\chi^3}.

Finally, we describe the response of polarization to the strain gradients in the absence of electric fields. Writing polarization as P(G)=P0+μG+wG2P(G)=P_0+\mu G+wG^2, the linear and quadratic responses to the applied strain gradient μ=dPdG∣G=0\mu=\left.\frac{dP}{dG}\right|_{G=0} and w=12d2PdG2∣G=0w=\left.\frac{1}{2}\frac{d^2P}{dG^2}\right|_{G=0} can be obtained by differentiating Eq. (8) with respect to GG:

[ddG(dFP∗dP−2qGzP−fG)]G=0=[(d2FP∗dP2−2qGz)dPdG−2qGzP−f]G=0=μχ−Cdzχ−f=0,\left[\frac{d}{dG}\left(\frac{dF_P^*}{dP}-2qG_zP-fG\right)\right]_{G=0} = \left[\left(\frac{d^2F_P^*}{dP^2}-2qG_z\right)\frac{dP}{dG}-2qG_zP-f\right]_{G=0} = \frac{\mu}{\chi}-\frac{Cd z}{\chi}-f=0,
μ=χf+Cdz,\mu=\chi f+Cd z,
[d2dG2(dFP∗dP−2qGzP−fG)]G=0=[d3FP∗dP3(dPdG)2−4qzdPdG+(d2FP∗dP2−2qGz)d2PdG2]G=0=−2κχ3μ2−2CdzχP0μ+2wχ=0,\begin{aligned} \left[\frac{d^2}{dG^2}\left(\frac{dF_P^*}{dP}-2qG_zP-fG\right)\right]_{G=0} = \left[\frac{d^3F_P^*}{dP^3}\left(\frac{dP}{dG}\right)^2-4qz\frac{dP}{dG}+\left(\frac{d^2F_P^*}{dP^2}-2qG_z\right)\frac{d^2P}{dG^2}\right]_{G=0} \\ &=-\frac{2\kappa}{\chi^3}\mu^2-\frac{2Cd z}{\chi P_0}\mu+\frac{2w}{\chi}=0, \end{aligned}
w=CdzP0μ+κχ2μ2=CdzP0(χf+Cdz)+κχ2(χf+Cdz)2,w=\frac{Cd z}{P_0}\mu+\frac{\kappa}{\chi^2}\mu^2 =\frac{Cd z}{P_0}(\chi f+Cd z)+\frac{\kappa}{\chi^2}(\chi f+Cd z)^2,
P(G)=P0+(χf+Cdz)G+[CdzP0(χf+Cdz)+κχ2(χf+Cdz)2]G2.P(G)=P_0+(\chi f+Cd z)G+\left[\frac{Cd z}{P_0}(\chi f+Cd z)+\frac{\kappa}{\chi^2}(\chi f+Cd z)^2\right]G^2.

We estimate the experimentally extracted polarization by the average value within the cantilever:

ΔPeff=1h∫−h2h2P(z) dz−P0=μeffG+weffG2,\Delta P_{\mathrm{eff}}=\frac{1}{h}\int_{-\frac{h}{2}}^{\frac{h}{2}}P(z)\,dz-P_0=\mu_{\mathrm{eff}}G+w_{\mathrm{eff}}G^2,
μeff=χf,weff=κf2+C2d2h212P0(1+κP0χ2),\mu_{\mathrm{eff}}=\chi f,\qquad w_{\mathrm{eff}}=\kappa f^2+\frac{C^2d^2h^2}{12P_0}\left(1+\frac{\kappa P_0}{\chi^2}\right),

obtaining Eq. (2) of the main text.

Finally, taking a simple double-well potential FP∗(P)=αP22+βP44F_P^*(P)=\frac{\alpha P^2}{2}+\frac{\beta P^4}{4} with κ=−3χ22P0\kappa=-\frac{3\chi^2}{2P_0}, we obtain Eq. (3) of the main text.

Sample preparation and characterization

BaTiO3_3 and Sm-PMN-30PT ceramics

BaTiO3_3 ceramics were prepared by solid-phase sintering. BaCO3_3 (99.99%, Aladdin), TiO2_2 (99.99%, Aladdin) powders as the raw materials by molar ratio matching. BTO powders are formed by pre-firing at 1000 °C for 3h. These powders were granulated with polyvinyl alcohol solution and pressed into shape at a pressure of 20 MPa. Then, the BTO ceramics were sintered at 1250 °C for 6h in Muffle furnace. The samples were cut into pieces with thicknesses of 0.5, 0.7, 1, and 1.5 mm using a diamond cutting machine after the samples were sintered. They were then annealed through a tube furnace at 800 °C for 20 h in an air atmosphere to eliminate oxygen vacancies caused by sintering. The measured coefficients for BaTiO₃ ceramics (Supplementary Material S2) were P0=10P_0=10 μC/cm², Q=q/C=0.0425Q=q/C=0.0425 m⁴ C⁻², and C=150C=150 GPa. The flexocoupling coefficient is f∼1f\sim1 - 10 V [20]).

Sm-PMN-30PT ceramics were prepared using a simple mixed oxide method, while a precursor method was used to fabricate the MgNb₂O₆ precursor powders [27]. Specifically, samples were calcined at 850 °C for 2h and sintered at 1230 °C for 2h. Thermal annealing was conducted at 900 °C for 10h to eliminate residual PbO on the surface of the ceramics. The sample was cut into long strips with a thickness of 0.7 mm.

PMN-PT (100) single crystals were commercially acquired (TOPVENDOR, Beijing, China) and double-side polished. All the samples were applied an electric field to make them generate an initial upward polarization, for the measurement of flexoelectric performance.

BaTiO₃/SrRuO₃/SrTiO₃/mica films

High-quality BaTiO₃ (BTO) thin films were grown on mica (001) substrates by pulsed laser deposition. SrTiO₃ (STO) was first deposited on the substrate surface as a buffer layer, followed by the deposition of SrRuO₃ (SRO) as the bottom electrode. During the deposition of the STO buffer layer and SRO electrode layer, the substrate temperatures were set to 700°C and 680°C, respectively. The oxygen partial pressure was maintained at 100 mTorr, while the laser fluence and repetition rate were 1.38 mJ/cm² and 10 Hz, respectively. During the deposition of the BTO thin films, the substrate temperature was maintained at 750 °C, the oxygen partial pressure was kept at 10 mTorr, and the laser fluence and repetition rate were 1.38 mJ/cm² and 8 Hz, respectively.

Characterization for macroscopic materials

The functional parameters of materials (include the coefficients of Eq. (2)) were measured as follows:

XRD: The phase composition and crystallographic orientation of our cells was characterized by XRD using a D/Max-IIIC X-ray diffractometer (Rigaku, Tokyo, Japan) with Cu Kα\alpha radiation to understand the crystallization of the perovskite.

SEM: The microstructures of the natural surfaces and acid-etched surfaces were observed by a scanning electron microscopy (SEM, Model JSM 6710F, Jeol, Japan).

Dielectric: The dielectric properties of the ceramics were measured by a broadband dielectric spectrometer ((DMS3000, Partulab, Germany)) and an impedance analyzer (E4294A, Agilent, Palo Alto, USA).

Ferroelectricity: The ferroelectric hysteresis loops (P-E loops) were analyzed using a ferroelectric integrated test system (TF Analyzer 3000, aixACCT, Germany).

Electrostriction: The electrostrictive performances were analyzed using the ferroelectric integrated test system as well, the test temperature is 200 °C to ensure that the sample is in the paramagnetic phase.

Elastic constant: Force curve measurements on BaTiO3_3 ceramics were performed using a Cypher ES atomic force microscope (Asylum Research, USA) operating in contact mode under ambient conditions. A CDT-NCLR conductive diamond probe (Nanosensors, Switzerland) was employed for all measurements, nominal spring constant: 72 N/m, resonant frequency: ∼\sim210 kHz, tip radius: ∼\sim100 nm) were employed to ensure electrical conductivity and mechanical robustness. To ensure representative data, force curves were acquired at 10 randomly selected regions across the polished surface of BaTiO3_3 ceramic samples, with 5 individual curves recorded per region. All force curve data were processed using Asylum Research's Igor Pro-based software, and extraction of key mechanical parameters such as Young's modulus.

DFT for BaTiO3_3

To further deepen our understanding of the experimental findings, we performed first-principles calculations to investigate the relationship between polarization and strain gradient in bulk BaTiO3_3 (BTO). The whole calculations were conducted using the Perdew-Burke-Ernzerhof (PBE) with Projector-Augmented Wave (PAW) potentials, as implemented in the Vienna ab initio simulation package (VASP) [28]. The MonkhorstPack scheme employed for k-point grid sampling in the irreducible Brillouin zone was set to 4×2×14 \times2 \times1. A plane-wave cutoff energy of 500 eV was employed. Atomic positions were relaxed using the conjugate gradient algorithm until the residual forces were smaller than 0.005 eV/Å. We adopted the computational model proposed by Xu's approach [29], allowing for eliminating polarization contributions arising from non-equivalent surface terminations.

Figure 2 shows the established configuration with 5×1×65 \times1 \times6 unit cells, with its energy fully relaxed by conjugate gradient algorithm. Periodic boundary condition is applied in all direction. There is only one unit cell in the yy direction in the supercell. Strain gradient is created by displacing every atom along xx axis, according to a cosine displacement function ε11=εmax⁡cos⁡(2πz/h)\varepsilon_{11} = \varepsilon_{\max} \cos(2\pi z / h), εmax⁡\varepsilon_{\max} being the maximum strain and hh standing for the thickness of the model (see Figure 2A), after that, the energy of the whole structure is fully minimized, while maintaining the Ba atom in a fixed position. Since the experimentally measured polarization includes both piezoelectric and flexoelectric contributions, we selected the unit cell where both strain and strain gradient are nonzero (highlighted by the trapezoidal frame in Figure 1A-C) to evaluate the relationship between polarization and strain gradient. With the derivative of ε11\varepsilon_{11} with respect to zz, the magnitude of the strain gradient ∂ε11∂x3\frac{\partial\varepsilon_{11}}{\partial x_3} is computed to be ∼107 m−1\sim10^7\ \mathrm{m}^{-1}, attainable in experimental measurements. Born effective charges are evaluated by using a density-functional perturbation theory (DFPT) [30]. With them, the polarization can be defined as: P=eΩc∑jwjZj∗ujP = \frac{e}{\Omega_c} \sum_j w_j Z_j^* u_j, where Ωc\Omega_c is the volume of the unit cell considered, uju_j denotes the atomic coordinates in the zz direction in the unit cell. Index jj spans all atoms contained in the unit cell. Zj∗Z_j^* is the Born effective charges of the bulk BTO. ww means weights that are set to 1/8 for Ba, 1 for Ti, and 1/2 for O.

Characterization for BaTiO3/SrRuO3/SrTiO3/mica\mathrm{BaTiO_3/SrRuO_3/SrTiO_3/mica}

Structural characterization: X-ray diffraction measurements, including 2θ2\theta–ω\omega scans, ϕ\phi scans, reciprocal space mapping (RSM), and X-ray reflectivity (XRR), were performed using a high-resolution X-ray diffractometer (SmartLab, Rigaku, JPN) equipped with a Cu Kα\alpha radiation source (λ=1.54\lambda= 1.54 Å). The 2θ2\theta–ω\omega scans were used to determine the out-of-plane lattice parameters of the films and membranes, while ϕ\phi scans were conducted to examine the in-plane epitaxial relationship between the film and substrate. RSM measurements around selected asymmetric reflections were carried out to further evaluate the strain state, lattice relaxation, and crystallographic coherence of the films. In addition, XRR measurements were performed to assess the film thickness, density, and interfacial roughness.

Ferroelectricity: Piezoresponse force microscopy (PFM) measurements were performed using an atomic force microscope (Asylum Research Cypher ES, Oxford Instruments, UK) to investigate the local ferroelectric properties of the BTO thin films. Conductive ASYELEC-01-R2 probes with a Ti/Ir coating were used for the measurements.

Strain gradients: switching spectroscopy piezoelectric force microscopy (SS-PFM) measurements were performed on the BTO thin film. A fixed position on the film surface was selected, and different normal loads ranging from 100 nN to 1000 nN were applied to the AFM tip. ASYELEC-01-R2 (2.8 N m−1^{-1} in force constant) were used in the measurements, the corresponding pressure can be estimated with a tip radius of 25 nm (ASYELEC-01-R2) under the assumption that the tip-sample contact is represented by a disc with radius equivalent to the tip radius.

PFM measurements for BaTiO3_3 film: under the 6 V DC bias condition, PFM phase hysteresis loops were recorded at each applied load to investigate the influence of mechanical force on the ferroelectric switching behavior.

The flexoelectric polarization vs strain gradient for BTO film

For BTO(111)(111) film, under the action of the strain gradient, the total polarization can be expressed as,

{P1=P0+μeffG+weffG2,w<0P2=−P0+μeffG−weffG2,w>0(10)\begin{cases} P_1 = P_0 + \mu_{eff}G + w_{eff}G^2, \quad w < 0 \\ P_2 = -P_0 + \mu_{eff}G - w_{eff}G^2, \quad w > 0 \tag*{(10)} \end{cases}

P1P_1 and P2P_2 respectively represent the total polarization when the ferroelectric polarization is upward and downward, which is the result of the coupling of spontaneous and flexoelectric polarization. At this point, the quadratic coefficient originates from non-thickness term, weff=−3χ2f22P0w_{eff}=-\frac{3\chi^{2}f^{2}}{2P_{0}}. As the applied voltage increases, when the polarization direction of the electric domain changes, it indicates that the external electric field will be equal to the electric field formed by the P1P_{1} or and P2P_{2}.

(1) When the polarization is positive, the second-order tortuosity coefficient weff<0w_{eff}<0 and

G′=−μeff2weff<0.G'=-\frac{\mu_{eff}}{2w_{eff}}<0.

Therefore, when the tip is pressed to apply a negative strain gradient, the hysteresis electric field Vd1V_{d1} continues to decrease, the corresponding absolute value of total polarization intensity P1P_{1} also decreases continuously accordingly.

(2) When the polarization is negative, the second-order tortuosity coefficient weff>0w_{eff}>0 and

G′=−μeff2weff>0.G'=-\frac{\mu_{eff}}{2w_{eff}}>0.

Therefore, when the tip is pressed to apply a negative strain gradient, the hysteresis electric field Vd2V_{d2} will first increase. Then decrease after the strain gradient exceeds G′G'. The corresponding absolute value of the total polarization intensity P2P_{2} maintains the same trend. The extreme point G′G' was obtained through fitting to be approximately 1.2×106 m−11.2 \times10^{6}\ \mathrm{m}^{-1}, which is comparable to the theoretical value.

Strain gradients and flexoelectricity for cantilever system

The strain gradient in a cantilever system can be written as:

∂ε11∂x3=3w(L)L2(1−xL)(11)\frac{\partial\varepsilon_{11}}{\partial x_{3}}=\frac{3w(L)}{L^{2}}\left(1-\frac{x}{L}\right) \tag*{(11)}

where w(L)w(L) is the vertical deflection defection delivered by the piezoelectric actuator at the end of the cantilever, which was collected by a laser displacement sensor. LL is the length of the cantilever and xx is the horizontal position along the cantilever.

When the change in curvature (strain gradient) with time during mechanical vibration can be expressed as G=G0sin⁡2πfvtG=G_{0}\sin2\pi f_{v}t, the flexoelectric polarization in the materials can be expressed as

Pf=weff(G0sin⁡2πfvt)2+μeffG0sin⁡2πfvt(12)P_{f}=w_{eff}(G_{0}\sin2\pi f_{v}t)^{2}+\mu_{eff}G_{0}\sin2\pi f_{v}t \tag*{(12)}

The flexoelectric signal was measured through a dynamic thermomechanical analysis system (DMA850, TA Instruments, USA). We used DMA to deliver an oscillatory bending, The displacement signal is collected by a laser displacement sensor (KEYENCE, LK-G30, Japan). According to the natural plane assumption of the beam, the bending makes the material be stretched above the neutral plane, and compressed under the neutral plane. That results in a strain gradient uniformly distributed along the thickness of the cantilever beam. The bend-introduced charges generated of the sample through the charge amplifier (SA-PE01, Shiao Technology, Wuxi, China) and were collected by the oscilloscope (MDO3024, Tektronix, USA). Therefore, the polarization can be obtained by the total electric charge collected on the electrode:

P3=Q′S(13)P_3 = \frac{Q'}{S} \tag*{(13)}

where Q′Q' is the charge, SS is the electrode area.

Polarization form through electrode topology design

For the capacitor poled parallel to the strain gradient (capacitor 1) the relationship between the bending-induced polarization and strain gradient is (Figure S15)

Pf(1)=wG2+μG(14)P_f(1) = wG^2 + \mu G \tag*{(14)}

Meanwhile, for the capacitor poled antiparallel to the strain gradient (capacitor 2), the relationship between the flexo-polarization and the strain gradient is (Figure S15)

Pf(2)=−wG2+μG(15)P_f(2) = -wG^2 + \mu G \tag*{(15)}

If the a and c, b and d electrodes are joined (Abbreviation: ac-bd), the total flexoelectric polarization induced by bending is the sum (Figure S15)

Pf=Pf(1)+Pf(2)2=μG(16)P_f = \frac{P_f(1)+P_f(2)}{2} = \mu G \tag*{(16)}

Instead, if the a and d, b and c electrodes are joined, the total flexoelectric polarization induced by bending is the sum (Figure S15):

Pf=Pf(1)−Pf(2)2=wG2(17)P_f = \frac{P_f(1)-P_f(2)}{2} = wG^2 \tag*{(17)}

SEM and XRD for BaTiO3\mathrm{BaTiO}_{3}

The SEM images were recorded on the polished and thermally etched surfaces of BaTiO3\mathrm{BaTiO}_{3} ceramic (Fig. S1-Figure 1). The sample exhibited dense microstructures with minimal visible porosity. As quantified using Nano Measurer software, the average grain size of BaTiO3\mathrm{BaTiO}_{3} was approximately 12 μm\mu\mathrm{m}.

The sample is sintered densely (Method II). The Rietveld refinements were performed to confirm the phase purity and examine the detailed evolution of the lattice parameters. An initial model for the tetragonal ABO3ABO_{3}-type perovskite with a P4mmP4mm space group was adopted. The measured XRD data exhibit excellent agreement with the calculated results, thereby confirming the reliability of the structural simulation.

SEM image and grain-size distribution for BaTiO3

Figure S1-1. (A) SEM for BaTiO3\mathrm{BaTiO}_{3}. (B) Grain size for BaTiO3\mathrm{BaTiO}_{3}.

XRD pattern for BaTiO3

Figure S1-2. XRD for BaTiO3\mathrm{BaTiO}_{3}.

The parameters of BaTiO3_3

Material parameters of BaTiO$_3$

Figure S2. Material parameters of BaTiO3_3 used to calculate the second-order flexoelectric coefficient: (A) temperature-dependent dielectric constant, (B) polarization-electric field hysteresis loop yielding P0P_0, (C) elastic constant CC, and (D) electrostriction coefficient Q33Q_{33}. The longitudinal electrostrictive coefficient QzzQ_{zz} is related to the transverse one by Q31/Q33=νQ_{31}/Q_{33}=\nu. Since the electrostrictive coefficient is directly proportional to the piezoelectric coefficient (see Methods I), so ν=d31/d33\nu=d_{31}/d_{33}, where ν\nu is effective piezoelectric Poisson's ratio. Such as, for poled BTO ceramic, ν∼0.6−0.8\nu\sim0.6-0.8. For BTO single crystal, ν∼0.3−0.4\nu\sim0.3-0.4 [S1].

The thickness of BaTiO3\mathrm{BaTiO}_{3} film

Since the BTO film deposited with 300 pulses is too thin to allow accurate thickness fitting directly from X-ray reflectivity (XRR), a linear relationship between the pulse number and film thickness was established using thicker samples. The linear fitting results show that 1000 pulses correspond to a film thickness of approximately 29 nm. Based on this relationship, the thickness of the BTO film deposited with 300 pulses is estimated to be approximately 8.5 nm.

Thickness calibration of BTO films prepared with different numbers of laser pulses

Figure S3. Thickness calibration of BTO films prepared with different numbers of laser pulses. (A) XRR profiles and fitting curves of BTO films prepared with different pulse numbers; (B) Dependence of the BTO film thickness on the number of laser pulses.

Structural characterization for BTO/SRO/STO/mica

Figure S4A and S4B show a typical out-of-plane 2θ/ω2\theta/\omega scan of the BTO/SRO/STO/mica heterostructure. On the mica(001) substrate, only the BTO and SRO (111) and (222) diffraction peaks are observed, indicating that the BTO and SRO films exhibit a highly preferred out-of-plane orientation, with no obvious secondary-phase peaks detected. The BTO(111) peak of the film grown on the mica substrate is close to the bulk BTO value, suggesting that the film is in a relaxed strain state. This conclusion is further supported by the reciprocal space mapping (RSM) results. Figure S4C shows the reciprocal space map of the heterostructure around the asymmetric BTO(121) and SRO(121) reflections. The BTO(121) and SRO(121) diffraction peaks are nearly aligned along the in-plane QxQ_x direction, indicating a good in-plane coherent relationship between the BTO film and the SRO bottom electrode layer.

Structural characterization of the BTO/SRO/STO/mica heterostructure

Figure S4. Structural characterization of the BTO/SRO/STO/mica heterostructure. (A) Typical out-of-plane 2θ/ω2\theta/\omega scan; (B) Enlarged view of the BTO(111) and SRO(111) diffraction peaks; (C) Reciprocal space map around the asymmetric BTO(121) and SRO(121) reflections.

Ferroelectricity for BaTiO₃ film with mica substrate

The ferroelectric switching behavior of BTO thin films grown on STO and mica substrates was investigated by Piezoelectric Force Microscopy (PFM). Figure S5 in supplementary show the PFM phase and amplitude images after domain writing. A box-in-box domain pattern was written by applying different DC biases to the conductive AFM tip. First, a 3×3 μm23 \times3\ \mu\mathrm{m}^{2} region was poled by scanning with a +6 V tip bias. Subsequently, the central 2×2 μm22 \times2\ \mu\mathrm{m}^{2} region was switched by applying -6 V, followed by writing the central 1×1 μm21 \times1\ \mu\mathrm{m}^{2} region again with +6 V. The PFM phase image exhibits a clear 180∘180^\circ contrast between oppositely poled regions, while the corresponding amplitude image shows distinct contrast at the domain boundaries, confirming the reversible ferroelectric polarization switching of the BTO thin film.

AFM topography, PFM phase, and PFM amplitude images of the BTO/SRO/STO/mica heterostructure

Figure S5. (A) AFM topography images of the BTO/SRO/STO/mica heterostructures; PFM out-of-plane amplitude and phase images electrical poling. (B) Phase, (C) Amplitude images of BTO/SRO/STO/mica heterostructure.

The relationship for AFM tip force and strain gradient of BaTiO3_3 film

PFM measurements for BaTiO3\mathrm{BaTiO}_{3} film

Subsequently, switching spectroscopy piezoelectric force microscopy (SS-PFM) measurements were performed on the BTO thin film. A fixed position on the film surface was selected, and different normal loads ranging from 100 nN to 1000 nN were applied to the AFM tip. Under the same dc bias condition, PFM phase hysteresis loops were recorded at each applied load to investigate the influence of mechanical force on the ferroelectric switching behavior (Figure S7).

PFM phase hysteresis loops for BaTiO3 film under different applied loads

Figure S7. PFM phase hysteresis loops were recorded at applied load from 100 nN to 1000 nN.

Flexoelectric coefficient measurement device

Flexoelectric coefficient measurement device

Figure S8. (A) The end of the sample is clamped, and the front end is driven by an actuator to undergo reciprocal cyclic motion. (B) When the sample bends downwards, the top surface is subjected to tension while the bottom surface is compressed, forming a positive strain gradient. Conversely, when the sample bends upwards, a negative strain gradient is formed.

The flexoelectric coefficients were measured through a dynamic thermomechanical analysis system (DMA850, TA Instruments, USA). We used DMA to deliver an oscillatory bending. According to the natural plane assumption of the beam, the bending makes the material be stretched above the neutral plane, and compressed under the neutral plane. That results in a strain gradient uniformly distributed along the thickness of the cantilever beam.

As the strain gradient increases, the waveform changes in real time for BaTiO3_3

Waveforms for BTO ceramic with a thickness of 0.5 mm

Figure S9-1. For BTO ceramic with thicknesses of 0.5 mm. The strain gradient gradually increased from 0.28 m−1^{-1} to 2.5 m−1^{-1}, the second-order information of the output signal gradually intensified.

Electrical signals for BTO ceramic with a thickness of 0.7 mm

Figure S9-2. For BTO ceramic with thicknesses of 0.7 mm. (A) in a very small strain gradient range (G1<G′G_1 < G'), the output electrical signal is at the same frequency as the vibration. (B) The frequency-doubling response observed under large strain gradients is demonstrated.

Electrical signals for BTO ceramic with a thickness of 1 mm

Figure S9-3. For BTO ceramic with thicknesses of 1 mm. (A) in a very small strain gradient range (G1<G′G_1 < G'), the output electrical signal is at the same frequency as the vibration. (B) The frequency-doubling response observed under large strain gradients is demonstrated.

Graphs showing displacement and electrical signal responses for BTO ceramic

Figure S9-4. For BTO ceramic with thicknesses of 1.5 mm. (A) in a very small strain gradient range (G1<G′G_1 < G'), the output electrical signal is at the same frequency as the vibration. (B) The frequency-doubling response observed under large strain gradients is demonstrated.

Fourier analysis for polarization intensity of BTO

S11. The relationship between the second-order polarization and the strain gradient for different thicknesses of BTO.

Temperature spectrum of output waveforms for BaTiO3\mathrm{BaTiO}_{3}

Output waveforms of BaTiO3 at 122 °C and 130 °C, shown in panels A and B

Figure S12. The output waveforms of BTO at (a) 122 °C and (b) 130 °C.

Nonlinear flexoelectricity in xxSm-PMN-PT

In order to extend the generality of the effect, another ferroelectric material, PMN-PT was selected, its piezoelectric coefficient was further increased by doping xx mol% Sm2+^{2+} ions (xxSm-PMN-PT, x=0,1,2x = 0, 1, 2) [S2]. Figs. S13-Figure 1 to S13-Figure 3 presents the dielectric, ferroelectric properties and electrostrictive coefficients of xxSm-PMN-PT, respectively. The results (Fig. S13-Figure 4) show that, when the strain gradient has not passed the G′G^{\prime}, the polarization introduced by bending and the displacement are of the same frequency. Beyond G′G^{\prime}, the frequency of the bending-induced polarization is twice that of displacement. Similarly, with the increase of the initial piezoelectric coefficient, the strain gradient corresponding to the extreme value of the polarization curve becomes smaller. This result is the same as that of BTO-based materials. More importantly, for the 2Sm-PMN-PT material, the second-order coefficient reaches up to 10 times that of the first-order coefficient.

Temperature dependence of the dielectric constant and dielectric loss in PMN-PT, 1Sm-PMN-PT and 2Sm-PMN-PT ceramics

Figure S13-1. Temperature dependence of the dielectric constant and dielectric loss in the (A) PMN-PT, (B) 1Sm-PMN-PT and (C) 2Sm-PMN-PT ceramics.

The P-E loop for xSm-PMN-PT

Figure S13-2. The P-E loop for xxSm-PMN-PT (x=0,1,2x = 0, 1, 2).

Electrostrictive performance curves for PMN-PT, 1Sm-PMN-PT, and 2Sm-PMN-PT

Figure S13-3. Electrostrictive performances for (A) pure PMN-PT, (B) 1Sm-PMN-PT, (C) 2Sm-PMN-PT.

Real-time displacement signals for pure PMN-PT, 1Sm-PMN-PT, and 2Sm-PMN-PT

Figure S13-4. The real-time signals of displacement for (A, B) Pure PMN-PT, (C, D) 1Sm-PMN-PT and (E, F) 2Sm-PMN-PT.

Nonlinear flexoelectricity in PMN-PT single crystals

Real-time displacement signals for PMN-PT single crystals with thicknesses of 0.7 mm and 1.5 mm

Figure S14. The real-time signals of displacement for PMN-PT single crystal with thickness of (A) 0.7mm and (B) 1.5mm. The second-order coefficient increases as the thickness increases.

The relationship between the flexoelectric polarization and the strain gradient for the topological electrode structures

Real-time Signals for BaTiO3\mathrm{BaTiO}_{3} with topology design

Real-time Signals for 2Sm-PMN-PT with topology design

Real-time output of electrical signals and vibration displacements for PMN-PT with topological electrode structures

Figure S17-1. For PMN-PT with topological electrode structures, the real-time output of electrical signals and vibration displacements when connecting a single-sided electrode.

Output signals and displacement for linear and high-order effects in series ac–bd and ad–bc

Figure S17-2. (A) For linear effects in series ac–bd, the output signal has the same frequency as the displacement. (B) For high-order effects in series ad–bc, the frequency of the output signal is twice that of the displacement.

References

  1. [31]Weilie Zhong. Ferroelectricity Physics. Science Press (2019).
  2. [32]F. Li, M. J. Cabral, B. Xu, Z. Cheng, E. C. Dickey, J. M. LeBeau, J. Wang, J. Luo, S. Taylor, W. Hackenberger, L. Bellaiche, Z. Xu, L. Chen, T. R. Shrout, and S. Zhang, Giant piezoelectricity of Sm-doped Pb(Mg₁/₃Nb₂/₃)O₃-PbTiO₃ single crystals. Science 364, 264–268 (2019).DOI

Paper details

vlnc:W_nchhs7nk
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Physics and Astronomy

Not recorded for this paper

View statement
Looking for related papers…
Contents