Introduction

The coupling of ferroic order parameter with external stimuli gives rise to thermo-, elasto- and electro-optical effects, which offer versatile properties in a single system1,2,3,4,5,6,7,8,9. Interestingly, the optical interaction with ferroic order parameters creates the possibility of remotely controlling the ferroelectric functionalities in a non-destructive way10,11,12. Optical control of ferroelectric properties allows for photostrictive actuation, photostrictive magnetometer, optically controlled memory effect, and light-controlled gas storage properties13,14. In memory applications, an optical stimulus can be utilised for read, write and erase operations, enabling the miniaturisation of wireless integrated circuits by combining image detection, sensing and computing10,11,12,15. It is also important to note that controlling the resistance state via the optically controlled polarisation reversal in ferroelectric materials could be a step forward towards developing energy-efficient neuromorphic computing sensors for information technology and artificial intelligence11.

In this aspect, the exploration of light interaction with ferroelectric polarisation has been reported by demonstrating light-modulated domain wall motion in single crystal BaTiO3 (BTO)16,17, photovoltaic and photostrictive effects in BiFeO3 crystal3,18, polarisation switching via light absorption in MoS2/BTO and MoS2/PbZr0.4Ti0.6O3 heterostructures12,19, controlling the resistance with light in imprinted BTO films11, optical control of domain wall memory in 2D α-In2Se320, optical and thermal control of the polarisation state in PbZr0.2Ti0.8O3 (PZT) thin films21 and organic enantiomeric crystals10. It was also revealed using Density Functional Theory (DFT) that photo-excited carriers in the conduction band can raise the energy of the soft modes and consequently modify ferroelectricity in BTO22,23. It has been established that the optical manipulation of the polarisation response of ferroelectric materials can mostly be correlated with the coupling of light absorption with structural properties, imprint effect, band alignment, interface effect and transport properties10,11,12,24,25. It was observed that 2D-transition metal chalcogenides (TMDs)/ferroelectric-based field-effect transistors with optically controlled, polarisation-dependent electronic transport exhibit several advantages over commercial FeRAMs: low operating voltage, non-destructive data readout, and non-volatile writing and erasing operation by electrical and optical means19,26,27. Also, Long et al. recently demonstrated that optical manipulation of the polarisation state is even possible in ferroelectric BTO tunnel junctions, opening a new possibility to develop energy-efficient memory devices using a simple device architecture11.

However, the reported optical interaction with ferroelectric polarisation is slow, varying from a few minutes to >10 s to switch the polarisation response under illumination in the 2D-TMDs/ferroelectric heterostructure12. In clamped BTO and PZT thin films, it is found to occur at much slower speeds, only after 10–20 min of illumination11,28. Therefore, the observed slow domain switching response poses a significant obstacle to advancements in the development of optically controlled computing. This could be associated with the film-clamping effect, which constrains the outcomes due to the substrate-induced strain, resulting in a slow optically controlled domain switching response. In contrast to clamped films, unconstrained freestanding membranes have already been demonstrated to show superior performance in electromechanical and photo-actuation properties29,30. Also, the observed fast light-mediated domain wall motion in unstrained 2D α-In2Se3 ferroelectrics was attributed to the absence of dangling bonds across their surfaces and their flexible assembly without substrate strain20. Therefore, this naturally leads to the question of whether the rate of optical switching of ferroelectric polarisation would be increased in a freestanding film, and what mechanisms underly optical domain switching. Study of such behaviour can contribute substantially to the understanding of the interactions between ferroelectric domains, strain and substrate clamping. Building this understanding is critical to understand whether future systems could achieve optical control of polarisation reversal at the same rate as electrical switching—a crucial step that would be required for future applications.

In this work, we present optically controlled domain switching in freestanding epitaxial BTO film on the subsecond time scale, surpassing the response times of any observed in ferroelectric oxide systems. In contrast, the domain switching response in a clamped epitaxial BTO film on Nb-doped SrTiO3 (Nb:STO) substrate is substantially slower and is not completely switched even after 30 minutes of illumination. Combining experimental and atomistic simulation, we demonstrate that the coupled effect of the reduction of the energy barrier for domain wall motion under optical excitation and the presence of an imprint bias voltage (VIMP) in the freestanding films initiates the light-induced polarisation reversal. It is also demonstrated that ferroelectric fatigue does not even occur after 75 cycles, measured on a single pixel of a freestanding BTO film, revealing the robustness of the material for developing remotely controlled advanced optoelectronic devices.

Results

Film transfer, structural and electrical properties

A freestanding BTO membrane was fabricated by transferring an epitaxial BTO (46-unit cell) film from an STO substrate to Indium tin oxide (ITO)-coated glass, by dissolving the water-soluble layer Sr4Al2O7 (SAO). The schematic representation of the film transfer process is displayed in Fig. 1a. Despite the occurrence of cracks (in the optical image), most of the BTO membrane remains flat; the topographic image (Fig. 1b) of the freestanding films illustrates average and RMS roughness of 0.98 and 1.2 nm, respectively over a 10 × 10 µm2 area. X-ray diffraction (XRD) (Fig. 1c) illustrates that the epitaxial BTO film (a ~ 3.905 Å and c ~ 4.16 Å) grown on Nb:STO substrate exhibits ∼2.2% compressive biaxial strain state with respect to bulk (a ~ 3.992 Å and c ~ 4.036 Å from JCPDS No. 01-081-2203). The non-linear optical signal is detected at 400 nm for BTO/Nb:STO films using a second-harmonic generation (SHG) experiment under 800 nm excitation. This demonstrates the breaking of inversion symmetry in the system, validating the ferroelectric nature of the film (see Supplementary Fig. 1). The reciprocal space mapping of BTO/STO film demonstrates that BTO and substrate share the same in-plane lattice constant (see Supplementary Fig. 2a). In contrast, for BTO/SAO/STO, the strain state of the BTO film is partially relaxed (c ~ 4.08 Å, see Supplementary Fig. 2b). The freestanding BTO film is almost relaxed (c ~ 4.02 Å) and displays tetragonality close to the bulk value (c ~ 4.036 Å). To analyse the presence of in-plane domains in the system, lateral piezoresponse force microscopy (LPFM) images (see Supplementary Fig. 3) were measured at 0° and 90° tip rotations31. The phase and amplitude responses exhibit a multidomain structure with identical LPFM images for both 0° and 90° tip scanning conditions. Therefore, the domain structure of the BTO membrane shows a c-domain-dominated state with relatively weak in-plane polarisation, which is consistent with the c value deduced from XRD data. Such domain structure is caused by the residual strain, which is not fully relaxed after the release process using SAO films32. The out-of-plane orientation of the domain (c-domain) is further confirmed by the out-of-plane PFM phase hysteresis (Fig. 1d) and butterfly-shaped amplitude loop (Fig. 1e). It is also important to note that the phase and amplitude loop is shifted towards the left (VC− > VC+; VC+ and VC− represent the positive and negative coercive voltage). This indicates the existence of a negative imprint bias voltage (Vimp = −0.55 V) in the freestanding membrane, which results in a preferred downward polarisation (PDOWN)11,33. Additionally, the effective d33 (d33eff) loops of clamped film and freestanding membrane demonstrate the typical saturated hysteresis-type behaviour of the films (see Supplementary Fig. 4). The increase in the remnant d33eff response in the freestanding film can be explained by the fact that clamped film is rigid to the substrate, while freestanding membranes have a vertical degree of freedom for motion30. This allows them to experience larger deflections under electric field, which results in an enhanced piezoelectric effect.

Fig. 1: Fabrication process of freestanding films, and structural and electrical properties of clamped and freestanding films.
figure 1

a Schematic of the film transfer process from as-grown BTO/SAO/STO film onto the ITO substrate. Note that the schematic image illustrates a film with one unit cell, where films of 46-unit cells are used in this experimental study. The optical image shows the released 46-unit cell BTO membrane on the ITO substrate. b Topography image of freestanding films. The inset displays the line profile of the topography image. c XRD scan of BTO/Nb:STO, BTO/SAO/STO, freestanding BTO/ITO and ITO substrates. d Phase and e amplitude loops of freestanding films, displaying positive (VC+) and negative (VC−) switching voltage.

Full size image

Electrical and optical control of domain switching

Poling the film by applying a bias of −10 and +10 V to a PFM tip produced a box-in-box switching pattern with polarisation direction in the upward (bright contrast) and downward (dark contrast) states, respectively, as depicted in Fig. 2a. This shows the electric control of ferroelectric polarisation in the freestanding film. The observation of non-zero amplitude response (see Supplementary Fig. 5a) and 180° phase difference between domains of opposite polarity in the PFM images confirms that ferroelectricity is retained after the film transfer process30. Additionally, outside the poling region, the freestanding film shows a PDOWN domain indicating an imprint field (EIMP) pointing downwards. This is in agreement with the shift of phase and amplitude loops in the negative voltage direction observed in Fig. 1d, e. Upon illumination at 391 nm (0.63 W/cm2; see Supplementary Fig. 6 for experimental setup), the upward polarisation (PUP) domain is now switched to the PDOWN domain, as shown in Fig. 2b. Simultaneously, the PFM amplitude value also increased by the photo-piezoelectric effect in freestanding BTO membranes (see Supplementary Fig. 5b)34. It is to be noted that the optically controlled domain switching response is also demonstrated over a 20 × 20 µm2 area (see Supplementary Fig. 7). In contrast, in the clamped film, the PUP domain is not completely switched to PDOWN even after 30 min of illumination, as shown in Supplementary Fig. 8. The photo-piezoelectric effect of the freestanding membrane is also validated from the d33eff loops measured under dark and illumination conditions (see Supplementary Fig. 9). The increase in the remnant d33eff response is evident in the photo-piezoelectric effect in the freestanding films. The electrostatic artefacts arising from surface charges are minimised by performing measurements at a specific read voltage that corresponds to the surface potential values in both dark and illuminated conditions (see Supplementary Note 9).

Fig. 2: Electrically and optically controlled domain switching in freestanding BTO films.
figure 2

a PFM phase response demonstrating PDOWN (dark) and PUP (bright) regions written by applying +10 and −10 V to the tip over a 10 × 10 μm2 area, respectively. b After 391 nm illumination, PUP (bright) regions switched to PDOWN (dark). c–f Change of phase (Grey line) and amplitude (Orange line) with frequency at positions (1) and (2) c, d after electrical switching and e, f under optical illumination. Current vs voltage response at positions (1) and (2) g, h after electrical switching and i, j under optical illumination.

Full size image

The domain switching of the freestanding film under illumination is also confirmed by a spectroscopy analysis35. A constant drive voltage (VAC = 2 V) is applied to detect the phase and amplitude signal by keeping the force between the tip and sample constant. Frequency-dependent responses are measured at two positions: (1) PDOWN domain and (2) PUP domain during a high-to-low frequency sweep. At position (1) a change in phase from −90 to +90° is induced at the contact resonance frequency (271 kHz, Fig. 2c). Whereas position (2) shows a phase change from +90 to −90° (Fig. 2d). Subsequently, the phase response with respect to frequency is acquired at both the points under illumination. At both positions, the phase now displays a change from −90 to +90° near contact resonance frequency (Fig. 2e, f), demonstrating that the PUP domain has switched to a PDOWN domain state under illumination. The amplitude response is also found to increase under illumination conditions (Fig. 2c–f). The amplitude response is measured over a longer period under dark and light illumination conditions (see Supplementary Fig. 10). To rule out the electrostatic artefacts in the PFM signal, the amplitude response obtained under dark and illumination conditions is fitted with the simple harmonic oscillator (SHO) function. The symmetric fitting of the amplitude response is plotted in Supplementary Fig. 11. Nevertheless, the effect of surface photovoltage on the amplitude response cannot be eliminated completely36.

Furthermore, the I–V characteristics obtained from c-AFM measured in PDOWN and PUP domains reveal the difference in resistance state associated with the polarisation reversal in BTO membranes under illumination. In accordance with previous reports, PDOWN corresponds to the low resistance state (Fig. 2g), and PUP corresponds to the high resistance state (Fig. 2h)12. At PDOWN condition, the polarisation in the membrane is directed towards the ITO electrode, which has a smaller work function than the Pt tip37, a large conduction current is observed in the reverse bias. In this condition, charge can flow easily from the bottom ITO to the BTO membrane due to favourable band alignment, resulting in a low resistance state. Conversely, at PUP condition, the conduction current decreased due to the higher Schottky barrier height formed between BTO membrane and the Pt tip, resulting in a high resistance state. Under illumination, the resistance of the PUP domain decreases significantly but that of the PDOWN domain stays almost unchanged (Fig. 2i, j). The OFF/ON ratio of the membrane is estimated as ∼280 at −3 V with optical switching, which is close to the electric switching OFF/ON ratio of ∼272. Controlling the resistance state by both electric field and illumination demonstrates the potential of the BTO membrane for dual optical–electrical controlled computing.

To exclude optical heating contributions on the domain switching response, we simulated the optical heating distribution maps using the finite element method38. A power heat source from the light beam diameter (200 μm) having 0.63 W/cm2 intensity is simulated in 18.4 nm (∼46-unit cells) of the freestanding BTO system in an ambient atmosphere. Overall, this indicates that between 0.25 and 0.5 K temperature change occurs due to illumination (see Supplementary Fig. 12), which is insufficient to cause any drastic change in the polarisation or domain structure. It should be noted that the applied 0.63 W/cm2 light intensity is nearly 76 times lower than the intensity used for the previously reported optically driven domain-switching response of a BTO thin film11. It is also important to note that the optically controlled domain switching behaviour is only observed for above bandgap excitation; under 529 nm illumination, the PUP domain does not switch to the PDOWN domain state even after 30 min, as shown in Supplementary Fig. 13. Therefore, we observed that sub-bandgap light does not influence the domain switching response.

In order to ascertain the domain switching time under illumination, the PFM phase and amplitude response were measured in the electrically poled region shown in Fig. 3a, turning the light on halfway through the PFM scanning procedure. Here, within the first 64 s, the scan of the upper half of the system shows the same written-in domain contrast as in Fig. 2. After 64 s the illumination starts, and the lower half is sampled. As soon as the light was turned on, the bright contrast of the PUP domains vanished immediately, indicating the PUP domain changed to the PDOWN domain (Fig. 3a). As a result, the amplitude response also increased immediately after light illumination (Fig. 3b). The time taken to measure in vertical directions across 10 × 10 µm2 is shown in Fig. 3a. Line scans of the phase and amplitudes along the yellow (Light off) and purple (Light On) lines are shown in (Fig. 3c, d). However, the optical-driven domain switching time cannot be ascertained accurately from the PFM image analysis under light ON and OFF conditions.

Fig. 3: Determination of domain switching time in freestanding BTO.
figure 3

a Phase and b amplitude images under dark (top) and illumination (bottom) for the initial domain structure in Fig. 2. The yellow and purple line scans under dark and illumination conditions, respectively, are shown in c and d. The time to complete the scan from top to bottom is 128 s, as indicated. e–h Series of phase versus frequency responses at the contact resonance frequency measured under e dark and f–h illumination conditions at times just before and after light was switched on. The optical image and the time when the phase response is captured are displayed in the inset of the graph. i Change in the phase response at 310 kHz with time (numbers in pink) extracted from the series of phase vs frequency responses before and after light was switched on.

Full size image

Alternatively, monitoring the frequency-dependent out-of-plane PFM phase response over time during repeated high-to-low frequency sweep while switching from dark to light illumination allows us to extract the switching time. The hold time during the measurement to capture the phase vs frequency response is 250 ms. In the dark condition, the PUP state shows a phase change from +90 to −90° at the contact resonance frequency (from high to low frequency; equivalent to that shown in Fig. 2d), which completely flips and changes from −90 to +90° at the contact resonance frequency under illumination (equivalent to that shown in Fig. 2f). Furthermore, the phase response illustrates the optically switched PDOWN domain maintains its downward polarisation state throughout optical illumination time. Phase vs frequency response just before and after illumination conditions are displayed in Fig. 3e–h. Before and after illumination, the phase value over time at a specific frequency (f = 310 kHz) is monitored (Fig. 3i), which demonstrates that the phase response changes within 500 ms. This estimate can be verified by monitoring out-of-plane PFM amplitude versus frequency response before and after illumination conditions (see Supplementary Fig. 14). The time measured from the moment of illumination to the point at which the amplitude response increases from its minimum value is defined as domain switching response which is estimated to be <500 ms. Excitingly, this domain switching response time in the freestanding film is nearly 1200 times faster than the previously reported BTO thin film11 and it is also faster than any optically driven domain switching response reported in the literature on ferroelectric thin films21,25,28. A summary of the domain switching time in various ferroelectric thin films is tabulated in Supplementary Table 1.

Repeatability tests of the electrical and optical cycle show that the optically switched domain can be completely retrieved by the application of the electric field. Figure 4h illustrates the optically switched PDOWN domain fully retrieving to the PUP domain after electrical writing, demonstrating reversible electrical and optically controlled domain switching in the freestanding BTO membrane. Also, the domain switching response is captured at a single point in several electrical and optical cycles, rather than scanning the tip, to prevent the influence of local electrostatic artefacts on the repeatability test caused by the PFM tip scanning. Therefore, the phase versus frequency response is captured before and after illumination (Fig. 4j, k) in several cycles and at a particular point. After every illumination process, an electrical pulse with −10 V amplitude was applied for 50 ms to orient the PDOWN domain to PUP domain. This process was repeated for up to 75 cycles (Fig. 4l, m). The phase value at a particular frequency (f = 320 kHz) was recorded in each cycle and plotted in Fig. 4n. The stable phase value after electrical (∼+90° ± 10°) and optical switching (∼−90° ± 10°) at 320 kHz illustrates that ferroelectric fatigue behaviour is not observed in the freestanding BTO film after 75 electrical and optical illumination cycles. However, after 75 electrical and optical cycles, we observed that the contact resonance frequency is found to be lower than (<10 kHz) the first cycle (Fig. 4j, k). This behaviour may relate to the change in mechanical properties due to the damage of the measured pixel caused by multiple electric cycles, yet the contribution of electrostatic artefacts and tip damage cannot be neglected completely39,40. We also tested the electrical and optical domain switching response up to 10 cycles over 1600 pixels (1 pixel size \(\approx\) 25 × 25 nm), in order to examine any fatigue effects averaged over a larger area. Note that these measurements are performed in contact mode with tip lift-off between pixels to avoid local electrostatic force originating from the several scanning processes of the PFM tip over the sample surface. The density plots of the PFM phase response with frequency at 1600 pixels for 1st and 10th cycles are plotted in Supplementary Fig. 15, showing no evidence of ferroelectric fatigue effects.

Fig. 4: Cyclic domain switching response of freestanding films under electric field and illumination.
figure 4

a–c PFM after 1st electrical cycle. d–f PFM under illumination afterwards, and g–i PFM after 2nd electrical cycle for the same sample area. From top to bottom: Topography, phase and amplitude images. For the point marked by a red arrow in e the phase versus frequency response is shown j and l after −10 V electrical pulse for 50 ms and k and m under illumination. After the first set of cycles from j to k,  from l to m, there are 75 electric and optical cycles. n Phase value at 320 kHz for all measurement cycles. The yellow and purple crosses indicate the value of phase at 320 kHz frequency in the dark and after illumination.

Full size image

Atomistic simulations

In order to explain the domain-switching by illumination and its dependence on clamping, we now turn to ab initio simulations. The simulated polarisation in relaxed BTO is decreased by photo-excited charge carriers (Fig. 5a). At the same time, tetragonal strain (c/a) and energy difference between tetragonal ferroelectric and cubic phase (with P = 0) decrease. Without illumination, the tetragonal phase with c/a = 1.023 is about 17 meV/f.u lower in energy compared to the paraelectric one. In the optically excited material with 0.03 or 0.08 e/f.u. in the conduction band, c/a decreases to 1.019 and 1.006 e/f.u., respectively. For 0.08 e/f.u. in the conduction band, the energy difference between both phases is reduced to 1 meV/f.u22. We find that the ferroelectric tetragonal phase is no longer metastable for 0.1 e/f.u. This gradual loss of the ferroelectric state with light-induced charge excitation can be understood by the screening of long-range Coulombic interactions and the resulting weakening of the phonon instabilities related to ferroelectric ordering22. The opposite trend is induced by 2.2% compressive epitaxial strain, i.e., perfect clamping to the substrate, where tetragonal distortion, ferroelectric instability and ferroelectric polarisation increase with the magnitude of strain, as expected41. Without illumination, this compressive strain induces a state with P of about 59 µC/cm2 and increases the energy difference between paraelectric and ferroelectric states approximately by a factor of three (58 meV/f.u.). Even if 0.08 e/f.u are excited to the conduction band, this value of strain stabilises a ferroelectric state with c/a = 1.084 and 57 µC/cm2 (Fig. 5a) which is 46 meV/f.u lower in energy than the strained state with P = 0. Thus, the strain completely compensates the reduced ferroelectric stability induced by illumination. Although the charges partly screen the long-range interactions, the large polarisation—strain coupling dominates the subtle interplay of different energy contributions.

Fig. 5: Polarisation changes under illumination simulated by DFT and domain wall properties at finite temperatures by MD simulations.
figure 5

a Change of polarisation in clamped (compressive strain 2.2%) and relaxed BTO (tetragonal phase, no temperature, no clamping) by photo-excited carriers. Atomic structure of b BaO-centred (ground state) and c TiO2-centred (metastable state, energy barrier for wall motion) domain walls excerpted from the 16 × 1 × 1 unit cell used in DFT simulations. Arrows illustrate the local polarisation of each Ti-centred unit cell (arb. scale). d Polarisation profiles across BaO-centred (blue) and TiO2-centred (red) tetragonal 180° domain walls and their changes with the concentration of excited charges in e/f.u. (symbols). e Energy (∆E) landscape diagram of imprint ferroelectrics. The variation of energy landscape diagram of clamped (compressively strained) and freestanding (relaxed) films under dark and illumination. f Examples of polarisation profile 〈Pz〉x (average of Pz for all unit cells at the same x position) across a tetragonal BaO-centred 180° domain wall under 0.4% and 0.6% biaxial compressive strains found in finite temperature MD simulations. g Change of the critical field strength Ecr for domain wall motion under biaxial strains (negative/positive for compressive/tensile strain) found in MD simulations. The critical field strength of the unclamped material is shown by the star symbol.

Full size image

As homogenous switching from P to −P has to pass the state with P = 0, the photo-excited carriers systematically reduce the energy barrier for field-induced polarisation switching, while this barrier increases under compressive strain. Ferroelectric switching under moderate field strengths is however dominated by the thermally activated motion of domain walls rather than by homogenous switching of the polarisation41,42. Therefore, it is useful to analyse the domain wall properties in the presence of excited charges.

In order to determine the energies and polarisation profiles of BaO (Fig. 5b) and TiO2 (Fig. 5c)-centred domain wall using DFT, a supercell with 16 × 1 × 1 f.u. is considered as a model system. Blue lines in Fig. 5d show the polarisation profile across a tetragonal 180° domain wall centred on a BaO-plane. This wall configuration is energetically most favourable, and we obtain a domain wall energy of about 10 mJ/m2 without illumination, in agreement with previous work43. The energy barrier to shift the domain wall from one plane to another BaO-plane is given by the energy of TiO2-centred walls. While BaO-centred walls are atomically sharp with a slight reduction of polarisation on both interfaces, the TiO2-centred wall on the other side is in the less favourable state with one P = 0 layer, as indicated by the red lines in Fig. 5d. Without charge excitation we find an energy of about 38 mJ/m2 for the unfavourable wall position and energy barrier for wall motion of about 28 mJ/m2. The energy of the TiO2-centred domain wall depends on strain and charge carriers as the energy differs between paraelectric and ferroelectric states. Without strain, the energies of BaO and TiO2-centred walls for 0.03e/f.u. excited carriers are reduced to about 1 and 2 mJ/m2. Therefore, the energy barriers to nucleate and move domain walls strongly decrease under illumination. The schematic representation (Fig. 5e) of the energy landscape for the clamped and freestanding films displays the change in polarisation and barrier height under illumination. It demonstrates the reduction of barrier height in relaxed films due to the removal of substrate-induced strain, which is further reduced under illumination. Note that, our results only give insights into the well depth and not into the slopes of the energy curve.

As domain wall motion is a thermally activated process, we now turn to molecular dynamics (MD) simulations at a temperature where unclamped bulk shows tetragonal phase. Figure 5f shows the change of the polarisation profile across a tetragonal 180° wall by compressive strain at this temperature. In agreement with the DFT predictions, the thermalised wall is flat and corresponds to a BaO-centred wall, without layers with P = 0. Furthermore, the polarisation in the domains increases with compressive epitaxial strain. Note that thermal fluctuations reduce the magnitude of polarisation and the polarisation in the unclamped bulk at the given temperature is about 29 µC/cm2. Without an external field, thermally activated wall motion by one layer is possible but a rare event at the given temperature. Under a finite electrical field parallel to the polarisation in domain, the field stabilises this domain and can be distinguished in three different field regimes: (1) Rare motion of domain walls if the field is below the critical field strength Ecr. The energy barrier to shift the domain wall is too high and the domain walls are pinned. (2) Field-induced switching by domain wall motion whose velocity increases with the field strength and (3) ultra-fast homogeneous nucleation and switching starting in the domain in the high field regime42,44.

The microscopic process of the moving wall is observed in regime (2) for a plane which is initially polarised antiparallel to the applied field (−P shown in blue) in front of the moving wall (see Supplementary Fig. 16). With time quasi-2D nuclei (red, +P) form and expand45. The intermediate state and energy barrier for the motion of the domain walls is thus neither the state with macroscopic P = 0, nor the Ti-centred domain wall with one layer with polarisation equal to zero. Rather, it is the energy barrier for small quasi-2D clusters to switch via Pi = 0 and nucleate a reversed cluster. Analogously, regime (3) is characterised by the nucleation of reversed clusters in the domain which expand in all three directions.

For a bulk-like material without clamping, we find Ecr = 70 kV/cm (red star symbols in Fig. 5g) for the chosen temperature, while a field of about 300 ± 10 kV/cm activates the ultrafast switching in the domain. Note that the field strengths cannot be compared quantitatively to the experiment46,47. More relevant is the change of these field strengths with strain. Clamping without strain reduces the critical field strength by about 20% (50 kV/cm), while homogeneous nucleation already is induced at 200 kV/cm. The clamping reduces tetragonal distortion and polarisation in the external field. Further reduction of polarisation and critical field strength is observed for small values of tensile strain. With compressive strain, the critical field strength increases non-linearly and reaches more than 1000 kV/cm for 1% compressive strain, i.e., increases by more than a factor of 10 compared to the unclamped bulk. Importantly, this is related to a large increase in tetragonal distortion and polarisation (45.6 µC/cm2 in the +P domain for 1000 kV/cm and 1.1% strain). For a given strength of the external field or internal bias field, significant changes in switching behaviour occur due to strain. Clamping to compressive strain reduces domain wall velocity and can even result in full pinning for moderate field strengths. As shown by the DFT simulations above, the same trends with strain are to be expected under illumination.

Discussion

Optically controlled domain switching in ferroelectrics is an effect of one or more of the following phenomena: (a) coupling of optical absorption with imprint effect11, (b) charge screening by the photo-generated charge carriers at a Schottky junction11,23, (c) charge screening by the injected electrons at the interface12,19, (d) light-induced reversible structural change due to photo-isomerization10, (e) reduction of the ferroelectric instability by excited carriers22, and (f) optical heating effect. In our work the optical heating effect has been eliminated by the finite element calculation. Our DFT simulations show that photo-excited charge carriers systematically reduce the energy barriers to nucleate and move ferroelectric domain walls. Depending on the strength of the applied field or of the imprint field, ferroelectric switching can be activated. Generally, the origin of the imprint field is correlated with several factors, such as the growth during thin film deposition, electrostatic boundary conditions, and surface termination, which favours a preferred polarisation direction48,49,50. In our system, the imprint field creates a downward polarisation in the clamped film, indicated by dark PFM contrast in the unpoled area (see Supplementary Fig. 8a), and it is maintained in the freestanding film even after the film transfer process (see Fig. 2a). The important finding in our study is that the domain switching effect is only seen in the freestanding BTO membrane and not in the clamped BTO film, despite the observation of imprint effect and the production of photogenerated charge carriers in both systems. The only observable difference we noted is that the clamped BTO exhibits around 2.2% compressive strain resulting from substrate-induced strain, whereas the freestanding BTO is relaxed. These differences can be reproduced by DFT and MD simulation. With increasing compressive strain, the energy barrier for polarisation reversal systematically increases and can compensate the effect of the optically excited charge carriers. For the strain of 2.2%, the critical field strength to active domain wall motion increases by more than one order of magnitude. Therefore, from the MD simulation, we observed that the tensile strained and relaxed BTO systems exhibited significantly faster domain wall motion under the influence of an electric field, in contrast to the compressive strained BTO (see Supplementary Fig. 17). We also demonstrated that the motion of the domain wall occurs exclusively under above bandgap illumination conditions. As a result, domain switching takes place with 391 nm illumination, while it does not occur with 529 nm. Combining the PFM and DFT with MD simulation results, we find that the optical-driven polarisation reversal phenomenon in freestanding films is triggered by three factors: (i) removing the substrate-induced strain, which reduces the energy barrier for domain wall motion and increases the domain wall velocity, (ii) lowering the polarisation and therefore the energy barrier for domain wall motion under illumination and (iii) coupling of imprint field with the above factors.

In summary, we have demonstrated that the combined effect of imprint field and reduction of energy barrier of domain wall motion under illumination influences switching of ferroelectric polarisation and the resistance state in freestanding BTO film. The freestanding film shows dual stimulus (electric field/light) control of the resistance state due to the change in the polarisation response. The elimination of substrate-induced strain in the freestanding BTO film has a significant impact on increasing domain wall velocity, consequently resulting in much faster domain switching speed than the clamped films under illumination. It is established that the observed optically controlled domain switching speed in our work is in the subsecond timescale, which is nearly 1200 times faster than the previously reported clamped BTO film11. We demonstrated 75 cyclic electric and optical measurements to alter polarisation switching in the material, confirming the robustness of the observed properties in the membrane. An important outcome of the understanding that we have obtained is that an alternative and more feasible approach to achieving complete optical control of polarisation in the ferroelectric device could be obtained by fine-tuning the strain of a device formed from ferroelectric membranes. One possible way to control the strain of the device is by employing external strain through substrate engineering. In future work, the observed phenomenon could also be efficiently utilised by exploiting narrow bandgap ferroelectrics, such as BiFeO3, [KNbO3]1−x [BaNi1/2Nb1/2O3−δ]x membranes, for visible-light controlled wireless optoelectronic applications.

Methods

Film fabrication

BTO films were grown on (001) TiO2-terminated single-crystalline STO substrate with SAO buffer layers by oxide molecular beam epitaxy. SAO and BTO layers were deposited in sequence at a substrate temperature of 850 °C with an oxygen pressure of 1 × 10−6 Torr. In addition, BTO films were grown on Nb:STO substrate without SAO layer for comparison.

Film transfer on ITO

As-grown epitaxial BTO film (46-unit cell) on STO substrate with a Sr4Al2O7 (SAO) sacrificial layer is used to transfer onto transparent ITO substrate using a homemade clamping device. The BTO/SAO/STO sample is placed upside down onto a 10 × 10 mm2 ITO-coated glass substrate so that the BTO side is kept in contact with the conducting ITO base. After water treatment, the whole system was detached, and the freestanding BTO was left behind on the ITO substrate. The experimental details of the film transfer process can be found elsewhere51.

XRD analysis

BTO crystalline structures were characterised by high-resolution XRD with a Bruker D8 Discover diffractometer. 2θ–ω measurements were performed to confirm the high crystallinity of BTO/SAO/STO, BTO/Nb:STO and freestanding BTO samples.

PFM and C-AFM measurements

PFM measurements were performed using Bruker ScanAsyst Dimension icon AFM in the contact resonance mode. The SCM-PIT-V2 (Pt/Ir coated Si) cantilevers were used for the measurements. Prior to the measurement, a common calibration practice was followed using a standard ferroelectric BTO crystal52,53. The SS-PFM measurements were performed in order to obtain the phase and amplitude loops in the bias-off conditions. The c-AFM measurements were performed using the same tips used for PFM in the tip ground condition. Prior to the measurements, the current amplifier (1 pA to 100 nA sensitivity) was calibrated using a 10 MΩ resistor. The current was directly measured in the contact mode through the calibrated built-in current amplifier. Photo-induced PFM and photocurrent measurements were carried out under bottom illumination using Redoxme variable LED light source connected to the multimode optical fibre. The photo-induced PFM measurements were performed under top illumination conditions for BTO/Nb:STO sample using 378 nm (5 mW) Laser illumination (Vortran Laser Technology). The intensity of the fibre output and laser was determined by the PM100A power metre (Thorlabs) connected to the S120VC (Thorlabs) optical sensor. The repeatable test is performed in the CR-PFM module for 10 cycles over 1600 pixels, where the measurement of each pixel is performed in the contact mode, and the time for each pixel measurement is 15 ms. In order to move to the next pixel, the tip moved from one position to another position in the lift-off conditions.

Density functional theory simulations

DFT simulations of the tetragonal phase of bulk BTO were performed using the abinit programme package54 together with Wu–Cohen potentials55 and Fermi–Dirac smearing of about 0.1 eV. An energy cutoff of 952 eV and a 8 × 8 × 8 k-mesh together with thresholds of 5 × 10−7 eV/Ang and 2.5 × 10−5 eV/Ang for differences in forces between self-consistency steps and forces on ions in structural optimisation, respectively, guarantee the high accuracy of the simulations.

Following the approach of Ref. 22 above bandgap optical excitations were modelled by 0–0.1 e/f.u. electrons excited to the conduction band. The local polarisation in each Ti-centred/f.u. was determined by \({P}_{i}=\frac{1}{\Omega }{\sum}_{j}{w}_{j}{Z}_{j}\Delta {z}_{j}\), with Ω the volume of the unit cell, j running over all nearest neighbours of atom i and wj their weight factor (1/8 for Ba and 1/2 for O), with ∆zj the shift of ion j relative to the paraelectric state, Zj the Born effective charge of the ion direction taken from (ZBa: 2.74e, ZTi: 7.492e, ZO1 = ZO2 = −2.150e, ZO3 = −5.933e)56,57. Additionally, supercells with 16 × 1 × 1 f.u. with two domains (P and −P) each with a width of 8 u.c separated by two 180° [100] domain walls were studied. It has been reported that this domain size is sufficient to avoid finite size effects58. Cell shape, volume and the positions of the central layers in each domain were fixed to their bulk values. As the TiO2-centred wall is only metastable, also Ti and O atoms on the central plane of each domain have been fixed. For the supercells, the k-mesh were reduced to 2 × 8 × 8 and the thresholds for force relaxations have been increased by one order of magnitude.

Molecular dynamics simulations

The domain wall dynamics in BTO were simulated using the open-source feram code59 utilising the effective Hamiltonian by Zhong et al. parametrised with DFT calculations for BTO56,60,61. In this coarse-grained approach, the energy surface is expressed as a function of local optical displacement vectors, i.e., dipole moments, and local acoustic displacement vectors, i.e., local strains, and a global homogeneous strain. Due to the coarse-graining and the additional internal optimisation of strain, instead of the 15 atomistic degrees of five atoms, only the 3 degrees of freedom of the local dipole moment are explicitly considered in each unit cell62.

The model predicts the paraelectric to ferroelectric phase transition of a cooled sample without strain at about 280 K. This underestimation of the ferroelectric transition temperature is well understood and does not influence the qualitative changes of the material properties with strain63. Without loss of generality, we equilibrated the system well in the tetragonal phase at 260 K using the Nosé–Poincare thermostat with a time step 1 fs64.

To mimic strained films, we fixed the strain in the (001) plane to η1 = η2 and η6 = 0 while influences of interfaces and surfaces, including depolarisation, interface termination, and atomic relaxation were neglected. We sampled the strain range of −1.1 to 0.2% relative to a = 3.986 Å which is the lattice constant for the metastable cubic phase at 0 K in our model, using a 0.1% grid.

We used periodic boundary conditions in combination with system sizes of 164 × 164 × 46 f.u. Each simulation cell contained two domain walls with initial distances of about 32 nm. The system was initialised with polarisation along ±z-direction and two 180° equidistant domain walls as follows: local dipoles were randomly initialised and pre-poled in local fields of ±100 kV/cm along z, (see Supplementary Fig. 18) for 30 ps. After removal of these fields, equilibration continued for 30 ps; finally, an external field was instantaneously applied along +z direction.

The dependence of the field response on the field strength was sampled over 10 ps with a time resolution of 1 ps and a field resolution of ΔE = 10 kV/cm in 3 independent runs for each field strength. We determine the domain wall velocity by the shift of the domain wall centre x0 with time. The domain wall centre is estimated by fitting the mean-polarisation per x-layer\(\,{\left\langle {P}_{z}\right\rangle }_{x}\) every 1 ps via \({\left\langle {P}_{z}\right\rangle }_{x}={P}_{0}\cdot {tanh} (\frac{x-{x}_{0}}{{d}_{{DW}}})\,+\varepsilon (x)\). Here, P0 and dDW saturation polarisation and the width of the wall, respectively. The second term ε(x) accounts for polarisation response on the applied field within domains, including linear dielectric response and transient negative capacitance44. Non-equilibrium effect may initially boost the wall velocity for this instantaneous field applications44. Therefore, the first 1 ps were excluded from the analysis. We classify a field strength as regime 1, rarely moving walls, if the wall in the centre of the simulation cell moved less than 1 f.u. in 2 of 3 runs. Note that the critical field strength cannot be directly compared to experiments as simulations on the idealised material commonly overestimate the critical field strengths for domain wall motion and polarisation switching.

COMSOL simulation

To decouple the role of light-induced heating in the thin films, we simulated the temperature distribution maps using the finite element method using the Heat Transfer Module of the commercial COMSOL Multiphysics software38. The heat transfer model in solids (ignoring the heat convection of air) is governed by the following equation,

$$\rho {C}_{p}\partial T/\partial t+\nabla .q=Q$$

Where the expression for heat flux q is given by

$$q=- \! \! k\Delta T$$

Here ρ is the material density, Cp is the heat capacity, T is the absolute temperature, Q is the power density of the heat absorbed from the incident radiation (we assume that all the incident light fluence is absorbed by the sample) and k is the thermal conductivity of the materials.

To consider the radiation from boundaries to the ambient surroundings, we used a boundary condition following the Stefan–Boltzmann law.

$$-n\,.\,q={{\rm{\epsilon }}}{{\rm{\sigma }}}({T}_{{amb}}^{4}-{T}^{4})$$

Where ε is the surface emissivity, σ is the Stefan–Boltzmann constant (5.67 × 10−8 W/m2K4) and Tamb is the ambient temperature (Tamb = 300 K) of the surroundings. The values of the physical parameters of the materials used in the simulation are given in Table 1.

Table 1 Physical parameter of materials used in the COMSOL thermal simulation11,65
Full size table

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.