Disentangling electronic and thermal contributions to light-induced resistance switching in BaTiO₃ ferroelectric tunnel junction

Ferroelectric materials are of interest for diverse applications, from sensing and powering devices to data storage and logic functionalities, such as nonvolatile memristors in an artificial neural network.¹ A ferroelectric tunnel junction^2–4 consists in a few nanometers thick ferroelectric layer sandwiched between two metal electrodes. In this junction, switching the polarization direction or its magnitude results in a change of resistance.⁵ Therefore, by controlling ferroelectric polarization with external electric stimuli, it is possible to modify the resistance state of the device, mimicking a memristive behavior. Most commonly, ferroelectric polarization is controlled by the application of suitable electric fields, which allows the modulation of the polarization magnitude and its direction.⁴ In a practical realization, this process requires charging and discharging the ferroelectric capacitor and the switching time is not limited by the intrinsic switching and motion speed of ferroelectric domains but convoluted with the time constant of the measuring circuit. In an alternative approach, the polarization state can also be affected by light with promises of faster response,⁶ and the manipulation of ferroelectric polarization by light has been observed in several systems.^7–16 The response of polarization to a suitable illumination can have different physical origins, mainly classified into two categories: (i) electronic and (ii) thermal effects. In the former case, photogenerated carriers would destabilize polarization and, in the presence of a symmetry-breaking electric field, a reversal of polarization can take place.^17–20 In the latter case, light radiation can locally increase the temperature that may lead to a reduction of the coercive voltage and, again in the presence of a symmetry-breaking electric field, to a polarization reversal.²¹ Whereas the description above refers to radically different scenarios, in practice, however, both effects can operate unisonally, leading to a limited understanding of the microscopic mechanisms at operation and challenging the control of optical switching of polarization.

In a broader context, the response of ferroelectric capacitors to optical stimuli may not only involve the electronic-thermal effects occurring in the ferroelectric layer, but also the optical responsivity of electrodes themselves. For instance, it has been reported that the light-dependent depletion layer width in doped semiconducting electrodes (Nb:SrTiO₃) produced changes in electroresistance (ER) across Nb:SrTiO₃/Sm_0.1Bi_0.9FeO₃ tunnel junctions.²² In contrast, the photocarrier generation in narrow bandgap intrinsic semiconductors (MoS₂) in a MoS₂/BaTiO₃ (BTO) device is claimed to lead to changes in screening properties with subsequent impact on the electroresistance of the devices,¹⁹ while persistent changes of conductivity in ferroelectric films and ferroelectric capacitors have been reported and attributed to metastable charge trapping²³ or, when using energetic photons, to the formation of photoinduced ionic defects.²⁴ The polarization-dependent response in the mentioned Nb:SrTiO₃/Sm_0.1Bi_0.9FeO₃ and MoS₂/BaTiO₃ samples above^19,22 indeed suggests different scenarios.

In a recent example, the optical switching of ferroelectric polarization and subsequent reversal in ferroelectric tunnel junctions based on BaTiO₃ (BTO)/SrTiO₃(STO) bilayer has been shown to lead to switching from a low-resistance state (LRS) to a high-resistance (HRS) state.²⁰ It was proposed that photocarriers generated within the ferroelectric layer overrule thermal effects. Here, we aim at further disclosing the prevalent mechanism for optically induced polarization reversal. We report on the role of the wavelength of light on the observed response, explore the time dependence of the photoresponse, and elucidate the impact of possible thermal effects. We conclude that the effect is mainly dominated by photogenerated charges rather than heating and we discuss why large light intensity or long illumination times are required to observe the optical polarization/resistance switch of the ferroelectric tunnel device.

All samples mentioned in this work were grown by Pulsed Laser Deposition (PLD). The structures are STO//La_1/3Sr_2/3MnO₃ (LSMO, 30 nm)/BTO (4 nm) (results shown in the supplementary material), STO//LSMO (30 nm)/STO (1 nm)/BTO (4 nm), and STO//LSMO (30 nm)/BTO (70 nm). The heterostructures were deposited in a single process. The LSMO layer was grown at 725 °C, at an oxygen pressure of 0.1 mbar and a laser frequency of 2 Hz. The BTO was grown at 700 °C, at an oxygen pressure of 0.02 mbar and 2 Hz. STO dielectric layer was grown at 700 °C, at an oxygen pressure of 0.02 mbar and 2 Hz. Further experimental details and structural characterization can be found elsewhere.²⁵ The role of the STO layer in the device is to enhance the ER and to reduce the leakage as discussed elsewhere.²⁰ Top Pt electrodes (20 nm thick) were deposited by sputtering through a stencil mask, obtaining arrays of circular contacts of 20 μm diameter in the STO//LSMO(30 nm)/BTO (4 nm) and STO//LSMO (30 nm)/STO (1/nm)/BTO (4 nm) samples, and square contacts of 60 × 60 μm² in the STO//LSMO (30 nm)/BTO (70 nm) sample.

Electrical characterization was conducted as follows: the bottom electrode (LSMO) was grounded while the V(t) signal was applied to the top electrode (Pt). Resistance was extracted by evaluating it at 0.5 V from I(V) curves collected in the ±0.5 V range. As mentioned in the text, the resistance was measured after applying trapezoidal voltage pulses of amplitude (V_W) and time width τ_write = 100 μs, where τ_write is the rise time, plateau time, and decay time of the trapezoidal signal. Electroresistance (ER) loops were obtained after successive application of V_W and resistance measurements, where V_W was varied following a bipolar triangular shape. All measurements were done at room temperature using an AixACCT TFAnalyser 2000 platform. Piezoelectric force microscopy (PFM) measurements were performed with an MFP-3D ASYLUM RESEARCH microscope (Oxford Instrument Co.), using the AppNano silicon (n-type) cantilevers with Pt coating (ANSCM-PT-50) and these are shown in the supplementary material.²⁶ Optical illumination was performed with blue light (λ = 405 nm) and red light (λ = 638 nm), with a maximum illuminating power density of ≈9 W/cm²) driven by a CPX400SA DC power (AimTTi Co.). The illumination time is denoted as τ_light. The laser source is installed on the anti-vibration experimental table with a fixed inclination (45°). The diameter illumination spot is ∼250 μm, thus covering the whole capacitor area. The COMSOL finite-element thermal model has been used to calculate thermal gradients and the results are included in the supplementary material.

ER loops were collected in dark and under illumination in a LSMO/STO/BTO (4 nm)/Pt junction as illustrated in Fig. 1 (inset). The ER loop measured in dark (Fig. 1, squares) shows a high-resistance state (HRS, ∼6 × 10⁵ Ω) corresponding to polarization down (P_DOWN, toward LSMO), and a low-resistance state (LRS, ∼2 × 10⁵ Ω) corresponding to polarization up (P_UP, toward Pt) with ER = (HRS – LRS)/LRS ≈ 200%. Note the obvious existence of an imprint field, as revealed by the shift toward negative voltage of the ER loop (blue vertical line). The imprint field is −4.4 MV/cm toward LSMO, indicating that P_DOWN is the most stable state. The ER loop collected under illumination during 1 min after approximately 10 min under illumination using a λ = 405 nm laser beam of a power density of 9 W/cm² is also shown (Fig. 1, circles). It can be readily appreciated that ER is virtually suppressed (ER ≈ 0), being definitely smaller than the ER recorded in dark. Note also that the resistance value of the ER loop collected under illumination roughly equals to the HRS of the ER loop in dark. Finally, the ER loop is collected again in dark (Fig. 1, triangles). It can be observed that the ER loop recovers its initial shape and magnitude (ER ≈ 200%).

Resistance as a function of time [R(t)] has been measured when the sample was illuminated using different light wavelengths (blue λ = 405 nm, red λ = 638 nm) [sketches in Figs. 2(a) and 2(b)]. Pre-poling was conducted in dark before illumination, by using a trapezoid pulse of amplitude V_W = ±8 V and τ_W = 100 μs, to set a specific initial state [either HRS(P_DOWN) or LRS(P_UP) indicated by red/blue triangles in Figs. 2(c)–2(f)]. The power density for each light wavelength is fixed at 1.4 W/cm², time of illumination lasts for ∼200 s [bluish background in Figs. 2(c)–2(f)]. In Fig. 2(c), we show R(t) after setting the LRS initial state. Data show that under blue illumination, the resistance gradually rises from an initial LRS to a final HRS that closely coincides with the HRS dark value at ∼100 s. In sharp contrast, when an HRS is set in the dark as initial state, illumination does not modify the resistance state at all, i.e., it remains at HRS [Fig. 2(d)]. Considering that the resistance of these ferroelectric tunnel junctions is determined by remnant polarization, the results shown in Figs. 2(c) and 2(d) indicate that light has induced a polarization reversal from P_UP to a final state P_DOWN. Importantly, the polarization direction in the final state (P_DOWN) is dictated by the direction of the imprint field (pointing down toward LSMO) as shown in similar systems.²⁷ In Figs. 2(e) and 2(f), we plot R(t) of similar experiments but collect using the red light. Irrespectively, if the initial state is set to LRS or HRS, light does not appear to produce any significant change. Direct evidence of light-induced reversal of the remnant polarization from P_UP to P_DOWN has also been observed in a LSMO/BTO(4 nm) sample using PFM (in the supplementary material). In Fig. 3, we show the resistance measured after a series sequence of stimuli [V_W (τ_write = 100 μs) of different polarities, and blue or red illumination]. In Fig. 3(a), the HRS (P_DOWN) state is written by V_W = +8 V pulse (step 1); then, the LRS (P_UP) state is written by a V_W = −8 V pulse (step 2); LRS remains at the same resistance level after illumination of red light (step 3) and remains at the same level in dark (step 4); HRS recovers (step 5) after illumination of blue light and remains in HRS in dark (step 6). Repeating the previous operation (from step 3 to step 5) in the following step 6 to step 9, no substantial change is observed. Complementarily, in Fig. 3(b), the LRS (P_UP) state is written by V_W = −8 V pulse (step 1); then, the HRS (P_DOWN) is written by V_W = +8 V pulse (step 2); then, the subsequent data are collected following the same protocol as in Fig. 3(a), and no substantial change has been observed. Thus, the collected data illustrate that an optical switch from LRS (P_UP) to HRS (P_DOWN) only occurs using blue light. Therefore, the light wavelength and the initially set state determine the occurrence of a resistance switch under illumination.

In order to further investigate the photoresponse dependence on wavelength, we measured photocurrent generated under blue and red-light illumination. Figure 4(a) shows the short-circuit photocurrent as a function of time. Data show that under blue illumination [bluish region in Fig. 4(a)], a clear j_sc ≈ 12 μA/cm² develops. The I–V curve under illumination is shown in S2 in the supplementary material. Instead, no short-circuit photocurrent develops under red illumination [reddish region in Fig. 4(a)] for blue illumination. Note that the generated photocurrent is very small, and, thus, it does not affect significantly the measured resistance (corresponding to a current density of ≈150 mA/cm²) shown in Figs. 1 and 2. The difference in photoresponse depending on the used wavelength is expected from the fact that the blue light wavelength is 3.06 eV, near a BTO bandgap (E_g = 3.3 eV). Although the photon energy is somewhat smaller than the bandgap of BTO, deep or shallow levels can contribute to a significant photon absorption.^29,30 In the case of red light, with a photon energy of 1.9 eV, photoabsorption and, thus, photocarrier generation should be much depressed, which is consistent with the absence of short-circuit photocurrent. In addition, the absence of any effect of the device resistance when using red illumination indicates that photon-induced thermal effects are also negligible. A plausible scenario is the one sketched in Fig. 4(b), where the energy profile of the LSMO/BTO/STO/Pt device is depicted. The direction of the imprint field E_imp and the initial state of polarization P_up (LRS) are indicated. The electron–hole pairs generated by blue photons within BTO come to screen P_up and P are switched to P_down following E_imp. To further crosscheck the possible contribution of heating in the reported resistance change temperature, distribution maps have been simulated. These are based on the finite elements calculation of photoinduced thermal effects. On them, a power heat source (diameter 250 μm) of 9 W/cm² is simulated in the STO(001)//LSMO/STO/BTO system on the top of a PVC stage surrounded by air T_amb = 20.15 °C. The heat source has been placed at the BTO surface. The initial temperature of the model is equal to T_amb (20.15 °C). We see from the simulated result in Fig. 5(a) that there is a thermal gradient from the beam center to the borders, but the temperature change is limited to 4 °C. Figure 5(b) shows the temperature dependence on time of A and B points market in Fig. 5(a). It can be inferred that within 1000 s, the temperature rises and saturates. We find that at point A ΔT ≈ 4 °C, while at point B ΔT ≈ 2 °C. Note that the result is based on the assumption that all light fluence can be completely absorbed by the sample; thus, less temperature increase is expected to occur. More details about the finite-element thermal model are described in S3 in the supplementary material. The simulation results allow us to conclude that the sample temperature increase could not be larger than 4 °C; thus, the temperature increase is unlikely to produce any polarization reversal. Therefore, the scenario depicted in Fig. 4(b) and already described is the most plausible one.

An additional salient observation in Fig. 2(c) is the remarkably slow nature of the switching process (≈100 s). This slow response is not expected from a photovoltaic response because carrier generation is an exceedingly faster process.³¹ To get a hint into the ultimate reason for this relatively slow process, we compare the photoinduced change of resistance in BTO ferroelectric capacitors where the thickness of the photoabsorbing layer (BTO) largely differs. In Fig. 6(a), we show data collected in LSMO/STO/BTO(4 nm)/Pt and LSMO/70 nm BTO(70 nm)/Pt junctions, under blue photon illumination at 0.1 W/cm² power density. Note that the STO(1 nm) layer is not present in the thicker film, which does not affect the conclusions discussed as follows. Consistent with the results of Fig. 2(c), for the 4 nm BTO junction, an illumination time of ≈100 s is needed to switch gradually from LRS to HRS. In contrast, the 70 nm BTO junction responds faster to illumination and the required time for LRS to HRS transit is only ≈70 s. Note that in this, the thicker film resistance is most probably controlled by the Schottky barrier height, which itself is modulated by the ferroelectric polarization, rather than tunneling.³² In any event, the data above suggest that the concentration of photocarriers, which is ultimately dictated by the BTO thickness, is a crucial parameter to account for the different time responses. An additional insight can be derived from R(t) measurements under different laser powers. In Fig. 6(b), we show R(t) data collected in STO/BTO(4 nm) junction at different power densities: 0.1, 1.2, and 9 W/cm². Data clearly show that by increasing light power, the time scale for resistance state change shortens from ∼200 s to ∼50 s. Similar effects are observed in BTO(70 nm) capacitors (S4 in the supplementary material).

Therefore, Figs. 6(a) and 6(b) show that a faster change of resistance is observed in thicker BTO film and using higher light power intensity. Therefore, it can be inferred that the total amount of photogenerated charges is a relevant parameter determining the time scale for polarization and resistance switching under illumination, because photogenerated charges increase both for thicker BTO films and when using larger light power density. These results agree with the data shown in Fig. 6(c), where it is shown that the short-circuit photocurrent increases with power and sample thickness. According to earlier reports by Dimos and Warren and Land and Peercy, photogenerated charges play a role as screening polarization in perovskite materials.^7–16 

To evaluate the role of the photogenerated charges in the photoresponse, we have calculated the charge generated under illumination, using the reported absorption and efficiency of BTO films (α = 500 cm⁻¹ and η = 0.6%)³³ following the procedure described in S5 in the supplementary material. Figure 7(a) shows the calculated time-dependent accumulated charge density (Q/A) under the electrodes of area (A), using blue photons and 0.1 W/cm² power density for 70 and 4 nm films. The accumulated charge is obtained from the integration through the time of the simulated photocurrent. For the 70 nm film, the illumination time to achieve P ≈ 25 μC/cm², which is similar to the remnant polarization of bulk BTO, is about ≈ 30 s, which indicates that the time to screen the polarization of bulk BTO would be τ_screen ≈ 30 s. For the 4 nm film, τ_screen ≈ 500 s is obtained. The different τ_screen values obtained in thick and thin BTO capacitors are fully consistent with the experimental values. In Fig. 7(b), the time-dependence of the accumulated charge for different light power is shown. For 9 W/cm², τ_screen ≈ 10 s is obtained, whereas for 1.2 W/cm², τ_screen ≈ 50 s, and for 0.1 W/cm², τ_screen ≈ 700 s. The trend and order of magnitude of calculated results shown in Figs. 7(a) and 7(b) coincide with observations shown in Figs. 6(a) and 6(b). In Fig. 7(c), accumulated charge for different α and η values is shown. We can see that the modification of both parameters has an important impact on τ_screen. In particular, τ_screen decreases with efficiency and absorption increase. Calculations shown in Figs. 7(a)–7(c) indicate that the slow changes of resistance result from the low quantity of generated photocarriers due to the low absorption and conversion efficiency of the BTO layer. This result also gives the hint about obtaining faster optical switching of resistance using more absorbing/more efficient ferroelectric materials.

We have studied light-induced polarization switching and concomitant resistance changes in BTO-based capacitors where the BTO layer is thin enough (≈4 nm) to allow tunnel transport or thick enough (≈70 nm) to impose interface-limited conductance, f.i. Schottky barriers. In both cases, it has been shown that, in the presence of an imprint field, photoabsorption promotes a unidirectional resistance switch from LRS to HRS, which we have attributed to polarization reversal. Wavelength-dependent experiments indicate that the observed light-induced polarization reversal suppression is related to the photovoltaic effect, and it is operational using highly energetic photons (blue) but absent while using less energetic photons (red). Moreover, data and simulations show that a possible thermal contribution to polarization reversal is negligible. We have shown that the relatively slow time response of polarization and resistance switching (of the order of 100 s) results from the reduced photon absorption of BTO at blue (3.06 eV) as the bandgap of BTO is 3.3 eV and poor photon-carrier conversion efficiency. We have argued that light-induced photocarriers (electron–holes) are generated within the ferroelectric photoabsorber, and at some point, the density of the induced photocarriers (σ_ph) is such that the polar nature of BTO is lost (in other words, the polar displacements are suppressed by free carriers) and /or these photocarriers, rather than the electrodes contribution to screen polarization. In the first case, when light is suppressed, polarization is restored along the direction dictated by imprint and thus polarization switching may take place, as observed. In any event, the time required for polarization switching is that required to accumulate at the ferroelectric film surface, a charge equal to the polarization σ_ph = P. Naturally, the lower the absorption in the ferroelectric film, the longer the time required to reach the P value. It is not a surprise that for a thin BTO layer, which is rather transparent at visible, a substantial time is required until the compensation condition σ_ph = P, is achieved. This study shall be a guidance for future implementation and development of fast, optically controlled memristor or visual neural networks.

See the supplementary material for additional transport, PFM, and photoresponse characterization.

Financial support from the Spanish Ministry of Science and Innovation (No. 10.13039/501100011033), through the Severo Ochoa FUNFUTURE (Project No.CEX2019-000917-S funded by MCIN/AEI), Project No. TED2021-130453B-C21 funded by MCIN/AEI and European Union NextGenerationEU/PRTR, and Project Nos. PID2020-118479RB-I00, PID2020-112548RB-I00, and PID2019-107727RB-I00 funded by MCIN/AEI, and from CSIC through the i-LINK (No. LINKA20338) program are acknowledged. Project supported by a 2020 Leonardo Grant for Researchers and Cultural Creators, BBVA Foundation. Xiao Long and Huan Tan are financially supported by the China Scholarship Council (CSC) under Grant Nos. 201806100207 and 201906050014. X.L. and H.T.’s work has been done as a part of their Ph.D. program in Materials Science at Universitat Autònoma de Barcelona.

The authors have no conflicts to disclose

Xiao Long: Investigation (equal). Huan Tan: Investigation (equal). Florencio Sánchez: Investigation (equal). Ignasi Fina: Conceptualization (equal); Investigation (equal). Josep Fontcuberta: Conceptualization (equal); Investigation (equal).

The data that support the findings of this study are available from the corresponding authors upon reasonable request.