Metal halide perovskite (MHP) is a promising photovoltaic technology that has produced very high efficiencies with solution-processed methodologies (Kim et al., 2020). MHPs can be described using an ABX₃ structure, where A = monovalent cations (i.e., methyl ammonium—MA), B = divalent cations (i.e., Pb²⁺), and X = halide anions (i.e., I⁻), with MAPbI₃ (MAPI) being the most studied configuration. In addition to remarkable electrooptical semiconductor properties, MHP semiconductors show mixed ionic-electronic conduction by ionic defect displacement (Azpiroz et al., 2015; Lopez-Varo et al., 2018; Senocrate and Maier, 2019; Zhang et al., 2020). This causes intrinsic memory effects (hysteresis) in the current–voltage ( I − V ) characteristics (Almora et al., 2016; Tress et al., 2016; Rong et al., 2017; Wu et al., 2018; Alvarez et al., 2020; Bisquert et al., 2021; Bisquert and Guerrero, 2022a) at both short and long time scales, forming an effective memristor. A memristor is a two-terminal device whose resistance depends on the history of current and voltage applied to the device. Memristors enable the storage of information by metastable modification of the device conductivity, (Pershin and Di Ventra, 2011; John et al., 2018a; Rahimi Azghadi et al., 2020; Fang et al., 2021; Kang et al., 2021; Kwak et al., 2021) and are the main candidates to realize artificial synapses for neuromorphic computation algorithms (Mehonic and Kenyon, 2016; Rahimi Azghadi et al., 2020; John et al., 2021a; Bou and Bisquert, 2021; Gogoi et al., 2021; Gong et al., 2021; Kang et al., 2021; Kwak et al., 2021; Christensen et al., 2022). Arranged in a crossbar format with a differential configuration, memristors allow synaptic propagation to be realized facilely via Kirchhoff’s Current Law and Ohm’s Law, enabling efficient in-memory computing (John et al., 2021b).

While conventional memristors are activated by purely electrical stimuli, photo-stimulation of memristive responses provides new opportunities for a variety of applications such as light-gated and electro-photo-sensitive memristors for opto-neuromorphic and arithmetic computing, (Maier et al., 2016; Tan et al., 2017; John et al., 2018b; Emboras et al., 2020; John et al., 2020; Ma et al., 2021) integrated photonic neural networks, (Stark et al., 2020; Shastri et al., 2021), and preprocessing of image data in artificial vision before the transfer to the computing unit (Yang et al., 2020; Gong et al., 2021; Kim et al., 2021). Direct processing of visual information with such optoelectronic memristors may enable simpler architectures, mitigating the need for additional electro-optical converters for signal transduction and communication. MHPs are attractive materials for developing such novel electrooptical neuromorphic platforms because of their excellent optical absorption coefficient, charge transport, bandgap tunability extending across the UV, visible, and IR spectra, and high photoluminescence (PL) quantum yield, and high extinction coefficients. Combining these outstanding electronic properties with their intrinsic ionic conduction, MHPs would allow easy implementation of electrooptical synaptic elements for information storage and computing.

MHP memristors emerge from solar cells, but their behavior in the current-voltage regime is very different. In solar cells, the inverted hysteresis is a property preferably minimized (Tress et al., 2016; Yang et al., 2017; Wu et al., 2018), while for memristors, it is amplified to permanent and reversible changes of the conductivity (Berruet et al., 2022). These characteristics have not been systematically studied yet, entailing significant attention.

Here, we aim to provide a clear understanding on the multiple features of the very complex I − V curves of different devices and obtain fundamental insight about the ionic-electronic processes that cause the characteristics, in combination with insights from Impedance Spectroscopy (IS) measurements. We build different devices based on MAPI, the most simple and well-understood MHP formulation with low activation energy for ion migration that makes it suitable for memristor applications. First, we reproduce the standard solar cell (SC) configuration (FTO/c-TiO₂/m-TiO₂/MAPI/spiro-OMeTAD/Au). Two different MHP memristor devices are chosen for comparison as model systems that simplify the outer interfaces by removing the spiro-OMeTAD layer. The effect of the charge carrier selectivity of the perovskite/contact interface is studied by comparison with devices that contain PEDOT (FTO/PEDOT:PSS/MAPI/Au) and TiO₂ (FTO/c-TiO₂/mp-TiO₂/MAPI/Au). Measurements are completed in the dark and under illumination to show how the photovoltaic effects influence their operational properties. The present work establishes the link in the electrical response between solar cells and memristors. We unravel the poorly understood connection between the different regimes observed in the I − V curve and the effect of light, first of its kind to the best of our knowledge.

Since the hysteresis of the current–voltage (Almora et al., 2016; Tress et al., 2016; Rong et al., 2017; Wu et al., 2018; Alvarez et al., 2020; Bisquert et al., 2021; Bisquert and Guerrero, 2022a) is a central characteristic of the memristive devices, we summarize in Figure 2 on the basic idea of capacitive and inductive hysteresis as described recently (Bisquert et al., 2021; Bisquert and Guerrero, 2022a). The type of hysteresis and IS response is highly connected with the presence of capacitive and inductive currents. The hysteresis is denoted “regular” as observed in solar cells when the current of the photovoltaic quadrant is more positive in the forward scan, Figure 1B. Alternatively, hysteresis is “inverted” as observed often in memristors when the positive current of the forward scan is lower than the positive current of the backward scan, Figure 1D (Li et al., 2022; Shen et al., 2017). Regular hysteresis has previously been correlated with the presence of capacitive behavior in the IS response while the inverted hysteresis indicates the presence of chemical inductors in the response. The chemical inductor (Bisquert and Guerrero, 2022a) is a general denomination for a class of dynamical phenomena often based on a chemical reaction that produces a formal inductive response in impedance and transients without an underlying electromagnetic effect. The chemical inductor observed in halide perovskite solar cells is related to the slow ion-controlled electronic phenomena that occur at the interfaces of the devices. More details will be given in the following experiments.

We note that the hysteresis effects have been already observed in emerging solar cells, where ion motion is significant (Contreras et al., 2016; Elbohy et al., 2019). For example, the dye-sensitized solar cell contains a liquid electrolyte, and the strong capacitive hysteresis in I − V curves is clearly observed in Figure 2, where the effect of scan rate, temperature, and use of different solvents modify the hysteresis response. Similarly, the crossing effects in the I − V curves of Figure 3 show a transition between capacitive and inductive hysteresis that strongly depends on the mobile ionic species.

It has been reported that the resistive switching mechanism can be modified by using different top contacts (Han et al., 2019). Here, we focus on a low reactivity Au contact which is widely used in solar cells and the switching is controlled by the interface. In halide perovskite solar cells, migrating ions that reach the external contacts generate a capacitive current as the contacts compensate for their charge providing electroneutrality in the form of a capacitor (Figure 4). In MHP, this capacitance also depends on the light intensity with a rising light-induced capacitance, as previously reported (Juarez-Perez et al., 2014; Guerrero et al., 2021). It is known that halides vacancies (V_I ⁺), cations A (i.e., methyl ammonium), and metal collectors all can migrate at different rates under an external voltage bias. Halides vacancies are the fastest migrating species in MHP. When we apply a positive voltage at the FTO electrode (Figure 4A), halide vacancies will go toward gold and effectively create I⁻ excess concentration at the FTO/TiO₂ interface. The opposite is true if we apply a positive voltage at the gold contact, Figure 4B. The capacitive and inductive effects of these ions at the interfaces can be monitored by IS (Guerrero et al., 2021; Berruet et al., 2022; Taukeer Khan et al., 2022) as discussed in the latter sections.

The connection between light stimulation and production of photovoltage and photocurrent is well understood for a typical solar cell, Figure 5A. (Bisquert et al., 2004; Bisquert, 2020) Photogeneration in the semiconductor produces separated electron and hole carriers that diffuse to the metal collectors and the selectivity of contacts causes a directional photocurrent. When the voltage is increased in the forward direction, recombination is enhanced, and the photocurrent decreases until it vanishes at the open-circuit voltage ( V o c ) point. In the reverse direction, there is a depletion of carriers which leaves only a small current, hence the solar cell shows a diode or rectification property. (Bisquert, 2020) However, a strong reverse bias may produce a breakdown effect with raising current in the opposite direction (Bowring et al., 2018).

Figure 6 shows the device configuration of the SC device and the electro-optical response. Our PV performance of ∼15% is well aligned with reported values for the chosen formulation (Figure 6A). Hysteresis is typically observed in the current density-voltage ( j − V ) curve for this configuration with separation of the forward and backward scans even for measurements conducted in dark (Figure 6B). The hysteresis behavior is complex with marked differences as a function of the scan rates, presence of crossing points of the curves at about +0.6 V, and change in the sign of the current far away from 0 V (spikes in the semilog curves) (Li et al., 2022).

Figure 7 shows photovoltaic performance of the solar cell device under a monochromatic blue light at different illumination intensities, with negative current for photovoltaic effect. As expected, the photogenerated current is enhanced with light intensity thanks to the excellent carrier generation and extraction. Very interestingly, in the complete SC there is a crossing in the forward and reverse scans at V o c conditions with a change in tendency from regular ( V < V o c ) to inverted ( V > V o c ) hysteresis. In addition, the capacitive hysteresis also increases with illumination level in terms of the separation of the forward and reverse scan on the y-axis (Figure 7B). The capacitance takes values in the order of 0.2–0.9 mF/cm².

Figure 8 shows the j − V characteristic of the solar cell device at 50 mW/cm² blue light illumination. Below V o c , the capacitive behavior shows regular hysteresis and above V o c the inductive behavior is manifested in the j − V as inverted hysteresis as in Figure 7A. Figures 8B,C confirm this statement, showing two capacitive arcs in the complex plane plot. Only at high dc voltages, beyond V o c , it is possible to observe an inductive behavior (Bisquert and Guerrero, 2022a; Berruet et al., 2022), with low frequencies crossing to the fourth quadrant of the complex plane (Figure 8B inset). We note that the inductive loop is not well developed as in the next experiments clearly indicating that spiro-OMeTAD is affecting the impedance response.

The model systems which do not contain spiro-OMeTAD have been fully characterized with two types of selective contacts, PEDOT:PSS (selective to holes), and TiO₂ (selective to electrons). The dark I − V characteristics of the two memristor devices at different voltage sweep rates are shown in Figure 9. Memristor performance of the device containing Spiro-OMETAD is well aligned with previously reported results (Berruet et al., 2022). Both show inverted hysteresis loops when the voltage is positive at the Au contact similar to the SC configuration above ∼0.5 V in Figure 7A. The onset voltage for transition to the higher conductance state is larger for higher sweep rates, as explained in a recent work (Berruet et al., 2022) that will be summarized in the Model section. The loop is steeper for the PEDOT memristor, due to the fact that the compact/mesoporous TiO₂ is a better selective contact for electrons than PEDOT, which reduces the hysteresis effect in the TiO₂ memristor case. We note that some of the most recent reported memristors show performances higher than those reported here. However, top performance devices contain buffer layers (i.e., PMMA) at both perovskite/contact interfaces and reactive contacts (i.e., Ag) that complicate the device configuration and interpretation of operational mechanism. Here, we work with a simplified configuration to provide an understanding of the operational mechanism of memristors and devices that have been fabricated without a buffer layer in the perovskite/top contact interface.

It is also important to highlight that the response in the dark is only inductive at positive voltages, there is no crossing for both model systems as compared to the SC for measurements in the dark. This result clearly points to the capacitive currents due to the migrating I⁻ ions with Spiro-OMeTAD in the dark of the SC configuration. Otherwise, only PEDOT memristors show inverted hysteresis when the voltage is negative. For TiO₂ memristors, the capacitive hysteresis is found in this region, as shown in solar cell devices in Figure 7B. This difference in PEDOT memristor performance is due to irreversible reactions taking place between ions and TiO₂ at this interface. For example, for the interface TiO₂/MAPI it has been reported that Ti−I−Pb bonds may form, as identified by Raman Spectroscopy, which can easily accommodate excess ionic charge when a negative bias is applied (Carrillo et al., 2016). This configuration could lead to neutral interfaces that would be in agreement with the low capacitance measured in the current work for the memristor containing TiO₂. Comparing these characterizations with the one previously discussed in Figure 6B, it is possible to affirm that the TiO₂ device shows closer performance to an MHP solar cell than the PEDOT device in dark conditions.

To investigate the effect of light on the memristor devices, we performed the current-voltage characterization at different illumination levels and the results are shown in Figure 10. In principle, the looping curves are like those in the dark in Figure 9. But there are also significant differences. The PEDOT device shows a symmetric opening of the curves V = 0 for increasing light intensity, Figure 10B. This is the effect of the light-induced capacitance, previously mentioned, that is well known in halide perovskites (Juarez-Perez et al., 2014). Indeed, the measured capacitance increases with light intensity as indicated in Figure 10C. For the TiO₂ device, the photoinduced capacitive current is also observed, but it is clearly displaced to negative values of the current since the TiO₂ is selective to electrons and it is able to extract the photogenerated current. In consequence, the TiO₂ device shows diode-like characteristics and a positive photovoltage at all light intensity values. The difference between the two devices is explained by the mechanistic diagrams in Figure 5. The PEDOT device has a negligible effect on contact selectivity, hence capacitive current prevails in the photocurrent. For TiO₂ devices, the metal oxide contact acts as a selective contact that extracts preferentially electrons, hence the photogeneration gives a dominant-negative photocurrent for both signs of the voltage sweep direction. Thus, this performance is similar to the one developed by a solar cell, shown in Figure 7A.

For further analysis, we focus on a particular illumination, Figure 11. For PEDOT devices, it is observed that there are two loops on right and left of Figure 11A, where the current in forward and backward makes crossing pathways. As mentioned earlier in Figure 3 this property indicates a change from regular to inverted hysteresis caused by the onset of dominant inductive property in the impedance spectra (Bisquert and Guerrero, 2022a; Berruet et al., 2022). The presence of the chemical inductor (the loop in the fourth quadrant of the complex plane) is in fact obtained in the measurements of impedance spectroscopy in Figures 11B,C. While this change from capacitive to inductive behavior is shown in Figure 11E at high positive voltages, this is not obtained when negative voltages are studied in TiO₂ memristors (Figure 9B), corresponding with the loops in Figure 8D. This trend is similar to the one shown by the solar cell device, Figures 8B,C, with inductive properties significant only at high dc voltages.

In order to explain the experimental observations in a unified framework, we present a dynamical model (Berruet et al., 2022) that includes several capacitive mechanisms and a chemical inductor feature. It is formed by the system of equations I t o t = I p h + i c 0 f + u R b + + C m d u d t + Q m d f d t , (1) τ k d f d t = ( 1 − f ) − e − u − V T V m f . (2)

The model has three independent variables: I t o t , u , f . The first two are the external current and voltage and f is an occupation function ( 0 ≤ f ≤ 1 ). The variable f indicates the state of the mechanism that controls the large injection current in the activated state, with the onset potential V T . V m is an ideality factor with a dimension of voltage. i c 0 is the saturation value of the injection current. Our model provides an alternative pathway to those approaches based on drift-diffusion simulations. (Li et al. 2022; Shen et al., 2017) This simple two-dimensional model already describes simultaneously the current-voltage curves and impedance characteristics. Furthermore, the model enables full analysis of time transient techniques. (Bisquert and Guerrero, 2022b).

Eq. 1 describes the components of I t o t : the photocurrent I p h , the injection current i c 0 f , a small ohmic current of constant resistance R b , a capacitive charging of the interfaces with capacitance C m , and a voltage-dependent capacitance correlated to f with a maximum surface charging Q m .

If the time derivative in Eq. 2 is suppressed, we obtain the following steady-state solution: f s s ( u ) = 1 1 + e − ( u − V T ) / V m . (3)

Eq. 2 is a delay equation that states that the formation of the conduction state at a certain voltage, f s s ( u ) , lags behind the changes of applied voltage with a characteristic time τ k . This slowing down of the response gives rise to the chemical inductor in the equivalent circuit.

The steady-state current is I a p p = I p h + u a p p R b + i c 0 f s s ( u ) . (4)

In the solar cell a saturation regime f s s ≈ 1 may not be observed. Then the current in the injection regime at large voltage has the exponential form I a p p ≈ I p h + i r e c 0 e u app / V m , (5) where i r e c 0 is a recombination current parameter. The impedance model is obtained by small perturbation of Eqs. (1, 2). It has the expression (Berruet et al., 2022). Z ( s ) = [ C m s + R b − 1 + 1 R a + L a k s + 1 R 2 + 1 C 2 s ] − 1 , (6) R a − 1 = d i c d u = i c 0 V m f s s ( 1 − f s s ) , (7) L a d = τ d R a , (8) R 2 = L a k i c 0 Q m , (9) C 2 = Q m R a i c 0 . (10)

The equivalent circuit is shown in Figure 12.

Based on this model we can form several regimes of responses observed in the experimental evidence of the preceding Section.

In Figure 13 we show the effect of a constant capacitor C m that opens the steady-state I − V in capacitive hysteresis as in Figures 1, 7B. Figure 14 shows the effect of voltage-activated capacitor C 2 that becomes active at V o c . It causes capacitive spikes close to V o c as in Figures 2, 3A.

Figure 15A shows the inductive hysteresis loop that is observed at a large injection current in the measuring devices. To provide a more complete description of the inductive hysteresis we extend the model as follows: I t o t = I p h + u R b + i c + + C m d u d t + Q m d f d t , (11) τ d d i c d t = i c 0 f − i c , (12) τ k d f d t = e α u − V T V m ( 1 − f ) − e ( α − 1 ) u − V T V m f . (13)

This more general three-dimensional model has four independent variables: I t o t ,   u , f ,   i c . Here the injection current i c is a variable in addition f . In steady-state, they satisfy the relationship i c = i c 0 f s s . (14)

As described before (Berruet et al., 2022), Eq. 12 represents a diffusion or migration time of ions that introduces a delay i c with respect to the external perturbation, by the characteristic time constant τ d . This temporal parameter indicates the time necessary to establish the configuration of high f that produces the large electronic current i c 0 . Eq. 13 is an extension of (2) with a Tafel coefficient α . It is observed in Figure 15C that the expanded model provides different shapes in which the go and return currents make parallel tracks.

Finally, in Figure 16 the capacitive hysteresis is combined with an inductive hysteresis at high voltage. There is a crossing point from capacitive to inductive hysteresis in Figure 16, as observed in Figure 7A. In the impedance spectra, the low-frequency capacitive arc is transformed into an inductive feature in the fourth quadrant of the complex plane.

The previous analysis shows that a voltage-activated current with capacitive and inductive kinetic properties provides a satisfactory explanation of the current-voltage hysteresis and impedance properties of the different devices studied. We obtained a good description of the experimental results. Models with very few variables describe a large span of observed phenomena. The question is to give a definitive interpretation to the transformation equations in terms of physical mechanisms. One recent example is the combination of bidirectional photocurrent and hysteresis that has been discussed by modification of the contact layer. (Li et al. 2022).

We considered three types of devices: a solar cell with excellent selective contacts, a PEDOT memristor with both contacts selective to holes, and a hybrid TiO₂ device with partially selective contacts. In Figure 5 we showed elementary models to explain the mechanism under light conditions close to zero applied voltage that leads to the generation of a photocurrent. In the experimental figures, we have observed the transition to a large current when the Au electrode is positively biased. This large current and the associated memory effect is the key operational characteristic of a memristor. In addition, we showed that the presence of regular/capacitive hysteresis under illumination is connected with the surface ionic polarization by the high capacitance measured by IS and the large capacitive current that scales with the light intensity.

Alternatively, for measurements in the dark, large inductive currents have been observed in all types of devices measured. We can discuss the origin of the large current rise in the dark injection regime based on Figure 17.

In the case of a solar cell with very good selective contacts, the large current is due to recombination as shown in Figure 17A. This current regime is usually not analyzed in solar cells as the regime beyond V o c is uninteresting compared to the extraction regime below V o c . However, we have shown in the previous sections that the exponential current is well described by Eq. 5 as a recombination current that may correspond to electron-hole recombination via trap states as often reported for Light Emitting Diodes. Numerous analyses have discussed the effective ideality factor in relation to the recombination mechanisms. (Wetzelaer et al., 2015; Tress et al., 2018; Caprioglio et al., 2019).

For the operation of memristors, the injection domain is an important feature of the device operation. It is therefore essential to obtain a physical description of the variable f that provides the exponential rise of the injection current at a certain voltage. Based on the model discussion, we have several options:

1) The memristor device injects electrons and holes by partially selective contacts as in Figure 17A, producing a recombination current and f representing the ion-activated recombination sites as suggested in solar cells (Pockett and Carnie, 2017; Shen et al., 2017).

But another possibility is that there is no contact selectivity and the current is due to just one carrier going through the sample as in Figure 17B, i.e., electron-only devices. (Li et al., 2022) In the current work both PEDOT:PSS and Au are contacted more selective to holes than electrons due to their energy levels and have often been used for this purpose in solar cells and LEDs research. Therefore, it is likely that if the second possibility is responsible for the inductive currents and the device would be hole-only. The question is to determine the physical origin of the activation function f in this case. The activation mechanisms must predict an exponential rise and a saturation effect. There are several options outlined in Figure 18:

2) The activation of surface conduction sites by reaction of the incoming ions, Figure 18A. Here is the interpretation of the function f in Eqs. 1–2 is direct: It is the fraction of surface-activated sites. The chemistry of the perovskite/contact perovskite is very complex and it is possible that the presence of halides at the interface and oxidizing/reducing conditions may lead to the formation/rupture of electrically conductive sites. A representative example is the case of the interface MAPI/Au where gold halides (i.e., AuI₃) may form reversibly during polarization at positive bias and create an interface with reduced resistance to extract the charge (Pospisil et al., 2019). A negative bias, metallic Au would be recovered with increased charge transfer resistances. A similar, effect has been reported for devices containing MAPI/Ag (Solanki et al., 2020).

The present experimental results and modeling do not distinguish between these options, or between models in Figure 17 and Figure 18, and additional discriminatory experimental work is needed.

The operation of solar cells and memristors depends crucially on the structure of contacts. We found different phenomena like directional photocurrent, photocapacitance at low bias voltage, and alternation of capacitive and inductive hysteresis domains can be explained by a model in which the rising current is activated by voltage. In the case of a solar cell, this current is well explained by the ionic-dependent recombination model. But for the memristor, different options appear since the devices are less affected by photogeneration and may become single carrier dominated. We conclude that solar cells and memristors can have very different underlying mechanisms, but more investigation is needed for the attribution of the rising current and the memory effect.