Introduction

Beneficial from the switchable spontaneous polarization with external stimuli, ferroelectrics are holding great promise for numerous devices, i.e., non-volatile memory1,2,3, logic device4,5,6,7, actuators8,9,10, and ferroelectric photovoltaic (FePV) applications11,12,13,14,15,16. The inherent nature of ultrafast switching, ultralow power consumption, and ultrahigh retention makes ferroelectrics highly potential for developing energy-efficient neuromorphic computing, overcoming the Von Neumann bottleneck. To meet the unprecedented device requirements of increased storage density, mimicking complex neuromorphic dynamics, as well as overcoming the intrinsic bi-stability, diverse switching kinetics have been designed based on various material phases and structures, enabling a plethora of ferroelectric devices17,18,19,20,21. Taking advantage of polarization switching kinetics has provided a promising solution for designing novel multifunctional ferroelectric devices.

Based on the theory of nucleation and ___domain growth, polarization switching kinetics is typically characterized by a waiting time (t0) of nucleation and/or ___domain coalescence. When ___domain coalescence dominates the polarization reversal process, polarization switching follows ___domain-wall-motion mechanism (Fig. 1a, e and Supplementary Note 1), and is well described by the Kolmogorov–Avrami–Ishibashi (KAI) model22. While nucleation dominates the polarization reversal process (Fig. 1b, f), the nucleation-limited-switching (NLS) model is responsible for switching kinetics23. These statistic characteristics of ___domain nucleation and coalescence gives rise to various switching processes, and thus enable different device applications. For example, the switching time for ___domain coalescence in a single crystal is nearly constant, and thus a binary digital switching that the adopted pulse width is either insufficient to reverse the polarization or deterministic for switching completely, is easy to trigger, facilitating the non-volatile memory. While the statistics of nucleation in polycrystalline ferroelectrics is a Lorentz distribution function, providing an efficient way to achieve multiple polarization states that are necessary for synapse mimicking. In addition, designing a multi-peak distribution function, such as in ferroelectric superlattices, even makes controlled layer-by-layer switching of individual ferroelectric layers feasible (Fig. 1c, g), paving way for realizing multi-level non-volatile memory24. To date, although various switching kinetics have been designed for constructing different functional ferroelectric devices, the device with specific ferroelectric materials and architecture generally owns a single switching kinetics (Fig. 1a–c). Alternatively, achieving configurable kinetics control of polarization switching in a pre-defined material or device holds great potential for enriching ferroelectric device applications, but remains a significant challenge, as the switching process is usually initiated at crystal boundaries or defects, closely related to the crystal phase/structure.

Fig. 1: Progress of polarization switching in different ferroelectrics.
figure 1

a Diagram of polarization switching in a single crystal, where ___domain coalescence dominates the switching process and is described by the KAI model. b Diagram of polarization switching in polycrystalline ferroelectrics, where ___domain nucleation dominates the switching process and follows the NLS model. c Diagram of polarization switching in ferroelectric superlattice, where individual ferroelectric layers can be switched layer-by-layer. d Diagram of anomalous polarization switching in CIPS. While Cu ions undergo intralayer hopping and interlayer hopping progress during the migration, polarization alignment with and against the applied electric field occurs sequentially. e Time-dependent changing polarization of KAI model, with a constant of switching time t0. f Time evolution of changing polarization based on the NLS model, where switching time is characterized by the Lorentzian function. g Time-dependent changing polarization in ferroelectric superlattices, where a quadruple-peak function determined by four individual ferroelectric layers is responsible for the distribution of switching time. h Time-dependent changing polarization in CIPS, where intralayer and interlayer switching associated double-peak function is responsible for the distribution of switching time.

Full size image

Van der Waals (vdW) layered ferroionic CuInP2S6 (CIPS) owns the unique coupling effect of ferroelectricity and ionic conductivity25,26,27,28,29,30,31,32. Importantly, polarization switching and ionic migration cross the vdW gaps are both derived from Cu ions hopping progress33,34,35. Notably, the interlayer crossing of Cu ions even enables polarization aligns against the applied electric field (Fig. 1d, h). These extraordinary physical characteristics make CIPS a potential platform for exploring rich polarization switching processes and developing configurable multifunctional ferroelectric devices. However, the huge leakage current, resulting from ion current with Cu ions long-term migration and electron current with electrons drift or tunneling, hinders the switching kinetics exploring in CIPS thin films2,36,37,38,39. In this work, making use of polarization-determined FePV behaviors, and systematically investigating polarization switching kinetics of CIPS, we discovered three distinct switching modes depending on the Cu ions migration dynamics in ferroionic CIPS. Through precisely tuning the ion dynamics via the amplitude and width of poling voltage (Vp), we achieved configurable polarization switching, including digital switching, gradual switching, and accumulative switching in a single device. This work unlocks new possibilities for innovative ferroelectric device types based on the ferroionic material system.

Results

Anomalous ferroelectric photovoltaic switching in CIPS

For characterizing the polarization switching kinetics in CIPS thin films, the conventional positive-up-negative-down (PUND) method is invalid40, as the polarization-switching current is obscured by leakage current2,36,37,38,39. To avoid the effect of ion migration and electron tunneling, FePV measurement is adopted to characterize the polarization states (see Methods for details). When polarized CIPS is upon the laser illumination, a polarization state determined short current (Isc) is generated16,41. FePV Isc actually provides an effective way to quantify the remanent polarization16 (discussed in Supplementary Notes 2 and 3). For this reason, few-layered graphene is utilized as the transparent electrodes, and FePV measurement of a sandwiched graphene/CIPS/graphene device is performed. The device fabrication is described in the methods, and CIPS crystal-growth and the materials characterizations are shown in Supplementary Fig. 1 and Supplementary Fig. 2. During the measurement, both poling and reading voltage are applied to the bottom graphene electrode (Supplementary Fig. 3), and the sweeping range of read voltage is limited to 0.1 V to avoid disturbing the polarization. A positive pulse of 3 V (Vp = 3 V) or a negative pulse of −3 V (Vp = −3 V) is carried out to poling the device. The pulse width is 2 ms, a typical duration required for polarization switching in CIPS30,42. After that, the current-voltage (I-V) curves of the device under illumination are measured. We can see from the results in Fig. 2a that Isc is negative at 3 V Vp but positive at −3 V Vp. Moreover, as Vp decreases from 3 to −3 V and then increases back to 3 V with step of 0.6 V, the monitored photovoltaic current changes from negative to positive one at about −2 V Vp and then changes back to negative one at about 2 V Vp (Fig. 2b). The extracted Isc as function of Vp (Isc-Vp) is depicted in Fig. 2c, showing a typical ferroelectric switching associated clockwise hysteresis behavior. The amplitude of coercive voltages (Vc), related to switching from upward polarization (Pup) to downward polarization (Pdown) or switching from Pdown to Pup, are both about 2 V.

Fig. 2: Anomalous FePV switching behavior in CIPS.
figure 2

a I-V characteristic of device under illumination after applying 3 V (−3 V) Vp, plotted by the red dots (blue dots). The black dots of I-V are measured in the dark condition. b Changing behaviors of photovoltaic current as Vp scans from 3 to −3 V and back to 3 V, with magnitude illustrated by a color map. c Extracted Isc as a function of Vp from (b). d I-V characteristic of device under illumination after applying 11.4 V (−11.4 V) Vp, plotted by the red dots (blue dots). e Changing behaviors of photovoltaic current as Vp scans from 11.4 to −11.4 V and back to 11.4 V with a step of 0.6 V. f Extracted Isc as a function of Vp from (e). g Isc versus time of several typical polarization states in (f). h Retention characteristics of Pdown (set by 11.4 V Vp, red line) and Pup (set by −11.4 V Vp, blue line) states. i Cycle-to-cycle measurement of Isc at Pdown and Pup states in (h).

Full size image

Surprisingly, while Vp increases from 3 to 11.4 V, in contrast to the result in Fig. 2a, a completely reversed I-V characteristic is observed. Isc is positive at 11.4 V Vp but becomes negative at −11.4 V Vp (Fig. 2d). Given that Isc is opposite to the direction of polarization dipoles, the results in Fig. 2d indicate Pdown and Pup are even achieved through applying a positive and a negative Vp, respectively. To further characterize such anomalous FePV behavior, a cycling scanning Vp ranged from 11.4 to -11.4 V with a step of 0.6 V is performed. It is found that the phase of Isc even switches six times during Vp cycling scanning, as shown in Fig. 2e. The corresponding open-circuit voltage (Voc) as a function of Vp also shows a 6-times switching behavior (Supplementary Fig. 4). The extracted Isc-Vp in Fig. 2f shows an interesting feature of three clear hysteresis windows. As Vp backward decreases from 11.4 to −11.4 V and then forward increases from −11.4 to 11.4 V, Isc sequentially changes from state i to state iv (i→ii→iii→iv) first, and following changes from iv back to i (iv→v→vi→i). Here, featuring Isc from state ii to state iii with Vp backward scanning and Isc from state v to state vi with Vp forward scanning, forms a hysteresis window. Such hysteresis window can be well tuned by changing the pulse width of Vp (Supplementary Fig. 5). It is found that the difference of Isc between state v and state ii decreases and even near disappears as pulse width increases to 40 ms, but the critical voltage at which Isc sharp changes almost keep unchanged, with amplitude of ~2 V. The value is similar with the Vc in Fig. 2c, indicating the intrinsic ferroelectric switching should be responsible for the hysteresis window near 0 V Vp.

To further confirm the photoelectric response of the device, the time-dependent Isc was also monitored, providing an effective method to demonstrate the retention and stability of polarization. Isc versus time of several typical states in Fig. 2f are illustrated in Fig. 2g. The results indicate that the polarization remains almost unchanged while applying laser illumination. Furthermore, Fig. 2h characterizes the retention of CIPS FePV, where the polarization was set as Pdown and Pup by 11.4 and −11.4 V Vp, respectively. Isc remains well more than 1000 s, indicating a great potential for achieving long optoelectronic storage. Through turning on and off the laser repeatedly, the cycle-to-cycle measurement of Isc was performed. The result in Fig. 2i demonstrates that such anomalous Isc is fairly robust, with a negligible change over 250 cycles. The detailed mechanism will be discussed in the following section.

Mechanism of anomalous FePV behavior

To further confirm the FePV characteristics of CIPS, ten-times repetitive Vp with various amplitudes have been employed to poling the device. As shown in Fig. 3a, the device was first reset by 11.4 V Vp with a width of 2 ms (11.4 V, 2 ms) to obtain Pdown state. It is found that for 3 and −11.4 V Vp, once the first voltage pulse is applied, Isc sharply decreases from 2 nA to one less than −2 nA, indicating the polarization is being switched from Pdown to Pup. Meanwhile, Pdown keeps almost unchanged while -3 V Vp is applied repetitively. Thus, it demonstrates that two Ec with opposite polarity, including a polarization alignment parallel Ec1 with small amplitude and a polarization alignment anti-parallel Ec2 with large amplitude, are responsible for the switching from Pdown to Pup. Similarly, two Ec are also observed as polarization is switched from Pup to Pdown (Fig. 3b). These two coercive fields associated polarization switching, are also identified by the switching spectroscopy piezoresponse forece microscopy (SS-PFM) and PFM mapping measurement (Supplementary Fig. 6).

Fig. 3: Switching mechanism of the anomalous FePV behavior in CIPS.
figure 3

a Response of Isc while applying repetitive Vp with decreasing from 9 to −11.4 V. A Vp (11.4 V, 2 ms) is used as reset voltage to obtain an initial Pdown state, denoted by red arrow. b Response of Isc as the repetitive Vp increases from −9.6 to 11.4 V, where a Vp (−11.4 V, 2 ms) is adopted to reset the device, forming another initial Pup state, denoted by blue arrow. c Evolution of Cu ions distribution while applying cyclic scanning Vp, where the focused Cu ions are denoted by the black dashed box to present ion migration paths. Cu ions in LP and HP states are depicted by the blue and red spheres, respectively, and downward polarization (−LP/−HP) and upward polarization (+LP/+HP) are uniformly named Pdown and Pup, illustrated in the inset.

Full size image

Theoretical and experimental studies demonstrate that due to the Cu ions migration, Cu ions hopping in CIPS happens not only within the single layer but also across the vdW gaps, namely intralayer hopping and interlayer hopping, respectively. Combining with the two hopping progresses, Cu ions long migration happens. Such long migration and accumulation of Cu ions on one side of electrode, even induce a photovoltaic Isc (Supplementary Fig. 7), but the polarity of Isc is opposite to that of case at large Vp in Fig. 3a, b. By contrast, as Cu ions hop across the vdW gap merely, polarization aligns against the direction of the applied electric field32, leading to a depolarization electric field whose direction is in agreement with the polarity of Isc observed in Fig. 2d. Thus, the individual interlayer hopping rather than Cu ions long migration is more desirable for the anomalous FePV Isc (Supplementary Fig. 8). Although the energy barrier for hopping across the vdW gap is much higher than hopping within the single layer, nearly half that could be reduced through sequential switching35. Indeed, the intermediate states beneficial to sequential switching are widely observed in our CIPS device (Supplementary Fig. 9). Besides, a plentiful supply of Cu vacancies and interstitials can also effectively facilitate such interlayer hopping33,35. Therefore, it is reasonable to infer that the two coercive fields, Ec involved switching characteristics origin from intralayer and interlayer switching progress, respectively. That is, the polarization alignment parallel to Ec1 is responsible for intralayer switching and the polarization alignment anti-parallel to Ec2 is responsible for interlayer switching, respectively.

Based on the above analysis, the physical scenario of anomalous FePV behaviors can be illustrated in Fig. 3c. In our device, the obtained Pdown and Pup states actually contain low polarization (LP) and high polarization (HP)28, as supported by PFM mapping measurement in Supplementary Fig. 9. Here, for simplicity, the downward polarization (−LP/−HP) and upward polarization (+LP/+HP) are uniformly named Pdown and Pup, respectively. To conveniently display the Cu ions' intralayer and interlayer hopping progress, we focus on the initial Cu ions distribution in layer-1, denoted by the black dashed box, while applying cyclic scanning Vp. After applying a stimulus of 11.4 V Vp, the initial polarization state is set as Pdown, with most of Cu ions accumulate at the bottom of layer-1. When a positive Vp with an electric field amplitude larger than Ec1, is applied subsequently, these Cu ions hop from the bottom to the top of layer-1. These Cu ions keep located at top sites until a negative Vp with electric filed amplitude larger than Ec1 is applied, at which these Cu ions hop back to the bottom of layer-1 (Fig. 3c). The almost same Isc at state (i) and state (iii) confirms that the corresponding polarization states of them are similar. Next, as a more negative Vp with electric field amplitude larger than Ec2 is applied, these Cu ions at the bottom of layer-1 hop even across the vdW gap to the top of layer-2. At that time, polarization is unusually switched from Pdown to Pup, against the applied electric field. After that, applying forward scanning Vp, similar with the progress from state (i) to state (iv), Cu ions sequentially undergoes hopping from the top to bottom of layer-2 (iv→v), hopping from the bottom to top of layer-2 (v→vi), and hopping from the top of layer-2 across to the bottom of layer-1 (vi→i). Thus, the hysteresis window near zero Vp in Fig. 2f attributes to the intralayer switching. Importantly, during the progress from state i to state ii (from state iv to state v), the intralayer switching dominates, in contrast to the progress from state vi to state i (from state iii to state iv), where interlayer switching dominates. As a result, a small Vp is required for the former while a large Vp is required for the latter, forming two hysteresis windows. Such a typical hysteresis loop with six-times switching behaviors, is also observed in the SS-PFM measurement (Supplementary Fig. 6), further providing cross validation that polarization switching between upward and downward directions alternatively occurs six times while applying cyclic scanning voltage. So far, the anomalous FePV behaviors root in distinguishable Cu ions intralayer and interlayer hopping progress, indicating it is possible for characterizing the intralayer and interlayer switching kinetics.

Configurable polarization switching in CIPS

To further shed light on the switching kinetics of Cu ions intralayer and interlayer hopping progress, time-dependent Isc measurement under various Vp were performed (Fig. 4). The device is initially set by 11.4 V, time evolution of Isc under different Vp presents a tendency that Isc gradually changes from positive to negative and then changes back to positive value (Fig. 4a). As Vp increases, the changing process becomes faster. Such a time-dependent twice switching process is further depicted in Fig. 4b. It indicates the ferroelectric polarization is switched from initial Pdown to Pup for a short time, and gradually changes back to Pdown for a long time. In other words, for a short time, the intralayer switching that from initial Pdown to Pup dominates, but for a long time, interlayer switching gradually emerges and induces polarization changing back to Pdown. The corresponding modes of polarization switching are named as Type I and Type II, respectively. Given that the dipole in CIPS owns independent switching progress, the NLS model is more appropriate for polarization switching kinetics23 (see Supplementary Notes 2,3). The extracted Lorentzian distribution functions at 4 V Vp is displayed in Fig. 4c. It is founded that the distribution function presents unique double-peak features, and the extracted switching time t1 is one eighth of t2. The evolution of log t1 and log t2 as a function of the inverse electric field follows the characteristic Merz’s law20,43 (Supplementary Fig. 10). Actually, during Type II switching progress, intralayer and interlayer switching are both involved. To reveal the individual interlayer switching progress, a reset voltage of −11.4 V is adopted to obtain the initial Pup. One can see from Fig. 4d, e that Isc under different Vp monotonically increase from negative to positive value, with ferroelectric polarization is being switched from the initial Pup to Pdown (named Type III switching mode). At 4 V Vp, the extracted switching time t3 is about three times larger than t2, as plotted in Fig. 4f. The magnitudes of these extracted switching time value are consistent with the polarization switching in bulk CIPS29 (Supplementary Figs. 10,  11). In addition, the extracted activation field and lower limit of switching time of these three switching modes, indicates interlayer switching should own the biggest energy barrier among them, which is further confirmed by the temperature-dependent FePV behaviors (Supplementary Fig. 12).

Fig. 4: Polarization-switching kinetics in CIPS.
figure 4

a Measured Isc as a function of pulse width at various Vp, where a Vp (11.4 V, 2 ms) is used as reset voltage to obtain an initial Pdown state. The solid lines are the fitting curves. b Mapping plot of Isc derived from (a), highlighting two distinct switching process that intralayer switching for a short time and intralayer-interlayer coupling switching for a long time. c Extracted Lorentzian distribution from (a) at 4 V Vp. d Measured Isc versus pulse width at different Vp, with an initial Pup state through applying reset voltage Vp (−11.4 V, 2 ms). e Mapping plot of Isc derived from (d). f Extracted Lorentzian distribution from (d) at 4 V Vp. g Diagram of Cu ions intralayer and interlayer hopping progress, of which dominates the switching process, resulting in the corresponding switching mode. h Time-dependent switching model of Type I, Type II, and Type III. The focused Cu ions on the bottom layer are used to describe their migration pathway in different switching modes.

Full size image

Due to the obtained polarization state is determined by the final static distribution of Cu ions, polarization reversal in CIPS should be determined by which of intralayer and interlayer hopping of Cu ions dominates, as depicted in Fig. 4g, h. That is to say, the switching kinetics of CIPS strongly depend on the distribution of Cu ions. For example, at an initial Pup condition, most of the Cu ions are located at the top of a single layer, and thus, the intralayer hopping does not occur due to the absence of bottom Cu ions. As pulse width increases, the interlayer hopping gradually emerges and dominates polarization reversal progress, forming Type III switching. By contrast, at an initial Pdown condition, most of the Cu ions are located near the bottom of a single layer. For a Vp with short width, the intralayer hopping happens first and dominates the polarization reversal progress, resulting in Type I switching. While extending the duration of Vp, Cu ions hop from the bottom to top of this single layer and a part of these top Cu ions further hop across the vdW gap towards upper layer simultaneously, resulting in the intralayer and interlayer coupled Type II switching. Under this switching mode, the coupling between the intralayer and interlayer, depending on the Cu ions occupation distribution, can be feasibly tuned by the Vp amplitude and width, resulting in various polarization states (Fig. 4h), beneficial for ferroelectric multifunctional design.

Based on the understanding of configurable ferroelectric switching kinetics of CIPS, the polarization switching phase diagram depending on the Vp amplitude and width is theoretically simulated as shown in Fig. 5a (see Supplementary Note 2). Through configuring the Vp pulse amplitude and width, three modes of intralayer switching (Type I), intralayer-interlayer coupling switching (Type II), and interlayer switching (Type III) are available for designing ferroelectric devices. Utilizing Type I switching mode by applying 3 and −3 V Vp with a width of 2 ms, a typical digital memory with sharp switching between two distinct polarization states of Pup and Pdown is achieved (Fig. 5b). Here, the negative Isc and positive Isc are used as the logic state “0” and logic state “1”, respectively. Importantly, departs from conventional ferroelectric digital memory, such Pup and Pdown involved binary digital memory even can be tuned through applying unidirectional Vp as depicted in Fig. 5c, d. As the polarization is switched to Pup states by a setting voltage Vp (3 V, 2 ms), a Vp with a longer width (3 V, 30 ms) or a Vp with a larger amplitude (10 V, 2 ms) is adopted as reset voltage to force Pup to switch back to Pdown states. In this way, the switching behaviors that from Pup to Pdown driven by Type III switching or from Pdown to Pup driven by Type I switching alternately happens. Notably, the endurance of the anomalous interlayer switching in Type III is over 104 cycles (Supplementary Fig. 13), facilitating the logical application.

Fig. 5: Device operation under configurable kinetics of polarization switching.
figure 5

a Simulated phase diagram of time and field-dependent polarization switching, where intralayer switching (Type I), intralayer-interlayer coupling switching (Type II), and interlayer switching (Type III) can be achieved by configuring Vp amplitude and width. b Typical digital memory of two distinct polarization states based on Type I switching model. c Digital memory behavior achieved by applying 3 V Vp with different durations. d Digital memory behavior achieved by unidirectional Vp with different amplitudes. e, f Gradual switching based on Type II switching mode through tuning the pulse width (e) and amplitude (f), respectively. g, h Accumulative switching achieved by designing pulse width (g) and amplitude (h), where Cu ions are gradually accumulated, driven by Type I switching, and subsequently released, driven by Type III switching.

Full size image

The unique intralayer-interlayer coupling switching mechanism provides an effective way to precisely tune the distribution of Cu ions at the top and bottom layers, facilitating the achievement of multiple intermediate polarization states. As a series of Vp with increasing width from 1.6 to 2.1 ms is applied to the device, interlayer hopping progress is continuously enhanced while intralayer hopping progress is continuously reduced, making the total polarization states downward gradually. By contrast, through applying a series of Vp with decreasing width from 1.3 ms to 0.8 ms, interlayer hopping progress is gradually suppressed while intralayer hopping progress is increasingly enhanced, forming an upward polarization state (Fig. 5e). Similar behaviors are also achieved by applying a series of Vp with changing amplitude (Fig. 5f). Such gradually switching behavior is promising for implementing long-term potentiation (LTP) and long-term depression (LTD) of synaptic device.

In addition, configuring Vp with a larger range of increasing width (or a larger range of increasing amplitude), Cu ions could be tuned to continuously accumulate to top of layers through Type I switching and then release to upper layer through Type III switching, achieving an accumulative switching behavior (Fig. 5g, h). One can see that when the width of Vp increases progressively from 0.06 to 24 ms, Isc gradually decreases from 3.5 to −3 nA within ten pulses, then increases back to 3.5 nA. This accumulation and releasing progress involved switching behaviors paved the way for mimicking biological neurons44,45,46. These cyclic intralayer and interlayer switching behaviors cause Cu ions to gradually accumulate or be released. The three modes of ferroelectric switching provide a platform for achieving binary switching, gradual switching, and accumulative switching in a single device.

Discussion

In summary, taking advantage of FePV behaviors, polarization switching kinetics of CIPS have been systematically investigated. It is found that FePV in CIPS presents an anomalous switching behavior, in which the phase of Isc switches six times for cyclic scanning Vp. Such anomalous FePV behavior origins from the ion hopping with the intralayer process at a small Vp and the interlayer process at a large Vp. Polarization-switching kinetics in CIPS presents three switching modes, including intralayer switching, intralayer-interlayer coupling switching, and interlayer switching. Configuring the Vp amplitude and width provides a way to tune the switching kinetics in ferroionic CIPS, achieving multiple polarization switching modes in a single device. The ion migration modulated polarization switching kinetics paves the way for the development of advanced multifunctional ferroelectric devices, highlighting the potential of combining the ferroelectric and ion transport characteristics.

Methods

Characterization of CIPS single-crystal

The as-grown crystals were characterized by X-ray diffraction (XRD) patterns. Energy dispersive X-ray spectroscopy was adopted to characterize the homogeneity and composition of the CIPS films obtained by mechanical exfoliation. The ferroelectric phase of as-grown CIPS crystals is further confirmed by the Raman spectroscopy (HORIBA LabRAM Odyssey Raman spectrometer).

Fabrication of the device

The CuInP2S6 crystals were synthesized via chemical vapor transport method (Supplementary Fig. 1). Mechanical exfoliation was employed to prepare hexagonal boron nitride (h-BN), few layer graphene and CIPS flakes, and a vertical structure of graphene/CIPS/graphene/h-BN was fabricated by stacking these flakes on 285 nm SiO2 substrate. After that, metal electrodes (Cr 8 nm/Au 50 nm) were introduced on graphene by lithography and electron beam evaporation.

Electrical measurement

The electrical measurement was performed by using an FS-Pro 380 semiconductor parameter analyzer, with the device placed in a high vacuum probe station (less than 1 × 10−3 Pa). The specific measurement steps of ferroelectric photovoltaic (FePV) are as follows: A poling voltage (Vp) was applied at the bottom graphene electrode to set the polarization state in the dark firstly, while the top graphene electrode was grounded. Subsequently, with laser illumination, I-V sweeping with a limited range of 0.1 V was adopted to obtain the FePV characteristic curves. In addition, the time-dependent short-current (Isc) at zero bias voltage was also monitored. Upon the 405 nm laser illumination, numerous carriers were generated, and Isc sharply increases from the noise level (1 × 10−13 A) to the magnitude of nA level. Given the CIPS bandgap of ~2.8 eV (Supplementary Fig. 2b), a 405 nm laser diode with a fixed light intensity of about 1.2 mW was used as the excitation source.

Atomic force microscope (AFM) measurement

AFM was performed on the Asylum Research Cypher S system using the PPP-EFM probe from NANOSENSORS™, with a spring constant of 2.8 N/m and a free resonance frequency of 75 kHz.

Piezoelectric force microscopy (PFM) measurement

PFM was performed on the Asylum Research Vero ES system using the PPP-EFM conductive probe, with a spring constant of 2.8 N/m and a free resonance frequency of 75 kHz. The Vero ES PFM is equipped with a quadrature phase differential interferometry (QPDI) detector, and makes measuring the accurate displacement of the probe tip available. Notably, the data presented in Supplementary Fig. 6f were acquired using the Asylum Research Cypher S system, employing the same type of PPP-EFM conductive probe with identical specifications. In this work, datasets obtained from the Cypher S system are explicitly indicated in the figure captions. Unless otherwise specified, all other PFM measurements were performed using the Asylum Research Vero ES system.