Introduction

HfO2-based ferroelectric (FE) thin films demonstrate promising non-volatile memory potential due to robust polarization retention at sub-10 nm thicknesses, inherent compatibility with state-of-the-art complementary metal-oxide-semiconductor (CMOS) technology, and high-density integration capability1,2,3,4. However, HfO2-based FE exhibits wake-up effects and fatigue issues, which may originate from depolarization fields or interface-related problems, consequently imposing challenges on widespread device applications5,6. To address these issues, Pešić et al. proposed a concept of antiferroelectric (AFE) nonvolatile memory (NVM), demonstrating that the built-in bias field can shift the hysteresis loop, allowing AFE materials to switch between polarized and non-polar states—thus laying a crucial theoretical and technical foundation for high-endurance memory development7,8. Recent experimental studies have indicated unique advantages of HZO-based AFE materials in NVM devices over FEs, including enhanced endurance and faster operational speeds9,10,11. However, to realize such performance, special device processes are required to engineer asymmetric work-functions (WFs) for top and bottom electrodes and thereby shift the polarization-voltage (P–V) curve to achieve FE-like characteristics and nonvolatile behavior, despite the inherent volatility of AFE films7,12,13. This WF-engineering technique involves a series of complex fabrication steps—such as deposition, annealing, wet etching, and re-deposition—which not only increase process complexity but also raise serious concerns, such as wet-etching-induced surface defects14 and degradation in device reliability, since the device performance is highly sensitive to the interfacial condition and properties. Thus, developing asymmetric electrodes (by using different materials to achieve distinct WFs) for AFE memories while maintaining high interfacial quality and device reliability remains highly challenging. Furthermore, the nondestructive readout of such nonvolatile AFE memories remains unresolved so far. The use of electrical pulses in conventional two-terminal devices for readout operations is inherently destructive, necessitating post-read rewriting, which significantly reduces the lifespan of HZO15.

Optical readout is a well-known nondestructive strategy. Nevertheless, reading the polarization states of HZO optically is challenging due to the inherent optical transparency of HfO2 (bandgap ≈5.8 eV)16. An effective approach is to couple the pyroelectric properties of HZO films with the photothermal response of the sandwiching electrodes. HZO exhibits pyroelectric behavior arising from its spontaneous polarization, where subtle thermal fluctuations induce changes in polarization strength, generating transient surface charges17,18,19. When this pyroelectric effect is combined with the photothermal response of the electrodes, a light-heat-electricity conversion mechanism can be established20,21,22: the photo-absorption of the electrodes causes a temperature rise in the device, which subsequently induces a measurable electrical response via the pyroelectric effect in the HZO films. Since the pyroelectric responsivity of HZO films depends on their polarization state8,23, this photo-pyroelectric mechanism enables optical readout for non-volatile memory devices. Notably, because it does not require any externally applied electrical pulses, this photo-pyroelectric readout is self-powered and thus inherently nondestructive. To date, photo-pyroelectric readout for detecting polarization states in hafnium-based ferroelectric devices has been explored, typically using bulky substrate electrodes (e.g., Si or Nb:SrTiO3) as the optical absorption layer24,25. However, beyond the required transparency of the top electrode and HZO layer, the temperature rise in such configurations is significantly suppressed due to efficient heat dissipation through the substrate’s high thermal conductivity, thereby degrading the photo-pyroelectric sensitivity. Moreover, it has been found that the remnant polarization (Pr) of FE HZO films is reduced compared to conventional metal–ferroelectric–metal (MFM) structures26,27. Thus, the applicability of the photo-pyroelectric readout scheme to HZO-based memory remains uncertain—particularly for AFE HZO memory, where its feasibility has not yet been examined.

In this work, we demonstrate that van der Waals integrated graphite nanosheets (GNS) can serve both as top asymmetric electrodes for AFE Hf0.1Zr0.9O2 NVM devices and as efficient photothermal harvesting layers28,29 to trigger the photo-pyroelectric effect, thereby enabling optically sensing functionality. The use of GNS as the top electrode facilitates efficient thermal energy transfer to the Hf0.1Zr0.9O2 layer under photo illumination. Transient photothermal excitation induces rapid temperature changes in the Hf0.1Zr0.9O2 before heat dissipates into the substrate, generating polarization-dependent short-circuit photocurrents measurable by external circuits. This optoelectronic response exhibits binary correlation with the programmed polarization states, establishing an optical readout mechanism that preserves structural integrity during non-destructive memory interrogation. Our devices exhibit wake-up-free operation, enhanced fatigue resistance, and voltage-programmable photo-pyroelectric response. The asymmetric GNS/W electrodes, combined with post-deposition annealing (PDA), establish a built-in bias field that shifts the polarization hysteresis loop of AFE Hf0.1Zr0.9O2 to enable two discrete polarization states. Compared with conventional work-function engineering approaches that require complex interface modifications, our method significantly reduces fabrication complexity while increasing photothermal conversion efficiency and, consequently, the reliability of photo-pyroelectric sensing. The two-terminal architecture offers superior integration potential over three-terminal field-effect transistor (FeFET) counterparts. This advance combines robust polarization switching with optical readability, addressing a critical trade-off between non-destructive readout capability and device scalability in HZO-based systems. Our demonstration provides a strategy for multifunctional memory devices that combine electrical programmability, photonics-compatible interfaces, and logic operations.

Results and Discussion

Figure 1a shows the GNS/Hf0.1Zr0.9O2/W capacitor architecture: A 7.5 nm-thick Hf0.1Zr0.9O2 film sandwiched between a 25 nm GNS (top electrode/optical absorber) and 40 nm tungsten (W, bottom electrode). The simple architecture and mature ALD-based processes offer manufacturing scalability (fabrication details in Experimental Section and Supplementary Fig. 1). Figure 1b displays cross-sectional high-resolution transmission electron microscopy (HRTEM) imaging of the full stack, confirming the 7.5 nm built-in bias field engineering AFE layer with sharp GNS/Hf0.1Zr0.9O2 and Hf0.1Zr0.9O2/W interfaces. Atomic-resolution high-angle annular dark field (HAADF) scanning transmission electron microscope (STEM) image and simultaneously acquired integrated differential phase contrast (iDPC) image are presented in Fig. 1c, d, revealing the O-phase structure of the imaged nanocrystalline. The iDPC-STEM image, in particular, shows that half of the oxygen columns deviate from the central position of the four nearest Hf/Zr atoms, a typical character of the O-phase when viewed along the [010] zone axis1,6,30,31. The projected structure models are overlaid on the experimental images for comparison. Elemental analysis via energy-dispersive X-ray spectroscopy (EDX) mapping reveals the device composition. The elemental distribution of W, Hf, Zr, O, C, Cr, and Au elements is provided in Supplementary Fig. 2, showing co-localization between zirconium and oxygen. The spectrum from our devices (Supplementary Fig. 3) shows only the highly prominent G and 2D peaks characteristic of pristine graphite. This absence of spurious peaks indicates that our dry transfer process, which utilizes a PDMS stamp without any intermediate polymers, avoids chemical contamination32. The built-in bias field-engineered HZO heterostructure, realized through post-deposition annealing (PDA) and asymmetric electrodes, enables non-volatile storage in AFE HZO. Figure 1d demonstrates this built-in bias field engineering approach through a voltage-axis shift of the P–V curve.

Fig. 1: Built-in bias field engineering in graphite nanosheets (GNS)/Hf0.1Zr0.9O2/W devices.
Fig. 1: Built-in bias field engineering in graphite nanosheets (GNS)/Hf0.1Zr0.9O2/W devices.The alternative text for this image may have been generated using AI.
Full size image

a Schematic structure of the GNS/Hf0.1Zr0.9O2/W stack. Cross-sectional (b) HRTEM characterization. c HAADF image. Hf/Zr atoms visible only. d iDPC image. Hf/Zr and Oxygen atoms visible. e Built-in bias field-induced P–V curve shift in antiferroelectric (AFE) capacitor. f Voltage-programmable pyroelectric modulation through polar/non-polar control.

According to the report by Pešić et al., the built-in bias field exhibited by WF engineering and fixed oxide charges can cause AFE hysteresis shift, enabling switching between polarized and non-polarized states in a nonvolatile manner7,12. We demonstrate the built-in bias field engineered Hf0.1Zr0.9O2 heterostructure, realized through PDA and asymmetric electrodes, enabling nonvolatile switching between upward polarization and non-polarization states at lower voltages. Figure 1e shows this built-in bias field effect through a voltage-axis offset of the P–V curve. As shown in Supplementary Fig. 4, the leakage J–V characteristics of the GNS/Hf0.1Zr0.9O2/W device are lower and exhibit greater asymmetry compared to that of the W/Hf0.1Zr0.9O2/W device. The energy band diagrams of the W/Hf0.1Zr0.9O2/W and GNS/Hf0.1Zr0.9O2/W devices are presented in Supplementary Fig. 5a and b. As illustrated in Fig. 1e, our built-in bias field-engineered Hf0.1Zr0.9O2 capacitor with GNS top electrode demonstrates voltage-controlled polarization switching: positive voltage pulse reconfigures the polycrystalline Hf0.1Zr0.9O2 matrix into a non-polar configuration, whereas negative pulse induces a polarized state. Remarkably, the pyroelectric response only appears in the polarized configuration due to its non-centrosymmetric crystal symmetry. This phase-dependent functionality enables selective photo-pyroelectric generation under instantaneous optical excitation—photon absorption in the GNS electrode induces rapid thermal transients within the polarized lattice, producing detectable photocurrent pulses through polarization-coupled pyroelectric conversion. Such opto-ferroelectric coupling enables nondestructive optical readout of programmed polarization states.

Figure 2a presents the P–V hysteresis loops of clamped post-metallization annealing (PMA) processed W/Hf0.1Zr0.9O2/W capacitor under different sweep amplitudes. The PMA-treated film exhibits an AFE-like behavior with double hysteresis loops, originating from reversible field-driven non-polar→polarized transitions30,33. Figure 2d shows current density-voltage (J-V) curves of PMA-processed W/Hf0.1Zr0.9O2/W device, revealing two switching/backswitching current pairs of AFE properties (labeled as I/I* and II/II*), where asterisks denote backswitching processes. Figure 2b exhibits AFE-like hysteresis and a small non-zero Pr at zero electric field in non-clamped PDA-processed W/Hf0.1Zr0.9O2/W device. Figure 2e exhibits a partial shift in J-V characteristics relative to Fig. 2d, demonstrating that the PDA process introduces built-in bias fields by modulating interfacial asymmetry34. The PDA-processed GNS/Hf0.1Zr0.9O2/W device shows a significant 2Pr value of ~30 μC cm−2, as shown in Fig. 2c. This value is much higher than the 2Pr value in HfO2-ZrO2 thin films with metal-ferroelectric-insulator-semiconductor (MFIS) structure24,26,27. P-V and J-V curves for PDA-processed Hf0.1Zr0.9O2 with Cr, Ni and Pt top electrodes are provided in Supplementary Fig. 6. Compared with traditional metal electrode capacitors, GNS electrode capacitors exhibit enhanced polarization values, which may be due to the van der Waals contact between the graphite electrode and Hf0.1Zr0.9O2 interface, as well as the chemically inert surface, where defects at the interface are relatively fewer35,36. Consistently, the GNS-electrode device shows the lowest leakage current among all tested samples (Supplementary Fig. 4), further confirming the reduced interfacial defect density at the van der Waals interface. The J–V curves in Fig. 2f reveal two notable FE-like polarization switching peaks. These effects originate from the electrode asymmetry37 and PDA-mediated interface modulation34 working together to create a substantial built-in bias field. The work function mismatch in asymmetric electrodes (W: 4.55 eV vs. GNS: 4.6–4.7 eV) directly establishes this built-in field38,39. In addition, we have directly measured the work function of our GNS using Kelvin Probe Force Microscopy (KPFM), which yields ~4.68 eV (Supplementary Fig. 7), in agreement with the literature. This mechanism is further evidenced by the pronounced coercive voltage (Vc) asymmetry observed in PDA-processed GNS/Hf0.1Zr0.9O2/W thin films, confirming the strong built-in electric field.

Fig. 2: Evolution of hysteresis loops of the AFE Hf0.1Zr0.9O2-based devices.
Fig. 2: Evolution of hysteresis loops of the AFE Hf0.1Zr0.9O2-based devices.The alternative text for this image may have been generated using AI.
Full size image

P–V hysteresis loops: (a) post-metallization annealing (PMA)-processed W/Hf0.1Zr0.9O2/W, (b) post-deposition annealing (PDA)-processed W/Hf0.1Zr0.9O2/W, and (c) PDA-processed graphite nanosheets (GNS)/Hf0.1Zr0.9O2/W devices. J–V switching characteristics: (d) PMA-processed W/Hf0.1Zr0.9O2/W, (e) PDA-processed W/Hf0.1Zr0.9O2/W and (f) PDA-processed GNS/Hf0.1Zr0.9O2/W devices.

Furthermore, systematic measurements of P-V characteristics in PDA HfxZr1-xO2 and WF-engineered Hf0.1Zr0.9O2 films with GNS top electrodes reveal that PDA-processed Hf0.1Zr0.9O2 achieves maximum remanent polarization, outperforming WF-engineered counterparts as shown in Fig. 3a. In addition, PDA-processed Hf0.25Zr0.75O2 and ZrO2 devices achieve higher remanent polarization compared to WF-processed devices (Supplementary Fig. 8). Notably, our PDA method circumvents the electrode replacement step required in conventional WF engineering, simultaneously simplifying device fabrication and avoiding surface defects induced by Standard Cleaning 1 processes. Next, Positive-up-negative-down (PUND) measurements were performed to verify the reliable FE-like storage characteristics of the PDA GNS/Hf0.1Zr0.9O2/W capacitors. Figure 3b shows the input voltage (double pulse) and current response during PUND measurement of the PDA GNS/Hf0.1Zr0.9O2/W device. As expected, the current corresponding to the first positive/negative pulse is higher than that of the second positive/negative pulse, confirming significant polarization switching in the PDA Hf0.1Zr0.9O2 layer. The current generated by the first positive/negative pulse includes contributions from both polarization switching and non-switching currents (such as leakage and capacitive currents)40. In contrast, only non-polarization switching currents are observed during the second positive/negative pulse, as no polarization switching occurs at this stage.

Fig. 3: Electrical Properties of GNS/Hf0.1Zr0.9O2/W Capacitors.
Fig. 3: Electrical Properties of GNS/Hf0.1Zr0.9O2/W Capacitors.The alternative text for this image may have been generated using AI.
Full size image

a The P–V hysteresis loops of the post-deposition annealing (PDA) HfxZr1-xO2 and work-function (WF)-processed Hf0.1Zr0.9O2 films with graphite nanosheets (GNS) top electrodes (TEs). b The time-domain voltage and current density waveforms in the PUND measurement of PDA-processed GNS/Hf0.1Zr0.9O2/W capacitor. c Pr of PDA-processed GNS/Hf0.1Zr0.9O2/W at various cycle numbers. d Retention characterization of PDA-processed GNS/Hf0.1Zr0.9O2/W. e Write/read scheme and the current density responses. f Distribution of each intermediate polarization state repeated 100 times.

As critical reliability parameters for nonvolatile memory applications, endurance and retention characteristics were characterized in the PDA GNS/Hf0.1Zr0.9O2/W capacitors (measurement scheme detailed in Supplementary Fig. 9). Figure 3c depicts the remanent polarization (Pr) evolution during bipolar cycling under 100 kHz bipolar cycling using 4 V triangular waveforms. The corresponding P–V and J–V curves are shown in Supplementary Fig. 10. Remarkably, stable cycling of 109 cycles was achieved, demonstrating wake-up-free operation and enhanced fatigue resistance. Supplementary Fig. 11 shows additional endurance measurements on three more devices under the same conditions, all demonstrating the same enhanced fatigue resistance. The remarkable endurance likely originates from polarization-state modulation between polar and nonpolar states, rather than full reversal of polarization, which is consistent with the mechanism proposed by Pešić et al. 7. Retention testing (Fig. 3d) confirms stable polarization states with extrapolated data retention exceeding 10 years, meeting industry requirements for NVM applications. At working temperatures of 353 K, the device also maintains good retention characteristics (supplementary Fig. 12).

Figure 3e displays polarization switching characteristics of PDA GNS/Hf0.1Zr0.9O2/W capacitor under varying write pulses. The device was initialized by a triangular pulse, and the last two pulses read the remaining polarization switched by the written pulse. Voltage-dependent intermediate states demonstrate precise control through bias amplitude modulation. To confirm consistency, we repeated each voltage pulse sequence 100 times, and measured the switchable polarization results for each state; as shown in Fig. 3f, PDA GNS/Hf0.1Zr0.9O2/W capacitor exhibits reliable consistency under conventional voltage control.

To probe the intrinsic pyroelectricity of the PDA GNS/Hf0.1Zr0.9O2/W device, Sharp-Garn measurements (TA = 1.8 K, T = 15 s; Methods) under zero-bias (Vbias = 0) conditions reveal voltage-controlled phase responses41,42. As shown in Fig. 4a, at −4 V poling (polarized state), a pure pyroelectric current density (~6 nA cm−2) emerges with near-ideal phase coherence (Φ ≈ 94°, theoretical limit 90°), yielding a giant pyroelectric coefficient p = −39.2 ± 0.5 μC m−2 K−1. This measured coefficient notably aligns with the commonly reported range (−30 to −70 μC m−2 K−1) for HfO2-based ferroelectric films19. The pyroelectric coefficient p quantifies current density generation per unit temperature change, and the negative sign confirms polarization decreases with heating (dP/dT < 0)43. At +4 V poling (non-polar state), the pyroelectric response decreases significantly, with only a small remaining value of −4.1 ± 0.4 μC m−2 K−1. This bipolar modulation evidences voltage-controlled polarization switching. Figure 4b shows hysteresis loops in pyroelectric coefficients versus voltage, matching Landau-Devonshire theory where p exhibits proportional dependence on both remnant polarization and relative permittivity19. In summary, the pyroelectric properties exhibited by Hf0.1Zr0.9O2 thin films enable non-destructive light detection by utilizing the material’s pyroelectric response and GNS light absorption.

Fig. 4: Multilevel states and pyroelectric properties of the PDA graphite nanosheets (GNS)/Hf0.1Zr0.9O2/W device.
Fig. 4: Multilevel states and pyroelectric properties of the PDA graphite nanosheets (GNS)/Hf0.1Zr0.9O2/W device.The alternative text for this image may have been generated using AI.
Full size image

a Pyroelectric current induced by sinusoidal heating (Vbias = 0 V). b The zero-field pyroelectric coefficient as a function of poling electric field. Working mechanism, (c) transient response and (d) schematic structure under light irradiation with negative poling voltage.

The self-powered photoresponse was studied systematically. As shown in Fig. 4c, d, light pulses induce heat generation in the GNS, causing temperature spikes in Hf0.1Zr0.9O2 that reduce polarization (P → P - ΔP), thereby creating current density spikes (Jpy)23. When exposed to continuous light, the pyroelectric current vanishes as temperature stabilizes. Similarly, when the light is turned off, rapid cooling generates inverted current peaks (-Jpy). In both cases, the pyroelectric current density follows Jpy = p·dT/dt, and dT/dt is the temperature change rate from light absorption44. Reproducible testing using multiple illumination pulses showed no noticeable degradation in current density peaks, confirming the stability of the photo-pyroelectric current density generation (Supplementary Fig. 13). Additionally, the response times within a switching cycle were recorded at 17 ms, as shown in Supplementary Fig. 14. Control experiments with W and graphene top electrodes (Supplementary Fig. 15) showed similar P–V loops but no obvious photoresponse, confirming GNS’s specific light absorption enables the photo-thermal-polarization coupling required for pyroelectric charge generation. This effect requires simultaneous light variation and polarization dynamics—current only occurs when both light changes and polarization exist. To investigate this phenomenon further, the surface temperature change induced by pulsed-light illumination was simulated. The details of the simulation are provided in Supplementary Fig. 16 and related text (Supplementary Information).

The device’s self-powered response characteristics were tested at Vbias = 0 V under varying light power. As shown in Fig. 5a, current density peaks grew significantly from ~23 to 79 nA cm−2 as 475 nm light power increased from 0.5 to 2 mW cm−2. This power dependence comes from stronger temperature changes and faster heating/cooling rates under higher light intensity, which amplify polarization changes. Figure 5b displays the responsivity (R = Ipy/(A·P), where A is the effective device area (~23,000 μm2) and P is light intensity) under 475 nm illumination. The maximum value reaches ~46.2 μA W−1 across different power densities. The self-powered instantaneous response was systematically investigated at Vbias = 0 using 475, 530, 585, 635, 980, and 1550 nm light sources with same power densities (2 mW cm−2). The dynamic response shown in Fig. 5c, d reveals a wavelength-dependent response from 475 to 1550 nm, indicating varying photoactivity. The highest pyroelectric current density peaks (Jpy ≈ 79 nA cm−2) at 475 nm likely results from enhanced light absorption, consistent with the GNS and HZO absorption spectra in Supplementary Fig. 17. The van der Waals GNS electrode retains stronger remnant polarization than conventional semiconductor absorption layers, while avoiding complex lithography required by subwavelength metamaterials45. The broadband optical absorption properties of graphite suggest potential for wide-spectrum photodetection. This cost-effective configuration achieves superior photothermal conversion efficiency, presenting a viable alternative for optimizing photo-pyroelectric device performance.

Fig. 5: Photo-pyroelectric performance in graphite nanosheets (GNS)/Hf0.1Zr0.9O2/W device.
Fig. 5: Photo-pyroelectric performance in graphite nanosheets (GNS)/Hf0.1Zr0.9O2/W device.The alternative text for this image may have been generated using AI.
Full size image

a Light-power-dependent photoresponse under negative poling voltage (−4.0 V, w = 10 μs). b Corresponding Responsivity (R) as a function of light intensity. c Photo-pyroelectric current density under different laser wavelengths (475–1550 nm) at the same light intensity. d Variation of the pyroelectric current density peaks in the 475–1550 nm range. e The transient photo-pyroelectric response after applying different voltage pulses (−4 V → +4 V and +4 V → −4 V, w = 10 μs). f Working mechanism, short voltage pulses cause distinct effective polarization states in the device, resulting in different photo-pyroelectric responses. g Variation of responsivity from (e) as a function of voltage pulse. (Vbias = 0 V).

In addition, transient current density measurements at zero bias showed how polarization states affect light response. As shown in Fig. 5e, f, when applying +1 V to +4 V pulses (10 μs), current density peaks decreased from ~79.1 to 5.5 nA cm−2 as the material phase changed from polar to non-polar. The polar state generates electricity through temperature-induced polarization changes (pyroelectric effect)23, while the non-polar state lacks this capability. Conversely, negative pulses (−1 V → −4 V) induce non-polar→polar state restoration with Jpy recovery to its initial ≈ 79 nA cm−2. Figure 5g demonstrates a FE-like loop in responsivity, where maximum R ≈ 39.5 μA W−1 emerges at −4 V (Hf0.1Zr0.9O2 upward polarized state), sharply contrasting with the R ≈ 2.8 μA W−1 minimum at +4 V (non-polar state). These test results are basically consistent with the pyroelectric response test results of voltage programming in Fig. 4b. These measurements demonstrate voltage-polarity switching of the photothermal-pyroelectric response, with positive pulses deactivating (OFF) and negative pulses activating (ON) the output. Supplementary Fig. 18 quantifies the ON/OFF states after 109 switching cycles, revealing <15% variation in both response amplitudes—consistent with the exceptional endurance shown in Fig. 3c. Our device achieves reconfigurable photo-pyroelectric control with >109 stable switches, outperforming existing systems (Supplementary Table S1). Notably, characterization of endurance over repeated voltage switching cycles is currently absent from the literature. This voltage-controlled photoresponse behavior enables non-destructive polarization readout, demonstrating potential for in-memory sensing and computing applications.

We demonstrate a photo-responsive AFE-like memory device based on 8-nm-thick Hf0.1Zr0.9O2 thin films with van der Waals GNS and W electrodes. The engineered built-in electric field, established through PDA and asymmetric electrode configuration, enables polarization switching between distinct states—polar and non-polar—for digital logic operation. The resulting GNS/Hf0.1Zr0.9O2/W capacitor exhibits wake-up-free polarization switching and excellent fatigue endurance. The van der Waals GNS electrode suppresses interfacial defects to enhance ferroelectric properties while functioning as a photothermal conversion component, enabling direct optical readout of polarization states through photocurrent responses under 475–1550 nm illumination. Mechanistic analysis reveals that the GNS layers efficiently convert incident light into transient thermal changes, which in turn drive the pyroelectric response without requiring external bias. Crucially, this design integrates electrical programmability and optical sensing. This work establishes a material platform for optically readable nonvolatile memory based on hafnium–zirconium oxide ferroelectrics and introduces a previously unexplored photothermal–pyroelectric coupling mechanism. These findings enable development toward next-generation, low-power, multifunctional memory devices.

Methods

Sample preparation

The 40 nm-thick W bottom electrodes were deposited on the SiO2/Si substrates via sputtering. HfxZr1-xO2 (x = 0.25, 0.1, 0) thin films with 8 nm thickness were subsequently grown by atomic layer deposition (ALD) at 230 °C. [(CH3)(C2H5)N]4Hf (TEMAH), [(CH3)(C2H5)N]4Zr (TEMAZ), and H2O were used as Hf precursor, Zr precursor, and oxygen source, respectively. The HfxZr1-xO2 films were processed through three distinct protocols: (1) For the PDA process, the HfxZr1-xO2 films underwent rapid thermal annealing (RTA) at 600 °C for 30 s in N2 ambient, then two distinct top electrode configurations were employed: mechanically exfoliated GNSs were dry-transferred as top electrodes, whereas photolithography-patterned 40-nm sputtered W top electrodes (100 × 100 μm2) served as control samples; (2) For the PMA Hf0.1Zr0.9O2 film, 40 nm W top electrodes were sputter-deposited prior to RTA crystallization; (3) For the WF process, W top electrodes capping followed by the same RTA, then the W layer was etched away via Standard Cleaning 1 wet etching in a 50:2:1 H2O/H2O2/NH3 solution at 50 °C for 5 min, and finally the GNSs were transferred onto the bare HfxZr1-xO2 film.

Details of GNS transfer: The dry transfer of GNS was performed on a dedicated two-dimensional material transfer station. Prior to the transfer, the target substrate was thoroughly cleaned by sequential ultrasonic cleaning in acetone, ethanol, and deionized water. Graphite nanosheets were prepared via mechanical exfoliation from a bulk graphite crystal using polymer tape, and transferred onto the PDMS surface. The GNS-loaded PDMS stamp was aligned and brought into contact with the substrate under optical microscopy. To facilitate GNS release, the PDMS was gradually pressed from one side against the substrate, allowing progressive contact formation, and then slowly peeled from the same edge to minimize bubble formation. Following transfer, the sample was immersed in acetone for 30 min to remove organic residues, followed by sequential rinsing with ethanol and deionized water. Finally, the sample was dried under a stream of nitrogen gas.

Microstructural characterization was performed using scanning transmission electron microscopy (STEM) with energy-dispersive X-ray spectroscopy (EDX). Cross-sectional samples were prepared via focused ion beam (FIB, FEI Helios) milling, followed by high-resolution TEM analysis. TEM imaging confirmed the polycrystalline nature of the ferroelectric Hf0.1Zr0.9O2 layer. The HRTEM imaging was performed on a field emission transmission electron microscope operated at 200 kV (Talos F200i, Thermo Fisher Scientific). Aberration-corrected HAADF and iDPC imaging were performed on a double aberration-corrected scanning/transmission electron microscope operated at 300 kV (Themis Z, Thermo Fisher Scientific). For atomic-resolution STEM imaging, the convergence angle was set to 21.4 mrad under aberration-free conditions, with the DPC signal collected between 10 and 37 mrad and the HAADF signal collected from 39 to 200 mrad. Raman spectroscopy characterization was performed using a HORIBA HR Evolution Raman spectrometer with 532 nm laser excitation. Using Molecular Vista Photoinduced Force Microscopy (Kelvin Probe Force Microscopy mode), we measured the contact potential difference (CPD) between the GNS and a gold reference electrode.

Device characterization

Ferroelectric properties were evaluated using a Keithley 4200A-SCS parameter analyzer equipped with the 4225-PMU module. Polarization-voltage (P–V) hysteresis loops and current density-voltage ( J–V) characteristics were measured at 1 kHz with ±4 V bipolar excitation. The endurance tests applied 4 V square-wave voltages at 100 kHz frequency, with the bottom electrode grounded throughout measurements.

Pyroelectric and photo-pyroelectric measurements

Pyroelectric measurements were carried out with the Sharp-Garn method under zero bias, applying sinusoidal thermal modulation41,42. The pyroelectric current was quantified using:

$$p={I}_{A}{{\cdot }}\sin (\varPhi )/({T}_{A}{{\cdot }}A{{\cdot }}2{{{\rm{\pi }}}}f)$$
(1)

where IA is the measured current amplitude, Φ is the phase shift between the temperature and current signal, TA is the temperature amplitude, A is the capacitor area, and f is the temperature frequency. Hence, a phase shift of Φ = 90° represents a pure pyroelectric signal, and Φ = 0° is a non-pyroelectric signal. Keithley 4200A-SCS was used to measure the output current. The optical pulses were generated by a function generator (RIGOL DG1022Z) driving light-emitting diodes (λ = 475, 530, 585, 635, 980, 1550 nm). All light response data were acquired under zero bias voltage. The incident light intensity was measured by the light power meter (THORLABS). Visible absorption spectra were measured with a Filmetrics F40 thickness gauge.