Interfacial Engineering of Two-Dimensional Perovskite Memristors toward Reliable and Flexible Optoelectronic Memory Arrays

Abstract

Two-dimensional (2D) Ruddlesden–Popper halide perovskites are promising candidates for flexible optoelectronic memories owing to their tunable composition, reduced power consumption, and intrinsic structural anisotropy. Yet, achieving long-term reliability and environmental stability in perovskite-based memristors remains a major challenge. Here, we demonstrate highly oriented (PEA)₂MA₄Pb₅I₁₆ thin films integrated into optoelectronic memristors, where interface optimization with softly deposited Ag nanoparticles (NPs) enhances device performance without damaging the active layer. The resulting forming-free ITO/Ag NPs/(PEA)₂MA₄Pb₅I₁₆/Pt NPs/ITO devices exhibited robust binary resistive switching, with a large memory window (∼10⁵), excellent endurance (>10¹¹ cycles), and a low SET voltage (∼0.2 V). The confined conducting filament (CF) formation, induced by the vertically aligned crystalline domains, combined with the energetically favorable atom extraction from Ag NPs, ensured stable and low power switching under both electrical and optical stimuli. Numerical simulations were further conducted, elucidating the origin of the enhanced switching behavior and the role of crystal orientation in mitigating perovskite degradation pathways. The device conductance can be also modulated by applying light pulses, yielding a low energy consumption (∼100 nJ per programming event at 740 nm), whereas multicolor perception was achieved using additional wavelengths (450 and 550 nm). A 10 × 10 flexible crossbar array was fabricated, demonstrating high device yield, reproducible performance under bending, and excellent environmental stability over an extended period. This work establishes (PEA)₂MA₄Pb₅I₁₆ as a benchmark material for reliable, low-power, and optically programmable memristors and provides a scalable strategy for next-generation perovskite-based optoelectronic architectures.

1. Introduction

The continuous demand for faster, denser, and more energy-efficient data storage has fueled the search for next-generation nonvolatile memory technologies capable of surpassing the physical and scaling limits of conventional CMOS-based flash architectures. Among emerging candidates, resistive random-access memory (RRAM) has attracted significant attention owning to its simple metal–insulator–metal configuration, fast switching speed, high endurance, and compatibility with three-dimensional integration. In filamentary RRAM devices, nonvolatile resistance states arise from the formation and rupture of conductive filaments (CFs), enabling both data storage and in-memory computing functionalities for neuromorphic applications. Despite rapid progress, the stochastic nature of CFs remains a major challenge for reliable memristive operation. In conventional metal oxide systems, random nucleation sites and uncontrolled ion migration lead to device-to-device variability, cycle-to-cycle fluctuations, and unpredictable switching thresholds. , Such variability limits precise synaptic weight modulation and hinders scalability in neuromorphic architectures. Two-dimensional (2D) Ruddlesden–Popper (RP) perovskites offer a promising route to mitigate such stochasticity, as their layered crystal structure and vertical orientation can spatially confine ionic migration pathways, promoting reproducible CF dynamics. − Moreover, interfacial engineeringsuch as introducing metallic nanoparticles (NPs) or tailored buffer layerscan further localize electric fields and control filament geometry, paving the way toward deterministic switching and enhanced operational stability. ,

Compared with their three-dimensional (3D) counterparts, 2D perovskites exhibit markedly enhanced structural and environmental stability, making them highly attractive for next-generation device applications. , The incorporation of hydrophobic organic spacer layers in 2D perovskites effectively suppresses moisture entrance and lateral ion migration, leading to exceptional long-term environmental and operational stability. Their quantum-well-like structure allows tunable quantum confinement, which can be optimized to balance charge transport and exciton binding energy. In RP phases with higher layer numbers (n > 1), the reduced confinement enhances carrier mobility, while the presence of large organic cations assists in defect passivation and minimizes nonradiative recombination losses. These synergistic effects result in improved photocarrier dynamics, stronger light–matter interactions, and superior resistance to degradationproperties that collectively position 2D perovskites as a robust and versatile alternative to traditional 3D perovskite systems for optoelectronic memory platforms.

However, crystal orientation and grain boundary distribution critically affect the CF properties and overall stability of 2D perovskite-based memristors. Vertically aligned crystalline domains promote uniform electric field distribution and controlled CF formation, leading to reproducible switching behavior and enhanced device reliability. In contrast, randomly oriented grains and abundant grain boundaries introduce local electric field distortions and defect sites that can act as uncontrolled nucleation centers for CF growth, resulting in variability, premature breakdown, and poor endurance. Moreover, grain boundaries often serve as diffusion pathways for ions and metal species, accelerating CF degradation and compromising retention stability. The majority of the reported works in the literature dealing with 2D perovskite memristors have demonstrated limited endurance and retention performance due to their rich polycrystalline nature and the existence of a large number of grain boundaries. − Therefore, precise control over crystal orientation and interfacial structure is essential to achieve localized filament formation and long-term operational reliability.

In parallel, to translate these advances into practical technologies, it is essential to realize flexible and mechanically robust memristive platforms. Integrating perovskite-based memristors on plastic substrates enables conformable, lightweight, and large-area neuromorphic circuits suitable for wearable and biointerfaced electronics. However, achieving stable switching in flexible devices remains difficult due to interfacial degradation, strain-induced defects, and environmental sensitivity of the perovskite lattice. Addressing these challenges will unlock the full potential of perovskite memristors for flexible, low-power, and optically reconfigurable neuromorphic systems.

Along these lines, in this work, a highly stable and multifunctional 2D metal halide perovskite memristor array on a transparent flexible polyethylene naphthalate (PEN) substrate was demonstrated based on the hybrid compound (PEA)₂MA₄Pb₅I₁₆ (where MA denotes methylammonium cation), achieving operational stability exceeding five months (Figure ). The introduction of the Pt and Ag NP layers was found to play a crucial role in suppressing the crack formation in the electrode materials, promoting uniform CF formation and enhancing switching reliability. A numerical model was further employed to elucidate the mechanisms underlying the improved switching behavior and to assess how crystal orientation affects the degradation pathways of the perovskite material. Moreover, the use of hydrophobic large organic cations controlled the crystal orientation growth toward a vertically aligned RP phase, as was confirmed through grazing-incidence X-ray diffraction (GIXRD) measurements. In this way, the resulting crystal structure provided efficient vertical channels for silver ion migration, ensuring high endurance (10¹¹ cycles) and long retention (10⁵ s). The device exhibited light-tunable conductance with ultralow energy consumption (∼10 pJ per programming event at 740 nm) and multicolor perception at 450 and 550 nm, while a 10 × 10 flexible crossbar array demonstrated high yield and stable operation under mechanical bending. These advances highlight the importance of structural engineering and interfacial design in 2D perovskite systems, offering a viable pathway toward flexible, reconfigurable, and low-power memory architectures for next-generation neuromorphic and optoelectronic applications.

1.

(a) Diagram showing the 10 × 10 perovskite array constructed on a flexible substrate. (b) Optical photograph of the crossbar array. (c) Close-up view of the 10 × 10 array.

2. Results and Discussion

2.1. Growth and Characterization of Vertically Oriented 2D Halide Perovskite Films

A comprehensive characterization was performed to establish the identity, quality, and crystallographic orientation of the synthesized perovskite film. The collective data provides a consistent and robust model of the material as a high-quality, (PEA)₂MA₄Pb₅I₁₆ (n = 5) RP phase. To confirm this identification, a (PEA)₂PbI₄ (n = 1) control film was also synthesized and analyzed. The film’s structural identity and crystallographic orientation were established by performing GIXRD measurements. The GIXRD pattern (Figure a) confirms the successful formation of the (PEA)₂MA₄Pb₅I₁₆ (n = 5) RP phase. The pattern reveals that the film is not randomly oriented but instead possesses a complex, highly oriented structure with two distinct orientations. The pattern displays a strong (002) peak at 5.3°, which is the definitive signature of the “c-axis” orientation. This was validated by a (PEA)₂PbI₄ (n = 1) control film, which showed an GIXRD pattern (Figure S1) almost exclusively dominated by a very intense peak at this same 5.3° position, confirming it as the hallmark of c-axis growth. However, in our (PEA)₂MA₄Pb₅I₁₆ film, a second reflection, the (111) peak at 14.1°, is even more intense, establishing it as the dominant feature in the pattern.

2.

(a) GIXRD pattern, (b) steady-state photoluminescence (PL) spectra and (c) UV–vis spectra of the (PEA)₂MA₄Pb₅I₁₆ film.

The relatively strong PbI₂ diffraction peak at 2θ ≈ 12.6° arises from our deliberate stoichiometric design, in which 1 mmol of PbI₂ and 0.4 mmol of PEAI create an inorganic-rich environment; this residual PbI₂ passivates grain boundaries and reduces nonradiative recombination. The large full width at half-maximum (fwhm) of the (002) and (111) peaks reflects the nanocrystalline nature intrinsic to RP (n = 5) perovskites rather than poor crystallinity. According to the Scherrer equation, broader peaks correspond to smaller crystallite sizes, consistent with structural confinement by bulky PEA⁺ spacers that limit the growth of inorganic octahedra into narrow quantum-well domains. Similarly, the increased surface roughness observed in atomic force microscopy (AFM) images is a consequence of vertically oriented grains, which are characteristic of quasi-2D perovskites and facilitate efficient out-of-plane charge transport. Therefore, both the GIXRD broadening and platelet-like surface morphology are expected physical features of high-quality n = 5 quasi-2D films optimized for electronic performance.

To further evaluate the environmental stability of the perovskite films, optical photographs and X-ray diffraction (XRD) measurements of the (PEA)₂MA₄Pb₅I₁₆ films were recorded immediately after fabrication and after 150 days of storage under vacuum conditions (Figure S2). No discernible change in film color or transparency was observed, indicating the absence of significant degradation, which would otherwise manifest as yellowing due to PbI₂ formation. Consistently, the characteristic (002) and (111) diffraction peaks remain well preserved after long-term storage, confirming the retention of the quasi-2D crystal structure and demonstrating the excellent chemical stability of the films over time.

As established in the literature, the orientation of the inorganic components is key to device performance. A perfect vertical alignment (where the inorganic layers are exactly perpendicular to the substrate) is signified by a dominant (202) peak, while a tilted vertical orientation is represented by the (111) peak. In our film, the (202) reflection at ∼28.5° is significantly suppressed, while the (111) reflection is clearly dominant. While the (111) texture represents a “tilted” orientation rather than a perfect vertical alignment, it marks a significant deviation from the insulating c-axis (00l) orientation. The dominance of these (111) domains suggests that the inorganic slabs are oriented at an angle that allows for continuous percolation pathways between the bottom and top electrodes. Therefore, the GIXRD data reveal a highly ordered film structure composed of a mixture of domains: c-axis oriented (002) domains and, crucially, a dominant fraction of (111)-oriented tilted vertical domains. , This specific (111) texture is the key structural feature enabling vertical charge conduction in the device. This structural identification is supported by X-ray photoelectron spectroscopy (XPS) (Figure S3), which confirmed the film’s chemical composition, verifying the successful incorporation of both PEA⁺ and MA⁺ organic cations, as well as the Pb²⁺ and I^– ions.

The optical and electronic properties of the film are in excellent agreement with the complex, bimodal structure identified by GIXRD. To provide a clear reference, a (PEA)₂PbI₄ (n = 1) control film was analyzed, which showed only a single, sharp PL emission at 520 nm (Figure S4a), confirming this as the signature of the n = 1 phase. This baseline makes the PL spectrum of the (PEA)₂MA₄Pb₅I₁₆ (n = 5) film (Figure b) clear to interpret. The n = 5 film exhibits a classic energy cascade with two distinct features: a sharp, high-energy peak at 520 nm and a dominant, broad emission centered at 705 nm.

This dual–emission profile is fully consistent with the bimodal structure observed in GIXRD. The sharp 520 nm peak can be assigned to the n = 1-like “quantum-well” (QW) domains. The dominant 705 nm emission corresponds to the fundamental optical transition of the target n = 5 “bulk” domains. The GIXRD pattern shows the coexistence of c-axis (002) domains (which are structurally n = 1-like) and the desired “tilted” (111) domains (which represent the bulk n = 5 phase). Therefore, both PL and GIXRD confirm that the film is not a single, perfect phase but rather a complex, mixed-phase material composed of both n = 1 QW domains and n = 5 bulk domains. This dual–emission profile is strongly corroborated by the UV–vis absorption spectrum (Figure c), which displays a sharp excitonic peak at 517 nm (correlating with the QW domains) and a distinct absorption shoulder at approximately 710 nm (correlating with the n = 5 bulk bandgap). ,

Furthermore, ultraviolet photoelectron spectroscopy (UPS) provided the film’s absolute energy levels (Figure S4b), revealing a valence band maximum (VBM) at −5.63 eV and a conduction band minimum (CBM) at −3.71 eV. The CBM was determined using the value of the optical bandgap (E _g), being 1.92 eV, derived from the Tauc plot of the UV–vis absorption spectrum (Figure S4c). Critically, the work function was also measured and from this, the Fermi level (E _F) was determined to be at −4.81 eV. This position, located almost perfectly in the middle of the 1.92 eV bandgap, confirms the material is nearly intrinsic and possesses a very low background doping density. , AFM measurements were also conducted, indicating the relatively low surface roughness of the perovskite film (Figure S5), which is of paramount importance for device fabrication, as it promotes a high-quality top–electrode interface and enables a uniform electric field distribution.

2.2. Resistive Switching Performance and Modeling

Figure illustrates the device architecture and corresponding electrical characteristics of the proposed perovskite-based resistive switching element. As shown in Figure a, the vertical heterostructure comprises indium tin oxide (ITO) bottom and top electrodes sandwiching a hybrid perovskite film, (PEA)₂MA₄Pb₅I₁₆, that accommodates both metallic Ag and Pt NPs as conductive nucleation sites. The perovskite matrix provides a mixed ionic–electronic environment that facilitates the formation of percolating CFs across the film thickness, while the distribution of Ag nanoclusters (NCs) and Pt NPs governs the local electric-field concentration and filament stability. The cross-sectional SEM image in Figure b reveals a compact, uniform interface between the hybrid perovskite and the bottom electrode (BE) layer.

3.

(a) Schematic illustration of the vertical device structure and the percolating CF. The ITO electrodes have a thickness of 60 nm, while the perovskite active layer is approximately 400 nm thick. Reducing the perovskite thickness resulted in increased switching variability and reduced operational stability, whereas thicker ITO layers led to crack formation during strain measurements, adversely affecting the mechanical reliability of the devices. (b) Cross-section scanning electron microscopy (SEM) image of the (PEA)₂MA₄Pb₅I₁₆ interface. The scale bar corresponds to 100 nm. (c) I–V hysteresis patterns for the reference and (PEA)₂MA₄Pb₅I₁₆ samples with the application of 10 mV/s. A constant I _cc of 100 μA was enforced for both samples. The numbers and arrows in the graph signify the switching direction. Similar hysteresis patterns were obtained by starting the voltage scans from 0 V to either positive or negative biases. Devices from five fabrication batches exhibit consistent bipolar switching with well-defined hysteresis loops, indicating excellent device-to-device and batch-to-batch reproducibility (Figure S8a–e). (d) I–V consecutive cycling behavior after the implementation of 200 DC endurance cycles. The variation of both HRS and LRS during consecutive cycling can be found in Figure S8f. (e) Measured and calculated current responses under the application of a constant I _cc of 100 μA (f) Measured current responses under the application of various I _cc ranging from 1 μA to 100 μA. (g) Measured and calculated distribution of the LRS as a function of I _cc. (h) Pulse endurance measurements under the application of alternative pairs of +1 V/1 μs and −0.5 V/1 μs square pulses. (i) Retention performance at elevated temperature. The retention measurements were conducted under vacuum conditions (10^–1 mbar). The endurance and retention characteristics were evaluated on a total of 10 independent devices fabricated on the same plastic substrate. The results, summarized in Figure S9, demonstrating good device-to-device reproducibility with limited performance variability. All responses were recorded at a read-out voltage of 100 mV.

The current–voltage (I–V) characteristics presented in Figure c–f demonstrate a reproducible and controllable resistive switching behavior without applying any electroforming step. In Figure c, the reference device incorporating bulk Ag displays a typical bipolar hysteresis. The transition from the high resistance state (HRS) to the low resistance state (LRS) takes place at about V _SET ∼ 210 mV, while for the opposite effect, a V _RESET of about −100 mV is required. On the contrary, the sample containing Ag NPs exhibits reduced V _SET (∼180 mV) and sharper switching transitions, indicating facilitated extraction of Ag atoms when Ag NPs are placed on top of the perovskite structure (Figure S6). Indeed, metal NPs display markedly enhanced surface reactivity and atomic mobility, a direct consequence of their high surface-to-volume ratio. This structural feature lowers the energy barrier for atom migration or extraction from the NPs, effectively making the release or rearrangement of metal atoms energetically more favorable. As a result, the formation and dissolution of CFs within devices can occur efficiently under minimal energy input. −

The advantageous impact of Ag NPs is further evident from the direct current (DC) endurance measurements. More specifically, the reference sample exhibited irregular hysteresis behavior (Figure S7a), with the V _SET showing substantial variability. Such inconsistencies may arise from residual Ag incorporated into the perovskite matrix during deposition, as previously demonstrated in similar systems, , leading to the unpredictable formation of CF in each switching cycle. In contrast, samples containing Ag NPs displayed highly reproducible hysteresis patterns across consecutive cycles, highlighting their stabilizing effect on device performance (Figure d). The role of both Ag and Pt NPs is clarified in Figure S7b, where a comparison between three different device configurations is presented. These results indicate that Ag NPs facilitate stable CF formation at reduced energy input, whereas Pt NPs enhance local electric fields via increased interfacial roughness, collectively leading to improved switching uniformity and endurance. The beneficial influence of Ag NPs on device performance is further illuminated through numerical modeling of the switching behavior. A previously reported self-consistent model was employed to reproduce the experimental patterns, assuming a thermally accelerated migration mechanism for Ag ions. The resistive switching behavior was modeled by describing Ag⁺ migration through the perovskite with drift, diffusion, and thermodiffusion fluxes, while vertically aligned RP phases were treated as preformed channels with enhanced ionic mobility and preferential nucleation, enabling accurate reproduction of CF formation and the memristive response via coupled drift–diffusion, current continuity, and Joule heating equations.

The filamentary nature of the switching is supported by the independence of the LRS from the total device area (Figure S10), in agreement with recent experimental observations of CF formation in mixed organic–inorganic halide perovskites. According to our model, Ag atoms can accumulate and coalesce into NCs due to their elevated solid solubility and induced material precipitation, which facilitates the controlled formation of CFs. This behavior contrasts with the reference sample, where irregular hysteresis and large V _SET variations likely arise from uncontrolled Ag incorporation during deposition, leading to stochastic CF formation. In devices containing Ag NPs, the pre-existing NPs act as nucleation sites, stabilizing CF formation and yielding reproducible hysteresis over consecutive switching cycles. Additional details on the simulated device geometry, boundary conditions, and switching dynamics are provided in Figures S11–S13. The measured I–V curves are in excellent agreement with numerical simulations based on filamentary conduction models (Figure e). The model is also able to capture the transient responses of our devices (Figure S14), while insights about the origins of the transition from the HRS to LRS under negative bias polarity are also provided (Figure S15). Furthermore, varying the current compliance (I _cc) from 1 μA to 100 μA (Figure f) enables multilevel conductance states, demonstrating the tunability of the LRS through controlled CF growth.

The correlation between LRS resistance and I _cc, depicted in Figure g, follows a power–law relation with a slope close to 1, consistent with a self-limited metallic filament formation process. Alternating current (AC) endurance measurements under pulsed operation were also conducted (Figure h), confirming stable switching over more 10¹¹ alternating pulses of +1 V and −0.5 V with pulse widths of 1 μs, without observable degradation of either the high-resistance state HRS or LRS. Finally, the retention measurements at elevated temperature (85 °C, Figure i) verified excellent nonvolatile properties, with both resistance states remaining stable for over 10⁵ s. A benchmark analysis is provided in Tables S3 and S4, indicating the enhanced performance of our prototypes with various PEA-based devices − and other material configurations that have been fabricated on flexible substrates.

The model was further adapted to investigate the stability characteristics of vertically oriented 2D perovskites. The simulations incorporated intrinsic degradation mechanisms specific to the layered architecture, including interlayer ion migration and anisotropic phase evolution, described within the Cahn–Hilliard framework. − Consistent with experimental observations (Figure b), negligible deterioration of the memory elements was found under typical coupling coefficients, confirming the structural resilience of the vertical configuration over extended operation periods. In contrast, under extreme coupling or elevated bias conditions, the model predicts a mode transition from bipolar to volatile switching (Figure S16). However, no such behavior was experimentally observed at any I _cc. From an ionic transport perspective, the anisotropic crystal orientation of vertically structured 2D perovskites plays a crucial role in suppressing cation migration and phase segregation. The layered structure, comprising inorganic slabs separated by bulky organic cations, introduces energy barriers that hinder long-range diffusion of mobile species, including halide ions and Ag⁺. Ion transport is thus predominantly confined along vertical pathways defined by the perovskite octahedral framework, enabling localized and reproducible CF formation. In addition, the organic spacer layers mitigate electric-field-driven halide redistribution, limiting compositional inhomogeneity and suppressing iodine-rich or lead-deficient domain formation. Consequently, vertically oriented 2D RP perovskites preserve chemical uniformity and stable switching behavior under prolonged operation, rendering them highly suitable for reliable memristive devices.

4.

(a) Schematic illustration of the device passivation against humidity. (b) I–V hysteresis patterns at a period of 5 months. The devices were primarily stored under vacuum conditions and were exposed to ambient atmosphere only during the electrical characterization. (c) Histogram depicting the HRS and LRS state for 100 (10 × 10) different cells. (d) CDF plots for both HRS and LRS in the fabricated array. (e) Optical image of the flexible crossbar array under strain conditions. Measurements were performed on devices located at the center of the substrate to ensure uniform strain. The samples were tested under a three-point bending configuration with motor-controlled cantilever displacement, and the effective bending radius experienced by an individual memristor device was approximately 5.2 mm. (f) I–V hysteresis patterns under the application of a constant strain load of 1.2%. The statistical distribution of V _SET, V _RESET, and the memory window as a function of the bending cycles is provided in Figure S17f. The results indicate that these key electrical parameters remain largely stable, confirming the excellent mechanical robustness of the proposed devices.

2.3. Reliability Assessment of Vertical Perovskite Memristors on a Plastic Substrate

Figure a illustrates the effectiveness of the passivation layer in protecting the perovskite-based memristive stack against environmental degradation. The organic spacer layers in the 2D lattice act as hydrophobic barriers, effectively preventing the ingress of moisture and oxygen, which are major causes of perovskite decomposition. In the vertically aligned configuration, these organic layers are oriented parallel to the electrodes, forming compact interfaces that hinder water penetration while maintaining efficient out-of-plane charge transport. Additionally, the strong van der Waals bonding between the organic and inorganic sublayers suppresses lattice strain accumulation during resistive switching cycles, minimizing the generation of defects and halide vacancies that typically accelerate chemical instability. This encapsulation strategy preserves the crystal integrity and maintains the defect landscape responsible for controlled resistive switching. Consequently, the device retains its electrical performance even under long-term ambient exposure, as verified by the stability measurements presented in Figure b. Furthermore, the statistical evaluation of 100 individual memory cells (Figure c) reveals consistent behavior with a narrow distribution of HRS and LRS values. The cumulative distribution function (CDF) plots shown in Figure d further support the uniformity of the resistive switching characteristics, exhibiting small confidence of variance (σ/μ) ratios of 0.12 and 0.25 for the HRS and LRS, respectively, indicative of reliable device-to-device reproducibility across the array.

Mechanical flexibility was also evaluated to assess the feasibility of integrating the devices into wearable electronics. As shown in Figure e, the crossbar array retains its structural integrity when subjected to tensile bending, with no observable fracture across the electrode stack. The mechanical robustness is preserved even under repeated bending cycles, indicating that the device design can effectively accommodate mechanical strain without compromising the integrity of the active layers.

The electrical characteristics recorded under a constant tensile strain of 1.2% (Figure f) demonstrate only negligible variations in both the SET/RESET voltages and the ON/OFF current ratio. This minimal degradation highlights the intrinsic strain tolerance of the interface-engineered device structure, enabling stable resistive switching during continuous mechanical deformation. Importantly, the incorporation of the metallic NP interfacial layer significantly contributes to this enhanced mechanical reliability. As supported by the structural analysis in Figure S17a–e, the NP layer acts as a crack-inhibiting buffer, distributing mechanical stress more uniformly and thereby mitigating crack formation in brittle electrode materials, and preventing the propagation of strain-induced defects. The preservation of the I–V hysteresis behavior under mechanical stress highlights the mechanical robustness of the layered perovskite architecture and the stability of the metal–halide framework, confirming its potential for application in flexible and wearable memristive systems.

2.4. Optoelectronic Programming of Flexible Crossbar Array

Figure a illustrates the multicolor optical programming capability of the flexible perovskite crossbar array. As schematically depicted in Figure b, the device structure enables selective photoactivation through red (R) (740 nm), green (G) (550 nm), and blue (B) (450 nm) illumination using an optical mask, each inducing distinct resistive states within the array. The statistical analysis presented in Figure b demonstrates the reproducibility and separation of the resistance levels corresponding to the three illumination wavelengths across 100 individual cells. The narrow distribution and minimal overlap between the HRS and LRS under different colors confirm the reliable and independent optical programming of each channel.

5.

(a) Schematic illustration of the flexible perovskite crossbar array enabling multicolor optical programming using red (740 nm), green (550 nm), and blue (450 nm) illumination using an optical mask. (b) Statistical distribution of HRS and LRS values for 100 cells under the three wavelengths, demonstrating clear separation between the various optical memory states. Resistance maps showing wavelength-dependent programming, reset, and reprogramming of character “A” in the array. The width of the optical pulses was 1 s and a light intensity of 1.4 mW/cm² was delivered. (c–h) Uniform reprogramming behavior under repeated RGB-light exposure. Retention of optically written states after (i) 10, (j) 60, and (k) 120 min, confirming excellent nonvolatile stability. The latter measurements were conducted under air conditions. In all cases, a read-out voltage of 100 mV was used.

As shown in the resistance mapping plots (Figure c–h), the array exhibits clear and uniform switching behavior during the programming, reset, and reprogramming cycles. The spatially resolved resistance patterns validate the feasibility of encoding optical information in a matrix configuration, where each cell can be selectively addressed and rewritten using different wavelengths of light. For the erase process, a constant negative bias of −0.3 V was applied, while the read-out process was performed at 100 mV for each pixel. The relatively high ON/OFF ratios permit the storage of the desired character with a sharp contrast. The electrical energy consumption of the generated photoresponses was calculated using the following equation

$$E_{\mathrm{e}\mathrm{l}\mathrm{e}\mathrm{c}\mathrm{t}\mathrm{r}\mathrm{i}\mathrm{c}\mathrm{a}\mathrm{l}}=V\times I\times t$$ 1

where V is the voltage pulse amplitude (100 mV), I is the measured current, and t _spike is the pulse duration (∼1 s). The resulting energy per event was approximately 100 nJ per programming event for red light irradiation, which was increased to about 350 nJ and 1 μJ during green and blue light irradiation, respectively.

To assess reconfigurability and retention, additional optical cycling and time-dependent measurements were performed. The array maintains consistent reprogramming behavior under repeated blue-light exposure (Figure f–h), exhibiting stable conductance levels without noticeable fatigue. Furthermore, the resistance states remain unchanged for at least 120 min after optical writing (Figure i–k), confirming the nonvolatile nature and temporal stability of the photoinduced memory states. These results highlight the robustness of the perovskite-based optoelectronic memory platform, which combines wavelength-selective tunability, long-term retention, and mechanical flexibilitykey attributes for future applications in photonic logic circuits, neuromorphic computing, and flexible information storage systems. ,

The multicolor optical programming behavior arises from photoassisted Ag⁺ ion migration from the Ag NPs. − Illumination with different wavelengths generates photocarriers of varying energies and densities in the (PEA)₂MA₄Pb₅I₁₆ layer, which modulate the energetics of Ag atom oxidation and Ag⁺ injection at the Ag NP/perovskite interface (Figure S18). Shorter-wavelength light (blue and green) produces higher-energy carriers and larger carrier concentrations, enhancing local conductivity and partially screening the Coulombic binding of Ag atoms within the NPs, thereby lowering the activation energy for Ag⁺ release and accelerating ion drift through the vertically oriented RP domains. In contrast, longer-wavelength red light generates fewer and lower-energy carriers, resulting in weaker Ag⁺ activation and more limited filament growth. Wavelength-dependent optical absorption further influences the spatial distribution of Ag⁺ migration: blue and green light are absorbed near the Ag NP/perovskite interface, promoting localized Ag⁺ injection and rapid filament thickening, whereas red light penetrates deeper, producing a more distributed photocarrier profile and reduced Ag⁺ accumulation. Once injected, Ag⁺ ions migrate along preferential vertical pathways defined by the RP crystalline domains and are reduced near the Pt NPs interface. The interplay of carrier-energy effects, selective absorption, and controlled CF termination at Pt NPs results in wavelength-dependent, nonvolatile, and reproducible multilevel resistance states, highlighting the critical role of interface and crystal-orientation engineering in achieving deterministic multicolor optoelectronic memory behavior.

3. Conclusions

In summary, we successfully fabricated high-performance optoelectronic memristors by integrating softly deposited Ag NPs with a highly oriented quasi-2D (PEA)₂MA₄Pb₅I₁₆ (n = 5) film, confirmed by GIXRD to possess dominant tilted vertical (111) crystalline domains. This specific crystal and interface engineering strategy yielded devices that are forming-free and exhibit benchmark performance, including an exceptionally large memory window (∼10⁵), superior endurance (10¹¹ cycles), and a low SET voltage of ∼0.18 V. We revealed that these vertical domains are critical for guiding confined CF formation, which, coupled with Ag NPs-mediated ion dynamics, ensures stable switching under both electrical and optical stimuli. Crucially, the devices demonstrated advanced optoelectronic functionality, allowing for precise, multicolor-responsive conductance modulation with ultralow energy consumption (∼100 nJ). Validated by a robust, flexible 10 × 10 crossbar array, our findings establish this highly oriented (PEA)₂MA₄Pb₅I₁₆ as a benchmark material for reliable, reconfigurable neuromorphic hardware and present a general, scalable strategy for next-generation perovskite-based optoelectronics.

4. Methods

4.1. Materials

Lead iodide (PbI₂, 99.99%) was supplied by TCI EUROPE N.V., Methylammonium iodide (MAI, 99.99%) and Phenethylammonium iodide (PEAI, 99.99%) were acquired from Greatcell Solar Materials. N,N-dimethylformamide (DMF, 99.8%), dimethyl sulfoxide (DMSO, 99.8%) and Isopropanol, (IPA,99.5%) were purchased from Acros Organics B.V.B.A.

4.2. Materials Characterization

The UV–vis transmittance spectra of the perovskite films were recorded with a Carry 60 Agilent UV–vis spectrometer. Steady-state photoluminescence spectra were recorded on a Horiba Fluoromax+ spectrofluorometer, equipped with a 150 W Xe lamp as the excitation source and a Hamamatsu R13456 PMT (185–980 nm nominal) as the emission detector. For the time-resolved fluorescence spectra, a time-correlated-single-photon-counting (TCSPC) method via a Fluorohub single photon counting controller, a Delta-hub laser diode as an excitation source (404 nm, pulse duration <150 ps) and the PMT detector were applied, along with a 450 nm optical filter intervened between the sample and the PMT. The lifetimes were determined by exponential fitting of the as-collected data with the DAS6 software. The XPS data were collected with a PHOIBOS 100 (SPECS) hemispherical analyzer using Mg Ka X-ray source with photon energy 1253.64 eV. Voigt functions were used for the fitting analysis after standard Shirley background subtraction. TEM characterization was carried out on an FEI CM20 instrument operated at 200 kV. SEM imaging was conducted using a Jeol JSM 7401F system with an accelerating voltage of 2 kV. The AFM measurements were conducted using the Park NX-Hivac equipment from Park Systems. The GI-XRD measurements were performed on the films using a Bruker D8 Advance diffractometer equipped with a Cu Kα radiation source (λ = 1.5406 Å) operated at 40 kV and 40 mA, and a LYNXEYE XE-T detector. The data were collected in grazing-incidence geometry over a 2θ range of 1–30°, using a step size corresponding to 0.05°/s and 1–2 s/step scan speed, with a fixed incident angle of θ = 0.25°, chosen to enhance surface crystallinity and minimize contributions from the substrate.

4.3. Perovskite Film Fabrication

Indium-doped tin oxide (ITO) substrates were used. The substrates were cleaned by sonication washing with detergent (Helmanex III 2% in deionized water), deionized water, acetone, and isopropanol for 15 min. Before perovskite deposition, the cleaned ITO substrates were treated with ultraviolet ozone for 30 min and then were transferred immediately into a N₂ glovebox. The precursor solution of perovskite was prepared by was prepared by dissolving 1 mmol of PbI₂ and 0.4 mmol of PEAI in 1 mL of DMF/DMSO (4:1 v/v, premixed). For the MAI solution, 0.126 mmol of MAI in 1 mL of IPA is used to obtain the final solution. Upon solvent addition, the mixtures were stirred on an orbital shaker for approximately 2 h. 60 μL of the PbI₂/PEAI precursor solution was spin-coated at 3000 rpm for 30 s, followed by annealing at 70 °C for 10 min and cooling to room temperature. The resulting films exhibited a bright yellow color. Subsequently, the as-prepared (PEA)₂PbI₄ films were coated with the MAI solution by spin-coating at 3000 rpm for 30 s. At the 15th second of spin-coating, 150 μL of the corresponding MAI solution was dynamically dripped onto the spinning substrate. The films were then annealed at 100 °C for 20 min. The final films appeared dark red to brown.

4.4. Device Fabrication and Optoelectronic Measurements

The arrays of the ITO (60 nm) BEs were deposited onto flexible PEN substrates at room temperature through a shadow mask by sputtering, using a high-purity ITO target (99.99%) under an Ar atmosphere. The base pressure was 2 × 10^–6 mbar, while the deposition pressure was 4 × 10^–4 mbar, with an Ar flow of 20 sccm. The width of each metallic line was 100 μm, while their pitch and length were 650 μm and 14.5 mm, respectively. Then, Pt NPs with an average diameter of ∼3 nm and an average density of ∼5 × 10¹¹ NPs/cm² were deposited through the gas condensation terminated technique. In this method, NP formation takes place in a small condensation chamber connected to the main chamber through a narrow aperture (∼5 mm, serving as the expansion zone). A DC magnetron source was employed to sputter a high-purity Pt target, generating Pt atoms that enter the high-pressure condensation region, where their mean free path is significantly reduced and rapid thermalization occurs. NPs are subsequently formed as these thermalized atoms move toward the expansion zone. The deposition was carried out under an Ar flow of 60 sccm, with the chamber pressure maintained at 3 × 10^–3 mbar. The DC current was set at 0.1 A. The substrates were subsequently cleaned by sequential sonication in detergent solution (2% Helmanex III in deionized water), deionized water, and isopropanol for 15 min each. Prior to perovskite coating, the cleaned ITO surfaces were exposed to ultraviolet ozone for 30 min and immediately transferred into a N₂ glovebox. The resulting perovskite films exhibited a thickness of approximately 400 nm. To complete the 10 × 10 array structure, first Ag NPS and then 60 nm patterned ITO as a capping layer were deposited through a shadow mask via DC and RF sputtering, respectively. The Ag NPs exhibited an average diameter of ∼5 nm and an average density of ∼3 × 10¹¹ NPs/cm². The deposition was conducted under an Ar flow of 60 sccm, with the chamber pressure maintained at 3 × 10^–3 mbar. The DC current was set at 0.35 A. Electrical characterization was carried out at room temperature using a Keithley 4200 semiconductor parameter analyzer (4200-SCS) in combination with a SUSS MicroTec probe station, and the RGB LED array was driven by an HP 8116A pulse generator. The intensity of the UV beam was measured using the Gentec-EO laser power and energy meter with display (model: MAESTRO). The shadow masks were fabricated on 50 μm-thick aluminum foil. Patterning was carried out using a continuous-wave 1064 nm laser source (II–VI Laser Enterprise) operated at a power of 1 W. The laser beam was focused to a 30 μm spot and scanned across the aluminum surface at a speed of 600 mm/s.

The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/acsami.6c00377.

• GIXRD, XRD, XPS, PL, UPS, and AFM measurements for the perovskite film, TEM measurements and elemental analysis of the Pt and Ag NP arrays, I–V hysteresis patterns of the samples containing bulk silver as TE and the reference sample, device-to-device and cycle-to-cycle variability, variability of the endurance and retention properties, dependence of the device resistance from the total area, geometry of the simulated cell, details of the switching model, simulated profiles during the SET and RESET transitions, transient responses, temperature profiles, details of the segregation model, distribution of the operating electrodes’ resistance under strain conditions, energy diagram configuration for the sample under consideration, comparison in the memory performance of various PEA-based and other material configurations devices that have been reported in the literature (PDF)

The manuscript was written through contributions of all authors. All authors have given approval to the final version of the manuscript.

The open access publishing of this article is financially supported by HEAL-Link.

The authors declare no competing financial interest.