Synthesis of Large‐Area 2D Prussian Blue as Ion‐Transport Channels for Non‐Volatile Memristors

TL;DR

A new method creates large 2D Prussian blue films that work well in memory devices due to their efficient ion transport and structural stability.

Contribution

A scalable synthesis method for high-quality, large-area 2D Prussian blue films with tunable thickness and excellent ion transport properties.

Findings

The 2D Prussian blue films show reliable bipolar electrical switching with high Roff/on ratios and long retention.

Ion transport across 1.5 µm lateral gaps was achieved, much larger than conventional distances.

The synthesis method can be extended to other coordination polymers for various applications.

Abstract

2D Prussian blue and its analogues hold great promise for applications in catalysis, energy conversion, sensing, and memory devices, thanks to their open frameworks, surface activity, and directional ion transport. However, synthesizing high‐quality and large‐area 2D films remains a major challenge. Here, we present a robust and scalable liquid‐liquid interfacial synthesis that enables the formation of continuous, 2D flakes of Prussian blue (Fe3+[Fe2+(CN)6]0.75) with tunable thicknesses from ∼2 nm to several hundred nanometers. The controlled reduction of [Fe3+(CN)6]3− to [Fe2+(CN)6]4− enables slow, directed growth of 2D‐FeFe layers. Unlike films formed from nanoparticles, this method yields high‐quality flakes suitable for integration into devices. As a demonstration, we incorporated these films into Ag filament‐based electrochemical metallization memristors. The 2D‐FeFe devices ≥50 nm thick exhibited reliable bipolar electrical switching, with high Roff/on ratios (∼106), >6 h retention, and stability over 150 cycles. Strikingly, switching was observed across 1.5 µm lateral gaps, far exceeding conventional silver filament formation distances, highlighting the superior ion transport and structural integrity of these 2D frameworks. This scalable approach to 2D Prussian blue, which has the potential to be extended to other related coordination polymers, offers exciting opportunities beyond memristors, enabling integration into technologies where thin‐film compatibility, directional ion transport, and high surface activity are critical, such as catalysis, energy storage, and neuromorphic computing.

A robust liquid‐liquid interfacial synthesis enables continuous, large‐area 2D Prussian blue films with tunable thickness from ∼2 nm to hundreds of nanometers. Controlled reduction guides slow, directional layer growth, yielding high‐quality flakes. These films show exceptional ion transport and device integration, demonstrated in memristors with long‐range switching, high Roff/on ratios, and strong cycling stability.

Figures

(a) Scheme of the synthetic approach for the synthesis of 2D‐FeFe and its deposition on SiO/SiO2 (285 nm). (b) Optical microscope image of the 2D‐FeFe flakes. (c) Representation of the memristive device built with 2D‐FeFe.

Optical microscope images (left) and AFM images (right) of 2D‐FeFe flakes deposited on a Si/SiO2 (285 nm) substrate after 30 to 120 min of reaction.

(a) Raman spectra measured with a 532 nm excitation wavelength of 2D‐FeFe with different reaction times, and bulk FeFe. (b) Fe2p XPS spectra of 2D‐FeFe fished at different times compared to bulk FeFe. (c) percentage of Fe2+ extracted from the Fe2p region as a function of the reaction time for the 2D‐FeFe. (d) Temperature‐dependent magnetization of 2D‐FeFe30 min, 2D‐FeFe120 min, and bulk‐FeFe. The M values of 2D‐FeFe30 min were multiplied by 1500, and the values of 2D‐FeFe120 min for 200, for a more visual comparison with its bulk counterpart.

(a) Scheme of the Ag filament formation/fracture in our device. (b) Bipolar resistive switching measured in the −3 to 3 V voltage range for the 2D‐FeFe60 min. (c) Bipolar resistive switching measured in the −7 to 3 V voltage range for the 2D‐FeFe90 min. (d, e) Double‐logarithmic I‐V curves for the SET process of 2D‐FeFe60 min and 2D‐FeFe90 min, respectively, with data fitted to identify different conduction regimes: (A) Ohmic contact, (B) SCLC, (c) TFL in the HRS, and (D) Ohmic contact in the LRS.

(a) SET and RESET Voltages of 20 measured devices of 2D‐FeFe30 min. (b–d) Variation of the LRS and HRS Resistance with the successive cycles (b), during 450 s measured continuously at 0.2 V (c), and more than 104s measured for 450s every 2 h. at 0.2 V (d). (e) Statistical analysis of VSET and VRESET from 10 different devices displayed in box plots. (f) Box plots depicting the Roff/on across the 10 devices, showing the median value and IQR.

Ven diagram of the Ron/off against the channel length for the representative memristive examples summarized in Table S1.

Full text

Introduction

1

Prussian Blue (PB, Fe_4_[Fe(CN)6]3·zH_2_O) is a cyanide‐bridged coordination polymer of Fe^2^ ^+^/Fe^3^ ^+^ ions linked by CN^−^ ligands. Its open cubic framework provides cavities and ionic channels, while redox‐active and mixed‐valence sites endow it with excellent electronic and ionic transport [1, 2, 3, 4, 5, 6, 7, 8] These features, combined with structural robustness and tunable charge transfer, make PB and its analogues (PBAs, A_x_M[M′(CN)6]ᵧ·nH_2_O; A = alkali cations; M, M′ = transition metals) a highly versatile materials family [8, 9, 10, 11, 12, 13].

Since its discovery in 1707, PB has found diverse uses: first as the world's earliest synthetic pigment [14], later as a medical antidote for thallium and caesium poisoning [15], and as a model in molecular magnetism [3]. Over time, its adaptable framework has enabled applications in fields as varied as catalysis and electrochemical energy storage [16, 17, 18, 19, 20, 21, 22, 23, 24]. Today, this evolution has reached a critical frontier: PB and PBAs are emerging as active components in electronic devices, from switching elements to neuromorphic architectures. Yet, despite their structural and electronic promise, the field faces a fundamental bottleneck—poor processability and limited integration into device‐ready architectures.

Recent work illustrates both the promise and the challenge. Ruthenium‐based PBAs in nanocrystalline films exhibit reversible conductance switching over four orders of magnitude, driven by electrochemically tunable mixed valency and intervalence charge transfer between Ru centers [25]. These devices operate as brain‐like artificial synapses with retention nearly two orders of magnitude higher than organic polymers. However, their nanocrystalline nature imposes significant limitations. Grain boundaries and weak coupling between particles suppress conductivity. Similarly, early demonstrations of PB/PBA memristors—resistive switching devices that store information by changing their resistance state and are considered key building blocks for low‐power memory and neuromorphic computing—confirmed resistive switching and ion‐induced insulator‐to‐metal transitions [26, 27, 28, 29]. Yet device preparation was extremely challenging, as they relied on integrating dispersed nano‐ or microcrystals into architectures that are fundamentally incompatible with practical electronics. This poor structural control and non‐scalable design hindered reproducibility, limited uniform ionic transport, and prevented straightforward device fabrication

In this context, we envisage the preparation of PB as a promising alternative to true 2D materials. Unlike films formed from aggregated particles, 2D materials offer continuous structures with micrometre‐scale lateral dimensions and minimal grain boundaries. These yields enhanced uniformity and more efficient electronic and ionic migration pathways, which are critical factors for stable and high‐performance electronic devices [30, 31]. Nevertheless, synthesizing PB/PBAs in 2D form is non‐trivial. Their inherently 3D framework connectivity resists exfoliation and favors bulk crystallization. Previous attempts have focused on synthesizing Prussian Blue within layered matrices, but these approaches suffer from the inability to isolate the material from the host matrix. Direct solvothermal methods have also been explored [32]; however, they typically yield small nanostructures—more akin to nanoparticles than true 2D platelets[33, 34].

To address this challenge, we employed interfacial synthesis, a technique that has shown great promise for producing 2D coordination frameworks such as MOFs and COFs [35, 36, 37, 38, 39, 40]. In this method, metal and ligand precursors are dissolved in immiscible liquid phases, and the reaction occurs at the interface. This confines growth to a thin, planar region, promoting layer‐by‐layer assembly instead of bulk precipitation. Successful applications of this strategy include 2D coordination polymers [41], such as planar nickel bis(dithiolene) [42], bis(dipyrrinato)zinc(II) [43], or the family of electrically conductive M‐BHT [44]. Despite its success, this approach has rarely been applied to PB and its analogues or other strongly coordinated 3D polymers, largely due to their rapid nucleation kinetics and lack of intrinsic layered structures.

Herein, we overcome these challenges and report a robust liquid–liquid interfacial synthesis for producing high‐quality 2D systems of PB (Fe[Fe(CN)6]0.75, FeFe), and demonstrate its application in Electrochemical Metallization (ECM) based memristive devices. This method enables the fabrication of continuous 2D materials and, for the first time, brings the advantages of interfacial growth to the PB family. The resulting 2D‐FeFe span thicknesses from a few nanometers to several hundred nanometers, representing a significant advancement in the structural control and processability of PB materials. In horizontal ECM devices, FeFe films (∼50 nm thick) exhibited stable and repeatable resistive switching over 150 cycles, with on/off ratios exceeding 10^6^ and excellent memory retention. Remarkably, reliable switching was maintained even across wide electrode gaps of 1.5 µm. Such long‐range ionic migration is enabled by the continuous and uniform nature of the 2D‐FeFe films, in sharp contrast to isolated nanoparticles or microcrystalline films. By minimizing grain boundaries and structural discontinuities, our 2D systems facilitate efficient ion transport over much larger distances than previously attainable in PB‐based materials.

Experimental Section

2

Materials and Methods

2.1

All chemical reagents were purchased and used without further purification. Tetrabutylammonium bromide (TBA‐Br), Iron (III) Nitrate, and potassium hexacyanoferrate were purchased from Sigma‐Aldrich. Ultrapure water (18.2 MΩ), and Chloroform were used in the following syntheses.

‐Synthesis of TBA_3_[Fe(CN)]6: TBA_3_[Fe(CN)]6 was prepared by following a literature procedure. It is based on the cationic exchange that was achieved by mixing 20 mL of an aqueous solution 0.1 M of K_3_[Fe(CN)6] with 20 mL of a 0.1 M aqueous solution of TBA‐Br. After 24 h under stirring, 40 mL of CHCl_3_ was added and stirred with the aqueous phase for 1 h. After the extraction of the CHCl_3_, it was left to crystallize for one week, obtaining yellow crystals of TBA_3_[Fe(CN)]6. The TBA exchange was demonstrated by infrared spectroscopy, showing the presence of the TBA vibrational modes below 2000 cm^−1^, Figure S1.

‐Synthesis of FeFe PB bulk: The bulk reference sample of the FeFe was synthesized at room temperature by adding simultaneously two aqueous solutions: (i) 40 mL of FeNO_3_ (10 mM) and (ii) 40 mL of K_3_[Fe(CN)6] (10 mM). Then, the mixtures were stirred for 30 min before being centrifuged at 11000 rpm for 20 min. The final solution was centrifuged at 13000 rpm for 15 min thrice (in water).

Synthesis of FeFe 2D sheets: The FeFe 2D sheets were synthesized by gently adding 2 mL of a 30 mM aqueous solution of Fe(NO_3_)3 over a 30 mM TBA_3_[Fe(CN)6]solution in CHCl_3_. After 30 min an orange/blue film is formed at the interphase, which is fished with a Si/SiO_2_ (285 nm), substrate previously washed with basic piranha three times, and twice with water. After the fishing, the substrate is immersed in water for 2 min, this process is repeated twice to ensure the complete removal of unreacted precursors. The fishing was also done after letting the reaction occur for 1 and 1 h 30 min.

Physical Characterization

2.2

Optical microscope images were conducted on a Zeiss Axio A1m imager using white light illumination at normal incidence with a 20, 50, or 100× objective (0.9NA). Scanning Electron Microscopy and Energy Dispersive Spectroscopy studies were carried out on a Hitachi S4800 microscope. Transmission Electron Microscopy studies were carried out on a Jeol 300 arm microscope operating at 80 kV, and a Technai G2 F20 microscope operating at 200 kV. Samples were prepared by dropping suspensions on lacey formvar/carbon copper grids (300 mesh). Raman spectroscopic characterization was carried out on a Horiba LabRAM HR Evolution confocal Raman microscope. The measurements were conducted in a backscattering configuration excited with a solid‐state green laser (λexc = 532 nm) that was focused using a 100× objective (0.8 NA), thus leading to a laser spot with a diameter of ca. 1 µm. Power X‐Ray Diffraction (PXRD) patterns were collected in a PANalytical Empyrean diffractometer using copper radiation (Cu K α λ = 1.5418 Å) with a PIXcel detector, operating at 40 mA and 45 kV at room temperature. X‐ray Photoelectron Spectroscopy (XPS) measurements were done with a KALPHA Thermo Scientific spectrometer. All spectra were collected using Al Kα radiation (1486.6 eV), monochromatized by a twin crystal monochromator, yielding a focused X‐ray spot (elliptical in shape with a major axis length of 200 µm) at 30 mA and 2 kV. The alpha hemispherical analyzer was operated in the constant energy mode with survey scan pass energies of 200 eV to measure the whole energy band and 50 eV in a narrow scan to selectively measure the particular elements. XPS data were analyzed with Avantage software. A background function was used to approximate the experimental backgrounds. Charge compensation was achieved with the system flood gun that provides low‐energy electrons and low‐energy argon ions from a single source. Magnetic Measurements: Magnetic data were collected with a Quantum Design MPMS XL‐5 susceptometer. DC FC magnetization measurements were performed under an applied magnetic field of 100 Oe. The isothermal Magnetization at 2 K was performed between −7 and +7 T. The I‐V traces were measured using a Keithley 2450 Source Meter. It acts both as the voltage source and the current sensing unit. The voltage source has a delay of 0.1 s for it to stabilise. I‐V traces were measured at a voltage ramping rate of both 10 and 20 mV/s. 20 mV/s was found to give the best memristive responses. A compliance current of 10 µA was applied to avoid shorting the devices for the 2D‐FeFe90 min devices, due to the high voltage required for the filament formation, while no compliance current was used for the 2D‐FeFe60 min. The equipment and measurements were controlled using the QCoDeS package with Python.

Results and Discussion

3

Synthesis of Bulk

3.1

To understand the structure of the bulk materials, we first synthesized them by mixing the cyanide precursors TBA_3_[Fe(CN)6] with Fe(NO_3_)3 in water. This reaction produced a blue solid identified as Fe[Fe(CN)6]0.75 (FeFe). In the system, the coordination of Fe by cyanide is confirmed by infrared spectroscopy, through the splitting of the ∼2100 cm^−1^ ν(C≡N) band of the uncoordinated [Fe^III^(CN)6]^3−^ into two bands at 2060 and 2170 cm^−1^, indicating the formation of Fe^II^(CN)‐Fe^III^ and Fe^III^(CN)‐Fe^III^ species(Figure S1) [45]. The presence of Fe^2^ ^+^ is attributed to an intrinsic autoreduction process mediated by cyanide ligands in the presence of water [46, 47]. The detection of NH_4_ ^+^, a characteristic byproduct of cyanide ligand oxidation during autoreduction, further corroborated this mechanism. NH_4_ ^+^ formation was confirmed and quantified using a modified indophenol blue method, which revealed a distinct absorption band after 24 h, corresponding to NH_4_ ^+^ production (Figure S2). Structurally, FeFe consists of nanocrystalline cubes with a size of hundreds of nanometers (Figure S3). The FeFe compound exhibits the expected cubic crystalline structure, with cell parameter a = 10.1993(2) Å, as obtained from Pawley refinement of the PXRD data (space group Pm‐3m, Rwp = 1.73%, GoF = 2.7, Figure S4).

Synthesis of 2D Films

3.2

To synthesize the compounds as 2D systems, we confined their growth along the z‐axis using a liquid/liquid interphase method. TBA_3_[Fe(CN)6] was dissolved in CHCl_3_, while Fe(NO_3_)3 was dissolved in water. The aqueous Fe^3+^ solution was then carefully layered on top of the organic phase, enabling coordination to occur at the interface (Figure 1a).

(a) Scheme of the synthetic approach for the synthesis of 2D‐FeFe and its deposition on SiO/SiO2 (285 nm). (b) Optical microscope image of the 2D‐FeFe flakes. (c) Representation of the memristive device built with 2D‐FeFe.

To optimize 2D film formation at the liquid/liquid interface, different precursor concentrations (10, 30, 50, and 100 mM) of TBA_3_[Fe(CN)6] and Fe(NO_3_)3 were systematically evaluated. A visible blue film appeared after approximately 30 min at 30 mM, after ∼20 min at 50 mM, and only after ∼6 h at 10 mM, indicating a non‐linear dependence of nucleation time on precursor concentration. At 100 mM, however, the reaction became non‐selective, leading to bulk precipitation rather than confined interfacial film formation.

To assess morphology and verify interfacial confinement, the films were directly transferred onto TEM grids. As shown in Figure S5a, at 30 mM after 30 min of reaction, thin, low‐contrast flakes are observed, indicative of nanoscale thickness. At 50 mM and also 30 min, the films exhibit higher contrast, thus, increased thickness, with improved coverage but reduced uniformity (Figure S5b). At 100 mM, uncontrolled homogeneous nucleation predominates, resulting in the aggregation of Prussian Blue nanoparticles rather than well‐defined 2D sheets (Figure S5c). In contrast, the films obtained at 10 mM after 6 h display minimal TEM contrast, suggesting ultrathin morphology, which was further confirmed by Atomic force microscopy (AFM) to have thicknesses in the range of 2–5 nm with micrometer‐scale lateral dimensions (Figure S6). Among all tested conditions, the 30 mM precursor concentration provided the best balance between nucleation time, interfacial selectivity, and film uniformity, resulting in reproducible, large‐area continuous films suitable for transfer and further characterization. As shown in Figure S7, the interfacial film appears continuous across the entire contact region, indicating the potential for ultralarge‐area coverage with optimized deposition protocols. Consequently, we focused our study on the 30 mM precursor concentrations. Moreover, as demonstrated in Figure S8, the film thickness could be further tuned by extending the reaction time (60–120 min), offering a level of morphological control that is not achievable at higher precursor concentrations.

The reaction time is essential for the formation of 2D‐FeFe because it relies on the reduction of [Fe^3^ ^+^(CN)6]^3^ ^−^ to [Fe^2^ ^+^(CN)6]^4^ ^−^—a relatively slow process, also observed in bulk synthesis. As previously discussed, Prussian Brown, which corresponds to the fully oxidized Fe^III–^Fe^III^ form, is thermodynamically unstable in aqueous environments. It undergoes gradual partial reduction to generate Fe^2^ ^+^ species, leading to its transformation into Prussian Green/Blue. This autoreduction mechanism is our driving force for controlling the reaction kinetics and forming our 2D‐films, while using Fe^2^ ^+^ directly results in rapid precipitation of Prussian blue microcrystals rather than the formation of 2D sheets (Figure S9). After 60 and 90 min of reaction, 2D‐FeFe forms large sheets of the coordination polymer that continue to thicken over time, as indicated by increasing contrast in TEM images (Figure S8).

The large lateral dimensions of the flakes were confirmed by scanning electron microscopy (SEM) of the 2D‐FeFe deposited on Si/SiO_2_ (285 nm) substrates at different reaction times (Figures S10). 2D‐FeFe flakes spanning hundreds of micrometres were observed after 30 min of reaction, with lateral size increasing over time. Energy‐dispersive X‐ray spectroscopy (EDS) confirmed the presence of Fe at the flakes' surface (Figures S10).

From this point forward, films composed of the FeFe compound collected after 30, 60, 90, and 120 min of reaction will be referred to as 2D‐FeFe_30 min_ to 2D‐FeFe_120 min_, respectively. The evolution of flake morphology with reaction time was further examined using optical microscopy, which provided complementary information to the SEM analysis. As shown in Figure 2, Optical images of 2D‐FeFe_30 min‐120 min_ revealed large flakes with lateral dimensions extending several hundred micrometres, consistent with SEM observations (for more images, see Figures S11–S13). In this context, SEM and optical microscopy images show that after 30 min of reaction, the 2D‐FeFe film collected from the liquid/liquid interface consists of fractured flakes several microns in size. In contrast, extending the reaction time beyond 60 min produces a more continuous film spanning the millimetre scale.

Optical microscope images (left) and AFM images (right) of 2D‐FeFe flakes deposited on a Si/SiO2 (285 nm) substrate after 30 to 120 min of reaction.

Furthermore, a progressive increase in optical contrast across the series also suggested an increase in the flake thickness. To quantify this, we performed AFM measurements on the flakes (Figure 2; Figures S14–S17). Specifically, the measured thicknesses were approximately 13 ± 9 nm for 2D‐FeFe_30 min_, 43.5 ± 24 nm for 2D‐FeFe_60 min_, 76 ± 36 nm for 2D‐FeFe_90 min_, and 160 ± 70 nm for 2D‐FeFe_120 min_. These measurements confirm that the thickness of the 2D films increases with reaction time, following an almost linear trend (Figure S18), which indicates that the crystal growth is predominantly interface‐reaction‐controlled. Therefore, the data were fitted to the linear growth model: h (t) =  k × (t − *t_ind_ *), Where h(t) is the film thickness, k is the growth rate, and t_ind_ induction (nucleation) time. The fitting yields a k approximately 1.2 nm/min, consistent with a linear growth regime. The negative extrapolated intercept indicates the presence of an induction/nucleation period before measurable film growth begins (*kt_ind_

• = −24.2). By setting h(t) = 0 in the fitted equation, we estimate that nucleation requires approximately 20 min, which is in good agreement with the 30 min that are visually required to observe a film formed at the liquid/liquid interphase. After this induction time, film growth proceeds linearly with time. It must be noted that the values obtained from the fitting carry large uncertainty due to variability in AFM thickness measurements.

Chemical and Structural Characterization

3.3

We explore the structure and chemical composition of the different flakes by Raman spectroscopy, X‐ray photoemission spectroscopy (XPS), Surface X‐ray diffraction (SXRD), and TEM electron diffraction (ED). Raman spectroscopy is commonly used for characterizing cyanide‐based compounds, thanks to the characteristic cyanide stretching mode above 2000 cm^−1^, which provides information on the coordination environment of the cyanide molecule. As seen in Figure 3a, the cyanide precursors Bulk‐FeFe exhibit a prominent single band located at 2157 cm^−1^, corresponding to the 1A_g_ν(CN) stretching vibration band characteristic of PB [14, 48], which appears significantly shifted compared to the iron cyanide precursor (Figure S19) [49]. The 2D‐FeFe flakes consistently exhibit a distinct band at 2157 cm^−^ ^1^, regardless of their thickness, Figure 2a. Furthermore, the 2D‐FeFe flakes demonstrate remarkable long‐term stability. The films preserved both their morphology (Figure S20) and characteristic Raman fingerprint after one year under ambient conditions (Figure S21), underscoring their robustness and suitability for device integration without encapsulation. In contrast, their response to thermal stress reveals that when the same 2D flakes were heated in air at 150°C for 2 h, the characteristic A_9_ ν(CN) stretching band disappeared and was replaced by a clear photoluminescence signal (Figure S21), indicative of degradation of the cyanide‐bridged framework. Although bulk TGA suggests apparent stability within this temperature range (Figure S22), this transformation can be attributed to the loss of coordinated water molecules. In the bulk material, water plays a crucial role in completing the coordination sphere of otherwise unsaturated iron centers originating from intrinsic CN defects and Fe–CN vacancies. In the 2D films, these defects are even more prominent due to their high surface‐to‐volume ratio, making the surface‐coordinated water molecules particularly important for maintaining structural integrity. Upon dehydration, the removal of these stabilizing water ligands destabilizes the coordination environment—an effect well documented for vacancy‐rich Prussian Blue analogues—leading to partial framework collapse and facilitating oxidative decomposition into iron oxides upon continued heating [50].

(a) Raman spectra measured with a 532 nm excitation wavelength of 2D‐FeFe with different reaction times, and bulk FeFe. (b) Fe2p XPS spectra of 2D‐FeFe fished at different times compared to bulk FeFe. (c) percentage of Fe2+ extracted from the Fe2p region as a function of the reaction time for the 2D‐FeFe. (d) Temperature‐dependent magnetization of 2D‐FeFe30 min, 2D‐FeFe120 min, and bulk‐FeFe. The M values of 2D‐FeFe30 min were multiplied by 1500, and the values of 2D‐FeFe120 min for 200, for a more visual comparison with its bulk counterpart.

The oxidation states of the 2D systems, particularly 2D‐FeFe, as a function of reaction time, were investigated by X‐ray photoelectron spectroscopy (XPS). As shown in Figure 2b, all 2D‐FeFe samples exhibit clear signals for iron in two oxidation states: Fe^2^ ^+^ at approximately 709 eV (purple band) and Fe^3+^ at around 711 eV, along with its characteristic satellite features (orange band) [51, 52, 53]. Notably, the proportion of Fe^2+^ increases with reaction time, confirming the progressive in situ reduction of Fe^3+^ centers and the gradual formation of the conventional FeFe Prussian blue structure. In the N 1s region, the 2D systems exhibit the same N species. On the one hand, a strong peak at ∼398 eV, which is associated with cyanide nitrogen (Figure S23). On the other hand, two additional peaks, at ∼400 and ∼402 eV, are attributed to nitrogen from residual TBA and nitrate species originating from the precursors.

SXRD was used to corroborate that the crystalline structure of the 2D‐Flakes matches their bulk counterparts. In this regard, the measurements were performed on 2D flakes deposited on ITO substrates, which offer better signal clarity than SiO_2_ due to their transparency and lower crystallinity. A clear (002) reflection was observed at 17.4° for 2D‐FeFe_30 min‐120 min_, corresponding to a d‐spacing of ∼0.51 nm, consistent with bulk materials (Figure S24). No other peaks were detected, likely due to their low intensity. For the thinnest sample 2D‐FeFe_30 min_ SXRD signals were too weak. Instead, we used TEM‐Electron diffraction (ED) to assess their crystallinity. ED patterns collected from individual flakes confirmed the presence of the (002) diffraction spot with a d‐spacing of ∼0.51 nm (Figure S25), consistent with the expected cubic structure of Prussian blue in an isolated flake. Furthermore, high‐resolution TEM was performed on the 2D‐FeFe_30 min_, 2D‐FeFe_60 min_, and 2D‐FeFe_90 min_ samples, revealing crystalline planes with interplanar spacings consistent with the ∼0.51 nm expected for Prussian Blue and its analogues (Figure S26). Notably, in the 2D‐FeFe_30 min_ sample, we were able to resolve the characteristic square arrangement associated with the cubic PB framework. In contrast, the thicker 2D‐FeFe_60 min_ and 2D‐FeFe_90 min_ samples show only linear periodic features because their greater and uneven thickness prevents precise low‐dose pre‐alignment before the films’ amorphization.

Finally, given the well‐known magnetic properties of Prussian blue analogues, we investigated the magnetic behavior of 2D‐FeFe using SQUID magnetometry. Measurements were performed on flakes deposited on Si/SiO_2_ (285 nm) substrates and compared to their bulk counterparts. As shown in Figure 3e, 2D‐FeFe exhibits a clear ferromagnetic response, similar to bulk FeFe PB. However, the Curie temperature (Tc) of the 2D material is significantly lower. In 2D‐FeFe, magnetization increases gradually with decreasing temperature and shows a sharp rise below 5 K, marking the onset of ferromagnetic ordering.

This size‐dependent suppression of magnetic ordering is also reflected in isothermal magnetization curves (Figure S27). While bulk FeFe exhibits a small hysteresis of 40 Oe, no hysteresis is observed in 2D‐FeFe, even in flakes as thick as 100 nm, as is the case with 2DFeFe_120 min_. Interestingly, although bulk FeFe consists of nanoparticles 100–200 nm in size, comparable to the flake thickness in 2DFeFe_120 min_, their magnetic behavior differs. This discrepancy is attributed to the amount of Fe^2+^ present in the 2D films, which is lower than its bulk counterpart. As shown in Figure 3c, bulk FeFe contains significantly more Fe^2+^ than 2D‐FeFe, which enhances the ferromagnetic pathways and makes the magnetic interactions stronger [3]. This intrinsic ferromagnetism, even with a low Tc, is particularly appealing for multifunctional device applications. In emerging spintronic memristors, ferromagnetic ordering and magnetic‐field‐dependent transport can be exploited to offer an additional degree of control over resistive states, enabling magnetically tunable synaptic weights and hybrid memristive–spintronic functionality [54, 55].

Electrical Response and Device Fabrication

3.4

After confirming the successful synthesis of producing high‐quality 2D‐FeFe, we integrated them into the ECM memristive. The devices use an asymmetric Ag/2D‐PB/Au structure, Figure 4a. We fabricated the devices using a fingerprint‐like layout with a central Ag electrode surrounded by six Au electrodes placed at distances of 1.5, 3.5, and 5.5 µm (Figure S28a,b). This setup works well for ECM devices because the Ag electrode can be oxidized and reduced when a voltage is applied. During voltage cycling, Ag^+^ ions are created at the Ag electrode and move through the porous 2D‐PB layer next to it and are reduced at the Au electrode when the set voltage (Vset) is applied. In this scenario, both electrodes are connected, giving rise to the low resistance state (LRS), Figure 4a. Applying a reset voltage (Vreset), the Ag filament is re‐oxidized, breaking it and returning the initial high resistance state (HRS). These threshold voltages correspond to the ‘SET’ (write) and ‘RESET’ (erase), similar to digital memory. In this context, we mainly focus on films thicker than 10 nm (e.g., 2D‐FeFe_30 min‐90 min_) because they offer better continuity, cover larger areas, and allow more working electrodes per chip (Figure S28c,d).

(a) Scheme of the Ag filament formation/fracture in our device. (b) Bipolar resistive switching measured in the −3 to 3 V voltage range for the 2D‐FeFe60 min. (c) Bipolar resistive switching measured in the −7 to 3 V voltage range for the 2D‐FeFe90 min. (d, e) Double‐logarithmic I‐V curves for the SET process of 2D‐FeFe60 min and 2D‐FeFe90 min, respectively, with data fitted to identify different conduction regimes: (A) Ohmic contact, (B) SCLC, (c) TFL in the HRS, and (D) Ohmic contact in the LRS.

To evaluate device performance, Ag/2D‐PB/Au structures were tested through current‐voltage (I‐V) switching cycles and current‐time measurements at fixed voltage to assess memory behavior and stability. Nevertheless, we tested a device made with the thinner films 2D‐FeFe_30 min_ to corroborate that thicker and more continuous films are preferable, observing that 2D‐FeFe_30 min_ did not show the expected bipolar switching (Figures S29). This may be attributed to two main reasons: (1) As they are isolated flakes, to get a flake properly contacting both electrodes is challenging; (2) due to structural breakdown during Ag^+^ migration in such thin flakes. Similar to the effect of cation inclusion, such as K^+^ induces lattice distortions blocking charge transfer effects in other PBAs [4, 12]. In contrast, devices using thicker flakes 2D‐FeFe_60 min_, and 2D‐FeFe_90 min_ exhibited clear bipolar resistive switching when measured across 1.5 µm gaps. However, devices with larger electrode gaps of 1.8 µm and 2.4 µm did not show bipolar switching, likely because Ag^+^ migration was insufficient over the increased distance (Figure S30). Both the Ag/2D‐FeFe_60 min_/Au and Ag/2D‐FeFe_90 min_/Au devices exhibit the expected bipolar resistive switching characteristic of ECM memristors, Figure 4b,c. Note that the electrical switching was perfectly maintained during several cycles for both systems (tested up to 150 cycles for the thinner 2D‐FeFe_60 min_, Figure S31). However, a significant difference was observed in the SET voltage (Vset) required to trigger the low‐resistance state.

Specifically, the Ag/2D‐FeFe_60 min_/Au device consistently switches at lower voltages (Vset = ca. 1.5 V), whereas the Ag/2D‐FeFe_90 min_/Au device requires much higher voltages, with the first SET event occurring near −5 V. This discrepancy was unexpected, as the only structural difference between the two devices is the thickness of the 2D‐FeFe layer. In this context, the flake thickness is not expected to play a significant role in the switching behaviour. In contrast, the cyanide defects and vacancies in the thicker film may play a major role in the higher Vset observed. To further investigate the switching behavior, we analyzed the conduction mechanisms in the LRS and HRS by fitting the first SET I–V curves on a double‐logarithmic scale. As shown in Figure 4d,e, both devices exhibit an ohmic region in the LRS (Region A) with a slope close to 1, indicating metallic conduction through a continuous Ag filament, consistent with ECM‐type memristors [56, 57]. In the HRS, conduction differs before filament formation. For Ag/2D‐FeFe_60 min_/Au, the curve shows a single region with a slope of ∼1.44, characteristic of trap‐controlled space‐charge‐limited conduction (SCLC). This suggests partial trapping of carriers in shallow states within the 2D layer [58]. In contrast, Ag/2D‐FeFe_90 min_/Au displays three HRS regimes: an initial near‐ohmic region (slope ∼0.9, Region B), likely due to thermally generated carriers; a second region (slope ∼1.7, Region C) consistent with trap‐controlled SCLC; and a final trap‐filled limit (TFL) regime (slope ∼6, Region D), where all traps are filled and current rises sharply. Only then can Ag^+^ ions move freely and form the conductive filament needed for the SET transition. This complex, trap‐dominated conduction in the thicker 2D‐FeFe_90 min_ film suggests a higher density of trapping centers, which increases the activation energy required for filament nucleation. These trap sites arise from cyanide defects, generated mainly by the autoreduction mechanism ([Fe^III^(CN)6]^3^ ^−^ → [Fe^II^(CN)5]^3^ ^−^, CN^−^ loss) and intrinsic vacancy compensation in Prussian Blue structures [59]. These defects and vacancies act as trap sites and reduce the availability of CN^−^ groups for Ag^+^ coordination and migration [22, 60]. As a consequence, Ag^+^ ions require higher voltages to nucleate and grow conductive filaments. However, it must be noticed that other microstructural factors may also contribute.

Therefore, we focused on the Ag/2D‐FeFe_60 min_/Au device, as the most promising candidate. Upon applying a complete voltage sweep (−2 V → 2 V → −2 V), the device transitioned from HRS (∼10^1^ ^1^ Ω) to LRS (∼10^5^ Ω), corresponding to the set process. A subsequent positive sweep restored the HRS at around 1.5 V, completing the reset operation (Figure 4b). As shown in the histogram in Figure 5a, the average Vset and Vreset after 150 cycles were −1.4 ± 0.7 V and 1.1 ± 0.9 V, respectively, demonstrating its good stability. The device also demonstrated excellent data retention, exhibiting Roff/on ratio of ∼10^6^, perfectly stable when measured at −0.2 V with successive cycles and over time in both conductive states, Figure 5b,c, respectively.

(a) SET and RESET Voltages of 20 measured devices of 2D‐FeFe30 min. (b–d) Variation of the LRS and HRS Resistance with the successive cycles (b), during 450 s measured continuously at 0.2 V (c), and more than 104s measured for 450s every 2 h. at 0.2 V (d). (e) Statistical analysis of VSET and VRESET from 10 different devices displayed in box plots. (f) Box plots depicting the Roff/on across the 10 devices, showing the median value and IQR.

Furthermore, we examined the device's behavior after a period of idle (0 V), followed by remeasurement at −0.2 V, to confirm the non‐volatility of the HRS and LRS. As shown in Figure 5d, both states remained stable for over 2 × 10^4^ s (∼6 h), underscoring the device's potential for non‐volatile memory applications [61, 62]. To assess device‐to‐device uniformity, I‐V measurements were performed on ten randomly selected fingerprinted electrodes from the same device. This demonstrates that, thanks to the large coverage of the transferred 2D‐FeFe_60 min_ layer, it is feasible to obtain multiple functional devices straightforwardly. As shown in Figure 5e, all devices consistently exhibit bipolar resistive switching, with Vset and Vreset values falling within a similar range. Furthermore, the ten devices display well‐differentiated LRS and HRS states (Figure S30d), achieving a median Ron/off ratio of ∼10^5^, Figure 5f. In addition, we demonstrated that the device can operate in both digital and analog modes. By applying a single rectangular pulse of −3 V for 0.1 s, the device switches completely from the HRS to the LRS, corresponding to a binary “0” to “1” transition, suitable for digital memory applications. In contrast, when applying multiple lower‐amplitude pulses (e.g., −2 to −1 V), the device can reach intermediate resistance states between the HRS and LRS (Figure S32a–c), enabling gradual, tunable conductance changes that are ideal for analog computing or neuromorphic applications.

Notably, the Ag/2D‐film/Au memristors reported in this work exhibit competitive performance metrics—including R_off/on_ ratios, endurance, and retention—comparable to some of the best‐performing devices based on porous materials, perovskites, oxides, and conventional 2D materials (Table S1). A key distinction, however, lies in the device architecture. While most memristors rely on nanoscale electrode gaps (typically tens to hundreds of nanometers), our devices achieve stable resistive switching across a substantially larger lateral gap of 1.5 µm. This is particularly significant because longer ion‐migration paths usually introduce higher mechanical stress and instability, often hindering reliable switching, as evidenced in Figure 6. The fact that the 2D‐FeFe films maintain consistent switching under these conditions, at distances much larger than those used in conventional oxide or perovskite memristors or electrochemically grown PB/PBA films, indicates that the 2D‐FeFe framework effectively supports and stabilizes ionic transport over micrometer‐scale paths.

Ven diagram of the Ron/off against the channel length for the representative memristive examples summarized in Table S1.

Conclusions

4

In summary, we report the first synthesis of a true 2D system of Fe[Fe(CN)6]0.75 using a liquid/liquid interfacial strategy. By confining crystal growth at the interface, we produced continuous FeFe layers. The formation of 2D‐FeFe requires controlled, slower kinetics to avoid rapid precipitation of FeFe microcrystals. This was achieved by combining Fe(NO_3_)3 and [Fe^3+^(CN)6]^3−^, with 2D‐FeFe forming via spontaneous reduction of [Fe^3+^(CN)6]^3−^ to [Fe^2+^(CN)5]^3−^ in air. Remarkably, film thicknesses are tunable from ∼2 nm for 2D‐FeFe up to several hundred nanometers by adjusting the reaction time and precursor concentrations. Unlike conventional polycrystalline films, this method yields large‐area, high‐quality 2D flakes suitable for device integration.

In this context, we demonstrated their application in ECM memristors, where films thicker than 50 nm exhibited clear bipolar electrical switching driven by reversible Ag filament formation. Notably, the rigid 2D‐FeFe crystalline framework enabled superior performance, with high Roff/on ratios (∼10^6^), stable operation over multiple cycles, and retention exceeding 6 h. Importantly, reliable switching was achieved across lateral gaps as large as 1.5 µm, significantly larger than the conventional tens to hundreds of nanometers, highlighting the material's superior ability to support long‐range ion transport compared to nanoparticle‐based films. Furthermore, unlike common 2D materials such as MoS_2_, graphene, or h‐BN, the intrinsic cavity network of 2D‐FeFe effectively guides Ag‐ion migration, which markedly enhances the R_off/on_ ratio even at extended channel lengths.

Beyond memristors, this scalable approach to 2D PB frameworks, which can be extended to PBA and potentially to metal organic frameworks via slowing reaction kinetics, opens new avenues for integration into applications where surface activity, directional transport, and thin‐film compatibility are essential, such as catalysis, energy conversion, and chemical sensing.

Author Contributions

R.T‐C. conceived the work, developed the synthetic protocol, carried out most of the synthesis, characterization, and electrical measurements, and was responsible for data analysis and manuscript drafting. H.C. designed the electrical devices. J.C‐G. performed the crystallographic characterization. D.H. and M.A‐A. assisted with the optimization of synthetic conditions. C‐S.A. conducted the TEM‐ED characterization of the flakes. G.F. and G.G. contributed to the electrical characterization of the flakes. E.C. and L.B. contributed to manuscript writing.

Conflicts of Interest

The authors declare no conflicts of interest.

Supporting information

Supporting File: smll72407‐sup‐0001‐SuppMat.docx.