Nanofluidic sensing inspired by the anomalous water dynamics in electrical angstrom-scale channels

Nanoconfined water has unique physical and chemical properties, often leading to the discovery of unexpected phenomena^1,2,3,4,5, which play key roles in various applications such as sensing⁶, filtration⁷, energy conversion⁸, and catalysis^9,10. Compared to the conventional control of confined water transport by varying the pH, temperature, or ionic species^3,11, which suffers from low stability and poor controllability, the emerging electrical control strategy overcomes these drawbacks and attracted extensive attention^12,13. Electrically controlled confined water has shown precise manipulation of flow rate in graphene oxide membrane and transistor-like electrohydrodynamic effect in hexagonal boron nitride (hBN)^12,13, respectively. Although such reports attracted extensive attention, a deep mechanistic consensus is still lacking because the confined water phenomenon under voltage seems distinct from channel to channel. Despite the theoretical simulations can be employed to understand similar processes, the obtained results are often conflicting^14,15,16. Therefore, further efforts are necessary to enrich the current understanding and then to better support the applications.

In 2018, an interesting electrically controlled water permeation was reported in graphene oxide membranes in which the confined water permeation rate was decreased following the increase of voltage until complete blocking at ~2.0 V¹². Such breakthrough discovery is thought of opening an avenue for developing smart membrane technologies. However, two remaining challenges need to be addressed: (1) to realize the precise and fast switch between speeding up and slowing down the water permeation rate by voltage control in a membrane. (2) to make the effective voltage-control range lower than the theoretical water-splitting voltage (1.23 V)¹⁷. To address the above challenges, we proposed tuning the confined water flow dynamics through the combination of ionic and electric factors.

2D transition metal carbides (MXenes) are potential candidates for studying confined water in nanospaces due to their unique layered structure, excellent electrical conductivity, and hydrophilicity^{18,19,20,21,22}. Since the original channel spacing of multilayer MXenes is smaller than the dynamic diameter of water molecules, ion intercalations are often used to enlarge their interspacing for applications^23,24,25. Here, based on the most widely reported MXene (Ti₃C₂), we reported an anomalous permeation performance of electronically controlled water on the metal cation-intercalated Ti₃C₂ membranes, achieving precise and fast switch between speeding up and slowing down the water permeation rate in a low-voltage range. A maximum water permeation rate was obtained at 0.9 V. In 0~0.9 V, the water permeation rate increased following the rise of voltage, which can be attributed to the orderly denser arrangement of confined water caused by low currents that induced the polarization of water molecules. Subsequently, the reduced free energy barrier of metal cations weakened the suppression of current, leading to a sudden change in current and a sharp decrease in permeability at ~0.9 V. The reason is that the high current drove the dissociation and agglomeration of confined hydrolysis into macromolecular clusters in 2D channels, hindering the flow of water molecules. In situ Fourier transform infrared spectroscopy (FT-IR) further revealed the equilibrium transformation of confined water between three structural types: monomer ⇋ dimer ⇋ cluster. This work provides a distinct way to precisely manipulate nanofluids.

Design and fabrication of Ti₃C₂ MXene-based permeation membranes

Figure 1A briefly shows the process of electrically controlled water permeation in the device: water flows through the membrane device with the aid of its own gravity and external pressure. Subsequently, the structure and dynamics of the confined water are manipulated by an external power source to regulate the overall permeation rate in a regular manner. Our device consists of a Ti₃C₂ MXene permeation membrane (0.5 mm in thickness, 38 mm in diameter) and bilateral porous Au electrodes (Fig. 1B). The membrane was fabricated by assembling a porous gold film with multilayer MXene layer by vacuum-assisted filtration. Such thin gold film is porous, which allows water to flow through naturally. A pump and an electrochemical workstation were employed for regulating the internal pressure field and electric field environment during tests (Supplementary Fig. 1). More details on membrane fabrication and permeation measurements are shown in the “Methods” section.

A 3D schematic illustration of an electrically controlled water permeation device. B Photographic images of the top and side view of Au-MXene-Au permeation membrane. The thickness and diameter of the MXene permeation layer is approximately 0.5 and 38 mm, respectively. Pink and blue regions represent respectively MXene membrane and Au electrode. C OM and SEM images show the top morphology of the MXene membrane. High-resolution HAADF-STEM and corresponding fast Fourier transform (FFT) images of multilayer Ti₃C₂ (D) and K-Ti₃C₂ (E). Scale bars: 10 nm⁻¹ in inserted FFT images in (D) and (E).

Full size image

To obtain natural 2D nanochannels, layered Ti₃C₂ was synthesized by selectively etching away the “Al” layer from the parent phase Ti₃AlC₂^26,27,28. However, the X-ray Diffraction (XRD) patterns showed that the layer spacing of Ti₃C₂ is ~9.57 Å (Supplementary Fig. 2), indicating the effective channel is only 1.98 Å which is smaller than the diameter of water molecule (~2.65 Å). This narrow 2D channel is difficult for water molecules to enter and will impede the movement of water molecules. Therefore, metal cations (K⁺ and Li⁺) were adopted as spacers for further separating adjacent Ti₃C₂ nanosheets. The disordered structure and morphology of MXene membrane were characterized by an optical microscope (OM), scanning electron microscope (SEM) and high-angle annular dark field scanning transmission electron microscope (HAADF-STEM), as shown in Fig. 1C, D. High-resolution HAADF-STEM images and Bragg’s formula demonstrated that the layer spacings of K⁺ and Li⁺ intercalated Ti₃C₂ (K-Ti₃C₂ and Li-Ti₃C₂) were expand to ~12.28 and ~12.24 Å, respectively (Fig. 1E, Supplementary Fig. 2 and Fig. 3).

Electrical water permeations of membranes

To evaluate the effect of an electric field on the water movement behavior, we measured the water permeability of membranes using the typical gravimetric method. Each measurement was maintained for over 6 h to ensure the accuracy of the data. In this device, the water permeates through by two pathways: the internal 2D channels and localized voids between the MXene particles. For the Ti₃C₂, the flow mainly comes from the voids between the particles, whereas the metal cation-intercalated Ti₃C₂ has both operating modes. Figure 2A shows the permeation rate of water per unit time for various membranes with no external pressure. For each independent experiment, we applied different voltages on membranes and used a precision electronic balance to record the increase in water weight. At 0 V, the permeation rates were 3.01, 4.53, and 4.98 g h⁻¹ cm⁻² for Ti₃C₂, K-Ti₃C₂, and Li-Ti₃C₂, respectively. Considering the effective channel size, the relatively faster water permeation rates for K-Ti₃C₂ and Li-Ti₃C₂ are reasonable. When the voltage was applied within 0~0.9 V, the water permeation rates in the three membranes increased with different slopes (K-Ti₃C₂≈Li-Ti₃C₂ » Ti₃C₂) following the voltage rise. Such phenomena are totally distinct from the previous report case on graphene oxide in which the permeation rate was continuously decreasing with the increase of voltage¹². The permeation performance of the K-Ti₃C₂ and Li-Ti₃C₂ membranes exhibited strong voltage sensitivity and reached the maximum permeation rates of 24.12 and 22.87 g h⁻¹ cm⁻² at ~0.9 V.

A Water permeation rate through K-Ti₃C₂, Li-Ti₃C₂, and Ti₃C₂ membranes. Inset of (A) records the I–V characteristics of membranes. The error bars in (A) represent the maximum error range of permeation rate in fifteen independent experiments. B The increase in weight of water in the bottom container changes with time at 0, 0.5 and 1 V. Inset of (B) shows the stability of the K-Ti₃C₂ membrane. Time evolution of water molecules flow through nanochannels in K-Ti₃C₂ with parallel current of 0 A (C), 0.028 A (D), and 0.402 A (E). Red, gray, and blue atoms in (C–E) represent oxygen, hydrogen, and potassium atoms, respectively. F Water flow rate from AIMD simulations in K-Ti₃C₂ and Ti₃C₂. The parallel and vertical to the membrane surface were denoted as || and vertical ⊥, respectively. G I–V characteristics of K⁺, OH⁺, and F⁻ ions in the 2D channels of K-Ti₃C₂ by AIMD simulations.

Full size image

Unexpectedly, the permeation rates of K-Ti₃C₂ and Li-Ti₃C₂ membranes started dropping sharply from 0.9 V and then quickly returned to the initial voltage-free value. The water permeation through the K-Ti₃C₂ and Li-Ti₃C₂ membranes further decreased at 10 V. The permeation rates of water were only 0.84 and 2.74 g h⁻¹ cm⁻², respectively. Different from the intercalated membranes, the permeation rate in the Ti₃C₂ membrane always increases very slowly following the voltage increase within 0~10 V, which confirmed that the water in particle voids is almost unaffected by external voltage. The I–V characteristics of membranes are depicted in the inset of Fig. 2A. The permeation rates and currents of K-Ti₃C₂ and Li-Ti₃C₂ undergo a sudden change almost simultaneously, suggesting that there might be a special correlation between the water permeation rate and current. In this study, the electrically controlled permeation rates of both K-Ti₃C₂ and Li-Ti₃C₂ exhibited similar trends, thus we selected K⁺ ion-intercalated and original Ti₃C₂ for further study.

To eliminate the influence of surface metal cations on water permeation rate, the multilayer Ti₃C₂ was first annealed at 500 °C for 1 h in Ar atmosphere (C500) and then immersed into a KOH solution (C500-K). The XRD results confirm that the effective interlayer size of C500 was only 1.76 Å and the (002) characterization peak of C500-K sample didn’t shift to the left (Supplementary Fig. 5A, B), indicating that K⁺ didn’t intercalated into the interlayer spacing. EDS elemental mappings confirm the homogeneous distribution of K element on the surface of Ti₃C₂, as shown in Supplementary Fig. 5C. The experimental results didn’t show the gating phenomenon, and no obvious difference in water permeation rate was observed on C500 and C500-K under voltage (Supplementary Fig. 5D).

In addition, since the pure 2D channel is provided by the single-layer MXene permeation membrane, it can be used to further confirm that this anomalous phenomenon occurs in the 2D channel. Dimethyl sulfoxide (DMSO) was used for layering multilayer Ti₃C₂ and excluding metal cations in the control sample, which was then filtered into a membrane (S-Ti₃C₂) (Supplementary Fig. 6). Subsequently, the obtained membrane based on single-layer Ti₃C₂ was immersed in KOH solution for more than 10 hrs (S-Ti₃C₂-K). The XRD and SEM results confirm that the d-spacing and thickness of both S-Ti₃C₂ and S-Ti₃C₂-K were ~6.58° and 5 μm (Supplementary Fig. 7), respectively. EDS line scans show that K ions are uniformly dispersed on the S-Ti₃C₂-K membrane (Supplementary Fig. 8). In Supplementary Fig. 9A, B, the water permeation rate of S-Ti₃C₂-K increases with increasing voltage in 0~0.9 V and then the permeation rate sharply decreases and the current suddenly changes at 0.9~5 V, which are consistent with the K-Ti₃C₂. In contrast, Supplementary Fig. 9A shows water permeation rate of S-Ti₃C₂ kept increasing in 0~5 V. These results further indicate that the anomalous water permeation phenomenon originates from confined water within the 2D channel, and the sharp decrease in water permeation rate at high potentials depends on the current mutation caused by metal ions in 2D channels.

To confirm that the surface defects of the material hardly affect the water permeation process, STEM was employed to investigate the size of the surface defect. Firstly, we didn’t observe any through defects in multilayer Ti₃C₂ and K-Ti₃C₂ (Supplementary Fig. 10). Subsequently, the STEM images of single- or few-layer Ti₃C₂ and K-Ti₃C₂ confirm that the defects are quite localized and <1 nm (Supplementary Fig. 11). To further confirm that a few large-sized defects will not cause accidental water leakage, we adopted the 100 mL of 0.5 μg mL⁻¹ Evans Blue (EB) dye flows through the S-Ti₃C₂ and S-Ti₃C₂-K membranes. Since the size of EB (Size: 3.7 nm × 1.2 nm) is between the 2D channel and the larger surface defect, the impact of surface defects on the water permeation process can be determined by detecting the concentration of EB in the permeate²⁹. The experimental results showed that the retention rates of EB in S-Ti₃C₂ and S-Ti₃C₂-K membranes were 95.0% and 94.3%, respectively (Supplementary Fig. 12). Therefore, the factor of localized surface defects is negligible for the leakage of water molecules.

In Fig. 2B, the permeation of the K-Ti₃C₂ membrane was repeatedly tested under 0, 0.5, and 1 V, and showed excellent reversibility. The positive stability of water permeability in K-Ti₃C₂ membranes was also demonstrated at 0.1 V for over 30 hrs (Inset of Fig. 2B). In addition, the membrane permeation rates were carefully investigated under different membrane thicknesses and pressure conditions. Supplementary Fig. 13A, B show consistent water permeation rates.

Investigation of water dynamics by simulations and in situ spectroscopy

Subsequently, the confined water dynamics in the channel were investigated using ab initio molecular dynamics (AIMDs) simulations. Based on the experimental results, the currents were set within 0~0.402 A in simulations. Figure 2C–E and Supplementary Fig. 14A, B show the sequential flow process of water molecules in the 2D channels of K-Ti₃C₂. It was discovered that water molecules were able to occupy the channel within ~200 ps at both 0 and 0.028 A, but the number of water molecules in the steady state is inconsistent. In comparison, it takes about 400 ps for a water molecule to fill the channel at 0.402 A. For Ti₃C₂, the flow rate of water molecules in the 2D channel only undergoes a slight change when the current increases from 0 to 0.028 A (Supplementary Fig. 15A–C). In Fig. 2F, we further estimated the flow rate of water molecules in the channel and defined it as the weight of water passing through a fixed cross-section per unit of time. The computational results exhibit a high level of concordance with our experimental study, irrespective of the direction of the applied electric field. Furthermore, we elucidated the underlying mechanisms of the current mutation. Simulations were constructed to examine the possible effects of different ions in 2D nanochannels on the voltage–current under water-filled conditions (Supplementary Fig. 16). As shown in Fig. 2G, the K⁺ ions induced a sudden alteration in the overall current within 0.9~1 V, while the current changes generated by other ions can be ignored. The reason is that the free energy barrier has an inhibitory effect on the current, whereas the decrease in the free energy barrier of K⁺ ion forms a rich 2D conductive pathway at ~1.0 V. Consequently, a rapid surge in current was observed, which is consistent with the previous report³⁰. Nevertheless, despite the correlations between the structural variation and the dynamical evolutionary process of confined water are discovered, the discrepancy in flow rates remains perplexing. Hence, it is imperative to insight into the mechanisms behind the electrical manipulation of confined water.

Previous experimental results indicated that the notable disparity in the permeation performance between Ti₃C₂ and K-Ti₃C₂ can be attributed to the electrical control of water molecules in the 2D channels. However, confined water within the materials is more difficult to discern than surface water³¹. Infrared technology is suitable for detecting water molecules inside materials as the band of the infrared spectrum has a particularly strong signal corresponding to water molecules³². Typically, attenuated total refraction infrared spectroscopy (ATR-IR) was employed at micrometer depths, posing a challenge in effectively probing the internal structure and dynamics of water molecules due to the limited penetration of infrared light³³. In this study, we explored the structure and intermolecular interactions of confined water in the electrical 2D channel by transmission infrared spectroscopy. Compared to ATR-IR, the advantage of FT-IR is to obtain overall information including surface and bulk phases^33,34,35,36. A self-designed transmission-type infrared cell was used to detect the water molecules in situ under different electric fields and atmospheres (Supplementary Fig. 17).

As shown in Fig. 3A, the infrared spectra of the K-Ti₃C₂ sample were recorded in ~ 100% RH at different voltages. Two characteristic peaks can be discovered in Fig. 3A: an intensity peak at ~1633 cm⁻¹ originating from the bending vibration of water molecules and another infrared band located above 3000 cm⁻¹ corresponding to the symmetric and antisymmetric stretching modes of water molecules. Simultaneously, the water background was subtracted from all as-obtained infrared spectra to eliminate the interference of gaseous water in the environment. Subsequently, the stretching vibrational bands can be deconvoluted into five distinct components by Gaussian fitting (Fig. 3A). The peaks at approximately 3600 cm⁻¹ and 3520 cm⁻¹ correspond to the monomer, at approximately 3413 cm⁻¹ correspond to the dimer, and at approximately 3312 cm⁻¹ and 3200 cm⁻¹ correspond to the cluster³⁷. At voltages <0.9 V, we observed that both the bending vibrational peaks and stretching vibrational bands of the confined water in K-Ti₃C₂ increase with voltage rise, indicating a potential alteration in the amount of water within the nanoconfined space. By contrast, the infrared spectra of Ti₃C₂ exhibited minimal variation across different voltage conditions (Supplementary Fig. 18A). Subsequently, we subjected the K-Ti₃C₂ membrane at different voltages for 1 h under 100% RH and found that their final weights changed considerably. As shown in Fig. 3B, the weight of the membrane at 0.9 V is approximately doubled than that at 0 V.

A In situ infrared spectra of confined water inside the 2D channels of K-Ti₃C₂ under different voltages. The water stretching vibrational bands are fitted as the monomer (green and yellow), dimer (blue), and cluster (red and cyan). B Electrically controlled K-Ti₃C₂ membrane weight. The error bars in (B) represent the maximum error range of membrane weight in five independent experiments. C Radio changes of water monomer, dimer, and cluster from (A) in situ infrared spectra. The pink region represents the part of the number of clusters that changes dramatically. Red and gray atoms in (C) represent oxygen and hydrogen atoms, respectively. Schematic of the water distributions in the 2D channel of K-Ti₃C₂ at 0 V (D) and low-voltage range (E). The yellow arrows in (D) and (E) represent the flow direction of water molecules. Red and gray atoms in (D) and (E) represent oxygen and hydrogen atoms, respectively.

Full size image

Changes in the number of water molecules within the channels could be firstly attributed to changes in effective channel size or surface-wetting ability. Herein, we performed in situ XRD measurements with voltage at high humidity and contact angle tests. The XRD results indicate that the layer spacing of K-Ti₃C₂ and Ti₃C₂ didn’t change at <0.9 V (Supplementary Fig. 19A, B), which means that the alteration in permeation rate is independent of the channel size. In addition, the contact angle at 0 potential was respectively stabilized in the range of 45~50° and 54~57° with voltage increase (Supplementary Figs. 20 and 21), indicating the wetting properties of K-Ti₃C₂ and Ti₃C₂ membranes had no changes. Ultimately, the phenomenon was attributed to the directional polarization and electronic polarization of the confined water caused by voltage. To confirm it, after achieving a large amount of water adsorption with different voltages, we recorded the stretching vibration band of water on K-Ti₃C₂ under 10 s after power outage. At the same time as the power outage, we closed the testing chamber to avoid the entry of gas. After the power outage, the intensity of the water stretching vibration band remained almost constant. This indicates that although water molecules in the 2D channel are crowded in the absence of voltage, the water molecules didn’t leave the 2D channel rapidly since there is no pressure-driven water permeation process and no externally driven gas fluid (Supplementary Fig. 22). Supplementary Fig. 23 further evaluates the ratio of water monomer, dimer and cluster in K-Ti₃C₂ at different voltages. At 0.3 V, only ~3% of the clusters and ~4% of the dimers were broken. In contrast, ~32% of the clusters were broken at 0.9 V. Therefore, we believed that although the number of H-bonding increases with voltage (Fig. 3C), the increase in voltage promotes the break of more intermolecular H-bonding due to the rapid increase in the number of water molecules.

Figure 3D, E depict the transition states of water molecules from random to ordered. At 0 potential, the spatial utilization efficiency of K-Ti₃C₂ in 2D channels is not high. At this point, the water molecules are entrapped and the disordered water molecules lead to more spatial gaps in the channel. With the increase of voltage, some hydrogen bonds are bent and locally broken due to the polarization effect, which eventually leads to a regular distribution of electrical confined water in the channel. The densification of water molecules in the electrical channel is greatly improved compared to the disordered water molecules. To further validate this viewpoint, the dynamics of water molecules in the channels were investigated using in situ infrared spectroscopy (Supplementary Fig. 24). Initially, the samples were exposed at approximately 0 RH and observed the infrared absorption spectra intensity at different voltages, demonstrating that the infrared spectra won’t interfere by the voltage itself. Subsequently, the bubbling method was employed to continuously transport water molecules into the sample cell, and the gas flow rate throughout the entire testing process was controlled at 1000 ± 10 sccm. It was discovered that the water absorption peaks were stabilized at almost the same time (120 s) under different voltage conditions, but showed stronger water absorption peaks with the increase of voltage. Therefore, the enhancement in permeation capability application of an electric field can be attributed to a modification in the amount of water molecules traversing per unit size.

We then conducted an in-depth analysis for the unexpected sudden change when the voltage >0.9 V. Since a sudden change in current was observed while the flow rate changed, we first considered the partial water evaporation loss caused by Joule heating³⁸. For this purpose, the surface temperature of K-Ti₃C₂ membrane was recorded near the current mutation point with an infrared imager (Supplementary Fig. 26). The observed change in the membrane surface temperature was only 2.2 °C, corresponding to an increase in voltage from 0 to 5 V. The experimental results effectively eliminated the influence of the Joule heating effect on the water permeation rate. In situ XRD measurements were used to analyze the 2D channel size changes (Supplementary Fig. 19A). Indeed, at ~1 V, a 0.08 ° shift of the 002 peaks to a large angle was recorded and the layer spacing narrowed by approximately 0.15 Å. However, as the voltage continued to increase, the (002) characteristic peak promptly reverted to its original position, hence failing to elucidate the persistent decline in the water permeation rate. Furthermore, in Supplementary Figs. 24F and 25, the dynamic features of water molecules corresponding to potentials were recorded by in situ infrared spectroscopy. The intensity of the stabilized water absorption peak remains almost constant with increasing voltage, but the time to reach a steady state continues to increase. In contrast, the changes in the infrared spectra of Ti₃C₂ are negligible (Supplementary Fig. 25). Therefore, it is reasonable to believe that the decrease in the permeation rate depends mainly on the changes in flow rate.

The Gaussian-fitting spectra reveal (Fig. 3A, C) that the number of water clusters significantly increases to ~38% of the amount of confined water and the size of the cluster is increased (3200 cm⁻¹) at 1 V, while the intensity of the water stretching bands has no significant change. With the increase of voltage, the number of water clusters once increased to ~42%. Therefore, we assume that there is an equilibrium transition of confined water inside 2D nanochannels under electrical excitation: monomer ⇋ dimer ⇋ cluster. Figure 4A depicts the dynamic reorganization process of confined water. Since the conductive pathway formed by K⁺ within a 2D channel creates a strong electric field in and around it¹², a model was designed to confirm its ability to ionize a large amount of water molecules (Supplementary Fig. 27B), as detailed in the “Methods” section. Herein, each conductive pathway can be considered as a 2D wire with a thickness equal to the effective channel size of K-Ti₃C₂. Supplementary Fig. 28A shows that in the area near the positive electrode (1 V), the electric field can reach 10⁹. Within a range of 200 nm from the positive electrode, we observed the electric field inside and near the conductive wire exceed 5.0 × 10⁶ V m⁻¹ at 1 V, and they are almost unaffected by length. In contrast, we found that the maximum electric field strength within the 2D channel is <10⁵ V m⁻¹ at 0.8 V (Supplementary Fig. 28B), far below the maximum electric field strength at 1.0 V (10⁹ V m⁻¹). In addition, its electric field strength at 0.8 V rapidly decreases to below 10⁰ V m⁻¹ along the negative electrode direction. Previous reports have confirmed that water molecules begin to ionize ~10⁶ V m⁻¹ and rapidly ionize >10⁸ V m^−1 12,39. Such a significant change confirms that water molecules undergo almost no ionization at <0.9 V, while the strong electric field generated at >0.9 V is sufficient to promote the water molecules’ dissociation into hydrated (H₃O⁺) and hydroxyl (OH⁻) ions, thereby increasing the cohesion of water.

A Schematic illustration of dynamics for water structure from low to high potential. Schematic illustration of water molecules passing through 2D slits in K-Ti₃C₂ nanosheets at low (B) and high (C) potentials. Red and gray atoms in (A–C) represent oxygen and hydrogen atoms, respectively. D The number of water molecules in the K-Ti₃C₂ channel versus AIMD simulation time is based on different H₃O⁺ and OH⁻ ion concentrations at 0 potential. 0, 1%, 2%, 3%, 4%, 5%, and 10% represent the proportion of ions (H₃O⁺ and OH⁻) to water molecules. E The normalized flow rate of water molecules with different H₃O⁺ and OH⁻ ion concentrations.

Full size image

Density functional theory (DFT) calculations confirmed that the charged ions (H₃O⁺ and OH⁻) have a strong binding ability with the surrounding water molecules (1.32 and 1.14 eV) (Supplementary Fig. 29). The strong intermolecular electrostatic attraction induces spontaneous aggregation of the surrounding water molecules to form strong associative molecular chains, thus hindering the transportation of confined water in the nanoconfined space (Fig. 4B, C). Subsequently, AIMD simulations further confirmed the effect of dissociated H₃O⁺ and OH⁻ on the water permeation rate. In Fig. 4C, D, water molecules were disassembled into H₃O⁺ and OH⁻ to simulate the mobility of water molecules in 2D nanochannels at 0 V. The simulation results show that the flow rate decreases with increasing ion concentration, which is consistent with the infrared spectroscopy results. Therefore, the confined water permeation under heavy current can be mainly attributed to the large-sized hydrated ion clusters that block the 2D nanochannels.

Nanofluidic sensing

Inspired by the anomalous gating phenomenon, the samples were evaluated as sensitive materials for humidity detection to explore the nanofluidic sensing paradigm owing to 2D confined water that can arouse the resistance changes of MXenes⁴⁰. Here, the humidity response was defined as Response = (R_a/R_Ar) × 100%, where R_a and R_Ar represent the resistance change and the resistance exposure to pure argon gas, respectively. The optical image in Fig. 5C shows a micro-scale sensing substrate based on our photolithography fabrication with 2 μm resolution. Briefly, the manufacturing process involves the use of photolithography technology to create interdigital electrode patterns at the micron level on silicon wafers. Subsequently, non-conductive resin is transformed into glass carbon with desirable conductivity by subjecting to high-temperature pyrolysis.

A Dynamic response curves of Ti₃C₂, K-Ti₃C₂, and Li-Ti₃C₂. B Comparison between the Ti₃C₂, K-Ti₃C₂, and Li-Ti₃C₂ in terms of sensor response, response time (T_res), and recovery time (T_rec) under electrical control. C Stability of the humidity-sensing performance at 86.4% RH. The error bars in (C) represent the maximum error range of response in three independent experiments. Inset in (C) is a micro-scale sensing substrate. Scale bars: 100 μm in inserted SEM images in (C). D Selectivity of mixed gas testing for Ti₃C₂, K-Ti₃C₂, and Li-Ti₃C₂ to interference gases (Methanol, Toluene, Formaldehyde, NO, NO₂, H₂, N₂, and CO₂).

Full size image

The humidity-sensing measurement device is shown in Supplementary Fig. 31A, the alteration in humidity inside the test chamber was accomplished by accurately manipulating the proportion of dry and humid argon gas (Ar), and the gas flow rate was controlled at 1000 ± 10; sccm. In the process, each branch was controlled by a flow valve. A commercial humidity sensor with an accuracy of ±3% was installed to record the environmental humidity. Figure 5A records the dynamic resistance variation curves of Ti₃C₂, K-Ti₃C₂, and Li-Ti₃C₂ with different voltages in 5.6~86.4% RH. Specifically, pure Ar (dry) was introduced into the testing chamber to create a low humidity environment (5.6% RH), which was recorded as the initial humidity (Response: 0). Subsequently, pure Ar was purged through a washing bottle filled with deionized water and then blown into the testing chamber until the humidity reached 86.4% RH. The experiment results indicate that the K-Ti₃C₂ (1.27) and Li-Ti₃C₂ (1.12) have significant improvements in humidity responses, which are about 10 and 8 times higher than Ti₃C₂, respectively. This can be attributed to the introduction of metal cations which enlarged the effective interlayer spaces, allowing more water molecules to enter into the channel interior. Supplementary Fig. 32B illustrates a simple linear relationship between current and voltage under both humidity conditions. The results indicate that the sensitive layer/electrode interface has excellent conductivity, without obvious potential barriers or resistance, and charges can freely across the interface.

Subsequently, we further analyzed the humidity-sensing performance of the three samples with electrical manipulation. Figure 5B shows a comparative analysis of the modulation effect of the electric field on the humidity-sensing performance of Ti₃C₂, K-Ti₃C₂, and Li-Ti₃C₂. The experimental results didn’t show an obvious change in the humidity response of three samples at different voltages. Noteworthy that when the voltage increases from 0.3 to 0.9 V, the response and recovery speed of K-Ti₃C₂ (Response time: 113 → 49 s; Recovery time: 158 → 97 s) and Li-Ti₃C₂ (Response time: 65 → 38 s; Recovery time: 195 → 132 s) are greatly improved (Fig. 5B and Supplementary Fig. 32). In addition, the dynamic response curves of Ti₃C₂, K-Ti₃C₂ and Li-Ti₃C₂ and corresponding calibration curves were showed at 10.7%, 21.2%, 35.1%, 48.5%, 62.6% and 86.4% RH (Supplementary Fig. 33), where the response at 5.6% RH was recorded as 0. We also recorded 100 data points as signal-to-noise ratio tests at 5.6% RH and defined the minimum detection limit as more than 3 times the signal-to-noise ratio. The experimental results indicate that Ti₃C₂ with metal cation has a fast response/recovery speed, stronger sensitivity, and lower LOD for humidity, and surpasses that of previously reported state-of-the-art humidity-sensing materials (Supplementary Table 2). Three samples were exposed to low and high-humidity environments for extended periods of time without observing obvious sensitivity changes, suggesting excellent stability (Fig. 5C). Figure 5D shows the anti-interference ability of the sensor in response to typical gases. Humidity changes from 5.6% to 46.4% in all selective experiments. The results confirmed that the Ti₃C₂, K-Ti₃C₂, and Li-Ti₃C₂ have excellent selectivity for humidity. Therefore, electric and ionic manipulations are promising methods for developing nanofluidic sensing applications.

In summary, our work reports an anomalous gating phenomenon on the ion-intercalated Ti₃C₂ membranes under electric field conditions and accurately controlled the variation of water permeation rate in the range of 0.83~24.12 g h⁻¹ cm⁻² at atmospheric pressure. By the combination of in situ infrared spectroscopy and AIMD simulations, the evolution of confined water structure was elucidated. The results indicate that the directional polarization of confined water under low-current settings generated more space to accommodate water molecules, while hydrated ions under heavy-current conditions blocked the channel and hampered the transmission of water molecules. Inspired by the anomalous gating phenomenon, a nanofluidic humidity-sensing paradigm was developed and the response/recovery speeds were greatly improved by manipulating voltage. This work is anticipated to bring new insights into membrane separation fields such as air purification, catalysis, sensing, and water treatment.

Chemicals and materials

MAX parent phase Ti₃AlC₂ (300 mesh, 99.5 wt.%) was purchased from Foshan XinXi Technology Co., Ltd. Potassium hydroxide (KOH, Meryer, 85%) and Lithium hydroxide (LiOH, Meryer, 99%) were used as received. Hydrofluoric acid (HF, >50%), hydrochloric acid (HCl, 36.0–38.0%), and ethanol (C₂H₆O, ≥99.7%) supplied by Sinopharm Chemical Reagent Co., Ltd. Dimethyl sulfoxide (DMSO, ≥99.95%) were provided by Aladdin. Evans blue dye was purchased from Sigma-Aldrich (EB, E2129-10G).

Synthesis of multilayer Ti₃C₂ MXene

Multilayer Ti₃C₂ was synthesized by etching the “Al” layers from Ti₃AlC₂ (MAX phase) with 10% HF. Briefly, a 7.5 mL 50% HF was added into 22.5 mL deionized (DI) water in a 100 mL Teflon beaker, and then 3 g Ti₃AlC₂ powders were slowly added in. With magnetic stirring, the reaction proceeded for 24 h at RT. The sediment was washed with DI water, centrifuged at 5000 × g for 5 min, and the supernatant was poured out. This process was repeated 6–7 times until the pH >5. The as-obtained multilayer Ti₃C₂ powder was collected by drying under a vacuum at 60 °C for 12 h.

Synthesis of the K-Ti₃C₂ and Li-Ti₃C₂

K-Ti₃C₂ and Li-Ti₃C₂ were successfully synthesized following the reported method²⁴. An aliquot of 2 g multilayer Ti₃C₂ was immersed in 2 M 50 mL KOH and LiOH solution, and the multilayer Ti₃C₂ was stirred at RT for 1 h. The as-obtained Ti₃C₂T_x powders are denoted as “K-Ti₃C₂ and Li-Ti₃C₂”. DI water was added to the mixed liquid, and the resulting product was then centrifuged at 5000 × g for 5 min to remove the supernatant. This procedure was repeated once after washing the product with DI water until pH >5. The K-Ti₃C₂ and Li-Ti₃C₂ powders were vacuum-dried at 50 °C for 12 h.

Synthesis of C500 and C500-K

Firstly, the multilayer Ti₃C₂ powder was annealed in an Ar atmosphere at 500 °C for 1 h, and the heating rate was 5 °C min⁻¹. Subsequently, the annealed Ti₃C₂ samples were immersed in a 2 M KOH solution, and the product was collected after centrifugation, filtration, and drying.

Synthesis of the single-layer Ti₃C₂ and K-Ti₃C₂ membranes

Ti₃C₂ nanosheets were synthesized by using the DMSO. Specifically, 1 g Ti₃C₂ powder and 25 mL of DMSO solution were mixed and stirred for 12 h at RT. Subsequently, the mixed solution was washed by centrifugation (6800 × g for 5 min) three times. The single-layer Ti₃C₂ membrane was prepared by vacuum filtration of the sediment. Finally, the S-Ti₃C₂ membrane was immersed in 2 M 20 mL KOH solution and then washed with DI water.

Fabrication of permeation membranes

The permeation membrane consists of the electrode layer and permeation layer. Firstly, the electrode layer was obtained using standard ion-beam sputtering deposition. In detail, the working time of the low vacuum coating machine (Germany, Leica EM ACE 200) and the distance between the membrane and the target material were pre-set at 5 min and 9 mm, respectively. Next, the deposition of a gold layer on a commercially available porous mixed cellulose ester substrate (MCE, 50 mm diameter, 0.22 μm pore size) was achieved by sputtering a gold target with an energy of 2.5 kV and an ion-beam current of 30 mA at a <5 Pa residual atmospheric pressure. To obtain uniformly sized MXene particles, the collected samples were screened using 400 mesh sieves, respectively. The as-obtained MXene was uniformly deposited on the gold-deposited cellulose membrane through vacuum-assisted filtration to fabricate the MXene permeation layer. The membrane thickness was controlled to be quite uniform while ensuring the same mass density. Finally, the top Au electrode was fixed onto the as-fabricated permeation layer and then dried at room temperature (RT) for 24 h (Au/MCE-MXene-Au/MCE).

Material characterizations

The morphologies and microstructures of materials were analyzed by optical microscope (OM, GP-680V), field emission scanning electron microscopy (FESEM, Gemini SEM 500), and atomic force microscope (AFM, FM-Nanoview1000AFM). The crystal structures of materials were characterized by transmission electron microscopy (TEM). The atomic-resolution scanning transmission electron microscopy (AC-STEM) was performed using ThermoFisher Talos F200X (FETEM, 200 kV). High-angle annular dark field (HAADF)-STEM images were recorded using a convergence semi-angle of 11 mrad, and inner- and outer collection angles of 59 and 200 mrad, respectively. Energy dispersive X-ray spectroscopy (EDS) was carried out using 4 in-column Super-X detectors. X-ray diffraction (XRD, Rigaku D/Max-2550 diffractometer) patterns were recorded with Cu-Kα radiation (λ = 1.54059 Å) with a scanning step of 0.02° and a step time of 0.5 s. X-ray photoelectron spectroscopy (XPS) spectra were obtained using an ESCALAB 250Xi system with Al Kα as the X-ray source.

Permeation measurements

The effect of electrical control on the permeation performance of MXene membranes was investigated by pressure- and gravity-driven water permeation experiments. The experimental setup for permeation testing was first designed based on a vacuum-assisted filtration device. Specifically, two glass containers were clamped from both sides to create a confined space and sealed with an O-ring. Subsequently, a vacuum filtration pump and pressure gauge were connected to the lower space of the device to regulate the pressure difference between the environment and the outside. Herein, it was discovered that the as-fabricated Au/MCE-MXene-Au/MCE could withstand a pressure difference of ~1 bar, which indicates that the permeate membrane is robust. In addition, to ensure that the copper conductor would not short-circuit during testing by coming into direct contact with the lower film, we wrapped the entire copper conductor with plastic tape. Next, a typical gravimetric method was used to evaluate the water permeation rate. The weight loss per unit of time was estimated from a precision electronic balance and the volume change of the aqueous solution. Different voltages were applied to the samples with a CHI 660E potentiometer. To detect the current-induced Joule heating effect, the surface temperatures of the membranes were recorded by an infrared imager (HM-TPH11-3AXF, Beijing Precede Technology Co., Ltd, China). In addition, a blank sample was set up to deduct the effect of evaporation of surface-adsorbed water molecules on the experiment. The entire test procedure was carried out in a glove box with an Ar atmosphere so that the water permeation rate in the permeation experiments was almost independent of the water pressure. In addition, the influence of surface defects on the water permeation process was verified through a dye experiment. Specifically, 0.5 µg m⁻¹ solution of Evans blue (EB) dye was prepared as the feed solution. In the experiment, 100 mL of the feed solution was placed on top of the membrane under testing in our pressure-driven permeation setup. After 12 h, all the solution passed through the membrane (permeate). Subsequently, we soaked the permeable membrane in 100 mL of DI water for >24 h to investigate the optical absorption spectroscopy of EB dye in the retentate.

In situ XRD measurements

The interplanar spacing of the sample was analyzed by XRD (D8 Advance, Bruker) with Cu-Kα radiation, and the 2θ range of XRD pattern was taken from 3° to 75° with 0.02° step size and 6 min/° step speed. Next, to achieve high humidity (82.5% RH), the water molecules were continuously transported to the testing chamber through the typical bubbling method. Specifically, we introduced pure Ar through a glass bottle containing deionized water and blew it around the sample. Under such high-humidity conditions (82.5% RH), the sample can be considered as a 2D channel filled with water. For low humidity environments (6.2% RH), we directly introduced pure Ar into the surrounding area of the test bench for >5 min to drive away the water in the air. Two copper wires were connected to both ends of the membrane, and a voltage was applied to the membrane after reaching the target humidity. A commercial sensor with a detection range of 0~100% RH with an accuracy of 3% was used to monitor the real-time humidity and the bias voltage at both ends of the samples was controlled by an electrochemical workstation (CHI 660E) in all experiments. XRD patterns were collected under different voltage conditions at RT. The c lattice parameter of samples was calculated by Bragg’s equation:

where θ represents the angle between the incoming ray, reflected ray, and reflected crystal surface, and λ is the wavelength of the incident wave.

In situ infrared measurements

To probe the structure and dynamics of water molecules in 2D channels of MXene, the transmission-type infrared spectroscopy was used. The infrared spectra were recorded on a Perkin-Elmer spectrum 3 FT-IR spectrometer with a resolution of 4 cm⁻¹ at ambient conditions. The measurement substrate was fabricated by screen printing techniques. Typically, Platinum (Pt) conductive paste and Calcium fluoride (CaF₂, 0.5 mm thickness) sheets were first selected as conductive inks and substrates. Subsequently, the ink was coated on the template and scraped into the holes of the template using a scraper to form an electrode pattern. Finally, the as-fabricated substrate (0.5 × 2 cm) was collected by drying and welding. One wire was connected to the test substrate and another wire was connected to the top of the membrane. This method can mimic the surrounding electric field of water permeation experiments as much as possible without blocking the infrared signal. A measurement substrate was immersed in ethanol for ultrasonic treatment, and Ti₃C₂, K-Ti₃C₂, and Li-Ti₃C₂ were chosen as measurement samples, respectively. The obtained ink was dropped onto substrates and dried in a vacuum at 50 °C for 1 h. The bias voltage at both ends of the samples was controlled by an electrochemical workstation (CHI 660E). For Fig. 3A and Supplementary Fig. 18A, to obtain the humidity, we introduced pure argon gas into a wash bottle filled with deionized water and blew it into the testing chamber. Until the humidity reached close to 100% RH, we applied different voltages and recorded the experimental data. In Supplementary Figs. 24 and 25, we used a similar bubbling strategy to blow wet Ar into the test chamber and applied different voltages on the membrane before the experiment. Importantly, the gas flow rate was controlled at 1000 sccm for all experiments to ensure a consistent number of water molecules were introduced at the same time in each experiment. A commercial sensor with a detection range of 0~100% RH with an accuracy of 3% was used to monitor the real-time humidity in the testing chamber.

Device fabrication and sensing measurement

SU-8 TF 6000.5 was spun coated on Si wafer at 2000 rpm for 30 s, prebaked at 110 °C for 1 min, and patterned with TERA-Fab (TERA-print, LLC, Skokie, IL, USA) using 20X objective and 365 nm light source. The built-in digital light processing system consists of 1024 × 768 micromirrors which can create a single projection using maskless design, which has a minimum resolution of 1.25 μm and a maximum area of 1.28 × 0.96 mm². To acquire an interdigital electrode, 9 stitches were utilized and every projection was patterned using 20% intensity for 1.3 s. After patterning, the resulting substrate was post-baked at 110 °C for 2 min before developing in SU-8 developer for 3 min. Final acquired non-conductive electrodes were rinsed with ultra-pure water and dried with nitrogen. Such prepared electrodes were then pyrolyzed in a tube furnace at 1000 °C for 1 h with a ramp-up speed of 2.5 °C/min. This high-temperature pyrolyzing process converted non-conductive resin into glassy carbon. Sensitive films based on Ti₃C₂, K-Ti₃C₂, and Li-Ti₃C₂ were fabricated on the substrate surface by a simple drop-coating method. A 20 mg amount of Ti₃C₂, K-Ti₃C₂, and Li-Ti₃C₂ were added to 200 μL DI water respectively, and the mixture was ultrasonicated. The products were evenly coated onto the interdigital electrode, and the mass density of sensitive films was 2 ± 0.2 mg cm ⁻², before vacuum drying at 50 °C for 2 h. To obtain the humidity, pure argon was introduced into a gas-washing cylinder with DI water. The target gas and pure argon were respectively introduced into the specially designed gas mixing apparatus. The humidity-sensing measurement was accomplished by accurately manipulating the proportion of dry and humid argon gas, and the total gas flow rate was controlled at 1000 ± 10 sccm. In the process, each branch was controlled by a flow valve. A commercial humidity sensor with an accuracy of ±3% was installed to record the environmental humidity. Selective testing is performed by introducing both Ar with water molecules and the interference gases were introduced into the test chamber at the same time, the flow rate of each gas was kept at 500 sccm. Specifically, when conducting selectivity experiments on water molecules and VOCs, both water molecules and VOCs are simultaneously blown into the testing chamber through pure Ar. For other interference gases (1000 ppm), Ar with water molecules and other interference gases are directed purged into the test chamber. Response time and recovery time are defined as the time required for a 90% alteration in resistance.

Simulation of the electric field around conductive pathways

Firstly, we uniformly divided the K-Ti₃C₂ membrane into four equal parts. Although these parts had a resistance approximately four times that of the parent membrane, each individual part had a conductivity similar to that of the original membrane (Supplementary Fig. 27A). It indicates the uniform distribution of the conducting wires within the K-Ti₃C₂ membrane and thus can be regarded as a parallel resistance model of the conducting wire structure. The electric field within the current-carrying conductor was simulated using the COMSOL Multiphysics finite-element-based solver (https://www.comsol.com/). The ‘Electric currents’ module was used and the electric field E was computed as the opposite gradient of the electric potential V as follows E = −∇V. The point form of Ohm’s law states that: J = σE where σ is the electrical conductivity (SI unit: S/m). The static form of the equation of continuity then states: ∇·J = −∇ · (σ∇V) = 0. Here, each 2D conductive pathway can be regarded as a 2D wire, with its half-width (a) a being half of the effective channel size, its length recognized as half the lateral size of a single nanoparticle (L), and b is interlayer distance. The Laplace equation is valid for any point at a distance r between a and b. Based on XRD and STEM results (Supplementary Figs. 2 and 3), we set the a and b to 0.235 nm and 0.615 nm, respectively. The electric conductivity of the current-carrying conductor was taken at 0.8 V and 1 V (248 and 4025 S m⁻¹) through the I–V curve in Fig. 2G. The electric conductivity of the outer medium was assumed to be 0.025 S m⁻¹ according to the I-V curve in Supplementary Fig. 27A. The boundary conditions for the above scenario are V (r, z) = 1 V when r ≤ a and z = L, V (r, z) = 0 V when r = b or r ≤ a and z = 0. All of the above estimates are based on a simple model of a single straight conducting wire. Due to the close arrangement of 2D channels, the complex structure of conductive wires may generate higher electric fields, especially near the positive electrode.

DFT calculations

All calculations were performed with the density functional (DFT) theory using the Vienna ab initio simulation package^41,42. The Perdew-Burke-Ernzerh (PBE) exchange-correlation functional was adopted as the generalized gradient approximation⁴³. The plane wave energy cut-off was set at 500 eV, based on the the projected augmented wave (PAW) potentials⁴⁴. The threshold of energy and force convergence were set at 10⁻⁶ eV and 0.01 eV/Å, respectively.

The formation energy can be obtained by the following formulas.

where \({{{\rm{E}}}}_{{{{\rm{H}}}}_{2}{{\rm{O}}}+{{{\rm{H}}}}_{3}{{{\rm{O}}}}^{+}/{{{\rm{OH}}}}^{-}}\), \({{{\rm{E}}}}_{{{{\rm{H}}}}_{2}{{\rm{O}}}}\) and \({{{\rm{E}}}}_{{{{\rm{H}}}}_{3}{{{\rm{O}}}}^{+}/{{{\rm{OH}}}}^{-}}\) represent the binding and the individual energy of H₂O and H₃O⁺ (or OH⁻), respectively.

Molecular dynamics simulations

To investigate the water molecular dynamics in charged 2D channels, molecular dynamics (MD) simulations were carried out using the LAMMPS software package. The computation model consisted of parallel MXene (Ti₃C₂) nanosheets, H₂O, K⁺, and ionized H₃O⁺ and OH⁻. The MXene nanosheet repeat unit is originally from the Ti₃C₂ crystal structure covered by surface groups such as -O, -OH, and -F. The ratio of -O, -OH, and -F is 0.3/0.4/0.3 as shown by XPS spectra (Supplementary Fig. 4). The XPS peak shows that the atomic ratio of O and F in Ti₃C₂ and K-Ti₃C₂ are 5.89/27.81 and 6.39/31.79, respectively (Supplementary Table 1). In addition, Supplementary Fig. 4C, D shows O 1s region of Ti₃C₂ and K-Ti₃C₂, the bindings of O 1s at 533.3, 532.5, 532.0, 530.9, and 530.1 eV were attributed to adsorbed H₂O, C-O, -OH, -O, and TiO₂, respectively. The proportion of -OH and -O to the overall O element in Ti₃C₂ (-OH: 25.6% and -O: 19.6%) and K-Ti₃C₂ (-OH: 27.7% and -O: 20.8%) were obtained by Gaussian fitting, respectively. By the XRD and STEM analysis, the d-spacing of 2D channels in Ti₃C₂ and K-Ti₃C₂ were set to 9.61 and 12.38 Å, respectively. According to the water permeation results of K-Ti₃C₂ under different pressures (Supplementary Fig. 13B), the external pressure is set to 10,000 Pa (~100 mbar). The box size in Ti₃C₂ and K- Ti₃C₂ is 12 × 4.89 × 3.02 nm. The H₂O molecules were adopted by the standard SPC/E model. To balance the negative charge of MXene channels, the K⁺ ions were limited to the slit channel. The Lennard–Jones potential with a 10 Å cut-off radius and Coulomb potential were used together to describe intermolecular interactions. In addition, the Verlet algorithm is used for the Numerical integration of equations of motion. The entire running program undergoes multi-step steepest-descent energy minimization before the calculation begins. The equilibrium MD simulations were conducted in the common NVT ensemble using the LAMMPS package. The universal force field parameters for Ti₃C₂, K⁺, and H₂O were derived from CVFF that can be applied to inorganic systems, and periodic boundary conditions are applied in the y and z directions. In NVT simulations, the time step is 1.0 fs. Subsequently, 10 ns NVT equilibrium simulations were performed. Temperature of the system was maintained at 300 K by the V-rescale thermostats. The pressure of the system in the NPT ensemble is 1 standard atmospheric pressure. Simulation results were performed and analyzed using GROMACS and VMD software. All MD simulations in this work were repeated three times.