Introduction

Selective and tunable ion transport is crucial in advanced material applications, including ‘lab-on-a-chip’ devices1,2, energy harvesters3,4, desalination membranes5,6 and neuromorphic devices7,8. Among the many external stimuli affecting ionic transport, electrostatic gating appears to be highly reversible. Inspired by the remarkable efficiency of biological channels that rely on gated ion transport to activate certain functions9, researchers have explored the effects of electrostatic gating in several confined systems10. Although few experimental groups have achieved reasonable gate modulation at dilute ion concentrations, little progress has been made at physiologically and practically relevant concentrations11,12,13 (i.e., 100 mM or higher). To be effective at high ion concentrations, the channels need to overcome electrostatic screening, which essentially restricts the channel size available for ion transport. The Debye length estimation at a 1000 mM KCl concentration suggests that the channel size should be smaller than 6 Å. These challenging size requirements explain the absence of works addressing high concentrations. Nevertheless, gating studies at dilute concentrations revealed several intriguing effects: dehydration of ions14, modification of interlayer interactions15, concentration polarization16, and ion sieving17,18. In the past, several confined systems have been explored, such as silica nanochannels and nanowires11, unintercalated graphene oxide (GO laminates12), MXenes13, MXene-GO [ref. 19,20] and TiO2 nanochannels21. However, the majority of these systems have channel sizes > 7 Å. The best gate modulation effect reported thus far is \( < \,\)1000% at concentrations as small as 10 mM.

At a high salt concentration (1000 mM), significant ion interaction effects are expected if the channel size is comparable to the Debye length, as demonstrated by molecular dynamics (MD) simulations22,23. At this length scale, ion-ion and ion-water-ion interactions are likely to occur because of the close proximity of ions, water, and surfaces. Moreover, experiments can detect and study these effects. For example, the knock-on mechanism of ion transport due to Coulomb repulsion could lower the energy barrier and enhance the ionic conductance24. Electrostatic gating of the fluidic surface can change the ionic concentration and their transport in highly confined channels. In addition to gating, the valency of the ions can also affect their interactions. In the very few studies, electrostatic gating was also found to alter ion˗water interactions14.

Modulating ion transport at sea salt concentrations of 600 mM or higher is extremely difficult because of the requirement of highly confined fluidic channels, since the short Debye screening length at this concentration makes electrostatic gating ineffective. The focus of this paper is to modulate ion transport at these extreme concentrations, which is rarely discussed. There are many advantages of having high ion concentrations; for example, a high concentration ensures high ionic conductivity and hence a better signal-to-noise ratio. High concentrations may provide better stability against environmental fluctuations.

With this motivation, we designed channels with an overlapping electrostatic double layer (EDL) at high salt concentrations to assess the ion modulation and interaction effects. For this purpose, we fabricated laminates with angstrom-sized fluidic channels using earth-abundant and cost-effective vermiculite clay material. In vermiculites, the interlayer cations can be easily exchanged, thus providing a tunable transport height from \(\sim\)3 to 5 Å. The atomically small interlayer space ensures that the Debye layer overlaps even at a 1000 mM concentration, enabling the selective and tunable transport of smaller cations such as K+ via electrostatic gating. The insulating membrane itself can serve as a gate dielectric, eliminating the need for an extra layer. The membrane is stable under aqueous conditions25 and can operate at elevated temperatures26.

Results and Discussion

Vermiculite crystals that occur naturally are composed of layers of magnesium aluminosilicate. Each layer is made up of three sheets: one octahedral sheet occupied by the Al3+ ion at the octahedral site can be substituted with Mg2+ or Fe2+, which is sandwiched between two tetrahedral sheets occupied by the Si4+ at the tetrahedral site (Fig. 1a) [ref. 27,28]. Additionally, Al3+ ions substituted one-fourth of the Si4+ ions in tetrahedral sites, causing an enhanced negative charge in the layered structure. These excess charges are balanced by cations such as Mg2+ or Ca2+ ions, which reside in the space between these layers. In our study, these Mg2+ cations were exchanged with other cations via the 2-step heat exchange approach described in the methods section. We intercalated K+, Ca2+, and Al3+ ions into vermiculite (V) laminates to achieve variable interlayer spacing. To estimate the interlayer spacing (d) of vermiculite laminates, we performed X-ray diffraction (XRD) on free-standing Li‒V, K‒V, Ca‒V, and Al‒V membranes in both dry and wet states (Supplementary Fig. 1a). We observe no significant difference in the peak positions of both states. Since we performed our ion transport measurements in a wet state, it is appropriate to discuss the XRD results in the wet state. The XRD data revealed a ‘d’ spacing of 12.2 Å for K‒V. After subtracting the space occupied by the single magnesium aluminosilicate layer, which is 9.6 Å [ref. 29], we obtained a transport height of 2.6 Å for ion transport, which is occupied by solvated K+ ions. Similarly, for the Ca‒V and Al‒V membranes, the transport height is 5.6 Å and 5.1 Å, respectively, occupied by hydrated Ca2+ and Al3+ ions. In the case of the Ca‒V and Al‒V membranes, higher-order peaks were observed, indicating complete laminated frameworks. Scanning electrone microscope (SEM) cross-sectional images further confirmed the laminated structure (Supplementary Fig. 2) and helped us estimate the thickness of the membrane (i.e., \(\sim\)3.5 μm). The cross-sectional energy dispersive spectroscopy (EDS) analysis of the K-intercalated vermiculite membrane revealed the main elements of vermiculite, such as O, Si, Mg, and Al, along with the intercalant K (Fig. 1b). The EDS maps of Ca‒V and Al‒V similarly show the intrinsic elements of vermiculite along with the intercalant (Supplementary Fig. 3a, b). However, intercalant Al is hard to distinguish from intrinsic Al. The membrane consists of monolayers with a thickness of 14 Å, as estimated from the AFM height profile (Supplementary Fig. 4). The mica-like structure adsorbs 1 or 2 layers of water molecules on its surface and hosts intercalated ions on its surface30. This results in a slightly larger thickness of 14 Å instead of 9.6 Å.

Fig. 1: The structure of vermiculite and its composition.
figure 1

a Schematic of the vermiculite layers. The unit cell consists of tetrahedral and octahedral layers with Mg2+ occupying the octahedral sites and Si4+ and Al3+ the tetrahedral sites with shared oxygen. Exchangeable cations are occupied between the layers to balance the excess negative surface charge and are replaced with K+/Ca2+/Al3+ cations. b Energy dispersive spectroscopy (EDS) analysis of the elements across the layers of the K‒V membrane indicates the presence of O, Si, Mg, Al, and the intercalant K. The scale bar indicates 1 µm. c High-resolution X-ray photoelectron O 1s spectra (XPS) of the K‒V, Ca‒V, and Al‒V membranes. The grey open circle indicates the raw data of O 1s, and the blue line is the fitted data. Two deconvoluted spectra (green line and violet line) were derived from the raw data. The two black dotted straight lines indicate the peak position of -OH (violet) and Si-O-Si(Al)/Si-O-Mg (green) for K‒V membranes. A clear shift of peak position towards higher binding energy is observed with increasing cation valence. Source data are provided as a Source Data file.

Full size image

Surface charges play a crucial role in controlling nanofluidic transport because the charged solid surface and electrolytes form an EDL [ref. 31], as proposed by Helmholtz in 1853. V-laminates feature a negative surface charge that draws cations to the interlayer space from the electrolytes, creating an EDL, which is confirmed by our zeta potential (Supplementary Fig. 5) and transport measurements. We employed several charged cation intercalants to control these zeta potentials. The zeta potential measurements of the K‒V, Li‒V, Ca‒V, and Al‒V solutions indicate that the K‒V layers have a larger zeta potential of –50 ± 3 mV than the Ca‒V (–19 ± 3 mV) and Al‒V (–4 ± 1 mV) layers. We also measured the zeta potential of the membranes; the values were very similar to those of the dispersed solution. The higher zeta potential value of the K‒V solution indicates that more K+ is needed to neutralize the surface than Ca2+ and Al3+. This observation also suggests that in vermiculite, the surface charge can be easily controlled by the appropriate exchange of ions. The vermiculite membrane exhibited mechanical stability, with a tensile strength of ∼35 MPa and a fracture strain of ∼2% (Supplementary Fig. 6). This demonstrates that vermiculite is flexible and suitable for use in aqueous conditions.

Furthermore, we carried out a Fourier transform infrared (FTIR) spectroscopy study to confirm the presence of functional groups in the vermiculite. As described in Supplementary Fig. 7, the broad absorption peaks at 3360 cm-1 and 1643 cm-1 indicate the stretching and bending vibrations of the -OH groups, respectively32. The absorbance peak at 970 cm-1 corresponds to the asymmetric stretching vibration of Si-O-Si. The two peaks at 735 cm-1 and 652 cm-1 are related to the Al-O-Al and M-O-Si (M is Mg, Al, and Fe) bonds, respectively29,32. We also performed Raman analysis for all the cation-intercalated membranes, which revealed the characteristic peaks of Mg-O/Al-O and Si-O (Supplementary Fig. 8) [ref. 33].

We performed X-ray photoelectron spectroscopy (XPS) analysis on all the cation-intercalated membranes. The XPS survey of the K‒V, Ca‒V, and Al‒V samples provided in Supplementary Fig. 9a shows common elements such as O, Si, Mg, and Al, along with intercalant atoms. The depth analysis of the intercalated cations indicates successful intercalation within the layers (Supplementary Fig. 9b–d). The O 1s deconvoluted spectrum has two peaks for K‒V, one at 531.0 eV and the other at 532.3 eV, which corresponds to the Si-O-Si(Al)/Si-O-Mg and hydroxyl groups, respectively (Fig. 1c)34,35. Both peaks shifted toward higher binding energy values with increasing cation valence, and the shift was maximal for Al-intercalated vermiculite, i.e., at 531.5 eV and 533.3 eV. The higher binding energy suggests that the electrostatic interaction between Al3+ and the layers of vermiculite is considerable among all the intercalated vermiculite, which is in the order of Al3+‒V > Ca2+‒V > K+‒V. A similar shift is also reported in the case of cation-intercalated montmorillonite membranes33. A detailed analysis of the spectra corresponding to the elements of vermiculite, Si, Al, and Mg is provided in Supplementary Fig. 10.

Electrical modulation of ionic conductance measurements of angstrom capillary devices

We fabricated a voltage-gated device using various cation-exchanged vermiculite membranes. The membrane was initially cut into several pieces of size 5 × 4 mm2. These membrane pieces were further utilized to make the device. For the gate voltage, a gold wire was fixed onto a silver strip made in the middle of the membrane, which ensures a uniform gate voltage across the membrane (Fig. 2). This sample was then encapsulated between two acrylic blocks with the help of an epoxy (Loctite Stycast 1266), which isolated the gold gate electrode from the electrolytes. To expose both ends of the membrane, the sample was polished with P1000 emery sandpaper, and the final dimensions were confirmed via optical microscopy. The polished sample was sandwiched between two circular acrylic pieces with a prefabricated hole to ensure that ion transport occurred only through the vermiculite membrane (Fig. 2). The fabricated device had a length, l of 3 mm, and width, w of 4 mm. We performed all our measurements with these devices.

Fig. 2: Fabrication of a voltage-gated sample.
figure 2

Schematics of various steps involved in the fabrication of voltage-gated vermiculite samples for studying gated ion transport characteristics. Here, h, w, and l are the thickness, width, and length of the membrane, respectively. We used a gold wire to apply the gate voltage. The images are not drawn to scale.

Full size image

We performed ionic transport measurements in a homemade PEEK (polyether ether ketone) cell. The device was placed between two reservoirs containing aqueous salt solutions. Two homemade Ag/AgCl electrodes, labeled as the source and drain, were used for the ionic measurement. The ions were transported by applying a voltage, Vds (drain to source voltage), and the resultant current, Ids (drain to source current), was monitored. We used another source meter to apply a gate voltage (Vg) to the nanofluidic device, with the source as the ground terminal. A schematic of the measurement setup is shown in the inset of Fig. 3a. For the transport studies, we choose the right salt-exchanged membrane and solutions, for example, K‒V membranes with KCl solutions, Ca‒V membranes with CaCl2 and Al‒V membranes with AlCl3, to avoid any ion exchange during the experiments. We prepared all the salt solutions with Milli-Q deionized water, and the final pH of the solution was 5‒7; however, AlCl3 is an exception, with a pH of 2.7 for a 1 mM chloride concentration. The pH values of the Li+, K+, Ca2+, and Al3+-intercalated vermiculite flakes dispersed in a 1 mM chloride solution were 7.2, 6.8, 7.1, and 5.6, respectively. It is clear from various characterizations that small changes in pH do not affect d-spacing or stability since equilibrium is achieved due to cation exchange. The sample temperature was maintained at 298 K for all the measurements, unless otherwise specified.

Fig. 3: Electric field modulated ion transport through the K-V membrane.
figure 3

a Modulation of the ionic current, Ids with different Vg values from −2 V to +1 V in steps of 0.5 V in 1000 mM KCl solution. Inset: schematic of the voltage-gated device used for the transport measurement. b Variation in conductance with gate potential for 1 mM, 10 mM, 100 mM, and 1000 mM KCl. Data are presented as mean values ± standard deviation. N = 3 independent samples. c Gating ON/OFF ratio between conductance at −2 V to +1 V for the several concentrations of KCl from 1 to 1000 mM. Data are presented as mean values ± standard deviation. N = 3 independent samples. d Arrhenius plot for 1000 mM KCl showing the activation energy at several Vg values, the solid lines represent a linear fit. Source data are provided as a Source Data file.

Full size image

We first studied the effect of the gate voltage on the ion transport characteristics of a K‒V sample. For this purpose, Ids–Vds characteristics were measured at different gate voltages, with Vg ranging from −2 V to +1 V. The characteristics were mostly linear at low Vds (Fig. 3a) for a 1000 mM KCl solution. There is a visible change in the slope of the curve with various Vg values. A finite zero-current voltage is observed for a larger negative Vg, plausibly indicating an asymmetric charge distribution. The larger positive current with negative Vg is attributed to the increased K+ density and mobility, as supported by MD simulations presented later. The ionic conductance was extracted from the slope of the Ids–Vds plot. The KCl conductance across the membrane was reduced from 5.0 μS to 1.5 μS when Vg increased to +1 V (Fig. 3b). Conversely, the conductance value increased to 22.9 μS after applying a gate voltage of –2 V. The conductance increased by 1400% when Vg changed from +1 V to −2 V at a concentration of 1000 mM, which, to the best of our knowledge (comparison with other works is provided in Supplementary Table 1), is the highest ever reported (additional data can be found in Supplementary Fig. 11a, b). The conductance was highly reversible when Vg was tuned back to +1 V from −2 V (Supplementary Fig. 12a, b), a clear indicator of the stability of our devices. Furthermore, at the end of each gate voltage cycle, we measured the water conductance to check the integrity of our devices. More than 90% of our devices were found to be stable in this gate voltage range.

To understand the role of the EDL in ion transport, we examined the magnitude of voltage modulation at several KCl concentrations in the range of 1 mM‒1000 mM. For this purpose, we defined a gating ON/OFF ratio, which is the ratio of conductance at Vg = −2 V to Vg = +1 V (Fig. 3c). The gating ON/OFF ratio is smaller at a 1 mM concentration; however, it increases and saturates in the concentration range of 10 mM‒1000 mM, which is very similar to the zero Vg conductance versus concentration behavior (Supplementary Fig. 13). The increase in the gating ON/OFF ratio with increasing concentration is unusual and strikingly different from the findings of other reports12,13. It is therefore important to understand the reasons for the significant gate modulation effects at high KCl concentrations. For this, let us first consider the situation of zero gate voltage. The cations (K+) from the KCl solution are drawn to the negatively charged vermiculite surface (Supplementary Note 1) and are screened by the K+ ions when the channel surface is not subjected to any gate voltage. The electrolyte solution concentration and ion valence control the screening length, also known as the Debye length, \({\lambda }_{D}\,\), which can be computed via the following formula given as

$${\lambda }_{D}=\sqrt{\frac{\varepsilon {RT}}{{\sum }_{i=1}^{N}{F}^{2}{Z}_{i}^{2}{C}_{i,0}}}\,$$
(1)

where \(\varepsilon\) is the permittivity of the solution; R and T are the gas constant and temperature, respectively; F is the Faraday constant; N is the total number of ionic species; and \({Z}_{i}\) and \({C}_{i,0}\) are the valence and bulk concentrations of the \(i\)th ionic species, respectively. The above expression implies that the Debye length decreases with increasing ion concentration. For example, at a 1000 mM KCl concentration, \({\lambda }_{D}\) is 3 Å. The transport height for the K‒V laminate is 2.6 Å, which means that for KCl concentrations ranging from 1 mM to 1000 mM, the EDL overlaps. Because of this, the variation in ionic conductance with concentration is not significant (Supplementary Fig. 13). For example, the conductance only slightly increased from 1.31 μS to 5 μS when the KCl concentration was varied from 1 mM to 1000 mM. This relatively small change in conductance, despite a three-orders-of-magnitude increase in concentrations, suggests that ion transport is dominated by surface effects rather than bulk diffusive transport, G \(\propto\) C0 [ref. 36].

We tested whether the ion hydration shells are squeezed or the laminated membrane structure expanded upon an applied gate voltage (Vg). The in-situ XRD data (Supplementary Fig. 1b) confirm that the laminated structure remains intact after a high negative Vg is applied. We tried to quantify the softness of the ion hydration shell, i.e., the extent of ion‒water interactions and their modification under the influence of a gate voltage. We performed ionic conductance measurements at several temperatures and gate voltages to extract the activation energy (\({E}_{a}\)). Figure 3d displays the Arrhenius plot between ln(G) and 1/T, at different Vg values. \({E}_{a}\) is calculated using the following formula

$$G\propto \exp \left(\frac{{-E}_{a}}{{RT}}\right)$$
(2)

where R is the universal gas constant, and the sample temperature T, varies from 25 to 60 °C. Our diffusion study revealed that K‒V laminates are cation-selective due to the high density of negative surface charges (Supplementary Fig. 14a, b & Supplementary Table 2). Therefore, we can readily ascribe the activation energy, \({E}_{a}\), to that of K+ ions. At 0 Vg,\(\,{E}_{a}\) is 298 meV. When Vg is +1 V, the \({E}_{a}\) value increases to 309 meV, which explains the reduced ionic conductance observed for positive Vg (Fig. 3b). Interestingly, when Vg = −2 V, \({E}_{a}\) decreased from 298 meV to 287 meV. The existence of a threshold voltage (Vg ≈ −1 V) and the lower \({E}_{a}\,\)(hence increased conductance) for Vg < −1 V suggests a modification in ion‒water (hydration) interactions around K+ ions.

We examined the impact of the gate voltage on the transport of higher valency cations using the electrolytes CaCl2 and AlCl3, and the conductance vs. concentration characteristics at 0 Vg are displayed in Supplementary Fig. 15a, b. Figure 4a shows the I‒V characteristics at different Vg values for 500 mM CaCl2 solutions. There is a visible change in the slope of the I‒V curves with Vg. However, the current at zero voltage is lower than that of KCl solutions, especially for negative Vg. Figure 4b shows the influence of the gate potential on the ionic conductance of CaCl2 solutions at a concentration of 500 mM. For positive Vg, we observe a reduction in the CaCl2 conductance upon an increase in Vg, similar to that of KCl solutions. However, for negative Vg, the conductance reaches a maximum at Vg \(\approx \,\)−1 V, beyond which it decreases. We performed these measurements at several chloride concentrations ranging from 1 mM to 1000 mM (Fig. 4b). When we increased the solution concentration from 1 mM, the maximum conductance, which occurred at around Vg \(\approx\)−1 V for all the concentrations, also increased. We also wanted to confirm whether this observation holds for a trivalent cation such as Al3+. For this purpose, we chose aqueous solutions of AlCl3 with chloride concentrations ranging from 1 mM to 1000 mM. In this case, we also observed a maximum in the conductance, but at Vg \(\approx \,\,\)0 V (Fig. 4c, d). The maximum conductance increased when we increased the solution concentration, a trend similar to that of CaCl2 solutions.

Fig. 4: Voltage-gated ion transport through Ca‒V and Al‒V devices, and the effect of ion valence.
figure 4

a Variation in the ionic current, Ids with the applied potential, Vds, for 500 mM CaCl2 solutions at various Vg. b Conductance vs Vg plot for 0.5 mM, 50 mM, and 500 mM CaCl2 solutions. Data are presented as mean values ± standard deviation. N = 3 independent measurements. c Variation in the ionic current, Ids with the applied potential, Vds, for 333 mM AlCl3 solutions at various Vg. d Conductance vs Vg plot for 0.33 mM, 33 mM, and 333 mM AlCl3 solutions. Data are presented as mean values ± standard deviation. N = 3 independent measurements. Source data are provided as a Source Data file.

Full size image

The zero Vg conductance of CaCl2 and AlCl3 is lower than that of KCl. Our estimated transport height for Ca‒V and Al‒V is 5.6 Å and 5.1 Å, respectively. It can accommodate 1–2 layers of water molecules as the bare sizes of Ca2+ and Al3+ ions are 2.02 Å and 1.06 Å, respectively37. Theoretical studies suggest that the first hydration layer is very strong in the case of Ca2+ and Al3+, which is difficult to remove as penalties of −1504 kJ/mol and −4665 kJ/mol energy are needed, respectively37. Therefore, major modification in the ion‒water interaction is unlikely even with the largest Vg that we have applied. Like in KCl solutions, the CaCl2 and AlCl3 conductance saturates at high concentrations (Supplementary Fig. 15a, b). Contrary to the KCl results, we observed a decrease in conductance with negative Vg, which suggests the emergence of a different interaction. Moreover, the maximum conductance decreases with decreasing ion concentration. Our diffusion data (Supplementary Fig. 14a, b) and zeta potential measurements (Supplementary Fig. 5) indicate that the cation selectivity of vermiculite is in the order of KCl > CaCl2 > AlCl3.

We observed that the gate voltage dependent ionic conductance behavior of vermiculite laminates that are intercalated with KCl, CaCl2, and AlCl3 solutions is strikingly different from other reported membranes that are unintercalated, such as on MXenes13. The possibility of water splitting at extreme values of Vg is ruled out simply because the conductance of CaCl2 and AlCl3 solutions decreases with Vg beyond −1 V. In the case of K+ ion transport through K‒V laminates, we see a nonmonotonic increase in conductance with negative Vg. Recently, an atomic transistor consisting of rGO showed an increase in K+ transport when a gate voltage was applied14. In this case, the density of ions increases, and the barrier energy for ions decreases with Vg, which results in increased conductance. The barrier energy is exponentially related to the ion mobility (\({{{\rm{\mu }}}}\)) as

$${{{\rm{\mu }}}}={{{{\rm{\mu }}}}}_{0}\,{exp }^{\left(\frac{-{E}_{a}}{{RT}}\right)}$$
(3)

where µ0 is the bulk ion mobility (i.e. for K+, µ0 = 7.62×10-8 m2V-1s-1). A decrease in the barrier energy clearly leads to an increase in ion mobility and hence in the conductance.

We also checked whether the gating ON/OFF ratio depends on the cation’s energy barrier. For this purpose, we chose a Na‒V membrane with NaCl solutions at a concentration of 1000 mM. We found that the gating ON/OFF ratio decreased to \(\sim\)1100% (Supplementary Fig. 16a, b), which could be due to its (Na+) slightly larger hydration barrier than K+. Additionally, with increasing membrane thickness (K‒V, h = 8 µm), the overall ionic conductance increases, but the gating ON/OFF ratio decreases to \(\sim\)1100% (Supplementary Fig. 17a, b). This result suggests that the gating effect might be more pronounced in thinner than in thicker membranes.

To understand the ion transport mechanism under gate voltages for K‒V and Ca‒V systems, we conducted all-atom MD simulations. The MD simulation setup is illustrated in Fig. 5a. We considered transport heights of 2.6 Å and 5.6 Å for the 1000 mM KCl and 500 mM CaCl₂ systems, respectively, matching the experimental values. To emulate the gating effect, a net charge \(\Delta {Q}_{{{{\rm{net}}}}}\) was introduced to the membrane atoms38 using the distribution equation as

$$\Delta {q}_{i}={\Delta Q}_{{{{\rm{net}}}}}\frac{\left|{q}_{i}\right|}{{\sum }_{i}\left|{q}_{i}\right|}$$
(4)

where \({q}_{i}\) represents the partial charge of atom \(i\) in the membrane. The charge modification applied was within 1.5% of the original partial charge. Ideally, the surface charge density, \(\sigma\), has a linear relation with \({V}_{g}\) which is given below

$$\sigma=\frac{{\Delta Q}_{{{{\rm{net}}}}}+{Q}_{0}}{A}=C{V}_{g}+{\sigma }_{0}$$
(5)

where A is the surface area, \({C}\) is the capacitance, and \({Q}_{0}\) and \({\sigma }_{0}\) are the net charge of the membrane and the surface charge density at \({V}_{g}=0\), respectively. An external electric field corresponding to 1 V was applied to induce ion transport. Additional methodological details of the MD simulations are provided in the Methods section.

Fig. 5: Molecular dynamics (MD) simulation of ion transport through Å-scale 2D vermiculite membranes.
figure 5

a Snapshot of the simulation systems for 1000 mM KCl (top) and 500 mM CaCl₂ (bottom) solutions. b Effect of the gating voltage on the ion distribution and transport through the vermiculite interlayer spacing. c-d Cation current as a function of surface charge density for c 1000 mM KCl and d 500 mM CaCl₂. The error bars for the current were estimated as the standard deviation of the slope obtained from a linear regression of the flux-time data, sampled every 50 ps over a total simulation time of 200 ns. The centre of the error bar is the mean. e, f Density of intercalated ion as a function of surface charge density for e 1000 mM KCl and f 500 mM CaCl₂. The error bars for the ion density were estimated as the standard deviation of the ion density, sampled every 50 ps over a total simulation time of 200 ns. The centre of the error bar is the mean. g, h Mobility of intercalated ions as a function of surface charge density for g 1000 mM KCl and h 500 mM CaCl₂. Source data are provided as a Source Data file.

Full size image

The surface of the vermiculite membrane is terminated with oxygen atoms, naturally generating negative electrostatic dipoles directed toward the interlayer space, thereby providing binding sites for cations. This locally negative electrostatic environment within the highly confined vermiculite interlayer attracts cations, even when the membrane is electrically neutral. Figure 5b shows snapshots of ions in K‒V and Ca‒V systems, demonstrating that a negative gate voltage generally increases the population of intercalated cations, whereas a positive gate voltage decreases it. Note that cation transport through this highly confined space involves a sequential hopping process between binding sites created by the oxygen-terminated surface. Such a sequential, barrier-leaping ion transport mechanism has been historically studied in the context of biological nanochannels39 and has also been documented in synthetic membranes40. Our MD simulations for cation currents in the K‒V and Ca‒V systems show qualitatively consistent trends with the experimental observations (Fig. 5c, d). In the K‒V system, the cation current increases with decreasing σ (decreasing Vg), whereas the Ca‒V system exhibits a peak current at a specific σ value. We calculated average ion density within the interlayer space as a function of surface charge density (see Fig. 5e, f). In the K‒V system (2.6 Å), only K⁺ ions intercalate, whereas in Ca‒V systems (5.6 Å), both Ca²⁺ and Cl⁻ ions occupy the interlayer space. Figure 5g, h shows the average mobility of ions between the membranes. In the K‒V system, K⁺ mobility increases with decreasing surface charge. In the Ca‒V system, the Ca²⁺ mobility peaks at σ ≈ −0.4 C/m², whereas the Cl⁻ mobility decreases with decreasing σ.

We closely examined ion trajectories at different gate voltages in our MD simulation, leading to the following insights into ion transport mechanisms (illustrated in Fig. 5b): For the K‒V system at negative Vg, the density of K+ increases and K⁺ ions undergo frequent, short-distance hopping, resulting in high current (Supplementary Movie 1). At a positive Vg, K⁺ ions become depleted and exhibit infrequent hopping over relatively long distances, leading to a low current (Supplementary Movie 2). For the Ca‒V system, Ca²⁺ ions become less mobile at strongly negative Vg because of the strong Ca‒V interaction (Supplementary Movie 3). At positive Vg, Ca²⁺, and Cl⁻ ions move in opposite directions, interfering with each other’s transport and resulting in a congested hopping process that reduces Ca²⁺ mobility (Supplementary Movie 4). As a result of reduced mobility at both the negative and positive Vg limits, the Ca2+ current reaches its maximum at the intermediate Vg value (Supplementary Movie 5). Theoretically, the mobility of K⁺ ions can also be reduced at even higher negative surface charges. To further understand the effect of gating on ionic mobility within these Å scale channels, an energetic analysis of each ion may be necessary, although this is beyond the scope of this work.

We also investigated a commonly used GO system to infer the gating effect at 1000 mM, where the Debye length does not overlap with the transport height ( ~ 7 Å). We found no evidence of a gating effect (Fig. S18a–d).

The significant electric field effect observed in vermiculite membranes is a result of (i) the considerable length of the gate electrodes (\(\sim \,\)2 mm), which enhances cation accumulation when negative gate voltages are applied, and (ii) the \(\sim \,\)1 nm thick insulating vermiculite layers capacitively induce charges, and the presence of 1 to 2 layers of water ensures weaker electrostatic screening, resulting in a field effect on the fluidic channels. Theoretical simulations indicate that thinner dielectric layers induce higher surface charge and zeta potential41. Recent reports of the electric field effect in conducting membranes such as GO and MXenes without the use of a gate oxide for gating are puzzling; however, it may be argued that there are thin insulating layers in the form of functional groups, which might help induce surface charges and hence tunability of conductance or permeation. Our device configuration is inspired by solid-state field-effect transistors; thus, the interpretation of the results is straightforward. Mica has a very high breakdown strength of 12 MV/cm [ref. 42], which is expected to be similar for vermiculite. Therefore, the probability of the formation of conducting filaments in our membranes is very unlikely since our maximum applied voltage is only 2 V across a membrane thickness of 3.5 μm.

The gated ion transport device demonstrated in our study is effective in modulating the conductance of monovalent ions and, as such, holds promise for several advanced applications. First, the precise control over ion flow in angstrom-scale channels makes these devices excellent candidates for lab-on-chip systems, where accurate dosing and real-time regulation of ionic species are critical. For instance, in controlled drug delivery, the ability to selectively modulate the transport of monovalent ions could be leveraged to trigger downstream chemical or biological processes with high temporal resolution. Additionally, the distinct response observed for monovalent vs. multivalent ions‒evidenced by the contrasting transconductance behavior under negative gate voltages‒suggests that these devices could be engineered to separate ions based on their valence. This selective ion separation could have significant implications in water purification, where the removal of specific ion types is desired, as well as in the design of fuel cells and batteries. In these energy applications, energy-efficient ion gating is essential to optimize performance and reduce losses. The inherent modularity of the vermiculite membrane system allows for further functionalization. For example, by incorporating inorganic pillars (such as alumina) to expand the van der Waals gap, the device architecture could be tuned to accommodate larger species43. This would enable the transport of biomolecules (e.g., DNA, RNA, proteins), finding applications in biosensing and gate-controlled sequencing and detection platforms.

We demonstrated how gate voltage alters ion-water and ion-surface interactions at room temperature in highly confined vermiculite laminates. Negative gate voltages increased the cation concentration inside the channels, resulting in unusual ion gating effects. We observed a sizeable ionic conductance modulation of \(\sim\)1400% in 1000 mM KCl solutions. In the same voltage range, much-reduced conductance modulation was observed in the CaCl2 and AlCl3 solutions, resulting from ion-surface interactions inferred from the MD simulations. Future research must focus on high salt concentrations and extreme confinements to understand and explore the whole dynamics of ion interaction effects. This study reveals how to treat industrial waste of high salt concentrations efficiently. Overall, two-dimensional materials offer endless possibilities for fabricating tunable channels at the Å-scale.

Methods

Chemicals

Natural vermiculite crystals (2‒3 mm in size), potassium chloride (KCl, \(\ge \,\)98.5%), sodium chloride (NaCl, \(\ge\) 99.0%), lithium chloride (LiCl, \(\ge\)99.0%), calcium chloride (CaCl2, \(\ge\)98.0%), aluminum chloride (AlCl3, \(\ge\)97.0%), PVDF (0.22 µm pore size), and silver paste were purchased from Sigma Aldrich. All the chemicals purchased were used as received. Gold wire (30 μm in diameter) was purchased from Tanaka K.K. (Japan).

Fabrication of vermiculite laminates

The natural vermiculite crystals obtained were thermally expanded, and the cations were exchanged via a two-step method25. Briefly, in the first step, 100 mg of vermiculite crystals were soaked in a 200 mL saturated NaCl solution and refluxed for 24 h at 100 °C, followed by washing with deionized water 8 to 10 times to remove excess salts. During this process, interlayer Mg2+ cations are exchanged with Na+ ions. In the next step, the sodium-exchanged vermiculites were soaked in 200 mL of 2000 mM lithium chloride (LiCl) solution, refluxed for an additional 24 h, and again washed with DI water until excess chloride ions were removed. The resulting Li-exchanged vermiculite (Li‒V) crystals were dried and dispersed in water at a 1 mg/mL concentration and sonicated for 30 min. The monolayers were obtained by centrifuging at 1509 × g for 15 min or allowing the large flakes to settle in the solution overnight under gravity. We used the supernatant containing the monolayer to prepare the vermiculite membrane via vacuum filtration assembly. We used PVDF of 0.22 µm pore size as the support for vermiculite membranes, which easily peeled off after drying the samples under an IR lamp for 10 min. These free-standing membranes were dipped into 1000 mM chloride concentration of KCl, CaCl2, and AlCl3 solutions for 24 h, which resulted in the intercalation of the cations K+, Ca2+, and Al3+, respectively, into the vermiculite interlayers. Our previous study25 reported that the above cations make vermiculite membranes highly water-stable. We subsequently washed the cation exchanged membranes with water to remove excess salt from the surface and dried them under an IR lamp. These materials were further characterized and used for ion transport studies.

Characterization

A Rigaku Multipurpose X-ray diffractometer (XRD) with Cu Kα radiation (λ = 1.5406 Å) was used to determine the interlayer spacing of the cation-exchanged vermiculite membranes. The thickness of the exfoliated vermiculite flake was determined using atomic force microscopy (Bruker Nano wizard Sense AFM), the membrane thickness was determined from cross-sectional SEM images (JEOL JSM-7900F), and a zeta potential measurement was carried out via DLS and Zeta Seizer instrument (Nano ZS Malvern Instrument). Energy dispersive spectroscopy analysis (EDS, JEOL JSM-7900F) was performed to determine the elements present in the vermiculite membranes. Fourier transform infrared spectroscopy (FTIR) (Perkin Elmer), and Raman spectroscopy (WITec) were used to confirm the functional groups present in vermiculite. For Raman spectroscopy, 532 nm laser light with 5 mW of power was used for excitation at room temperature. X-ray photoelectron spectroscopy (XPS) was also carried out to determine the elemental details. XPS spectra were acquired with an ESCALAB 250 XI (Thermo Fisher Scientific, source: Mg Kα 300 W, pass energy: 40 eV) system, where the analysis chamber was pumped down to ultrahigh vacuum (UHV \(\sim\)5 × 10-10 mbar). The mechanical testing was carried out using a universal testing machine (UTM, model number- Kappa SS_CF100).

Molecular dynamics simulation

All-atom molecular dynamics (MD) simulations were performed to investigate ion transport through Å-scale two-dimensional (2D) vermiculite membranes. The simulations were conducted using periodic rectangular simulation boxes with dimensions of 3.66 nm × 4.25 nm × 15.0 nm (x, y, z) for the 1000 mM KCl system and 4.56 nm × 4.25 nm × 15.0 nm for the 500 mM CaCl₂ system. Three layers of vermiculite membranes were stacked along the x-direction and positioned at the center of the z-coordinate system, as depicted in Fig. 5a. The vermiculite membranes and ions were modelled using the Clay Force Field44 and water molecules were described by the standard simple point charge (SPC) model45. To emulate the gating effect of the membranes, a net charge (\({Q}_{{{{\rm{net}}}}}\)) was added to the membrane atoms using the distribution equation described in the main text. The systems were solvated with water molecules and ions with additional cations added to neutralize the system. The positions of the vermiculite membrane atoms were restrained using a harmonic spring. Prior to production runs, the system underwent an initial energy minimization, followed by equilibration under the NPT ensemble with a variable z-box size at 300 K and 1 bar for 2 ns. The barostat and thermostat relaxation times were set to 2 ps and 0.1 ps, respectively. Following equilibration, an external electric field of 0.067 V/nm was applied along the z-direction, corresponding to a 1 V transmembrane bias, and the simulation was conducted under the NVT ensemble. Ion transport simulations were performed for 200 ns, and the last 100 ns of trajectory data were used for analysis. All MD simulations were performed using a GPU-accelerated MD code with GROMACS46,47 version 2024.2, and atomic visualizations were generated using OVITO48.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.