MoO_x powder, carbon black (Super P, SZKejing, Shenzhen), and polyvinylidene fluoride (PVDF, SZKejing, Shenzhen) binder were dispersed in a N-methyl-2-pyrrolidone (NMP, AR, Aladdin, Shanghai) with a mass ratio of 8:1:1. Then, the mixture was stirred under a speed of 200 rpm at room temperature for 8 h to form a homogeneous slurry. The slurry was blade–coated on carbon nanotube (CNT) film and dried at 60 °C for 12 h, achieved MoO_x/CNTs film electrode. The thickness of the doctor blades was 100 μm, and the mass loading of MoO_x in MoO_x/CNTs film was 1.5 mg cm^-2. As for active carbon (AC)/CNTs composite electrode, AC powder (XFNANO, Jiangsu), carbon black, and PVDF binder were dispersed in a NMP with a mass ratio of 8:1:1. Then, the mixture was stirred under a speed of 200 rpm at room temperature for 8 h to form a homogeneous slurry. The slurry was blade–coated on CNT film and dried at 60 °C for 12 h, formed the AC/CNTs composite electrode. The thickness of the doctor blades was 300 μm, and the mass loading of AC in AC/CNTs film was 4 mg cm^-2.

PVA/H₃PO₄ electrolyte (60 µL) was drop–coated on a piece of positive AC/CNTs electrode. Then, the above electrode was placed in a vacuum oven to make more polymer electrolyte infiltrating into electrode. The CAPode was assembled by pressing one AC/CNTs electrode coated with PVA/H₃PO₄ electrolyte and another negative MoO_x/CNTs electrode together.

Scanning electron microscopy (SEM) images were obtained by high-resolution field emission scanning electron microscopy (FESEM, Hitachi S–4800, Japan) with an accelerating voltage of 5 kV. Transmission electron microscopy (TEM) images and selected area electron diffraction (SAED) patterns were taken out by high-resolution transmission electron microscopy (HRTEM, JEOL–2010, Japan). The X-ray diffraction patterns were characterized by X-ray diffraction spectroscopy (XRD, D8 Advance, Bruker, Germany) with Cu–kα radiation (V = 30 kV, I = 25 mA). X-ray photoelectron spectroscopy (XPS) spectra were collected by using Shimadzu Kratos, AXIS Ultra DL (Japan), and the binding energy data were corrected to C 1s = 284.6 eV. Renishaw Raman spectrometer (Britain) equipped with a 514 nm laser is used to characterize the structure of electrode materials.

The electrochemical performance was measured by an electrochemical working station (CHI 760 E, Shanghai Chenhua). The current-voltage (I-V) characteristics of the devices were measured on a source meter (2400, Keithley, America). Unless specifically mentioned, all the electrochemical performances is calculated based on active materials (i.e., MoO_x, AC), and the related formulas are shown in Supplementary Materials.

MoO_x nanobelts containing abundant oxygen vacancies were prepared by a one–step hydrothermal method (see the experimental section for details). The crystal morphology of MoO_x nanobelts was investigated by HRTEM [Figure 2A and Supplementary Figure 1], from which mutually perpendicular lattice stripes of MoO_x nanobelts can be observed. The lattice spaces of 0.18 and 0.2 nm [Figure 2A] were corresponded to the (002) and (100) crystal planes at 90° angle in the SAED diagrams [Figure 2B], respectively. Many discontinuities or distortions in the MoO_x lattice stripes can be observed in the HRTEM images, which indicated the presence of oxygen vacancies in the MoO_x nanobelts. The SEM image [Figure 2C] and energy-dispersive X-ray spectroscopy (EDS) elemental mapping [Figure 2D and E] showed that the elements of molybdenum and oxygen were uniformly distributed in the MoO_x nanobelts. From Supplementary Figure 2, it can be seen that the as–synthesized MoO_x had a nanobelt structure with a width of about 300 nm and a length of 10 μm. The XRD pattern of MoO_x [Supplementary Figure 3] showed obvious diffraction peaks at (020), (040) and (060) crystal planes, which well matched with the reference card of MoO₃ (PDF#05–0508), confirming its monoclinic phase derived from an orthorhombic phase with Panm space^[24,25].

The Raman spectrum of MoO_x in Supplementary Figure 4 appeared typical fingerprint vibrational modes at 291, 666, 819 and 996 cm^-1. More specifically, the Raman peak at 291 cm^-1 represented the swing mode of the double bond O=Mo=O, whereas the Raman peaks at 666, 819, and 996 cm^-1 responded to the stretching modes of the triple–coordinated oxygen (Mo–O_s), double–coordinated oxygen (Mo–O_a), and terminal oxygen (Mo=O_t), respectively. Figure 2F showed the XPS full spectrum of MoO_x, which appeared typical peaks of Mo 3p, Mo 3d, and O 1s. The XPS spectrum of O 1s in MoO_x [Figure 2G] showed the presence of oxygen defects in addition to lattice oxygen. The oxygen vacancies in MoO_x nanobelts can expand the Van der Waals gap and regulate the electronic structure of Mo atoms, which will result in enhancement of the charge storage kinetics and the capacity of MoO_x^[26,27]. The fine spectra of MoO_x [Figure 2H] had a twin peak of Mo 3d_3/2 at 236.23 eV and Mo 3d_5/2 at 233.07 eV. The XPS result showed that MoO_x was mainly composed of Mo⁶⁺ ions (90.11%) and a small percentage of Mo⁵⁺ ions (9.89%).

CNT films were used as current collectors to load active materials of MoO_x and AC, and PVDF used as binders. From Figure 2I to L, all the I-V curves of both MoO_x/CNTs electrode and AC/CNTs electrode well remained as they were bent to different angles and after different bending cycles, indicating no obvious changes of their electrical conductivities. The SEM results [Figure 2M and N, Supplementary Figures 5 and 6] showed that the conductive networks composed of MoO_x nanobelts and AC almost unchanged after 1,000-3,000 bending cycles. The excellent structural and electric stabilities of electrodes play an important role in the following demonstrated flexible CAPode devices.

The adsorption energies of H⁺ and PO₄^3- ions at various active sites in MoO_x were calculated using density functional theory (DFT), providing a theoretical evaluation of the ion sieving capacity of MoO_x. Supplementary Figure 7 present the adsorption energies of H⁺ and PO₄^3- ions on two types of oxygen sites. The adsorption energies for H⁺ ions on O_t and O_a are about -0.66 and -0.88 eV, respectively, which indicates that H⁺ ions can readily interact with both terminal and asymmetric oxygen sites, facilitating their intercalation into MoO_x. In contrast, the larger-size PO₄^3- ions and their electrostatic repulsion with negatively charged oxygen atoms result in much higher adsorption energies of 6.99 eV on O_t and 8.52 eV on O_a. As a result, the small ions of H⁺ in the electrolyte can easily intercalate the MoO_x active sites, making the CAPode chargeable under forward bias.

As shown in Figure 3A, the CAPode was assembled by directly pressing one electrode of MoO_x/CNT and another electrode of AC/CNTs together with the polymer electrolyte in between. The cross-sectional SEM image of the whole device [Supplementary Figure 8] showed that the CAPode had a typical sandwiched structure, and the thickness of MoO_x and AC was about 12 and 50 μm, respectively. The thickness of PVA/H₃PO₄ was about 50 μm. Figure 3B and C showed cyclic voltammetry (CV) curves of the CAPode, which exhibited a typical asymmetric shape. From the CV curves, it can be seen that the CAPode prossessed a unidirectional charging characteristic with a threshold voltage of about 0 V. This unidirectional charging behavior can be attributed to the selective intercalation/deintercalation of ions in/out MoO_x. In the negative voltage range from -0.95 to 0 V, the small cations of H⁺ can intercalate into and de-intercalate from interlamination of MoO_x, which was corresponding with pseudocapacitive redox peaks in CV curves. In contrast, there was negligible response current in the voltage range from 0 to +0.95 V, because only limited large-size PO₄^3- anions were absorbed on the external surface of MoO_x electrode. The large difference of response current between the positive and negative voltage regions revealed the excellent ion–sieving ability of the MoO_x electrode. When the scan rate increased from 10 to 60 mV s^-1, the shape of CV curves remained almost the same, the peak current increased linearly with the threshold voltage stabilized at about 0 V, suggesting the rapid intercalation and deintercalation process of H⁺ ions. The rectification capability of CAPode is evaluated by RR_I and RR_II. RR_I is defined as the ratio of the redox peak currents in negative and positive voltage ranges, and RR_II is defined as the ratio of the capacitances in negative and positive voltage ranges^[11]. Benefiting from the charge storage capability of MoO_x under negative bias, the RR_I of the developed CAPode reached a maximum value of 14.62 at a scan rate of 10 mV s^-1, and remained as high as 10.82 even at a higher scan rate of 60 mV s^-1, indicating excellent rectification performance [Figure 3C].

As shown in Figure 3D and E, the galvanostatic charge–discharge (GCD) curves of the CAPode under different current densities had a shape of quasi-triangle, and also exhibited a unidirectional charging behavior. Almost all of the capacity of the CAPode was contributed by the potential range from -0.95 to 0 V, which was highly consistent with the results of the CV tests. The specific capacitance and RR_II of the MoO_x–based quasi-solid-state CAPode were further calculated according to the GCD curves. As shown in Figure 3E, the specific capacitance of the MoO_x–based CAPode reached 139.32 F g^-1 at the current density of 0.5 A g^-1, and the RR_II of the CAPode was as high as 96.18%. Even the charge/discharge current density further increased to 7 A g^-1, the MoO_x–based quasi-solid-state CAPode also exhibited a high specific capacitance of 41.34 F g^-1 and a RR_II of 90.73%, indicating its excellent rate capability.

To further reveal the charge storage mechanism and valence bonds’ changes of MoO_x electrode during process of charging/discharging, in-situ XPS of MoO_x was conducted, with MoO_x as working electrode, AC as counter electrode, Ag/AgCl as reference electrode, and 0.2 M H₃PO₄ solution as electrolyte, respectively. From XPS spectra of O 1s and Mo 3d in MoO_x electrode at different states [Figure 3F-I], it can be seen that Mo⁶⁺ and Mo–O species decreased at discharged state, while the Mo⁵⁺, Mo⁴⁺ and Mo–O–H species increased obviously. It can be concluded that the redox centers were the molybdenum atoms in MoO_x electrode, and Mo–O bonds also played an important role to allow H⁺ intercalation.

As shown in Figure 3J, the MoO_x–based quasi-solid-state CAPode can maintain 83% of the original capacity and almost 100% of the original coulombic efficiency after 2,000 charge/discharge cycles at a high current density of 2.0 A g^-1, which indicated the CAPode with excellent cycling performance [Supplementary Figures 9 and 10]. Compared with other cations, we used PVA/H₃PO₄ as the gel electrolyte, due to H⁺ ions can provide relatively fast ion transport kinetics and high energy storage ability. For comparison, several commonly used gel electrolytes (PVA/LiCl, PVA/KOH, PVA/H₂SO₄, and PVA/HClO₄) were also investigated to validate their feasibility for CAPodes. As Supplementary Figures 11 and 12 showed, the response currents of CAPodes based on both electrolytes of PVA/LiCl and PVA/KOH in the potential range from -0.95 to 0 V were much lower than that of CAPode based on PVA/H₃PO₄ electrolyte, resulting in lower rectifications (RR_II = 79.52% and 84.86%), respectively, due to the sizes of Li⁺ and K⁺ larger than H⁺. The CAPode based on electrolytes of PVA/H₂SO₄ and PVA/HClO₄ delivered RR_II ranging from 96.12% to 95.26% [Supplementary Figures 13 and 14], but their cycling stabilities [Supplementary Figure 15] were rather low, because the strong acidity of H₂SO₄ and HClO₄ will inevitably corrode the MoO_x electrode material. Although the polymer chains may hinder ions movement of polymeric electrolytes^[28,29], our CAPodes achieved a well balance between energy storage efficiency and rectification performance through tuning the ionizable components of the electrolytes. For comparison, the constructed CAPode not only possessed high energy storage capacity and stability, but also exhibited comparable rectification capability with reported CAPodes and ion diodes [Figure 3K]^{[9,11,12,23,30-39]}.

Mechanical flexibility is essential to meet the demands of flexible and wearable electronics^[40]. To evaluate the flexible stability of the developed CAPodes, we investigated their rectification ratio and capacitance retentions as the devices under different bending angles and bending cycles. As Figure 4A showed, there were no obvious structural delamination or damage as the prepared device was bent under different angles. It can be seen that both CV [Figure 4B] and GCD [Figure 4C] curves of the CAPode almost overlapped as the device was bent from 0° to 180°. From the electrochemical impedance spectroscopies (EIS, Figure 4D), the series resistance and ionic diffusion ability of the CAPode almost unchanged under different bending angles, which was the reason why the device can well maintain its electrochemical performance under bending states. Even after 3,000 repeated bending cycles from 0° to 180°, all the CV, GCD and EIS curves [Figure 4E-G] still maintained as its original states, which indicated the CAPode with excellent flexible stability.

Figure 5A and B schematically showed the simplified equivalent circuits of the CAPode under forward bias and reverse bias, similar to the unidirectional conduction characteristics of semiconductor diodes. A series of EISs under biases from -0.95 to +0.95 V were measured to analyze the rectification performance of the CAPode. From Figure 5C, the impedance in low–frequency region of the Nyquist plot obviously decreased as the bias gradually decreased from 0 to -0.95 V, which revealed the efficient acceleration of ion migration in this process^[41,42]. Correspondingly, the phase angle of the Bode phase diagram in the low–frequency region became smaller and smaller with the bias decreasing from 0 to -0.95 V [Figure 5D]. In contrast, both impedance and phase angle slightly increased as the bias increased from 0 to +0.95 V, which revealed that there were few ions contributed to energy storage process because of their large size. The results were well corresponding to the electrochemical behaviors discussed above, which further confirmed the working mechanism of the developed CAPodes.

To evaluate the feasibility of MoO_x–based CAPode for electronic circuits and logic gates, the current rectification capability was investigated by the standard chronoamperometry (CA). Before CA test, the initial open-circuit voltage of CAPode was set at 0 V for a period of time to eliminate the disorderliness of ions until the value of current was stable. As Figure 5E showed, the instantaneous current was as high as 20.22 mA at -0.95 V, while the instantaneous current was as low as 1.86 mA at +0.95 V. After removing of the loading bias voltage, the current response under +0.95 decreased faster than under -0.95 V, due to its mechanism of ions surface adsorption rather the mechanism of intercalation/deintercalation. As a result, a rectification ratio of the CAPode between the response currents under negative and positive bias voltages reached a maximum value of 33.6 at 2.2 s and gradually decayed and stabilized at 15.7 after 600 s [Figure 5F], which was obviously different from that of semiconductor diodes (constant rectification ratio after applies bias voltage). Figure 5G showed the current-time (I-t) curves of the CAPode with alternating voltages of -0.95 and +0.95 V applied for 1 s once a time for a few cycles. It can be seen that the final current at -0.95 V was about four times of that at +0.95 V. Both currents will increase with the applied pulse time increasing, but the final current at -0.95 V was always higher than that at +0.95 V [Supplementary Figure 16], due to the intrinsic feature of the CAPode. From I-V curves shown in Figure 5H, the CAPode was on with a charging current of -22.8 mA under -0.95 V, while it was off with a charging current of 1.9 mA under +0.95 V. Distinct “ON” and “OFF” states of the CAPode can also be realized under low forward and reverse bias voltages (e.g., ±0.6 and ±0.3 V, Supplementary Figure 17), suggesting its excellent unidirectional charging capability.

The rectification characteristic of the CAPode endows it to be used for transmitting, regulating, and establishing signaling patterns of ions and/or biomolecules^[43]. By using two identical CAPodes, logic gates of “AND” and “OR” were constructed and investigated. Figure 6A and B showed the equivalent circuit diagrams of an “AND” logic gate and its corresponding truth table, respectively. In case of the “AND” logic gate [Figure 6C], two CAPodes with the same direction were applied a negative bias voltage (-0.95V, denoted as “1”), the entire circuit conducted and output a high current (denoted by “on”). As a positive voltage (+0.95 V, denoted as “0”) was applied on two CAPodes, the entire circuit was off, almost without current passing through (denoted by “off”). With one CAPode applied a positive voltage and another CAPode applied a negative voltage, the circuit can also achieve a state of “off”, but needing long time. Figure 6D and E showed the equivalent circuit diagram of an “OR” logic gate and its corresponding truth table. For the “OR” logic gate, as a negative bias voltage of -0.95 V was applied to either CAPode A or CAPode B, the circuit was conductive, but the output current was lower than that of both CAPodes applied with a bias voltage of -0.95 V [Figure 6F]. As a positive bias voltage of +0.95 V was applied on both CAPode A and CAPode B at the same time, almost no current can pass through the circuit.

To validate the feasibility of the CAPodes for practical applications, logic gates were constructed to turn on/off a light emitting diode (LED) bulb, with the CAPodes as the power supply at the same time. To meet the voltage requirement of the used LED, two identical CAPode connected in series that had a widen voltage of 1.9 V [Figure 6G and Supplementary Figure 18] were used as one unit of ionic diode to build the logic gates. CA curves of the logic gates (“AND” and “OR”) by alternately applying -1.9 V and +1.9 V under different inputs of [(0, 0), (1, 1), (0, 1) and (1, 0)] were measured to evaluate the response frequency of the whole logic gates. Before each external bias voltage was applied, a 0 V interval was set to make logic gates return to their initial states. As Supplementary Figures 19 and 20 showed, the current under -1.9 V decreased more slowly than that under +1.9 V because of the unidirectional energy storage feature of CAPode. Under -1.9 V, final value of current (1, 1) was always higher than that under +1.9 V. While the logic gates (“AND” and “OR”) showed close current responses as the bias voltages were applied under (0, 1) and (1, 0) inputs for different time [Supplementary Figures 21 and 22]. The results demonstrated that the response speed of logic gates based on CAPodes was fast and stable.

As shown in Figure 6H, a LED was switched by an “AND” logic gate based on the CAPodes, where it can be lit up only when both units were under negative bias voltages (1, 1). As for the “OR” logic gate [Figure 6I], the LED can be switched on as the logic gate under states of (1, 1), (1, 0) and (0, 1), and it will be switched off when the both units were applied with positive bias voltage in a state of (0, 0). Furthermore, the outputs of logic gates did not change as the CAPode units were bent to 180° [Supplementary Figure 23], and the I-t curves of logic gates (“AND” and “OR”) under different inputs (different bias voltages applied for 5 and 10 s) almost overlapped as the CAPode units were bent to different angles [Supplementary Figure 24], indicating the excellent flexible stability of the logic gates derived from CAPodes. The CAPode–based logic gate integrated with both functions of current rectification and energy storage does not require other external conducting wires, which can efficiently simplify the structure of logic circuit and reduce the power consumption.