## Introduction

Herein, by assembling an n-Si/CoOx Faradaic junction photoelectrode and a MnOx counter electrode into a two-electrode device (Supplementary Fig. 1b), we find a photovoltage memory effect with the dark output voltage of the Faradaic junction device equal to the photovoltage. Further studies suggest that a Faradaic junction device with the photovoltage memory effect can minimize the interface energy loss and leads to much higher output performance of the device than one without the effect. In order to investigate the mechanism on the photovoltage memory effect, we develop an open circuit potential method to real-time monitor the photo charge and dark discharge processes of a photoelectrode and a counter electrode, respectively. According to the method, we obtain a prerequisite for photovoltage memory effect in a Faradaic junction solar rechargeable device. Moreover, by using a semitransparent counter electrode instead of an opaque one, we remarkably simplify the configuration of the device and reduce the volume, which achieves the highest dark volumetric energy density among all reported two-electrode solar rechargeable devices.

## Results

### Coupled electron-ion transfer in a Si/CoOx photoelectrode and a MnOx counter electrode

We obtained a photoelectrode by electrodepositing a CoOx layer on monocrystalline n-type silicon and then calcinating in air at 150 oC for 1 h. Scanning electron microscopy (SEM) images show that CoOx nanosheets vertically grow on the surface of the Si and the thickness of the CoOx film is 600 ~ 700 nm (Fig. 1a and Supplementary Fig. 2). We then performed the cyclic voltammogram (CV) curve of Si/CoOx photoelectrode in a three-electrode setup in the dark and under illumination (Supplementary Fig. 3 and Fig. 1b). The Si/CoOx indicates negligible current in the dark, whereas obvious photoanodic current is observed at the potential range of −0.05 VSCE to 0.35 VSCE under illumination. When the light is off, symmetrical dark cathodic current is obtained. These results suggest that Si/CoOx can be photo-charged and dark discharged at the same potential range. Moreover, a galvanostatic charge-discharge (GCD) curve exhibits continuously adjustable potentials are obtained in the Si/CoOx19,23 (Supplementary Fig. 4), which is a prerequisite for a bias-free two-electrode solar rechargeable device19. To investigate the electrode reaction process of a Si/CoOx Faradaic junction, Raman spectroscopy was carried out (Fig. 1c)24. For the dark discharged sample, Raman peaks at 520, 597 and 688 cm−1 are observed, which correspond to Si, Co(OH)2 and Co3O4, respectively25,26. When the Si/CoOx is photo charged, a Raman peak at 597 cm−1 of Co(OH)2 disappears and two new peaks appear at 466 and 668 cm−1, which are assigned as CoOOH26. UV-Vis spectra also confirm that the reflectance of Si/CoOx decreases after Co(OH)2 is oxidized into CoOOH (Supplementary Fig. 5). X-ray photoelectron spectroscopy (XPS) was used to further investigate the Faradaic reaction in Si/CoOx (Supplementary Fig. 6). After photo charge, the intensities of satellite peak of Co2+ and OH species decrease remarkably, and Co 2p core level emission shifts to higher binding energy27, which suggest that Co is oxidized and electron transfer happens during photo charge. In order to further study ion transfer in the Si/CoOx heterojunction, we performed isotope-labeling experiments by using H218O as the solvent and time-of-flight secondary-ion mass spectrometry (TOF-SIMS) to analyze the depth profiles of 18O ions (Fig. 1d). Compared with the photo-charged Si/CoOx, the intensity of 18O ions in the sample after dark discharge increases obviously at the depth with sputtering time < 100 s, which suggests O2− ions exchange between CoOx and H218O. Therefore, coupled electron and ion transfer happens in Si/CoOx, following a possible Faradaic reaction, Co(OH)2 + h+ +OH ↔ CoOOH + H2O.

The counter electrode was prepared by electrodepositing MnOx on fluorine-doped tin oxide (FTO) substrate. SEM image indicates that the MnOx film is flake-like microstructure (Fig. 1e). The MnOx electrode exhibits nearly symmetric shape at a potential window of −0.05 ~ 0.85 VSCE (Fig. 1f) and linearly adjustable potential in GCD curve (Supplementary Fig. 7), which indicates typical capacitive features23. To further study the reaction process on MnOx electrode, we performed Raman spectra of the electrode during charge and discharge process in the dark (Fig. 1g), similar to the Si/CoOx photoelectrode. Three Raman peaks at 496, 561 and 627 cm−1 in the discharged MnOx are assigned to MnOOH28. After dark charge, the peak intensities decrease and two new Raman peaks of Mn3O4 are observed at 305 and 657 cm−1 29. Moreover, a new Raman peak at 282 cm−1 of MnOOH also appears after dark charge28, which possibly comes from the disproportionation reaction of Mn3O430. Transmission electron microscopy (TEM) images also show that the lattice spacings of MnOOH are observed in the discharged sample, and the lattice spacings of Mn3O4 are observed after dark charge (Supplementary Fig. 8). To investigate the reaction process of ions in the MnOx electrode, we used D2O as the solvent and TOF-SIMS to analyze the depth profiles of D ions (Fig. 1h). The intensity of D ions increases remarkably after charge and decreases to negligible after discharge, which suggests that Faradaic reactions intercalation and de-intercalation of D ions (3MnOOH + e + H+ ↔ Mn3O4·2H2O)31 happen in MnOx electrode.

### Photovoltage memory effect and performance of Faradaic junction solar rechargeable devices

We then fabricated a two-electrode solar rechargeable device by a Si/CoOx photoelectrode and a MnOx counter electrode (Supplementary Fig. 1b). The I-t curve of Si/CoOx/KBi(aq)/MnOx under photo charge and dark discharge without bias was measured and the result is shown in Fig. 2a. An initial photocurrent density of 1.64 mA cm−2 is observed and then the photocurrent decays rapidly in the device under illumination, which is similar to the charging behavior of a conventional supercapacitor under voltage-constant mode32. The anodic photocurrent suggests electrons flow from the Si/CoOx photoelectrode to the MnOx counter electrode through the external circuit. The photo-generated holes and electrons are stored in CoOx and MnOx, respectively. When the light is off, the device indicates an initial opposite current density of −0.40 mA cm−2 without bias, which suggests that the stored charges can go back. The asymmetric current behavior between photo charge and dark discharge comes from higher electronic conductivity33 and lower charge transfer resistance after illumination (Supplementary Fig. 9 and Supplementary Table 1). Moreover, the cycle stability of the device during photo charge and dark discharge was also measured and the results are shown in Fig. 2b and Supplementary Fig. 10. The initial coulomb efficiency is about 100% and remains stable after 80 cycles, which suggests that side reactions are negligible during photo charge and dark discharge. Therefore, the Si/CoOx/KBi(aq)/MnOx realizes reversible photo charge and dark discharge under zero bias.

We further investigated the relationship between the dark output voltage and the photovoltage of the Faradaic junction device (Fig. 2c). The Si/CoOx/KBi(aq)/MnOx device indicates a residual open circuit voltage (VR) of 0.16 V in the dark, which comes from different dark equilibrium potentials (VE) of the photoelectrode and the counter electrode. When light is on, photovoltage is generated in the Si/CoOx Faradaic junction, leading to a reverse open circuit voltage of −0.29 V. Therefore, the photovoltage can be calculated as 0.45 V. During photo charge process in short circuit, the photo-generated holes are stored in the CoOx layer and photo-generated electrons in the MnOx counter electrode, respectively, which results in an open circuit voltage decreasing gradually19. When the potential of the photoelectrode is equal to that of the counter electrode, the photo charge process ends and the open circuit voltage is 0 V. And then the light is off, the Si/CoOx/KBi(aq)/MnOx device exhibits a dark output voltage of 0.45 V, which is the same value as the photovoltage in the Si/CoOx Faradaic junction mentioned above. In contrast, a commercial silicon solar cell indicates a photovoltage of 0.58 V, but a negligible dark output voltage. Therefore, the Faradaic junction device shows a photovoltage memory effect, which is intrinsically different from the photovoltaic effect in a classic p-n junction34. Furthermore, neither such photovoltage memory effect is observed in previously reported three-electrode or four-electrode solar rechargeable devices, in which self-discharging voltages are usually lower than photovoltages35,36,37,38,39. It is possibly due to the mismatch of operation voltages between solar cells and energy storage units35,36, or high internal resistance37,38,39.

In order to investigate the effects of the photovoltage memory effect on the performance, we also replaced the MnOx counter electrode with CoOx or MoOx as reference. After optimizing the loading amount of the Faradaic materials (Supplementary Fig. 11), the photovoltages and dark output voltages of the devices with MnOx, CoOx and MoOx counter electrodes are measured and shown in Supplementary Fig. 12a and Fig. 2d, respectively. Different from Si/CoOx/KBi(aq)/MnOx, the device with CoOx counter electrode indicates lower dark output voltage than the photovoltage, while Si/CoOx/KBi(aq)/MoOx device shows a higher dark output voltage than the photovoltage (The reasons are clarified in Supplementary Figs. 1821). Therefore, the devices with CoOx and MoOx counter electrodes do not have the photovoltage memory effect. The charge quantities of photo charge and dark discharge in the three devices are calculated by integrating I-t curves23 in Supplementary Fig. 12b. The Si/CoOx/KBi(aq)/MnOx shows much higher stored charge quantity than the Si/CoOx/KBi(aq)/CoOx and Si/CoOx/KBi(aq)/MoOx (Fig. 2e), which leads to the Si/CoOx/KBi(aq)/MnOx indicating the energy density over 30 times and 1.9*104 times higher than the devices with CoOx and MoOx counter electrodes, respectively (Fig. 2f).

### A mechanism of the photovoltage memory effect

Since a Faradaic junction device with the photovoltage memory effect indicates higher performance than the devices without the effect, it is very significant to study the mechanism of the photovoltage memory effect in the Si/CoOx/KBi(aq)/MnOx device. We therefore developed an open circuit potential (OCP) method40 to real-time monitor the potentials of the photoelectrode and the counter electrode at different working stage (Fig. 3a). The measurement details are shown in Supplementary Fig. 13. When an electrode is immersed into the electrolyte, electrochemical equilibrium is established between the electrode surface and the environment. The OCP method can measure the equilibrium potential with respect to a reference electrode. The voltage of a device is the difference between the equilibrium potentials of the two electrodes41. From Fig. 3a, the Si/CoOx photoelectrode and the MnOx counter electrode are disconnected in the dark (Stage I), the OCPs of Si/CoOx and MnOx are about 0.27 VSCE and 0.13 VSCE, respectively. When Si/CoOx and MnOx are connected directly by a Cu wire in the dark (Stage II), the OCP of MnOx does not change, while the OCP of Si/CoOx is adjusted to the same potential of MnOx due to much higher density of storable charge in MnOx than Si/CoOx in the dark (Supplementary Fig. 14). When the light is on (Stage III), the OCPs of Si/CoOx and MnOx both decrease gradually from 0.13 VSCE to 0.02 VSCE, which suggests that the photo charge process happens between the two electrodes. According to Fig. 1b, f, the Si/CoOx photoelectrode and the MnOx counter electrode cannot drive water splitting reactions in the potential range of 0.13 VSCE to 0.02 VSCE, which further confirms that side reactions are negligible under zero bias. The results can understand the nearly 100% coulomb efficiency in Fig. 2b. When the light is off and the two electrodes are disconnected from each other (Stage IV), the OCP of Si/CoOx jumps to a positive potential of 0.45 VSCE, while the OCP of MnOx remains the same potential at Stage III, which leads to the output dark voltage of 0.43 V. When the photo-charged device is discharged in the dark (Stage V in Fig. 3a), the potentials of the photoelectrode and the counter electrode can recover to the initial values.

We then plotted the band diagrams of the Faradaic junction device under disconnected and connected modes (Fig. 3b). The energy band positions of the Si semiconductor and the Faradaic potential windows of CoOx and MnOx were measured by electrochemical methods. A bare Si indicates a flat-band potential of −0.11 VSCE after anodic cycling activation process due to the formation of SiOx on the surface42 (Supplementary Fig. 15). The Faradaic potential windows of MnOx and CoOx on FTO were measured by CV and GCD methods, as shown in Fig. 1f, Supplementary Figs. 7 and 16. In the Si/CoOx Faradaic junction, the Fermi level of Si is adjusted by CoOx through interface charge transfer and an equilibrium potential is established (Fig. 3b, Stage I). As mentioned above, the Si/CoOx/KBi(aq)/MnOx device indicates a residue voltage (VR) of 0.16 V (Fig. 2c), which comes from the different OCPs of Si/CoOx and MnOx. When Si/CoOx and MnOx are connected by a Cu wire in the dark, the Fermi level of Si is adjusted to the equilibrium potential of MnOx (Fig. 3b, Stage II). After the light is on (Fig. 3b, Stage III), photo-generated holes in Si oxidize Co2+ into Co3+, which shifts the potential of CoOx positively. A photovoltage (Voc) is generated in a Si/CoOx heterojunction, which is intrinsically different from the photovoltage from a buried p-n junction in our previous study19. When the hole quasi-Fermi level in Si is the same with the potential of CoOx, the photo charge process ends. On the other hand, photo-generated electrons transfer from Si to the MnOx counter electrode by a Cu wire. Since the electron quasi-Fermi level in Si always keeps the same potential with MnOx due to the short circuit connection, the photo-generated electrons shift the potential of MnOx negatively, as well as the Fermi level of Si. We experimentally observe the quasi-Fermi levels of both electron and hole in a semiconductor at the same time. The results are in good agreement with our previous study43, in which a Faradaic junction indicates a characteristic of isoenergetic interfacial charge transfer and can minimize the interface energy loss. Therefore, the electron and hole quasi-Fermi levels in Si under illumination are exactly recorded by MnOx and CoOx, respectively. Both MnOx and CoOx indicate high Faradaic capacitance, the potentials of MnOx and CoOx after photo charge can be stored even when the light is off (Fig. 3b, Stage IV), which is the reason for the photovoltage memory effect in the Faradaic junction device. Based on above analysis, we plot a working mechanism schematic of the device with the photovoltage memory effect (see Supplementary Fig. 17). Moreover, by comparing with the devices without the photovoltage memory effect (Supplementary Figs. 1821), we conclude that the high performance of Si/CoOx/KBi(aq)/MnOx comes from the longer storage time and higher charge quantity of the Faradaic layer than the electric double layer23,44.

We further propose a prerequisite to realize the photovoltage memory effect, as shown in Supplementary Fig. 22. When a photoelectrode is connected with a counter electrode in short-circuit in the dark, its Fermi level will be adjusted to the equilibrium potential of the counter electrode since the photoelectrode indicates much lower storable charge density than the counter electrode in the dark. The hole quasi-Fermi level in a semiconductor should locate within the Faradaic potential window of a Faradaic material on the semiconductor for hole storage. On the other hand, the equilibrium potential of a counter electrode should locate within its Faradaic potential window for electron storage. Therefore, to achieve the photovoltage memory effect in a Faradaic junction device, the electron and hole quasi-Fermi levels in a semiconductor should locate within the potential windows of Faradaic materials in the two electrodes, respectively.

### Demonstration of a portable Faradaic junction solar rechargeable device

Finally, we developed a portable Si/CoOx/KBi(aq)/MnOx device with a significantly reduced volume (Fig. 4a and Supplementary Fig. 23). Because the MnOx counter electrode remains semitransparent during charge and discharge (Fig. 4b), it allows sunlight to incident across the counter electrode onto the photoelectrode (Supplementary Fig. 1b). A surlyn membrane was used as separator in the portable device and the electrolyte was injected between the two electrodes. We investigated the effects of the membrane thickness on the performance of the Si/CoOx/KBi(aq)/MnOx Faradaic junction device and the results are shown in Supplementary Fig. 24. As the thickness increases, the initial charge/discharge current density of the device decreases, which possibly comes from faster mass-transfer in the thinner device. The portable device exhibits a dark volumetric energy density over 30 times higher than the reported device in our previous study19 and reaches a record value among all reported two-electrode solar rechargeable devices (Supplementary Table 2). In addition, the Faradaic junction device with the photovoltage memory effect can be used for direct solar energy storage. After illumination, five Si/CoOx/KBi(aq)/MnOx units connected in series indicates a dark output voltage of 2.3 V (Fig. 4c), which can power a light emitting diode (LED) bulb in the dark under zero bias (Fig. 4d and Supplementary Movie 1). For practical application, the Faradaic junction solar rechargeable device should be further upscaled. For this purpose, substrate ohmic loss, ionic transport resistance, parasitic adsorption of Faradaic materials should be decreased. Some possible methods, such as conductive silver frame as collector11, further reducing the distance between the two electrodes and introducing back-side illuminated photoelectrodes45, can be used. Moreover, it is also significant to develop large-scale deposition methods for Faradaic materials on the semiconductors and FTO substates11.

## Discussion

In summary, we find the photovoltage memory effect in the Si/CoOx/KBi(aq)/MnOx solar rechargeable device, which achieves the highest dark volumetric energy density among all reported two-electrode solar rechargeable devices. The photovoltage memory effect is only observed in the Faradaic junction device, but not in p-n junction solar cells or three-electrode/four-electrode solar rechargeable devices. The effect in the device comes from the electron and hole quasi-Fermi levels in the Si semiconductor under illumination being exactly recorded by MnOx and CoOx, respectively. Moreover, we propose a prerequisite to realize the photovoltage memory effect in a solar rechargeable device, the electron and hole quasi-Fermi levels of a semiconductor within the potential windows of Faradaic materials in the two electrodes, respectively. The two-electrode Faradaic junction device with the photovoltage memory effect can minimize the interface energy loss, which has a potential to achieve higher performance than three-electrode and four-electrode solar rechargeable devices. These results pave an avenue to device integration and provide the possibility to construct other new photoelectric devices.

## Methods

### Preparation of n-Si/CoOx photoelectrodes

Before depositing CoOx, phosphorous-doped (100) n-type Si wafers with a resistivity of 1–10 Ω cm−1 (Shunsheng electronics, China) were cut into 1 × 1 cm2. The wafers were ultrasonically cleaned in acetone, ethanol, and deionized water in turn for 10 min, and then were immersed into mixture solution of H2SO4 and H2O2 (3:1, v/v) for 10 min, and cleaned by deionized water. Indium particles were welded onto the n-type Si to form ohmic back contact and a Cu wire was attached to the indium for (photo-)electrochemical experiments. In order to prevent leakage current, the lateral and back sides of n-type Si were sealed by silica glue. The sealed Si photoelectrode was immersed in 5% HF solution for 30 s to remove the native surface SiO2 layer. An chronopotentiometry technique was used to electrodeposit CoOx on the surface of Si in 0.1 M Co(NO3)2 aqueous solution with - 0.1 mA cm−2. Unless otherwise specified, the loading amount of CoOx on Si was 30 mC cm−2. In the electrodeposition, a Si photoelectrode, a saturated calomel electrode (SCE) and a graphite rod were used as a working electrode, a reference electrode, and a counter electrode, respectively.

### Preparation of MnOx, CoOx and MoOx counter electrodes

An electrodeposition technique was used to coat MnOx, CoOx, MoOx on fluorine-doped tin oxide (FTO) substrates in a three-electrode cell. A FTO substrate, a saturated calomel electrode (SCE) and a graphite rod were used as a working electrode, a reference electrode, and a counter electrode, respectively. Aqueous solution of 0.1 M (CH3COO)2Mn, 0.1 M Co(NO3)2 and 0.05 M (NH4)6Mo7O24·4H2O were used as the electrolytes for the deposition of MnOx, CoOx and MoOx, respectively. MnOx and CoOx were electrodeposited by chronopotentiometry technique at +0.5 mA cm−2 for 200 s and −0.1 mA cm−2 for 300 s on FTO substrates, respectively. MoOx was electrodeposited on an FTO substrate at a constant potential of −0.8 VSCE with deposition charge of 100 mC cm−2.

### Assembly of solar rechargeable devices

The Si/CoOx was used as a photoelectrode. MnOx, CoOx and MoOx were used as counter electrodes. The electrolyte was KBi solution (0.2 M KOH and 0.4 M H3BO3) with pH = 9. Before assembling the device, Si/CoOx was activated by scanning from −0.3 VSCE to 0.6 VSCE for 25 cycles under illumination. A surlyn membrane with 60, 90 and 160 μm thicknesses was used as a separator between the Si/CoOx photoelectrode and the counter electrode, and then was dried in an Oven for 10 min to fix the two electrodes. Finally, the electrolyte was injected into the gap between the two electrodes.

### Characterization of samples

The morphologies of the samples were characterized by scanning electron microscopy (SEM, Nano Nova S230) with an accelerating voltage of 15 kV. The Raman spectroscopy (Horiba T64000, excitation wavelength ~ 532 nm) and transmission electron microscopy (TEM, Tecnai G2 F20) were used to characterize the samples during charge and discharge process. Transmittance and reflectance of the films were analyzed by using ultraviolet-visible-near infrared spectroscopy (UV-Vis-NIR, Perkin Elmer Lambda 950). X-ray photoelectron spectroscopy (XPS) was performed on a K-Alpha instrument operating with an Al Kα X-ray source. The binding energies were calibrated by the C1s peak (284.8 eV).

### Time-of-flight secondary-ion mass spectrometry (TOF-SIMS) experiments

After the Si/CoOx/KBi(aq)/MnOx devices were photo charged and dark discharged under zero bias for 100 s in KBi solution with H218O and D2O solvents, the samples were analyzed by TOF-SIMS (Münster, Germany) with a detection mode of negative ions. The Bi+ ions with 30 keV was used for TOF analysis in an area of 91 × 91 μm2, a beam of 1 keV Cs+ ions for sputter etching in an area of 250 × 250 μm2.

### Electrochemical and photoelectrochemical measurements

For half-cell measurement, the Si/CoOx, a saturated calomel electrode and a graphite rod were used as a working electrode, a reference electrode, and a counter electrode, respectively. The electrolyte was KBi solution (0.2 M KOH and 0.4 M H3BO3) with pH = 9. The light source was 1 Sun of simulated solar illumination by a Xe lamp with AM 1.5 G filter (100 mW cm−2). Cyclic voltammetry (CV) curves, galvanostatic charge-discharge (GCD) curves and electrochemical impedance spectroscopy (EIS) were recorded using an electrochemical workstation (CHI 760E, Shanghai Chenhua). The areal charge quantity (Q) and energy density (E) of a solar rechargeable device were calculated following Eqs. (1) and (2), respectively, where I is the charge/discharge current under zero bias, t is the charge/discharge time, and ΔV is the voltage difference of dark discharge23.

$$Q=\int I\,{dt}$$
(1)
$$E=Q\Delta V$$
(2)