Interlayer-Exciton Based Nonvolatile Valleytronic Memory

Analogous to conventional charge-based electronics, valleytronics aims at encoding data via the valley degree of freedom, enabling new routes for information processing. Long-lived interlayer excitons (IXs) in van der Waals heterostructures (HSs) stacked by transition metal dichalcogenides (TMDs) carry valley-polarized information and thus could find promising applications in valleytronic devices. Although great progress of studies on valleytronic devices has been achieved, nonvolatile valleytronic memory, an indispensable device in valleytronics, is still lacking up to date. Here, we demonstrate an IX-based nonvolatile valleytronic memory in a WS2/WSe2 HS. In this device, the emission characteristics of IXs exhibit a large excitonic/valleytronic hysteresis upon cyclic-voltage sweeping, which is ascribed to the chemical-doping of O2/H2O redox couple trapped between the TMDs and substrate. Taking advantage of the large hysteresis, the first nonvolatile valleytronic memory has been successfully made, which shows a good performance with retention time exceeding 60 minutes. These findings open up an avenue for nonvolatile valleytronic memory and could stimulate more investigations on valleytronic devices.

Van der Waals HSs stacked by TMDs monolayers enable the generation of longlived IXs with a large binding energy of about 150 meV 1 , and a long diffusion distance over five micrometres 2 , further extending the already appealing properties of the constituent TMDs monolayers. Since IXs are composed of electrons and holes that are resided in neighboring layers, their physical properties strongly depend on the layer configurations and external fields or dopings 3,4 . Through electrical field or doping, we can modulate the emission intensity and wavelength of the IXs 1 , and even switch its polarization 5 . Recently, IXs in the HSs stacked by other layered materials such as 2D perovskites and InSe with TMDs monolayer have been demonstrated and can be utilized in mid-infrared photodetections 6,7 .
In particular, IXs in TMDs-based heterostructures carry valley-polarized information and thus would find promising applications in valleytronics taking advantage of their long lifetime 8 . Previous studies have demonstrated that IXs exhibit a large valley-polarization degree that can be tuned in a wide range by external electric field 9 and magnetic field 10 . Although considerable progress has been made in valleytronics, nonvolatile valleytronic memory has not been achieved up to date, which is indispensable for valleytronic devices. To this end, it is urgent to explore possible strategies to efficiently store valley-polarized information for further development of valleytronic devices and chips. Here, we have successfully achieved an IX-based nonvolatile valleytronic memory, which would greatly prompt relevant investigations on valleytronics.
In this work, the HS device is formed by a monolayer WS2 (top) and a monolayer WSe2 (bottom), both of which are contacted with an electrode (Fig. 1a). By applying voltage between the electrode and the heavily-doped Si substrate, we can control the doping level of the device when performing optical measurements. Figure 1b shows the optical microscope image of the device. The WS2 and WSe2 sheets are mechanically exfoliated from their respective bulk crystals and then transferred on a SiO2/p ++ -Si substrate through dry-transfer technique 11 . The edges of the two sheets are intentionally aligned to improve interlayer coupling 3 . IXs in the WS2/WSe2 HS. Fig. 1c shows the PL spectra of the HS, from which we can observe severe PL quenching and redshift of the intralayer excitonic peaks, together with the appearance of a low-energy peak at 1.4 eV. The quenching and redshift of the intralayer excitonic peaks can be attributed to interlayer charge transfer 12,13 and modified dielectric environment 14,15 , respectively. We ascribe the peak at 1.4 eV to the IX emission according to previous reports 16,17 . The excitation-power dependent PL spectra further verify its interlayer nature (Fig. 1d). The IX emission peak shows a blueshift with the increase of excitation power, which is due to many-body effect arising  from 0 V to −60 V, then 0 V all the way to 60 V and finally back to 0 V (Fig. 2a). The IX emission peak shows a redshift and the emission intensity is enhanced with the decrease of Vg, and vice versa. The redshift of the IX emission peak with Vg can be ascribed to the Stark effect 5,20 , which is further verified by the opposite shift trend of the IX emission peak in devices with stacking order inversed (Fig. S1). Interestingly, the IX emission peak exhibits a strong hysteresis upon cyclic-voltage sweeping. As indicated by the black arrows in Fig. 2a, the peak energy of the IXs at middle 0 V (0V-2) cannot return to the same value of initial 0 V (0V-1), until a further upward scanning that is finally back to 0 V (0V-3). The gate-dependent photon energy and PL intensity can be seen more clearly in Fig. 2b and 2c. For a simple discussion, we only compare the states at 0V-2 and 0V-3. The photon energy of 0V-2 is blueshifted by about 20 meV with respect to that of 0V-3. Meanwhile, the PL intensity of 0V-2 is weaker than 0V-3 with a contrast ratio of about 1.7. It is worth to mention that the light intensity changes non-monotonously as Vg decreases from 0 V to −60 V, indicating the occurrence of chemical doping [21][22][23] , which will be discussed in the following.
As shown in Fig. 2d, the IX emission peak of 0V-2 can be decomposed to two Gaussian peaks (detailed fittings of the spectra are provided in Fig. S2). The energy difference of the two peaks is about 20 meV, which is consistent with the splitting energy of the conduction band of WS2 24,25 , strongly suggesting the occurrence of spintriplet excitons 26, 27 . This peculiar phenomenon can be understood from the chemicaldoping 21,22 induced band-filling effect 5,28 , as depicted in Fig. 2e and 2f. When the device is chemically n-doped, the Fermi level will be lift up and IXs will shift to the spin-triplet state (IX T ), which has an inefficient PL yield because of inversed spin. Contrarily, when the chemically-doped electrons are released, IXs will return to the spin-singlet state (IX S ). Therefore, the IX emission peaks in 0V-1 and 0V-3 spectra are attributed to IX S emission, and that in 0V-2 spectra is mainly resulted from IX T . The IX T and IX S peaks can be well resolved in PL spectra acquired by picosecond laser excitation (Fig.   S3a). In addition, the intensity ratio of IX T /IX S increases with the increase of Vg (Fig.   S3b), thus confirming the band-filling mechanism and IX T /IX S origins. We have also measured the gate-dependent lifetime of the IXs (Fig. S3, c-e). The lifetime of the IXs at 0V-2 is slightly shorter than at 0V-1 and 0V-3 rather than getting prolonged, further supporting the IX T /IX S origins 10 . The difference of light intensities between 0V-3 and 0V-1 might be due to different levels of chemical doping at the initial and final sweeping stages, as discussed in the following. X - Vac. Mechanism of the excitonic hysteresis. Electrical hysteresis has been observed in devices based on two-dimensional materials, such as graphene and TMDs based fieldeffect transistors 29,30 . Generally, electrical hysteresis is attributed to the chemical-doping effect by doping species (O2 and H2O) that are bound at the device/substrate interface, and/or on the surface of the device [31][32][33] . In our case, we propose that the excitonic hysteresis mentioned above is originated from the same scenario.
Since our measurements were performed in high vacuum (~10 -7 Torr), the influence of the molecules on the device surface can be safely neglected. Therefore, the excitonic hysteresis is more likely due to the O2/H2O molecules that are trapped at the interface between the HS and substrate. To clarify this, we examine the gate-dependent PL spectra of the individual WSe2 region (Fig. 3a), because WSe2 is in the bottom of the HS and directly contacts the SiO2/Si substrate. Additionally, we conducted a control experiment with WSe2 monolayer on a hydrophobic substrate (Fig. 3b).
The emission features of the intralayer excitons in WSe2 are closely correlated to that of IXs. As Vg decreases from 0 V to -60 V, the emission of positive trions ( + ) is gradually enhanced, while the peak of neutral excitons ( 0 ) is suppressed, indicating an efficient hole doping (detailed data is provided in Fig. S4). Peculiarly, as Vg increases from −60 V back to 0 V, the evolution track is asymmetric to that from 0 V to -60 V.
The trion emission peak is firstly weakened, then enhanced and redshifted with the increase of Vg. The asymmetric evolution strongly indicates the occurrence of negative trions ( − ), and suggests that the WSe2 is chemically n-doped 33,34 at 0V-2. When Vg increases from 0 V to 60 V, the − peak is redshifted further, but with emission intensity weakened because of Coulomb screening from the free electrons 35 . When voltage scans backward from 60 V to 0 V, the − peak shows a blueshift and the emission intensity becomes weaker while the 0 peak is gradually enhanced, indicating that the chemically-doped electrons have been released. All the above features are well consistent with the previously mentioned chemical-doping effect.
To further validate such hypothesis, we then focus on the PL spectra of a control device with monolayer WSe2 on a hydrophobic substrate (Fig. 3b). The evolution tracks of + and 0 emission are roughly symmetrical along the black dashed line at about -50 V. The slight deviation of the symmetry line at -50 V (rather than -60 V) might be due to trace O2/H2O molecules that are adsorbed on WSe2 before the transfer procedure.
Besides, in sharp contrast to Fig. 3a, the track of − is quasi-symmetrical along the dashed line at 60 V, suggesting that the excitonic hysteresis is largely suppressed.
Therefore, H2O molecules should play a critical role in our observations. The broad PL peak centered at about 1.65 eV might be due to local-state exciton ( ) 36 , which is out of the scope of this study.
The surface of SiO2 is usually covered with a layer of silanol groups (≡ Si − OH), especially after it is treated by piranha solution or plasma cleaner 21,33 . With these silanol groups, SiO2/Si substrates are easily bound by ambient O2 and H2O molecules 33 . As shown in Fig. 3c, the electrochemical potential of the redox couple (O2/H2O) is about -5.3 eV 21,37 , which is slightly higher than the valence band of WSe2 (about -5.46 eV) 38,39 .
Therefore, electrons spontaneously transfer from O2/H2O to WSe2, making monolayer WSe2 initially n-doped (detailed information is provided in Fig. S4), and resulting in the deviated symmetry at −50 V in Fig. 3b.
When applying negative gate voltages, electrons are forced to transfer further from O2/H2O to WSe2. Consequently, the Fermi level of the HS is lifted up, and IXs shift to the spin-triplet state (Fig. 2e). Those chemically-doped electrons balance out the gate modulation, resulting in the non-monotonic behavior of the IXs in 0 ~ −60 V range (Fig.   2c) and the excitonic hysteresis. The chemical-doping effect also explains why 0 emission maintains its intensity from 0 V to -60 V for WSe2 on the hydrophilic substrate ( Fig. 3a) but greatly suppressed on the hydrophobic substrate (Fig. 3b).
When applying positive gate voltages, the chemical-doped electrons are driven back from WSe2 to the O2/H2O redox couple. Therefore, IXs return to the spin-singlet state ( Fig. 2f) when Vg scans back to 0V-3. This control experiment further verifies the chemical-doping mechanism and well explains the origin of the excitonic hysteresis of IXs shown in Fig. 2. We have made dozens of HSs on hydrophobic and hydrophilic substrates, and we can only observe excitonic hysteresis (both photon-energy and PL-intensity hysteresis) in the samples on hydrophilic substrates. The hysteresis is largely suppressed in HSs stacked on hydrophobic substrates (Fig. S5). In addition, we have also fabricated a WS2/WSe2/hBN HS on a hydrophilic substrate with WS2/WSe2 HS partially separated from the substrate by a thin layer hBN. For this device, the excitonic hysteresis is observed in the region where WS2/WSe2 HS directly contacts with the substrate, but absent in the hBN-insulated region (Fig. S6), further supporting the chemical-doping mechanism. The hysteretic behavior is well reproducible in multiple repeating measurements and also in different samples. Therefore, we rule out the influence of random contamination.

Valleytronic hysteresis of the IXs.
To study the chemical-doping effect on the valley-polarized features of the IXs, we measured the helicity-resolved PL spectra of the device (Fig. 4a). Interestingly, the IX peak exhibits a negative circular polarization in contrast to that of intralayer excitons in WSe2 and WS2, which can be ascribed to the The DOCP of the IXs can also be electrically controlled by Vg, as shown in Fig.   4b (the full data set is provided in Fig. S7). The absolute DOCP is greatly suppressed at −60 V (p-doping), but enhanced at 60 V (n-doping). This phenomenon has been reported by Scuri and coworkers, and is attributed to changes in valley-depolarizationtime caused by electron/hole doping 41 . Similarly, we believe our observation can also ascribed to the charge doping from external applied bias and chemical doping (Fig. S3).
Interestingly, the DOCP and lifetime (Fig. S3e) of the IXs also exhibit a strong hysteresis, probably due to the carrier trapping and detrapping induced by the above-  To evaluate the retention time of the encoded information, we then prolong the reading-operation time, as shown in Fig. 5b. Surprisingly, the 1 and 0 excitonic states can persist for at least 60 minutes, holding great promise for nonvolatile valleytronic memories. As a matter of fact, the retention time should be much longer than 60 minutes, as can be seen in a logarithmic-timescale plot (Fig. S8a). We also note that the 0 (1) state varies dynamically before reaching a steady state. This is probably due to the charging/discharging process of the device, as confirmed by the features of timedependent leakage current (Fig. S8b). It is worth to mention that the information encoding ability of the device can persist up to 250 K, which is promising for high temperature valleytronic applications (Fig. S9).
Since the nonvolatile valleytronic memory has never been reported, it is hard to make an objective comparison. Nevertheless, the device is similar to photonic memory, thus we list the parameters of our device and other nonvolatile photonic memories in Table 1, which shows that our device is outperforming in comparison with peer memory devices. The PL ON/OFF ratio of the 1/0 states could be as large as 3.6 ( Fig. S10), which is larger than peer photonic memories [42][43][44][45] . The power consumption of the device is estimated to be about 74/56 nW for set/reset operation (Fig. S8b), which is extremely low in comparison with other phase-change photonic memories [42][43][44][45] . The switching time of our devices could be very short but limited by our testing system, since the hysteresis effect could be established in several microseconds according to previous reports 46 . Finally, we demonstrate the memory function of the device, which shows a good writing/reading/erasing ability with retention time exceeding 60 minutes. Our study provides a potential paradigm to achieve nonvolatile valleytronic memory and thus would greatly advance the development of valleytronic devices.

Sample Preparations.
Electrodes were fabricated by standard photolithography and thermal evaporation (50 nm/2 nm Au/Cr). The substrates with prefabricated electrodes are ultrasonic cleaned and plasma cleaned before the fabrication of the HS. WS2 and WSe2 monolayer flakes were first mechanically exfoliated onto polymethylmethacrylate (PMMA) stamps, and then transferred on a SiO2 (300 nm)/Si wafer using a dry transfer technique with the aid of an optical microscope and a nano-manipulator.
The hydrophobic substrates were prepared via immersing in HMDS vapor for 10 min and then rinsing with acetone for 30 s to form a hydrophobic layer on the substrate 47 .
All the samples are not treated by thermal annealing, because the thermal-annealing procedure can disable or deteriorate the performance of nonvolatile memory devices.
Optical Measurements. The as-fabricated devices were mounted in a continuous flow cryostat with 10 -7 Torr vacuum. For gate-dependent PL measurement, the sample was excited by a 532 nm laser (23 μW) at 78 K. For the helicity-resolved PL measurement, the sample was excited by a 633 nm laser with a power of 180 μW at 78 K. The time interval between two adjacent spectra is about 1 minute when performing gatedependent measurement. For the memory operation measurement, the spectra were acquired with Vg changing cyclically and laser keeping focused on the sample. Each spectrum was measured within 10 seconds. All the PL spectra were collected by a 50× objective lens (N.A. = 0.7) in a Raman spectrometer (Horiba HR550) with a 600 g/mm grating. A Keithley 2400 sourcemeter was used as the voltage source.

Supporting Information
The supporting information consists of: 1) Gate-dependent PL spectra for a stacking-order inversed HS (WSe2/WS2) on a hydrophilic substrate; 2)Lorentz and Gaussian fittings of 0V-1, V-2 and 0V-3 spectra; 3) Spin-singlet and spin-triplet IXs and their lifetimes; 4) PL spectra of the WSe2 monolayer under gate voltages from 0 V to −60 V shown in Fig. 3a; 5) Gate-dependent PL spectra for a WS2/WSe2 HS on a hydrophobic substrate; 6) Gate-dependent PL spectra of a WS2/WSe2/hBN HS; 7) Helicity-resolved PL spectra of the WS2/WSe2 HS; 8) Electrically controlled memory operation in the HS; 9) Memory performance of the WS2/WSe2 HS under different temperatures; 10) PL spectra of 0V-2 and 0V-3 states and the ON/OFF intensity Ratio.