Thank you for visiting Nature. The browser version you are using has limited support for CSS. For the best experience, we recommend that you use a newer version of the browser (or turn off the compatibility mode in Internet Explorer). At the same time, to ensure continued support, we will display sites without styles and JavaScript.

npj 2D Materials and Applications Volume 5, Article Number: 1 (2021) Cite this article

The most advanced memristors are mostly formed of vertical metal-insulator-metal (MIM) structures, which rely on the formation of conductive filaments for resistance switches (RS). However, due to the random formation of the filament, the setting/reset voltage of the vertical MIM memristor is difficult to control, resulting in poor temporal and spatial switching uniformity. Here, a transverse memristor based on electron beam irradiation of rhenium disulfide (ReS2) at both ends is realized, and a resistance switching mechanism based on Schottky barrier height (SBH) modulation is revealed. The device has a non-forming, stable gradual RS characteristics, while achieving small transition voltage changes (6.3%/5.3%) during positive and negative scanning. RS is attributed to the movement of sulfur vacancies caused by voltage bias in the device, which regulates ReS2/metal SBH. Compared with the sudden RS in the MIM-based memristor, the gradual SBH modulation stabilizes the time change. In addition, the device was used to demonstrate the simulation of the long-term synaptic plasticity of biological synapses, showing its potential as an artificial synapse for energy-saving neuromorphic computing applications.

The memristor has been extensively studied and is regarded as one of the candidates for artificial synapses for neuromorphic computing1,2,3,4,5,6. In these memristors, the switching mechanism mainly relies on the formation of conductive filaments in the insulating layer, such as valence change mechanism (VCM) and electrochemical metallization (ECM) 7. For VCM-based devices, the change in conductance is caused by the migration of vacancy anions, such as oxygen vacancies 8,9,10,11. However, due to the random distribution of vacant anions in the insulator, the formation of anion filaments is a random process7,12,13. The resistance switch (RS) of the ECM device is caused by the movement and metallization of the active electrode metal cations (such as Ag) (References 14, 15, 16). However, due to the random nature of these cation migration paths 7, 13, 17, it is difficult to control such highly mobile metal cations during the electroforming step. Therefore, for devices based on VCM and ECM, time (cycle to cycle) changes are inevitable due to the random formation and rupture of conductive filaments. To overcome this problem, Choi et al. Demonstrated epitaxial random access memory (epiRAM) based on single crystal SiGe18. Due to the limitation of silver wire and the precise control of dislocation density, epiRAM achieves a small set voltage change. However, the device still cannot avoid time changes during the filament eradication process. Therefore, finding a non-filamentary switching mechanism is essential to control cycle-to-week changes. In addition, the growth temperature of polymer beam epitaxy (MBE) is not suitable for the integration of such epitaxial RAM and complementary metal oxide semiconductor (CMOS) technology. In terms of back-end (BEOL) compatibility, due to the development of chemical vapor deposition (CVD) and low-temperature two-dimensional material growth of large-scale two-dimensional materials, two-dimensional (2D) materials have become an alternative transfer technology 19, 20, 21, 22 ,twenty three.

Many vertical memristors based on two-dimensional materials and their derivatives have been proven. Some of the switching layers are made of pure 2D materials (such as MoS2, hBN, WSe2), and the switching mechanism is based on natural defects in the material (such as sulfur vacancies and boron vacancies) and active metal wires (such as Ag, Ti, and Cu) 24, 25 , 26, 27. Both short-term and long-term synaptic plasticity have been simulated in these devices24,25. In addition, it is reported that vertical memristors based on 2D material derivatives (such as MoOx/MoS2, WOx/WSe2) have low switching voltages due to the thin oxide layer 28,29. This vertical memristor is suitable for device scaling to achieve high-density array integration 30,31. In addition, compared with the lateral memristor, the longitudinal memristor exhibits a smaller setting voltage 2, 15, 28 due to a thinner switching layer. However, their two-terminal structure is not suitable for multi-terminal biological synapse simulation. Compared with vertical memristors, lateral memristors are more versatile, and multi-terminal memristors 32, 33 can be realized by adding more electrodes. Recently, a MoS2-based lateral memristive device was reported, which relies on the movement of sulfur vacancies caused by voltage bias and the Schottky barrier height (SBH) modulation of the metal/MoS2 contact area 32, 34, 35. This switching scheme distinguishes these memristors from filamentous memristors and can reduce the changes caused by the random filament formation process. In addition, finding new materials that are more likely to generate sulfur vacancies can improve switching performance. Rhenium disulfide (ReS2) is a two-dimensional material with weak interlayer coupling, soft Re-S covalent bond and low sulfur vacancy formation energy. When subjected to an external bias, it may experience more significant sulfur vacancy movement 36 ,37,38.

Here, we start with single crystal 2D ReS2 and demonstrate the ReS2-based lateral memristor by introducing sulfur vacancies in the material. The device exhibits no shaping gradient RS characteristics. A small cycle change (6.3% and 5.3%) of the transition voltage during the positive and negative scans is realized. Contrary to the filament switch mechanism, RS is derived from the movement of sulfur vacancies in ReS2 caused by voltage bias, which regulates the metal/ReS2 SBH. Compared to the sudden switching of a filament-based memristor, the resulting gradual RS allows for small changes in direct current (DC) scanning. In this ReS2 memristor, we confirmed the feasibility of electron beam radiation (EBI) in introducing defects into ReS2 through transmission electron microscopy (TEM). In order to further study the influence of EBI on ReS2, a detailed material study was carried out. The rectifier-mediated switching characteristics are recognized, which is different from typical nonvolatile memristors and requires a fixed read voltage polarity32. In addition, the RS mechanism is analyzed based on the selective EBI between the contact and the active area of ​​ReS2 and the temperature-dependent current-voltage measurement. The effects of electron beam dose and sheet thickness on the RS ratio are also discussed. Finally, the device is used as an artificial synapse, which can simulate long-term potentiation (LTP), long-term inhibition (LTD), paired pulse promotion (PPF), paired pulse inhibition (PPD), spike amplitude-dependent plasticity ( SADP) and spike time dependent plasticity (STDP).

As shown in Figure 1, we fabricated a ReS2-based memristor device with two flat ends. Figure 1a shows the structure of the manufactured ReS2 memristor. The device is designed in a horizontal configuration, and the sheet is selectively irradiated by a focused electron beam, as shown in Figure 1b. The guidelines in ReS2 point out the deficiencies caused by EBI. The composition of the defect will be discussed in detail later. Scanning electron microscope (SEM) images of four ReS2 memristors fabricated on the same sheet are shown in Figure 1c. There are eight parallel electrodes from left to right, and each pair of electrodes forms a transverse memristor. The EBI doses of these devices are 8000, 6000, 4000, and 0 μC cm-2 (original sheet) from left to right.

a Schematic diagram of the structure of our ReS2 planar memristor. b Schematic diagram of EBI. Selective area irradiation is performed by focusing the electron beam. The red line in ReS2 shows the guidelines for defects. c SEM image of ReS2 memristor. The distance between the two electrodes is 1.5 μm. The highlighted square shows the exposure dose and area. Scale bar, 5 microns. TEM image of df cross-section of ReS2 memristor at 6000 μC cm-2 EBI. d In the contact area. The inset shows a magnified image of the disordered lattice (red) and the original crystal structure (green). The white arrow indicates the position of the lattice disorder. Scale bar, 5 nm. e In the active area of ​​ReS2. Scale bar, 5 nm. f At the contact edge. The yellow dashed line marks the location of the contact edge. The inset shows a magnified image (red) of the lattice disorder. Scale bar, 5 nm.

Next, we conduct TEM analysis on the same sheet to study the properties of ReS2 irradiated by electron beams to verify the feasibility of EBI to produce defects in the ReS2 sheet. Figure 1d shows the TEM image of the contact area with an EBI dose of 6000 μC cm-2. The thickness of the flakes is 3 nm, composed of five layers of ReS2, and the thickness of each single layer is 0.6 nm39. The inset shows a clear contrast between the lattice disorder (marked by the red outline) and the layer-by-layer structure (marked by the green outline). When the lattice structure is distorted by lattice disorder, the image of the lattice structure becomes blurred25,35. Since EBI is performed before metal deposition (see "Methods"), the contact area and ReS2 active area in Figure 1c are both irradiated. It was also found that the crystal lattice of the active region of ReS2 was disordered (as shown in Figure 1e). Supplementary Figure 1c describes additional TEM analysis of other prominent areas in Figure 1c with different doses. 1-4, and lattice disorder was also found in ReS2 flakes. Interestingly, lattice disturbances are found at all contact edges (see Figure 1f and Supplementary Figure 3), which may be related to the influence of electrical measurements, which will be discussed in the device characteristics section. Based on these images, we assume that an appropriate dose of EBI (4000-8000 μC cm-2) can produce defects in ReS2 flakes. The uneven distribution of defects in TEM images may be attributed to the difference in formation energy between defects at different positions in ReS2 (Reference 38). Further top-view TEM characterization will clarify the spatial distribution of lattice disturbances40.

In addition to TEM, material analysis was also performed to understand the effect of EBI dose on defect density and the mechanism of defect formation in ReS2. First, the dose dependence of defect density was studied through morphology and optical analysis. Figure 2a shows the atomic force microscopy (AFM) phase images of ReS2 flakes irradiated with different doses. The original sheet showed a uniform surface with almost no wrinkles, which may be caused by mechanical peeling. When the amount of EBI increases, the surface of the sheet becomes rough and forms some crack-like features. The density of this feature increases with increasing EBI dose. This phenomenon indicates that ReS2 flakes may have more defects when exposed to higher EBI doses. In addition to the topography analysis of AFM, Raman and photoluminescence (PL) analysis are also performed to further confirm the formation of defects. Compared with AFM, the existence and density of defects may be more sensitive. This is because the lattice vibration mode and the radiation recombination process in the material are highly dependent on the crystal structure and defect density. Figure 2b shows the Raman spectra of flakes between 125 and 225 cm-1 at different EBI doses. Since ReS2 is an anisotropic material, Raman spectra are recorded with the same flake orientation ([010]) to eliminate the influence of crystal orientation42. The two labeled peaks correspond to the in-plane (Eg) and out-of-plane (Ag) vibration modes 37, 42, 43 of ReS2, respectively. The full range of Raman spectra can be found in Supplementary Figure 5. The irradiated flakes show a blue shift of Eg and Ag peaks of about 1.26 cm-1 in ReS2 (see Supplementary Figure 5), and the EBI dose ranges from 0 to 8000 μC cm-2. This blue shift of the Raman peak is caused by the relaxation of atom-atom vibrations in the 2D structure, indicating that more defects are generated in ReS2 when exposed to larger doses43.

AFM phase images taken with different EBI doses. Scale bar, 50 nm. b Raman spectra of ReS2 with different EBI doses. The dotted line marks the blue shift of the Eg and Ag peaks. c PL spectra of ReS2 flakes with different doses. d, e XPS signals of Re 4f and S 2p peaks before and after EBI at a dose of 6000 μC cm-2. The dotted line marks the peak position shift in the XPS signal. f Schematic diagram of sulfur vacancies generated during EBI.

In addition, the PL spectrum scan on the same flake is shown in Figure 2c. Pristine ReS2 shows a wide signal at about 1.52 eV, which is consistent with previous PL studies. After EBI, the PL signal becomes wider and its intensity decreases. Although the PL intensity of different EBI doses has little difference, it is still obvious that EBI can form defects in ReS2, which may cause more non-radiative recombination processes to reduce the PL intensity. Based on all these characteristics, it is confirmed that EBI can design defects in ReS2 flakes, where higher doses increase defect density. The chemical composition of the defect has also been clarified. According to theoretical calculation results, sulfur vacancies have the lowest formation energy in ReS2 (Reference 38). Compared with other transient metal dichalcogenides (TMD), the sulfur vacancies in ReS2 are more likely to form 36,38 due to the relatively soft Re-S covalent bond and weaker interlayer coupling. In addition, previous studies on the defect engineering of ReS2 using electron beam, oxygen plasma and helium ions have shown that sulfur vacancies are the most common defects inside the sheet 37, 38, 40. X-ray photoelectron spectroscopy (XPS) was performed to test our hypothesis. Figures 2d and e show the XPS spectra of Re 4f and S 2p before and after EBI at a dose of 6000 μC cm-2. The peak of 162.4 eV may originate from the contamination of polydimethylsiloxane (PDMS) during sample preparation, as shown in Supplementary Figure 6. After EBI, the binding energy of Re 4f and S 2p shifted to a smaller value. The specific binding energy and displacement values ​​are shown in Supplementary Table 1. This shift in binding energy further proves that there are defects in the ReS2 flakes after EBI, which is consistent with the binding energy shift reported in defective MoS2 (Ref. 44). In addition, before and after irradiation, the atomic ratio of S atoms to Re atoms was extracted in ReS2. The extracted atomic ratios of S/Re in the stripped flakes and irradiated flakes were 1.99 and 1.87, respectively. The decrease in atomic ratio proves the formation of sulfur vacancies in ReS2. Based on the analysis of EBI dose and defect composition, Figure 2f shows a schematic diagram of how the electron beam introduces sulfur vacancies in ReS2, where the highlighted circle indicates the presence of sulfur vacancies due to the sputtering effect of the electron beam. Intuitively, a larger EBI dose will cause more electrons to bombard the flakes, resulting in a higher density of sulfur vacancies.

The schematic diagram of the current-voltage (IV) curve of this ReS2-based memristor is shown in Figure 3a. The black dashed line represents the full loop current during the voltage sweep. The black arrows with numbers indicate the voltage sweep sequence during the measurement. It should be noted that the RS characteristic, called the rectification-mediated switching characteristic 32, is different from the typical IV curve of a non-volatile memory. In a typical non-volatile characteristic, the current increases under one voltage polarity (for example, positive bias) and decreases by 34 under the opposite voltage polarity (for example, negative bias). The difference is that the measurement current in the ReS2-based memristor switches from low current to high current under positive and negative bias.

a Schematic diagram of the IV curve of linear scale. The black arrows indicate the voltage scan sequence. The red and blue curves are the two states in the small voltage range after the positive and negative voltage scans. b, c IV curves of forward and reverse diode states extracted from different voltage sweep ranges. d Band diagram before and after positive voltage sweep. The blue curve above shows the conduction band before the voltage sweep. The red dashed line indicates that the SBH at the right contact decreases and the SBH at the left contact increases. e Qualitative device model after positive voltage sweep. The size of the Schottky diode symbol refers to the different SBHs at the contacts. The bottom schematic shows the expected IV curve with a small voltage range in the forward diode state. f Energy band diagram before and after negative voltage sweep. The red curve above shows the conduction band before the voltage sweep. The blue dashed line indicates that the SBH at the left contact decreases and the SBH at the right contact increases. g Qualitative device model after scanning with negative voltage. The bottom schematic diagram shows the expected IV curve in the small voltage range in the reverse diode state.

According to previous MoS2 work, this gradual RS behavior may be related to the sulfur vacancy movement caused by the voltage bias effect, which regulates the SBH 32,34 of the contact area. The positively charged sulfur vacancies 32 in MoS2 have been reported, where the RS curve switches from high current to low current under positive and negative bias. However, the sulfur vacancies in ReS2 are negatively charged because these vacancies are reported to also serve as mobile species in MoS2-based memristive devices45, 46, 47. During the measurement, the right electrode is always grounded, and a voltage is applied to the left electrode. The red curve and blue curve in Figure 3a are respectively defined as the forward diode (FD) state and the reverse diode (BD) state. Figure 3b and c show the measurement data of the FD state and the BD state under different voltage sweep ranges. The full range IV curve will be discussed later (see Figure 4e). Here, a larger voltage sweep range results in a greater degree of rectification in the FD and BD states, indicating that a more effective SBH modulation is achieved under a stronger electric field. The current difference between the FD and BD states may be attributed to the change and distribution of sulfur vacancies caused by EBI. A similar slight asymmetry was also found in MoS2-based devices. Investigate deeply to find the root cause. As shown in Figure 3d-g, the detailed mechanism of SBH modulation and sulfur vacancy movement is described using energy band diagrams and qualitative device models. During the positive voltage sweep (processes 1 and 2 in Figure 3a), the negatively charged sulfur vacancies move from right to left contact (see the energy band diagram in Figure 3d). This leads to a decrease (increase) in the density of sulfur vacancies at the right (left) contact, thereby reducing (enhancing) the Fermi level pinning (FLP) effect, and resulting in a lower SBH at the right contact and a higher SBH at the left contact. high. The modulation of the SBH after the positive voltage sweep is marked with a red dashed line in the energy band diagram (see Figure 3d). The device can be seen as two Schottky diodes connected back-to-back with a series resistor in between (as shown in Figure 3e and g). After the device is stressed by a positive voltage sweep, the SBH of the right contact is smaller than the SBH of the left contact. This asymmetrical back-to-back connection of Schottky diodes results in an IV curve as shown in Figure 3e, which is similar to a single Schottky diode 48 under forward bias. After this positive voltage sweep, it is called the FD state. Next, during the transition period 3 and 4 in Figure 3a, the sulfur vacancies drifted from the left electrode to the right electrode, which enhanced (mitigated) the FLP effect and resulted in high SBH at the right contact and low SBH at the left contact (see Figure 3f). The change of SBH after the negative voltage sweep is marked by the blue dashed line in the energy band diagram (see Figure 3f). This produces the expected IV curve shown in Figure 3g, similar to a single Schottky diode reverse biased. After this negative voltage scan, it is called the BD state. The retention test shows that this FD state and BD state can be retained for more than 700 seconds without external bias (see Supplementary Figure 7 for details). Although the retention time is longer than that of the volatile memristor2, the retention time of our device is shorter than 1500 minutes compared to the memristor with typical non-volatile characteristics proposed by Sangwan et al.34. The short retention time of our device may indicate that the sulfur vacancies are very easy to move and the diffusion barrier is low, which is attributed to the low formation energy of sulfur vacancies in ReS238. In addition, this gradual decrease of the retention curve is consistent with the results reported in other lateral memristors, where the explanation is also based on sulfur vacancy diffusion.

a Schematic diagram of the IV curve of linear scale. The black curve arrow in the normal state indicates the transition from the reverse diode state to the forward diode state, and the black curve arrow in the negative state indicates the transition from the forward diode state to the reverse diode state. b IV curves of five consecutive positive voltage sweeps (0 → 5 → 0 V). The current increases as the number of scan cycles increases. c IV curve of the other five scan cycles (0 → -5 → 0 V). The current increases in the negative state and decreases in the positive state. d DC programmatic enhancement and suppression. The red point is extracted from the first 0.5 V in the different cycle scans in b, and the blue point is extracted at 0.5 V in the different cycle scans in c. The inset is a schematic diagram showing the current value extracted at the same base voltage of 0.5 V. e IV curve is measured under different voltage sweep ranges. The curve shown here is not measured continuously. f 0.5 V current value extracted in different scanning ranges. The inset shows that the calculated RS ratio increases as the voltage sweep range increases.

To verify this mechanism, we first considered the possibility of sulfur vacancy movement caused by voltage bias in ReS2. The electric field that can drift sulfur vacancies in polycrystalline MoS2 and MoS2 irradiated with helium ions is usually greater than 104 V cm-1 (References 32, 34, 45). In this work, the electric field is about 30,000 V cm-1, which is equivalent to the minimum activation field. In addition, due to the relatively soft covalent bonds in ReS2, the movement of sulfur vacancies in ReS2 may require a smaller electric field compared to MoS238. Next, we performed selective EBI on control samples to prove that this RS is caused by sulfur vacancies in the contact area. The results show that only the devices irradiated in the contact area or the contact area and the ReS2 active area show obvious RS behavior (see Supplementary Figure 8), indicating that this RS behavior is mainly derived from the sulfur vacancies at the contacts, among which the vacancies in ReS2 The active area promotes the movement of the vacancy to the contact area. Then, the movement of sulfur vacancies caused by the voltage bias is supported by Figure 1f and Supplementary Figure 3. Since EBI is performed before electrode patterning, the probability of producing defects in each area of ​​the sheet is equal. However, the results showed that defects were found at all contact edges, which means that sulfur vacancies may be induced during electrical measurements, rather than EBI or other device manufacturing processes. In addition, electron dispersive X-ray spectroscopy line scans (see Supplementary Figure 9) have also been performed, indicating that there is a non-van der Waals contact between the metal and ReS2, which is generally considered to be the origin of FLP in 2D material contact49,50. In addition, temperature dependence measurements were performed to extract the SBH at the right contact in the FD state and the BD state. When the device was switched from the BD state to the FD state, it was found that the right contact SBH was significantly reduced (see Supplementary Figure 10). Obviously, RS is caused by the movement of sulfur vacancies caused by voltage bias, which changes the SBH at the contact and further adjusts the measurement current.

With the confirmation of the RS mechanism, we further studied the rectification-mediated switching characteristics through different DC scanning schemes. As shown in Figure 4a, it is worth noting that the positive voltage sweep changes the device from the BD state to the FD state (called BF transition), and vice versa for the negative sweep (called FB transition). In addition, always use a small positive voltage to read the current and resistance values ​​of the BD and FD states. Figures 4b and c show examples of how the DC voltage sweep modulates the current level in the device. First, apply five consecutive positive voltage sweeps (0 → 5 → 0 V) ​​to the device, as shown in Figure 4b. The measured current appears to increase monotonically with the number of measurement cycles. This is because the SBH at the right contact is reduced during the cycle scan with positive bias. Then we extract the current value at the first 0.5 V of each cycle, as shown in Figure 4d. After five consecutive positive cycles, we immediately continue to measure another five negative scan cycles (0 → -5 → 0 V), as shown in Figure 4c. The negative voltage sweep changes the device from the FD state to the BD state, and reads the current under a small positive bias of 0.5 V. We notice that the current increase is because the SBH at the left contact is decreasing. In contrast, the SBH of the right contact increases under negative scanning, which causes the current to decrease when reading with the same positive bias of 0.5 V.

Figure 4d shows the current value at 0.5 V extracted from the cyclic measurement. By applying positive voltage and negative voltage scanning respectively, a clear increasing and decreasing current trend was observed. This DC-programmed synaptic enhancement and inhibition demonstrates the potential of using such devices for pulse-programmed synaptic behavior simulation. In addition, we compared the IV curves measured under different voltage sweep ranges, as shown in Figure 4e. In each scan range, the device is measured for several cycles, and the stable IV curve is shown in the figure. Obviously, the current is mainly affected by the voltage sweep range. Figure 4f shows the current drawn in the FD state and the BD state at 0.5 V. Obviously, the larger scanning range enlarges the difference between the FD state and the BD state. The calculated RS ratio is shown in the inset of Figure 4f. As the scanning range increases, it increases from approximately 20 times to 70 times, indicating that it can achieve synaptic plasticity that depends on the pulse amplitude.

For memristive devices, the RS IV curve under voltage sweep conditions is very important and should have the following characteristics: (i) The RS ratio between the high resistance state (HRS) and the low resistance state (LRS) is greater than 24,28, (ii) Set voltage (Vset) 18, 51 small time changes, and (iii) the stable and durable performance of HRS and LRS52. Figure 5a shows the typical non-forming gradient RS IV curve of the memristor after fabrication irradiated at 6000 μC cm-2. Unlike other filament-based memristors that usually show abrupt RS curves28,53, our device shows a gradual change in conductance without electroforming during the voltage sweep, which indicates a non-filament RS mechanism. In addition, the device shown in Figure 5a measures 100 DC cycles to extract time changes. Here, we extract the BF and FB transition voltages based on the nonlinearity of the IV curve 45, which are greater than 5 (I ~ Vk and k ≥ 5), as shown in Figure 5b. The details of the transient voltage extraction are described in Supplementary Figure 11. The changes and histograms of the transition voltages of BF and FB are shown in Figure 5b and c, respectively. Compared with other filamentous memristors, the main advantage of ReS2-based memristors is their high time switching uniformity. The minimum time changes of the transition voltages of BF and FB reach 6.3% and 5.3%, respectively, which are much lower than filamentous memristors and are comparable to the improvement of devices based on metal doping or ion transport limitation (see Supplementary Table 2) 51, 54,55. In addition, this low-voltage change is important for neuromorphic computing applications and can improve learning accuracy31,56.

a One hundred cycles of IV scan measurement. The black arrows show the voltage scan sequence. b BF and FB conversion voltage extracted during 100 DC scan cycles. The voltage change is calculated by the ratio of the standard deviation (σ) to the average value (μ). c The extracted histograms of the transition voltages of BF and FB, both of which follow the Gaussian distribution. d The resistance value taken at 0.5 V during 100 DC sweep cycles. The RS ratio is about 50 times. e The average RS ratio of 3 nm thick flakes at different EBI doses. From left to right, take the average of 3, 3, 7, 5, and 3 devices respectively.

In order to study the durability performance, the resistance values ​​of the FD state and the BD state were extracted at 0.5 V, as shown in Figure 5d. Proved that the RS ratio is about 50 times. The increase in FD state and BD state during durability is due to equipment degradation during measurement. Like other memristors based on the rearrangement of atoms in the switching layer, this kind of sulfur vacancy movement may cause damage to the sheet, leading to device degradation. Next, we will study the effect of EBI dose and ReS2 thickness on switch performance. Figure 5e shows the average RS ratio of 3 nm thickness devices with different EBI doses, indicating that 6000 μC cm-2 is the optimal irradiation dose. The IV curves of memristors treated with different doses in the same slice are shown in Supplementary Fig. 12, where the 6000 μC cm-2 condition still achieves the maximum RS ratio. This is because small doses result in the least sulfur vacancies that trigger RS ​​behavior. On the contrary, an excessive dose will seriously increase the series resistance of ReS2, causing the total resistance to be dominated by the series resistance of ReS2 instead of contacting SBH, resulting in a smaller RS ​​ratio. Supplementary Figure 13 shows the RS ratio graph of all devices with a dose of 6000 μC cm-2, where a critical thickness of 5 nm was found to achieve a tenfold increase in the RS ratio. Beyond this thickness, RS behavior does not exist because at the same EBI dose, thicker flakes have a lower density of sulfur vacancies to trigger switching. It should be noted that it is impractical to compare the space between devices (device to device) changes. This is because our devices are manufactured by mechanical peeling, which makes it difficult to control the thickness, size, and shape of each sheet that affects RS performance. However, further advances in CVD to produce large-area and uniformly grown ReS2 films will allow future work to study the spatial variation of such devices.

The simulation of biological synaptic behavior is realized by using the memristor made by us. For artificial synapses, our devices operate similarly to the human nervous system. In human synapses, when a spike input reaches a presynaptic neuron, ionic neurotransmitters will be released from the presynaptic neuron, which triggers another spike in the postsynaptic neuron. Similarly, in the ReS2-based memristor, the left and right electrodes are regarded as pre-synaptic and post-synaptic neurons, respectively. Sulfur vacancies mimic the neurotransmitter in biological synapses, and adjust SBH at the contact point through pulse drift. Figure 6a shows the programming pulse waveform used to modulate the synaptic weight of the device, which is extracted based on a 0.5 V base voltage.

a Schematic diagram of pulse waveform modulating synaptic weight. The synapse weight refers to the current read at a base voltage of 0.5 V. b Potential measurement. The 5 V and 0.5 V current readings increase together. c Depression measurement. The current reading at -5 V increases, but the current reading at 0.5 V decreases. d LTP and LTD are measured by continuous positive and negative pulses. The green line shows the linear fit. PSC is measured at a 0.5 V reference voltage. e Enhanced dynamic response of pulse trains with amplitudes of 2 V, 3 V and 5 V. The illustration shows the base current read at 0.5 V. f The suppressed dynamic response of pulse trains with amplitudes of -2 V and -3 V, and -5 V. The current shown in the figure was read at a base voltage of 0.5 V. g SADP characteristics. Ten consecutive pulses are provided to the device. The current before and after the pulse train is extracted to calculate the synapse weight change. All parameters except amplitude remain unchanged. h PPF and PPD measurement. Only two pulses are introduced into the device. The dashed line is the exponential fit of the data. i STDP measurement. The dashed line is the exponential fit. The illustration shows a schematic diagram of the equivalent pulse waveform design.

Since it takes time for the sulfur vacancies to diffuse back to their original position, this reduced FLP can be maintained when the next pulse arrives, leading to long-term synaptic plasticity. In the following measurement process, the voltage input is introduced through the left electrode, and the current is recorded as the post-synaptic current (PSC). Figure 6b shows an example of enhanced measurement in this type of device. When a positive pulse is applied, the measured currents at 5 V and 0.5 V increase together, which is consistent with the DC enhancement in Figure 4b. However, when a negative pulse is applied (see Figure 6c), the absolute current value at -5 V increases, but the current measured at the same base voltage of 0.5 V decreases, which is consistent with the DC suppression shown in Figure 6. 4c. LTP and LTD are demonstrated and shown in Figure 6d. A continuous positive voltage pulse sequence is applied to the device along with another negative voltage pulse sequence, and the PSC read at 0.5 V gradually increases or decreases during the positive pulse or the negative pulse, respectively. It is worth noting that at the beginning of the positive and negative pulse sequence, the PSC response can be fitted by a linear function, which is suitable for neuromorphic calculation 58,59. Then, the dynamic response of this artificial synapse is studied by measuring pulse voltage stress with different pulse amplitudes (see Figure 6e and f). In Figure 6f, in order to obtain the same starting current, a pulse sequence with an amplitude of 3.5 V is applied before the negative pulse measurement. Similarly, the PSC measured at 0.5 V increases during the positive pulse and decreases with the negative pulse.

In addition, the dynamic response exhibits amplitude dependence. A device with a larger positive pulse amplitude can obtain higher current and faster increase, and can more effectively reduce the conductance stimulated by a larger negative amplitude, which means that the device has the ability to respond to stimuli of different amplitudes, or SADP60 . To further study the SADP behavior, the synaptic weight change (the current change read under the base voltage) with different pulse amplitudes was measured (see Figure 6g). During the measurement, each point is extracted from the synaptic weight change between the first and tenth pulse. The results show that a larger pulse amplitude can modulate the synaptic weight more effectively, which is consistent with the DC voltage scan range dependence in Figure 4f. In biological synapses, when presynaptic neurons undergo spike frequency adaptation, spikes will be produced at variable frequencies. In this way, the spike rate dependence of the synapses is also important and is analyzed in Figure 6h. The data points are extracted from the synaptic weight changes between the first and second pulses, so for positive and negative pulse conditions, it is also reclassified into PPF and PPD, respectively. The results show that a smaller pulse interval (higher frequency) leads to a larger change in synaptic weight because the device has less relaxation time during a shorter pulse interval2,25. A similar phenomenon was also found in the dynamic response measurement process with different pulse intervals, indicating that pulse sequences with shorter intervals require less time to stimulate the synaptic weight to the same value (see Supplementary Figure 14). STDP is also an important function of spike neural networks3. Theoretically, STDP describes the changes in synaptic weights when pre-synaptic and post-synaptic pulses are applied to synapses with a certain time delay. Generally, the synapse weight increases with a positive time delay and decreases with a negative time delay. Here, equivalently designed pulses are introduced into the presynaptic neuron just to simulate the effect of two pulses, which are widely used in memristor synaptic applications3,25,63. The details of the pulse waveform design are explained in Supplementary Figure 15 and Supplementary Note 1. A typical STDP result is shown in Figure 6i, which shows the device's ability to be applied to a spike neural network with STDP learning rules. Through exponential function fitting to the data, the positive and negative delay time constants are extracted as 16 ms and 60 ms, respectively. This microsecond response is similar to human neuron64.

In conclusion, it is proved that the ReS2-based memristor irradiated by the electron beam has improved switching uniformity. The RS mechanism is attributed to the movement of sulfur vacancies caused by the bias voltage, which leads to SBH modulation in the contact area. The formation of sulfur vacancies has been confirmed by material analysis. In addition, the non-formed RS behavior with minimum transition voltage changes of 6.3% and 5.3% was achieved under positive and negative scanning, respectively. In addition, this type of device also achieves a comprehensive demonstration of synaptic function. The minimal time change and successful simulation of synaptic plasticity in ReS2-based memristors have great potential for deployment as artificial synapses in neuromorphic computing systems.

First, peel off several layers of ReS2 flakes on heavily doped p-type silicon wafers with 90 nm silicon oxide. Next, Raman spectroscopy was performed to align the crystal orientation ([010]). Then, EBI was performed by electron beam lithography (EBL). After that, the left/right electrode was patterned by EBL again. In addition, a metal electrode layer of 10 nm Ti and 90 nm Au was deposited on the wafer by an electron beam evaporator. The chip is then placed in acetone for a standard peeling process. Finally, the device is passivated by the PMMA layer.

The DC scan is measured at room temperature and environmental conditions using a probe station and Agilent 4155B semiconductor analyzer. During all IV measurements, the Si substrate is grounded. Pulse measurement is performed by 4200-SCS Keithley semiconductor analyzer and Cascade probe station. Cryogenic measurement is performed under vacuum by Lakeshore CRX-VF cryogenic probe station.

Analyze ReS2 samples with and without EBI by AFM (Bruker Dimension Fastscan), Raman spectroscopy, PL spectroscopy (Witec Alpha 300R), XPS (Quantera PHI II), TEM and STEM (Talos F200X). The excitation laser wavelength of Raman and PL spectrometers is 532 nm. The laser power is less than 100 μW to prevent additional damage to the sheet. For Raman spectroscopy, the peak position refers to the silicon peak at 520 cm-1. XPS is collected by a monochromatic Al Kα X-ray source. The binding energy of all peaks is based on C 1s (285 eV).

The data supporting the plot of this article and other findings of this research can be obtained from the corresponding author upon reasonable request.

Wang, Z. etc. A fully memristive neural network for the classification of unsupervised learning patterns. Nat. electronic. 1, 137–145 (2018).

Wang, Z. etc. The memristor with diffusion dynamics serves as a synaptic simulator for neuromorphic calculations. Nat. alma mater. 16, 101–108 (2017).

Kuzum, D., Yu, S. & Philip Wong, H.-S. Synaptic Electronics: Materials, Equipment and Applications. Nanotechnology 24, 382001 (2013).

Feng, X., Liu, X. & Ang, K.-W. Two-dimensional photon memristors beyond graphene: progress and prospects. Nanophotonics 9, 1579–1599 (2020).

Wang, R. et al. The latest development of volatile memristors: devices, mechanisms and applications. Advanced Intel. system. 2. 2000055 (2020).

Li, Y. & Ang, K. -W. Hardware implementation of neuromorphic computing using a large-scale cross array of memristors. Advanced Intel. system. https://doi.org/10.1002/aisy.202000137 (2020).

Waser, R., Dittmann, R., Staikov, G. & Szot, K. Redox-based resistive switching memory-nano-ion mechanism, prospects and challenges. Senior alma mater. 21, 2632–2663 (2009).

Kim, S. et al. Experimental demonstration of the second-order memristor and its biologically realistic ability to achieve synaptic plasticity. Nanolet. 15, 2203-2211 (2015).

Lee, M.-J. Wait. A fast, high-durability and expandable non-volatile storage device made of an asymmetric Ta2O5-x/TaO2-x double-layer structure. Nat. alma mater. 10, 625–630 (2011).

Parker, G.-S. Wait. Observe in situ the filamentary conductive channels in the asymmetric Ta2O5-x/TaO2-x double-layer structure. Nat. Community. 4, 2382 (2013).

Shibuya, K., Dittmann, R., Mi, S. & Waser, R. The effect of defect distribution on the switching characteristics of Sr2TiO4 thin film resistors. Senior alma mater. 22, 411–414 (2010).

Wong, H.-P. etc. Metal oxide RRAM. Process IEEE 100, 1951–1970 (2012).

Yang, JJ, Strukov, DB & Stewart, DR Memristive devices used in calculations. Nat. nanotechnology. 8, 13–24 (2013).

Kim, K.-H. Wait. A functional hybrid memristor cross-array/CMOS system for data storage and neuromorphic applications. Nanolet. 12, 389–395 (2012).

Yang, Y. et al. Observation of conductive filament growth in nano-scale resistance memory. Nat. Community. 3, 1-8 (2012).

Yang, Y. et al. Electrochemical dynamics of nanoscale metallic inclusions in dielectrics. Nat. Community. 5, 1-9 (2014).

Waser, R. & Aono, M. Nano-ion-based resistive switch memory. Nat. alma mater. 6, 833–840 (2007).

Choi, S. etc. SiGe epitaxial memory for neuromorphic computing with reproducible high performance based on engineering dislocations. Nat. alma mater. 17, 335–340 (2018).

Akinwande, D. etc. Graphene and 2D materials for silicon technology. Nature 573, 507–518 (2019).

Jurca, T. etc. Low-temperature atomic layer deposition of MoS2 thin films. Angie. Chemistry International. Ed. 56, 4991–4995 (2017).

Cui, F. etc. Tellurium-assisted epitaxial growth of a large-area, highly crystalline ReS2 atomic layer on a mica substrate. Senior alma mater. 28, 5019-5024 (2016).

Gao, L. et al. Face-to-face transfer of wafer-level graphene films. Nature 505, 190–194 (2014).

Kang, K. et al. Assemble two-dimensional materials layer by layer into a wafer-level heterostructure. Nature 550, 229–233 (2017).

Feng, X. et al. Fully printed flexible MoS2 memristive artificial synapse with flying focus switching energy. Advanced electronics. alma mater. 5. 1900740 (2019).

Shi, Y. etc. Electronic synapses made of layered two-dimensional materials. Nat. electronic. 1, 458–465 (2018).

Ge, R. etc. Atomic resistor: Non-volatile resistance switching in transition metal dichalcogenide atomic sheets. Nanolet. 18, 434–441 (2018).

Xu, R. etc. The vertical MoS2 double-layer memristor with electrochemical metallization acts as an atomic level synapse, and the switching threshold is close to 100 mV. Nanolet. 19, 2411–2417 (2019).

Besonov, AA etc. Layered memristive and memristive switches for printable electronic products. Nat. alma mater. 14, 199–204 (2015).

Hehe, W. Wait. Synaptic resistors based on phase-engineered two-dimensional heterostructures. Senior alma mater. 30, 1801447 (2018).

Pi, S. etc. A crossed array of memristors with a half-pitch of 6 nanometers and a critical dimension of 2 nanometers. Nat. nanotechnology. 14, 35–39 (2019).

Xia, Q. & Yang, JJ Memristive crossover array for brain-like computing. Nat. alma mater. 18, 309–323 (2019).

Li, D. etc. The MoS2 memristor exhibits variable switching characteristics for bio-realistic synapse simulation. ACS Nano 12, 9240–9252 (2018).

Zhu, X., Li, D., Liang, X. & Lu, WD Ion modulation and ion coupling effects in MoS2 devices for neuromorphic calculations. Nat. alma mater. 18, 141–148 (2019).

Sangwan, VK, etc. Multi-terminal memory transistor of polycrystalline single-layer molybdenum disulfide. Nature 554, 500–504 (2018).

Wang, L. et al. Artificial synapses based on multi-terminal memory transistors for neuromorphic applications. Advanced Features. alma mater. 29, 1901106 (2019).

Tang Gai, S. et al. Due to the decoupling of electrons and vibrations, single-layer behavior in bulk ReS2. Nat. Community. 5, 3252 (2014).

Shim, J. et al. High-performance two-dimensional rhenium disulfide (ReS2) transistor and photodetector processed by oxygen plasma. Senior alma mater. 28, 6985–6992 (2016).

Horzum, S. etc. The formation and stability of single-layer rhenium disulfide point defects. Physics Rev. B 89, 155433 (2014).

He, Q. etc. In-depth exploration of the anisotropic evolution of single-layer ReS2 large-scale growth. ACS application alma mater. Interface 12, 2862–2870 (2020).

Lin, Y.-C. Wait. Single-layer ReS2: A two-dimensional semiconductor with adjustable in-plane anisotropy. ACS Nano 9, 11249–11257 (2015).

Xu, K., Cao, P. & Heath, JR Scanning tunneling microscope characterization of the electrical properties of wrinkles in exfoliated graphene monolayers. Nanolet. 9, 4446–4451 (2009).

Chenette, DA etc. The in-plane anisotropy of single-layer and few-layer ReS2 was detected by Raman spectroscopy and scanning transmission electron microscopy. Nanolet. 15, 5667–5672 (2015).

Huang, W. etc. Ion beam defect engineering of ReS2/Si photocathode with significantly enhanced hydrogen evolution reaction. Senior alma mater. Interface 6, 1801663 (2019).

McDonnell, S., Addou, R., Buie, C., Wallace, RM & Hinkle, CL Defect-dominated doping and contact resistance in MoS2. ACS Nano 8, 2880–2888 (2014).

Jadwiszczak, J. et al. MoS2 memristor manufactured by local helium ion beam irradiation. ACS Nano 13, 14262–14273 (2019).

Song, SH, Joo, M.-K., Neumann, M., Kim, H. & Lee, YH Defect dynamics in a single layer of molybdenum disulfide probed by noise nanospectroscopy. Nat. Community. 8, 2121 (2017).

Noh, J.-Y., Kim, H. & Kim, Y.-S. Stability and electronic structure of natural defects in monolayer MoS2. Physics Rev. B 89, 205417 (2014).

Osvald, J. Asymmetrical Schottky diodes are connected back to back, and the series resistor acts as a single diode. Physical State Solidi A 212, 2754–2758 (2015).

Liu, Y. et al. Close to the Schottky-Mott limit in the van der Waals metal-semiconductor junction. Nature 557, 696–700 (2018).

Schulman, DS, Arnold, AJ & Das, S. Contact engineering of two-dimensional materials and equipment. Chemical society. Rev. 47, 3037–3058 (2018).

Liu, Q. et al. Implantation of metal ions is used to improve the resistance switching characteristics of ZrO2 based ReRAM. Proc in 2009. European Solid State Device Research Conference (IEEE, Athens, Greece, 2009).

Lanza, M. etc. Research recommended methods for resistive switching devices. Advanced electronics. alma mater. 5. 1800143 (2019).

Wei, Z. et al. Highly reliable TaOx ReRAM and direct evidence of redox reaction mechanism. 2008 IEEE International Electronic Devices Conference (IEEE, San Francisco, California, USA, 2008).

Yin, JH, etc. By inserting Ru nano-dots, the uniformity of the switching parameters of the TiO2 thin film resistor is greatly improved. Senior alma mater. 25, 1987–1992 (2013).

Chang, W.-Y., Lin, C.-A., He, J.-H. & Wu, T.-B. Resistance conversion behavior of ZnO nanorod layer. App Physics Wright. 96, 242109 (2010).

Zidan, MA, Strachan, JP & Lu, WD The future of electronic products based on memristive systems. Nat. electronic. 1, 22–29 (2018).

Wang, Z. etc. Resistance switch material for information processing. Nat. Priest. 5, 173–195 (2020).

Chen, P. -Y., Peng, X. & Yu, S. NeuroSim: An integrated device-to-algorithm framework for benchmarking synaptic devices and array architectures. 2017 IEEE International Electronic Devices Conference (IEDM) (IEEE, San Francisco, California, USA, 2017).

Yu, S. Neurally inspired computing with emerging non-volatile memory. Process IEEE 106, 260–285 (2018).

Chen, L. et al. Van der Waals synaptic transistor based on ferroelectric Hf0.5Zr0.5O2 and two-dimensional tungsten disulfide. Advanced electronics. alma mater. 6. 2000057 (2020).

Benda, J. & Herz, AVM A general model for spike frequency adaptation. Neural computing. 15, 2523–2564 (2003).

Bi, G. & Poo, M. Synaptic modification in cultured hippocampal neurons: dependence on spike time, synaptic strength, and postsynaptic cell type. J. Neuroscience. 18, 10464–10472 (1998).

Wu, Y. et al. AlOx-based resistive switching devices with progressive resistance modulation for neuromorphic device applications. The 4th IEEE International Memory Symposium in 2012 (IEEE, Milan, Italy, 2012).

Song, S., Miller, KD & Abbott, LF conducts competitive Hebbian learning through synaptic plasticity dependent on spike time. Nat. Neuroscience. 3, 919–926 (2000).

This research was supported by the A*STAR Science and Engineering Research Committee (No. A2083c0061) and the National Research Foundation of Singapore (NRF-CRP22-2019-0007). SL and BL contribute the same to this work. We thank X. Gong from the Department of Electrical and Computer Engineering, National University of Singapore, for his valuable assistance in equipment measurement.

These authors made equal contributions: Li Sifan, Li Bochang.

Department of Electrical and Computer Engineering, National University of Singapore, 4 Engineering Drive 3, Singapore, 117583, Singapore

Sifan Li, Bochang Li, Xuewei Feng, Li Chen, Yesheng Li, Li Huang, Xuyao ​​Fong & Kah-Wee Ang

Advanced 2D Materials Center, National University of Singapore, 6 Science Drive 2, Singapore, 117546, Singapore

Sifan Li, Bochang Li, Xuewei Feng, Li Chen, Yesheng Li, Li Huang, Xuyao ​​Fong & Kah-Wee Ang

You can also search for this author in PubMed Google Scholar

You can also search for this author in PubMed Google Scholar

You can also search for this author in PubMed Google Scholar

You can also search for this author in PubMed Google Scholar

You can also search for this author in PubMed Google Scholar

You can also search for this author in PubMed Google Scholar

You can also search for this author in PubMed Google Scholar

You can also search for this author in PubMed Google Scholar

The project was supervised and directed by K.-WASL, and K.-WA conceived and designed the experiment. SL and BL have the same contribution. SL carried out equipment manufacturing. Electrical measurements were taken for SL and BL. LC performed material characterization. All authors participated in the discussion and analysis of the results. SL and K.-WA wrote the manuscript.

Correspondence with Kwong Yaofang or Kah-Wee Ang.

The author declares no competing interests.

The publisher states that Springer Nature remains neutral on the jurisdiction claims of published maps and agency affiliates.

Open Access This article has been licensed under the Creative Commons Attribution 4.0 International License Agreement, which permits use, sharing, adaptation, distribution and reproduction in any media or format, as long as you appropriately indicate the original author and source, and provide a link to the Creative Commons license , And indicate whether any changes have been made. The images or other third-party materials in this article are included in the article’s Creative Commons license, unless otherwise stated in the material’s credit line. If the article’s Creative Commons license does not include the material, and your intended use is not permitted by laws and regulations or exceeds the permitted use, you need to obtain permission directly from the copyright owner. To view a copy of this license, please visit http://creativecommons.org/licenses/by/4.0/.

Li, S., Li, B., Feng, X. etc. Low-variability electron beam irradiated rhenium disulfide memristor for neuromorphic calculations. npj 2D Mater Appl 5, 1 (2021). https://doi.org/10.1038/s41699-020-00190-0

DOI: https://doi.org/10.1038/s41699-020-00190-0

Anyone you share the following link with can read this content:

Sorry, there is currently no shareable link in this article.

Provided by Springer Nature SharedIt content sharing program

npj 2D Mater Appl (npj 2D Mater Appl) ISSN 2397-7132 (online)