Volatile and Nonvolatile Resistive Switching in Lateral 2D Molybdenum Disulfide-Based Memristive Devices

Abstract

Developing electronic devices capable of emulating biological functions is essential for advancing brain-inspired computation paradigms such as neuromorphic computing. In recent years, two-dimensional materials have emerged as promising candidates for neuromorphic electronic devices. This work addresses the coexistence of volatile and nonvolatile resistive switching in lateral memristors based on molybdenum disulfide with silver as the active electrode. The fabricated devices exhibited switching voltages of ∼0.16 V and ∼0.52 V for volatile and nonvolatile operation, respectively, under direct-current measurements. They also displayed the essential synaptic functions of paired-pulse facilitation and short- and long-term plasticity under pulse stimulation. The operation mechanism was investigated by in situ transmission electron microscopy, which showed lateral migration of silver ions along the molybdenum disulfide between electrodes. Based on the experimental data, a macroscopic semiclassical electron transport model was used to reproduce the current–voltage characteristics and support the proposed underlying switching mechanisms.

Emulating the brain and its biological functions with circuits based on conventional complementary metal-oxide semiconductor (CMOS) technology requires complex circuitry, which leads to high energy consumption. , In this scenario, resistive switching (RS) devices are promising candidates for implementing the desired artificial neurons and multifunctional synaptic devices in neuromorphic hardware. Among the wide variety of RS devices, electrochemical metallization (ECM) memristors based on two-dimensional (2D) materials exhibit particularly promising features as vertical and lateral memristors. − ECM memristors are characterized by the formation and rupture of metallic conductive filaments (CFs) under an applied electrical field. − These CFs usually originate from the migration of active metal ions such as silver (Ag) or copper (Cu). Up to date, CFs composed of Ag ions have been visualized in several device architectures. ,− ECM memristive devices based on layered 2D materials (2DMs), including molybdenum disulfide (MoS₂), have demonstrated low energy consumption of approximately 10 fJ, low switching voltages of <0.3 V, and RS in subnanometer thicknesses. ,− However, the growth and evolution of metallic CFs in 2DM ECM memristors rely on random defects, which often lead to high cycle-to-cycle and device-to-device variability.

Lateral 2DM-based devices, in which 2DMs are placed on a dielectric substrate and contacted by metal electrodes, have been extensively studied as RS devices. ,,− Lateral memristive devices present three key benefits over conventional vertical stacks: (i) the selector functionality can be easily integrated; (ii) they enable simultaneous signal transmission and learning; and (iii) they support spatiotemporal information processing. , However, the switching mechanisms in lateral devices for volatile and nonvolatile operation (and their coexistence) remain largely unexplored. In fact, the coexistence of both RS modes by tuning the measurement system’s current compliance (CC) has been only shown for vertical devices based on metal oxides and hexagonal boron nitride. − The coexistence of both RS modes within a single device could enhance the implementation of both synaptic devices and artificial neurons for future neuromorphic systems.

Here, we investigate the coexistence of volatile and nonvolatile RS in lateral MoS₂-based memristors using Ag as the filament-forming electrode. We show the transition between volatile and nonvolatile switching modes by tuning the CC of direct-current (DC) electrical measurements. Furthermore, we demonstrate the coexistence of both modes under pulsed voltage stress (PVS). We also show short-term (STP) and long-term plasticity (LTP) in response to different pulsed stimulation and analyze the RS mechanisms for both volatile and nonvolatile switching modes. The growth and stability of the CFs depend on several factors, such as the filament morphology and the applied electrical field. ,, Therefore, we conducted in situ transmission electron microscopy (TEM) images to analyze Ag migration along the MoS₂. Finally, we applied a semiclassical electron transport model that reproduces the experimental current–voltage (I-V) characteristics and supports the formation of metallic Ag CFs as the switching mechanism.

Our lateral devices comprise multilayer MoS₂ contacted by palladium (Pd) and active Ag electrodes with a submicron channel length (see schematic in Figure a). First, MoS₂ grown by metal–organic chemical vapor deposition (MOCVD) on a 2” sapphire wafer was wet transferred onto a SiO₂/Si substrate. The Pd electrodes were defined using optical lithography, and the Ag ones were defined using a laser writer. The Ag electrodes were covered in situ with a 50 nm aluminum (Al) electron-beam-evaporated (e-beam) capping layer to avoid tarnishing. Reactive ion etching (RIE) was carried out to pattern the MoS₂ channels. More details about the device fabrication are available in the Materials and Methods section in the Supporting Information (SI), and a schematic process flow with each step is displayed in Supplementary Figure S1.

1.

Device structure and material characterization. (a) Cross-sectional schematic of the lateral MoS₂-based memristor. (b) HRTEM image of the multilayer MoS₂ film between the metal electrodes. (c) AFM image of the as-grown MoS₂ on sapphire. (d) Top-view optical microscopy image of a fabricated lateral device. (e) TEM image showing the cross section of a device. The channel length was approximately 250 nm. Al₂O₃ was e-beam-evaporated to protect the MoS₂ during FIB.

The devices were characterized by high-resolution TEM (HRTEM). The layered structure of the MoS₂ in a device in Figure b shows an interlayer distance of 6.4 ± 0.05 Å, in line with values reported in the literature. , The atomic force microscopy (AFM) image in Figure c reveals a polycrystalline nature of the as-grown MoS₂ on 2” sapphire. Additional AFM confirms the thickness of ∼4.4 nm of the MoS₂ film (Supplementary Figure S2a). In addition, we conducted Raman spectroscopy measurements of the as-grown MoS₂ on sapphire and after transfer on SiO₂/Si. The extracted E¹ _2g and A_1g peaks of MoS₂ coincide with the literature values for more than four layers or the bulk, which is in agreement with the AFM and HRTEM data (see Figure S2b). , Furthermore, the photoluminescence peak position of 1.82 eV corroborates the multilayered nature of the MoS₂ and the good crystal quality of the transferred material (Figure S2c-d). − Figure d shows a top-view optical microscopy image of a fabricated device. A lamella was cut from a device by focused ion beam (FIB) milling, and a cross-sectional TEM image was taken to verify the integrity of the lamella. The channel length between the metal electrodes was ∼250 nm in this specific device. For the in situ TEM measurements, the channels of the devices were covered with an additional 80 nm e-beam-evaporated aluminum oxide (Al₂O₃) film to protect the MoS₂ from the ion beam during the FIB lamella preparation. Moreover, the Al₂O₃ layer prevents direct contact between the platinum (Pt) protection layer used in FIB and both the electrodes and MoS₂ layers, which would otherwise lead to a short circuit.

We conducted DC electrical measurements on the Pd/MoS₂/Ag/Al devices with CC values varying from 100 nA to 1 mA. More information regarding the influence of the CC values and the devices’ behavior can be seen in Figure S3. Volatile RS was observed for CC values from 100 nA up to 100 μA (Figure a). Figure b shows ten I-V curves with a 1 μA CC measured in subsequent voltage sweeps in the positive polarity and plotted in logarithmic scale (see linear scale in SI Figure S4a). The first sweep is marked in blue. We attribute the forming-free nature of our device to the submicron channel length which can promote filament formation at lower voltages. The voltage bias was applied to the Ag active electrode, while the Pd electrode was grounded. The arrows indicate the voltage sweep direction. Initially, the device was in a high-resistance state (HRS) until the SET transition occurred at an on-threshold voltage of V _t,on = 0.19 V, when it switched to its low-resistance state (LRS). During the backward sweep, the device switched back to the HRS at a hold voltage of V _hold = 0.087 V. We plotted V _t,on and V _hold in histograms, and the data were fitted with Gaussian distributions that show reasonably low standard deviations (σ) of 0.04 and 0.03 V, respectively (Figure S5). The mean V _t,on of 0.16 V is the lowest reported among similar lateral MoS₂-based RS devices (see Table ). ,,,, In addition to the table, Figure S6 presents the key benchmarking values including the material, channel length and V _t,on in a concise graph. Figure c shows the normalized cumulative distribution functions (CDFs) of V _t,on and V _hold for 60 consecutive volatile cycles, with a standard error of 3.9% and 3.1%, respectively (see Figure S4b for the corresponding I-V curves). Furthermore, Figure S7a shows the behavior of V _t,on for 5 devices with different channel lengths. The V _t,on decreases with decreasing channel lengths which is in good agreement with the literature. ,,

2.

DC electrical performance of the lateral MoS₂-based devices. (a) Volatile RS at different CC values from 100 nA up to 100 μA. (b) 10 subsequent volatile switching curves recorded with positive voltage polarity for 1 μA CC. The arrows indicate the voltage sweeping directions, with the first sweep marked in blue. (c) CDFs of the V _t,on and the V _hold of 60 consecutive volatile I–V curves. (d) Nonvolatile RS at 500 μA and 1 mA CC. (e) 10 consecutive nonvolatile I-V curves for 1 mA CC. The inset shows the resistance values for ∼860 nonvolatile consecutive cycles. (f) CDFs of the V _SET and V _RESET of ∼860 nonvolatile cycles.

1. Comparison of Our Work with Other Similar 2D-Based Literature.

ref.	Device structure	Material	Thickness and Channel length	Forming voltage (V)	Volatile (Vt,on)	Nonvolatile (VSET)	In situ switching mechanism
This work	Lateral	MOCVD-grown MoS 2	4.4 nm	Forming-free	O (0.16 V)	O (0.52 V)	O
			250–900 nm
	Lateral	CVD-grown MoS2	Monolayer	∼20 V	–	O (3.5–8.3 V)	O
			2 μm
	Lateral	CVD-grown MoS2	Monolayer	∼4 V	O (1.2 V)	–	O
			500 nm
	Lateral	Mechanically exfoliated MoS2	<6 layers	1.8 V	O (0.35 V)	–	–
			18 nm
	Lateral	Mechanically exfoliated MoS2	4–30 layers	∼3.8 V	–	O (2 V)	O
			250 nm
	Lateral	CVD-grown MoS2	Monolayer	Forming-free	–	O (10–20 V)	O
			5 μm
	Lateral	Exfoliated MoS2 nanoflakes	>1 layer	–	–	O (∼6 V)	O
			15 nm
	Vertical	CVD-grown vertically aligned MoS2	∼23 nm	Forming-free	O (∼0.35 V)	–	–
			23 nm

Nonvolatile bipolar RS was observed for CC values above 100 μA (Figure d). We then performed ten consecutive I-V sweeps with 1 mA CC, with the first sweep marked in red (Figure e). The extracted mean SET and RESET voltages of these switching cycles were V _SET = 0.26 V and V _RESET = −0.36 V. The inset in Figure e shows the DC endurance of a single device for ∼860 consecutive nonvolatile RS cycles with an average HRS-to-LRS resistance ratio (R _HRS/R _LRS) of more than 10⁴. We further measured the retention time of both the HRS and the LRS of a device (Figure S8). The device remained in the LRS for at least 8 days and in the HRS for ∼12 h. Furthermore, we measured the retention of 5 different LRS states for at least 1000 s (Figure S9). Figure f displays the corresponding cumulative voltage distributions with a mean V _SET and V _RESET of ∼0.52 V and ∼−0.61 V, and standard deviations of σ = 0.3 V and σ = 0.4 V, respectively. Figure S10 shows the CDFs of V _set and V _reset for 4 devices with a standard error of 5.07% and 4.4% for V _SET and V _RESET, respectively. We further demonstrated that not only can we observe volatile and nonvolatile switching but also, we can control the transition between both in the lateral MoS₂ devices solely by modifying the CC value (see Figure S11).

We conducted PVS tests with different pulses to observe the volatile and nonvolatile dynamic responses of our devices. Figure a shows the volatile transient current response of a device to a 1 μs voltage pulse of 6 V. The switching time of t _on = 250 ns is the time required to reach 90% of the ON current of I _on = 12.5 μA. The relaxation time of t _r = 131 ns is defined as the time needed to reach 10% of the difference between I _on and the OFF current (I _off) when switching back to the HRS. At the end of the voltage pulse, the device spontaneously relaxes to its initial OFF state. Furthermore, we measured the endurance of a device under PVS for over 1060 volatile RS cycles (Figure S12). The on-state resistance (R _on) exhibits low cycle-to-cycle variability with a R _off /R _on ratio of over 12. Figure b, in contrast, shows the transient nonvolatile output current response to a 0.02 s voltage pulse of V_SET = 3.5 V and a 0.05 s voltage pulse of V_RESET = −5 V. Read voltage pulses of V _read = 0.2 and 0.1 V were applied to measure the resistances in the LRS and HRS, respectively. After the SET pulse, the current did not return to its initial OFF-state but remained in the ON-state. The current pulse levels for the nonvolatile switching are over 100 μA, which matches the levels applied in the DC CC-dependent measurements. The R _LRS after the SET pulse was ∼1.4 kΩ, whereas the R _HRS was ∼200 MΩ.

3.

Current response over time of a lateral memristive device upon PVS. (a) Voltage pulse of 6 V for 1 μs showing volatile RS. (b) Nonvolatile RS upon PVS with a 3.5 V/20 ms SET pulse and a −5 V/50 ms RESET pulse. (c) Dynamic response of a device during a sequence of 40 voltage pulses of 1 V for 50 ns showing short-term potentiation with forgetting upon stimuli removal. (d) Train of 20 consecutive 2 V/2 ms pulses showing long-term plasticity. The read voltage applied between each pulse was 0.3 V. The inset shows the gradual increase in the current after each applied pulse.

STP and LTP were investigated by applying a series of consecutive voltage pulses. First, a train of 40 consecutive 1 V/50 ns pulses with a 50 ns period between them was applied and the transient response of our device recorded. The current response in Figure c shows a gradual current increase with each pulse, which saturates at 13 nA after about 18 pulses. The current decreases slightly after each pulse and decays to its initial state after the final voltage pulse. The fast relaxation to the initial OFF-state with no memory retention and a relaxation time of t _r = 2.2 μs, emulates STP, i.e., a forgetting process observed in biological synapses. We also achieved LTP in the same device by tuning the programming parameters so the device current increased to a higher value, triggering the nonvolatile regime. Figure d shows the device current response to a train of 20 consecutive pulses of an increased voltage of 2 V and a much longer duration of 2 ms. A magnification of the current response in the inset shows a gradual current increase after each applied pulse, resembling the biological synaptic function known as paired-pulse facilitation (PPF). After the final pulse, the current decayed to an intermediate ON-state, in contrast to the previous experiment with 40 smaller, shorter pulses. The retention time of the LRS was recorded for over ∼100 s, more than the time displayed in Figure d (see SI Figure S13). This device behavior aligns with the LTP concept, as the LRS lasts for more than a few tens of seconds. − LTP, long-term depression (LTD) and the increase in postsynaptic conductance (PSC) with varying pulse amplitudes were further investigated. Figure S14 shows the measured synaptic characteristics, specifically the obtained PSC as a function of the number of applied pulses for different amplitudes and polarity configurations. The application of positive pulses (Figure S14a) leads to an increase in PSC, demonstrating LTP. On the contrary, the introduction of a train of negative pulses (Figure S14b) results in a decrease in the PSC, showing Ltd. Figure S14c finally depicts the PSC as a function of the number of pulses for different pulse amplitudes, ranging from 2 V up to 5 V. Controlling the transition from STP to LTP by tuning the programming parameters of the pulse train is in line with previous works on SiO₂ and Ag₂S-based memristors. ,

We further investigated the current conduction mechanisms of our memristive devices. We plotted the I–V curves from Figures b (volatile RS) and 2e (nonvolatile) in double-logarithmic scale in Figure a and Figure b, respectively. In the HRS, the I-V curves exhibit a square-law dependence, indicating space-charge-limited conduction (SCLC), , which has been previously reported in two-terminal lateral polycrystalline MoS₂-based devices. In the LRS, the linear I-V dependence indicates ohmic conduction. This transition from SCLC in the HRS to ohmic transport in the LRS supports the proposed formation of Ag CFs. Temperature-dependent I–V measurements confirmed these transport mechanisms (Figure S15). In the HRS, the current increases with increasing temperature, which is characteristic of transport through the MoS₂ channel and SCLC. In the LRS, the current decreases with increasing temperature. Such a negative temperature coefficient of resistivity is characteristic of transport through metallic CFs. Both volatile and nonvolatile RS mechanisms exhibit similar transport characteristics, suggesting the formation of a metallic CF in both cases. However, the applied CC plays a crucial role in determining the RS behavior. We propose that low CC values lead to volatile RS with a weak Ag filament that self-ruptures upon removing the applied voltage. In contrast, higher CC values lead to nonvolatile RS because more Ag ions diffuse into the channel and create a stronger Ag CF. Therefore, we deduce that Ag ions diffuse into the channel driven by the lateral electric field and eventually form continuous Ag filaments. A top-view schematic that illustrates the RS mechanisms for the HRS and the LRS is shown in Figure S16.

4.

I-V characteristics of (a) a volatile and (b) a nonvolatile operation. In both cases, HRS shows SCLC, while LRS shows ohmic transport.

We conducted in situ TEM on lateral Pd/MoS₂/Ag devices. Figure a shows the typical TEM results from a fresh device in the HRS (without any electrical measurement before), including a bright-field TEM image, high-angle annular dark-field (HAADF) images and the corresponding energy-dispersive X-ray spectroscopy (EDXS) elemental maps of Mo, S, and Ag. No traces of Ag are observed between the electrodes within the MoS₂ channel. This observation is confirmed in Figure S17a, where the measured EDXS line profile for Ag is flat across the multilayered MoS₂. The Pd electrode of the device lamella was then electrically biased in situ in the TEM while the Ag electrode was grounded (see Figure S17c). The cross-sectional TEM image in Figure b shows the lamella device after in situ nonvolatile switching to the LRS. The arrows on the TEM image indicate three different regions in which HAADF images and their corresponding EDXS maps were taken (Figure c-e). The presence of Ag is clearly visible in these elemental maps. Further, the EDXS line profile of Ag coincides with those of Mo and S, proving the presence of Ag within the MoS₂ channel (Figure S17b). This confirms the formation of an Ag CF following the SET transition. These findings experimentally prove Ag ion migration driven by an applied lateral electrical field, in line with previous reports in the literature. ,,

5.

In situ TEM measurements for the resistive switching mechanism analysis. (a) Cross-sectional bright-field TEM image, HAADF image and the corresponding EDXS elemental maps of Mo, S, and Ag from a lamella in the HRS before any electrical measurement. The blue arrow denotes the region where the EDXS elemental maps were acquired. (b) Cross-sectional TEM image of a lamella after nonvolatile RS in the LRS. (c–e) EDXS elemental maps of the lamella in the LRS taken on the regions denoted as 1, 2, and 3 in panel b, respectively.

We investigated lateral memristive devices with polycrystalline 2D multilayer MoS₂ channels with submicrometer lengths. We demonstrated the coexistence of volatile and nonvolatile RS in the same devices, attainable with both DC and PVS. Our devices display repeatable RS with low switching voltages for both volatile (0.16 V) and nonvolatile (0.52 V) operation. The volatile behavior may be utilized in artificial neurons. We also experimentally demonstrated PPF, STP and LTP behavior in our devices, showcasing their potential as artificial synapses for neuromorphic systems. We investigated the current conduction mechanisms in the HRS and LRS, concluding that SCLC and ohmic transport are the dominant mechanisms, respectively. This behavior can be explained by the formation of Ag conductive filaments, which was confirmed via in situ TEM switching experiments. The flexibility of volatile and nonvolatile switching in our lateral memristors can be used to implement both synaptic devices and artificial neurons for future neuromorphic systems.

The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/acs.nanolett.5c01992.

• Materials and methods including MoS₂ growth process, lateral device fabrication, material and device characterization, and electrical measurements; MoS₂ material characterization; schematic of the fabrication process; volatile I-V curves in linear and log scale; histogram and Gaussian fit for the on-threshold and hold voltages; transition from volatile to nonvolatile back and forth; LRS retention time; lamella and setup preparation for in situ TEM; benchmarking with previous works (PDF)

M.C.L and A.D. conceived and designed the project. S.C. carried out the material characterization, mask design, device fabrication, and electrical measurements. B.C. supported S.C. with the AFM measurements. H.K, A.V. and M.H. provided the polycrystalline MoS₂. K.R. and J.J. performed in situ TEM and EDXS experiments. M.D.G. supported the analysis of the electrical data and the writing. S.C. wrote the initial manuscript, and all authors cowrote it. All authors have given approval to the final version of the manuscript.

The authors declare no competing financial interest.