Emerging non-volatile memories such as memristive devices that can act as artificial synapses, have attracted huge interest recently in the field of neuromorphic computing, due to their unique capabilities, including high-density integration, fast write and read speed, and compatibility with the existing complementary metal oxide semiconductor process (Yang et al., 2013; Wang et al., 2017; Zidan et al., 2018; Xia and Yang, 2019). In memristive devices data can be written, processed, and erased by changing resistance states between one or multiple low resistance (LRS) and high resistance states (HRS), while the original resistance state, usually, remains unchanged (Raffone, 2017; Sato et al., 2021; Alshaya et al., 2022).

Several material systems, including transition metal oxides (TMOs), such as HfO₂ (Maldonado et al., 2023), Y₂O₃ (Kumar et al., 2022a; Kumar et al., 2022b; Kumar et al., 2022c; Gautam et al., 2023; Kumar et al., 2023; Kumbhar et al., 2024), Ta₂O₅ (Kim et al., 2022), and TiO₂ (Strukov et al., 2008)), and 2D transition metal dichalcogenides (TMDs), like MoS₂ (Naqi et al., 2022), WS₂ (Cao et al., 2022), and MoSe₂ (Duan et al., 2023), have been extensively employed to develop memristive devices for numerous applications (Cao et al., 2022; Naqi et al., 2022; Duan et al., 2023).

Notably, 2D TMD materials offer outstanding electronic, optical, and mechanical properties (Ge et al., 2018; Kim et al., 2018), as compared to conventional TMO materials, which further enables significant advantages (Marseglia, 1983; Yin et al., 2019; Luo et al., 2020), including low-power switching (Feng et al., 2019; Yan et al., 2019; Ranganathan et al., 2020; Lu et al., 2021; Li et al., 2022), device thermal stability (Tong and Liu, 2023), electrostatic tunability (Sangwan et al., 2015), and mechanical flexibility (Ge et al., 2018). Therefore, the integration of layered 2D TMD materials with TMO materials in memristive devices can further enhance thermal stability, induce switching at lower voltages, and enable low power/energy operation (Pickett et al., 2009; Wang et al., 2021; Li et al., 2023; Lin et al., 2024). Moreover, such memristive structures can be utilized to emulate artificial synapses by mimicking basic synaptic functions, such as potentiation, depression and voltage-dependent synaptic responses.

Several successful attempts have been demonstrated to physically model the metal oxide-based memristive devices. For example, Kim et al (Kim et al., 2013a) have reported a physical electro-thermal model of Ta₂O₅ bilayer-based resistive memory by utilizing COMSOL Multiphysics. Bocquet et al (Bocquet et al., 2014) and Traore et al (Traore et al., 2016) have presented the physics-based compact numerical models and density function theory (DFT)-based calculation by adopting generalized gradient approximation (GGA) and Perdew–Burke–Ernzerhof functional (PBE) for HfO₂-based RRAM (Traore et al., 2016), respectively. Kumar et al (Kumar et al., 2022d) have implemented an Y₂O₃-based nanoscale memristor emulating synaptic functionality. However, the aforementioned metal oxide-only based memristive devices have several disadvantages, including thermal instability (Korotcenkov and Cho, 2012), high switching voltage (Ielmini, 2016), and limited tunability of the device conductance.

In this work, we propose an integration of a number of 2D MoS₂ layers with ultrathin Y₂O₃ in a memristive device structure that is operated as artificial synapse. Here, it should be noted that the integration of 2D MoS₂ with metal oxide memristive structure can be implemented in optoelectronic memories and may also act as optoelectronic synapse for bionic visual applications. The proposed physical model is based on the quantitative thermodynamic numerical modelling of the memristive devices, while COMSOL Multiphysics is used to model the memristive device operation. COMSOL Multiphysics (Kim et al., 2013b) helps to solve the partial differential equations by utilizing Finite Element Method (FEM). The heat transfer and electrodynamic equations have been resolved by utilizing physical parameters of 2D MoS₂ and Y₂O₃.

Figure 2A illustrates the simulated RS behaviour of the 2D MoS₂ integrated thin Y₂O₃-based memristive device subjected to a voltage pulsing scheme in the range −1.5 V to +1 V in forward and reverse bias, delivered to the top electrode through a load resistance. The RS response is segmented into four phases: positive forming voltage (+V _F), positive SET voltage (V _SET), negative rupture voltage (-V _R), and negative RESET voltage (V _RESET).

During the formation process, all CFs are created within the MoS₂ and Y₂O₃ layer of the RS device, causing it to transition into the SET state when subjected to a voltage amplitude equal to or less than V _F (as illustrated in Figure 2A). For the single layer MoS₂ integrated Y₂O₃-based memristive device, V _F and V _SET are modelled to be +0.81 V and +0.91 V, respectively. Conversely, during the rupturing process, all previously formed CFs are disrupted, leading the memristive device to switch into the RESET state when exposed to a voltage amplitude equal to or less than -V _R. The modelled values for -V _R and V _RESET are −1.5 V and −0.89 V, respectively, while in the case of double layer MoS₂ with thin Y₂O₃ (shown in Figure 2B), the V _F is +0.83 V, V _SET is 0.62 V, -V _R is −1.18 V and V _RESET is −0.61 V. Subsequently, in triple layer MoS₂ (shown in Figure 2C), the V _F is +0.88 V, V _SET is 0.15 V, -V _R is −0.81 V and V _RESET is −0.34 V. From Figures 3A,B it is concluded that the SET and RESET voltage values decrease as the number of 2D MoS₂ monolayers increases. This proves that the presence of 2D MOS₂ plays a pivotal role in the switching process, as compared to primary metal oxide layer. It is well known that the 2D MoS₂ layer requires lower bias potential to switch the device in SET and RESET region, because it shows quantum confinement effect and high carrier mobility (Ling et al., 2023). Therefore, charge carriers can move more easily within the material under the influence of a lower electric field, facilitating faster switching at lower bias potentials. This attribute further enhances the low power computation capability of the presented memristive device structure (Xu et al., 2019).

Additionally, to demonstrate the synaptic plasticity characteristics of the memristive device, such as potentiation and depression, a series of positive and negative voltage pulses with an amplitude of ±1 V and a voltage rise rate (V _RR) of 100 V/s are administered to the device. Figure 4 shows the potentiation and depression functionality of single layer MoS₂ (Figure 4A), double layer MoS₂ (Figure 4B) and triple layer MoS₂ (Figure 4C). As depicted in Figure 4A, when subjected to positive voltage pulses, the synaptic weight or normalized conductance of the memristive device undergoes a continuous reinforcement. Conversely, in response to negative electrical stimuli the synaptic weight diminishes. This ongoing modulation of the device’s conductance closely mimics the synaptic plasticity mechanisms observed in the brain (Jo et al., 2010; Das et al., 2018).

As evident from Figure 4D, the synaptic weight (or device conductance) of the memristive device is linearly decreased with an increment in the number of MoS₂ layers within the device structure. It should be noted that in monolayer (∼0.65 nm thickness) MoS2, quantum confinement effects are more pronounced due to the atomic layer thickness of the material, resulting in enhanced carrier mobility (Ling et al., 2023). However, as the number of layers increases, these effects diminish, leading to reduced carrier mobility as well as conductivity. Additionally, this trend may be attributed to alterations in the lateral dimensions of the device. Moreover, in configurations with a higher number of MoS₂ layers, the role of the thin metal oxide layer in the switching process diminishes, with primary switching being predominantly influenced by the MoS₂ layers rather than the Y₂O₃ layer, which basically constitutes combination of interfacial and filamentary switching mechanism (Wang et al., 2016; Li et al., 2018; Ginnaram and Maikap, 2021; Pam et al., 2022). This shift in dominance significantly impacts the overall conductance of the device. In this study, a series of 100 identical pulses (comprising 50 positive and 50 negative pulses) with an amplitude of ±1 V and a pulse width of 10 ms are employed to analyse the synaptic characteristics. Consistent with findings in existing literature (Zhang et al., 2013; Jang et al., 2014; Liu et al., 2018; Kim et al., 2020), the metal/metal oxide interfaces demonstrate a gradual potentiation process and an abrupt depression process, as illustrated in Figure 4D. The sudden shift observed during the depression process is attributed to variations in the free energy at metal/metal oxide interfaces (Yu, 2017). Additionally, accounting for the metal/metal oxide free energy and induced oxide layer, non-identical spike pulses featuring varying pulse amplitudes or durations offer a potential solution to mitigate the high asymmetry ratio observed in potentiation and depression processes (Yu, 2017). However, employing non-identical spikes may introduce complexity to peripheral circuits and neuro-inspired computing systems (Yu, 2017).

Figure 5 depicts the impact of pulse amplitude on synaptic plasticity, delineating both potentiation and depression processes. The graphs illustrate how altering the pulse amplitude effectively modulates the device conductance for single (Figure 5A), double (Figure 5B) and triple layer MoS₂ (Figure 5A), akin to the way spikes in neural communication influence synaptic strength. Notably, successive potentiation spikes elevate memristive conductance, while subsequent depression spikes induce a cycle of conductance reduction. The different pulsing scheme (non-identical pulses) is utilized during the computation of Figure 5. Maldonado et al (Maldonado et al., 2023) have observed similar behaviour experimentally in HfO₂-based memristive devices. Therefore, considering the aforementioned analysis, it can be concluded that control over the MoS₂ number of layers affects the conductance but not the synaptic response of the Yttria-based memristive device.

We presented a physical model detailing the integration of a number of nanoscales 2D MoS₂ layers with a thin metal oxide-based memristive device, aimed at emulating artificial synapse functionalities. The simulations demonstrate a bipolar RS response across all different number of MoS₂ layers, with device switching voltages decreasing as the number of MoS₂ layers increases. This phenomenon enhances the feasibility of 2D materials combined with transition metal oxide-based memristive devices in low-power computing applications. Moreover, the MoS₂-modified memristive device structure successfully demonstrated synaptic plasticity in terms of potentiation and depression, while the device conductance can also be varied by tuning the amplitude of the input pulsing scheme. Notably, in monolayer MoS₂, quantum confinement effects are more dominant due to the atomic layer thickness of the material, resulting in enhanced carrier mobility and corresponding increment in the device conductance. Conversely, in double and triple-layer MoS₂, these effects are minimized, which further affects the synaptic response of the device. However, in double and tri-layer device structures, the bulk materials conductivity may play a more important role, affecting thus the device switching voltages more significantly, as compared to the overall effect on device conductance. Moreover, the modified memristive device structure with the integration of 2D MoS₂ can pave the way for optoelectronic memories development that may also act as optoelectronic synapses in bionic visual applications. Hence, the profound advantage of the herein presented physical model lies in its ability to inform researchers about the potential functionalities of MoS₂ integrated Y₂O₃-based memristive devices and guide them in developing novel low computing power 2D-TMD/TMO-based artificial synapses for tuneable neuromorphic computing applications.