Synaptic plasticity and learning behaviour in multilevel memristive devices

Abstract

This research explores a novel two-terminal heterostructure of the Pt/Cu₂Se/Sb₂Se₃/FTO memristor, which exhibited essential biological synaptic functions. These synaptic functions play a critical role in emulating biological neural systems and overcoming the limitations of traditional computing architectures. By repeating a fixed pulse train, in this study, we realized a few crucial neural functions toward weight modulation, such as nonlinear conductance changes and potentiation/depression characteristics, which aid the transition of short-term memory to long-term memory. However, we also employed multilevel switching, which provides easily accessible multilevel (4-states, 2-bit) states, for high-density data storage capability along with endurance (10² pulse cycles for each state) in our proposed device. In terms of synaptic plasticity, the device performed well by controlling the pulse voltage and pulse width during excitatory post-synaptic current (EPSC) measurements. The spike-time-dependent plasticity (STDP) highlights their outstanding functional properties, indicating that the device can be used in artificial biological synapse applications. The artificial neural network with Pt/Cu₂Se/Sb₂Se₃/FTO achieved a significant accuracy of 73% in the simulated Modified National Institute of Standards and Technology database (MNIST) pattern. The conduction mechanism of resistive switching and the artificial synaptic phenomena could be attributed to the transfer of Se^2- ions and selenium vacancies. The neuromorphic characteristics of the Pt/Cu₂Se/Sb₂Se₃/FTO devices demonstrate their potential as futuristic synaptic devices.

Fig. 1. (a) The human biological neural system consists of pre-synaptic neurons and post-synaptic neurons interconnected through synapses, as clearly depicted in the magnified image. (b) The schematic of the electronic multilayer synapse structure (Pt/Cu₂Se/Sb₂Se₃/FTO) that works like a human biological synapse.

Fig. 2. (a) The elemental composition of Cu₂Se and Sb₂Se₃ by using EDS. (b) Cross-sectional SEM image of the multilayer structure of Cu₂Se and Sb₂Se₃. In the inset, the magnified images of the Cu₂Se and Sb₂Se₃ layers can be distinguished easily. (c) The top-view SEM image of the multilayer structure. The inset image shows that the film was deposited uniformly. (d) The AFM image used to calculate the surface roughness.

Fig. 3. Electrical characteristics: (a) the I–V characteristics of the Pt/Cu₂Se/Sb₂Se₃/FTO device, indicating analog behaviour during the continuous cycles of SET and RESET. (b) The gradual increase and decrease in conductance with the application of repeated fixed voltage sweeps to SET and RESET the device. In the inset, a magnified area under the red circle shows the change in conductance during RESET. (c) The endurance test of the Pt/Cu₂Se/Sb₂Se₃/FTO device was conducted under DC conditions. The inset shows the initial 10³ pulse cycle, indicating that the device reached stability after 600 cycle. The inset of figure (c) presents the stability of the HRS and LRS states of the device when the cycles were further extended by up to 10³ cycle under the same conditions. (d) The probability distribution of HRS and LRS evidences that the LRS state is more stable after 600 cycle, but HRS fluctuates. In the inset, the error bar chart is used to check the dispersion of the LRS and HRS states.

Fig. 4. (a) The endurance of the multilevel switching characteristics of the Pt/Cu₂Se/Sb₂Se₃/FTO device during SET operation. (b) The possible combinations of the multi-bit resistance states (00, 01, 10, and 11) and each resistance state measured for 100 pulse cycles. A total of 400 pulse cycle were run to achieve multiple states. (c) The applied pulse patterns to achieve multilevel switching of the device.

Fig. 5. (a) The potentiation and depression characteristics of the Pt/Cu₂Se/Sb₂Se₃/FTO device. (b) The applied 200 SET train pulses (−1.5 V, 100 ms) demonstrate potentiation. (c) The applied 200 RESET train pulses (1.5 V, 100 ms) demonstrate depression characteristics. A total of 2000 pulse cycles were applied to demonstrate the reproducibility of potentiation and depression. (d) EPSC decay after the application of a single short pulse with varying pulse amplitudes (−2 V, −2.5 V and −3 V) and a fixed 100 ms pulse width. Inset image: the pattern of pulses with different amplitudes and a fixed pulse width. (e) The EPSC decay rate after the application of variable pulse widths (100 ms, 250 ms and 500 ms) and a fixed −3 V pulse amplitude. Inset image: the pattern of pulses with different pulse widths and a fixed pulse amplitude.

Fig. 6. In quadrants (II) and (IV), the synaptic weight change (Δw%) STDP is presented as a function of the relative time change (Δt) between the pre-and post-synaptic paired pulse propagation. In quadrants (I) and (III), schematic representations of the pre-synaptic and post-synaptic applied paired pulses are given.

Fig. 7. The MNIST pattern recognition simulation of the Pt/Cu₂Se/Sb₂Se₃/FTO devices. (a) The conductance plot was obtained in the LTP and LTD operations. (b) The calculated normalized conductance of the devices with curve fitting. (c) The depiction of the employed synapse neural network comprising an input layer of 400 neurons, a hidden layer of 100 neurons and an output layer of 10 neurons for MNIST pattern recognition. (d) The simulated accuracy curves of the ideal and Pt/Cu₂Se/Sb₂Se₃/FTO devices on the MNIST dataset.

Fig. 8. A schematic representation of the internal mechanism: (a) in the absence of an external power supply, no movement of ions and vacancies happens. (b) When the negative and positive terminals are connected respectively to the top and bottom electrodes, the migration of V_Se and Se²⁻ starts in the upper cell (Cu₂Se) and lower cell (Sb₂Se₃). The migration of V_Se forms the conducting filament during SET operation in both cells. (c) When the negative and positive terminals are connected respectively to the bottom and top electrodes, V_Se recombines with Se²⁻ ions in both cells during RESET operation.