Figure 2. VO2 volatile memristor analysis. a) Schematic diagram of the memristive device, which consists of two Au/Ti electrodes sandwiching a dielectric film (VO2) in a lateral device structure. b) Scanning electron microscopy (SEM) image of the VO2 device. c) Cross-sectional transmission electron microscopy (TEM) image and the elemental mapping of the materials in the system for Al, Au, Ti, V and O, respectively. d) Cross-sectional TEM image of the VO2 device. e) High-resolution TEM image of the VO2 layer. f) The diffraction pattern extracted by Fourier transform of Figure 2e. g) Current-voltage (I-V ) characteristics of the device repeated for 100 cycles. h) SimulatedI–V curves from the thermodynamic simulation. The right area of figure show heat distribution at each moment.
Based on the threshold switching characteristics in VO2memristors, oscillatory neuronal behavior can be implemented with a simple circuit, whose configuration is depicted in Figure 3 a. The VO2 volatile memristor is connected in series with a load resistor (R L), and the intrinsic capacitance of the VO2 memristor provides the dynamics for integration. Figure 3d shows the oscillation characteristics of the spiking neuron. When a suitable voltage (V in) is applied and a matched serial resistor (R L) is connected, the voltage dividing effect between the R L and VO2 memristor results in a voltage drop over VO2 memristor that exceeds its V th. Therefore, the VO2 device will switch from off to on state. Once the device is in on state, the voltage drops acrossR L and the VO2 device will be re-distributed, and the voltage gets lower thanV hold and hence the device returns to off state. To achieve such oscillation effect, the following requirements should be satisfied:
(1)
(2)
When Equations (1) and (2) are satisfied, the above spike event will be produced and repeated (Figure 3d). Figure 3d and 3e show the oscillating behavior of the spiking neuron, when the R L or input voltage is varied, respectively. As expected, the oscillation frequency decreases (0.9, 0.7, 0.55, 0.35 MHz) asR L increases (3.0, 3.6, 4.2, 4.8 kΩ) (Figure 3d), whereas the rate increases (0.45, 0.65, 0.8, 0.9 MHz) as the input voltage increases (4.2, 4.8, 5.4, 6.0 V) (Figure 3e). The dependence of the oscillation frequency on the load resistor (R L) is systematically tested under different input voltages (4, 5, 6 V), and the results are summarized in Figure 3b, further demonstrating that the oscillation frequency decreases asR L increases in each V incase. Similarly, Figure 3c systematically analyzes the dependence of the oscillation frequency on V in, when variedR L (3.0, 3.5, 4.0, 4.5 kΩ) is adopted, implying the same increasing trend of oscillation frequency with increase ofV in. Therefore, the oscillation frequency of the spiking neuron can be effectively modulated by the threshold voltage of the VO2 memristor and circuit parameters (R L and V in), which paves the way for the construction of an artificial sensory system.