Figure 3. Oscillation characteristics based on VO2 volatile memristor. a) Schematic diagram of the oscillation neuron. b) Oscillation frequency as a function of load resistance (R L) with different input voltages (4, 5, 6 V). c) Oscillation frequency as a function of input voltage (V in) with different load resistors (3.0, 3.5, 4.0, 4.5 kΩ). d) Neuronal response of the oscillation neuron under a constant bias voltage (5 V, 20 μs) and the influence of varying resistance (R L). The output frequencies are 0.9, 0.7, 0.55, and 0.35 MHz at R L of 3.0, 3.6, 4.2, 4.8 kΩ, respectively. e) Neuronal response of the oscillation neuron in series with a fixed resistance (R L=4 kΩ) and the influence of varying input voltage (V in). The output frequencies are 0.45, 0.65, 0.8, and 0.9 MHz atV in of 4.2, 4.8, 5.4, and 6.0 V, respectively.
2.2. Artificial haptic perception neuron based on VO2volatile memristor
In a biological sensory nervous system, the mechanoreceptors are responsible for sensory information transduction. When the external stimuli exceed the mechanical threshold of the mechanoreceptors, sensory information is being coded through an action potential at a certain frequency.[38, 39] To mimic the biological activity from mechanoreceptors, the spiking response due to a haptic event, we integrated the piezoresistive sensor with a VO2 memristor to emulate the haptic perception as an artificia mechanoreceptor. Given the dependence of the output frequency of VO2 oscillation neuron on the load resistor (Figure 2b,d), the haptic perception function can be achieved by replacing the fixed R L with a piezoresistive sensor. Figure 4a shows a schematic diagram of the artificial haptic perception neuron using VO2 volatile memristor and a commercially available piezoresistive sensor (the entire experimental setup is shown in Figure S7, Supporting information). Figure 4b shows the I–V curves of the piezoresistive sensor under different pressures, showing different resistance state in response to pressure/weight inputs (from 20 to 700 g). Moreover, Figure 4c summarizes the resistance response of the piezoresistive sensor under different pressures/weights. The stability and thermal characteristics of the piezoresistive sensor are further shown in Figures S8 and S9, Supporting Information. It can be clearly seen that the resistance gradually decreases as the pressure increases. This characteristic can be generalized by a power function:
(3)
where α = 95570, β = -0.64 are extracted by fitting the curve presented in Figure 4c. The range of resistance change under pressure is between ~1 kΩ and ~20 kΩ, which meets the series resistance required by the VO2 oscillation neuron. This piezoresistive sensor can thus be combined with the VO2oscillation neuron to emulate artificial haptic perception, where the sensor is used as a receiver of pressure signals and the resultant sensory signal is converted into spike trains by the VO2neuron.
Indeed, when different pressures/weights (100, 150, 200, 250 g) are applied to the sensor with a constant bias voltage (5 V, 20 μs), the oscillation neuron exhibits different output spike frequencies (0.45, 0.55, 0.75, 0.95 MHz), as shown in Figure 4d. The converted spike frequency increases as the pressure increases. In this way, the artificial haptic sensory neuron can directly respond to pressure signals and encode them into spikes, and the output spikes can then transmit information to spiking neural networks for further processing.
Haptic perception allows human to recognize objects, discriminate texture, and react appropriately in a social exchange.[2, 40] This essentially requires integration of multiple spatial correlated sensory stimuli. In order to achieve this, we combine two sensors in parallel as a proof of concept, and they are further connected in series with a VO2memristor (the circuit structure is shown in Figure S10, Supporting Information). This is utilized to recognize Braille characters in the present study. As shown in Figure 4e, the black circles represent convex patterns in the Braille characters, while the white circles indicate no convex, which correspond to the cases of sensing pressure and no pressure when being touched, respectively. These scenarios were emulated in experiment by applying 100 g for convex patterns and 0 g otherwise. The results in Figure 4e show that when only one of the two sensors are triggered, the output oscillation frequency of VO2 neuron is ~0.4 MHz. However, when the two sensors are triggered at the same time, the output oscillation frequency will be higher (~1.1 MHz). The output frequency is zero when neither of the sensors is triggered (detailed information is shown in Supplementary Video 1-3). Therefore, the Braille characters can be read out from the different patterns of output frequencies produced by the VO2 neurons. It should be pointed out that some Braille characters may not be fully distinguished based on the horizontal inputs, and in this case an additional process can be introduced to apply vertical inputs onto the device so as to further distinguish them.