Figure 1. Schematic diagram of the biological and artificial multisensory neurons. A biological multisensory neuron that is stimulated by pressure and temperature. Pressure applied onto mechanoreceptors change the potentials of receptors that are embedded in the skin. Temperature applied onto thermal receptor change the receptor potential. The cell body of the sensory neuron integrates the potentials and initiates spikes with coded pressure information and temperature information. Inside the orange dashed box, it is the schematic images of the artificial multisensory neuron consist of a piezoresistive sensor and VO2-based oscillation neuron.
Here, we report an artificial multisensory neuron consisting of a piezoresistive sensor and a VO2 based volatile memristor connected in series. Such artificial sensory neurons can be used to sense different pressure inputs and convert them into spike trains as a result of the voltage dividing effect between the piezoresistive sensor and VO2 memristor. Besides, the spiking neuron is also capable of sensing temperature, by taking advantage of the intrinsic thermal sensitivity of metal-insulator transition in VO2. The spiking neuron is utilized to recognize Braille characters using multiple piezoresistive sensors. Notably, the traditionally separate haptic and temperature signals can now be fused physically in the VO2 based sensory neuron when synchronizing the two sensory cues, which is able to recognize multimodal haptic/temperature patterns. Such multisensory neurons could provide a promising approach towards e-skin, neuro-robotics and human-machine interaction technologies.
2. Results and Discussion
2.1. Oscillation neuron based on VO2 volatile memristor
The perception and cognition ability of human brain assisted with associative biomechanical and temperature sensations are critical for acquiring somatosensory information. The brain encloses numerous neurons to receive the interactive signals in different modalities (e.g., mechanical, temperature signals) and implements cross-modal neuromorphic computation in the multisensory association area.[33, 34] Figure 1 presents the biological multisensory integration nervous system, and the corresponding artificial multisensory system that is constructed (as shown in the orange dashed box), which consists of a piezoresistive sensor and VO2-based oscillation neuron.
As one of the key components, the oscillation neuron was first built, and its characteristics were thoroughly analyzed. As schematically illustrated in Figure 2 a, the oscillation neuron consists of two Au/Ti electrodes sandwiching a VO2 film in a lateral device structure. The structure of the device is characterized by the scanning electron microscopy (SEM) image in Figure 2b, showing that the channel length of the device is approximately 400 nm. Figure 2c and 2d show transmission electron microscopy (TEM) images of the VO2 device, with spatial mapping of Al, Au, Ti, V and O elements using energy-dispersive X-ray spectroscopy (EDS). An EDS line scan of the cell is shown in Figure S1, Supporting Information. Figure 2e exhibits a high-resolution TEM image of the VO2layer. The clear lattice fringes and the corresponding Fast Fourier Transformation (FFT) result (Figure 2f) show that the VO2 has high-quality crystalline structure with a tetragonal phase.
Electrical measurements of the Au/Ti/VO2/Ti/Au memristors show that the devices have threshold switching (TS) characteristics without going through any electroforming process. As shown in Figure 2g, reliable TS characteristics can be obtained under voltage sweeping mode with 100 cycles. Specifically, when the applied voltage exceeds a threshold voltage (V th) of ~1.4 V, the VO2 device switches from high resistance state (HRS) to low resistance state (LRS), and automatically returns to HRS once the applied voltage drops below a holding voltage (V hold) of ~1.0 V (Figure 2g, Figure S3 further shows the stable TS characteristics without compliance current). Transient electrical measurements show that the switching speed of the VO2 device is <120 ns from off state to on state, and <50 ns from on state to off state (Figure S2, Supporting Information). We also examined the stability and uniformity of the VO2 memristor, including the endurance, the cycle-to-cycle and device-to-device variations of the device, which demonstrate that the VO2 memristors have endurance of >106 cycles as well as acceptable cycle-to-cycle and device-to-device variations (Figure S4-S6, Supporting Information), making them qualified for functioning as oscillatory neurons.
The TS characteristics in VO2 volatile memristors can be well interpreted by the Mott transition coupled with a structural phase transition.[35-37] Figure 2h schematically depicts the dynamic evolution of the device state during the threshold switching process through COMSOL simulation. The orange line shows the experimental current-voltage (I-V ) characteristics, while the bule line depicts the simulation curve. From state (1) to state (2), heat is generated in VO2 while the applied voltage increases. When the applied voltage exceeds V th, joule heating generated by the voltage induces formation of a filament through the VO2 gap, leading to a transition from HRS to LRS, and the filament expands from the channel to both sides as the process progresses (state (2) to state (4)). When the applied voltage decreases, the heat is gradually dissipated and the size of the filament gradually decreases, and the device undergoes a transition from state (5) to state (8), leading to switching from LRS to HRS.