Construction Of Memristive Device Based On WO_x Materials And Study On Synaptic Emulation

As an emerging electronic component,memristor has attracted researchers'much attention in recent years as potential candidate for the construction of neuromorphic networks.In terms of transmission mechanism,memristor's resistance can be tuned continuously by the applied voltage or current,showing high degree of similarity to a biological synapse.In terms of morphology and structure,memristor has great potential application in the field of synaptic stimulation due to its nanoscale cell size and easy integration.However,the development of synaptic stimulation used by memristor is still at primary stage,and it's restrained by limited physical model and unclear physical mechanism.In addition,it's necessary to construct multifunctional synaptic devices to simulate more complex and diverse synaptic learning functions.In this thesis,we developed a new memristive model,and fabricated the memristor based on WOx materials.The device's physical mechanism,synaptic simulations,and flexible applications have been further studied.The details are as follows:Memristive model and devcice:a depletion-region-width modulation memristive model has been proposed in our work:under the external electric field,the oxygen ions can migrate from films to electrode,which could change the concentration of oxygen vacancy of oxide's surface,modulate the width of deleption region,and further change the device resistance.Based on the model,a stable memristive device was fabricated by using amorphous WOx materials.We not only obtained continuously tunable memristive behavior,but also realized the memory of capacitance.Further,it is proved that the memristive behavior is derived from the depletion-region-width modulation through two methods:building devices with different contact type between electrode and films and sputting the films by Ar ions.The temperature dependent second order dynamic process confirms that the physical mechanism of the resistance change is the migration and diffusion of oxygen ions.Simulation of synaptic learning:based on the memristor,several essential synaptic learning functions can be achieved,including nonlinear transmission characteristic,learning experience,and long-term/short-term plasticity.Further,the more complex learning behaviors has been obtained:including non-associative and associative learning.Based on the rate-dependent plasticity of our second-order memristor,we degsined the stimulus with sequential superposition.We not only obtained associative learning and forgetting behaviors,but also reproduced stimulus-interval-dependent physiological processes.The realization of these synaptic learning functions corroborate that our memory device possesses abilities to stimulate the synapse,and offer a reliable choice for the future construction of memristive neural network.Application on pattern recognition:in our work,we achieved the pattern recognition using the analog memristor.A face image with high accuracy can be recognized,and the error with inout one is less than 5%.Further,the coexistence of analog and digital resistive switching was developed in our WO_x-based memristor.We compared their differences in the application of pattern recognition:the pattern learning with high accuracy was achieved in analog memristor because the change of resistance state is continuous and the fluctuation is small;the pattern learning with high efficiency was achieved in digital memristor because the change of resistance state is discrete and the fluctuation is large.The results show that different learning behaviors are obtained in the same device,which lays the foundation for the realization of the diversity of synaptic learning.Application on flexible devices:we demonstrated a new transfer method suit for the memristor devices based on inorganic materials.In our work,the flexible and transferable memristor based on amorphous WO_x film was fabricated using a simple water-dissolution method,in which NaCl supporting substrate was employed.The device transferred on flexible substrates possesses similar memristive characteristic with original device even undergoing bending with big algle and multiple number.Further,the devices can be transferred on diverse substates with conformal contact using this method,showing no degeneration of ability to imitate synapse.Our results lay a foundation for the future application of neuromorphic computers in flexible and pastable fields.