Synaptic Transistors Based On Charge Trapping Effect

As the basic unit for information processing in our brain,synapse keeps the whole system highly robust and fault-tolerant due to its plasticity.The research on electronic devices with synaptic plasticity is of great importance for the construction of artificial neural networks to achieve brain-like computing.Among various synaptic devices,low-dimensional semiconductor transistor based on charge trapping effect has great potential for research and application in future neuromorphic computing.Due to the good retention property of charge trapping effect,transistors can realize the ’inmemory computing’ function by maintaining the programmed state through nonvolatile charge storage while performing logic processing through gate voltage regulation.In addition,the excellent optical and electrical properties of low-dimensional semiconductors benefit the capture of external information,enabling the integration of perception,storage and computation functions in a single device.In this thesis,we adopted both organic and inorganic charge trapping layers to construct the synaptic transistors with charge trapping capability.The simulation of synaptic behavior was successfully realized by applying different gate voltage bias served as input signals.The excellent optical and electrical properties of two-dimensional and one-dimensional materials enable the device to emulate more complex synaptic behavior with the combination of optical and electrical stimulation.The main research content of this thesis is as follows.(1)Long-term plasticity plays a crucial role in the learning and memory function of human brain.It is crucial to evaluate the device performance of mimicking long-term plasticity.In this thesis,we introduced a polystyrene(PS)polymer electret film between the tungsten selenide(WSe2)nanosheet and SiO2 dielectric layer to build the WSe2 nanosheet synaptic transistor with charge trapping capability.Here,it was found that the charge storage capability of the device was greatly enhanced with 254 nm UV irradiation during the programming process.Simulation of excitatory and inhibitory synaptic behavior was realized by applying modulated gate voltage.Due to the excellent charge storage capability of PS polymer electret the programmed state could maintain for a long time,and the long-term plasticity was successfully simulated.Finally,the mechanism of light-assisted charge storage was utilized to emulate the advanced learning mechanism of the brain.Both the asymmetric and symmetric spike-timing dependent plasticity were stimulated by using the electrical signal and light signal as the pre-synaptic and post-synaptic signals,respectively.(2)Since 80%of the external environmental signals are captured and processed by human visual system,the simulation of the human visual system is of great significance for building artificial neural networks.If perception,storage,and computation functions can be integrated in a single device,the working efficiency of the whole system could be improved greatly.In this thesis,the perception,storage and computing functions are realized in a single transistor by combining the charge trapping effect with the persistent photoconductivity of zinc oxide(ZnO)nanowires.The irradiation of 365 nm UV light induced photogenerated carriers in the ZnO nanowire and also caused the desorption of oxygen originally adsorbed on the surface of the nanowire,which increases the channel current in the device and leading to the successful simulation of excitatory behavior.However,the opposite inhibitory behavior cannot be fulfilled by a pure optical signal.In this thesis,we successfully simulated the synaptic inhibitory behavior by applying positive gate voltage to push electrons into the charge trapping layer and reduce the channel current.Both the channel conductance increase caused by persistent photoconductivity of ZnO nanowire and the channel conductance decrease caused by the charge trapping effect have good retention abilities,thus enabling the simulation of synaptic long-term plasticity behavior such as spike-timing dependent plasticity.Finally,simulation of a neural network utilizing the consecutive reversible resistance change and nonvolatile characteristic of ZnO NW synapse was carried out.A three-layer artificial neural network constructed by MATLAB demonstrates a high recognition accuracy up to 92%after only 20 training epochs for recognizing the handwritten digits.In summary,based on the three-terminal field-effect transistor structure and charge trapping effect,in this thesis,we constructed a low-dimensional synaptic transistor with long-term plasticity,which provides a feasible strategy for building neuromorphic devices and neural networks.Besides,inspired by the human visual system,with the combination of charge trapping effect and persistent photoconductivity of low-dimensional nanowires,we further explored the potential of integrating perception,storage and computation functions in an individual transistor,which provides a new method for building artificial visual systems.