Research Of Synaptic Plasticity And Application For MoS₂ Floating-Gate Transistor

With the development of brain neuroscience,the field of neuromorphic computing has attracted widespread attention in the academic community.Neuromorphic computing has great potential applications in various research fields,such as image recognition,cancer diagnosis,and autonomous driving.In order to further improve computing efficiency,research of low-cost,high-speed,low-power electronic devices is needed.In this thesis,the synaptic properties of MoS₂ channel-based floating-gate transistor devices are investigated.And the parameters of the device's synaptic properties are substituted into a supervised neural network to simulate and implement the function of identifying handwritten digital data sets.The research contents and results of this thesis are as follows:(1)MoS₂ channel-based floating-gate transistor devices are fabricated and characterized.The three terminals of the device are the source electrode,drain electrode,and back gate electrode.MoS₂ is used as the channel material.BN is used as the tunneling layer material.And MoS₂ or Au nanoparticles is used as the floating gate layer material.Also,some commonly used device characterization methods are used to characterize the floating gate transistor devices.(2)We Explore and optimize the electrical characteristics of MoS₂ floating gate transistor devices.In this thesis,we explore the device's back-gate transfer characteristic curve and impulse response characteristic curve,and optimize the pulse response of the device by adjusting the amplitude of the pulse or using Au nanoparticles as the floating gate layer material.By designing the same triangular pulse waveforms,the device with floating gate material made of Au nanoparticles realize the spike-timing dependent plasticity function of biological exponential decay.Also,the holding characteristic under different channel conductance state of the device is explored.(3)We build a deep convolutional neural network based on MoS₂ floating gate transistor to realize the task of handwritten digit data set recognition.This thesis compares the non-ideal characteristics of devices with two kinds of different floating gate layer material.Considering the non-ideal characteristics of the device,when put the synaptic performance parameters of the device with the floating gate layer material MoS₂ into the convolutional neural network,the recognition rate of can reach 91.1%.When put the synaptic performance parameters of the device with the floating gate layer material Au nanoparticle into the convolutional neural network,the recognition rate can reach 94.7%.