Carbon-based Neural Network Based On Single-electron Transistor And Memristor And Its Application In The Design Of Brain-like Visual Perception Chips

With the continuous advancement of artificial intelligence research and semiconductor chip technology,more and more scientists pay attention to the research of brain like computing chip,in order to establish a new computing paradigm inspired by brain,different from the existing von Neumann computing architecture,integrating information storage and computing.The establishment of brain-like neural network visual perception circuit system based on single-electron transistor and memristor is of great significance for the development of brain-like carbon chip with high integration and low power consumption.In this paper,a carbon-based neural network circuit with single-electron transistor and memristor and the key technologies involved in its application in the design of brain-like visual perception chip are introduced.Firstly,the sensing circuit design of photoelectric vision system based on graphene field effect transistor is introduced and verified by simulation;Secondly,for ultra-low-power single-electron transistor neuromorphic devices and their neuron circuits,the Verilog-A language is used to model the devices in the Cadence ADE environment and design the circuit simulation verification;Then,the memristor with graphene quantum dot structure is modeled and the corresponding synaptic circuit is designed to verify its STDP plasticity mechanism and memristor learning algorithm;Finally,according to the application in the design of brain-like visual perception chip proposed in this paper,a circuit system integrating visual information acquisition and preprocessing,memristor learning and neural network associative memory is established,and the visual application scene function verification based on handwritten digit recognition is completed under the cosimulation of Cadence and MATLAB.The main work of this paper is reflected in:1.The graphene field effect transistor device model that can work under the Cadence Spectre simulator is established,and the corresponding photoelectric vision sensor circuit is designed to realize the simulation of the graphene vision system circuit.2.Established a single-electron transistor device model and designed a singleelectron spiking neuron circuit,which reduces the redundancy of CMOS neuron devices and improves the integration of neuromorphic devices and their neuron circuits.3.For the quantum dot memristor,1T1R synapse circuit,crossbar synapse circuit and bridge synapse circuit are designed,and the synapse learning process is realized,which meets the requirements of brain-like computing in memory.In addition,in brain-like visual perception application scenarios,the improved Hebbian learning algorithm greatly increases the storage capacity of the associative memory neural network and achieves a high recognition rate of handwritten digits.