Preparation And Performance Study Of InZnO Electrolyte Gate-controlled And UV-light-controlled Synaptic Transistor

Artificial intelligence,represented by deep neural networks,has profoundly changed human society in the past decade.However,human beings are still in the era of weak artificial intelligence,and how to move toward strong artificial intelligence relies on software-level algorithm enhancement and brain-inspired hardware improvement.The traditional CMOS logic circuit computer system based on the von Neumann architecture is limited by the process technology which has been reduced to atomic size and Moore’s law which is ending,the data processing and storage are separated from each other,and the storage speed is much lower than the computation speed,so there is an urgent need to break the bottleneck of "storage wall" for data transmission.By learning from biologists and neuroscientists,we can study how the brain processes data,and then develop brain-like chips to solve the key problem of increasing arithmetic power.Multi-terminal artificial synapses,as a single component of AI hardware,have great potential to integrate AI systems to meet the demand for high-performance,low-latency,low-power computing power in the era of big data.In this dissertation,artificial synaptic transistor devices were prepared using indium zinc oxide(InZnO)nanowires as semiconductors,combined with polymer electrolytes and UV lasers,respectively,to perform simulations of various biological synaptic plasticity and to complete simulations of handwritten digit recognition based on artificial neural networks,as follows.1.InZnO nanowire arrays of electrolyte-gated synaptic transistors(EGTs)have been prepared by a simple and easy electrostatic spinning technique combined with a nanowire transfer method.The nanowires are as low as 40-80 nm in diameter,providing a physical basis for further tight integration of single components.A series of simulations of biological synaptic plasticity such as excitatory postsynaptic currents,double pulse fugacity and long-range plasticity are performed.Based on the enhancement/inhibition properties of the experimental data,the nanowire array devices obtained up to 93.1% recognition accuracy by simulating the experimental results with the MNIST handwriting recognition dataset and artificial neural networks.This study demonstrates that a cost-controllable and efficient fabrication technique to obtain wellaligned nanowire arrays enables artificial synapses with low power and high accuracy.2.UV photo-controlled synaptic transistors with InZnO nanowires were prepared by electrostatic spinning process,and the devices processed and stored information in the form of current under UV laser pulse irradiation at 375 nm wavelength,and successfully simulated biological synaptic plasticity based on the inherent continuous photoconductivity effect of metal oxides to realize the transition from short-term to long-term memory,and obtained high PPF index(276%)and long LTP retention time(>7200 s).With the synergistic effect of photoelectricity,the PD line based on optical enhancement and electrical inhibition was simulated by Cross Sim artificial neural network to obtain up to 97.1% accuracy of handwritten digit recognition.The Ebbinghaus memory forgetting curve was successfully simulated and showed the potential of image memory application.