Study On Solid-State Three-Terminal Artificial Synaptic Transistor Devices

The era of Big Data has brought convenience to our lives.However,the computer based on the von Neumann architecture is constantly challenged by the massive data.Due to the separation between the central processing unit（CPU）and the memory,the traditional computer needs to transfer data frequently when processing information.The high energy consumption and delay caused by the separation between the memory and CPU seriously restrict the efficiency of information processing.It is urgent to build a new computer architecture to deal with the massive data.The human brain is a high parallel,low energy consumption and high efficiency intelligent computer.There are 10¹⁵ synapses in human brain,which process and store the information obtained by human beings.Constructing brain-like computer system to process information with high efficiency and low energy consumption,has been an important way to solve the bottleneck of traditional computer.The artificial synaptic device is the core component of the brain-like computer system.There are two main types of synaptic devices:two-terminal synaptic device and three-terminal synaptic device.Memristor is a typical two-terminal synaptic device,which is regulated by the conductive filaments（ion or vacancy channels）to simulate the functions of biological synapse under the external electric field.Two-terminal synaptic device has simple structure and can be integrated easily.However,the read and write processes cannot be operated at the same time in this kind device.In addition,the SET and RESET voltages have a wide distribution,which leads a low accuracy of image recognition.Electrolyte gated electrochemical transistor is a typical three-terminal synaptic device,which is regulated by the ion migration to simulate the functions of biological synapse.The read and writeprocesses can be carried out at the same time in this synaptic transistor.And it has a better linearity and stability during the weight update process,which leads to a higher image recognition accuracy.In addition,the three-terminal synaptic device has similar structure and dynamic process with the biological synapse.Therefore,the electrolyte gated synaptic transistors are studied in this doctoral dissertation.The transition metal oxide Mott insulator VO₂,quasi-two-dimensional α-MoO₃ and MoS₂ are adopted as the channel layer of the electrolyte gated synaptic transistors.These three kind channel layers are prepared by magnetron sputtering,which has low cost,and compatibility with semiconductor industry.The main research works of this doctoral dissertation are briefly introduced in the following:（1）A solid-state flexible Mott synaptic transistor based on H⁺modulation has been designed and studied.In this work,high quality VO₂ film is prepared on flexible polyimide substrate by magnetron sputtering and rapid thermal annealing.The resistivity of the VO₂ film can change about 4 orders of magnitude during the metal-insulator transition.A solid-state flexible synaptic transistor based on this VO₂ film is fabricated.And the metal-insulator transition of the VO₂ film is modulated by H⁺insertion.A variety of biological synaptic functions are mimicked,including STP,LTP/LTD and SNDP.In particular,the weight updating of the flexible synaptic device has a good linearity and stability.Finally,the MINST dataset and the CrossSim simulator are used for image recognition simulation.The results show that the synaptic transistor has very high image recognition accuracy of 98%and 92%in flat state and bending state,respectively.（2）A solid-state α-MoO₃ synaptic transistor based on H⁺modulation has been fabricated and characterized.The traditional method to obtain high quality two-dimensional layeredα-MoO₃ ultra-thin film is mechanical exfoliation.However,considering the small area and irregular shape of the two-dimensional layered film prepared by mechanical exfoliation,it is hard to realize large scale production for practical applications.In this work,high qualityα-MoO₃ ultra-thin film is prepared on SrTiO₃（100）substrate by magnetron sputtering.A solid-state synaptic transistor based on the α-MoO₃ film is fabricated.The resistance ofα-MoO₃ channel layer changes nearly 4 orders of magnitude by electric gating induced H⁺insertion.A variety of biological synaptic functions are simulated,and the CrossSim simulator is used for image recognition simulation.The result shows that the α-MoO₃ synaptic device has an image recognition accuracy of 88%for MNIST dataset.（3）A solid-state MoS₂ synaptic transistor based on Li⁺modulation has been prepared and studied.The two-dimensional layered MoS₂ has been studied extensively in electronic and photoelectric devices due to its excellent electrical and optical properties.In most of the reported works,MoS₂ ultra-thin film is usually obtained by mechanical exfoliation,which is unsuitable for practical applications due to its small area and irregular shape.In this work,large area MoS₂ films with thickness of 10 nm and 20 nm are prepared on sapphire Al₂O₃（0001）substrate by magnetron sputtering.Solid-state synaptic transistors based on the MoS₂ films are fabricated.Two synaptic devices with different thickness of MoS₂ channel layers both show multiple synaptic functions.And the performance of these two MoS₂ synaptic devices are compared in details.The simulation results show that the two MoS₂ synaptic devices have recognition accuracy of 97%and 95%for MNIST dataset,respectively.