Electric-double-layer Coupled Oxide-based Neuromorphic Transistors Studies

Human brain has a massively parallel and reconfigurable architecture with a complex network of ¹011 neurons and ¹015 synapses.Such architecture makes possible more robust,plastic and fault-tolerant learning/memory functions than any current digital computer.In the neural network,the neuron that receives spike inputs from thousands of synapses distributed across dendritic trees is often considered to be the computational engine of the brain.Synaptic plasticity is the biological process by which specific patterns of synaptic activity result in changes in synaptic strength and is thought to contribute to learning and memory.Therefore,emulation of essential synaptic plasticity and computation are viewed as a key step toward neuromorphic computing.Recently,synaptic electronics which is aimed at capturing synaptic plasticity and computing power by single electronic device has aroused widely interesting.At the outset,two-terminal devices such as memristors,phase change memory,and atom switches,etc,have been explored as the building blocks of neuromorphic systems.More recently,three-terminal neuromorphic devices such as ionic/electronic hybrid transistors and ferroelectric transistors have been demonstrated in the pursuit of the synaptic plasticity and computation in a single device.The reported three terminal neuromorphic devices reveal alternative potentials in acting more than a weight tunable connection,which can perform signal processing/computing serving as synaptic filter,integrator,etc,in neuromorphic circuits.The ion-coupled oxide-based electric-double-layer?EDL?transistors are intrinsically equivalent to the ionic/electronic hybrid transistors,which could be applied to tune the conducting characteristics of the semiconducting channel in short-term and long-term,respectively,due to the ion/electron-correlated electrostatic coupling and electrochemical processes.On this basis,this thesis focused on the essential synaptic plasticity emulations and neuromorphic computing applications by the ion-coupled oxide-based EDL transistors.The main content can be summarized from the following aspects:?1?Preparation and electrical performance research of electrolytes.In this thesis,nanogranular SiO₂,methylcellulose,and graphene oxide were successively prepared.All of the electrolytes exhibit perfect insulativity.For example the maximum leakage current for the nanogranular SiO₂ is as low as 0.6 nA.What‘s more,all of the electrolytes are good proton conducting films.The highest proton conductivity of 4.2×10-4 S/cm was obtained from methylcellulose films.These results indicate the electrolytes are perfect platform for formation of EDL.Huge EDL capacitances >1 mF/cm2 were observed and the highest capacitance was measured to 18 mF/cm2 in graphene oxide.?2?Fabrication and performance characterization of oxide-based EDL transistors.Oxide-based EDL transistors were fabricated by using the electrolytes mentioned in?1?as the gate dielectric.Good transistor performances were achieved with field-effect mobility higher than 20 cm2V-1s-1.The on-off ratio and subthreshold slope of the oxide-based EDL transistors gated by nanogranular SiO₂ are 2×107 and 114 mV/decade,respectively.The fabrication processes of the oxide-based EDL transistors gated by Nanogranular SiO₂ are compatible with the CMOS technologies,which indicate that such transistors can have the great potentials for building neuromorphic circuits/chips.The transistors gated by graphene oxide or methylcellulose exhibit good transistor performance.Such transistors deposited on flexible substrates are also demonstrated with perfect mechanical flexibility.No appreciable degeneration in performance can be observed in the flexible graphene oxide gated oxide-based EDL transistors even after bending test for thousands times.In that case,such transistors could potential applied for large scale flexible neuromorphic circuits.?3?The short-term behaviors of synapse were successfully mimicked.These short-term behaviors are excitatory postsynaptic current?EPSC?,paired-pulse facilitation?PPF?,spatiotemporal correlated dynamic logic,and short-term memory.What‘s more,a theoretical model for short-term synaptic behaviors emulations based on EDL modulation and stretched-exponential decay function was proposed.Such model is consistent well with the experimental results.As the theoretical model is not limited to the EDL transistors mentioned before,our results will provide a useful guideline for short-term emulations of other ion-coupled EDL transistors.?4?The long-term behaviors of synapse were successfully mimicked.The electrochemical doping/dedoping processes between protons in electrolyte and electrons in semiconducting channel can be observed under a high gate voltage?|V|?4.0?.The XPS measurements indicate that the oxygen vacancies in IZO were increased by the electrochemical doping process,which results in a long-term increase in channel conductance.On the contrary,the electrochemical dedoping process would result in a long-term decrease in channel conductance.On this basis,the long-term behaviors such as spike-timing-dependent plasticity?STDP?,long-term memory,and classical conditioning,were successfully mimicked.?5?Dendrite related functions were successfully realized.The huge EDL capacitance is formed at the interface between electrolyte and channel function as a nanoscale capacitor.Almost all the gate voltage is applied on such capacitor with nearly no potential difference across the electrolyte.Therefore,multiple gate inputs could paralleled coupled to channel.Nonlinear dendritic integration functions were realized,which is very similar to the biological experiments.What‘s more,the neural arithmetic behaviors were successfully mimicked by using additional gate electrode as the modulatory input.?6?Visual processing functions were successfully realized.In analogy to the structure of visual neural network,an artificial visual system was built based on oxide-based EDL transistors.Such visual system shows selectivity in response to edge orientations around zero degree.The orientation tuning function can be realized with a full width at half maximum of 44.5°.Then the collision avoidance behaviors of LGMD neuron were successfully realized based on an artificial visual system with 20×20 photodetectors and multiple-gate oxide-based EDL transistors.