Electric-Double-Layer Transistors For Neuromorphic Devices And Systems

The conventional computing paradigm based on CMOS logic and the von Neumann architecture have become less efficient when the amount of information keeps exploding.Compared to the von Neumann architecture,the human brain has the high density of-1015 neuron-synapse connections,which enables the parallel signal processing with learning/memory functions for lower power consumption.So,the brain inspired neuromorphic computing have become an exciting research area and have attracted considerable interest over the last few years.Hardware implementation of synapse/neuron by individual device is of great significance for realizing such neuromorphic computation.Several new-concept devices have been demonstrated to implement an artificial synapse in a single device,such as atomic switch and memristors,etc.Particularly,electric-double-layer(EDL)transistor is a promising candidate.It can emulate single-input and single-output synaptic transmission,because the ion dynamic process in the EDL dielectric can result in a synapse-like behaviors in the channel layer.It can also emulate multi-input neuron-like computations within a multi-gate structure.Aiming at researches on EDL transistors for the neuromorphic applications,the following works are carried out.1.Indium-zinc-oxide(IZO)based EDL transistors gated by proton conducting nanogranular SiO2 electrolyte are fabricated at room temperature.The devices are built with an in-plane-gate structure,and could be easily extended to a multi-gate structure.The nanogranular SiO2 film shows a high proton conductivity of 1.51×10-4 S/cm and a high capacitance of 3.45 ?F/cm2.Based on this huge EDL capacitive coupling between gate and the channel,the IZO-based transistors exhibits low operation voltage of 1.6 V,large current on/off ratio of 4×106 and a high field-effect mobility of 35 cm2/Vs.Such EDL transistors could emulate biological synaptic behaviors and functions,including excitatory postsynaptic current(EPSC)and paired-pulse facilitation(PPF).Furthermore,dendritic integration functions could be successfully mimicked in a dual-gate device.2.For the physical understanding and neuromorphic design of EDL transistors,an InGaZnO4-based EDL transistors were numerically simulated by TCAD tools.An ion drift-diffusion model is used to describe the capacitive coupling characteristic of the solid electrolyte dielectric.By employing this model,the simulation could well reproduce the transfer characteristic of the EDL transistor,with low operation voltage and large current on/off ratio,which suggests our simulation method is reasonable.Important synaptic behaviors,such as excitatory postsynaptic current(EPSC)and paired-pulse facilitation(PPF),are mimicked by the transient solution.By modifying this EDL transistors to a two-terminal device,the synaptic short-term facilitation and depression could be realized.Furthermore,high-pass and low-pass filtering characteristics are also successfully emulated not only for fixed-rate spike train but also for Poisson-like spike train.3.A behavioral model of an ion conductor-gated EDL transistor is developed by integrating charge accumulation/relaxation processes and the classical field-effect transistor characteristics.With this device model,the simulation can reproduce both DC behaviors and the dynamic synaptic functions,which suggests such behavioral model is reasonable for synaptic/neuromorphic simulations.The synaptic behaviors,such as EPSC and PPF,are realized in the simulation of a single-gate device.Based on the simulation of dual-gate devices,several spatio-temporal dendritic integration functions could be realized,such as the supralinear summation and shunting inhibition.Further,direction selectivity is successfully mimicked in the simulation of tri-gate devices4.The behavioral model of an EDL transistor,which is derived in the last chapter,is described in the Verilog-A hardware description language and then incorporated directly into the Hspice circuit simulator.In the Hspice circuit simulator,neuromorphic circuits/systems could be realized by combining the EDL transistors with other components,.Based on a tri-gate EDL transistor and a resistor,a McCulloch-Pitts neuron can be realized.Such an artificial neuron is demonstrated for the application of data classification of a two-dimension space.We also build a multi-layer neural network,which consists of four such artificial neurons.This neural network could perform not only linearly separable classifications,but also the nonlinearly separable classifications.