Emulations of synaptic plasticity in the planar transistor configuration for neuromorphic computing
CHEN Yan,ZHANG ChenXi,WANG LaiYuan,LI TengFei,ZHU Ying,YI MingDong,HUANG Wei
DOI: https://doi.org/10.1360/N972016-01359
2017-01-01
Chinese Science Bulletin
Abstract:Due to the expansion of data information, computers based on thetraditional von Neumann architecture have been unable to store orflexibly handle large amounts of information. Compared to this, thehighly parallel nonlinear information processing the capability ofbrain reveals an attractive advantage, which is considered as thetarget of computing development. The human brain is composed of neuronalnetworks which are connected by 10 14 –10 15 synapses. Synapses are the connection channels for the exchangeof information between neurons, characterized by synaptic plasticity.Synaptic plasticity functions play a crucial role in the transmissionof neural signals in the brain and it is the molecular basis of biologicallearning and memory. In order to emulate neuronal networks, it isessential to design a physical device possessing the functions ofbiological synapses, so that the neuromorphic circuits which can functionallyemulate the brain-like behaviors can be realized. The transdisciplinaryresearches integrated the novel insights of neuroscience, physics,chemistry, materials science and engineering are performed. Researchershave shown that in the transistor structure with adjustive memristivecharacteristics, considered as the signal transmission and regulationmodules, the conductive channel and gate function are analogous toionic conduction and neurotransmitter release process in the biologicalsynapse, respectively. In most researches, voltage pulses appliedon the gate electrodes are used to simulate the presynaptic spikesor external stimuli, the channel conductance is considered as synapticweight. So the simulation of biological synapses can be realized insuch field effect transistors. Meanwhile, as a considerably popularplatform with memory effect, the electron-type transistor memory whichis designed for binary data storage is fabricated characterized bygentle process of charge trapping, where the relaxation tendenciesof OFF (ON) state in electrical operation mode can be utilized tosimulate the biological facilitating (depressing) activities, respectively.So, we can emulate the functions of biological synapses in the electron-typetransistor memory, too. In this paper, recent simulations of synapticplasticity functions in transistor synapses with planar structuresare summarized, including short-term synaptic plasticity (such aspaired-pulse facilitation and inhibition, dynamic filtering (high-passfiltering/low-pass filtering), spatiotemporally correlated signalprocessing, adaptation, etc), long-term synaptic plasticity, and thetransition between them, spike-timing-dependent plasticity, shuntinginhibition, and so on. The characteristics and different simulationmethods of these functions are illustrated, the mechanisms and advantagesof the planar transistor configuration are explained, current weaknessesand challenges of transistor synapses are pointed out, the futuredevelopment in this field are also prospected.