Due to the expansion of data information, computers based on the traditional von Neumann architecture have been unable to store or flexibly handle large amounts of information. Compared to this, the highly parallel nonlinear information processing the capability of brain reveals an attractive advantage, which is considered as the target of computing development. The human brain is composed of neuronal networks which are connected by 10 14 –10 15 synapses. Synapses are the connection channels for the exchange of information between neurons, characterized by synaptic plasticity. Synaptic plasticity functions play a crucial role in the transmission of neural signals in the brain and it is the molecular basis of biological learning and memory. In order to emulate neuronal networks, it is essential to design a physical device possessing the functions of biological synapses, so that the neuromorphic circuits which can functionally emulate the brain-like behaviors can be realized. The transdisciplinary researches integrated the novel insights of neuroscience, physics, chemistry, materials science and engineering are performed. Researchers have shown that in the transistor structure with adjustive memristive characteristics, considered as the signal transmission and regulation modules, the conductive channel and gate function are analogous to ionic conduction and neurotransmitter release process in the biological synapse, respectively. In most researches, voltage pulses applied on the gate electrodes are used to simulate the presynaptic spikes or external stimuli, the channel conductance is considered as synaptic weight. So the simulation of biological synapses can be realized in such field effect transistors. Meanwhile, as a considerably popular platform with memory effect, the electron-type transistor memory which is designed for binary data storage is fabricated characterized by gentle process of charge trapping, where the relaxation tendencies of OFF (ON) state in electrical operation mode can be utilized to simulate the biological facilitating (depressing) activities, respectively. So, we can emulate the functions of biological synapses in the electron-type transistor memory, too. In this paper, recent simulations of synaptic plasticity functions in transistor synapses with planar structures are summarized, including short-term synaptic plasticity (such as paired-pulse facilitation and inhibition, dynamic filtering (high-pass filtering/low-pass filtering), spatiotemporally correlated signal processing, adaptation, etc), long-term synaptic plasticity, and the transition between them, spike-timing-dependent plasticity, shunting inhibition, and so on. The characteristics and different simulation methods of these functions are illustrated, the mechanisms and advantages of the planar transistor configuration are explained, current weaknesses and challenges of transistor synapses are pointed out, the future development in this field are also prospected.
Read full abstract