Bioinspired Memristive Neural Network Circuit Design Of Learning And Memory And Its Application

In recent years,inspired by the neural information processing mechanism of the brain and the biological basis of cognitive behavior,the establishment of brain-like intelligence driven by cognitive bionics has gradually become an important development path in the field of artificial intelligence.Learning and memory,as the basic cognitive functions of the brain,are the basis for realizing cognitive activities and the most concentrated embodiment of brain-like intelligence.However,the current neural network structure of brain-like learning and memory does not have a deeper fusion of biologically relevant theories,and the designed circuit is complex in structure,large in size,high in power consumption,as well as not close enough to the bionic characteristics of neurons and synapses.Thus,it is difficult to achieve large-scale,lowpower and scalable neuromorphic circuits,and it cannot simulate learning and memory functions more realistically.In addition,human cognition of the world begins with multiple sensory modalities.There are facilitative and inhibitory interactions between different sensory modalities to accomplish different cognitive functions and improve cognitive abilities,while the existing studies of brain-like learning and memory circuits do not consider the interactions between modalities.To address the above problems,this thesis uses the neural information processing mechanism and biological principles of the biological brain as the theoretical support,and uses a novel neuromorphic device,memristor,as the neuron and synapse to bulid the simple and efficient bioinspired memristive neural network,design unimodal and cross-modal brain-like learning and memory circuits,and implement multiple functions of brain-like cognition toward different application scenarios of language cognition and object cognition.It is of great research significance and practical application to the development of bionic intelligent devices and brain-like intelligence.The main research results of this thesis are as follows:(1)Drawing on the biological principles and the neural information processing mechanism of Pavlov’s associative memory experiment,and combining the associative learning mechanism during second language acquisition,the simplified biological neuron structure and the corresponding bioinspired memristive neural network are constructed by taking the text information of two kinds of words,Chinese as the first language and English as the second language,as the two different stimuli for unimodal visual.Based on this,the unimodal visual associative learning circuit,which consists of the synapse module,the neuron module and the control module,is designed to realize six brain-like classical conditioning functions,such as learning,overshadowing,blocking,long-term effect of overshadowing leading to blocking,recovery from overshadowing and recovery from blocking.The simulation analysis in PSPICE shows that the designed bionic circuit exhibits the mutual inhibition between different language stimuli,which provides a further reference for bionic intelligent devices such as social robots or educational robots that can address language cognitive impairment.(2)Drawing on the biological principles of Drosophila’s associative memory experiments and the relatively simple neural circuit mechanism of Drosophila,and combining the phenomenon of cross-modal interactions between vision and olfaction,image stimuli and odor stimuli are used as unimodal visual and olfactory stimuli respectively to construct bioinspired memristive neural network.Based on this,the cross-modal associative memory circuit between visual and olfactory conditioning,which consists of the threshold module,the synergy module,the synapse and neuron module,is designed to implement five kinds of brain cognitive functions such as unimodal visual learning with threshold,unimodal olfactory learning with threshold,cross-modal reinforcement,cross-modal facilitation,and cross-modal memory transfer.The simulation analysis in PSPICE shows that the designed bionic circuit presents cross-modal synergy,facilitation and transmission interactions between vision and olfaction,providing a new idea to realize brain-like learning and memory circuits comparable to the dynamic associative cognitive abilities of biological brains.