Memristive Learning And Memory Circuit Design And Its Applications In Brain-Like Emotion

In recent years,the establishment of human-level brain-like intelligence inspired by the working mechanism of the brain has gradually become a research hotspot in the field of artificial intelligence.Therefore,it is necessary to learn from the way the brain processes information and establish a neural network integrating storage and computing to realize brain-like intelligence with the ability of learning,memory and emotional generation.However,many current researches on brain-like intelligence have not built a brain-like information processing architecture based on basic units such as neurons and synapses,which leads to the inability to perform hierarchical and parallel information processing,and cannot effectively solve von Neumann bottleneck.Besides,due to the excessive simplification of the working mechanism of the brain,the realized brain-like functions are not bionic enough in current researches.In addition,most of the works,especially the researches on brain-like emotions,is based on software algorithms.Although different algorithms and architectures are being pursued to improve the efficiency of training,using analog circuits to realize some or even all of the functions of software algorithms can achieve more efficient acceleration.Therefore,in this dissertation,bionic neurons and synapse circuits are designed using memristive devices with the functions of memory and integrated storage and computing,and then brain-like learning and memory circuits are designed based on the biological principles of brain learning and memory.On this basis,make full use of emotional generation mechanisms in the limbic system of the brain,according to corresponding bio-inspired models such as brain emotion learning model,a brain-like emotional information processing architecture is designed,which used to realize the generation and evolution of brain-like emotions that integrate of storage and computing.The primary innovative works and research results of this dissertation are as follows:In order to build neuron and synapse modules that are functionally bionic,simple in structure and easy to cascade,based on the characteristics of memristor,such as plasticity of memristance and memory,and combined with the functions and connection structures of neuron and synapse in brain,memristive neuron and synapse circuits are designed respectively as the basic unit modules for subsequent realizations of brain-like functional circuits.The designed memristive neuron can realize the activation and resting process similar to biological neuron,and can complete the efficient processing of input information;the designed memristive synapse can continuously adjust the weights according to the information from the anterior and posterior neurons,which realizes the bionic learning and forgetting functions.In order to simulate the learning and memory mode of brain more bionically,based on the learning and forgetting principles in Pavlov associative memory experiment,the adaptive function expansion and cascade of the designed memristive neuron and synapse circuits are carried out,so as to realize a memristive associative memory circuit with multiple learning and forgetting functions,and the speed of learning and forgetting can be adjusted adaptively according to the number of learning times.That is,the process that repeated training can improve the learning speed and slow down the forgetting speed is simulated,thus realizing the brain-like learning and memory functions.Based on the learning and forgetting rules in the memristive associative memory circuit,combined with the emotional learning theory,a single-associative emotional learning circuit that can learn a similar emotion through associative learning of a certain emotion is designed.On this basis,based on the principles of emotional fusion,an emotional multi-associative learning mechanism is constructed.According to the constructed mechanism,by cascading multiple single-associative emotional learning circuits,a multi-associative emotional learning circuit that can continuously learn different new emotions based on various original emotions is designed,which can effectively mimic the processes of brain’s emotional development.In order to fully simulate the process of emotional generation in the brain in terms of function and structure,according to the brain emotion learning model and the biological long short-term memory model,a brain-like multi-modal emotional information processing architecture is constructed firstly.Then a bottom-up circuit design method is adopted,based on the designed memristive neurons and synapse circuits,different functional module circuits that simulate the functions of the nerve nuclei in the limbic system of the brain are established.And then the functional modules are cascaded according to the established brain-like emotional information processing architecture to realize the memristive circuit that can generate emotions bionically in the two-dimensional emotional state space based on multi-modal information such as vision,speech and text.Based on the constructed brain-like emotion information processing architecture,drawing lessons from the evolutionary theory of emotion based on temporal and spatial information,a brain-like emotional evolution model combining internal regulation and external stimulus is established.On this basis,through the adaptive functional expansion and cascade of the functional modules of neural nuclei in the memristive brain-like emotional generation circuit,an internal regulation emotional evolution memristive circuit which can realize unconscious emotional evolution based on time information is designed,and an external stimulus emotional evolution memristive circuit which can realize conscious emotional evolution based on multi-modal input information is designed too.They cooperate with each other and can efficiently simulate the emotional evolution behaviors of the brain.The research results of this dissertation provide a new idea and reference for the realization of brain-like emotional learning,generation and evolution in the way of analog circuit,lay the foundation for the hardware implementation of the larger scale brain-like emotional circuit,and bring new hope for the realization of brain-like emotional learning,generation and evolution system and other related functions of emotional robot.