Research And Design Of High-order Associative Memory Neural Network Circuit Based On Memristor

With the rapid development of artificial intelligence technology,people are committed to building hardware circuit systems with smaller size,lower power consumption,and faster computing speed.As the amount of data to be processed continues to increase,the requirements for hardware circuit systems are becoming more and more stringent,and Moore’s Law will gradually begin to fail.Meanwhile,the Von Neumann architecture is also limited.Existing semiconductor transistors have gradually been unable to achieve the functions required by users,and people are in urgent need of hardware with strong data processing capabilities,small size,and low power consumption.Memristors have become the best choice to solve the current challenges due to their excellent characteristics such as nanoscale size,low power consumption,and memory characteristics.The neural network circuit based on memristor is an important information processing method,which can speed up the calculation speed and greatly reduce the size and power consumption of the built circuit.Thus,it is of great significance to study memristive neural network circuits.Memristive neural network circuits can simulate advanced neural activities such as learning,forgetting,and associative learning of organisms.Meanwhile,it is also an important basis for realizing bionic functions in artificial intelligence technology.The proposed memristive associative memory neural network circuit only focuses on the learning and forgetting process of associative memory,ignoring some higher-order phenomena in the associative memory process.Since these phenomena are also very important for organisms,it is not conducive to the construction and application of biomimetic circuits.According to some higher-order effect in the process of associative memory,this thesis designs the corresponding associative memory neural network circuit.The main content and innovation of this thesis are as follows:(1)A memristor-based secondary conditioned reflex Pavlov associative memory neural network circuit is proposed.The secondary conditioned reflex is based on the existing conditioned reflex,taking the conditioned stimulus that had already established conditioning reflex relationships as an unconditioned stimulus and making it appear together with another conditioned stimulus.Thus,a conditioned reflex process that has never occurred before is achieved.Through secondary conditioned reflex,organisms can acquire more complex behaviors without resorting to unconditioned stimuli.The realization circuit of secondary conditioned reflex has two types of learning functions and two types of forgetting functions.The biggest difference from some previously designed associative memory circuits is a learning process.This learning process uses the conditioned stimulus after learning and the conditioned stimulus together,and this learning process does not require the help of the unconditioned stimuli.At the same time,secondary conditioned reflex is also used for the simple classification problem,and a simple classification circuit is designed.When the input features are unknown,the circuit can be used for learning to find the category corresponding to each unknown feature.Moreover,after the first learning process,the classification result can be achieved by only inputting the signal separately at the port corresponding to the condition that needs to be classified.The proposed circuit is verified by simulation on PSPICE.According to the actual input and output,the corresponding results are consistent with the theoretical analysis.(2)An associative memory neural network circuit with blocking phenomenon is proposed.The blocking process refers to the pretreatment of one element in the compound,thereby blocking or preventing the condition of another element.First,in the circuit structure,the circuit constructs a complete neural network circuit.The input neuron circuit composed of CMOS and other devices replaces the DC signal source,and this circuit can generate pulse signals.This input module makes the circuit more bionic.Secondly,in terms of circuit function,the circuit realizes advanced neural activities such as learning memory,associative memory and forgetting.At the same time,the circuit also achieves a function that has never been achieved in previous work— blocking,a higher-order effect in the associative memory process that is important to organisms.The circuit not only successfully simulates the occurrence of the blocking phenomenon but also realizes the disappearance process of blocking.The emergence of this circuit makes the associative memory process more complete and more in line with biological characteristics.Finally,the designed blocking circuit is simulated on PSPICE,and the output result of the circuit is consistent with the expected result.