Memristor-based Neural Synaptic Learning Function Simulation And Bionic Transmission

Memristors have become the first choice for artificial synapses due to their continuously adjustable resistance,nanometer size and bio-like synaptic advantages.Habituation and associative memory are the basic learning styles of biology.Building a circuit based on the memristor to simulate the characteristics of habituation and associative memory is the most important way to implement the memristive neuromorphic system.Improving the memristor model can provide a consult for guiding its design and construction process to enhance its bionic synapse performance.This thesis chooses tungsten oxide memristor with threshold characteristics and forgetting characteristics to replace common HP memristor as synapse,constructs neuromorphic circuits,and simulates biological learning behaviors and temperature sensitivity of biological synapses.The main body of the research is carried out in the following three parts:Firstly,the significance of habituation is affirmed,the deficiency of current habituation realization circuit is indicated and a habituation realization circuit that does not require a voltage recovery module according to the forgetting characteristics of the tungsten oxide memristor model is established.Different from the multi-input positive and negative pulses used in the past,the external excitation voltage of the circuit constructed in this article only needs to be set to a single-input positive pulse,which is more suitable for actual neural signal,and can achieve the simulations of short-term habituation,long-term habituation and de-habituation phenomenon.Secondly,the phenomena of the classic Pavlovian dog experiment and the delayed learning are studied in detail,and an associative memory circuit with a delay module is constructed according to the threshold value and forgetting characteristics of the tungsten oxide memristor.The newly added delay module is used to control whether associative memory occurs,and there is no need to set up another module to achieve voltage control.The simulation verifies that it can not only simulate the learning and forgetting characteristics of associative memory,but also the delayed learning characteristic.Finally,an improved temperature-based model of tungsten oxide memristor is proposed and used as a biological synapse coupling double HH neuron to simulate the temperature-sensitive characteristics of synaptic transmission.In order to improve the accuracy of fitting the actual forgetting curve of the memristor,a new expression of the retention value of the memristor conductivity is added,the influence of temperature on ion migration and diffusion is considered,and a temperature variable is introduced into the original model.The rise in temperature causes changes in the rate of oxygen vacancy migration and diffusion,which leads to an increase in the rate of change in the conductance of the memristor,and further excitatory postsynaptic membrane potential amplitude and firing times.The simulation phenomenon is consistent with the neurophysiological experimental phenomenon.In this thesis,a tungsten oxide memristor is used to simulate synapse to construct neuromorphic circuits,which can transfer learning rules into neuromorphic systems and open up a new way of thinking about the influence of temperature on synaptic transmission.