The Circuit Design Based On RRAM Array For Neuromorphic Computation

In recent years,the rapid development of artificial intelligence has greatly changed people's way of life,the progress of science and technology makes people have a desire to simulate the human brain.However,artificial intelligence which is used to simulate the human brain cannot break out the traditional Von Neumann architecture,the transfer of information between memory and processor still needs to be realized through the bus.When dealing with the mass of information like human brain,data transmission on the bus will generate a large number of power consumption,while the human brain processes the same data with much lower power consumption,the development of artificial intelligence tells us that we can't achieve the real intelligence of self-consciousness like the human brain by software programming,so people turn their attention to the hardware realization of neural network.The implement of neural network through hardware circuit is called neuromorphic computation.In the past,large scale neural network need a large number of synapses,but there were many disadvantages in the realization of the synapse,so the research progress of neuromorphic computation is slow.RRAM resistance switching device has the advantages of small size,low power consumption,but also it has similar electrical characteristics with the synapse,therefore it is considered to be used for the simulation of the synapse in neuromorphic computation.Due to the application of the RRAM resistance switching device which is used as the synapse to achieve the neuromorphic computation,it has the ability to breakout the traditional Von Neumann architecture to achieve the machine which can simulate the human brain working mechanism.Under this background,this dissertation carried out the related research work on the application of neuromorphic computation based on RRAM resistance switching devices.This dissertation studies the neural network of bionics,select the appropriate neural network model,neuron model,neural network learning rule,using the RRAM resistance switching device as the simulation of synaptic,design the neuromorphic computing architecture.Based on the RRAM device resistance phenomenon,we deeply study the conductive filament model,establish the RRAM resistive device model and verify the consistency of the model with resistive phenomenon and the actual device,and some suggestions are put forward through the use of device model and real object.For the circuit structure and the training mechanism to calculate the character 0-9 training function neural morphology is proposed,the training function of neural pattern calculation is realized,and verify the circuit training results.Based on the characteristics of the human brain work,the circuit is improved,and the yield of the device is analyzed by Monte Carlo method.Through the use of RRAM resistance switching devices,we made the circuit design in neuromorphic computing field,breaking the traditional Von Neumann architecture in artificial intelligence,provide some useful reference for the further research work in this direction.