Research On The Design Of Naive Bayes Classifier Based On Memristor

With the advent of the 5G era,the development of artificial intelligence and the data explosion have placed higher requirements on computer performance.But with Moores Law gradually reaching the physical limits,the von-Neumann architectures limit the speed of data transmission and calculations.As the most promising solution to break through the traditional computer architecture,the Process-In-Memory(PIM)architectures based on the Resistive random-access memory(Re RAM or RRAM)can realize the integration of storage and computing and greatly improve the performance of existing computers.It is still a research hotspot at home and abroad.Naive Bayes classifier is a simple probabilistic classifier based on Bayes theorem.Compared with other neural network algorithms,it has a simpler structure and stable performance.It has a wide range of applications in information retrieval,pattern recognition and other fields.How to combine the naive Bayes algorithm with the electrical characteristics of the memristor to realize a simple and efficient probabilistic classifier has important research significance.This research proposes and designs a naive Bayes classifier based on the memristor,which maps the calculation process in the naive Bayes algorithm to the characteristic that the conductance gradient of the memristor is a logarithmic function with the number of pulses.It can effectively overcome the asymmetry and nonlinear constraints of the conductance change of the memristor in the application of neural networks.In this paper,the electrical characteristics of Ti/Hf O₂/Qds/Pt devices are extracted,and the code of naive Bayesian classifier based on memristor is written through python,which is verified by MNIST data set and IRIS data set respectively.Then,by modeling the non-ideal factors of memristor,such as the limited conductance states,device variations and device errors etc.,and research the influence of several non-ideal factors on the performance of the classifier.The simulation results show that device errors have a greater impact on the performance of the classifier,and the conductance inconsistency,conductance nonlinearity,and conductance state number have less impact on the memristor-based naive Bayes classifier.Finally,in view of the above problems,this article further optimizes the above classifiers.Based on the naive Bayes classifier based on the memristor,a selective Bayes classifier based on the memristor is designed and studied by simulation.The impact of non-ideal factors on the classifier is discussed.The simulation results show that the Bayesian classifier based on the memristor has a great improvement in tolerance to the non-ideal factors of the device.In this research,a naive Bayes classifier based on memristor is designed to realize real"on-chip learning"completely on memristor,which provides a new idea for in-memory calculation based on memristor.