Research On Optical Differencial Artificial Neural Network

In the age of information explosion,the ability to process large amounts of data becomes crucial.The development and maturity of microelectronics technology is conducive to the emergence of the next-generation industry that uses emerging artificial intelligence services and high-performance computing.These data-intensive enterprises rely heavily on the advancement of hardware used for computing.As people rely on smart devices and data in their daily lives,companies also rely on data.Based on this characteristic,this paper designs a kind of optical ordinary differential equation neural network chip based on optical waveguide to replace the shortcomings of traditional neural network chip such as slow calculation speed and high heat generation.First of all,as to the most important part of a neural network-fully connected computing layer,the paper applies a fully connected neural network based on MachZehnder interferometer,with the analyzing of triangular network and the rectangular network.As to the weak optical nonlinearity,optical nonlinear components composed of photoelectric hybrid modulation are introduced.Finally,the paper combined these two parts to design an optical neural network.Secondly,to solve the problem of solving ordinary differential equation by optical devices,the paper analyzes two optical computing components to calculate optical integral and differential devices.Using the Micro-ring resonator,the paper designs a structure for solving ordinary differential equations optically.This structure can be used to calculate the solution of an ordinary differential equation.At last,the paper used components listed above to design an optical neural network of an ordinary differential equation.This neural network uses optical components to design an electronic neural network,compared to traditional electric network,optical neural network can compute with a speed of light which can compute faster.At the same time,in order to make a better use if the calculation unit of the Mach-Zehnder interferometer,this paper also designs an extended optical neural ordinary differential network,which effectively improves the utilization efficiency of the Mach-Zehnder interferometer neuron and do the classification task.In the experimental verification,the paper completed the experimental design of the two neural networks above and finally completed the experimental simulation verification.Through experimental analysis,the optical neural network structure studied in this paper has certain advantages over the existing neural network structure in all aspects.