Numerical Investigation On Storage Property Of Photonic Neuron Dynamics And Associative Learning Network Based On VCSELs

In recent years,photonic information processing technology has been applied to neuromorphic calculations due to its breakthrough in bandwidth limitations and all-optical digital pulse conversion functions.The photonic neuron based on vertical-cavity surfaceemitting lasers(VCSELs)has the inherent advantages of low manufacturing cost,easy integration in a 2-dimensional array,low power consumption,and efficient coupling to optical fibers.Therefore,the VCSELs-based photonic neuron provides a good prospect for the photonic neuromorphic system in information processing and computation.In this thesis,the following concrete researches are carried out based on the neural dynamics generated by VCSEL subject to different double polarization optical injections(DOI)and the photonic neural network composed of VCSELs subject to DOI.One is to study the phasic spike generated by VCSEL subject to double optical injection with pulsed orthogonal polarization(DIPOP)through theory and experiments,and study the effects of the injection intensity of the external pulsed optical signal,injection frequency detuning of the external pulsed optical signal,the pump current,the spontaneous emission noise and white Gaussian noise on the phasic spike generated by VCSEL.In addition,a scheme for short-term storage of phasic spike in two mutually coupled VCSELs with DIPOP and the theory of this scheme based on the spin-flip model(SFM)model are proposed and studied firstly.Then,we study the effect of the injection intensity and injection frequency detuning of the external pulsed optical signal,the pump current,the central frequency detuning,coupling strength,coupling delay between the two mutually coupled VCSELs on the storage time of phasic spike in two mutually coupled injected VCSELs with DIPOP.The results show that injection frequency detuning of the pulsed optical signal,pump current,the central frequency detuning and coupling delay between the two mutually coupled VCSELs are the main factor affecting the storage time of phasic spike.In addition,by increasing the coupling delay,the storage time of phasic spike in two VCSELs is increased,but the storage cycles of the phasic spike are hardly affected.Finally,the robustness of white Gaussian noise in the process of phasic spike storage is considered and studied.Second,we use three VCSELs subject to double optical injection with pulsed parallel polarization(DIPPP)to build a photonic neuromorphic network(PNN)that can simulate the process of Pavlovian association learning and forgetting.The theoretical model of PNN is also derived based on the spin flip model.According to the learning rules of photonic spike timing-dependent plasticity(STDP),the connection synaptic weights between neurons in PNN can be modified through ex-situ learning rule.In the study,the association between feeding food and bell ringing can be established through Long-term potentiation(LTP)window of STDP during the process of associative learning;and in the process of forgetting,the Long-term depression(LTD)window of STDP can be used to forget the association.In addition,we also studied the effect of the time interval between two pre-synaptic spikes on the speed of associative learning and forgetting.The research results show that the shorter the time interval between two pre-synaptic spikes,the faster the associative learning and forgetting,and vice versa.Besides,the speed of associative learning and forgetting based on the PNN is nearly three orders of magnitude faster than that based on electronic neuromorphic network.Finally,based on the photonic associative learning,pattern recall is proposed and realized numerically.The recovery effect of pattern recall on different types of incomplete patterns is analyzed in detail.The research results show that when the incomplete pattern contains all the contour information of the image,even if its incomplete rate reaches63%,the incomplete pattern can be recovered to its corresponding complete pattern.