Research On Reservoir Computing Based On Semiconductor Lasers

In the current information age,the amount of information has exploded.Finding a fast and efficient way of processing information has also become a general trend.Neuromorphic computing is a computing method that simulates the processing of information in the human brain.Neuromorphic computing has attracted great attention in recent years,which effectively solves the problems that are difficult to handle with traditional computing methods,such as pattern recognition,speech recognition,and motion detection.In addition,the rapid development of photonic technology and optical devices has also offered new opportunities for optical information processing.Reservoir computing(RC)is a simple and efficient neuromorphic computing framework similar to human cortical circuits,and thus,is suitable for processing timing signals.It is derived from several recurrent neural network models,including state echo networks and liquid state machines.The RC system consists of a reservoir layer for mapping inputs to high-dimensional space and an output layer for pattern analysis from high-dimensional states in the reservoir layer.Compared with the software implementation on traditional electronic computers,the hardware implementation of RC based on optical devices will be more conducive to ultra-high-speed and ultra-lowpower information processing.The hardware implementation of a reservoir layer by using a delayed feedback nonlinear dynamic system(the time-delayed RC)has attracted widespread attention due to its simple structure.According to time division multiplexing,virtual nodes are set up on the delay feedback loop to replace the nodes in the traditional reservoir,which greatly simplifies the structure of hardware RC.This thesis is based on the time-delayed RC and new type of semiconductor lasers(SLs)to carry out in-depth cross-over research.The vertical cavity surface emitting laser(VCSEL)and semiconductor nanolaser(SNL)are considered as nonlinear nodes to investigate RC systems based on different SLs.Specifically,a timedelayed optical RC system model is established based on the new SLs,and the impact of key parameters of the system on prediction,classification performance and memory ability are studied.The research results provide theoretical guidance for the realization of neuromorphic RC,and at the same time have reference to the brain information processing mechanism,so it has important physiological and scientific significance.In addition,inspired by brain information processing mechanisms,neuromorphic RC can also provide new ideas for laser network research.On this basis,a neuromorphic RC system is constructed based on a fiber optic platform,which expands the potential applications of lasers in the field of ultra-high-speed photonic neuromorphic systems and high-speed brain-like information processing.The main work and research results of the thesis are as follows:1.The time-delayed RC system using a VCSEL subject to polarized optical feedback as a nonlinear node was studied,and the parallel task processing in the time-delayed optical RC system based on a single VCSEL was proposed for the first time.Due to the unique polarization characteristics of VCSELs,not only single-task processing could be achieved on such RC system,but also parallel task processing could be achieved.In the VCSEL,parallel polarization feedback and orthogonal polarization feedback were considered.A chaotic time series prediction task and a waveform recognition task were used to evaluate the parallel task processing performance of the time-delayed optical RC system based on polarization multiplexed VCSEL under two polarization modes.The results showed that the time-delayed optical RC system based on polarization multiplexed VCSEL could obtain good performance in a wide parameter range under both feedback conditions.In addition,better performance could be obtained under the condition of parallel polarized optical feedback.Since the timedelayed optical RC system based on polarization multiplexed VCSEL allowed parallel processing of tasks in two polarization modes in a VCSEL,it was valuable for lowpower neuromorphic photonics systems.2.A time-delayed RC system based on a VCSEL with dual-polarized optical feedback was proposed,and on this basis,the memory capacity of the RC system was deeply studied.For comparison,the memory capacity of an RC system based on a VCSEL with single-polarized optical feedback was also considered.Numerical results showed that,compared with single-polarized optical feedback,the time-delay RC system of VCSEL based on dual-polarized optical feedback could obtain higher memory capacity.When two feedback delays were two-fold of the RC system,the memory capacity was much lower.Moreover,we found that a larger feedback delay or information processing time could lead to a wider region of higher memory capacity.Thus,such proposed VCSEL-based RC system provided a prospect for the further development of the neuromorphic photonic system based on RC.3.A four-channels delay optical RC scheme based on mutually coupled VCSELs(MDCVCSELs)was proposed and numerically studied.For the first time,the four channels were implemented in two orthogonal polarization modes of two VCSELs.At the same time,the chaotic time series prediction task was used to quantitatively evaluate the prediction performance of the proposed RC system.It was found that the fourchannels RC based on MDC-VCSELs could produce comparable prediction performance with one-channel RC,and it was possible to increase four times information processing rate by using the four-channels RC.Moreover,the influences of sampled period of input signal and the number of virtual nodes were also considered.The proposed four-channels RC based on MDC-VCSELs was valuable for further enhancing the information processing rate of RC-based neuromorphic photonic systems.In addition,we also extended the coupled VCSELs system to the deep timedelayed optical RC,by stacking multiple reservoir layers between the input layer and the output layer to achieve spatial depth,thereby achieving better RC performance.The proposed time-delayed optical RC based on coupled VCSELs provides a new idea for further improving the information processing rate of RC-based neuromorphic photosystems.4.A time-delayed optical RC system based on a SNL with single feedback was proposed and numerically studied.At the same time,the chaotic time series prediction task was used to evaluate the prediction performance of the time-delayed optical RC system based on SNL with single feedback.The effects of the Purcell cavity-enhanced spontaneous emission factor F and the spontaneous emission coupling factor ? on the proposed RC system were analyzed extensively.It was found that,in general,increased F and ? extended the range of good prediction performance of the SNLbased RC system.Moreover,the influences of bias current and feedback phase were also considered.Due to the ultra-short photon lifetime in SNL,the information processing rate of the SNL-based RC system reached 10 Gpbs.In addition,the dual feedback were introduced into the time-delayed optical RC system based on SNL,and a time-delayed optical RC system based on SNL with dual-feedback was proposed.The simulation results showed that,compared with the time-delayed optical RC system based on the SNL subject to single feedback,the time-delayed optical RC system based on the SNL with dual-feedback could obtain better prediction performance.The influences of bias current,input signal modulation depth,feedback strength,feedback delay and other factors were also considered.The proposed highspeed SNL-based RC system provided theoretical guidelines for the design of RCbased integrated neuromorphic photonic systems.5.Based on the fiber optics platform and numerical simulation,the experimental prototype system of the VCSEL-based time-delayed optical RC was built for singletask processing.The dynamic behavior of the VCSEL-based time-delayed optical RC system was studied through experiments,and a single chaotic time series prediction task or a single channel equalization task was realized in this system,respectively.Note that,dual polarization modes could coexist in a VCSEL under specific parameter conditions.The experimental platform of the time-delayed optical RC system based on a single VCSEL for dual-task processing was built.A universal solution for the time-delayed optical RC system based on a single VCSEL for dual-task processing was built.By enhancing the external optical injection strength,the chaotic time series prediction and channel equalization,were simultaneously realized in this system,which provided a reference framework for the future multi-mode RC research.