Font Size: a A A

Design And Optimization Of Plasmonic Devices Based On Machine Learning

Posted on:2021-04-10Degree:MasterType:Thesis
Country:ChinaCandidate:Q LiuFull Text:PDF
GTID:2381330632462267Subject:Electronic Science and Technology
Abstract/Summary:PDF Full Text Request
As a new two-dimensional material,graphene can support the excitation and propagation of the surface plasmons polaritons(SPP)with the advantages of low propagation loss and high binding force.Based on the unique properties of graphene and the optical properties of SPP,various nanostructure designs have emerged continuously.Graphene metamaterial(GM),which is composed of graphene nanoribbons(GNRs),has attracted much attention because of its relatively simple process flow.In recent years,for the inverse design of photonic devices,artificial neural networks(ANN)are often used to replace the traditional numerical simulation method,thereby greatly reducing the design time.However,compared with some traditional machine learning methods,ANN’s advantages are not obvious in accuracy and efficiency in the design of some simple photonic devices.In addition,in the performance optimization of photonic structures,only a single target is usually optimized,and multi-objective optimization algorithms are rarely used.Therefore,there is still a lack of a comprehensive and fast solution in the design and performance optimization research of GM structure.In response to the above problems,the specific work of this thesis mainly include the following aspects:1.A new GM structure composed of double-layer GNRs is proposed.When light is incident vertically,it can excite SPP mode on the graphene surface.Based on the mutual coupling of modes in the structure,the plasmon induced transparency(PIT)can be formed in the transmission spectrum.By adjusting the chemical potential of graphene,PIT can be dynamically tuned.Based on the above properties,this structure can be used to build high-performance photonic devices,such as optical filters,optical switches,slow-light devices,and so on.2.For the proposed GM structure,a forward transmission spectrum prediction and inverse design scheme is proposed by using a variety of machine learning algorithms.The results prove that,whether forward transmission spectrum prediction or inverse design,the prediction accuracy of all algorithms is above 91%,and the consumption time of training and prediction is shorter than 37 seconds.Among these algorithms,the random forest ensures that the prediction accuracy reaches 96%,and the algorithm processing time is shorter than 7 seconds,which is more advantageous in terms of prediction accuracy and efficiency than ANN.The above results indicate that for some simple photonic devices,traditional machine learning is better than the current research hotspot ANN in some performance aspects.3.Evolutionary optimization algorithms are used to optimize the transmission spectrum of the GM structure for single and multi-objective optimization to obtain better performance indicators.As for single-objective optimization,after multiple iterations of genetic algorithm,quantum genetic algorithm and particle swarm algorithm,the loss between the optimized transmission spectrum and the target transmission spectrum drops to 0.12,0.09 and 0.13,respectively,which achieves the overall optimization of the transmission spectrum.As for multi-objective optimization,the non-dominated sorting genetic algorithm-II is used to optimize the peak-valley difference of PIT in the transmission spectrum.After a lot of iterations,for a single transparency window(dual transparency window),the difference between transmission peaks and valleys increase to 0.86 and 0.97(0.87,0.83,0.79 and 0.69),respectively,effectively improves the steepness of the transmission spectrum.The results can be widely used in high-performance optical switches,refractive index sensors and slow-light devices.
Keywords/Search Tags:graphene metamaterial, transmission spectrum prediction, machine learning, transmission spectrum optimization, evolutionary algorithm
PDF Full Text Request
Related items