Research On The Application Of Inverse Design In Photonic Devices And Systems

The performance of photonic devices is significant to the performance of optical communication and interconnection systems,which affects the final bit error rate and communication capacity.Before manufacturing photonic devices,we need to design the structural parameters of photonic devices according to the target performance requirements.In the design process of photonic devices,we usually have to try through experience or spend a lot of time on numerical simulation,then compare the results with the target response,and constantly update the structure to obtain the final structural parameters.The whole process is based on iteration.On the other hand,as the structures of photonic devices and systems become more and more complex,the design process becomes much more challenging.In this scenario,the structure parameters of photonic devices involve high-dimensional parameter space,thus it becomes quite redundant and tedious to obtain the structure parameters of photonic devices corresponding to the target performance,and lack of controllable theoretical guidance,so it is difficult to design devices that strictly meet the performance requirements.Therefore,the design method of photonic devices to obtain the theoretical structural parameters corresponding to the target performance before manufacturing is very important.In this paper,we study and explore the application of inverse design in photonic devices and photonic systems.Inverse design is a method of applying various computational optimization techniques to device design.It is a design process from performance space to parameter space,while the conventional design process is a design process from parameter space to performance space.Compared with the traditional design schemes,inverse design can not only overcome the inherent problems in the design of various photonic devices and photonic systems,but also simplify the design process and obtain more accurate solutions.The frameworks of inverse design in photonic devices and photonic systems are divided into three categories: inverse design based on theoretical analysis,inverse design based on optimization algorithm and inverse design based on machine learning.Micro-ring resonator channel dropping filters,flat spectral amplifiers and nanostructures with functions of regulating spontaneous emission spectra are important parts of photonic system.For the practical problems and requirements of these key devices,the corresponding solutions are given.In this paper,we employ several algorithms such as the algorithm based on theoretical analysis,conjugate gradient method,artificial neural network and random forest to inversely design several key photonic device examples such as micro-ring resonator channel dropping filters,flat spectral amplifiers and nanostructures with functions of regulating spontaneous emission spectra,realizing the corresponding functions and solving the respective problems encountered in the design of these devices.The detailed research contents are as follows:1.Various inverse design algorithms in photonic devices and photonic systems are constructed,including inverse design based on theoretical analysis,inverse design based on optimization algorithm and inverse design based on machine learning.2.Microring resonator channel dropping filter is an essential device in the WDM system.Since the coupling between the ring-ring and the ring-straight waveguide is significantly different,the resonance frequency mismatch between the resonators is caused,which leads to the distortion of the filter spectrum.The general solution is to change the size,shape,or refractive index of each cavity to compensate for the resonant frequency of each cavity,and it is difficult to give the desired spectral response.In order to overcome these difficulties and make the system robust,we propose an inverse design scheme of microring filters based on a multi-cell series structure.Starting from the performance index,we solve the structural parameters that meet the performance index step by step based on theoretical analysis.1)A novel microring filter structure is constructed by cascading multiple cells(the unit cell is composed of input straight waveguide,single microring,and output straight waveguide).This structure will not cause spectrum distortion,and the frequency correction is straightforward.The theoretical performance of this structure is analyzed in detail,and the theoretical performance limit(including the roll-off characteristics and 3d B bandwidth)of the filter with this configuration is discussed.2)For this kind of structure,starting from the performance index,based on the theoretical analysis method,the design steps of the filter meeting the target performance index are given,including the filter order configuration,the acquisition of the initial radius of the microring,the coupling distance,and other parameters,as well as the fine-tuning of the radius of the microring to achieve the purpose of frequency shift correction.3.In DWDM optical network,signal loss is compensated by optical amplifiers.However,the gain spectrum of an optical amplifier usually changes significantly with the wavelengths,resulting in the power imbalance between channels.Since the worst signal-to-noise ratio of each channel affects the maximum transmission distance,optical amplifiers with relatively flat gain are very important.By using the inverse design algorithm based on conjugate gradient optimization,a new amplifier structure based on a structured waveguide is obtained.Numerical results show that the structured waveguide increases the 3d B bandwidth of the spectrum by 22.44% and the flatness by 32%,which effectively solves the problem of uneven gain of the amplifier.1)The physical mechanism of the influence of local density of states on the emission spectrum is derived,and the emission spectrum problem is transformed into a problem equivalent to the local density of states.2)Based on the conjugate gradient algorithm,a structured waveguide(a periodic structure with multiple defects)is constructed to meet the target local density of states.After obtaining the optimal structural parameters,the emission spectrum based on numerical simulation is observed to analyze the effectiveness of the inverse design.4.Based on artificial neural networks,the transmission spectrum prediction and inverse design of high-order microring filters are studied.This scheme can predict the transmittance directly and quickly,and obtain the structural parameters closest to the target spectral shape in the design space.This reverse design scheme effectively solves the problem that the design of a high-order microring filter is tedious and it is difficult to get the target response accurately.More importantly,in the process of inverse design,the problem of frequency offset compensation in filter design is automatically solved.1)For the scenario of transmission prediction,when training the network, the structure parameters in the training set are taken as the network input,and the corresponding transmission spectrum is taken as the network output so that the loss of the training set and the verification set is minimal and there is no over fitting,and the loss of the test set is tested to test the generalization performance of the network model.Finally,given the structure parameters,the transmission spectrum of the filter is predicted by using the trained network.2)For the inverse design scenario,when training the network,the transmission spectrum in the training set is taken as the network input,and the corresponding structural parameters are taken as the network output so that the loss of the training set and the verification set is minimal and there is no over fitting.At the same time,the loss of the test set is tested to evaluate the general performance of the network model.Finally,the target transmission spectrum is input to the trained network,and the network outputs the corresponding structural parameters for inverse design.5.Based on random forest algorithm,photonic nanostructures are inversely designed to explore the regulation of spontaneous emission spectrum.1)A core-shell photonic nanostructure is constructed to modify the spontaneous emission spectrum.Due to the interaction and coupling of modes between the layers of core-shell photonic nanostructures,the spectral regulation mechanism is realized.2)Based on the random forest algorithm,the core-shell photonic nanostructures are designed reversely to achieve the purpose of adaptive control of the emission spectrum and realize the flexible control of the emission spectrum.