Optimal Design Of Integrated Optical Phased Array Based On Deep Learning

Optical Phased Array(OPA)is a new beam steering technology which is derived from microwave radar.It can be seen as an extension of microwave phased arrays in the field of optics,and can be used in applications such as autonomous driving,laser display and neural probes.Since the first realization of large-scale on-chip optical phased array chips in 2013,the related research on optical phased arrays has developed rapidly.In recent years,optical phased arrays mainly focuses on improve the performances of high integration,miniaturization,narrow beam bandwidth,and high grating lobe suppression,which requires a lot of device design and system design optimization.However,traditional design methods are inefficient and complex,and have empirical limitations,which limit the further improvement of the performance of optical phased arrays.In recent years,deep learning has been widely used in image recognition,speech recognition,language processing and other fields due to its advantages of high intelligence,strong adaptability and high optimization dimension.It has also become a brand-new method in the field of photonic device design,which is expected to overcome the weaknesses of traditional design methods in optical phased arrays and promote the development of optical phased array technology.This dissertation is mainly based on deep learning theory,carries out related research on the design of micro-nano photonic devices and chip systems aiming at the problems of the device design and topology optimization in optical phased arrays.The main contents and related achievements are as follows:Firstly,an efficient reverse design method based on deep learning for micro-nano photonic devices is proposed.The size of the micro-nano photonic device is closely related to the integration of the optical phased array.This dissertation studies the reverse design technology based on deep learning,and proposes a genetic-deep neural network reverse design algorithm by organically combining genetic algorithm and deep neural network,while reducing the data collection burden in deep learning.Further developed a Graphical User Interface(GUI)which is intuitive and easy-to-operate.Compact beam splitting devices with special splitting ratio and ultra-broadband characteristics for optical phased arrays was designed,fabricated and tested.The experimental results show that the beam splitting device with special splitting ratio achieves the performance of different target splitting ratios under the device size of less than 2.5 μm×2 μm.The ultra-wideband multimode interference coupler can achieve the maximum insertion loss of 0.43 d B in the broadband range of 270 nm with the device size is only 2.5 μm×1.5μm.The above research realizes the design of a miniaturized optical phased array beam splitting device.Secondly,a one-dimensional optical phased array chip with high grating lobe suppression performance was trial-produced based on deep learning.Optical phased array with large aperture sparse antenna array has the advantages of narrow beam width and low power consumption.However,if the antenna spacing in the array is too large,the design effect of grating lobe suppression will be poor.This dissertation proposes an efficient grating lobe suppression design model based on deep learning theory,and realizes the high grating lobe suppression design with large aperture sparse antenna arrays.Further the device design and layout design of the optical phased array were completed,and a one-dimensional high grating lobe suppressed optical phased array chip was fabricated.The experimental test was carried out,and the high grating lobe suppression effect of 7.9 d B and the narrow beam scanning with a beam width of 0.29°were achieved.Thirdly,a compact 2D optical phased array architecture with serial structure is proposed,and a high grating lobe suppression design is based on deep learning.The directional coupler in a typical 2D optical phased array architecture usually increases the distance between the antennas,causing waste of chip design space and affecting the design effect of grating lobe suppression.This dissertation proposes a serpentine 2D optical phased array architecture that eliminates the use of directional couplers.In the small and medium design scale,the design space can be saved by more than 30%,and the integration degree of the two-dimensional optical phased array can be improved.The serpentine 2D optical phased array realizes the high grating lobe suppression design through the deep learning-based design model,and achieves high grating lobe suppression effects of 8.8 d B and 12 d B in the X and Y directions.