Research Of Silicon-based Reconfigurable Analog Optical Computing Chip

With the advent of the information age,the computing demands of electronic systems are growing rapidly,thus presents fresh challenges in performance.Optical computing relies on its advantages in analog signal processing and the characteristics of no delay,ultra-high speed,low power consumption,and parallel computing to compensate for the disadvantages of electrical computing.It is widely used in high-speed microwave signal processing,computing accelerators and matrix-vector multiplication computing in machine learning,and many other fields.In addition,the booming development of integrated photonics provides hardware conditions for on-chip optical computing.At the same time,drawing on the idea of programmable electrical circuits,the concept of large-scale programmable optical computing networks is proposed.The research of reconfigurable optical computing chip is oriented to a kind of general optical resource configuration hardware,which realizes the reconstruction of different functions through programming.It can meet the needs of multiple optical computing in a single structure and has become a research hotspot in the field of optical computing.Although the field of reconfigurable optical computing chips has made rapid development and great progress,it still faces some problems and challenges.First of all,the field has not enough research on reconfigurable and scalable photonic topology network chips.Particularly,how to use large-scale photonic topology networks to construct reconfigurable large-bandwidth photonic analog operations is a key challenge.Secondly,in existing optical computing networks,control of system process is poor,and the limited optical computing network can only process limited optical data,which shows that there is still a lack of research on the expansion of network computing capabilities.Finally,the automatic configuration ability of large-scale photonic networks is weak,and most of the solutions still rely on the internal monitoring of network and manual calibration.Aiming at the above-mentioned problems,this dissertation conducts theoretical analysis and experimental verification on the silicon-based reconfigurable analog optical computing chips from the three perspectives of network topology,expansion of network computing and intelligent self-configuration algorithm.The content of this article mainly includes:(1)The development trends in the field of integrated reconfigurable optical computing are introduced,especially the research background and research significance in the field of analog optical computing and linear matrix optical computing.The dissertation reviews the domestic and foreign research progress in the above-mentioned fields,and points out the current challenges and opportunities.(2)Using the transfer matrix method,a detailed theoretical analysis of the principles of silicon-based microrings and silicon-based MZI is carried out.Several tuning methods of silicon-based devices are introduced,including thermal tuning,electrical tuning and optomechanical force modulation.Finally,the preparation and testing methods of silicon-based chips are introduced.(3)A reconfigurable spectral shaper based on a microring with two-dimensional adjustment of amplitude and phase is proposed.The cascade of 16 microrings is used to realize a fixed loop reconfigurable optical signal processor,and the experiments verifies the application of fractional-order continuous tunable differentiator and optical wavelength selective switch.Afterwards,aiming at limitations of traditional spectral shapers in reconfigurability and processing bandwidth,a reconfigurable analog optical signal computing chip based on quadrilateral topology loops is designed and fabricated.By tuning the light path of the topological network,differential,Hilbert,and integral transformation can be realiezed,where the functions can be switched and the operation order can be tuned.By improving the network structure,the signal processing bandwidth can reach 40 GHz.(4)Aiming at the problem that the network computing capacity cannot be expanded in the optical computing networks,the two-dimensional tuning microrings with amplitude and phase control mentioned above are used to realize the explicit optical matrix calculation.A4×4 universal optical matrix computing unit based on the on-chip microring array is designed and fabricated.This solution realizes the expansion of transfer matrix from non-negative domain to full real number domain through balance detection;and realizes expansion of matrix computing to full real number domain,full complex number domian,larger size and convolutional operation through electrical process control.Thus the 4×4 real number domain transfer matrix core is expanded into a 16×16 full complex number domain one.The application of the chip in signal processing(Walsh-Hadamard transformation,discrete cosine transformation and discrete Fourier transformation)and image processing(sharpening,blurring and edge extraction)is experimentally demonstrated.(5)Aiming at the problem of poor automatic reconfiguration capabilities in optical computing networks,an intelligent inuversal optical linear matrix calculation chip based on gradient descent algorithm is designed and fabricated.An arbitrary matrix transmission network based on unitary transformation is constructed by using 20 MZIs and 48 thermooptic phase shifters.The highlight of the scheme is that the chip is considered as a "black box",while system training is carried out through external optical ports,and the principle of gradient descent algorithm is adopted.The chip can realize self-configuration based on electrical controlling signals.In matrix operation,three basic matrix equations are solved,and the Google Page Rank algorithm is verified experimentally.For analog optical signal processing,we realize the tunable multi-channel optical switches,two different types of MIMO descrambler and the application of tunable optical filter.