Study On Integrated Photonic Devices And Systems For Optical Computing

After human society entered the information era,the growth speed of information and the needs for information processing has reached unprecedented heights.Meanwhile,the deep fusion of artificial intelligence(AI)with various fields also puts forward urgent requirements for hardware computing power to support AI applications.Traditional electronic computing systems have been difficult to meet increasingly growing needs for information processing and computing power.Due to the benefits of optical systems’ high speed,high bandwidth,high parallelism,low loss,and low crosstalk,optical computing has recently been a research hotspot to address the traditional electronic computing systems’ speed and power consumption bottlenecks.At the same time,the integration of optical computing systems has become an important development trend.Integrated photonic devices and systems can significantly increase the processing speed and energy efficiency of optical computing systems because of the high integration and convenient scaleup.However,the size of most traditional integrated photonic devices is difficult to reduce due to the diffraction limit.In addition,traditional integrated photonic devices have the shortcoming of weak universality and have difficulties in designing high-integration and complex devices.On the other hand,reconfigurable integrated optical computing systems also suffer from the complex reconfiguration program and slow configuration speed.All these issues limit the development of integrated optical computing systems.With regards to the development requirements of integrated photonic devices with high integration and strong universality,as well as the problems in the configuration of reconfigurable integrated optical computing systems,this thesis will conduct related research on the three aspects of high-integration basic functional devices,multifunctional complex computing units,and fast-reconstruction integrated computing systems.The main contents and innovative points of this thesis are as follows:1.A compact device architecture based on the metal-insulator-metal(MIM)waveguide and coding resonator is proposed in terms of highintegration basic functional devices.This is based on the special property of surface plasmon polaritons(SPPs),which can overcome the diffraction limit and further enhance the integration density.Benefiting from the strong constraint of SPPs and flexible manipulation of the coding structure,the device’s core size is only 0.4×0.4=0.16μm2.Compared to traditional integrated devices with dimensions of tens of micrometers,this device architecture can greatly improve the integration degree of photonic integrated circuits.In the spectral range of 800～1700 nm,four filtering functions,namely,narrowband band stop,broadband band stop,narrowband band pass,and broadband band pass are realized by inversely designing the coding resonator,which possess the mode coupling between the coding resonator and the transmission waveguide.Besides,by changing the central wavelength setting,this device can generate plasmon-induced transparency(PIT)effects with Q values of 34.86,44.26,36.54,and 58.13 at 1100 nm,1200 nm,1300 nm,and 1400 nm,respectively.Finally,the generation mechanisms of the filtering mode and PIT effect are theoretically analyzed using the coupled-mode theory and the transfermatrix method.This compact and flexible device architecture proposed in this thesis provides a universal scheme for the design of high-integration photonic devices for optical computing.2.In terms of multifunctional complex computing units,by using multiple input ports and multi-objective optimization algorithms,and improving the coupling method of the coding resonator and MIM waveguide,seven basic logical operations are implemented in a universal multiport device.At the footprint of 0.8 μm×1.1μm,the maximum extinction ratios of AND,OR,NOT,NAND,NOR,XNOR,and XOR logical operations are 10.15dB,57.54dB,43.25dB,20.76dB,10.42dB,24.04dB,and 27.74dB,respectively,achieving advanced performance in related devices.By changing the operating wavelength setting,the device can achieve XOR operations at wavelengths of 1070nm,1190nm,1430nm,and 1550nm,with maximum extinction ratios of 28.62dB,18.36dB,26.32dB,and 26.78dB,respectively.Unlike most previously reported logic gates of the same type,this multiport device not only implements all types of complex logical operations under a universal design,but also eliminates the condition of cascading,providing a new approach to reducing the design difficulty of complex functional devices.3.In terms of fast-reconstruction integrated computing systems,this thesis has investigated the integrated and reconfigurable optical neural network computing architecture based on the silicon-based Mach-Zehnder interferometer(MZI)topology network.To solve the problems of being sensitive to manufacturing errors,complex calibration programs,and slow configuration speed in this architecture,a highly robust method using deep learning algorithms is proposed to rapidly configure the network.The simulation results show that the fidelity of this method can reach more than 0.95 when configuring low-dimensional MZI topology networks with manufacturing errors.Meanwhile,the configuration time(approximately 0.15 seconds)is several hundred times shorter than that of the method based on the forward propagation(approximately 100 seconds).In the MNIST handwritten digit recognition task,the proposed configuration method is applied in a 4×4 MZI topology network to configure convolution cores and complete the optical convolution calculation.The recognition accuracy of the whole optoelectronic hybrid convolution neural network is more than 97%,which is equivalent to that of the electric convolution neural network,and proves the effectiveness of this configuration method.This configuration method provides a new way to rapidly configure reconfigurable optical neural networks.