Silicon-based Fano Resonant Electro-optic Modulator For Photon Convolution Computin

The convolutional neural network for object recognition requires over 80% of computational resources for convolutional calculations.The convolutional implementation based on silicon photonics devices can effectively reduce computational power consumption and improve processing speed.Among them,the silicon optical modulator serves as the basic unit of computation.The Fano resonance effect can effectively enhance spectral resolution and,when combined with silicon optical modulators,can significantly reduce the driving power consumption of the modulators,providing a fundamental component for photonic neural network computation.Therefore,the main content of the paper is as follows:1.The photonic convolutional neural network was introduced.The characteristic equation of the basic mode of the optical waveguide is expounded.Theoretically,the principle and implementation of Fano resonance is analyzed,the basic theory and transmission characteristics of the microring resonator is studied,and the realization mechanism of the modulator of the three types of modulation schemes is summarized.There lay a theoretical foundation for subsequent device design.2.To enhance device reusability,a multi-port output Fano resonance modulator is designed.It can achieve four-port output adjustable Fano resonance with high extinction ratio and high spectral resolution.By using the transfer matrix method,the Mach-Zehnder interferometer-coupled micro-ring resonator structure is designed,and the ring coupling structure is introduced to enable Fano resonance with different performance at the four-port output.Simulation results show that the maximum extinction ratio of the device is 56.19 d B,and the highest slope is 2175.74 d B/nm.Due to errors introduced during the fabrication process,the experimentally measured maximum extinction ratio and slope are 26.09 d B and136.97 d B/nm,respectively,validating the feasibility of multi-port output Fano resonance modulators.3.To improve the energy efficiency of convolutional computations,a Fano resonance electro-optic modulator based on a T-shaped waveguide-coupled micro-ring resonator structure is designed.The influence of device structural parameters on its spectral characteristics is analyzed using the finite-time domain difference method,and the optimal device structural parameters are designed.The effects of doping ion concentration and doping structure on waveguide refractive index and loss modulation are analyzed,and the optimal doping section and doping concentration are selected.Compared with micro-ring modulators,the power consumption of Fano resonance modulators for achieving the same optical intensity modulation is only 0.07 times that of micro-ring modulators,greatly reducing power consumption.Theoretically,the modulation depth of Fano resonance modulators can reach 12.44 d B,and the insertion loss is only 0.41 d B.4.To verify the simulation design of the devices and convolutional computations,device fabrication,testing,and chip wafer production are performed.The fabrication process of silicon photonics passive devices is optimized,and Fano resonance devices are successfully fabricated,and the test results are analyzed.Fano resonance passive devices with the highest extinction ratio of 18.77 d B are fabricated based on the optimal structural parameters.The modulation depth of the fabricated Fano resonance modulator is 10.21 d B,and the slope is 17.246 d B/nm.When a modulation voltage of 3.4 Vpp is applied,the highest modulation efficiency can reach 100 pm/V.The convolutional multiplication operation is verified,demonstrating the great potential of Fano resonance modulators in photonic convolutional neural networks.