Theoretical And Experimental Study On The Optical Vector Matrix Multiplier

Vector matrix multipliers (VMMs) can be used to perform Fourier transform and solve linear equations. They have wide applications in image processing, beam forming, radar detection, wireless communication, etc. With the fast development of micro-electronic technologies during the past decades, the VMM computation speed has been improved significantly. However the continuous improvement in performances of CMOS chips encounters bottlenecks in both physical size and manufacturing process. As well known, photonics has advantages of broadband, high speed, and low power consumption. It is easier to realize parallel processing. Hence, photonic technologies may provide a solution to solve the bottlenecks. Nowadays, researches on optical vector matrix multipliers (OVMMs) have become a hot topic and attracted more and more attention in recent years.In this paper, an implementation method of OVMM based on semiconductor optical amplifiers (SOAs) is proposed and verified experimentally. The purpose of choosing gain-adjustable SOAs rather than normal optical modulator is mainly due to its ability to improve computation precision by compensating the optical power loss, as well as to get a higher modulation rate since SOA is an active device and has a rising/falling time of 500ps in theory. In this thesis, based on a typical Stanford multiplier structure, a 2×2 OVMM module using optical fibers is designed, and its feasibility is verified via experiment.Theoretically, the OVMM can carry out two-dimensional parallel optical computing. This thesis investigates the computing principles of the OVMM and its applications in dot product, convolution, and Discrete Fourier Transform (DFT). Since in practice it is impossible to achieve computing devices with an infinitely-high dimension, we propose several solutions to reach higher dimension and precision with dimension-limited OVMM.The passive optical devices in the OVMM module proposed in this thesis, except for SOAs, can be implemented on a silica chip using planar lightwave circuit (PLC) technologies with improved performance and reduced cost. With the development in hybrid integration technologies, SOA can also be integrated with passive devices to form an OVMM chip. This is another reason why we choose SOAs as switching elements in this study. Therefore, we also explore the fundamental processes and techniques in PLC in the thesis. So far, we can deposit cladding and core layer films with a stress less than 200 Mpa after annealing, which is of significance in the implementation of PLC. We also get preliminary results for mask layer deposition, photolithography and etching. Further work on process optimization to get a better performance for PLC waveguides is required in the future study.