Expansion and Virtualization of the Photonic Neural Network: building the framework for the future

Abstract

Neural networks are a powerful computational method that has grown in both use and complexity in recent years. These new methodologies have led to astounding com- putational power, at the expense of speed, low bandwidth, and size of these electrical networks. To mitigate some of these concerns, we turn to arti cial photonic networks that beat out electrical ones in all of these regards and more. Though these photonic networks have been on the rise, because of their cost and small scale applications, they have yet to transition from the research stage to large scale application. In order to do so, the individual parts of the network need to be expandable and easily duplicated for a larger number of neurons. In this thesis, I describe both the virtualiztion of and testing of a feed-forward layered neural network, as well as the design and implemen- tation of a current driver circuit that is key to expanding the network. In order for growth in the eld of photonic networks, the systems we use must be modular and easy to expand upwards.