Silicon Photonics for Next-Generation Optical Processing and Communications

The rapid growth of transistor and fiber-optic technologies has brought unprecedented advancements in computing and telecommunications. While fiber-optics are poised to grow in speed this year, growth in computing power continues to slow down as the limits of transistor downscaling are approached. This has given rise to the growing field of silicon photonics, where the well-established microelectronic fabrication process is being adapted to create integrated devices that bridge the electronic and optical domains for large performance gains in data centers and high-performance computers. However, the inherent cost of switching between the two domains, and the fact that much of on-chip data transfer is still carried out by low-speed, high-power electronics, are problems that scale with the growing global demand for bandwidth. In this thesis, a novel optoelectronic computing logic architecture based on the silicon photonics platform is presented. This architecture combines the best of optical data transfer and electronic control for the highest level of throughput presented in the literature— presenting a promising option for future, "Beyond Moore" computing and processing. The main driver of this architecture is the silicon microdisk modulator, which achieves the best combination of energy efficiency, operating speed/bandwidth, compactness, and cost of all previously demonstrated optoelectronic processing devices. New configurations of the microdisk modulator are introduced to further improve the performance, not just for optical logic, but for all optical processing and communications applications. One of these applications is a novel on-chip optical communications circuit, presented in this thesis, that achieves ultrahigh-bandwidth-density through efficient microdisk-based design. As integrated photonic circuits like these become more complex, with hundreds or thousands of components on the chip, the design process becomes lengthier and more expensive. Even for small circuits, like those presented in this work, the time taken from design to characterization is delayed by having to design components secondary to the main devices (e.g., for coupling light into the chip). As the final part of this work, a machine learning based photonic device modeller is created to accelerate the photonic simulation and design process by multiple orders of magnitude, with minimal input from the designer.