Large-scale programmable silicon photonics for quantum and classical machine learning

Abstract

Photonic technologies provide many unique physical advantages including ultra-high bandwidths, energy-efficient operations, and low coupling to environmental noise. Furthermore, recent advances in foundry-based manufacturing platforms have enabled the emerging field of integrated systems photonics. In contrast to their bulk optics counterparts, these systems can co-integrate dense ensembles of active photonic and electronic components on a single wafer with high phase stability and small device footprints. Initial demonstrations of each element in the integrated photonics stack—sources, processors, and detectors—motivate the development of wafer-scale photonic integrated circuit implementations, which are poised to form a key building block for fundamental advancements in computing, communications, and sensing.

The first part of this thesis will discuss the development and early system-level demonstrations of linear programmable nanophotonic processors in the silicon-on-insulator platform for applications in quantum and classical machine learning and information processing. Using our developed processor architecture, we then present a nanophotonic Ising sampler for noise-assisted combinatorial optimization. Subsequently, we present a novel, foundry-compatible platform for integrating telecommunication-wavelength artificial atom quantum emitters directly in silicon photonic circuits. Finally, we report a capacity analysis of a structured interferometric receiver implemented with a silicon photonic processor for detection of optical signals in photon-sparse communication links.