While optics was traditionally considered an excellent high-bandwidth conduit of in- formation, the increased demand for bandwidth in modern applications makes it a great candidate for not just signal transmission but also signal processing. In addition, prolifer- ation of silicon photonics in data centers accelerated the progress in photonic technology and research in the last decade. The combination of these makes optics a viable candidate for signal processing applications today. This thesis mainly explores two optical signal pro- cessing applications that utilize the nonlinear properties of silicon photonic devices. The first chapter focuses on developing a thresholder-like transfer function using silicon pho- tonic devices for improving signal readout in a cryogenic radio frequency receiver. It cov- ers the challenges associated with current state-of-the-art receivers and reports experimen- tal implementations of five devices based on silicon and silicon nitride platforms that can boost the signal contrast of return-to-zero pulses. The following chapters focus on devel- oping photonic devices that mimic functional units of photonic neural networks (PNNs). We cover demonstration of all-optical programmable nonlinear activation functions for artificial photonic neurons based on integrated silicon and silicon nitride platforms. We then present design and simulations of a novel photonic spiking neuron, based on a hybrid graphene-on-silicon photonic cavity, which offers a more scalable way of implementing spiking neurons in photonics. Finally, we explore two other devices for PNNS, such as a novel synapse based on a nanophotonic cavity and a bipolar junction transistor for signal amplification in the electronic link of optoelectronic neurons.
Adviser: Paul Prucnal