PhD Defence by Ali Cem

The last decade has brought about a revolution in artificial intelligence, with AI becoming an integral part of diverse industries and daily life. However, the rapid growth in AI applications is outpacing the capacities of traditional electronic computers. In response, photonic integrated circuits (PIICs) offer a promising avenue for innovation in the form of optical computers. Optical computers hold the potential to be faster and more energy-efficient than electronic computers. Yet, the complexities of modeling and controlling these circuits present ongoing challenges.

This thesis significantly advances the field of modeling photonic integrated circuits for optical computing through an exploration of various machine learning techniques. It begins by investigating the use of machine learning models designed for Mach-Zehnder interferometer-based optical matrix multipliers. The results reveal the substantial enhancement in modeling accuracy can be achieved through neural network models. Furthermore, the use of machine learning is extended to address the complex task of the thermal crosstalk compensation in programmable photonic processors. Using a microring resonator within a photonic processor, the study explores multiple approaches to model spectral shifts due to thermal cosstalk. Experimental results show that both linear regression models and analytical models are attractive options.

To summarize, this thesis signifies a significant step towards more efficient and powerful optical computers, where the integration of cutting-edge photonic technology with the capabilities of machine learning promises to shape the future of computing.