Characterization and Modeling of a Si–PolySi Microring Resonator with Application in Neuromorphic Computing

The objective of this thesis is to characterize and model an optical microring resonator (MRR) in the silicon-insulator-silicon capacitor platform (SISCAP) composed of an hybrid silicon–polysilicon (Si–PolySi) ring waveguides, here we investigate its potential for neuromorphic computing applications, specifically using the reservoir computing (RC) approach. The first part of the work focuses on the study and analysis of the non-linear optical behavior of the Si–PolySi microring resonator: at high optical power injection, two-photon absorption (TPA) generates free carriers, which modify the refractive index through free-carrier dispersion (FCD). This leads to a power-dependent shift of the resonance wavelength and a distortion of the transmission spectrum, revealing the strong non-linear response of the ring. The dynamics of the free carriers create a temporal memory effect in the ring, which can be exploited as the nonlinear node of a reservoir computing network. In this configuration, the response of the system to a given input bit depends on the residual carrier population generated by the previous bits, enabling the processing of temporal correlations within the input sequence. The experimental work includes a pump–probe setup, as it allows monitoring the carrier dynamics response of the device during operation in neuromorphic computing experiments. As in previous studies on fully silicon microrings, the objective is to verify whether the carrier dynamics in the Si–PolySi ring can provide the non-linear and memory responses necessary for RC tasks. The specific application investigated in this thesis is the one-bit XOR predictor. The main idea is to exploit the faster carrier dynamics in polysilicon, due to its shorter carrier lifetime compared to pure silicon, to achieve higher bitrates in temporal computing applications. However, laboratory measurements show that the prediction accuracy obtained from experimental traces is lower than that reported for full silicon rings in previous studies. At high bitrates, the ring response cannot follow the input accurately, as the trap-dominated dynamic of silicon prevails over the faster carrier response of polysilicon. Furthermore, the thermal transient (often neglected in theory) is actually not negligible, and the output response is distorted due to thermal effects. We therefore conclude that the two effects limit the performance of this device for reservoir computing applications. This work provides a comprehensive characterization and modeling of Si–PolySi microrings, highlighting the limitations imposed by thermal effects and offering insights for future improvements in high-speed photonic neuromorphic computing.