Abstract by Kyle Roth
Simulating optical waveguides using neural networks
Silicon photonics is an exciting alternative to current electronic computation and provides a medium for quantum computing. Simulating the movement of light through circuit components such as waveguides, ring resonators, and Bragg gratings requires computationally expensive use of Maxwell's equations, making circuit design and chip development a tedious process. Generally, objects like waveguide tapers are simulated by modeling cross-sectional segments separately and combining the results using eigenmode expansion. We seek to replace each segment model with a neural network trained on the parameters of the segment. Previous work has shown that neural networks can efficiently imitate the results from Maxwell's equations, and modeling segments using neural networks promises to bring computational efficiency to the task of simulating complex chip components.