Analysis of Photonic Circuit Losses with Machine Learning Techniques

TL;DR

This paper uses machine learning, especially linear regression, to analyze and improve waveguide losses in silicon nitride photonic circuits, providing interpretable insights into fabrication dependencies and loss reduction strategies.

Contribution

It demonstrates that simple linear regression with variable selection can effectively predict and interpret waveguide losses, outperforming more complex models in interpretability and accuracy.

Findings

Linear regression achieves low prediction error and high interpretability.

Identification of key fabrication parameters affecting losses.

Process improvements can significantly reduce waveguide losses.

Abstract

Low-loss waveguides enable efficient light delivery in photonic circuits, which are essential for high-speed optical communications and scalable implementations of photonic quantum technologies. We study the effects of several fabrication and experimental parameters on the waveguide losses of a silicon nitride integrated photonics platform using various machine learning techniques. Compared to more complex machine learning algorithms, our results show that a simple linear regression model with variable selection offers a lower prediction error with high interpretability. The high interpretability, along with our domain knowledge, led to unique identification of fabrication process dependencies on the final outcome. With these insights, we show that by improving the process flow, it is possible to improve the loss by mitigating the cause in a real experiment.