Machine Learning Aided Prediction of Fabrication Uncertainties in Integrated Multi-Ring Filters

Abstract

We propose a machine learning-based framework to predict the fabrication uncertainty and evaluate the effective-index shift in multi-ring integrated filtering elements. Excellent results are achieved in predicting each ring’s effective-index shift.

Thermally-controlled transparency in multi-ring cascaded filters

“…As previously stated MRR-based device require calibration and tuning setups to compensate for their intrinsic fabrication uncertainty sensitivity. Although uncertainty-aware design techniques can be considered to minimise this effect, 5 as well as advanced models can be trained to predict the individual ring shift based on their output, 6 the most basic tuning setup still requires an independent control mechanism for each MRR. We assume thermal control based on micro-heaters, as it represents one of the most common and available solutions, although the result can be replicated with any tuning method that introduces a modulation of the effective index.…”