Fabrication uncertainties strongly affect the performance of photonic devices but are often overlooked during design. We combine unsupervised dimensionality reduction and Gaussian process regression for the stochastic, multi-objective design of photonic devices. Applied to silicon-on-insulator fiber grating couplers, the method reveals marked differences in yield and worst-case performance among 86 nominally equivalent designs and identifies Pareto fronts of robust devices.
Read the paper: Scientific Reports 14, 7162 (2024), doi: 10.1038/s41598-024-57315-4
Figure reproduced from P. Manfredi, A. Waqas, and D. Melati, Scientific Reports 14, 7162 (2024), under a CC BY 4.0 license.