Scalable time-domain uncertainty quantification for passive photonic circuits via dimension-reduction framework

The high refractive index contrast of silicon-on-insulator (SOI) platforms enables dense photonic integration but increases device sensitivity to manufacturing variations. To ensure reliable photonic integrated circuits (PICs), process tolerance analysis at the design stage is critical. This paper proposes a scalable time-domain uncertainty quantification method that combines exact dimension-reduction modeling (EDRM) with contribution-degree analysis (CDA) to construct a decomposed baseband augmented scattering matrix (D-BASM) model. The proposed approach significantly mitigates the curse of dimensionality inherent in traditional BASM-based models and requires far fewer sample points during the modeling process, thereby greatly enhancing computational efficiency. Comparative experiments on a fifth-order microring resonator filter and an MZI-based demultiplexer demonstrate its superior accuracy and efficiency compared to traditional BASM and Monte Carlo methods.