Enabling Wavelength-Dependent Adjoint-Based Methods for Process Variation Sensitivity Analysis in Silicon Photonics

To reach its potential as an emerging technology platform, integrated silicon photonics needs accompanying design-for-manufacturability (DFM) methods and tools to assist the design of silicon photonic devices and circuits. Here, we explore spatial sampling in adjoint-based methods for analysis of the sensitivity of photonic components against key fabrication process variations. We apply and test these spatial sampling adjoint analysis methods to examine the impact of line edge roughness (LER) on a passive Y-branch at a given operating wavelength, achieving about 3% relative error. We extend the approach to also study LER variation sensitivity across a range of wavelengths and validate our results with ensemble virtual fabrication and FDTD simulations. The adjoint sensitivity and variance estimation of Y-branch transmission imbalance is seen to be highly efficient in comparison to direct ensemble simulation, with consistent results in the 95% confidence interval.