Adjoint Methods and Inverse Modeling for Process Variation Analysis in Silicon Photonics

With the recent growing interest and development in the field of integrated photonics, manufacturing process variation studies are required to launch the many photonic applications to massive production and commercial use. Specifically, we need models that predict the impact of process variations on photonic circuits, extraction of the process variation information from actual fabrication measurements, so that we can use these models and information for future robust design that achieves high performance and yield given actual manufacturing limitations.

Current studies in the area of process variation in silicon photonics are emerging but still very limited. A general problem in both modeling the impact of variations and extraction from measurement is the cost of time, either the computational cost from simulation, or the cycle length of actual fabrication process. Therefore, here we explore some of the powerful tools that can shorten the simulation or experiment time, provide alternative, efficient, or economic approaches for process variation studies while still maintain accuracy. In particular, this thesis discusses two groups of methods for analysis: the adjoint methods for the modeling of variation impact, and inverse modeling for measurement data analysis.

Together, we hope these methods and techniques can become some generic tools that speed up the process variation study in silicon photonics, so that more robust design based on these models and information of process variation can emerge in the future, providing the road to high performance and yield for industry level production of integrated photonic applications.

Thesis committee:
Prof. Duane S. Boning (Thesis supervisor)
Prof. Jelena Notaros
Prof. Juejun (JJ) Hu
Prof. Luca Daniel