Integrated silicon photonics offers great potential for monolithic integrated photonic and electronic components using existing integrated circuit fabrication infrastructure. However, understanding of the impact of IC process variations on performance of photonic components remains limited. Methods for analysis that identify sensitivity of photonic components to the variety of process variations encountered during fabrication are crucial to enable viable design and manufacturing of silicon photonic systems.
We present the application of the adjoint method to predict the impact of different types of particle defects on silicon photonic circuits. The adjoint method is applied for both component and circuit level analysis to reduce computational cost, and shows good consistency with direct simulations. The results for complicated device components and small circuits are shown and discussed. The model and results can be used to help generate layout design rules and critical area extraction methods, and to assist silicon photonics designers in predicting and optimizing yield of complex silicon photonics devices and circuits.
Silicon photonics offers the ability to fabricate and integrate photonic and electronic components using existing integrated circuit fabrication infrastructure. Recent work seeks to understand the impact of IC process variations on performance of photonic components. In particular, methods for analysis that identify sensitivity of photonic components to process variations are crucial to enable viable design and manufacturing of silicon photonic systems.
We present two different and complementary methods for understanding the impact of geometric process variations on photonics components: ensemble statistical virtual fabrication simulations, and adjoint methods. These are utilized to identify the most sensitive regions of a Y-splitter photonic component to line edge roughness (LER) due to inherent lithography and etch process variations. In the ensemble approach, we simulate multiple instantiations with random LER applied to specific sections of the Y-splitter. This enables localization and quantification of LER impact on transmission, phase imbalance, and excess losses. These evaluations, however, come at the cost of many simulations. In adjoint sensitivity evaluation, only one or two simulations can identify regions most sensitive to LER. While first-order linear sensitivity is extracted, the adjoint has challenges in quantifying mean variation impacts. Both methods reveal that the Y-splitter is most sensitive to LER in the input taper, accounting for over 95% of the imbalance transmission. These two methods can be combined to quantify mean, variance, and sensitivity of photonic device components in the face of statistical variations. Incorporated into future photonic process design kits (PDKs), these analysis methods will help designers predict and optimize photonic component performance and yield.