Machine-assisted design and stochastic analysis in integrated photonics

Integrated photonic devices are steadily making their way into many application fields including modern optical communication networks and advanced sensors. On the other hand, the design of photonic devices and circuits mostly remains a time-consuming process largely based on the designer experience. This limits the size and complexity of the parameter space that can be handled. Moreover, addressing the effect of manufacturing variability remains a fundamental challenge since small fabrication errors can have a significant impact on light propagation, especially in high-index-contrast platforms such as silicon-on-insulator. The analysis of this variability with conventional approaches (e.g. Monte Carlo) can become prohibitive due to the large number of required simulations. Recent advances in machine-assisted design methods are opening the possibility to vastly expand the number of design parameters, exploring novel functionalities and non-intuitive geometries. In this invited talk we discuss the use of machine learning methods for the design of integrated photonic devices. We show the existence of a large number of possible designs that are all equivalent with respect to a given primary design objective but with distinct properties in other performance criteria. We use pattern recognition to reveal their relationship and to reduce the dimensionality of the large design space by properly defining new design variables. Likewise, we show how efficient stochastic techniques allow a quick assessment of the performance robustness and the expected fabrication yield for each tentative device. We focus in particular on stochastic spectral methods that have been regarded as a promising alternative to the classical Monte Carlo method, achieving a considerable reduction of the simulation time. Together, the reduction in the design space dimensionality and efficient stochastic techniques allow for the integration of the fabrication tolerance considerations into the design process.