Improving Fabrication Fidelity of Integrated Nanophotonic Devices Using Deep Learning

Next-generation integratednanophotonic device designs leverageadvanced optimization techniques such as inverse design and topologyoptimization, which achieve high performance and extreme miniaturizationby optimizing a massively complex design space enabled by small featuresizes. However, unless the optimization is heavily constrained, thegenerated small features are not reliably fabricated, leading to opticalperformance degradation. Even for simpler, conventional designs, fabrication-inducedperformance degradation still occurs. The degree of deviation fromthe original design depends not only on the size and shape of itsfeatures but also on the distribution of features and the surroundingenvironment, presenting a complex, proximity-dependent behavior. Withoutproprietary fabrication process specifications, design correctionscan only be made after calibrating fabrication runs take place. Inthis work, we introduce a general deep machine learning model thatautomatically corrects photonic device design layouts prior to firstfabrication. Only a small set of scanning electron microscopy imagesof engineered training features are required to create the deep learningmodel. By making corrections to the design layout, the fabricatedstructure more closely aligns with the original intended design andtherefore results in improved optical performance. Without modifyingthe nanofabrication process, adding significant computation in design,or requiring proprietary process specifications, we believe that ourmodel opens the door to new levels of reliability and performancein next-generation photonic circuits.

Automated summary

Next-generation integrated nanophotonic devices achieve high performance and extreme miniaturization through advanced optimization techniques. However, small features generated by these techniques are not reliably fabricated, leading to optical performance degradation. In this work, a deep machine learning model is introduced to automatically correct photonic device design layouts, improving optical performance without modifying the nanofabrication process or requiring proprietary process specifications.