Designing integrated photonic devices using artificial neural networks

Abstract

We develop and experimentally validate a novel artificial neural network (ANN) design framework for silicon photonics devices that is both practical and intuitive. As case studies, we train ANNs to model both strip waveguides and chirped Bragg gratings using a small number of simple input and output parameters relevant to designers. Once trained, the ANNs decrease the computational cost relative to traditional design methodologies by more than 4 orders of magnitude. To illustrate the power of our new design paradigm, we develop and demonstrate both forward and inverse design tools enabled by the ANN. We use these tools to design and fabricate several integrated Bragg grating devices. The ANN's predictions match very well with the experimental measurements and do not require any post-fabrication training adjustments.

Grating Couplers on Silicon Photonics: Design Principles, Emerging Trends and Practical Issues

“…The inverse design approach is similar to optimizing a black box to achieve certain performance metrics, while the iterative optimization and performance verification are fully automated using predefined computer algorithms. Inverse design on GC has already been reported in several research works [ 70 , 71 , 88 , 127 , 128 , 129 , 130 , 131 ]. Most of these works offer an automated design approach for non-uniform 1D GC devices with design goals already addressed in previous works reviewed above, such as increased CE, broadband operation and polarization independence.…”

Deep learning: a new tool for photonic nanostructure design

“…Nanostructures and metadevices are beginning to play an important role in integrated photonics 71 besides the fact that silicon photonic devices 72 typically also contain features with sub-micron dimensions. 73,74 The use of nanoscale features in silicon photonics introduces a vulnerability to fabrication related variations and defects which need to be well quantied. Several recent reports in the literature have focused on the application of DL to design problems in integrated photonics.…”

Accelerating silicon photonic parameter extraction using artificial neural networks

“…We simulated over 100,000 grating configurations at 250 wavelength points from 1.45 µm to 1.65 µm resulting in approximately 26,000,000 training points. We swept through 10 different corrugation widths, 11 different ICBG lengths, and 961 different chirping patterns [16].…”