Using machine learning to optimize multi-junction photonic power converters

We have developed a machine learning empowered computational framework to facilitate design space exploration for optoelectronic devices. In this work, we apply dimensionality reduction and clustering machine learning algorithms to identify optimal ten-junction C-band photonic power converter (PPC) designs. We outline our framework, design optimization procedure, calibrated optoelectronic model, and experimental calibration devices. We report on top performing device designs for on-substrate and flat back-reflector architectures. We comment on the design sensitivity for these PPCs and on the applicability of dimensionality reduction and clustering algorithms to assist in optoelectronic device design.