Machine Learning Optimization of Surface-Normal Optical Modulators for SWIR Time-of-Flight 3-D Camera

Surface-normal optical modulators based on multiple quantum wells are attractive for an increasing number of applications, including photonic links such as on-chip optical interconnects. The design of such structures however is still based on intuition and experience rather than on a quantitative assessment of the device and system performance, due to the extreme complexity of the device behavior and the large number of design parameters involved. We developed a method for the systematic optimization of the modulator design, using a combination of analytical modeling and supervised machine learning. The global optimization is driven by an evolutionary algorithm, and the robustness of the final results is evaluated using variance-based sensitivity analysis. The optimization algorithm was tested on the case of time-of-flight three-dimensional camera (ranging) application, yielding two novel optimized designs which allow for a considerable improvement of the depth resolution of the system. Finally, we propose a figure of merit for comparing the modulation efficiency of surface-normal modulators.