UC San Diego

This dissertation discusses how generative-type artificial neural network (ANNs) enables the real-time inverse design process of reconfigurable EM megastructures. Specifically, we demonstrate two designs: 1. a 2-D beamformer with rotatable dielectric slabs and 2. a conformal EM coating responding to free-form design goals on reflection pattern in a dynamic environment, the latter being an ultimate ambition of EM scattering control and nearly impossible to realize with conventional methods. These two examples demonstrate the superiority of ANN methods in dealing with high-nonlinear EM design problems requiring fast responses. Furthermore, the proposed data-driven free-form inverse-design approach can be accommodated for other science/engineering tasks.