Design of W-band Multilayer LCP Substrate Via-holes Transition Driven by Machine Learning

Millimeter-wave circuit systems have become a hot research topic in the field of wireless communication due to their characteristics of large bandwidth, high speed, and low latency. In millimeter-wave multilayer circuit systems, via interconnect structure serves as a key unit for connecting components and transmission lines, and its performance directly determines the quality of the entire communication system. This article proposes a design method for W-band via interconnect structure based on the collaborative design of backpropagation neural network (BPNN) and genetic algorithm (GA). This method uses 140 sets of HFSS simulation data of via interconnect structures with different dimensions as training samples for machine learning, and obtains a BPNN model with an accuracy of over 99%. Meanwhile, the trained BPNN is embedded into the GA algorithm to optimize the structural size of the via interconnect structure toward low reflection coefficient. A prototype of via interconnect structure is designed and fabricated based on the four-layer liquid crystal polymer process. The test results show that the designed via interconnect structure exhibits S₁₁ less than -15 dB in the wide frequency band of 75 GHz ~100 GHz.