MPCDNet: A Deep Learning-Based Microwave Passive Component Design Network

To enrich design geometry and shorten the research and development cycle, this article proposes a deep learning-based microwave passive component design network (MPCDNet) for simultaneous and automatic modeling of varying structures and parameters. First, the modeling parameters are discretized to ensure consistency with the structural modeling so that the parameter and structure modeling process are represented as binary matrices. Then, the microwave passive components design is regarded as a sequence modeling task. Through a multilayer transformer model, the input sequence is deeply modeled to capture the complex mapping relationship between the input attributes and its response. Finally, the effectiveness of the proposed network is verified by differential transmission lines and differential couplers. The results demonstrate that the proposed network achieves rapid concurrent modeling of both structural and parametric characteristics of microwave passive components while delivering precise performance predictions, showing a <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$30\times $</tex-math> </inline-formula> faster modeling speed for differential transmission lines and <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$120\times $</tex-math> </inline-formula> acceleration for differential couplers compared with HFSS simulations while enabling structural diversity scalable to <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$2^{n}$</tex-math> </inline-formula> configurations (<italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">n</i> is the subcell number).