Design of SIW filters in D-band using invertible neural nets

Inverse design of microwave circuits and electronic systems is challenging due to the ambiguous mapping relationship from output response to input design parameters. In this paper, we apply invertible neural networks (INN) for addressing inverse design problems, where geometrical parameters are estimated given the desired performance. In the proposed approach, the bi-directional inference processes are learnt using an INN. During the inverse process, the posterior distribution of the design parameters are reproduced based on the target response, which is especially advantageous in cases where similar design performances can be achieved with different parameter combinations. The effectiveness of the method is demonstrated using an inverse design example of substrate integrated waveguide (SIW) filter in D-band.