Inverse Design of Embedded Inductor with Hierarchical Invertible Neural Transport Net

Heterogeneous integration of voltage regulators in power delivery networks is a growing trend that employs embedded inductor as a key component in significantly improving the power distribution. In this work, we propose a neural network framework called the hierarchical invertible neural transport for the inverse design of an embedded inductor. With this invertible method, we obtain the probability distributions of the parameters of the embedded inductor design space that most likely satisfy the desired specifications. We also learn the impedance response for free in the forward design. In the forward design, our results show a 2.14% normalized mean square error when compared with the output response of a fullwave EM simulator.