Stable HIPPO-Based Circuit Macro-Modeling

Behavioral modeling of analog circuits is an important step of the integrated circuit design flow. Indeed, closed-box behavior modeling allows users to replicate the behavior of circuit elements and devices without explicitly knowing the inner workings of the device. Prior works have automated the generation of behavioral models using machine learning (ML) at both the device and circuit level. More specifically, a recent work has used high-order polynomial projection operators (HIPPOs) to augment gated recurrent unit (GRU)-based macro-models. This new HIPPO-based model has been shown to outperform state-of-the-art GRU-based circuit macro-models. In this article, we introduce a new type of modified recurrent neural network (RNN) circuit macro-model that uses the HIPPO framework, called HIPPO-RNN. Additionally, we present a modified HIPPO-RNN (stable-HIPPO-RNN) model that is more suitable for enforcing input-to-state stability (ISS), and derive corresponding stability constraints. These constraints effectively guarantee ISS stability of the macro-model during transient simulation. We show the validity and superior performance of our macro-models on two circuit modeling examples.