Unified Time Domain Foundation Models for Hysteretic Passive Components

Passive power components such as magnetics, capacitors, and piezoelectric resonators have intrinsic characteristics that exhibit non-linear behaviors influenced by many factors such as temperature, dc-bias, memory effects, and waveform shapes. Traditional modeling methods are usually overly simplified and cannot fully capture their complex multi-scale, multi-physics behavior in the time domain. This paper presents the key concepts of time-domain foundation models for hysteretic transient behavior in passive power components. The key properties of a hypothetical time-domain foundation model may include: 1) frequency agnostic, 2) adaptive to universal time steps; and 3) a hypothesis that the system’s initial condition has an impact for a limited time horizon. A foundation model can be physics-driven or data-driven, or a hybrid of both. We present an example neural network that is simple, robust, and accurate as one implementation of the foundation modeling framework.