Neural Networks in circuit simulators

Artificial Neural Networks (ANN) are gaining attention in the semiconductor modeling area, as alternative to physical modeling of high speed devices. A fundamental issue when including ANN's in a circuit simulator is how to manage the time dependency. One elegant solution recently proposed is the Dynamic Neural Network concept, where neurons are instances of differential equations. In this work the dynamic approach and further variations has been compared with classical static ANN, applied to the modeling of high performance bipolar junction transistor.