A genetic algorithm for generating RF circuit models from calibrated broadband measurements

Abstract

A genetic algorithm has been developed to generate circuit models from calibrated broadband measurements of coaxial and on-wafer devices under test (DUTs). The algorithm randomly builds a population of circuit models from lumped as well as distributed elements and compares the simulated scattering parameters for those models with calibrated measurements. The genetic approach iteratively culls the circuit models that agree most closely with calibrated measurements. In order to "test-drive" this approach, the genetic algorithm was applied to calibrated measurements of several two-port 1.85 mm coaxial DUTs. The resulting circuit models are reasonable and their simulated response agrees closely with the measurements. Subsequently, the algorithm was applied to calibrated measurements of a two-port, on-wafer nanowire device. This algorithm automates exploration of potential circuit models for yet-to-be-modeled devices, including those that incorporate nanowires.

A robust correction model based neural network modeling framework for electromagnetic simulations and RF measurements

“…Typically, in order to accurately mimic a given device's nonlinear behavior, it is a practice to develop circuit models based on measurement data [7,8]. However, even if research infrastructure supports adequate measurements, the process of collecting measurement data proves to be exceedingly expensive.…”