Parameterized Surrogate Models of Microstrip Structures for Electromagnetic-Based Power Amplifier Design and Statistical Analysis

Integrated circuits operating in the millimetre-wave range require careful electromagnetic optimization of the passive networks to properly account for crosstalk and radiative effects, but also accurate and reliable statistical analysis to assess the impact of process induced variability on chip yield. These two requirements are typically computationally incompatible, since the simulation time required by electromagnetic analysis is too high to allow for multi-trial Monte Carlo statistical analysis. This work presents a parameterized surrogate approach to model the passive networks that allows for an efficient yet accurate electromagnetic-based variability analysis of microwave circuits. As a case study, we report the statistical analysis of a GaN/Si device at 28 GHz evaluating the impact of statistical variations of the matching network, under concurrent variation of two technological parameters, on device performance.