Fast noise prediction for process optimization using only standard DC and S-parameter measurements

This paper presents a method to select, amongst a series of devices differing on their process, the best device in terms of high-frequency noise characteristics, only from standard DC and S-parameter measurements and without the need of noise measurements. The method is developed and tested using synthetic input data and noise data generated with the MEXTRAM compact model and validated against on-wafer noise measurements on a series of devices with different process variations. A very simple equivalent circuit, with base, emitter and collector resistances as the only external components, is shown to be sufficient to rightfully sort devices in terms of their high-frequency noise performance. This allows to fully automate noise prediction procedure based on standard DC and S-parameter measurements without the need of manual data manipulation.