A method for a reliable characterization of the small-signal equivalent circuit and the noise model of Heterojunction Bipolar Transistors (HBTs) is presented. It allows the device equivalent circuit elements (in T-topology) and its noise parameters (NPs) to be extracted simultaneously, using only the measurements of its S-parameters and its noise figure (measured for a well-matched impedance). The procedure is based on a simultaneous estimation of the device S-parameters and noise figure, by fitting to the corresponding measurements. The NPs estimated from the device model are compared to the NPs estimated from the measured noise figure, providing an additional term in the error function to be minimized that guarantees physical results. Thus, the error function is composed by three terms: the root-sum of squares (RSS) of the differences between measured and estimated S-parameters, noise figure and NPs, respectively. Experimental verification of the extraction of the equivalent circuit elements and NPs of an HBT, up to 8 GHz, are presented, and the NPs are compared to those measured with an independent (tuner-based) method.
In this work we present a method to characterize broadband noise circuit-models of on-wafer microwave noise sources. The models are used to estimate the device noise temperature, and therefore to characterize its Excess Noise Ratio (ENR). Two types of devices are considered: a cold-FET (Vds=0V) with the gate reverse-biased, and an unmatched avalanche noise diode. As a first step, a noise analysis is performed from noisy networks theory to derive an expression for the device output noise-current spectral density as a function of the intrinsic noise sources. Then, using the device measured reflection coefficient (or S-parameters in the case of a cold-FET) and measured noise temperatures, a regression technique is applied for the best frequency-fit between the measured and estimated 'multiplier' factor M (for which a smooth frequency dependence is assumed) associated to the intrinsic noise sources, thus reducing the measurement uncertainty. The resulting estimation of the device ENR features a sensible reduction in the measured 'ripple', without loss of the inherent 'slow' frequency variations due to variations of the device output impedance. As an application, the characterized devices are used as on-wafer noise sources to fully calibrate noise measurement systems.