Low-frequency noise power spectrum density characterizing of SiGe HBTs

Abstract

The main purposes of this thesis are the Low-Frequency Noise measurement of Silicon-Germanium Heterojunction Bipolar Transistors and its Power Spectrum Density Characterizing. The new generation 375 GHZ SiGe HBTs were measured in this work. We show that most of PSDs of the new generation SiGe HBTs have very ”bumpy” spectra which is contributed by GR noise sources. We investigated their basic characteristics of LFN such as the dominant noise source, base-current dependence, emitter geometrical scaling dependence and noise variation. They have similar LFN characteristic with the elder generation SiGe HBTs except for the emitter geometrical dependence.

The most important contribution of this work is that we particularly focused on developing a totally automatic mechanism to fit the Low-Frequency Noise Power Spectrum Density of SiGe HBTs so that we can use the magnitude of the fitting curve as the

low-frequency noise level at any frequency. A model based predictive and autonomous method was engaged for this purpose. This method offers the possibility that we can automatically predict the noise sources of transistors to get good initial fitting parameters in advance instead of finding each of them by eyes. Experiments with the fitting method shows that:

Always good fitting for most of the cases;

Accurately locating each noise source;

Sometimes meaningless fitting parameters but still good fitting.

Therefore, by using this method, we can find out how each noise source acts on the spectrum, which noise source dominates the spectrum, etc. And some careful interpretations will be presented based on this fitting procedure. Further, this method still

leaves large space to be extended, so it is a good basis for future work on fitting.