Low-frequency noise characterization of silicon-germanium resistors and devices

Abstract

The main topic of this thesis is experimental low-frequency electrical noise characterization of semiconductor devices. In particular, we concentrate on applications of the silicon-germanium alloy (SiGe). Low-frequency electrical noise is a sensitive measure of defects and non-idealities in semiconductor devices, which directly or indirectly impact device performance and reliability. Thus, it is of prime importance to be able to characterize the noise in semiconductor devices. We compare the low-frequency noise from poly-crystalline silicon-germanium thin film resistors with different germanium content, film thickness and doping level. The noise level decreases with increasing doping density. We find that the germanium content and film thickness have little influence on the noise level. The noise was found to stem from mobility fluctuations in the depletion region of the grains. We compare the low-frequency noise of silicon based field-effect transistors with poly-crystalline gates, made from silicon and silicon-germanium. The output noise level for N-MOSFETs is independent of the gate material, whereas for P-MOSFETs the silicon-germanium gate material results in lower noise. Analysis of fluctuating physical quantities, points towards mobility fluctuations for P-MOS, and number fluctuations for N-MOS. We present results from measurement of the low-frequency electrical noise in Al- GaInP QuantumWell Lasers. Experimental evidence of a connection between the noise and device reliability is found, and hence, low-frequency noise measurements can be used as a non-destructive reliability indicator for laser diodes. The low-frequency noise in state-of-the-art silicon-germanium Heterojunction Bipolar Transistors (HBTs) is explored. Device geometrical down-scaling induces a deviceto- device noise variation, caused by small sets of noise generating traps, that are different from device to device. We use proton irradiation to introduce additional traps, and find that it can reduce the noise variation without increasing the noise level significantly. Aggressive down-scaling normally results in higher low-frequency noise. However, we find that the latest generation of SiGe HBTs (> 200 GHz) breaks this trend, and only a residual background noise remains, resulting in record values of low-frequency noise level and noise corner frequency. We present, and apply, recent statistical tools to probe for non-linear coupling between frequency components in a noise signal. These tools are applied to low-frequency noise time series with Random Telegraph Signal (RTS) noise from small geometry SiGe HBTs. The noise in small HBTs is shown to be non-Gaussian and non-linear. The nonlinearity is shown to originate from the RTS component of the noise.