We have investigated the influence of carbon concentration on the low frequency noise (LFN) of Si/SiGe:C
Heterojunction Bipolar Transistors (HBTs). The HBTs are supplied by ST-Microelectronics Crolles and are based on a
0.13 &mgr;m BiCMOS technology. Three types of transistors were studied; they only differ by the amount of carbon
incorporated. When carbon is incorporated, representative noise spectra of the input current spectral density, SiB, show
important generation-recombination (G-R) components, while no such components are observed in carbon free
transistors. When the 1/f noise component is unambiguously observed, the associated figure of merit KB has a very good
value close to 4.10-10 &mgr;m2. In this paper we focus on the analysis of the G-R components associated with the presence of the carbon. Most of the observed Lorentzians are associated with Random Telegraph Signal (RTS) noise. No RTS noise
is found in carbon free devices. The RTS noise appears to be due to electrically active defects formed by the addition of
carbon, typically observed for concentrations above the bulk solid solubility limit in silicon. The RTS noise, amplitude
&Dgr;IB and the mean pulse widths (tH, tL), are analyzed as a function of bias voltage and temperature. The RTS amplitude is
found to scale with the base current and to decrease exponentially with temperature, independently of the carbon
concentration. The mean pulse widths are found to decrease rapidly with bias voltage, as 1/exp(qVBE/kT) or stronger. Our results confirm that electrically active C-related defects are localized in the base-emitter junction, and the RTS amplitude is explained by a model based on voltage barrier height fluctuations across the base-emitter junction induced by trapped
carriers in the space charge region. The observed bias dependence of mean pulse widths seems to indicate that two
capture processes are involved, electron and hole capture. These C-related defects behave like recombination centers
with deep energy levels rather than electron or hole traps involving trapping-detrapping process.
We investigate the impact of body biasing on the low frequency noise (LFN) performances of NMOS transistors from a transistors 130 nm CMOS technology. The body-to-source voltage VBS was varied from - 0.5 to + 0.5 V for reverse and forward mode substrate biasing. A detailed electrical characterization was performed and the benefits of the body bias analysed in terms of current and maximum transconductance variations. Noise measurements were first performed at low drain bias VDS = 25 mV and VBS = 0 V in order to discuss the noise model. Results are in agreement with the carrier number fluctuation theory. Bulk bias dependence of the LFN was investigated at VDS = VDD = 1.2 V. Significant noise reduction is observed in the subthreshold regime when applying a forward body bias. In strong inversion, the noise level is found to be approximately independent of the substrate bias VBS.
An overview of the theoretical 1/f noise models is given. Analytical expressions showing the device geometry and bias dependence of 1/f noise in all conduction regime are summarized. Recent experimental studies on 1/f noise in MOS transistors are presented with special emphasis for PMOS from a 90 nm CMOS technology. Gate and drain noise sources are investigated. It is shown that in subthreshold regime drain current noise agrees with carrier number fluctuation model whereas in strong inversion the evolutions can be described by mobility fluctuation model. Gate current noise shows 1/f and white noise. White noise is very close to shot noise, and we have a quadratic variation of 1/f noise with gate current. Coherence measurements show that the increase of drain noise at high gate biases can be attributed to tunneling effects. Input-referred gate noise and the volume trap density can be used as figure of merit. Discrepancies with the ITRS roadmap are discussed.