After years of intensive research effort, the design of RF integrated circuits in CMOS has now reached a wide acceptance for industrial designs. This is due to the high unity gain frequency and low-noise performance of today's deep sub micrometer MOS transistors combined with the high level of integration and the low cost of CMOS. Accurate MOS models are critical in order to reduce design cycles and to achieve first time success in implementation. Noise and small-signal modeling provide critical information in the design of RF circuits, especially for LNA blocks. The LNA is typically the first stage of a radio receiver and needs to provide sufficient gain while introducing the least noise possible. Unfortunately, available MOSFET compact models are intrinsically derived assuming constant (i.e. field independent) mobility model and uniform doping, as we will see, which can miss many interesting noise and small-signal phenomena. This thesis explores noise and small-signal behavior of MOSFETs, addressing many fundamental issues. The main focus of the thesis is to present very general analytic modeling methodology for noise and small-signal phenomena. First an analytical noise modeling methodology taking mobility degradation and lateral non-uniformity into account has been presented and equivalence between commonly used compact noise modeling methodologies in the presence of mobility degradation has been established. Using this analytic framework first thermal noise is studied. Introducing a new relation between electric field and noise temperature, charge-based expressions of thermal noise, which are suitable for implementation in commercial circuit simulators, are presented and used to study the effect of different physical factor on thermal noise. It is shown that noise properties of lateral non-uniform MOSFETs are considerably different from the prediction obtained with the conventional methods which, depending on the bias and doping profile, can overestimate the thermal noise by 2-3 orders of magnitude. This effect can be clearly explained by the fact that the presence of lateral non-uniformity makes the vector impedance field (the quantity responsible for noise propagation) position and bias dependent. The exploration of flicker noise starts with comparing the relation between two most widely used noise calculation methodologies, Langevin and flat band perturbation method. It is shown that the flatband perturbation technique actually underestimates the flicker noise at non-zero drain bias and a clear link between the two methodologies is established. Then a charge based expression of flicker noise PSD including field dependent mobility which, at present, is not considered by any of the compact models, is derived. Finally, low frequency noise under large signal periodic excitation is discussed and it is shown that, for almost all practical purpose, an averaged time constant and an averaged trap density can model the cyclo-stationarity of RTS and flicker noise respectively. All of the advanced compact MOSFET models (PSP, MM11, HISIM, BSIM, EKV etc.) dedicated to even sub 100 nm devices are entirely built on the Ward-Dutton (WD) partitioning, which was originally derived for long-channel uniformly doped MOSFET. It is shown that in a velocity saturated or laterally nonuniform device, in general, it may not be always possible to define a partitioning scheme. However, for small-signal operation one can always define a partitioning scheme but it is different from the original WD and depends on the mobility model and doping profile. All small-signal conductance and transconductance can be derived from small-signal partitioning function. At very high frequency, both conductance and capacitance becomes frequency dependent which can only be described using a NQS model. A small-signal NQS model including the effect of flied dependent mobility is presented. It is shown that that the inclusion of velocity saturation causes the magnitude of the gate-to-drain admittance to start to increase (instead of decreasing) above a certain frequency and its real part also starts to increase in the negative direction instead of becoming zero. Then it is briefly pointed out how NQS noise model of Prorret and Enz can be extended to include effect of field-dependent mobility. Finally, static, dynamic and noise properties of emerging double gate MOSFET is considered. It is predicted that although although DG MOSFET is superior compared to the bulk MOSFET in terms of DC and small-signal properties, its noise properties are actually inferior.
EPFL2007
