Four-terminal modeling and parameter extraction of RF MOSFETs

Abstract

In this dissertation, we have examined a variety of examples of modeling and parameter extraction methods for RF MOSFET’s. Modeling and parameter extraction techniques popular in III-V FET modeling were reviewed and recent efforts to model the RF MOSFET and extract the model parameters were examined in light of the differences between the MOSFET and the III-V FET. We have found that the differences -the lossy silicon substrate and the four-terminal properties of a MOSFET -resulted in major difficulties of RF MOSFET modeling and parameter extraction. We have also discussed several works which attempt to successfully apply the established methodology for III-V FET’s to MOSFET’s with proper modifications, along with the shortcomings and limitations of this three-terminal modeling approach. Four-terminal modeling based on new equivalent circuits to address the high-frequency effects arising in a MOSFET was found to be very complicated and not practical for CAD applications, even without considering the substrate coupling terms. As a temporary alternative, the macro-modeling approach has been examined with various examples. A new physical small-signal equivalent circuit with very simple and accurate parameter extraction method for a four-terminal RF MOSFET model has been presented. The model is valid in strong inversion operation region. The parameters can be extracted directly from the real and imaginary parts of Y-parameters. It was found that the extracted parameters remain almost constant with frequency, verifying that the method is accurate and reliable. Without any complex fitting or optimization steps, the total modeling error was calculated to be only 0.89 %. We also showed that, without the intrinsic body node in the equivalent circuit, gate bias dependence of the substrate-signal-coupling effects cannot be modeled correctly. A new accurate four-terminal modeling approach associated with a novel parameter extraction method has been proposed. The model is v...