4.4 Methodology Overview

The 2D extraction procedure is described using an N-channel MOSFET. The same procedure can be applied to a P-channel device.

• Input parameters determination: Several important parameters values are obtained by independent experimental means. The oxide thickness () is first determined using one capacitance measurement in the accumulation region as suggested in [87]. The polysilicon gate length () and polysilicon gate concentration () are then extracted by matching experimental gate-to-channel quasi-static capacitances () and   simulated results that take into account the polysilicon depletion effect [42][33]. Finally, the source/drain diode donor profile is measured directly using a 1D direct technique such as secondary ion mass spectrometry (SIMS).  
• Starting guess 2D profile: An initial TPS representation   is generated by combining 1D channel and source/drain profiles assuming a subdifusion factor for the donors rotation that will result in the measured effective channel length. The 1D profiles are determined by inverse modeling from deep depletion MOS and source/drain diode reverse junction capacitance respectively. It is noted that a good initial guess is important to avoid the trapping of the solution in a local minimum and to limit the number of iteration before converging to the solution.  
• 2D extraction: Finally, an optimization to extract the TPS coefficients of the doping for both acceptors and donors is performed. The nonlinear least-squares solver is used to minimize the fit   criterion between experimental and simulated gate and source/drain diode capacitance values.

The method was applied to data collected from devices fabricated using a retrograde n-well, salicided dual-gate CMOS process [44][43].   All experimental C-V characteristics were obtained using an HP4145B parameter   analyzer and an HP 4275A LCR meter. The measurement frequency of the HP 4275A LCR meter was set to 100 kHz. The resolution of the system is around 0.1 fF. In order to reduce the noise level in the measured results, the experimental data for sub 0.5m devices were averaged for several measurements. As a result, the actual resolution of experimental data is better than 0.1 fF.