4.9.3 Method Limitations

As other electrical profile extraction methods, the inverse modeling technique   suffers similar limitations:

• The method's capability is limited to the extraction of the electrically active net doping, not the chemical concentration of species atoms. Moreover, the required electrical measurements can only be taken after the completion of processing up to the metal layers. This limits the applicability of the method during process development.

• Uncertainties in the device geometrical structure, such as nonplanar surfaces, have a direct effect on the extracted profile. These can be resolved by independent determination of the structural information using transmission electron microscopy (TEM) imaging. Other types of input uncertainties, e.g. errors in the S/D SIMS profile, can also influence the extraction. An estimate of the accuracy of the extracted profiles subject to the inherent errors in the inputs can be obtained via a Monte Carlo simulation analysis [101].

• The extent of the device region where the profile can be determined depends on the range of measurement voltage, the doping level, and the device characteristics. For example, the capability of the gate to deplete the S/D-gate overlap region of carriers under accumulation bias, without causing the breakdown of the gate oxide, limits the resolution of the method for high concentration S/D profiles.

• The net doping term in Poisson's equation results in a correlation between the donor and acceptor coefficients. This issue is addressed by using the coupled iterative extraction scheme as described in the previous section.

• The dopant concentration values near the SiO/Si interface are correlated to the values of the gate work function and the oxide charge density along the channel used in solving Poisson's equation. The work function and charge density values determined during the deep depletion extraction step are presently fixed during the 2D extraction.

• Modeling assumptions in solving Poisson's equations are also a     source of uncertainty. For instance, there are two main approaches available in the literature to account for the anomalous threshold voltage variation in short-channel devices. The Reverse Short Channel Effect (RSCE) can be explained by either the existence of oxide charges or doping profile variation [85][48]. The chosen method to model this phenomenon influence the extracted profile near the interface. Fig. 4.17 compares the net doping along the surface in two cases: In the first one, oxide Gaussian charges were introduced in the simulation to model the RSCE. In the second, the oxide charge value was fixed to that determined from long channel device data. As seen, the two profiles are clearly different especially in the channel region. The inclusion of other types of electrical data can supply extra information that could eliminate this uncertainty. In particular, the use of subthreshold current data in the extraction might clarify the issue. In addition, the extraction of multiple length device profiles using each of the two assumptions can further show which method is more consistent in explaining the devices electrical behavior.

 
Figure 4.17: N-channel net doping at the SiO/Si interface extracted with (solid line) and without (dashed line) Gaussian charges to model RSCE.