Some of these definitions are not practically useful because they are not applicable to deep-sub-micron devices (1,3). The threshold-current definition 2 is attractive because of its simplicity and is being widely used by different groups inside companies. The numbers obtained are comparable only for a limited range of technology parameters. The same applies to the model-fit definition 4. One severe drawback of both methods 2,4 is that the numbers obtained by different organizations are almost for sure not comparable.E.1
The linear threshold voltage definition 5 does not suffers from any of the aforementioned drawbacks: depends on no other definitions and it applies to a wide range of devices. The only restriction of is the assumption of a small drain-source voltage, which precludes the investigation of DIBL phenomena.E.2 The definition of the drain-source voltage 6 was developed in this work and can be used to characterize the dependence of the threshold voltage. Note, that this definition relies entirely on the unique definition of , which in turn relies on that of . This makes the definition of as unique and universal as that of . The value is usually in the range of several for and can vary considerably with other device parameters. For a long-channel device in saturation and are identical (this follows directly from the definition of ). The necessary difference between and is that is independent of and is valid also for where would be infinite, whereas reflects the drain voltage dependence well for .
For this work, parameter extraction algorithms for and (definitions 5 and 6) were developed using polynomial regression analysis of the IV data. For both and (i.e., for ) the IV data are analyzed pointwise to find an appropriate range using heuristic algorithms. From these data subsets third-order polynomials are determined, from which and can be computed.