Model Testing

Requirements for a MOSFET Model in Circuit Simulation

(1) It should include most or all of the important physical effects in modern

MOSFETs.

(2) The model should meet the requirements for accuracy and continuity of the I-V equations and charge conservation.

(3) It should give accurate values and ensure the continuity (with respect to any terminal voltage) of all small signal quantities such as trans conductance gm, gmb, gds and all capacitances.

(4) It should ensure the continuity of gm/Id , an important quantity for analog circuit design, when Vgs is varied.

(5) It should give good results even when the device operates non-quasi-statically, or at least it should degrade gracefully for such operation, as frequency is increased.

(6) It should give accurate predictions for both thermal and 1/? noise in the triode and saturation regions.

(7) It should meet the above requirement over the weak? moderate and strong inversion regions, including Vb s≠0.

(8) It should meet all of the above requirements over the temperature range of interest.

(9) It should ensure the symmetry of model at Vds=0 if the device itself is symmetric.

(10) It should pass the Gummel slope ratio test and the treetop curve test.

(11) It should do all of the above for any combination of channel width and length.

(12) One set of model parameters should be sufficient for all device channel lengths and widths.

(13) The model should provide warning or stop the simulation when the model is used outside its limits of validity.

(14) It should have as few parameters as possible, and those parameters should be linked as closely as possible to the device and process parameters.

(15) It should be conducive to an efficient parameter extraction method.

(16) It should be compact and computationally efficient.

Benchmark Tests

Two categories test: qualitative test and quantitative test

Qualitative tests: to check the general behavior of a model without comparison to the experimental data.

Quantitative tests: to check the accuracy and scalability of the model against measured data.

Qualitative tests:

1. Triode-to-saturation characteristics (around Vth ) of Ids and gds
2. Triode-to-saturation characteristics (in strong inversion) of Ids and gds
3. Strong inversion characteristics in a linear plot of Id and gm
4. Subthreshold characteristics on a logarithmic scale, Log(Id ) and Log(gm)
5. gm /Id characteristics
6. Gummel symmetry test
7. Gummel slope ratio test
8. Gummel treetop curve test

还有七种tests.

Quantitative tests:

(1) Triode-saturation characteristics of Ids and gds versus Vds (around Vth)

(2) Triode-saturation characteristics of Ids and gds versus Vds in strong inversion

(3) Strong inversion characteristics of Ids and gm in both linear and saturation

regions

(4) Subthreshold characteristics of Log(Ids ) and Log( gm) in both linear and

saturation regions

(5) gm/ Ids characteristics at different Vds, and different Vbs

(6) Characteristics of Vth versus channel length at different Vbs

(7) Characteristics of Vth versus channel width at different Vb s

(8) Saturation current Idsat versus channel length at different Vg s

Helpful Hints

1. Summary of the BSIM3v3 model test results.

(1) The BSIM3v3 I-V model passes the qualitative tests discussed above and

shows smooth transitions from the triode to saturation regions, and from the

subthreshold to strong inversion regions. No negative conductance, kinks,

glitches or discontinuities are observed.

(2) It passes the treetop curve and the Gummel slope ratio tests, and demonstrates

model symmetry at the first derivative level. These results validate the

physics basis of the model in both the strong inversion and subthreshold

regions.

(3) It models the current and transconductances of the devices accurately, and

has good scalability over a wide device geometry range.

(4) It has been tested with the measured characteristics of devices from different

sources [12.20, 12.21, 12.22].

(5) The model shows good temperature dependence up to 125°C as demonstrated

in Chapter 9.

2. Understanding some limitations and shortcomings of the present

BSIM3v3 model

As discussed above, the BSIM3v3 model is good for both digital and analog

applications. However, it still can be improved to meet even more strict

requirements. We discuss some shortcomings of the present model to make

users aware of its limitations.

(1) The I-V model ensures the continuity and symmetry at Vds=0 at the first

derivative level, but fails at the second derivative level [12.27].

(2) The C-V model is not symmetric at V ds=0 as we have shown in Chapter 5.

(3) The bias-dependent source/drain series resistance is treated as a virtual

parameter to derive the analytical model by assuming that the device is symmetric.

This makes it very difficult or impossible to simulate an asymmetric

device. Separate source and drain series resistances should be used in the

model.

(4) It has been found that the velocity saturation and hot carrier effects can

influence the thermal noise characteristics significantly in short channel

devices. The present model accounts for velocity saturation only in an empirical

manner and should be enhanced to improve the accuracy of the thermal

noise model.

(5) The temperature dependence of impact ionization is not included in the

present model.

(6) The present model needs to be improved to simulate the high frequency

behaviors for RF applications because it does not include the influence of

some parasitics such as the gate resistance and the substrate resistances.

3. Additional benchmark tests to validate a model

The benchmark tests discussed in this chapter can be considered the basic

tests for validating a model. They check some salient properties of the model.

However, more benchmark tests need to be developed. For example, benchmark

tests to examine the harmonic distortion behavior of the model are

needed.

Bsim3 学习笔记11