Appropriate Device Models

Thread Starter

Papabravo

Joined Feb 24, 2006
20,992
There are numerous discussions on the modeling of electronic devices. It might be convenient to believe that devices have unique models. Evidence suggests that this is not the case. When it comes to MOSFETS we start with the Shichman-Hodes model (level 1) and according to the following:

https://docslib.org/doc/11258428/hspice-elements-and-device-models-manual

which in 600+ pages documents more than a few MOSFET models.

When it comes to the BJT and more generally any 2-port network, there are multiple models used to describe the behavior including, h-parameters, g-paramters (inverse h), y-parameters, and z-parameters.

Given these examples, it seems unlikely that any device will have a unique model that is both valid and clearly superior in all cases.

The pedagogical question of how to organize and teach all of this material probably does not have a unique answer either.
 

WBahn

Joined Mar 31, 2012
29,850
The pedagogical question of how to organize and teach all of this material probably does not have a unique answer either.
It doesn't. As with nearly all questions in engineering contexts, the best answer is, "It depends."

The behavior of even the simplest devices, like a resistor, is actually very complex in reality. In addition to parasitics like lead wire inductance and effective parallel capacitance, there are temperature effects and the resistance is also generally a function of voltage and current as well. And that's just for a resistor! On top of everything else, there's the statistical nature in all of these parameters, not only in absolute terms, but in terms of mismatch between instances.

For most people, almost all of these have no relevance. Some people might need to account for the tolerance of the nominal value and others might need to account for temperature variations, but few are going to care about the change in resistance as a function of voltage, for instance. But some do. Models are developed with the needs of the users in mind and few model makers are going to go to the trouble to incorporate things that hardly anyone needs, if for no other reason than that the device manufacturers have likely not ever measured it for the same reason.

When I was first designing on the IBM 130 nm process, I had to delve into the transistor model (I don't recall why) and discovered that every transistor was instantiated as a subcircuit with over three-hundred components. At least that explained the painfully slow simulations! But those models where characterized to the nines and were extremely accurate.
 

Thread Starter

Papabravo

Joined Feb 24, 2006
20,992
It doesn't. As with nearly all questions in engineering contexts, the best answer is, "It depends."

The behavior of even the simplest devices, like a resistor, is actually very complex in reality. In addition to parasitics like lead wire inductance and effective parallel capacitance, there are temperature effects, and the resistance is also generally a function of voltage and current as well. And that's just for a resistor! On top of everything else, there's the statistical nature in all of these parameters, not only in absolute terms, but in terms of mismatch between instances.

For most people, almost all of these have no relevance. Some people might need to account for the tolerance of the nominal value and others might need to account for temperature variations, but few are going to care about the change in resistance as a function of voltage, for instance. But some do. Models are developed with the needs of the users in mind and few model makers are going to go to the trouble to incorporate things that hardly anyone needs, if for no other reason than that the device manufacturers have likely not ever measured it for the same reason.

When I was first designing on the IBM 130 nm process, I had to delve into the transistor model (I don't recall why) and discovered that every transistor was instantiated as a subcircuit with over three-hundred components. At least that explained the painfully slow simulations! But those models were characterized to the nines and were extremely accurate.
I myself, in recently learning about MOSFETS on a chip, encountered channel-length modulation. It was a totally new concept since I never had to design stuff that went on a chip. I don't know if there is an easy way to convey this stuff at an elementary level. The author (Erik Brunn) did a decent job of showing that the level-1 SH model was sufficient for initial hand calculations. These calculations were good to have when doing simulation with higher level models, like BSIM-3 (level 49).

Having a spice simulator makes this kind of learning a good deal easier. I don't think I could be on a chip design team tomorrow, but at least I could understand the discussion.
 
Top