The lack of temperature correlation in the forward biased diode V-I curve with the SPICE model appears to be a well know issue. The method to extract of N, IS, and R described in the article appears to be similar to the ‘Three-point I-V method’ shown on page 63 of the LTSpice wiki document SPICEdiodeModel.pdf (ltwiki.org).Richard Zarod
I’m not familiar with the LTSpice publication but took a brief look at it. In my brief look-through I did not see a page 63.
I have never seen any text documenting the poor aid that the XTI parameter is to the SPICE diode. It is barely worth even including a value to name. At least it moves the curves in the correct direction. I can’t help but wonder if a better approach would have been to translate the saturation current (IS) parameter linking it to the ambient temperature. Semiconductor theory tells us that the ambient temperature moves IS and if this could be hard linked to the ambient temperature it should give favorable results (I think).
My method is parameter extraction is nothing new, I have merely organized it into a useful fashion. All that we are doing is identifying the equations for two curves and the merging of the two. It is simple high school algebra.
What I find interesting is the derivation of the default values used in the various diode models. As an example, LTSpice uses an Onsemi 1N914. However, Onsemi publishes two PSPICE files (dated 2007), 1N914.mod and 1N914.lib, but N, IS, and R are not only different in each of these files, but the values do also not match the default values used by LTSpice.ibid
Many, many, many engineers have marveled at your observation over the years. It seems counter-intuitive because the arguments should at least resemble themselves across differing characterizations of the same component derived from identical source data. I have marveled at this since 1984 at my first use of SPICE 2.g.6. The numbers bear no resemblance but produce the same curves! I hate numbers and you should too. 🙂
There could still be considerable simulation error even when N, IS, and R are chosen to match the Onsemi ‘typical’ V-I temperature graphs. Note the maximum Vf is specified at 1.0V @ 10 mA at 25 deg C. This worst-case value is almost 200 mV higher than the typical value @ -40 deg C.ibid
Amazingly, I had another comment from a QEX reader to that same effect!!! This is what I call “…the hand is quicker than the eye.” Read my response to Gary on this.
It is interesting that the 1N4148 default N, IS, and R values do not match those in LTSpice or either of the Onsemi PSPICE files. This begs the question of how the default values are chosen in the various simulators.ibid
I know, I know! But you have to consider that industrial libraries do more than reproduce published curves. The main problem that they face is what my QEX article addresses. The SPICE diode is a poor performer for duplicating differing ambient temperatures.
The universal principle that the industry uses (it’s a good principle) is to make the diode work reasonable well across all specified temperatures. Consider that you don’t really need exacting accuracy with the diode simulation. Therefore, what the industry does (and I briefly touched on this in my QEX article) is characterize so that the characterization spreads across the entire ambient temperature spread. They add an offset moving all of the curves.
Many thanks, Rick, for you most excellent questions/observations.