I've analyzed 4291 VDMOS models used in the wild (found on the internet over the years) and I found that 63 of them have negative Rd, 25 have Rd=0 and 4185 have positive Rd, the actual distribution is in the attachment.
Is the negative Rd error or is it possible for it to be negative?
Physically it is of course impossible. Numerically it can be ok, as Rd will be in series with the channel impedance and Rs etc.. It is 'wrong' if you can detect the negative Rd in a valid application of that device for which the model line was constructed.
I agree that a negative Rd does not inspire confidence in that particular model / modelmaker.
-marcel
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Can you explain your analysis more specific. Which models you have investigated? You used the ngspice vdmos model (.model N1 vdmos)? Is it the model parameter Rd you meant? Or you take the differential of some voltage over some current? In case you mean dVds/dIds in the output characteristic then it is possible to have negative output resistance by self-heating effect.
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
I've downloaded thousands of spice models from various websites. You know you search for model for BC547 and on the website is zip with 50 other models so I took it. Few years and 100MB later, I have collected 4000 random VDMOS models. The one I was actually using (IRLZ44N) had negative Rd which cought my attention. So I analyzed them all and found that 4000 have positive Rd, between 0-1ohm, but 63 of those 4000+ have negative. Now I want to know if it is error or if it is possible to be negative. It is RD parameter in VDMOS model. For example I found:
Interesting - was never aware of that. Don't know what they want do with that. Perhaps a special temperature behaviour.
Nevertheless - ngspice is cutting this and set the conductance to 0.
Which kind of model are used? Is it vdmos from LTspice or Pspice? With level parameter?
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
I've analyzed 4291 VDMOS models used in the wild (found on the internet over the years) and I found that 63 of them have negative Rd, 25 have Rd=0 and 4185 have positive Rd, the actual distribution is in the attachment.
Is the negative Rd error or is it possible for it to be negative?
Last edit: simpletk 2023-06-01
Physically it is of course impossible. Numerically it can be ok, as Rd will be in series with the channel impedance and Rs etc.. It is 'wrong' if you can detect the negative Rd in a valid application of that device for which the model line was constructed.
I agree that a negative Rd does not inspire confidence in that particular model / modelmaker.
-marcel
Can you explain your analysis more specific. Which models you have investigated? You used the ngspice vdmos model (.model N1 vdmos)? Is it the model parameter Rd you meant? Or you take the differential of some voltage over some current? In case you mean dVds/dIds in the output characteristic then it is possible to have negative output resistance by self-heating effect.
I've downloaded thousands of spice models from various websites. You know you search for model for BC547 and on the website is zip with 50 other models so I took it. Few years and 100MB later, I have collected 4000 random VDMOS models. The one I was actually using (IRLZ44N) had negative Rd which cought my attention. So I analyzed them all and found that 4000 have positive Rd, between 0-1ohm, but 63 of those 4000+ have negative. Now I want to know if it is error or if it is possible to be negative. It is RD parameter in VDMOS model. For example I found:
.model IRLZ44N vdmos VTO=2.13 RS=0.02356 KP=681.267 RD=-0.0041 ...
Interesting - was never aware of that. Don't know what they want do with that. Perhaps a special temperature behaviour.
Nevertheless - ngspice is cutting this and set the conductance to 0.
Which kind of model are used? Is it vdmos from LTspice or Pspice? With level parameter?
It is IRLZ44N in this file http://bordodynov.ltwiki.org/standard.mos.XVII.txt
The actual LTspice distribution have not this entries. I assume AD has sorted out the models as nonsense.
I imagine this negative RD is an outcome of automated model
fitting, left to itself. Humans don't so often apply nonphysical
values.
Although I have seen some of that.