MOSEFT Modelling

Joined: 18 Aug 2009 Posts: 14

You allow the user to define a slope 'S' which defines the current gain (transconductance) of the mosfet when it is operating in saturation. However, in a real device this slope varies with the gate voltage (i.e. the current is a function of (Vgs-Vt)^2). I think that a more accurate model would allow you to define the slope 'S' in A/V^2:

I = S*(Vgs-Vth)^2

Where S is either found in a data sheet or from the material properties of the device:

S = un*Cox*W/L/2

I realize that I can make this model with a PWL voltage controlled current source but I'm not sure if that type of model would be slower.[/img]

Site Admin Joined: 15 Jan 2009 Posts: 25

Yes, that's exactly what I'm thinking about now: if it makes sense to add more complex FET model with automatic PWL approximation. Probably yes.

By the way, in 1.34 you can approximate arbitrary function in PWL window, as we talked some time ago.

Joined: 18 Aug 2009 Posts: 14

Yes, I noticed the function approximation feature. Very nice!

By the way, GaN transistors (which International Rectifier is starting to release) have negative thresholds (normally ON devices). Does the NL5 FET model have any problems accepting negative voltage thresholds?

Site Admin Joined: 15 Jan 2009 Posts: 25

darko wrote:

Does the NL5 FET model have any problems accepting negative voltage thresholds?

No problem. NL=No Limits Smile

Joined: 18 Aug 2009 Posts: 14

Great.

By the way, you should consider starting a section on you web site for user contributions so that registered contributors can add NL5 models of standard parts as well as documentation on how to use the simulator in difference ways. I would happily contribute a some of my own work.

Site Admin Joined: 15 Jan 2009 Posts: 25

Actually, this forum was supposed to be a place for such "exchange of knowledge". However, you can see that there is no much activity here. Why? There are three possible answers.

1) Nobody's using NL. This is not true: in addition to many licensed users there are a lot of people who are using NL without license, with <=20 components. I can't track how many, but I can see that with every new build there is a spike of downloads from the same IPs. So, NL is being used.

2) Users do have problems but do not ask for help. I believe this is true for many users. Most of my friends will try to solve the problem or find workaround, rather than calling customer support (which is usually time-consuming and useless anyway...). What is funny: they can easily call me or even stop by and ask, but many of them do not do that. When I ask them why, they say: don't know, just used to do it that way...

3) Users do not have problems at all. This is the most favorable answer for me, and in fact I'm going to believe it. The whole idea of ideal components is to make simulation and troubleshooting simple, clear, and transparent. If anything does not work right, or as expected, a cause of a problem can be easily identified due to very simple models.

Thank you.