Re: 2n7000 spice models

From: Winfield Hill (hill_a_at_t_rowland-dotties-harvard-dot.s-edu)
Date: 03/28/05 

Date: 28 Mar 2005 07:02:31 -0800

Winfield Hill wrote...
> Jim Thompson wrote...
>>
>> Win, WHAT is important to you in power-FET modeling, subthreshold?
>> Maybe we can mutually determine the "fix" for the models?
>
> OK, here's what I wrote early in the discussion, last December 31st,

Winfield Hill wrote...

>> Well, then surely you'll want viable Spice MOSFET models, because
>> you'll be using the FETs throughout their useful linear range, and
>> the output-stage crossover region is critical.
>>
>> Sadly most standard Spice library VMOS models simply don't do the
>> subthreshold linear region. For example, see the 10-decade plots
>> on page 123 of our book. A jellybean 2n7000 is rather similar to
>> the VN01 that we show in figure 3.14, and certainly a proper Spice
>> model should be able to make that plot. But I'd be surprised if
>> your standard Spice libraries work properly below say 5 to 20mA,
>> which is not that far below the FET's maximum current. Keep in
>> mind that linear power FET circuitry always operates well below
>> the maximum rated FET switching current, to keep power dissipation
>> junction heating under control.
>
> With respect to the Id-vs-Vgs curves on page 123 of our book, and
> the g_m plots on page 132, the spice models should be able to show
> this performance. At subthreshold currents a FET acts very much
> like a transistor with respect to transconductance, etc., and when
> a power FET is used in linear audio amplifiers, e.g. in class AB,
> it may go through this region during each cycle. So in using Spice
> to determine distortion and evaluate various design configurations,
> surely it's necessary for the FET model to smoothly simulate the
> subthreshold region, and properly progress to the current-saturated
> regions that are normally accurately modeled. I'm going to go in
> to the lab and take some detailed 2n7000 measurements later today.

 I've completed a set of measurements over the current range of 1pA
 to 10mA, and posted the results on a.b.s.e. (Above 10mA I'll have
 to take pulsed measurements, which requires a different setup.) As
 anticipated, the data looks similar to the VN01 data, AoE page 123.

 As expected, the 2n7000 has an exponential gate-voltage-programmed
 drain current in the under-1mA subthreshold region, similar to a BJT.
 It follows the formula Id = Is exp(Vgs/Vt) and for my tested 2n7000
 Is = 0.1pA and Vt = 78mV. The transconductance is g_m = Id / 0.078
 (by comparison BJT transconductance, g_m = Id / 0.025, is 3x higher).

 My measurements show conduction starting at 200mV, 1nA at Vgs = 0.7V,
 and 1.0mA at Vgs = 1.8V, with a 2.303 Vt = 178mV/decade slope (see
 plot on a.b.s.e.). This compares to 58mV/decade for BJT transistors.

 Above 0.1mA the 2n7000 drain current (and transconductance) begins to
 fall below that predicted by the exponential formula. I tried improving
 the model by reducing Vgs by a term Id*Rs. This worked over 0.1 to 5mA
 with Rs = 15 ohms, but above 5mA the term overcorrects and the formula
 predicted too little current.

 Note, the spice model I tried earlier showed 0pA up to 2.4V, where it
 suddenly and dramatically soared to about 100nA, and it continued with
 excessive transconductance up to over 100uA. It required Vgs = 2.51V
 for 1mA, so it was *very badly* off the mark on all counts, and nearly
 useless for linear modeling. 

-- 
 Thanks,
    - Win