Tweaking spice models to generate device parameter variation

[Paul Burridge]

Hi,

I need to check the efficacy of a new FET biasing scheme to see how
tolerant it is to the wide parameter variations one gets with
real-world FETs. Hence, the usual spice model for a FET with its
invariable Vgs(off) and Idss isn't going to be of any use. If I change
the parameter "Vto" (which I believe is the same as Vgs(off) in the
real world from between say -2 to-6 volts, will that alone suffice to
emulate real world manufacturing variations? I mean, will tweaking Vto
alone also change the Idss to the appropriately meaningful value for a
FET's characteristic quadratic transconductance curve?

THanks,

p.
--

"What is now proved was once only imagin'd." - William Blake, 1793.

 

[Kevin Aylward]

Paul Burridge wrote:

Hi,

I need to check the efficacy of a new FET biasing scheme to see how
tolerant it is to the wide parameter variations one gets with
real-world FETs. Hence, the usual spice model for a FET with its
invariable Vgs(off) and Idss isn't going to be of any use. If I change
the parameter "Vto" (which I believe is the same as Vgs(off) in the
real world from between say -2 to-6 volts, will that alone suffice to
emulate real world manufacturing variations? I mean, will tweaking Vto
alone also change the Idss to the appropriately meaningful value for a
FET's characteristic quadratic transconductance curve?

No. You need to change Beta (or Kp), the transconductance as well. SS
already has built guestimates if you use its WC run facility, so you
don't have to ask these questions, usually...

For frequency response you would also have to change the capacitance
parameters, again, WC are already defaulted in SS

Kevin Aylward
salesEXTRACT@anasoft.co.uk
http://www.anasoft.co.uk
SuperSpice, a very affordable Mixed-Mode
Windows Simulator with Schematic Capture,
Waveform Display, FFT's and Filter Design.

 

[Paul Burridge]

On Tue, 07 Sep 2004 17:59:52 GMT, "Kevin Aylward"
<salesEXTRACT@anasoft.co.uk> wrote:

No. You need to change Beta (or Kp), the transconductance as well. SS
already has built guestimates if you use its WC run facility, so you
don't have to ask these questions, usually...

For frequency response you would also have to change the capacitance
parameters, again, WC are already defaulted in SS

Thanks, Kev. These runs are only at a fixed LF, BTW.
--

"What is now proved was once only imagin'd." - William Blake, 1793.

 

[Paul Burridge]

On Tue, 07 Sep 2004 17:59:52 GMT, "Kevin Aylward"
<salesEXTRACT@anasoft.co.uk> wrote:

Paul Burridge wrote:
Hi,

I need to check the efficacy of a new FET biasing scheme to see how
tolerant it is to the wide parameter variations one gets with
real-world FETs. Hence, the usual spice model for a FET with its
invariable Vgs(off) and Idss isn't going to be of any use. If I change
the parameter "Vto" (which I believe is the same as Vgs(off) in the
real world from between say -2 to-6 volts, will that alone suffice to
emulate real world manufacturing variations? I mean, will tweaking Vto
alone also change the Idss to the appropriately meaningful value for a
FET's characteristic quadratic transconductance curve?


No. You need to change Beta (or Kp), the transconductance as well. SS
already has built guestimates if you use its WC run facility, so you
don't have to ask these questions, usually...

I've just had a look, Kev. Would I be right in thinking one selects
the component/properties and then chooses between "weak, nominal or
strong" to get the MC parameter variations for a FET? That's what
seems to be happening, as I note the values change according to which
of the three selections I make? Does this accurately represent the
manufacturing spread one would expect to see in real-world devices? It
would save me having to cobble up my own probably inaccurate
guesstimates, you see.
--

"What is now proved was once only imagin'd." - William Blake, 1793.

 

[Kevin Aylward]

Paul Burridge wrote:

On Tue, 07 Sep 2004 17:59:52 GMT, "Kevin Aylward"
salesEXTRACT@anasoft.co.uk> wrote:

Paul Burridge wrote:
Hi,

I need to check the efficacy of a new FET biasing scheme to see how
tolerant it is to the wide parameter variations one gets with
real-world FETs. Hence, the usual spice model for a FET with its
invariable Vgs(off) and Idss isn't going to be of any use. If I
change the parameter "Vto" (which I believe is the same as Vgs(off)
in the real world from between say -2 to-6 volts, will that alone
suffice to emulate real world manufacturing variations? I mean,
will tweaking Vto alone also change the Idss to the appropriately
meaningful value for a FET's characteristic quadratic
transconductance curve?


No. You need to change Beta (or Kp), the transconductance as well. SS
already has built guestimates if you use its WC run facility, so you
don't have to ask these questions, usually...

I've just had a look, Kev. Would I be right in thinking one selects
the component/properties and then chooses between "weak, nominal or
strong" to get the MC parameter variations for a FET?

Yes. To do a forced run. In most cases, you just run with the nominal,
and use the Worst Case or MC runs to do the actual variations
automatically (blur "SS" button, WC reruns tab)

It will look for a specific model variations in the library
(basename_XN, basename_XW, basename_XS). If it cant find one, it will
use default multipliers that are set in \defaults\SuperSpice.lib to
create one. Once it creates one it will use that one always afterwards.
If you reset the default multipliers you will need to manually change
the auto created one, or delete it so that it will be re-created.

That's what
seems to be happening, as I note the values change according to which
of the three selections I make? Does this accurately represent the
manufacturing spread one would expect to see in real-world devices?

Well, "accurately" is open to debate. I just checked what I had
SuperSpice.lib, and for the jfets the default variation for Vto is
probably a bit low. Its set to 1.25 and 0.75 scale factors. A better
guesstimate is probably about 2 and 1/2. I must have rushed this a bit.

If you press the help on the WC setup page, you wil get a description of
what happens.

It
would save me having to cobble up my own probably inaccurate
guesstimates, you see.

I'll go back over my defaults to make sure that they are all reasonable.

Kevin Aylward
salesEXTRACT@anasoft.co.uk
http://www.anasoft.co.uk
SuperSpice, a very affordable Mixed-Mode
Windows Simulator with Schematic Capture,
Waveform Display, FFT's and Filter Design.

 

[Kevin Aylward]

Kevin Aylward wrote:

Paul Burridge wrote:


It
would save me having to cobble up my own probably inaccurate
guesstimates, you see.

I'll go back over my defaults to make sure that they are all
reasonable.

Well, I must admit I am rather embarrassed by this. There was another
discrepancy. The Vt multipliers were the wrong way for jfets. That is, a
strong model should have a low vt, weak a high vt. Not sure how this
crept in but I did have:
*these demo models may not be accurate

*worst case models only illustrate functionality

In the WC setup file (SuperSpice.lib, so I did had a cop out

I have now updated the download with a corrected version that has better
defaults for the jfets.

Kevin Aylward
salesEXTRACT@anasoft.co.uk
http://www.anasoft.co.uk
SuperSpice, a very affordable Mixed-Mode
Windows Simulator with Schematic Capture,
Waveform Display, FFT's and Filter Design.

 

[Paul Burridge]

On Fri, 10 Sep 2004 08:01:58 GMT, "Kevin Aylward"
<salesEXTRACT@anasoft.co.uk> wrote:

Well, I must admit I am rather embarrassed by this. There was another
discrepancy. The Vt multipliers were the wrong way for jfets. That is, a
strong model should have a low vt, weak a high vt. Not sure how this
crept in but I did have:
*these demo models may not be accurate

*worst case models only illustrate functionality

In the WC setup file (SuperSpice.lib, so I did had a cop out

Thanks, Kev. I'm grateful to you for looking into this for me before I
carried out any runs. It's reassuring to see that you've been open
about this problem and dealt with it swiftly instead of sweeping it
under the carpet as others might have been tempted to do. Respect is
due!

I have now updated the download with a corrected version that has better
defaults for the jfets.

Many thanks indeed. I'll give it a whirl this afternoon. I had noticed
the spreads of your hi-med-low parameters were quite close together -
pity the manufacturers can't make the real thing to your tolerances!



--

"What is now proved was once only imagin'd." - William Blake, 1793.