Spice simulation

Jump to Latest
thanh

thanh

Member
Joined 2004
lineup said:
Interesting question.
If one model is not correct/bad = does not show close to the curves of the transistor,
can one amplifier using this model/transistor be good/better in Simulation?

We have to say YES.
Because if the model is not correct,
it can be 'too good model'. Better than Reality.
Or 'too bad data' for this transistor.

The only thing we can do is to try to verify a model, if not sure it is close to real.
And there are only 2 good ways to do this:

1. Check against real measured data/curves taken from a REAL Transistor.

2. Check against some data/curves, that some other person has measured.
Most often those curves in DATASHEET.
Click to expand...


:) Thanks Lineup again!:)
 
B

Bob Cordell

Member
Joined 2006
cordellaudio.com
Edmond Stuart said:


Hi Bob,

The only company, AFAIK, that publishes sub-threshold data (down to 1uA) is NXP (formerly Philips), see attached example.

Cheers,
Edmond.
Click to expand...


Thanks Edmond. This is a much more complete datasheet. A couple of interesting things are notable from the low-Ids data.

First, the low-current rate of change of Vgs vs temperature as Ids is held constant is about 8.1 mV/C. This is not unreasonable, as we have observed numbers like 5-6 mV/C at higher currents on the order of 150 mA for vertical MOSFETs, where the TC has probably already begun its journey to zero at high currents.

Secondly, the curves of log Ids vs Vgs at low currents are essentially straight lines, emphasizing the fact that the device characteristic goes to an exponential characteristic (just like BJTs) at very low currents.

While the slope of such curves is about 60 mV per decade for BJTs, it is about 277 mV/decade for these devices in the sub-threshold region. This means that these devices asymptote to a transconductance that is 0.22 times that of a BJT for a given current.

Finally, Philips use of the term Trench-MOS seems a bit confusing, suggesting that these are Trench-FETs like those made by Siliconix. I don't think that is so, as other manufacturers just describe the IRFT640 as a third-geberation vertical MOSFET technology (read HEXFET). Over the years, a number of manufacturers have devised other geometries to behave equivalantly to the HEXFET, and this may just be the name for Philips' approach. On the other hand, I suppose that Philips may be using a process that legitimately could be called a Trench-FET process that happens to yield behavior like that of a third-generation HEXFET.

Most TrenchFETs that I am familiar with have quite low threshold voltages combined with extremely high gm and very low Rds-on.

Cheers,
Bob
 
lineup

lineup

Member
Joined 2005
as usually
Robert Cordell gives us his facts, as seen by his old experienced eyes

and what he finds is hard to argue about, by anyone
just for us to try to explore and use, if we will deal with these devices he tries to describe

thanks bob
for trying to share your knowledge, once again
 
KSTR

KSTR

Member
Joined 2007
2SC5171 2SA1930 SPICE models

.MODEL Q2SA1930 PNP(
+ IS=10.000E-15
+ BF=210
+ VAF=78
+ IKF=10.000E-3
+ XTB=1.5
+ BR=.1001
+ VAR=100
+ IKR=10.000E-3
+ ISC=10.000E-15
+ CJE=3.252E-12
+ CJC=63.196E-12
+ MJC=.33333
+ TF=83.239E-12
+ XTF=10
+ VTF=10
+ ITF=1)

.MODEL Q2SC5171 NPN(
+ IS=10.000E-15
+ BF=210
+ VAF=100
+ IKF=10.000E-3
+ XTB=1.5
+ BR=.1001
+ VAR=100
+ IKR=10.000E-3
+ ISC=10.000E-15
+ CJE=2.0000E-12
+ CJC=38.866E-12
+ MJC=.33333
+ TF=83.239E-12
+ XTF=10
+ VTF=10
+ ITF=1)
Click to expand...
I don't remember where I found those, though...
 
aparatusonitus

aparatusonitus

Member
Joined 2004
Try this one...

.SUBCKT LM336Z_2V5 2 1
* MODEL FORMAT: SPICE3
* External Node Designations
* node 1: Vref OUTPUT
* node 2: Ground (Common)
VSENS1 1 10 0
LXL 10 11 10
EREF 21 32 19 2 1.0
ISET 2 15 1.0m
DON1 15 16 DMOD1
VSENS2 16 19 0
DON3 15 27 DMOD1
RSS 17 27 1000
FX 2 25 VSENS1 1.0
EON3 13 2 25 2 1.0
VDROP3 17 13 -1
DX2 2 31 DMOD1
RZZ 31 30 1
FSENSE 2 30 VSENS1 1.0
EBACK 2 32 31 30 1e6
RFWD 1 33 30
DFWD 2 33 DMOD1
RDFW 2 33 1MEG
EYY 30 35 1 2 0.1
DYY 2 35 DMOD1
RYY 30 36 1
DX1 36 2 DMOD1
RSET 19 2 RSET 2490
.MODEL RSET R TC1=4e-05
RS1 11 12 0.26
RSX 10 11 0.001
LSR1 12 21 4.15394e-05
RS2 10 21 20
RXY 25 2 RXY 18700
.MODEL RXY R TC1=0.001
RSTEP 19 18 22411
CTRAN 18 2 1.43863e-11
.MODEL DMOD1 D
*-- SPICE3 DIODE MODEL DEFAULT PARAMETER
* VALUES ARE ASSUMED
* IS=1e-14 RS=0 N=1 TT=0 CJO=0
* VJ=1 M=0.5 EG=1.11 XTI=3 FC=0.5
* KF=0 AF=1 BV=inf IBV=1e-3 TNOM=27
.ENDS LM336Z_2V5
 
E

Elektor78

Member
Joined 2008
Question with pspice

Hi to all :)

In a simulation of class ab amplifier with pspice I need to draw
the "famous" graph Thd% Vs Power. (as that of the Audio Precision.. :D )
it should not be too much difficulty...
do I perhaps have to calculate the FFT for every value of vsin? :cannotbe:
can it be done this in pspice? :bawling:

Thanks :)
 
PB2

PB2

Member
Joined 2004
linkedin.com/in/petebasel
john curl said:
Sounds good to me, who is going to buy one of these testers? ;) Then we can get really good models for every device.
Click to expand...

I have a client who's looking into parameter extraction and I just spoke to Agilent; the ICCAPs software and large rack of complex test equipment starts at $500,000! We are not buying that any time soon ...

There is the old Tek 370 which has been out of production for some time with no up to date replacement.

Anyone know of an up to date replacement for the 370/371?

Tektronix and HP/Agilent have no suggestions for a current system.

Pete B.
 
PB2

PB2

Member
Joined 2004
linkedin.com/in/petebasel
Thanks Syn08, was just typing my notes and forgot that the folks at Agilent mentioned this unit that starts at $40,000:
http://www.home.agilent.com/agilent...g&pageMode=OV&pid=153734&ct=PRODUCT&id=153734

I'll look into those that you mentioned but I think we're going to go with Tek or Agilent. Took a look at your suggestions for the Keithley's LabTracer software and Series 2600 and it looks impressive. LabTracer sounds like LabVIEW but probably not compatible. We will probably want computer control of this test bench and our initial thoughts were LabVIEW but very preliminary at this point.

Pete B.
 
andy_c

andy_c

Banned
Joined 2003
andycpublic.50webs.com/
On a slightly different subject, remember we were talking about how the beta vs. current of my models for the MJL3281a/1302a did not drop off nearly as sharply at high currents as the datasheet?

Recently, I've been messing around with making the BJT parameter fitting more automated by using VBA with Excel to implement the Gummel-Poon equations in Massobrio and Antognetti. This was to come up with models for the 2SC4793 and the 2SA1837. I found the PSPICE model for the 2SC4793 and noticed that it had a parameter called NK. This parameter is not documented in either the LTspice manual or in Massobrio and Antognetti. I found the PDF of the PSPICE manual online, and sure enough, NK is described as the "high-current roll-off coefficient". They also show the equations used.

It turns out that for equation 3 on this page of mine, the form gets changed from:

qb = 0.5 * q1 * (1 + sqrt(1 + 4 * q2))

to:

qb = 0.5 * q1 * (1 + pow((1 + 4 * q2), NK))

The default value of NK is 0.5, making it the same as the SPICE2 equation as described in Massobrio and Antognetti (the first equation above). By altering NK, one can make the high-current beta rolloff steeper than with the more restricted SPICE2 formula. I tried messing around with this in LTspice, and even though it is undocumented, it is supported. MicroCap supports it as well.

So another mystery is (hopefully) solved, as well as demonstrating that Massobrio and Antognetti is out of date in some areas. Quasi-sat is another area in which it is out of date.
 
andy_c

andy_c

Banned
Joined 2003
andycpublic.50webs.com/
Sorry to be talking to myself here, but I've had a chance to experiment with the use of NK in the 2SC4793 model. Prior to incorporating NK, I was having quite a bit of trouble fitting both the ln(Ic) vs. Vbe and beta vs. Ic curves. Results were way off from the datasheets no matter what I did. After trying an NK value of 0.8 and increasing Ikf, I was able to completely nail beta vs. Ic and greatly improve ln(Ic) vs. Vbe.

It looks like the NK parameter is really the key to getting accurate simulations of beta in the high-current region. Also, since it leads to increasing Ikf as well, the interaction of Ikf with the forward current parameters for the low current region is lessened, making the parameter fit that much easier.
 
You must log in or register to reply here.
Top Bottom