Evaluation results: MJL3281A/MJL1302A SPICE Models

[andy_c]

Originally posted by jcarr
(...)So what do you do when you are interested in Early voltages and other parameters that are _not_ stated in the data sheet?(...)

Good point. I decided that I wanted to try getting as near to "data sheet clone" performance in my tweaked model as possible. So I started by trying to duplicate the so-called "Gummel plots" of ln(Ic) and ln(Ib) vs Vbe described in Massobrio and Antognetti. So I take plots of Ic vs Vbe and plots of beta vs Ic to reconstruct these. However, the Gummel plots must be done at Vcb = 0, and the data sheet plots are done with Vce = 20 V. So the first thing I need to do is "back out" the Early effect from the data to normalize it to Vcb = 0. So I need to know the Early voltage. As soon as I try to get it from the standard curve tracer plots, I see that they don't fit the theory of all intersecting at a point at all. Not even close. Then there's the reverse mode parameters like reverse Early voltage and so forth. That stuff isn't available from the data sheet at all.

This ended up being rather time consuming, so I decided to just tweak the AC parameters to try to get the ft vs Ic curves to match the data sheets as well as I can.

On a positive note, messing around with the formulas for Early effect modeling in Massobrio and Antognetti was really an eye-opener. I must admit that I previously had some confusion regarding the "down and dirty" specifics of the Early effect. But after pounding my head into the wall for a while on this, the light finally came on. I highly recommend this book to anyone interested in transistor modeling. It's really good.

[andy_c]

Here's an update of the model tweaking. I adjusted the CJE, TF, XTF and ITF parameters of the On Semiconductor MJL3281A model to get the ft vs Ic curve to be as close as possible to the data sheet. I wasn't able to emulate the very sharp dropoff of ft at high currents, so the best I could do was split the error. This makes the ft start dropping off at lower currents than the data sheet, yet the final ft value at the highest current is larger than the data sheet value. However, it's possible to get very good agreement with the data sheet in the low-current region by tweaking only CJE and TF. The final model is included below. I've also included a new plot of ft vs Ic that includes this modified model. The modified model is the yellow trace. Next step is to do a similar thing for the MJL1302A.

.MODEL Qmjl3281a_mod npn
+IS=6.5498e-11 BF=139.247 NF=1.00176 VAF=46.776
+IKF=10 ISE=7.75232e-12 NE=3.34341 BR=4.98985
+NR=1.09511 VAR=4.32026 IKR=4.37516 ISC=3.25e-13
+NC=3.96875 RB=11.988 IRB=0.111742 RBM=0.102914
+RE=0.00127227 RC=0.209833 XTB=0.115253 XTI=1.03146
+EG=1.11986 CJE=1.0531e-08 VJE=0.4 MJE=0.450375
+TF=2.6464e-9 XTF=1000 VTF=2.06045 ITF=175
+CJC=5e-10 VJC=0.4 MJC=0.85 XCJC=0.959922
+FC=0.1 CJS=0 VJS=0.75 MJS=0.5
+TR=1e-07 PTF=0 KF=0 AF=1

[millwood]

I read this whole discussion with great interest, and I have found in the past that some simulations didn't make a whole lot of sense but I didn't suspect models.

Anyway, back to the topic. an actual device can have many different behaviors under difference circumstance. The model has just limited number of degrees of freedom so there is a compromise (in mimicing a complicated set of bahaviors from a limited number of variables).

so maybe the ft-current relationship is sacrificed in the process for other relationships deemed more important by On Semi?

[Fred Dieckmann]

i do not remember when and what models I posted but would imagine I did.

Here are the ones that came with my Spice program. They are modeled with the AREA factor a have subcircuit to include the base spreading resistance I believe.

.SUBCKT QSC3281 1 2 3
* TERMINALS: C B E
* 200 Volt 15 Amp SiNPN Power Transistor 12-03-1991
Q1 1 2 3 QPWR .67
Q2 1 4 3 QPWR .33
RBS 2 4 9.5
.MODEL QPWR NPN (IS=1.63P NF=1 BF=150 VAF=254 IKF=12 ISE=1.34N NE=2
+ BR=4 NR=1 VAR=20 IKR=16.5 RE=22.1M RB=4 RBM=.4 IRB=5.556U RC=4.84M
+ CJE=481P VJE=.6 MJE=.3 CJC=312P VJC=.22 MJC=.2 TF=5.33N TR=204N)
+ XTB=1.5 PTF=120 XTF=1 ITF=9.6)
.ENDS


.SUBCKT QSA1302 1 2 3
* TERMINALS: C B E
* 200 Volt 15 Amp SiPNP Power Transistor 12-03-1991
Q1 1 2 3 QPWR .67
Q2 1 4 3 QPWR .33
RBS 2 4 9.5
.MODEL QPWR PNP (IS=1.63P NF=1 BF=130 VAF=254 IKF=11 ISE=1.34N NE=2
+ BR=4 NR=1 VAR=20 IKR=16.5 RE=12.1M RB=4 RBM=.4 IRB=5.556U RC=4.84M
+ CJE=1.09N VJE=.6 MJE=.3 CJC=708P VJC=.22 MJC=.2 TF=5.33N TR=204N)
+ XTB=1.5 PTF=120 XTF=1 ITF=9.6)
.ENDS


I have not examined them very closely to see how good the models are.

The genral form for a BJT model in Spice is:

QXXXXXXX NC NB NE <NS> MNAME <AREA> <OFF> <IC=VBE, VCE> <TEMP=T>

http://newton.ex.ac.uk/teaching/CDHW...ec3.html#3.4.3

[andy_c]

Originally posted by Fred Dieckmann Here are the ones that came with my Spice program. They are modeled with the AREA factor a have subcircuit to include the base spreading resistance I believe.

That's an interesting way of doing the model. It looks like this double transistor approach has a similar effect to the XCJC parameter used to model the collector-base capacitance being distributed across the base spreading resistor. From Massobrio and Antognetti, "The capacitance XCJC*CJC is placed between the internal base node and the collector. (1 - XCJC)*CJC is the capacitance from external base to collector, while CJC is the total base-collector capacitance." So it looks like this is the equivalent of XCJC=0.33.

Looks like these are the same models you posted earlier as the data agrees with what I had. I previously missed the subcircuit usage though.

[Fred Dieckmann]

"I previously missed the subcircuit usage though."


More likely I failed to include it. Thanks for you work of shedding light on transistor models. It should be very helpful to the "Spice Guys"

Fred

PS Do you do jfets...........

[andy_c]

Originally posted by Fred Dieckmann (...)More likely I failed to include it.(...)

Looks like you included it and I missed it. Here's the original thread:

http://www.diyaudio.com/forums/showt...227#post174227

As to the JFETs, I haven't gotten to that chapter yet. I suspect that eventually I'll try my hand at them.

[andy_c]

Here's the modified MJL1302A model as I mentioned I'd do. I used a similar approach to the MJL3281A model mods, adjusting CJE, TF and ITF to try to match the data sheet's ft vs Ic as closely as possible. As you can see, the original On Semiconductor models were already pretty close. The new model is the yellow trace and the data sheet values are the red trace. I got rid of the PSPICE model, as it was so bad that it messed up the graph scaling. This is really a minor tweak, but it may be of interest, especially to those trying to get the most accurate small-signal model possible.

.MODEL Qmjl1302a_mod pnp
+IS=3.25053e-12 BF=60.3363 NF=0.992063 VAF=19.8199
+IKF=7.18352 ISE=3.25712e-12 NE=3.42487 BR=5.15499
+NR=1.03617 VAR=2.77936 IKR=9.38159 ISC=2.5e-13
+NC=3.89405 RB=0.776136 IRB=0.0998107 RBM=0.776136
+RE=0.000613663 RC=0.0424163 XTB=1.43773 XTI=1
+EG=1.05 CJE=1.0690e-08 VJE=0.728073 MJE=0.42161
+TF=2.9458e-9e-09 XTF=1000 VTF=4.11586 ITF=380
+CJC=1.79861e-09 VJC=0.814822 MJC=0.473271 XCJC=1
+FC=0.8 CJS=0 VJS=0.75 MJS=0.5
+TR=1e-07 PTF=0 KF=0 AF=1

[andy_c]

I had a look at the MJE15030 and MJE15031 models as well, with the thought that I'd have to tweak those too. It turns out that the ft vs Ic of these models agrees quite well with the data sheet curves of same.

One thing that I found interesting was the rolloff of ft at high currents for the NPN device is much more abrupt than for the PNP. The SPICE model tracks this sudden change quite well. But the model for the MJL3281A did not compare nearly as well to its data sheet rolloff of ft at high currents, despite my having made numerous attempts to tweak this. What's going on here?

Thinking about this question made me realize something I hadn't thought of previously. The parameter I was using to adjust the falloff of ft at high currents is called IKF. This interacts with another parameter, TF (the forward transit time). For a "semi-ideal" transistor, the ft increases with current at low currents, then becomes approximately constant at a certain current level, reaching a maximum value of 1/(2*pi*TF). Now the purpose of IKF is to make the forward transit time a function of current, rather than just being equal to TF. This then causes ft to drop off again at high currents. So in a nutshell, by tweaking IKF, one can adjust the rolloff of ft at high currents. But what does this really do? Well, it only adjusts the bandwidth of hFE. The rolloff of ft at high current is also influenced by the decrease of the low-frequency value of hFE at high currents (since ft is the product of the low-frequency hFE and the -3 dB bandwidth of hFE). This effect is dealt with strictly by the transistor DC parameters. So suppose the vendor didn't accurately model the falloff of DC beta at high currents. Guess what, the ft falloff at high currents will never be right either, no matter how you tweak IKF.

So I suspect that the inaccurate modeling of the falloff of ft at high currents for the MJL3281A is due to inaccurate modeling of the falloff of DC beta at high currents. But unfortunately I didn't check that. I assumed they couldn't possibly get the DC stuff wrong. Sigh.

At any rate, users of the On Semiconductor models for the MJE15030 and MJE15031 needing accurate modeling of ft vs Ic can be confident that the models predict this variation quite accurately out of the box.

[Christer]

Andy, do you have a definition of the effect of the IKF parameter?
My semiconductor physics book is too old to include it, since it
was added in later Spice versions. Howver, the info i have
managed to find about it on the net and in various docs. does
not suggest that it models any frequency dependence. My
understanding is that its purpose is only to model high injection
(aka. beta drop), ie. the decrease in hfe at high values of Ic,
which AFAIK is not frequency dependent. It seems that IKF
is the value of Ic where hfe has dropped by 50%.

BTW, have you managed to model a beta drop that coincides
reasonably well with datasheets? What I mean here is a
hfe vs. Ic diagram. Although I haven't looked at the particular
BJTs you are considering, I have looked at a number of others.
None of the models came even close to the datasheets. I have
also DIY'ed Spice models for some transistors I couldn't find
models for, and there seems to be no way to get a beta drop
that is large enough to coincide with datasheets. IKF does not
have a strong enough effect, which is somewhat surprising
since I think it is intended to model the high injection equation.
Similarly, I also have not found any models or managed to
design any myself that model the saturation region of a BJT
well. Have you checked this for the BJTs you are considering?
This should matter, I think, for the crossover-distorsion
simulations discussed in another thread and which you (I think)
simulated.