Modeling the Base Spreading Resistance

The transistor model that's normally used in LTSpice is the

and it does include an Rb, and apparently a parameter Irb.

NXP usually has the Gummel-Poon parameters for their transistors on their website. I've pulled them off from time to time.

LTSpice can also handle the VBIC model, but the IC stands for "inter company" and I've never been able to get hold of the parameters for the model of any transistor - they are "commercial in confidence" - and outsiders don't get them.

I use a fair number of those, and they're great parts. I'm slipstreaming them in as a replacement for the late lamented BFT25A. (They're not so very late-lamented round here, as we have 3 reels in stock, but for customer design-ins that isn't a big help.)

I've never done a careful A-B study of the model vs. a proto, though it would be a good thing to do.

Board-level models IME are never accurate enough to worry that much about.

Cheers

Phil Hobbs

Am 01.01.21 um 04:36 schrieb Artist:

Shouldn't this be the RB parameter in their spice model? RB = zero bias base resistance. It should be 1.17 Ohm then. That interacts with RBM / minimum resistance at high currents and JRB / current where it falls to half to its minimum value.

RBM and JRB are probably futile. Many parts used to be "characterized" with Orcad/Pspice parts.exe and that usually gave 5 Ohms or so IIRC.

I consulted Massobrio/Antognetti, but I'm only half awake.

cheers, Gerhard

Getrue [1] thought highly of SITCAP [2] decades ago. But, given your present predicament with model inadequacies, apparently nothing came from Getrue's high hopes. You've already discounted Ebers-Moll. And Paasschens [3] notes how Gummel-Poon won't work either. Paasschens advocates interpolation.

Perhaps you can create your own model to interpolate Fig 22 and Fig 23, from the BFU520A datasheet, in order to model noise, at least. Good luck! Please keep us informed of any progress.

Note.

Danke,

Not at board level, at least. Chip-level models are very good---they have to be, since no vaguely sane person is going to use a foundry they can't trust to build what they designed.

BJTs follow their noise models reasonably accurately. It's not necessarily the case that the same single value of Rbb' gives the best match to both the noise and high-frequency properties, though.

(Rbb' and Ccb set the collector-base time constant, which along with f_T sets the maximum useful frequency of the device.)

Cheers

Phil Hobbs

• Just add the resistor in series with the base. You can measure noise in the audio region of a (silicon) transistor over a wide current range, plot the result, and easily extrapolate to the multi-Mhz region and the nominal microamp region for a standard RF transistor. I have done this for RF transistors where there was no or scant data for a transistor made for RF work. The secret is, the GBW is roughly constant from microamps thru milliamps. Needs a bit of creativity for measuring in the microamp region, to then calculate the value of that resistor.

Twas a bit late 4 me, was not too awake and goofed.

GBW is rather linear WRT collector current, so if you measure it at one current you can calculate GBW at much higher or lower currents, 5 to

Spot NF is constant over current and rbb' even more so.

Both tricky to do in microamp and nanoamp region.

There's usually a very broad peak in f_T---the linear thing only applies at lowish currents, and is also affected by the rolloff of DC beta down there.

Cheers

Phil Hobbs

Exactly.