Greetings folks -

I wonder if there is a way to estimate any of the first order bjt parameters, such has dc beta or Ft for a model? Yes, a test bench could be created but I was wondering if one could read the model and do some basic mental arithmetic that would estimate how the model behaves. I am thinking, for example, of the bog-common 2N3904 which has a huge range of beta in real life. Knowing what beta the model maker chose for the model would help in doing the correlation between simulation behavior and bench behavior.

Many thanks
Jim Wagner
Oregon Research Electronics

On 22/02/2021 18:01, Jim Wagner wrote:

I wonder if there is a way to estimate any of the first order bjt parameters, such has dc beta or Ft for a model? Yes, a test bench could be created but I was wondering if one could read the model and do some basic mental arithmetic that would estimate how the model behaves. I am thinking, for example, of the bog-common 2N3904 which has a huge range of beta in real life. Knowing what beta the model maker chose for the model would help in doing the correlation between simulation behavior and bench behavior.

The two parameters (beta and Ft) you've chosen have no fixed value; they both vary with Ic, Vce and temperature as well as sample to sample. If you mean BF and not beta, then that's a different matter. Beta is the real-world value of Ic/Ib, which varies as described above. BF, on the other hand is a model parameter and fixed, but that doesn't necessarily make it equal to beta, as the "simulated" beta is affected by a host of other model parameters, like IKF, NK, IS and NF that modify how "beta" varies under different conditions. BF might equal the peak value of beta for some devices, but it might not for others. There is no one model parameter that defines Ft, although it is affected by TF. However, it is also affected by Ic and Cbe, which are also not model parameters but a physical ones.

But are you asking how model creators decide on what constitutes a typical device sample to base the model on? Or are you saying you want to minimally modify the default BJT to make it have a passing resemblance to a particular real device? It sounds a bit "hand-wavy".

--
Regards,
Tony

Sorry, I should have thought the question through, more carefully. Of course, beta and Ft depend on voltage, current, and temperature, I guess that what I’d like to figure out is where, in the parameter distribution the model makers chose to put the model. For example, the model COULD represent a minimum gain device or it could represent a “typical” device or any of a number of other choices. Rarely does the text of the model indicate this, so the user is left a bit blind about which of several models to choose for a particular device.

Thanks
Jim

I think Helmut's recommendation in this case would be to use the PSpice docs.? Helmut used PSPCREF.pdf which can be downloaded from many places on the net.? Then turn to the chapter on BJTs and read all about it.? There are pages of equations that describe how the Gummel-Poon model behaves.? But I think it does not answer your question directly; you still have to do the work.

The problem is that parameters such as Beta and Ft are best for making lab measurements of transistors, not for modeling them.? SPICE model parameters are for modeling them but not for measuring them.? There is some overlap but not a lot.

? ? "For example, the model COULD represent a minimum gain device or it could represent a “typical” device or any of a number of other choices."

In my experience, virtually all models for simple parts like BJTs are "typical".? The only time I've seen anything else, is when they were specified to be "best-case" and "worst-case" or other labels.? Without such a label, I would always assume typical.? But also remain skeptical because every model can be bad right from the start, or a poor selection from the distribution of parts, or an example of typical from 25 years ago before the process shifted.

Andy

And then there is the question of where your distributor got THEIR parts.? These days, some are rejects that fall off the datasheet.? Watch out for that.

Andy

开云体育

You mean like those “grab bags” that can be gotten for a pittance at a swap meet? ?Got a bunch of those and now they are mixed in with good stock :=(( Wonder where those that are marked with the Motorola logo might fall. What ever.

Bad 'black market' parts even fall into the mainstream distribution channels.? You need to know your sources very well.

Andy

Hi Jim,?

The approach I have taken is to gather the 2 port s parameters for a box full of say 3904. Then use the s data with a waveform arithmetic statement in LTSpice to get ft. As ft is an extrapolation of the low frequency gain on a 6dB per octave slope, it is pretty fast to evaluate. Recall, it is just a function of gm and Cbe; unity short circuit current gain crossing point. It does assist in validating the quality of some BIP models. I usually take the s data at the collector current where advertised ft peaks.?

Alan?

On 22/02/2021 20:47, alan victor wrote:

The approach I have taken is to gather the 2 port s parameters for a box full of say 3904. Then use the s data with a waveform arithmetic statement in LTSpice to get ft. As ft is an extrapolation of the low frequency gain on a 6dB per octave slope, it is pretty fast to evaluate. Recall, it is just a function of gm and Cbe; unity short circuit current gain crossing point. It does assist in validating the quality of some BIP models. I usually take the s data at the collector current where advertised ft peaks.

I'd be interested to know whose S-parameter data you've got for 2N3904?

Overall, your suggestion isn't worth much because S-parameters only model linear devices. The whole point about SPICE models is that they model the non-linear characteristics of devices. As a by-product they can also re-produce the S-parameters if the SPICE model is good. The converse is simply wishful thinking.

--
Regards,
Tony

OTOH, we can't talk about Ft without treating it as a linear(ized) amplifying device.

Andy

Hi Tony,?

Thanks for the feedback.?

I am afraid you missed my point. First, I never said I would use s parameters to emulate a non linear aspect of the model. However, the point of the validation exercise is to satisfy the criteria that the s parameters of the non linear model... Spice model... operating in a linear manner meet the s parameters of the measured device. Many of them do... spot on. And other models fail miserably.?

Second point, the use of s parameters actually can produce a very excellent STARTING point for successful non linear design. I have applied such techniques to both power amplifiers and one and two port oscillators.?

Again, I emphasize they provide an excellent initialized starting condition. That is to say, ?find embedding networks, matching elements and resonator development in oscillator applications.?

Alan?

开云体育

On 23/02/2021 02:24, alan victor wrote:

I am afraid you missed my point. First, I never said I would use s parameters to emulate a non linear aspect of the model. However, the point of the validation exercise is to satisfy the criteria that the s parameters of the non linear model... Spice model... operating in a linear manner meet the s parameters of the measured device. Many of them do... spot on. And other models fail miserably.

While we're missing points, you forgot to point to where S-parameters for the 2N3904 can be found. Of the thousands of different types of transistors in existence, probably fewer than 1% of them have published S-parameters.

Second point, the use of s parameters actually can produce a very excellent STARTING point for successful non linear design. I have applied such techniques to both power amplifiers and one and two port oscillators.

I agree about a starting point in the absence of non-linear parameters. But since these designs can only really be optimised by non-linear analysis, you need the non-linear parameters anyway - S-parameter analysis provides no information relating to how much power you're going to get, for example. Having said that, the great advantage of design by S-parameters is that linear analysis is orders of magnitude faster than non-linear analysis.

Again, I emphasize they provide an excellent initialized starting condition. That is to say, ?find embedding networks, matching elements and resonator development in oscillator applications.

-- 
Regards,
Tony

While many devices do not have S, they do have other parameter sets, h or y, ?and that requires conversion to S. Finally direct measure of S may be required.?For the case of the 3904, either measure the S directly or use h parameters and convert to S. Some y parameters exist for the 3904 as well and the same process, conversion to S is required. At that point, ?with an available non linear model, the comparison to measured as well converted to S data is needed before one can trust a non linear model.?

Instances exist where no model is available, either linear or non linear. Then measured data is required. Stated previously, beginning with S measured data does provide a gateway to a reasonable first pass approach to a non linear design. ?

Regards,?
Alan?

Alan wrote, "... they do have other parameter sets, h or y, ?and that requires conversion to S."

I think you meant to write "and that allows conversion to S."

I think Tony is still interested in the s-parameter data for the 2N3904.? Do you have a pointer to it?? Maybe when you wrote that you "gather" the s-parameters, you were speaking hypothetically?

Andy

Hi Andy,?

Yes, allows conversion to S. As far as the 3904, the data I have I measured from a variety of manufacturers.?
The conversion from manufactured y or h and a comparison to measured S of the same device in some cases is quite good and in others, not so.

Regards,?
Alan

Alan,

So, I guess you are unwilling to part with the 2N3904 data?

Why didn't you just say this?

Andy

Andy,

I'll be glad to share data.

As my time permits I will search for what I have and post. This was done quite some time ago.

Many of the files are on a HP Winchester CS 80? drive which was connected to a workstation and analyzer. So recovery and assembly?
of the data takes some doing.

Have a great day.?

开云体育

On 23/02/2021 15:44, alan victor wrote:

While many devices do not have S, they do have other parameter sets, h or y, ?and that requires conversion to S. Finally direct measure of S may be required.?For the case of the 3904, either measure the S directly or use h parameters and convert to S. Some y parameters exist for the 3904 as well and the same process, conversion to S is required. At that point, ?with an available non linear model, the comparison to measured as well converted to S data is needed before one can trust a non linear model.?

Instances exist where no model is available, either linear or non linear. Then measured data is required. Stated previously, beginning with S measured data does provide a gateway to a reasonable first pass approach to a non linear design.

I admire your dedication to S-parameters, but I can't help thinking going via Y-parameters (which are rarely given in modern datasheets) is more work than deriving from scratch a half-decent SPICE model directly from the datasheet. For example, the current On-Semi (was Fairchild) 2N3904 provides almost everything needed except an Ic vs. Vce plot. It even has some H-parameter graphs, but unfortunately magnitude only at 1kHz (except Hfe) , so not that useful.

-- 
Regards,
Tony

Tony,?

I agree.

Shame that the current On-Semi data does not provide an I-V plane plot.?

All that said, however, a circle back to small signal data set is still useful and unfortunate in many cases
that will require further measurements.?