bipolar (BJT) transistor families for audio power output stages

Compare the actual datasheet information for the MJL3281 and the 2SC3281 for those parameters most important for an output stage.


Peak ft for the MJL is about 60 MHz, while that of the 2SC is 30 MHz.

Now for the biggie: ft droop at high current and only 5V;
MJL is 30 MHz at 4A; 2SC is only 8 MHz at 4A.
ft of the 2SC3281 really crashes at high current.


Bob

how those Ft curves are misleading..at first, i thought " how tricky is the
semi industry" , but in fact, it s me who was dumb, as figures speak for themselves..
the Ft curves display the gain AS A FUNCTION OF THE CURRENT, this, at a frequency of 1 mhz, thus showing no more than the gain dispersion as a
function of the current...
a perfect device would have a straight line...

no mention that their datasheet says 30mhz typ. at 1A ,the curves show 50MHZ+ at this bias point...mystery of the us industry..but let s take it as granted an discuss using the higher figure...

what you see as qualities for the MJL3281 are in fact serious drawbacks..
if we look at the curves, we see that from 0.1 to 2A, there s a gain
dispersion of 150%....
the 2SC3281 has a 50% dispersion in the same range, which tell that
it s three time more linear....at 0.01 A , his 0.1A gain is reduced by a factor of more than 4...
of course, onsemi did stop their curve at 0.1A, but extending it show clearly that at 0.01 A, their device has a gain close to 1....in fact, they display only
the part of the curve that is marketable...a key figure for amps designer is thus missing....
indeed, these curves tell absolutly nothing about the switching speed
or the bandwith of the product, despite their (misleading) name...
remember that 2SC is a more than twenty years old device, and the MJL
has increased current capacity and TDP compared to the original..
the question was to compare with the 2SC5200, the actual general
purpose bjts fo amps, but even then, the old toshiba device is better than
the onsemi one in an amp designer point of view, not adding that they are actually faster than the "modern one"....

the onsemi, as proved by the curves are not linear at all, not suited to audio, and are good at switching, as they have low gain at low current and gain increasing dramatically with the current...lumba ogir is right on this point...

attached are the gain curves in fuction of frequency with a 4A current, in common emitter mode for a 5V Vce...impedance source is 2R, so the
Cob influence is reduced to little value...


regards,

wahab
 

Attachments

  • CE BW.zip
    5.5 KB · Views: 141
There are good known manufacturer for vintages power devices (2N series in TO-3 e. g.). One example is "MOSPEC" about
:: Welcome :: MOSPEC SEMICONDUCTOR
Who knows more about the quality standart ??
In this case I have also started this thread:
http://www.diyaudio.com/forums/pass...els-mospec-2n5876-2n5878-mj21193-mj21194.html

Independend of this, here (pdf attachement) one of the Case Studies of U.S.-Japan Technology Linkages in Semiconductors as read sample about Toshiba-Motorola - from the book
<< U.S.-Japan Strategic Alliances in the Semiconductor Industry: Technology Transfer, Competition, and Public Policy >>
 

Attachments

  • DOC Appendix A TOS MOT.pdf
    127.2 KB · Views: 150
AX tech editor
Joined 2002
Paid Member
using an ac source in sweep mode and running an AC analysis...
as the beta are different, i run three circuits that have optimised
base currents, so all the devices run exactly at 4.000A.....

OK thanks, so it was a simulation. Subject to accuracy (or lack thereof) of the models, of course. It looks too good to be true ;)

jd
 
the Ft curves display the gain AS A FUNCTION OF THE CURRENT, this, at a frequency of 1 mhz, thus showing no more than the gain dispersion as a
function of the current...
a perfect device would have a straight line...

No.

The fT variation of a BJT with current is described in equation (16) on this page. Note the dependency of fT on gm (listed as gmf, the forward gm, to distinguish it from the reverse gm). gm is given by Ic/VT, where VT=kT/q~26mV. The only way a BJT could have an fT that's independent of current would be if gm were independent of current, and that contradicts the physics of the situation. When gm becomes large, fT depends only on the forward transit time TF. But then TF starts to increase at high currents, causing fT do decrease above a certain point. If TF is made large (a slow device), then the variation of fT with current is decreased. If TF were zero, fT would be (almost) directly proportional to collector current.
 
Last edited:
No.

The fT variation of a BJT with current is described in equation (16) on this page. Note the dependency of fT on gm (listed as gmf, the forward gm, to distinguish it from the reverse gm). gm is given by Ic/VT, where VT=kT/q~26mV. The only way a BJT could have an fT that's independent of current would be if gm were independent of current, and that contradicts the physics of the situation. When gm becomes large, fT depends only on the forward transit time TF. But then TF starts to increase at high currents, causing fT do decrease above a certain point. If TF is made large (a slow device), then the variation of fT with current is decreased. If TF were zero, fT would be directly proportional to collector current.

what i said doesn t contradict your stances...
the question was : is the MJL3281 a good device for audio?...
i was responding to say that it s pointless to use the Ft as
used by onsemi as valuable norm to define wich transistor is good
and wich is not..
this doesn t tell us wich one has a high bandwith, since the test is
simply a measure of the gain at a fixed frequencyof 1MHZ..
this is as measure the gm(I) at this frequency...
 
btw, andy , can you explain me why the two devices have different
relative values of current increases part of the curves, since
the gm is linearly dependant of the current, at least in a part of the curves
, according to well established physic laws...

The easiest way to see that mathematically is on this page, equations (11) and (12). Equation (11) makes it look like fT is proportional to gm, but a closer look in equation (12) shows that Cbe in the denominator of (11) consists of the sum of two terms:

Cbe = gm * TF + Cje(Vbe)

So there's really an implied gm term in the denominator of equation (11). Let's neglect Ccb for the moment, as it will be much less than Cbe.

If the first term for Cbe above (= gm * TF) dominates as it does at high currents, then Cje(Vbe) can be neglected, and the gm terms in the numerator and denominator of the fT expression in equation (11) cancel (neglecting Ccb). In that case, fT is independent of current, having value 1/(2*pi*TF). So the slope really depends on the relative size of the two terms that make up Cbe above. If the second term dominates, fT will be nearly proportional to gm (and to Ic). I say "nearly proportional", because when Ic increases, Vbe increases also, which makes Cje(Vbe) increase somewhat.

Of course, real transistors don't approach a constant fT at high currents. Rather fT falls off once it hits the peak. SPICE models this by making TF itself vary with current.
 
Last edited:
Here's one other thought. Surely, the fact that fT varies with current indicates nonlinear behavior, right? But who's the culprit here? Let's imagine that Cbe were completely constant, independent of Ic and Vbe. Surely that's a requirement for linearity too. But equation (11) shows that for a constant Cbe and Ccb, fT is proportional to current! Indeed, the way to make fT independent of current is to arrange for Cbe to be proportional to current just as gm is. So the culprit of this variation of fT with collector current is the fact that gm is proportional to collector current.
 
I agree no simulation is going to tell you anything about these parts but one thing is for sure Onsemi datasheets have errors, Bob you have much better equipment than me, compare the capacitances as stated by onsemi, actually the the p cob is much higher as it should be, given the structure, little datasheet errors, please also keep in mind you compairing 30 year old japanese component to a recent onsemi one, and overall one shouldnt forget that the onsemi is a 1302 with a little bit bigger die, so will obviously have better SOA but worse cob. I can think of more modern bigger die japanese transistors which are better if you want to compare apples with apples, here youll see better SOA than ONsemi with half the cob. Also youll notice that japanese datasheets are very very accurate unlike the western datasheets which are very optimistic sometimes, use curvetracer and do some comparisons and youll get interesting results.

Onsemi devices are not bad, they are copies of the japanese with much better quality control than like comparing chinese manufactured copies.

Through the channels I get the japanese devices the onsemi would cost me more than double the price, the US goverment still have heafty own industry protection taxation on goods coming in from japan, through the channels I use they are on equal footing and this way the japanese parts are much cheaper so it all depends on where you live and if your goverment is trying to protect your own industry, on level footing its a very different ball game.

Theres some interesting things written about ,,buy local practices,, still being used in the US.
Through tiefbassuebertrlinks you can get some details about the toshiba motorola onsemi marriage.


I was just comparing the devices that were mentioned here as being the same device, and where the japanese part was alleged to be superior. What are the more modern japanese transistors that you are referring to? I always like to do apples-apples comparisons.

BTW, I don't have the perception that Japanese data sheets are more complete or more accurate in their curves. Just look at the 2SC3264 LAPT datasheet; just one page.

Crappy or not, I have not been able to get SPICE models for many Japanese transistors.

My bottom line is that it is still usually dangerous to make generalizations, even though all of us are occasionally guilty of it.

Don't get me wrong. The Japanese do great things very well. I drive a Lexus and will probably never buy an American car again, although that is probably unfair on my part.

Cheers,
Bob
 
No.

The fT variation of a BJT with current is described in equation (16) on this page. Note the dependency of fT on gm (listed as gmf, the forward gm, to distinguish it from the reverse gm). gm is given by Ic/VT, where VT=kT/q~26mV. The only way a BJT could have an fT that's independent of current would be if gm were independent of current, and that contradicts the physics of the situation. When gm becomes large, fT depends only on the forward transit time TF. But then TF starts to increase at high currents, causing fT do decrease above a certain point. If TF is made large (a slow device), then the variation of fT with current is decreased. If TF were zero, fT would be (almost) directly proportional to collector current.

Exactly! Indeed, the reduction of ft at low current is a reflection of the somewhat fixed portion of junction capacitance as ratio'd against gm.

BTW, in a power amplifier output stage, beta droop and ft droop at high current is a far bigger problem than at low current.

Cheers,
Bob
 
The only way a BJT could have an fT that's independent of current would be if gm were independent of current, and that contradicts the physics of the situation.

Oops. That statement is an error. If Cbe were directly proportional to collector current as it is in the mid-current region, and Ccb were negligible compared to Cbe, fT could be made constant with variations in collector current. This amounts to two nonlinearities canceling each other out. I probably should have said, "...independent of current in the low-current region...".