BJT vs MOSFET - taking the best of both

[Nelson Pass]

I would be willing to bet that with a single amp
you'll see more measured distortion. I have seen
and built Zens with the lower transconductance devices
and they all measured greater distortion, and to my ear,
did not sound as good.

One of the trade-offs is the lower capacitance, so that
with lower transcondance devices you can use a higher
input impedance, and this sometimes helps with sources
that have trouble driving the low input impedance.

If you are driving the speaker with 2 balanced Zens
(or better yet 2 pairs of parallel balanced Zens) you
can get considerably less distortion due to cancellation
of the dominant second harmonic.

[Circlotron]

Originally posted by subwo1
One useful feature about JFETs is that the pinchoff voltage is below the source voltage for an N-channel device, and the other way around for P-channel. This characteristic could help in driving them as voltage followers up to the power supply rails. Otherwise, I believe they behave much like mosfets.

Are JFETs only small signal devices or is there such a thing as a power JFET too?

[jcarr]

Circlotron:

>is there such a thing as a power JFET?<

At least there used to be, from the likes of Yamaha, Sony and NEC. These devices were referred to as vertical power FETs, and date back to the mid-70s, right around the time that Hitachi was introducing its MOSFETs (which originally had a completely different designation than 2SK134/2SJ48 et al). As far as I know, there were a number of application issues relating primarily to the gate drive which made these vertical power FETs tricky to use and easy to break.

In most cases, the devices were featured in commercial power amps from their respective manufacturers, and I would not be surprised if the audio divisions in each company had a major influence in getting these devices produced.

A quick list of commercial power amps incorporating these devices would include JVC's JM-S7, Sony's TA-5650 and TAN-5550, Sansui's BA-1000, and of course the B-1 from Yamaha. I am sure that there are many more that I have forgotten.

A few years ago, Tokin started manufacturing a line of static-induction transistors (SIT), which behave very similarly to the older Sony, Yamaha and NEC devices. However, I think that Tokin was subsequently acquired by NEC, and I haven't kept track of what happened to the SIT lineup.

2SK60/2SJ18 (Sony), 2SK70/2SJ20 (NEC), and 2SK77 (Yamaha) are some of the device codes that I remember. Apologies for any memory lapses.

regards, jonathan carr

[MarcelvdG]

Apparently everyone here is convinced that power bipolar junction transistor device capacitances are smaller than power MOSFET device capacitances. This may be true when you compare the bipolar part's junction capacitances to the MOSFET's gate and overlap capacitances, but as soon as you forward bias a bipolar device, you get a large diffusion capacitance between base and emitter (this capacitance models the minority charge storage in the base). For a typical epitaxial-base power transistor with 10MHz fT biased at 1A, its value is around 600nF, much greater than the oxide capacitance of a normal power MOSFET. In fact, this is why the fT of the transistor is only 10MHz despite of its huge transconductance.

[traderbam]

Interesting argument, Marcel. You may be right about the base capacitance being very large: how did you arrive at these figures?

One thing that may be of more importance is the energy required to cause a change in output current. For example, for a FET and BJT with similar Pd and Imax, how much charge is required to change their output currents from, say, 1A to 2A in 1us? You would have to make some assumption about the change in collector-base/drain-gate voltage to take account of charging Ccb/Cdg.

Figuring out the gate charge is relatively easy because the datasheets usually show a graph of Id vs Q as FETs are most often used in switching applications. Trickier to find the equivalent for a BJT, perhaps.

[AKSA]

Marcel,

Accepted, but you refer only to base/emitter capacitance, which is not too significant, particularly in a common collector since the Vbe does not much change during operation.

Of far greater significance is the depletion capacitance across the base/collector, the so-called Miller capacitance, which is profoundly influential in the common emitter configuration.

Your comment is interesting, but less relevant in this context, since the gate capacitance of a mosfet is far higher wrt the drain, and this effectively mandates use of muscular drive, not so much a problem with a bipolar output stage and in any case ameliorated by use of a double emitter follower.....

In closing I'd suggest the propensity to self-oscillation of the mosfet introduces other problems, necessitating a gate stopper which to some extent negates the use of strong drive at the gate.

I apologize for the subjective comments; I don't have capacitance figures to hand, but am flying blind. I do know the gate capacitance of a P type IRF9140 is 600pF, and this capacitance is against the drain, a real PITA for a source follower.

Cheers,

Hugh

[subwo1]

Hello, I noticed I mistakenly said go with all MOSFETs. I meant use them in the power output stage.

[MarcelvdG]

Regarding Traderbam's question: in the following I will assume you bias your transistor in the normal forward active region, not in saturation or in reverse or anything funny. When you know a transistor's transconductance and fT for a given bias point, the sum of the base-emitter and base-collector capacitances is given by the equation:

cpi+cmu~=gm/(2*pi*fT),

where cpi is the base-emitter capacitance, cmu is the base-collector capacitance, gm is the transconductance, fT is the extrapolated frequency at which the current gain drops to unity and pi is 3.14159265358979....
Ideally, neglecting some second-order effects which occur at high current densities, the transconductance of a bipolar transistor equals:

gm=IC*q/(kT),

where IC is the collector bias current, q is the electron charge (1.6022E-19 C), k Boltzmann's constant (1.38065E-23 J/K) and T is the absolute temperature.

Assuming 10MHz fT, 1A bias current and kT/q=26mV (which is true just above room temperature):

gm~=38.4615 S,

cpi+cmu~=612.134nF.

The collector-base capacitance cmu is just pure junction capacitance, the base-emitter capacitance cpi is a combination of junction and diffusion capacitance. So after subtracting the junction capacitances, you are left with the base-emitter diffusion capacitance.

Regarding Hugh's/AKSA's remarks, it is certainly true that capacitance between base and collector is much worse than an equal capacitance between base and emitter. This does, however, not mean that diffusion capacitance is necessarily negligible.
For example, without any cascoding, base-collector capacitance is gm*Rload+1 times as bad as base-emitter capacitance, due to the Miller effect in common-emitter stages or the bootstrapping effect in emitter followers. When gm=38 siemens and the load resistance Rload=8 ohm, gm*Rload+1=305. But a 600nF cpi then still has a comparable influence as a cmu of 600nF/305, or almost 2nF. The situation gets worse at lower load resistances.

To make matters worse, I think that in most practical cases in audio amplifiers the static base current required by bipolar output transistors is even larger than the capacitive currents. Of course you can solve that by using double emitter followers, at the expense of a more complicated high frequency behaviour of the feedback loop, but you could do the same in an amplifier with a MOSFET common-drain output stage as long as gate stoppers don't spoil the fun.

By the way, has anyone ever tried RC damping networks between gate and source instead of gate series resistors to stop parasitic oscillations in power MOSFET's? I mean connecting a resistor in series with a capacitor and then putting the whole thing between gate and source. I haven't tried it, but I have often used circuits like that in unstable RF IC's.

Yours sincerely,
Marcel

[subwo1]

Hello all again. I have not noticed mention of an advantage of using mosfets instead of BJTs as class A output devices. They are much more tolerant of operating at high temperatures than bipolar transistors.

[x-pro]

Hi all,

for 10 years I was making amplifiers with N-channel only MOSFETs in the output and really like these devices. Main reason is that the modern "switching" logic level MOSFETS, like BUK555-100 and later similar devices, are incredibly tightly matched inside one batch (less than 1% difference in capacitances, threshold voltage and transconductance). So you have nice symmetry for push-pull without spending much money (BUK555 was about $1 each) . The main advantage of the MOSFET ouput stage, in my view, is that it has low output impedance even without NFB, and does not change the load of the voltage amplifier as much as BJT output does with the load change.

Al